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

[BrEpBrEpBrEpBrEp]

Slow motion though

I wish... might be easier to keep track of it that was the case.

 

[voodooless]

I wish... might be easier to keep track of it that was the case.

well, in that case it’s a very long train.. a normal train wreck just takes a few dozen seconds.

 

[BrEpBrEpBrEpBrEp]

well, in that case it’s a very long train.. a normal train wreck just takes a few dozen seconds.

Yes, I believe you've just hitched the 3842nd car.

 

[Raindog123]

Ultrasonic content is very low compared to normal music in the baseband.

I don't think there is any controversy here.

Dimitry, without slightest provoking, but why do you say that?

Yes, strumming an acoustic guitar will not produce too many high-energy overtones, and neither will a finest falsetto singer... Yet my son synthesizes square waves and triangular-envelope spectra (not decreasing but increasing with frequency triangular!) all the time - while composing music for his computer games with his friends (using FL Studio). Eg, listen to any modern [Nintendo] Mario (“8-bit-like”) or Metroid game soundtrack - anything goes! The limit is only the human hearing, not the sound source/origin. And we - or our kids - are definitely way past the 'old school', whether it's Mozart or The Eagles.

 

[symphara]

Dimitry, without slightest provoking, but why do you say that?

Yes, strumming an acoustic guitar will not produce too many high-energy overtones, and neither will a finest falsetto singer... Yet my son synthesizes square waves and triangular-envelope spectra (not decreasing but increasing with frequency triangular!) all the time - while composing music for his computer games with his friends (using FL Studio). Eg, listen to any modern [Nintendo] Mario game soundtrack - anything goes! The limit is only the human hearing, definitely not the sound source/origin. And we - or our kids - are definitely way past the 'old school', whether it's Mozart or The Eagles.

It started with a claim about 35kHz. Which I bet not even your son can hear.

He's just stating the obvious. What's the opposite? That ultrasonics are in fact a significant proportion? The majority? It's just not true.

 

[Hayabusa]

Yes - they could have done the same lossy compression on the same portion of the ultrasonics and just not stored it (and the metadata/flags etc) in the audible band. It could have been stored in the PCM file header (as most PCM formats allow pretty much anything to be stored in the header), where it would be ignored by non-MQA-capable equipment, and recognized and used by MQA-capable equipment.



Yes - the problem with the above scenario is that it would have reduced the file-size savings, and while it would have worked with digital redbook files, it would not have worked with CDs, as it would have been really cumbersome to figure out where to put the extra data on the disc, and with any CD that had a very long program of music on it, there wouldn't be enough room on the disc to write all the extra data.

So the solution of storing it in the audible band is a more elegant and compatible solution - the issue, of course, is that it's a solution to a problem that doesn't exist. Or to be more generous to MQA, it's a cure that's worse than the disease.

Maybe storing the extra data as metadata in the FLAC file is a solution?
Its fully compaitible with current solutions. (2 channel FLAC)

 

[Raindog123]

you look at your tweeters they are much smaller than your woofers.

The reason is that there is much less energy in the HF.

...and yet that 'much smaller' tweeter can/will hurt your ear significantly more. I do not want to oversimplify - as the spectral sensitivity is most definitely at play here too - but the smaller size of the tweeter is simply due to a shorter wavelength, and not necessarily less energy/power:

Spectral density

Sound energy

 

[earlevel]

...I'll put myself on the spot here (gladly, as it will allow true experts to throw rocks.)

But while RF-frequency ('gigahertz class") DACs' state-of-the-art today is about 16+ bits, I thought that low-frequency audio DACs (and digital filters of DAC SoCs) are at least 24bit, and might even be 32 now...? While still 'of consumer prices'..?

No rocks...it depends if you mean 24-bit (or 32) mathematically. It really doesn't mean much when you have a minimum noise floor dictated by physics, coupled with a limited maximum signal dictated by consumer and pro standards and practicality. So really we can't do 24, we just make DACs that accept 24-bit input and go through the motions. If we want to be (overly) generous, we might say 22 but all things considered 20 is more honest as the best we're going to get.

And we're at the practical limit of doing better. Electronics can't get quieter without getting a lot colder...and I mean everything...and going up significantly in voltage swing for more signal is impractical. Fortunately (I guess), it would also be ludicrous considering the dynamic range limitations of our ears. (People get fooled by the numbers for the range of human hearing, because we're rarely in a soundproof enough environment to actually hear the quietest things, which are limited by internal body noises anyway, and the stated top end of hearing will damage our ears permanently in seconds. Just because we can hear it doesn't mean we should. So, when people say we need 24 bits because our ears have a dynamic range of 140 dB, they are being naive.)

 

[Raindog123]

It started with a claim about 35kHz. Which I bet not even your son can hear.

He's just stating the obvious. What's the opposite? That ultrasonics are in fact a significant proportion? The majority? It's just not true.

Thanks, you are actually making my case for me.

There are two rather independent statements going on: (1) "there is little ultrasonics in the 'music' " and (2) "humans can't hear ultrasonics"...

Personally, I absolutely subscribe to the second. So yes, my son and his friends cannot hear 35kHz.

However, when people say "The 'music' spectrum envelope follows the 'MQA triangle' - and this is how I interpret Dimitry's "ultrasonic content is very low compared to normal music in the baseband" - this is what I was trying to question/disagree with by referring to my son's electronic ['square-wave'] music synthesis experiments.

To reiterate, I agree that we should not care about ultrasonics in our music playback designs. But not for the 'significant vs insignificant [ultrasonic] proportion' reason - which in my mind (and in my examples) is questionable. But simply 'because humans just can't hear it'.

 

[Raindog123]

Yes, I believe you've just hitched the 3842nd car.

"Welcome to Snowpiercer, a train that is 3,850 cars long..."

 

[BrEpBrEpBrEpBrEp]

"Welcome to Snowpiercer, a train that is 3,850 cars long..."

And it goes round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round, and round...

Damn that's a good analogy...

 

[symphara]

Thanks, you are actually making my case for me.

There are two rather independent statements going on: (1) "there is little ultrasonics in the 'music' " and (2) "humans can't hear ultrasonics"...

Personally, I absolutely subscribe to the second. So yes, my son and his friends cannot hear 35kHz.

However, when people say "The 'music' spectrum envelope follows the 'MQA triangle' - and this is how I interpret Dimitry's "ultrasonic content is very low compared to normal music in the baseband" - this is what I was trying to question/disagree with by referring to my son's electronic ['square-wave'] music synthesis experiments.

To reiterate, I agree that we should not care about ultrasonics in our music playback designs. But not for the 'significant vs insignificant [ultrasonic] proportion' reason - which is in my mind (and in my examples) questionable. But simply 'because humans just can't hear it'.

There's a difference between can and is. Just because your son experiments doesn't mean that research in millions of tracks that show a certain distribution is suddenly invalid.

I don't think I understand your point.

 

[oursmagenta]

There's a difference between can and is. Just because your son experiments doesn't mean that research in millions of tracks that show a certain distribution is suddenly invalid.

I don't think I understand your point.

What about a track that contains portions that really push the limit of "natural" music/sound?
The first practical exemple that comes to my mind is a movie soundtrack (and I don't mean the score), especially a sci-fi one.

 

[pjug]

On audibility of ultrasonics and "ringing", here is an easy exercise. Maybe there is something wrong with this -- let me know.

In a 96000 Audacity project, create a burst of 23900 sine wave like the top track in image below. If you play this you will hear two snaps from the sudden transition at beginning and end of the burst, essentially two steps. We don't hear the 23900, unless we are gifted with hearing above 20KHz. Save this and resample this to 48000 (I used no dithering) and this pretty much removes the 23900 and what remains looks like two linear phase impulse responses, middle track [edit: actually the middle track shown is inverted]. Resample this to 9600 and save. Now make a difference track (invert one track and mix) to get the bottom curve.

The bottom curve is silent compared to the other tracks, even though it looks like it has pre and post ringing. But if you look at the spectrum of the bottom file there is nothing <20KHz. So the top and middle track contain the same audible information and in time alignment. The "ringing" from downsampling is just a proper representation of the sound.

 

[awdeeoh]

I'll suggest you better training your ears to undecoded MQA.

 

[Raindog123]

I don't think I understand your point.

No, I do not think you do…

Imagine, someone in the year 2000 conducting a survey, a very thorough survey on “What is the average downlink speed of consumer internet?” Well, the answer would be “about 1Mbps”…. If the same survey conducted today will get the answer “50Mbps”… And both then and now, there were cases significantly exceeding these statistical average ‘distributions’.

My somewhat obvious and trivial points are: What we used to listen - and what is captured in those millions of today’s tracks (1) is evolving - electronic synthesis enters ‘the stage’ more and more. (2) As the 'chicken-egg', the analysis of older material is both limited by and limits the yesterday’s technology choices. And if/when these limitations are removed, the results of the [music spectral content] analysis will eventually change (see “1”). And finally, (3) in the matter like this, a good research must show both ‘the average’ and ‘the standard deviation’ - to assess how many cases deviate from this average, even today…

Just my quick thoughts (not 'the ultimate and unchallenging truth').

 

[sandymc]

It is not currently as no one in the industry has reacted to my quest.

Snake oil salesmen signing up for scientific testing?

Something about turkeys and thanksgiving comes to mind.

 

[tmtomh]

I actually think this enormous thread, especially the last several pages, has produced some very helpful clarity despite the periodic acrimony.

That clarity comes in a very particular area: the entire debate about whether "real music" contains high levels of ultrasonic content is a debate that has arisen because of the fundamentally questionable claim at the heart of MQA's design. That claim is the claim that true high fidelity music reproduction requires a sample rate greater than 48kHz, because according to Stuart (in his AES paper), digital hi-fi music reproduction requires bandwidth up to 26kHz.

So is @Raindog123 correct that "real music" includes electronic music that can have increasing amplitude as frequency increases up to and beyond 20kHz? Yes, that appears to be correct.

Is @DimitryZ correct that generally speaking music produces ultrasonic tones at lower amplitude than tones in the audible band? Yes, that appears to be correct as a general proposition even though there are exceptions per Raindog's comment.

Is @amirm correct that if you are going to create, market, and distribute high-res music, then you have to deliver it in 24-bit form since 16-bit is the next practical step down from that, and that you more or less have to deliver it in at least 88.2k/96k form since 44.1k is not high-res and 48k is not consistently considered high-res by audiophile consumers? Yes, he is correct about that. We here can certainly critique that audiophile conventional wisdom, but Amir is basically correct about those market realities in terms of the available tiers of bit depth and sample rate.

Are Amir, @John Atkinson , and Bob Stuart correct that a fully "empty"/"100% clean" 24 bits of depth is not necessary to produce a sufficiently low noise floor that meets or exceeds the limits of human hearing? Yes, they are correct. While Amir's, Atkinson's, and Stuart's particular arguments about this have some slight variations from each other, we can all acknowledge that 6.02 x 20 = 120.4, which means that 20 bits' worth of depth provides the necessary -120dB noise floor and therefore it is not necessarily a problem to store data in the lowest 4 bits, if there is some value or good reason to do so.

So none of these arguments is, in my view, wrong - they are all substantially correct as far as they go.

The underlying issue, of course, is that there is no evidence that ultrasonic frequencies are necessary at all - at whatever amplitude - for hi-fi music reproduction. And since there is no need to preserve ultrasonic frequencies, there is by definition no sonic value in doing anything - no matter how clever or allegedly harmless - to store them inside the digital data that encodes/stores the sound in the audible band. In other words, the way to reduce the file/bandwidth size of high-sample-rate digital content without compromising fidelity is simply to downsample it to the minimum feasible sample rate whose Nyquist frequency remains above 20kHz (plus some buffer for filtering).

So a lot of the stuff we've been arguing about doesn't actually need to be argued. The only real question from a sonic point of view is:
Is there potential harm caused by the way MQA stores ultrasonics and other encoding info/data/flags in digital files? I would say Yes, for two main reasons:

1. It stores some of that data within the first 16 bits. That means decoded, 24-bit MQA may have compromised bit depth because its effective bit depth is not contiguous in the most significant bits. It also means that undecoded MQA has a noise floor inferior to 16-bit redbook, which is a problem not only for the millions of playback systems that lack MQA capability, but also for the fact that undecoded MQA content has begun to pollute the pool of redbook digital music, and if/when undecoded MQA files begin to circulate outside Tidal (very likely/possibly already happening, because we know how lazy/indifferent the labels are about this stuff), they will not be marked as such and we'll have no way of knowing where they are or how numerous.
2. It creates reconstruction-filter challenges for playback equipment. Hopefully this will be solved more fully as time goes on, but as of now there are lots of DACs out there that either emit audible clicks when switching between MQA and non-MQA filters on the fly, or else just keep MQA fllters on permanently. As MQA filters are slow/leaky, they produce undesirable aliasing. If you can't turn them off for non-MQA content, or you can turn them off only at the cost of audible clicks between MQA and non-MQA tracks, that's a problem.

IMHO the market case for MQA is very limited and rather weak. There is no significant streaming (or IMHO digital-file-for-purchase) demand specifically for files with bit rates above 88.2/96k. And MQA provides no significant size savings compared to 24/88.2 or 24/96 FLAC. Moreover, the ability to have one MQA file that can be served in multiple resolutions is also a highly overrated feature, since in practice that "multiple" is only two: redbook and high-res. There is no market for - and never will be a market for - a more split-up set of tiers like redbook vs 24/48 vs 24/96 vs 24/192, etc. Never going to happen.

In this vein, one can make a strong argument that the main - and only significant - force working to create a market for super-high sample rates like 352.8k these days is MQA itself, which is ironic since MQA does not deliver that sample rate and is totally incapable of ever doing so.

 

[cochlea]

"Welcome to Snowpiercer, a train that is 3,850 cars long..."

I’m just hanging around in hopes of making the 4,000th post so I’ll have something to tell my grandkids about the part I played in The Great MQA Debate of ‘21.

 

[ebslo]

I actually think this enormous thread, especially the last several pages, has produced some very helpful clarity despite the periodic acrimony.

That clarity comes in a very particular area: the entire debate about whether "real music" contains high levels of ultrasonic content is a debate that has arisen because of the fundamentally questionable claim at the heart of MQA's design. That claim is the claim that true high fidelity music reproduction requires a sample rate greater than 48kHz, because according to Stuart (in his AES paper), digital hi-fi music reproduction requires bandwidth up to 26kHz.

So is @Raindog123 correct that "real music" includes electronic music that can have increasing amplitude as frequency increases up to and beyond 20kHz? Yes, that appears to be correct.

Is @DimitryZ correct that generally speaking music produces ultrasonic tones at lower amplitude than tones in the audible band? Yes, that appears to be correct as a general proposition even though there are exceptions per Raindog's comment.

Is @amirm correct that if you are going to create, market, and distribute high-res music, then you have to deliver it in 24-bit form since 16-bit is the next practical step down from that, and that you more or less have to deliver it in at least 88.2k/96k form since 44.1k is not high-res and 48k is not consistently considered high-res by audiophile consumers? Yes, he is correct about that. We here can certainly critique that audiophile conventional wisdom, but Amir is basically correct about those market realities in terms of the available tiers of bit depth and sample rate.

Are Amir, @John Atkinson , and Bob Stuart correct that a fully "empty"/"100% clean" 24 bits of depth is not necessary to produce a sufficiently low noise floor that meets or exceeds the limits of human hearing? Yes, they are correct. While Amir's, Atkinson's, and Stuart's particular arguments about this have some slight variations from each other, we can all acknowledge that 6.02 x 20 = 120.4, which means that 20 bits' worth of depth provides the necessary -120dB noise floor and therefore it is not necessarily a problem to store data in the lowest 4 bits, if there is some value or good reason to do so.

So none of these arguments is, in my view, wrong - they are all substantially correct as far as they go.

The underlying issue, of course, is that there is no evidence that ultrasonic frequencies are necessary at all - at whatever amplitude - for hi-fi music reproduction. And since there is no need to preserve ultrasonic frequencies, there is by definition no sonic value in doing anything - no matter how clever or allegedly harmless - to store them inside the digital data that encodes/stores the sound in the audible band. In other words, the way to reduce the file/bandwidth size of high-sample-rate digital content without compromising fidelity is simply to downsample it to the minimum feasible sample rate whose Nyquist frequency remains above 20kHz (plus some buffer for filtering).

So a lot of the stuff we've been arguing about doesn't actually need to be argued. The only real question from a sonic point of view is:
Is there potential harm caused by the way MQA stores ultrasonics and other encoding info/data/flags in digital files? I would say Yes, for two main reasons:

1. It stores some of that data within the first 16 bits. That means decoded, 24-bit MQA may have compromised bit depth because its effective bit depth is not contiguous in the most significant bits. It also means that undecoded MQA has a noise floor inferior to 16-bit redbook, which is a problem not only for the millions of playback systems that lack MQA capability, but also for the fact that undecoded MQA content has begun to pollute the pool of redbook digital music, and if/when undecoded MQA files begin to circulate outside Tidal (very likely/possibly already happening, because we know how lazy/indifferent the labels are about this stuff), they will not be marked as such and we'll have no way of knowing where they are or how numerous.
2. It creates reconstruction-filter challenges for playback equipment. Hopefully this will be solved more fully as time goes on, but as of now there are lots of DACs out there that either emit audible clicks when switching between MQA and non-MQA filters on the fly, or else just keep MQA fllters on permanently. As MQA filters are slow/leaky, they producing undesirable aliasing. If you can't turn them off for non-MQA content, or you can turn them off only at the cost of audible clicks between MQA and non-MQA tracks, that's a problem.

IMHO the market case for MQA is very limited and rather weak. There is no significant streaming (or IMHO digital-file-for-purchase) demand specifically for files with bit rates above 88.2/96k. And MQA provides no significant size savings compared to 24/88.2 or 24/96 FLAC. Moreover, the ability to have one MQA file that can be served in multiple resolutions is also a highly overrated feature, since in practice that "multiple" is only two: redbook and high-res. There is no market for - and never will be a market for - a more split-up set of tiers like redbook vs 24/48 vs 24/96 vs 24/192, etc. Never going to happen.

In this vein, one can make a IMHO very strong argument that the main - and only significant - force working to create a market for super-high sample rates like 352.8k these days is MQA itself, which is ironic since MQA does not delivery that sample rate and is totally incapable of ever doing so.

Great post, and I agree with the vast majority of it. Of course I have to nit-pick at least one point. IIRC, the space savings of MQA WRT 24/88.2 FLAC encoded PCM seemed of significance, hence the discussion of 20-bit 88.2kHz or 96kHz. But if we accept the market will not accept a 20-bit format, and if we accept that we need 26kHz of bandwidth for "high-res", then I have to wonder how well FLAC would compress 24-bit 88.2kHz or 96kHz PCM that had been aggressively filtered to a 26kHz pass band. And would it be any more disingenuous to market such content as "88.2kHz" or "96kHz" than it is to market MQA as "24-bit"?

Also, there is one important part of the discussion that I feel is missing from this, which is the fidelity of the 16-bit 44.1kHz MQA format (ie. the MQA-CD format) fully decoded and rendered vs. regular 16-bit 44.1kHz PCM.