MQA is Vaporware

[Shadders]

Just now, Lee Scoggins said:

 

Thanks. Links?

Hi Lee,

One forum i am aware of - and the site provides lots of free educational material :

https://www.dsprelated.com/forums

 

Others that are popular :

https://dsp.stackexchange.com/

https://forums.ni.com/t5/LabVIEW/DIGITAL-SIGNAL-PROCESSING/td-p/2553427

 

Some forums expect people to have some level of technical capability - so you may get redirected.

 

Regards,

Shadders.

[Shadders]

1 hour ago, FredericV said:

Quoted Text "Bob Stuart’s temporal de-blurring filters that eliminate one of the biggest monsters of bad digital: pre and post ringing."

Hi FredericV,

Thanks for this - this establishes what temporal blur actually is, when MQA state it - pre and post ringing of a filter.

 

Did this quote come from the MQA site, or is this someone else's statement of what they think MQA temporal blur is ?

 

Thanks and regards,

Shadders.

[Shadders]

5 minutes ago, FredericV said:

https://parttimeaudiophile.com/2018/01/09/mqa-a-fresh-take-why-the-big-labels-are-converting-their-catalogs/

 

When you kill all pre-ringing and limit postringing to 1 cycle, you get aliasing. This filter sounds very tight with bass, and the attack of the bass kicks more, but you miss a sense of depth and sounstage is thin.

This is both observed when using SoX as described in part 4 here:
 

http://archimago.blogspot.be/2017/12/howto-musings-playing-with-digital_23.html

 

and when listening to MQA on a Mytek vs the source DXD.

It sounds like the decay of instruments and voices are shortened with less depth.

Using Archimago's intermediate phase filter with a tiny bit of preringing and mainly postringing, this proves independent researchers can design a better sounding filter than corporations like Meridian / MQA.

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

This filter just sounds so more fluent and natural than MQA's filter. The decay of instruments and soundstage is better. And the best part: filter can be implemented using open source.
 

Hi FredericV,

Thanks. I have not read the articles, but, i thought that the temporal blur has already occurred to the master file - and as per the MQA wikipedia listing, that MQA had used a image sharpening algorithm (http://icms.org.uk/downloads/BtG/Dragotti.pdf) to removal the "temporal blur" - or is the wiki mistaken ?

 

If all MQA is, for the claimed correction of temporal blur, is a different filter applied to the existing master (MQA encoded) - then this is a complete con. It is irrelevant whether SoX or anyone else can produce a better filter - since MQA is committing fraud. (temporal blur still exists in the MQA file).

 

Is there any evidence that they process the master file to remove the temporal blur ?

 

Regards,

Shadders.

[Shadders]

2 minutes ago, FredericV said:

 

 

I believe MQA claims that you'll need to use products which can do the second unfold (=upsampling + dithering + killing postringing).

 

 

What I believe is happening is the fact that MQA has mapped all kind of transient response errors into a map of 32 pre-defined filters, where the encoder detects which one will be the best estimate, and the first unfold passes this parameter to the second unfold, which is what makes the MQA sound thinner than the original master file that was used to encode the file.

 

As MQA is a black box, we will need further reverse engineering.

Hi FedericV,

OK- thanks. So, we will never know until reverse engineered whether they are implementing any processing on the master file. The references to image sharpening study may be just a red herring - to make it look involved and detailed - subterfuge.

If the SoX filters at least have the same subjective effect, then MQA is not required.

Maybe DAC manufacturers can implement the SoX filters and claim pseudo-MQA without the cost, DRM, and lossy coding, and phones have an app that does the same.

Regards,

Shadders.

[Shadders]

4 minutes ago, daverich4 said:

 

Chris, I don't know if you can see my activity down to what I click on and read but if you can you'll see that when I do look at an MQA thread I skip large chunks of what has been posted since the last time I looked at the thread. Most of what information about the topic is expressed is covered in the first 20-30 pages. Repeating it over and over again over hundreds of more pages doesn't really serve any additional purpose. One big downside of this is that Industry representatives are slammed the minute they show up. Lee Scoggins has allowed himself to be a punching bag way more often than I ever would have. I frequently don't agree with what is published in Stereophile and TAS but I still subscribe to them both to keep up with what those in the industry are saying. I see value in that but apparently you and many of the posters here don't. So be it.

Hi,

The repeats are because different people post on the forum at different times.

When i read the posts by people who promote, or are for MQA, then they do not seem to engage with the actual topic in detail.

 

There is no technical sparring - where one criticism of MQA is challenged with evidence addressing the criticism. The visitors repeat the standard MQA technical statement, but they do not seem to understand the detail, so cannot engage in the discussion at the technical level.

 

This is frustrating to people who have worked to determine the details of MQA, and realise the false claims, and have proven them to be false.

 

Regards,

Shadders.

[Shadders]

51 minutes ago, Fokus said:

 

Sigh.

 

Deep sigh.

 

Very deep sigh.

 

Where does all this nonsense come from?

 

What MQA does is really rather simple.

 

Example:

 

Given a 192kHz recording. It will have ringing at 96kHz, because it has been produced with an industry-standard half-band AA filter in the ADC.

 

MQA will downsample this to 96kHz. After all, 96k is all that can pass through their pipe.

For this operation they do not select a standard AA filter, but a tailored one. Their filter will reach null at 96kHz (to kill the recording ADC's ringing), but otherwise it will be very leaky, i.e. it causes aliasing in the target 0-48kHz band. How much aliasing? According to their original literature (and f...ing please go and read the AES papers, instead of coming here with new fabrications every time), the intention is (was?) to select a filter such that the aliasing that falls below 20kHz always remains below the recording's noise floor (or a similar criterion). (This selection was to be done by humans, on a track by track base, ideally under supervision of dead artists, i.e. the touted white glove treatment.)

 

Thus they obtain a 96k file. This can be folded to 48k and back, in what we know is conceptually a lossless operation (in reality it is lossy, because the 24 bits of the container have to be distributed over the two bands).

 

Upon playback, after unfolding to 96k, the stream is upsampled to the original 192 by once more a leaky filter, one of the 32 that Mansr/Archimago documented. These rendering filters add little or no ringing, and according to the original plans each filter ought to be specifically tailored to the DAC in use. (But we know now that this is not happening, as totally different DACs have been seen using the same filter coefficients*.)

 

Voila: a 192k recording is passed through a 48k channel without having been subject to steep AA or AI filtering.

 

 

For masters of higher resolution the same applies, but more.

 

For masters of lower resolution the above does not work.

 

For a 96k master (ringing at 48kHz) they presumably** insert a short low pass filter, thus removing the 48kHz component (some would call this apodising).

 

For a 48/44.1k master they cannot do this, because the result would show a reduced treble in the audible band. They do have a funny patent about applying an all-pass filter with severe phase lag above 20kHz, so that any pre-ringing is literally taken up and dropped down again after the main impulse. Oh, the ways these people go in their chase of ghosts!

 

 

Oh, and what about all these MQA references to sparse sampling, new insights since Shannon, triangular kernels, ...   Red fish. Smelly red fish.

 

 

(* And now we have dCS crying that they really are doing something really very special, you have to believe us!)

 

(** When I analysed digital rips of Tidal signals I forgot to look into this aspect.)

 

Hi Fokus,

It is not fabrication, but maybe misunderstanding.

 

Re wiki : "The audio stream is sampled and convolved with a triangle function, and interpolated later during playback. The techniques employed, including the sampling of signals with a finite rate of innovation, were developed by a number of researchers over the preceding decade, including Pier Luigi Dragotti and others.^[11]"

 

Are you stating that MQA does not implement the process as per the wiki, i.e. the wiki may need revising ?

 

The presentation as per Dragotti indicates that their method is used for removal of blurring - slide 4.

 

Regards,

Shadders.

[Shadders]

17 minutes ago, Fokus said:

Shadders, do you know a red herring when you see one?

 

Hi Fokus,

Yes - i know what a red herring is. As per my previous post on page 261:

 

"So, we will never know until reverse engineered whether they are implementing any processing on the master file. The references to image sharpening study may be just a red herring - to make it look involved and detailed - subterfuge. "

 

That is why i asked the question regarding the wiki, which you responded to. Since no one knows what the processing entails, it is possible that they have implemented the algorithm/process.

 

We cannot discount the information, as we do not know what the MQA process is in detail - someone stated it was a black box.

 

Brian Lucey stated that there are harmonics in the MQA results - pleasing to the ear ? Is this a by product of the stated convolution with the triangle function ? (we do not know).

 

Regards,

Shadders.

[Shadders]

30 minutes ago, Fokus said:

 

Convolving with a triangle is filtering. Any filter has a time domain representation, and their leaky filters will have fairly simple representations. But the referral to the story of sparse sampling and image deblurring and whatever, that is bullshit. Just say that you use a weak filter, don’t dress it up to make it look sciency.

 

Hi Fokus,

That is the issue - we do not know. We have no inside knowledge of the process of MQA, and the process artefacts may be harmonics. MQA may use a derivation of the deblurring study.

 

The AES paper i have located is : http://www.aes.org/tmpFiles/elib/20180112/17501.pdf

 

Reference [23] is the Dragotti paper, and this is referred to in Section 4.1, page 9. It does not state it is definitely implemented, but section 4.1 indicates that it is part of the MQA process - else, why describe such an implementation.

 

Even so, any such process is just an effect, an does not guarantee a perfect result (i.e.what the artist heard), but we should not ignore it, although it makes life more difficult to challenge MQA.

 

Regards,

Shadders.

[Shadders]

1 minute ago, crenca said:

 

Um, the X axis is the signal db & the Y is signal's frequency - standard audio signal stuff...perhaps you are asking something else?

Or,  or....

The X is the frequency, and Y is the signal power in dB.

[Shadders]

Just now, crenca said:

 

Ha...somehow I could still edit!

Of course, i couldn't wait the obligatory 30 minutes.

[Shadders]

16 hours ago, Fokus said:

(and f...ing please go and read the AES papers, instead of coming here with new fabrications every time),

Hi Fokus,

As per your comment, i have read the literature : http://www.aes.org/tmpFiles/elib/20180113/17501.pdf

 

Section 4.1 lists the relevance of the Dragotti paper - page 9.

 

Therefore my referral to the wiki is correct.

 

Have you challenged the AES with their mistake in endorsing the MQA paper ?. That is, the MQA paper in listing and stating the MQA process uses the Dragotti paper (and others), is a false statement by MQA ?.

 

Have you written to MQA (Bob Stuart and Peter Craven e-mail address is on the paper) to notify them of this mistake ?

 

Thanks and regards,

Shadders.

[Shadders]

16 hours ago, Fokus said:

Given a 192kHz recording. It will have ringing at 96kHz, because it has been produced with an industry-standard half-band AA filter in the ADC.

 

MQA will downsample this to 96kHz. After all, 96k is all that can pass through their pipe.

For this operation they do not select a standard AA filter, but a tailored one. Their filter will reach null at 96kHz (to kill the recording ADC's ringing), but otherwise it will be very leaky, i.e. it causes aliasing in the target 0-48kHz band.

Hi Fokus,

I am new to DSP aspects here - so i have a few questions.

 

When i examine the spectrums of those presented in Hifi News, i do not see the energy spike which would be the ringing of the filter used in the ADC.

Why is that ?

 

Why is the ringing exactly at 96kHz, if the ADC filter cut off is at perhaps 86kHz ? (or another value)

 

Regards,

Shadders.

[Shadders]

2 minutes ago, Fokus said:

 

1) because 'ringing' is not a distortion of the system's frequency curve or steady state signal spectrum. It can only be seen in amplitude-versus-time, or (slightly) in spectrum-versus-time.

 

2) the majority of ADCs are half-band and thus cut at (about) 96kHz.

Hi Fokus,

If the ringing can only be seen in the time domain, then that means that the ringing energy is in the signal.

 

That then means the FFT will extract/present this energy at the frequency of the ringing. So we should see it - it will be in the FFT output.

 

Why do we not see it in spectral plots as per Hifi News or others ?

 

Regards,

Shadders.

[Shadders]

12 minutes ago, Fokus said:

3) As for challenging the papers. Why??? It is not even a given that these were ever reviewed before acceptance. I don't know the standards for AES journals.

 

When I started my academic carreer the first advise I got was: and remember, papers are mostly lies. It is even worse today than back then. This said, I never knowingly lied in any of my papers ;-) Also not implying that MQA are lying. Just telling you that a paper is not necessarily authoritative.

Hi Fokus,

What evidence do you have that MQA have NOT implemented the process/algorithm as per the Dragotti paper ?

Thanks and regards,

Shadders.

[Shadders]

12 minutes ago, Fokus said:

MQA may or may not use a triangle in some cases. That is not important. What is important is that these filters are leaky, and that MQA count on the aliasing falling below the noise. The triangles are not magic, the theory in the Dragotti paper bears no relevance to audio, and the blurring on the image in the D paper has nothing to do with the blurring of an audio master in the MQA story. That's why these are red herrings. It all looks impressive, but it has little or no relevance.

 

As an aside, bringing lore from image processing to audio is generally a bad idea. Image processing is resource constrained, audio processing is not. Therefore much of what happens in image processing is about efficiency and finding computational shortcuts, not about absolute quality of result. In audio you can run a Sinc filter. In imaging not even remotely.

Hi Fokus,

If MQA are using a triangle kernel, and this is not taken into account in your analysis of MQA, then your analysis is insufficient ? You are missing something ?

 

Although image processing such as a filter may be a 2D spatial process, the edge enhancement, or deblurring, still happens in both dimensions - X and Y. So, is it not therefore possible to adapt a 2D algorithm to a 1D domain, such as an audio file, to implement the same process (albeit adapted) ???

 

As you have admitted, you do not know what the internal MQA processes are (black box), so maybe you cannot state MQA does not use the Dragotti paper etc.

 

Regards,

Shadders.

[Shadders]

Hi Bob,

Thanks - i have not seen, but as per first few paragraphs, there is no energy at the cut off frequency of the filter - so ringing does not occur in the real world. I will read more thoroughly later - thanks

Regards,

Shadders.

[Shadders]

39 minutes ago, crenca said:

I thought it was a little amusing that Lee suggested that mqa algorithm "requires" encryption.   Is there any such thing in the wild, a computational algorithm that "requires" encryption for it to be solved?

Hi,

Bitcoin requires digital signing using public and private keys :

https://www.coindesk.com/math-behind-bitcoin/

Regards,

Shadders.

[Shadders]

Just now, Samuel T Cogley said:

 

The whole point of Bitcoin is anonymity through encryption.  I think @crenca's point is how does audio DSP require encryption?

Ahhhh, but he never specified that.......

[Shadders]

3 minutes ago, crenca said:

 

Well, my point was that does there exist a computational algorithm, for which to be completed, "requires" encryption.

 

Well of course, and that would be encryption algorithms!  However, everything else (including MQA's encoding scheme) would not "require" encryption because the point is not encrypted itself.  Encryption is present in MQA to serve the purpose of "management" of end users (and DAC manufactures, mixing engineers, etc.) digitally.  Lee is suggesting, ostensibly because he was told by MQA's PR people but I actually doubt that - he is just making it up as he goes along, that somehow the folding process of MQA would not work without it being buried behind encryption.

 

I am not sure even Bitcoin qualifies (could be wrong), because in a world populated with angels (as opposed to men), bitcoin could work without the need for encryption....

Hi,

Hmmm, scrambling a data stream requires a pseudo random generator - which is the basis for encryption. Not sure if the scrambled data stream is scrambled or encrypted.

 

Of course MQA operation does not require encryption since it is possible to just scramble the data stream, so every knows the key (to prime the PRNG/PRBS), and anyone can decode.

 

Regards,

Shadders.

[Shadders]

2 hours ago, Tony Lauck said:

When encoding information that is correlated to the music (and this includes the folded high frequency content) the encoded information must be scrambled or "whitened" into pseudo-noise to convert distortion artifacts into noise artifacts. This can be done using pseudo-random number generators which involve no cryptographic algorithm, it can be done using cryptographic one-way hash functions which accomplish the same effect with more confidence, or it can done with encryption algorithms which use a key to control access to the encoded information.  All three of these methods will achieve the same technical result from an audio quality perspective, albeit with different hardware implementation costs.

Hi Tony,

This would not be possible - as you have stated, it is a one way hash function, and the data that goes in, is not reversible (in practice), and the hash output is generally a lot smaller than the input data stream.

Regards,

Shadders.

[Shadders]

1 hour ago, Tony Lauck said:

Snippet "can be done using cryptographic one-way hash functions which accomplish the same effect with more confidence"

 

I was discussing the need for the scrambling function needed to whiten the low order bits that represent the folded high frequency information.  Since these bits appear in the code space for the undecoded playback they need to appear to the receiver and listener as uncorrelated with the music.  That way, they will be heard as random noise, rather than distortion.  (It's actually slightly more complicated because changing the low order bits to random values will still be correlated to the music in the form of noise modulation, but these are details of the dithering algorithms and are unrelated to the method of generating the pseudo-randomness.)

Hi Tony,

A cryptographic hash function does not scramble. A cryptographic hash function operates on an input data stream (such as a file) and produces a finite output which is the same size whatever the data stream input size is - examples are MD5, SHA etc.

 

With regards to randomness, Sample Rate Converters dither using the Triangle Probability Density Function. Essentially, dither is based on a pseudo random algorithm, but the randomness has a probability density function (specific distribution) which can be triangular, gaussian (white noise) etc (see : https://en.wikipedia.org/wiki/List_of_probability_distributions the section "Continuous distributions"). Weakness in the algorithm (pseudo random) can expose encrypted links etc., to attack.

 

Regards,

Shadders.

[Shadders]

3 minutes ago, Samuel T Cogley said:

 

This is not correct in my experience in using these hashes.  For example, an SHA512 hash returns a 512 bit (64 byte) value regardless of the size of the input stream:

 

host $> openssl sha512 01\ Cleveland\ Orchestra\ -\ Symphony\ No.\ 1\ in\ B\ flat\ major\ \(\'Spring\'\)\,\ Op.\ 38\;\ 1.\ Andante\ un\ poco\ maestoso\ -\ Allegro\ molto\ vivace.flac
SHA512(01 Cleveland Orchestra - Symphony No. 1 in B flat major ('Spring'), Op. 38; 1. Andante un poco maestoso - Allegro molto vivace.flac)= 5def2b4e3fc826a005b5bc5795b9e07b7f3007ef229bf5c354cdda4dcaad2593c70a519a13f36ca32f4432699a2f0a0a99e2abf844406f994b04c044f94056aa

host $> ls -l 01\ Cleveland\ Orchestra\ -\ Symphony\ No.\ 1\ in\ B\ flat\ major\ \(\'Spring\'\)\,\ Op.\ 38\;\ 1.\ Andante\ un\ poco\ maestoso\ -\ Allegro\ molto\ vivace.flac
-rwxrwxrwx  1 user  group  53218211 Nov 16 18:31 01 Cleveland Orchestra - Symphony No. 1 in B flat major ('Spring'), Op. 38; 1. Andante un poco maestoso - Allegro molto vivace.flac

Note that the input file is 53MB, and the hash is 64 bytes.

 

What you're describing sounds more like a One TIme Pad.

Hi,

What i have stated is correct - the hash is the same size, whatever the input stream size.We are in fact, in agreement. The size is finite = constant size = same size.

Regards,

Shadders.

[Shadders]

Hi,

Referring back to the Hifi News opinion piece about MQA on CD. I have not seen any negative statements about MQA. It does state that there is aliasing above 48kHz, but no negative impact statement.

 

Hifi News is a popular magazine in the UK - and it does provide technical studies/reports, but again, nothing negative about MQA is printed.

 

If the press continue to be uncritical of MQA, then the readership will eventually believe what they are told. It is after all subjective reviews that people take away from a magazine review.

 

And we all know that audiophiles are easily swayed as to the latest and greatest - cables, cable lifters, special USB electronics etc. Many people need to be told what they should be hearing.

 

Peoples understanding and evidence about the shortcomings of MQA are scattered in the pages across this forum. If one were to direct anyone to this site, then the relevant information is so sparsely distributed, that all people will eventually take away from the MQA threads are the exchanges (sometimes vitriolic) between people rather than the information content.

 

I think this has been said before, that what is required is the centralisation of the technical criticisms of MQA in an easily readable format. This can be a thread which is added to as new aspects and technical understanding evolves, and does NOT include exchanges of views which detract from the science. The statements in the thread MUST be backed up with evidence, and not rhetoric.

 

Then all that is required (hopefully) is the direction to the thread for anyone who wants to see the details - which can include referrals to other sites that expose MQA for what it is ?

 

Or should we just link to : https://www.xivero.com/downloads/MQA-Technical_Analysis-Hypotheses-Paper.pdf

 

Makes life a lot easier.

 

Regards,

Shadders.

[Shadders]

14 minutes ago, Fair Hedon said:

I find it shocking he has not been banned for telling someone to F OFF, and  calling another person a J $$$.

Hi,

If you ban Lee, then you won't be able to contradict statements.

 

He does have valid points in some of the posts.

 

Regards,

Shadders.

[Shadders]

5 minutes ago, Fair Hedon said:

I fail to see his valid points, and if they are there, they are wrapped in a bouquet of misleading statements.

HI,

A quick one without having to trawl through all the posts - he stated if he likes the sound of MQA then he will purchase - same for others throughout the world.

 

Despite MQA shortcomings, and technical inferiority to other high resolution files, if people like MQA they will buy it. People purchase MP3 as opposed to CD - despite the superiority of CD.

 

Regards,

Shadders.