MQA Deep Dive - I published music on tidal to test MQA

[Maki]

I think we should stop and acknowledge the original poster here. I am not competent to judge the technical technical merits of the codec, but I can read. And thanks to the original poster the "lossless" marketing lies have been exposed and even the defenders of MQA have retreated to a you can't evaluate a lossy codec that way argument. Which may in fact be true, but what is the benefit to the consumer? The only thing I see is the increasingly irrelevant smaller file size. Prove that your lossy codec sounds better or go away. Widely used and open lossless codecs are common, in fact much more so than this stuff. I have made my decision, and I won't pay a penny more for MQA anything, what about the rest of you?

I don't even care about lossless vs lossy, the fact that it's closed is more than enough for me to avoid it like the plague.

 

[scott wurcer]

The technical arguments are moot at this point. Many years ago the powers that be at my company bought a startup that had patents on techniques to make pictures on a 640X480 CRT VGA display appear perceptually better. I guess they thought hi-res flat panel displays were "decades" from practicality (the money went down the drain). MQA made the assumption that lossy codecs will continue to be necessary into the future, not the best of bets.

BTW no one has answered my question, what does better than lossless mean?

 

[awdeeoh]

The technical arguments are moot at this point. Many years ago the powers that be at my company bought a startup that had patents on techniques to make pictures on a 640X480 CRT VGA display appear perceptually better. I guess they thought hi-res flat panel displays were "decades" from practicality (the money went down the drain). MQA made the assumption that lossy codecs will continue to be necessary into the future, not the best of bets.

BTW no one has answered my question, what does better than lossless mean?

MQA

 

[pjug]

At this point I really think discussion of MQA sound quality should include MQA-CD. Since Tidal and Warner are puking out MQA-CD as redbook replacements. 2L has some tracks in both MQA and MQA-CD to look at. Spectra of first 30 secs of test track 106 below [worth checking my work since audirvana and vb cable tools are new to me. I don't know if these curves contain glitches that I introduced, or maybe I botched something]

While the first unfold with MQA is OK to 20KHz, at least, MQA-CD seems to have issues below 20KHz. 2nd DXD curve with offset is to show the real hires curve clearly.

 

[amirm]

Why not? If you want to know it's limits?

If you want to know its limits, you find music that performs that functionality, not test tones. We have such a list for loss perceptual codecs for example. Same could be developed for MQA and then there would not be arguments as to why such was chosen.

 

[amirm]

So, the 24/96 will be equally superior, if one subscribes to the "there is still something musical and worth preserving above 22kHz", correct?

There is on a number of files I have analyzed.

And for those subscribing to the "human listening is limited by 20kHz", this preserving is just useless waste (just like reproducing anything at 100kHz...)?

Only so if you can prove that original high res music converted to 16/44.1 is perceptually lossless. MQA folks post a peer-reviewed paper/blind test that showed this was not the case for some filtering at least. And unless you use noise shaping, we can demonstrate that the noise floor of 16 bits is audible.

 

[mieswall]

There's a few things I simply cannot understand about you specifically. The first, is the vaccuous claims about DSD's superiority or inferiority compared to MQA. When you do a comparison. What is the benchmark metric being used here? Your subjective listening, or are you using some performance accepted metric like noise or frequency response observation or what?

The second, and more puzzling, is you admit MQA dun goofed about the lossless claim (only a doormat would grant them the benefit of doubt by allowing them to co-opt the widely held understood meaning of "lossless" and chalking the incident up as a mistake where they didn't properly define the term). I don't want their definition, and the fact that they absorbed the word into their earlier statements and used it in a similar context as everyone else does, but with the hidden intent of supposedly "meaning something else" -- that sort of behavior isn't something you see amongst adolescence beyond comedic attempts. Certainly not the behavior you expect for adults, or purported specialists in the field of audio. Like imagine me co-opting the word lossy for years, but in the end saying "oh well I meant simply high amounts of bit rate bloat this whole time". It's ridiculous at face value, and quite frankly an insult to most sensibilities.

The third issue is, while they have dropped the lossless lingo (though not really as Tidal hasn't when I looked). They still hold to this equally as troubling claim of "original" and "true to the original source" and things of that nature. I'm just not see how this is actually being delivered. I'm not saying I don't see how they CAN deliver -- I'm saying I cannot see how they're actually DOING SO with their current product offering. I don't want to be pointed to patents, and their claims of ADC profiling, as none of that proves that's what MQA IS CURRENTLY doing, since MQA can't be evaluated for the most part (both literally, and also according to you personally, since it seems all probing instances are riddled with falsity in your view).

If this is actually the case, then beyond the realm of psychologically troublesome (by way of placebo and such), what could you yourself be possibly doing that warrants the idea that allows to to assume you're taking a rational position in the assertion that MQA is superior above something actually lossless for example if the goal is to listen to music as intended by the publisher? If we know less in virtue of the claims you make about how we're getting the evaluation wrong by the means we've gone about exploring; why would anything but pure agnosticism with respect to objectively verifiable means not be the default position, and then hard-skepticism, due to MQA's nonsensical behavior and history? I still haven't seen the inference structure that gets anyone out of this ordeal. Beyond of course the aforementioned appeal to anecdotes about end-user experiences as a consumer of MQA.

The reason you're being called a troll by some, is because you subscribe to this perplexing, and needlessly more complexity forming position of some sort of revisionist of diction. Like when you say things like "When the thing is, it transforms noise, but leave music untouched, imho." Even if it were true that it left music "untouched", I don't understand why you feel a transformation of noise (whatever the heck that actually means) is somehow mutually exclusive to such an operation as to somehow now not have an effect on "the music". OP already showed in his example (as flawed as you percieve it to be), his entire file, along with parts that had music in it, were all polluted, and done so in a predictable manner to some degree based on the level of ultrasonic content present.

It's supremely irritating to have to deal with someone where you need to ask him what they take every single word of their statements to actually mean definitionally. "Transformation of noise", that could literally mean anything at the end of the day for example.

Likewise when you make either the ignorant or simply unconventionally put claim about how MQA isn't "lossy" processing. See, for this, the main frustration is that you seem to not understand why the claim is irrelevant. EVEN IF the encoder was doing a lossless process, for whatever reason, the end-result it ends up spitting out a file that isn't identical. Now you claim "that's what happens when you plop in a test tone". But the normal person wouldn't care, as I could plop in test tones all day in lossless formats, and there wouldn't be an issue beyond perhaps with something that rubs up against the idea of encoders and decoders themselves. Why is it that MQA can purport "original true to source" claims, yet their encoder shits itself if a test tone of any kind is present? What is it about this clearly stated issue do you seem to actually protest particularly?

You then talk about "try MQA folks, and see if you like it, instead of throwing stones because you were told it losses information". That's not why stones are being hurled, and if you truly take this to be the reason why, then there is a concerning lack of observational power on your end. MQA could sound wonderful for multiple reasons, likewise it could sound like nothing different (I actually asked if anyone is able to blind test any MQA content folded vs fully unfolded because I wasn't able to blind test a difference in the few albums I realized I had). But none of this has anything to actually do with why stones are being hurled as you characterize. The reason the stones are being hurled, is because corporations swapping in and out of shadows, making claims that can be falsified currently, and not opening themselves up to proper scrutiny that could lay their entire ordeal to rest. Instead they proceed on, like some deranged maniac that believes all Press is good Press. How MQA content is perceived by someone is entirely irrelevant to the history of critique about the company and it's offering. EVEN IF some people can enjoy MQA content because it perhaps offers a version of an album never seen before or something. That has nothing to do with invalidating the claims OP made. The company still flees and works from the shadows as it sees fit, and that will forever keep it in everyone's bad graces.

I urge you to read your post. Virtually every premise you posit, is an assumption of some sort. You're careful not to make any seriously hard-claim, and keep stating on how you're just presenting opinion based loosely on the brute-force acceptance of the company's assertions when referenced technical claims they've purported outside of MQA strictly speaking through extrapolations from patents for example. Or when you say: Also things like: "I think it is obvious: the MQA process probably define conversion parameters in advance". How can something be obvious, but then conclude "well it's probable", and is simply asserted with an explanation that is definitionally the only option available as an explanation (yet not at all elaborated on, as to how such a thing is possible, or why this needs to be presumed). This is a claim you make, that isn't even positted by MQA themselves. Not that it would matter, because if they can't demonstrate how they're doing it, it would be as good as claiming they're magically doing it somehow... All I see in the entire post is a constant stream of "I guess" "I think" "it wouldn't be hard to envision" etc... Like goodness, so much hip fire. The amount of plausibility one would have to grant is staggering.

Lastly, you claim there is personally audible benefit you experience from MQA files. I am willing to actually grant this possibility, if it is merrited by the file being MQA as the reason for the benefit. And not that you have a glass of wine before engaging. Can you please show me (since everyone outside of MQA isn't privy to the entire technical picture) the results that prove conclusively the aspect present within MQA files, that accounts for your preference for them from an audible perspective? Keep in mind, this has to be something that isn't possible with any other format (lossless files run through DSP can't emulate this, otherwise MQA loses it's exclusive beneficial property). What is this thing you are hearing precisely, and where is the evidence you can actually discern it. I would like the results of studies for a preference ideally that matches to MQA as statistically significant, since I have issues with MQA inherently since it's not really based on any psychoaccoustic model. I need to be shown all these beneficial results by means of something like double/tripple blind tested individuals. Lossless files for example are all virtually the same, but I can still demonstrate the benefit of one versus the other in other metrics. Since MQA is lossy, it has it's work cut out for it.

Actually one more thing I have to get on record at the very least. Do you have any vested financial interest with MQA or any friendships among the staff or something? I don't take people to be shills very often at all, but it's getting so bad that for the leaps you take into asserting the things you do, with quite ambiguous wording (generalized and vague on purpose it seems to me), so I just have to ask if you have any affiliation with anyone from the company or anyone involved with business with the company? You are obviously aware of how much presupposing you're actually doing about MQA itself. Surly someone knee deep into it's purported benefits, would be troubled by all the holes in their stories and explanations - by now must have requested a ton of answers from MQA themselves so that you have the technical and objective backing of the knowledge to argue properly for MQA's defence beyond "well it sounds good to me, and I don't see a reason to assume they're lying/false about much of anything". Or is it truly your position that MQA is not something that warrant skepticism at this junction of the strongest kind?

Numbers according your paragraphs:

1- Not claiming superiority of any format. The opposite: I'm just saying that it is possible for DSD to be audibly better., or not. As for now, and without proper testing, we need to trust our ears, and mine tell me that MQA sounds really good; best than any other? I don't know, perhaps no, but certainly way better than Redbook.

2- Yes, I agree MQA hasn't handle this issue properly. I suspect MQA IS lossless dealing with the music "slice" of the file (good testing may confirm or refuse that), but totally, completely "lossy" below the noise floor and above the music threshold (because it is designed, precisely, to be that way). If the definition of "lossless" is correctly applied? I believe it is in the context of music, which to be honest, is the thing that matters to me.

3,4- I think that lingo is intended to claim MQA as "lossless" about the music, and more specifically, regarding the content captured from analogue sources in the studio, with the coming out from your DAC as close as possible to that. I agree, nevertheless, that it would be a claim difficult to prove, and so, again, the use of the word has been a bit careless MQA's part. But it is equally misleading to analyze this problem as a capability to feed any digital content to the converter and get the exact same image in the reversed process. That's the job of a compressor, or an archival device, but not of a music container system aimed for music *reproduction* (not archival, that is handled internally by the studio in other way), as long as this system preserves that music content intact. Which is the claim that MQA should have done instead of this "lossless" mess, imho, and which I think it delivers.

5-the manipulation of noise is a concept easy to understand: what's below the noise threshold the recording process, further steps in ADC conversion, what your DAC and system adds, etc, is totally useless data ( you can't even listen it). The size of that is very significant (several bits in size for each sampling); a part of it is even measured in this very site in every DAC Amir reviews!; and this is, I think, the one thing clearly explained in MQA talks and papers.
MQA stores there not that original noise, but the information it captured above 24 Khz, but dithered for enough degree of randomness that a non-MQA device still see it as noise, but that a MQA decoder is able to filter to get the real data hidden in it. Of course, that content is absolutely different than the original noise in that area (if you are making a loopback comparison), and so, the process is "lossy" in strict terms. But that "lossyness" is not only not bad, it is what makes MQA better in the first place.

6- Which premises are wrong in my post? Basically There are (for now that we are not dealing with time smearing part of MQA, just the folding process) only two premises in my post:

A) How MQA handles noise, as explained above; I fail to see the assumptions you point. Otherwise, almost everything that's measured here would be equally wrong.

B) The increasingly larger headroom above music with unused space, that MQA also reshape for other uses. In the space of the post it wold be too large to demonstrate in detail how this is also a fact (that higher octaves in music contain less amplitude than lower ones, thus requiring less bit-depth to be properly registered). Also fail to see any "assumption" here.
I urge you to google for graphs of the harmonic amplitude distribution ANY music instrument (any, even the one you think it could be the most demanding), and you will see that after the fundamental of say, a A3 musical note (220 Hz), each successive harmonic is of lower amplitude (the pattern of amplitudes is maintained in every note, it is what defines the timbre [sorry, tSpanish?] of that instrument). When you rich the 5th harmonic @3520 hz (2:440, 3:880, 4:1760, 5:3520) the amplitude of that sinusoidal wave is perhaps 20 or 30 db lower that the fundamental @220 hz. And so, it requires much less bits to be quantized. But in a standard PCM file (the one we are supposed to be lossless for not to bother our audience here) uses exactly the same amount of bits to registered than in the 220 Hz fundamental (16 bits for Redbook, and with some of them being noise as previously said) than for that 3520 hz sinusoid. MQA uses less bits for the latter, and reshape the rest of that bit-depth as room for the folding process. As higher you go in pitch, the less the bit-depth required. In ultrasonic frecuencies is even lower (and the key reasons of why you need that ultrasonic content is another issue that exceeds this lossy discussion).
Obviously that 16-bit part of that sample @3520 will be different than Redbook. This difference is what the tests of Archimago and GoldenOne are showing, but it doesn't means that MQA has lost any music data; it has just used the spared space more usefully. But, horror!, their analysis is showing a "lossy" process!! I hope this lengthy explanations clarifies how absurd is the term applied here, in the first place.
MQA pre-assumes that the higher you go in frequency, the less amplitude you should expect to find, because NO instrument generates as high amplitude as the fundamentals (although that's not exactly true but not relevant for our purposes, some instruments have their first harmonics a bit higher, but that happens at lower frecuencies because that happens only at the first harmonics).

If you then feed this algorithm with a lot of high amplitude at those higher frequencies (like in a square wave), obsviously the conversion process will be fooled. But you can't do that with music (it is just impossible), only with test tones. Thats' why (mango a lot of other reasons) these tests are so fundamentally flawed.

7- How do you want me to show you that what I listen in A is better than B? I do not have the means to test that, not even the knowledge. I understand you can't believe me. The best way to clear the issue is that you, yourself, to make this comparisons. I explained the aural tests I do in a previous post. I may provide example of musical pieces that I find MQA sounding better, but perhaps the better ones are the myriad of files at different formats available in 2L Audiophile's site.

8- Absolutely no relation with MQA, Meridian or Tidal. In fact, I live at least 15 thousand km apart from their headquarters...jisus! If you smash the guys, you are a white dove; but if you disagree with bad tests, you are paid by Meridian?!

 

[samsa]

And unless you use noise shaping, we can demonstrate that the noise floor of 16 bits is audible.

Well that's a definitive argument against the (16/44.1) MQA-CD format.

They use the 3 LSBs to encode the MQA data, claiming (correctly, I hope) that -- being pseudo-random --- it looks just like dither. But you can't monkey with the spectral content of the MQA data, ergo you can't noise-shape. Ergo the noise-floor is audible?

 

[Raindog123]

Only so if you can prove that original high res music converted to 16/44.1 is perceptually lossless. MQA folks post a peer-reviewed paper/blind test that showed this was not the case for some filtering at least.

Regarding the "perceptually lossless 16/44.1", I was coming for your expertise. And now you are asking me for proof... In my area, if some information is outside sensor's band, this information generally cannot be accessed through some post-processing. But I only intercept very-fast-flying thingies, and don't build consumer audio.

MQA folks post a peer-reviewed paper/blind test that showed this was not the case for some filtering at least.

Can you please point me at this peer-reviewed paper by the MQA folks? Thanks.

And unless you use noise shaping, we can demonstrate that the noise floor of 16 bits is audible.

If even 16 bits lead to audible noise floor (and, yes, I have seen this argument), then it will be especially silly to further borrow some of these bits for MQA's ultrasonic signaling, aka unfolding (are we still talking below 22kHz?)

 

[amirm]

Can you please point me at this peer-reviewed paper by the MQA folks? Thanks.

Sure: https://www.audiosciencereview.com/forum/index.php?threads/statistics-of-abx-testing.170/

Original paper: https://www.aes.org/e-lib/online/browse.cfm?elib=17497
“The audibility of typical digital audio Filters in a high-fidelity playback system,” Convention Paper, Presented at the 137th AES Convention 2014

 

[amirm]

If 16 bits leads to audible noise floor (and, yes, I have seen this argument), then it will be especially silly to further borrow some of this bits for ultrasonic signaling...

It is also used like HDCD to expand the dynamic range beyond 16 bits.

 

[Raindog123]

Well, all that comes to mind is:

"But I was thinking of a plan
To dye one’s whiskers green,
And always use so large a fan
That they could not be seen."

 

[Raindog123]

Numbers according your paragraphs:

@mieswall How about an answer to my question? Can even be slightly less elaborate than that of yours to Tks.

 

[awdeeoh]

That must be a recent development. There used to be none whatsoever.

Jay-Z's back catalogue is already in 16/44.1 MQA form even before Warner started replacing those plain 16/44.1 with MQA ones.

Some are 16/44.1 in 16/44.1 container, some are 16/44.1 in 24/44.1 container.

 

[vkvedam]

Why isn't this thread been promoted to the home page @amirm, this should be right up there as a sticky and kudos to @GoldenOne

 

[usersky]

B) [...] I urge you to google for graphs of the harmonic amplitude distribution ANY music instrument (any, even the one you think it could be the most demanding), and you will see that after the fundamental of say, a A3 musical note (220 Hz), each successive harmonic is of lower amplitude (the pattern of amplitudes is maintained in every note, it is what defines the timbre [sorry, tSpanish?] of that instrument). When you rich the 5th harmonic @3520 hz (2:440, 3:880, 4:1760, 5:3520) the amplitude of that sinusoidal wave is perhaps 20 or 30 db lower that the fundamental @220 hz.

You tell us that the higher we go in the frequency domain, the less relevant the content is for real music. Let's forget for few minutes that here on ASR some guys climax for 3db under -112 and admit that your figures are totally relevant for music reproduction. If by 3520Hz some harmonics get quite low, I can imagine how irelevant the content at 20kHz might be (for me personally it is totally irelevant since my hearing demonstrably stops at 17kHz So by your argument at 20kHz "real music" have very little relevant content (even if we choose to ignore repeated studies that prove >20KHz signals do not produce audible impression in humans). Now this seems quite contradictory with great deals of effort MQA does to preserve content above 20kHz or even 30kHz, sacrificing quality in audible band (as OP shown with harmonic signal at 1kHz).

 

[dmac6419]

At this point I really think discussion of MQA sound quality should include MQA-CD. Since Tidal and Warner are puking out MQA-CD as redbook replacements. 2L has some tracks in both MQA and MQA-CD to look at. Spectra of first 30 secs of test track 106 below [worth checking my work since audirvana and vb cable tools are new to me. I don't know if these curves contain glitches that I introduced, or maybe I botched something]

While the first unfold with MQA is OK to 20KHz, at least, MQA-CD seems to have issues below 20KHz. 2nd DXD curve with offset is to show the real hires curve clearly.

View attachment 124886

You forgot Universal music group,lots of MQA CDs in Japan.

 

[dmac6419]

Jay-Z's back catalogue is already in 16/44.1 MQA form even before Warner started replacing those plain 16/44.1 with MQA ones.

Some are 16/44.1 in 16/44.1 container, some are 16/44.1 in 24/44.1 container.

Most of the music on Tidal is still Redbook was listening just yesterday.

 

[Mountain Goat]

So if I, as an artist, decide to include a test tone in my song, or something similar to a test tone for a specific effect, MQA will decide it isn't music and barf all over the track. How is that h-fi? How is that the "original true to source as intended by the artist" ?

Somebody needs to make some test-tone dubstep. Tidal won't be able to tell the difference. I'd do it if I hadn't broken my fool arm today.

 

[sandymc]

It is just that you can't test lossy codecs, specially layered ones like MQA in this manner.

This is true. But as I understand it, GoldenOne's intent in testing was to test MQA against its marketing claims, not as a lossy codec. As an aside, I'd be very interest to see how MQA does compared the various other lossy codecs on the market. But until MQA codecs are easily available, it seems that would be difficult.