MQA pt II

[DimitryZ]

OK, in case it helps I've just modified the contents of the zip at
http://jcgl.orpheusweb.co.uk/temp/IQResults.zip
I've added filetype extensions to the files and also added a short text file saying what the other files contain.
The data is essentially all CSV. The program is a 'C' one.

Sorry for any confusion. I'm used to being able to DnD any file into a general file editor and it will show it as 'text' regardess. The graphs I put up were all pngs so should be fine anyway.

Output0.csv contains four columns. The first one is likely frequency. The next four are columns of identical numbers (252.9297. 0. 252.9297. 0.). I don't think this is spectral output

In the Output1.csv, the four columns are variable and can be plotted. What are their units and what GO cases do they represent? It's a good idea to add this information into the excel file.

Since this is your work, isn't it better for you to do your own plots (with axes labeled) and make them available in a format that is easily displayable? I suspect there is something different in the Unix png definition...

This is the plot of your output1 file, without the last column, which appears to be phase. Is this what you intended us to see and what are we looking at? The grey line is input against itself, I presume. I plotted it as log-semilog, since the Y axis obviously wasn't log.

https://drive.google.com/file/d/1ipWotvJbWpvqNXmLZ7Civ56R47EYA8gF/view?usp=drivesdk

 

[Jim Audiomisc]

FLAC is way less effective as the recording frequency increases. It is grossly less effective at the 352 DXD frequency, which is loaded with random ultrasonic noise - the current darling of the "stupid lossless" crowd.
.

That's why it makes sense to noise shape and cut down excess bits under the 'sea of noise' first as I've been pointing out.

DSD (SACD) only 'works' because it does this.

Once you use the shaping and sea depth reduction flac becomes more efficient. However it also then shows that in some/many cases there is 3/4 of SFA in terms of real 'ultrasonic' information anyway. 8-]

 

[Jim Audiomisc]

MQA offers one decoding process, which it calls unfolding.

Perhaps you can reach out to GO directly. If I was doing this work I would want to know exactly how these files were handled and if they went through an ADC, what equipment he used, etc. But that's me.

Perhaps someone who is a member at ASR (paging mansr) can invite GO to join PFM.

The MQA patents make clear that they offer *two* methods in combination:

1) The 'we-must-not-call-them-leaky' resampling filters.

2) The use of a set of (undefined in the patents) 'primitive shapes' to represent some of the features lost at HF.

For all I know (2) are as much of a boojum as HDCD's claimed 'filter switching' which was in the patents but never implimented, though. Alas, layers of Secret Sauce tend to make it hard to find out!

Given the 'provenance' problem I suspect the only way I can know I'm talking to GO and not someone else would be for me to join, or someone I can trust to act as a checkable go-between to confirm our identities. I may give up at some point and join ASR, and then may leave it again later once this is all dusted. But at present I'm simply analysing what we have and will report. That takes time and effort in itself and my mind is mainly on analysis... and cooking dinner.

 

[tuga]

Output0.csv contains four columns. The first one is likely frequency. The next four are columns of identical numbers (252.9297. 0. 252.9297. 0.). I don't think this is spectral output

In the Output1.csv, the four columns are variable and can be plotted. What are their units and what GO cases do they represent? It's a good idea to add this information into the excel file.

Since this is your work, isn't it better for you to do your own plots (with axes labeled) and make them available in a format that is easily displayable? I suspect there is something different in the Unix png definition...

What browser are you using? Can you see this:

 

[DimitryZ]

What browser are you using? Can you see this:

Chrome in windows and yes I can. Also visible in Opera in Android.

 

[DimitryZ]

That's why it makes sense to noise shape and cut down excess bits under the 'sea of noise' first as I've been pointing out.

DSD (SACD) only 'works' because it does this.

Once you use the shaping and sea depth reduction flac becomes more efficient. However it also then shows that in some/many cases there is 3/4 of SFA in terms of real 'ultrasonic' information anyway. 8-]

Jim, that is your well-intentioned wish list, not actual reality.

And if you filtered the 352 DXD to something reasonable like 96 that can be efficiently FLACd, the "stupid lossless" crowd will scream "don't touch my lossless master, that's what MQA does!"

 

[DimitryZ]

The MQA patents make clear that they offer *two* methods in combination:

1) The 'we-must-not-call-them-leaky' resampling filters.

2) The use of a set of (undefined in the patents) 'primitive shapes' to represent some of the features lost at HF.

For all I know (2) are as much of a boojum as HDCD's claimed 'filter switching' which was in the patents but never implimented, though. Alas, layers of Secret Sauce tend to make it hard to find out!

Given the 'provenance' problem I suspect the only way I can know I'm talking to GO and not someone else would be for me to join, or someone I can trust to act as a checkable go-between to confirm our identities. I may give up at some point and join ASR, and then may leave it again later once this is all dusted. But at present I'm simply analysing what we have and will report. That takes time and effort in itself and my mind is mainly on analysis... and cooking dinner.

Whatever they do in the decoding process is simple, since it can be done in Windows, IOS, Android and DAC firmware on the fly. Once the decoded file is LPCM, they attach the simple upsampling rate and filter choice (there are 32 of them and mansr hates them all) and hand it off to the DAC. Encoding is likely much more involved.

I would not advise to take a smug posture and expect that in an audio engineering competition with Bob Stuart (one of the heavyweights in audio signal processing for decades and the man behind Dolby TrueHD) you will easily show him up to be an incompetent fool or a clever fraudster. My understanding is that your work in audio is limited and it's well a known phenomenon that folks from adjacent (can't talk about) fields always think they are best qualified for brilliant discoveries in the simple, commercial areas.

 

[Rosewind]

Given the 'provenance' problem I suspect the only way I can know I'm talking to GO and not someone else would be for me to join, or someone I can trust to act as a checkable go-between to confirm our identities.

I would suggest that GO answer you by way of another video if he is willing to do so. He is able to read this forum without joining and could be pointed to relevant posts in this MQA II thread and in the original MQA thread. That would help him navigate around too many posts that are pointless and unhelpful.

 

[DimitryZ]

I would suggest that GO answer you by way of another video if he is willing to do so. He is able to read this forum without joining and could be pointed to relevant posts in this MQA II thread and in the original MQA thread. That would help him navigate around too many posts that are pointless and unhelpful.

Please, no videos. They are a terrible way to do engineering, like doing automotive repair with choreography. JimA is seemingly attempting to do something that is at least informally peer-reviewed. This requires several timely bi-directional communication cycles.

It's a good indicator of GO's intentions if he agrees to participate. I hope he does.

I will try to use this as an insentive to gain basic knowledge in the area of digital signal processing. Missiles, radars, sonars and torpedoes, all day, every day...can get boring

 

[DimitryZ]

While you European folks were sleeping, I used my time to run a comparison of the weird MQA species - 44.1KHz master vs. its' ostensible LPCM counterpart. This time I used headphones, because my daughter upstairs complained that her gaming friends couldn't hear her over my music. Tidal and Qobuz were the streaming services used. Ostensibly, this is a comparison of the MQA process in its' purest form - the stuff they do with the master LPCM file that is so disturbing to some audiophiles.

I used this Melody Gardot's track from her Currency of Man Album. As always, you blokes can thank me for dropping this devastating music into this boring thread:

When I started I thought MQA had a slight edge: vocals were more distinct and intelligible and the recording had more depth. But as I switched back and forth a few times, I appreciated the more upfront presentation of LPCM and it's edgier sound that seemed more aligned with the music. MQA had a slightly lower volume. Overall MQA sounded like you were in the middle of the club and LPCM was more front row. LPCM was more Rega and MQA was more Linn.

Verdict: DRAW, preference dependent on your system and personal taste. Audible in the audiophile sense, but not dramatic.

PS. There were no compression, distortion, tonal anomalies or other audible issues audible with either format. Though the lower average volume of MQA hinted that the format is attempting to fill the available dynamic range to its' capacity with no compression. This would be well known to photographers working in Photoshop or PaintShopPro (and adherents of Ansel Adams' Zone System, of course).

One wonders if the transfer function they use is a perfectly straight one or they bend it for each recording. To those beginning to scream, I would ask you to consider that all pro and most serious amateur photographers (including yours truly) bend it for every image. In this case a photographer is both an artist and a master engineer. We bend a curve to make the final print to be what we feel the original scene was.

There is zero controversy, just normal professional work flow. All pro image software tools have this capability. Photographers are taught and encouraged to use it. Modern cameras have this as a highlight choice, altering the "original" at the point of registration. Can anything be more evil?

Sounds familiar?

I will be happy to post "bent IO curves" landscape photos on the photo forum as a "spillover" MQA discussion (and I will post the RAW images actually captured by the camera sensor). I won't compete with superb photographers posting there, but I am no slouch. And all serious photographers will tell you they manipulate digital image data to get the final image (that would be LPCM going to a DAC) closer to what they see as the original scene.

As an MIT aerospace undergraduate in the late 80s I was taught unified engineering - a concept that united many engineering disciplines under an umbrella of common mathematics. In my career, I found this to be very helpful in allowing me to work across multiple disciplines, from structures to thermal to materials to control systems. Sensibly, the Institute has evolved this concept to more engineering disciplines:

https://ocw.mit.edu/courses/aeronau...ngineering-i-ii-iii-iv-fall-2005-spring-2006/

I find there are many commonalities between the current lossless/MQA debates in the audio realm and the common and widely accepted workflows in the image world.

 

[kaolin]

While you European folks were sleeping, I used my time to run a comparison of the weird MQA species - 44.1KHz master vs. its' ostensible LPCM counterpart. This time I used headphones, because my daughter upstairs complained that her gaming friends couldn't hear her over my music. Tidal and Qobuz were the streaming services used. Ostensibly, this is a comparison of the MQA process in its' purest form - the stuff they do with the master LPCM file that is so disturbing to some audiophiles.

I used this Melody Gardot's track from her Currency of Man Album. As always, you blokes can thank me for dropping this devastating music into this boring thread:

When I started I thought MQA had a slight edge: vocals were more distinct and intelligible and the recording had more depth. But as I switched back and forth a few times, I appreciated the more upfront presentation of LPCM and it's edgier sound that seemed more aligned with the music. MQA had a slightly lower volume. Overall MQA sounded like you were in the middle of the club and LPCM was more front row. LPCM was more Rega and MQA was more Linn.

Verdict: DRAW, preference dependent on your system and personal taste. Audible in the audiophile sense, but not dramatic.

PS. There were no compression, distortion, tonal anomalies or other audible issues audible with either format. Though the lower average volume of MQA hinted that the format is attempting to fill the available dynamic range to its' capacity with no compression. This would be well known to photographers working in Photoshop or PaintShopPro (and adherents of Ansel Adams' Zone System, of course).

One wonders if the transfer function they use is a perfectly straight one or they bend it for each recording. To those beginning to scream, I would ask you to consider that all pro and most serious amateur photographers (including yours truly) bend it for every image. In this case a photographer is both an artist and a master engineer. We bend a curve to make the final print to be what we feel the original scene was.

There is zero controversy, just normal professional work flow. All pro image software tools have this capability. Photographers are taught and encouraged to use it. Modern cameras have this as a highlight choice, altering the "original" at the point of registration. Can anything be more evil?

Sounds familiar?

I will be happy to post "bent IO curves" landscape photos on the photo forum as a "spillover" MQA discussion (and I will post the RAW images actually captured by the camera sensor). I won't compete with superb photographers posting there, but I am no slouch. And all serious photographers will tell you they manipulate digital image data to get the final image (that would be LPCM going to a DAC) closer to what they see as the original scene.

As an MIT aerospace undergraduate in the late 80s I was taught unified engineering - a concept that united many engineering disciplines under an umbrella of common mathematics. In my career, I found this to be very helpful in allowing me to work across multiple disciplines, from structures to thermal to materials to control systems. Sensibly, the Institute has evolved this concept to more engineering disciplines:

https://ocw.mit.edu/courses/aeronau...ngineering-i-ii-iii-iv-fall-2005-spring-2006/

I find there are many commonalities between the current lossless/MQA debates in the audio realm and the common and widely accepted workflows in the image world.

In your opinion and understanding, is MQA lossless or not?

 

[DimitryZ]

In your opinion and understanding, is MQA lossless or not?

In the strictly technical sense, it is. However, as the higher frequency LPCM files are 90% ultrasonic noise, the definition of lossless becomes increasingly meaningless, in the audiophile sense or in the functional sense.. "Lossless" means that 90% of your file is noise.

In the name of "lossless" do you want high volume upper ultrasonics to pump through your system Stressing your amps and speakers?

I don't.

 

[tuga]

In the strictly technical sense, it is. However, as the higher frequency LPCM files are 90% ultrasonic noise, the definition of lossless becomes increasingly meaningless, in the audiophile sense or in the functional sense.. "Lossless" means that 90% of your file is noise.

In the name of "lossless" do you want high volume upper ultrasonics to pump through your system Stressing your amps and speakers?

I don't.

This measurement shows a tweeter's high-level resonant peak at 27kHz producing/modulating harmonic distortion at 13.5kHz and 9kHz:

Now look at the aliasing produced by MQA's leaky diaper:

The fact that all that distortion is signal-correlated might explain why some people like MQA. Vinyl and valves (when pushed) also produce signal-correlated distortion.
Maybe it sounds more "vibrant" and "real"...


It's cråp design.

 

[EPear]

...

Sounds familiar?

eh, yes..

we might agree on some points after all, let me expand from my comparisons, in keeping with similar perception tools:

somewhere at the back - front row
veil - no veil
curtain - no curtain
muddy water- clear water
congested- more air
vocals like some sweet cake - proper patisserie

do I need to clarify which is which?

I have also never heard actual distortions with MQA e.g. unlikely you would hear additional twists in the electric guitar solos, these are expressed in the above differences I would imagine..

a tiny little point - it’s one thing to bend your own curve as much as you like until you believe that’s how you saw reality in that ingenious (no doubt) photo of yours, and a completely different matter bending Gursky’s and lying and pretending you know better than him without asking him. You did the same with my humble post telling everyone what you know I meant with every line, so you know what I mean here)

obviously, poor Melody Gardot but at this rate, soon we will see you at the Picasso MQA bending workshop to .. well, bend more stuff. Or the Frida Kahlo - to tell her what pain is? After all, they had no clue what they meant, right? People like you know better and surely, MQA knows best, just needs more preaching.

Come to think of it, anything can be MQA-ed! (or as you say - bended, manipulated and all that).

 

[Jim Audiomisc]

Output0.csv contains four columns. The first one is likely frequency. The next four are columns of identical numbers (252.9297. 0. 252.9297. 0.). I don't think this is spectral output

In the Output1.csv, the four columns are variable and can be plotted. What are their units and what GO cases do they represent? It's a good idea to add this information into the excel file.

Since this is your work, isn't it better for you to do your own plots (with axes labeled) and make them available in a format that is easily displayable? I suspect there is something different in the Unix png definition...

This is the plot of your output1 file, without the last column, which appears to be phase. Is this what you intended us to see and what are we looking at? The grey line is input against itself, I presume. I plotted it as log-semilog, since the Y axis obviously wasn't log.

if you read the 'C' code you can see that the columns are:

frequency, I, Q, Amp, Ang

i.e. the output is the real (I = In phase) and imaginary (Q = Quadrature) components. These are then used to generate the amplitude and phase.

PNG is easily displayable. It is a platform independent bitmap used on millions of webpages. So far as I can tell most everyone else hasn't a problem with it.

Your plot expects me to use Google. Nope, I don't use them as I am not their product. if you want me to see something, put it on the web or somewhere else that is open.

 

[Jim Audiomisc]

Chrome in windows and yes I can. Also visible in Opera in Android.

I don't use those so can't advise. But they should be able to load and display a PNG without needing it embedded in HTML. All the browsers I use do this OK. And I think others here are also OK.

 

[Jim Audiomisc]

Jim, that is your well-intentioned wish list, not actual reality.

And if you filtered the 352 DXD to something reasonable like 96 that can be efficiently FLACd, the "stupid lossless" crowd will scream "don't touch my lossless master, that's what MQA does!"

No. wrt point 1: It is what Information Theory tells us. And is routinely used by audio engineers. Indeed, many DACs and ADCs use it internally.

wrt your second point: you're arguing with someone else.

 

[Jim Audiomisc]

I would not advise to take a smug posture and expect that in an audio engineering competition with Bob Stuart (one of the heavyweights in audio signal processing for decades and the man behind Dolby TrueHD) you will easily show him up to be an incompetent fool or a clever fraudster. My understanding is that your work in audio is limited and it's well a known phenomenon that folks from adjacent (can't talk about) fields always think they are best qualified for brilliant discoveries in the simple, commercial areas.

You keep thinking of this in terms of "the man" instead of "the ball". You are also once again drifting back into taking for granted what I will conclude.

My only 'conclusion' this far is the one that is blindingly obvious to those who understand the application of IT to information storage efficiency. That the combination of appropriate bitfreezing and noise-shaping can make FLAC more efficient. To the point where there is no real need in general for any closed alternatives that may discard real info or add components that weren't in the source material.

 

[Jim Audiomisc]

In the strictly technical sense, it is. However, as the higher frequency LPCM files are 90% ultrasonic noise, the definition of lossless becomes increasingly meaningless, in the audiophile sense or in the functional sense.. "Lossless" means that 90% of your file is noise.

In the name of "lossless" do you want high volume upper ultrasonics to pump through your system Stressing your amps and speakers?

I don't.

WRT your first point you are, again, showing a lack of understanding wrt IT. But, yes, wideband recordings generally are swampted by noise if we simply use 24 bit LPCM. That's why bitfreezing/sample-depth reduction/noise-shaping can give us higher eficiency without losing real info.

Random noise is not 'information'. And you only need 1-bit's worth of noise to linearise.

Your last point can be used to argue *against* MQA.

 

[adamdea]

In the strictly technical sense, it is. However, as the higher frequency LPCM files are 90% ultrasonic noise, the definition of lossless becomes increasingly meaningless, in the audiophile sense or in the functional sense.. "Lossless" means that 90% of your file is noise.

In the name of "lossless" do you want high volume upper ultrasonics to pump through your system Stressing your amps and speakers?

I don't.

Dmitry
I hope that take this in good part. Let us assume for a moment that each person participating in this discussion is a grown up sincere person who is genuinely interested in advancing his understanding and in understanding whether the other participants have relevant information which he either does not have or has not yet fully taken into account. Let us for the moment discard all of the voluminous evidence to the contrary and proceed for a while on this basis.

It is troubling me that
1) you appear to wish convey a revelation that MQA is in fact about (or at least involves) discarding unnecessary HF information the reproduction of which may be damaging to downstream components or otherwise undesirable.
2) you regard the filter which Jim has referred to as "leaky" as being an apodising filter.

Now this leads me to suppose that either

• you have some information which I do not have and which (I hope I will be forgiven for this presumption) I think @Jim Audiomisc @mansr and other do not have;
• you may have misunderstood a lot of what they and others have been saying

Now I am nowhere near as smart as @Jim Audiomisc @mansr @Werner or various other people on this forum but I can with my hand on my heart say that my main interest in hifi forums is actually trying to understand stuff.

It also seems to me that there is a real opportunity for people to achieve some mutual understanding here and not just go round in circles. Since ex hypothesi you are a sincere and interested person just trying to get to the truth you must surely be interested in sorting out this misunderstanding: It may or may not be (as I suppose) the principle obstacle to progress in this discussion but it certainly must be a key point; and if you are right it provides a chance for Jim, mansr and otherwise to rethink their position. Equally if you are mistaken you will surely wish to reconsider yours.

My understanding of the leaky filter reference is that it relates both the the anti aliasing filter used to downsample hi sampling rate material to low sample rate material (ie the container) and the "anti imaging" filter used to reconstruct the audio signal from that data. Now you must have read this article which Jim wrote 5 years ago and which has been extensively referred to in pretty much every discussion I am aware of, as well as this thread.
http://www.audiomisc.co.uk/MQA/origami/ThereAndBack.html
Whilst I can understand that the word "leaky" may seem perjorative, it seems to express very well the point that the process involves allowing through the filter elements which conventional wisdom (and the sampling theorem) suggests should be filtered out

Two obvious points arise-
1) this "filter" (or non filter) creates spuriae both at the lower sample rate of the container and the higher sample rate into which it is unfolded.
2) I can't see any basis on which this could be described as an apodising filter. Now that suggests to me that either you are talking about a different filter or using the word in a sense I don;t understand.

Are you saying that the MQA process does not involve the process described by Jim in the article above? If not can you explain why and also explain how, as you understand it, MQA will process a DXD file into a container and then reproduce the thereby encoded audio signal.

Either way, can you explain what the filter is (in terms of frequency and phase behaviour) that you are referring to and where it is used in the MQA process (both packing and unpacking/reconstruction) ? Can you explain why it is properly described as apodising and whether it does or does not permit aliasing in one direction and imaging in the other?

Can you explain how that filter assists in removing unnecessary hf spuriae from the ultimate dac output compared with
1) an assumed DXD input as the original recording file
2) a conventionally downsampled (ie downsampled from dxd using a steep linear phase filter at or around nyquist) 96 or 44.1 Khz Fs flac file.

If you can explain this for me at will at least help me to understand what the point of disagreement (or conceivably misunderstanding) is. I hope it will help others, but if I'm wrong at least it will have helped me.
Anyway- I'm here to learn, as I am sure you are and I really feel that going through this process will help me (and I think others) to understand you. I appreciate this may take some time, for which I apologise.