MQA Review - Technical Analysis

Discussion in 'General Audio Discussion' started by Woland, Apr 15, 2021.

  1. SSL

    SSL Friend

    Friend
    Joined:
    Nov 12, 2015
    Likes Received:
    1,277
    Trophy Points:
    93
    Guys, PCM is lossless in the audible band when considering the frequency domain. This is a mathematical fact, and the founding information theory behind MQA hasn't changed it. Bit depth is just quantization error is just noise floor is just dynamic range. Nothing complicated. We can hand-wave and repeat "information theory" over and over again, but at the end of the day, the frequency domain is a solved problem.

    Where PCM is "lossy" is in the time domain. MQA calls this "temporal blurring" and points to the dreaded pre- and post-ringing on impulse response reconstructions as examples. These artifacts introduce uncertainty about where the impulse occurs in the sampling interval, and about the phase information of the harmonic components.

    And by the way, this isn't a new concept. Minimum phase filters, apodizing filters, DSD, have all been attempts to get around this limitation. MQA just seems to be the first to have explicitly referred to this time-domain uncertainty as "lossless vs lossy" for the purposes of marketing.

    Now, MQA claims to be lossless by "fixing" this uncertainty in the time domain once and for all. This is literally part of their marketing. Apparently, the impulse response of MQA is better than that of air, which I would guess is where their claims of lossless temporal resolution originate.

    Okay fine, but how. MQA says, a combination of a non-rectangular sampling kernel and an encoding that, while proprietary, can be inferred to depart from the Fourier model in favor of something like wavelet theory. Why? Because this allows the format to localize ("identify") both frequency AND time domain information without any uncertainty. Time domain information can be fully localized, and amplitude information below the resolution of the sampling interval can be retrieved.

    The problem, as I mentioned before, is that their technique relies on a fully proprietary MQA production process. Including hardware and software, from recording to playback. Moreover, the reduction in bit depth that results from the MQA magic sauce folding process is supposedly offset by their special sampling method, which again implies it needs to be present to offset the loss in fidelity. This is all straight from the horse's mouth, not just me inventing.

    But, who here thinks this is happening for any of the music being distributed as MQA, let alone a majority of the catalog? Even if it were, how is the consumer supposed to know? Moreover, should anyone care? It is yet to be established whether pre- and post-ringing artifacts even matter.
     
    • Like / Agree Like / Agree x 6
    • List
  2. dasman66

    dasman66 Self proclaimed lazy ass - friend

    Friend
    Joined:
    Mar 13, 2018
    Likes Received:
    1,149
    Trophy Points:
    93
    Location:
    NW Pennsylvania
    My non-scientific, listener based view of lossy vs lossless (which I assume a lot of regular listeners probably ascribe to as well):

    Lossy - the source changes the data stream
    Lossless - the source faithfully reproduces the data stream

    If MQA is adding noise and taking a 16 bit file and converting to 14bit, then it is changing the original data stream. I don't care what frequency its doing its thing, its changing it. If its adding noise that wasn't there originally, then its changing it. We don't need to get into the semantics and technical definitions of lossy vs lossless.

    I want the data stream to remain unchanged. If I then choose to add color to the data/music by choice of dac, choice of filter, choice of amp, choice of transducer, then fine. I make those decisions. But the data I'm starting with is the same no matter who/what/where is playing it.

    Yes, I'm sure I'm simplistic, but my guess is that a lot of us are.
     
    • Like / Agree Like / Agree x 10
    • List
  3. purr1n

    purr1n Building Magnis part time because it's peaceful.

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    75,339
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Simple argument makes the best sense here.

    Bob Stuart knows the deeper down the rabbit hole people go with the techno-babble, the more confused people will get. This is exactly what he wants.

    There's an art to maintaining the lie: misdirection.

    I say follow the money and do blind tests between PCM 24/96 and MQA using the same master.
     
    • Like / Agree Like / Agree x 8
    • List
  4. earnmyturns

    earnmyturns Smartest friend

    Friend
    Joined:
    Sep 25, 2016
    Likes Received:
    2,932
    Trophy Points:
    113
    Location:
    Palo Alto
    Home Page:
    With all due respect, this is wrong, as it ignores the effect of quantization errors on frequency reconstruction (not just on noise floor). That's why there's significant argument around different quantization (dithering) techniques -- best observed in the visual domain through comparisons of periodic artifacts (Moiré patterns) depending on different quantizers. "Lossless" is a marketing term that became popular as a contrast between PCM and perceptual audio coders (such as MP3 or AAC), but it does not reflect what happens between the original analog waveforms and the rest of the chain. One can say that a codec is "lossy" relative to a source encoding, or "lossless" relative to the same, but not "lossless" in an absolute sense since measurement and quantization error are always involved in any analog-to-digital conversion, even when Nyquist-Shannon sampling conditions are met (which they never are, but that's another story).

    The reason this matters in discussing MQA is that the whole smoke-and-mirrors of "post Shannon" is based on arguing for manipulations of the encoder away from standard PCM quantization. In other words, in replacing one "lossy" (quantized PCM) by another (MQA) which then becomes a matter of taste and marketing, rather than a mathematical absolute -- given that MQA blocks any accurate measurements of their chain from analog to digital back to analog.
     
    • Like / Agree Like / Agree x 1
    • Respectively Disagree Respectively Disagree x 1
    • List
    Last edited: Apr 17, 2021
  5. Pancakes

    Pancakes Almost "Made"

    Contributor
    Joined:
    Aug 13, 2020
    Likes Received:
    325
    Trophy Points:
    63
    Location:
    Atl
    We can circle jerk about semantics until the cows come home. The reality is, to a music listener (be they audiophile or not), "lossless" implies "unaltered". It also implies high rez but that's a different subject. In this context, MQA is not lossless. The equipment will never be lossless but the source (file, disk, etc.) should be if the claim is that it isn't altered.
     
    • Like / Agree Like / Agree x 7
    • Epic Epic x 1
    • List
    Last edited: Apr 17, 2021
  6. purr1n

    purr1n Building Magnis part time because it's peaceful.

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    75,339
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    ^ This.

    I didn't know PCM is now a lossy format. It's not even a format.

    Fuck all of you. It's clear that Bob Stuart has won.
     
    • Like / Agree Like / Agree x 5
    • Epic Epic x 5
    • List
    Last edited: Apr 17, 2021
  7. earnmyturns

    earnmyturns Smartest friend

    Friend
    Joined:
    Sep 25, 2016
    Likes Received:
    2,932
    Trophy Points:
    113
    Location:
    Palo Alto
    Home Page:
    Unaltered from what? The whole recording chain is all about tasteful alteration :D We have to be more precise. What riles many about MQA is that they make vague gestures towards being more faithful to the "source" but to the best of our knowledge, MQA tracks are just run-of-the-mill PCM tracks processed by some blackbox MQA encoder without any added information about the math from original capture to the source PCM. Thus, by information theory's data processing inequality, MQA can never contain more information about the original than the PCM track it was derived from. And as we suspect, it's likely to contain less.

    See, we can be more precise in our terms and still arrive to the correct conclusion ;)
     
    • Like / Agree Like / Agree x 2
    • List
  8. purr1n

    purr1n Building Magnis part time because it's peaceful.

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    75,339
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    1. Hires PCM - what it is.
    2. FLAC - compressed container that can be used on hires PCM, when uncompressed looks exactly like original highres PCM hence lossless - that is lossless compression.
    3. MQA - compressed lossy codec for hires PCM when uncompressed does not look like the original hires PCM - and according to @GoldenOne's testing, even has shit which has aliased down to the audible region from ultrasonics.
    Important note:

    I am not aware of any "MQA" ADC that would address quantization errors of PCM. Besides, quantization errors of 24-bit PCM are effectively nil and limited by analog. The only thing theoretically better than 24-bit PCM with respect to quantization error in the audio band is DSD - but again, limited by analog circuits, way way before the theory comes into play.
     
    • Like / Agree Like / Agree x 10
    • Epic Epic x 1
    • List
    Last edited: Apr 17, 2021
  9. purr1n

    purr1n Building Magnis part time because it's peaceful.

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    75,339
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Any time we attempt to argue against Bob Stuart's claims to lossy nature of PCM or the smeared nature of PCM, we lose.

    Don't you guys get it?

    Fuck, Bob Stuart should be running China, not Xi.

    Remember guys:
    1. Follow the money
    2. Do blind tests between FLAC(PCM 24/96) and MQA using the same master.
     
    • Like / Agree Like / Agree x 5
    • List
  10. Pancakes

    Pancakes Almost "Made"

    Contributor
    Joined:
    Aug 13, 2020
    Likes Received:
    325
    Trophy Points:
    63
    Location:
    Atl
    I agree with your entire post but I think it's making things more complicated than they need to be.

    Unaltered from the finished file that comes out of the studio is the only realistic thing we can hope for and therefore that's the standard by which we should be measuring.
     
    • Like / Agree Like / Agree x 5
    • List
  11. Thad E Ginathom

    Thad E Ginathom Friend

    Friend
    Joined:
    Sep 27, 2015
    Likes Received:
    9,483
    Trophy Points:
    113
    Location:
    India
    Wait, I guess I haven't done my homework properly. I realise I'm not clear on what is the input to the MQA rendering process? Assumptions are bad, and I'd assumed that it was the existing PCM material. ie the finished product is put through another process, compressed to MQA just as we might compress it to FLAC.

    But if MQA is something that happens in the studio, something that becomes the finished product, then that's a whole different pile of shit.

    Depend on what I just realised I don't know. Yes, FLAC/etc compression is an "IT bit/byte" thing, albeit optimised to the sort of data patterns found in audio. But if MQA is something that is applied at the source source, then of course it's a whole different ballgame. And if Mr BS has won, he's poisoning the well.
     
  12. Josh83

    Josh83 Friend

    Friend
    Joined:
    Mar 1, 2017
    Likes Received:
    1,254
    Trophy Points:
    93
    Location:
    Charlottesville
    Just as a survey:

    Pohlmann, Principles of Digital Audio:
    "Lossless" is "coding in which file size is compressed, but upon playback the original uncompressed file is restored. Because the restored file is bit-for-bit identical to the original, there is no change in audio fidelity.”

    Watkinson, An Introduction to Digital Audio:
    "In lossless coding, the data from the expander are identical bit-for-bit with the original source data."

    Zolzer, Digital Audio Signal Processing:
    "Audio coding can be divided into two types: lossless and lossy. Lossless audio coding is based on a statistical model of the signal amplitudes and coding of the audio signal (audio coder). The reconstruction of the audio signal at the receiver allows a lossless resynthesis of the signal amplitudes of the original audio signal (audio decoder). On the other hand, lossy audio coding makes use of a psychoacoustic model of human acoustic perception to quantize and code the audio signal. In this case only the acoustically relevant parts of the signal are coded and reconstructed at the receiver. The samples of the original audio signal are not exactly reconstructed."

    It seems pretty clear to me that MQA is not lossless according to all of these definitions. Indeed, it specifically meets Zolzer's definition of lossy, but Stuart is trying to spin it as "perceptually lossless."

    Also, @crenca was referring to Miska, the developer of HQPlayer, as supposedly saying that "as near as we can tell MQA is in fact lossless in the audioband (20-20)." That didn't comport with what I've read from Miska about MQA, so I reached out to him to ask him if he's said MQA is lossless in the audio band. His response: "No I have never said it is lossless because it isn't."

    Bottom line: There's no commonly accepted definition of "lossless" that MQA meets.

    As others have noted, Stuart is trying to claim that PCM audio as we know it isn't actually lossless ('cuz "blurring), and that if MQA were applied from end-to-end (recording to playback) its overlapping triangles would be truly "lossless." This is a fundamentally different claim, and it seems incredibly dubious based on everything we know about MQA.

    However, it's also an irrelevant claim, since 99.999% of MQA material on Tidal is simply Redbook or hires PCM encoded to MQA. In that sense, it is clearly lossy. MQA cannot add information that's not there in the original PCM files, and it's not reproducing them bit perfectly.
     
    • Epic Epic x 9
    • Like / Agree Like / Agree x 6
    • List
  13. wormcycle

    wormcycle Friend

    Friend
    Joined:
    Aug 13, 2016
    Likes Received:
    1,541
    Trophy Points:
    113
    Location:
    Toronto, ON, Canada
    The same guy who six years earlier was raving how 192/24, that "we need higher sampling rates to increase the time resolution up to that of our auditory system."
    And maybe he was right, maybe technically lossless in the audioband (20-200 is not lossless at all if the original recording was digitized to 192/24
     
    • Like / Agree Like / Agree x 1
    • List
  14. spoony

    spoony Spooky

    Friend
    Joined:
    Sep 28, 2015
    Likes Received:
    494
    Trophy Points:
    63
    MQA doesn't make any sense if your source material is already a digital PCM master.
    /thread

    Also, I cancelled my Tidal subscription because I don't want my bills to fuel this dumpster fire, I was using the 'HIFI' profile thinking it was good 'ole Redbook, not doing any A/B against FLAC because I don't fucking have time for that. Going back to lossy Spotify and a few offline files until they get their HD shit together (No Qobuz, Deezer or Amazon HD in my country).
     
    • Like / Agree Like / Agree x 3
    • List
  15. SSL

    SSL Friend

    Friend
    Joined:
    Nov 12, 2015
    Likes Received:
    1,277
    Trophy Points:
    93
    It's confusing because MQA is doing a lot of different things at once, which is probably by design.

    My point all along has been that MQA in the real world right now has nothing to do with audio quality, and that calling MQA presently in the wild "lossless" is wrong by their own definition.
     
    • Like / Agree Like / Agree x 2
    • List
    Last edited: Apr 17, 2021
  16. wormcycle

    wormcycle Friend

    Friend
    Joined:
    Aug 13, 2016
    Likes Received:
    1,541
    Trophy Points:
    113
    Location:
    Toronto, ON, Canada
    I think @crenca is right using the definition of lossless implicit in the Shannon-Nyquist theorem, which, as I understand it , is that any digital file with sampling rate 2x of the frequency of the initial signal can used to recover the original signal in the DA process.
    But this, in the case of digital processing of the original sample file, it raises a lot of questions.
    Are we sure that if we digitize a tape to 192/24 it does not matter what type of digital processing we apply in the middle, as long as the final file includes 2x the sample rate it will render the original signal in the DA conversion? I do not now for sure but I doubt it very much.
    But I also agree with @crenca that any definition of lossless that is not implicit in the Shannon Nyquist theorem is just an opinion. But it does not mean that it is automatically an invalid opinion.
    in the digital domain the only definitions of lossless that makes sense to me is close to Zolzer: if the original file samples cannot be reconstructed, then encoding is not lossless.
     
    • Like / Agree Like / Agree x 1
    • Respectively Disagree Respectively Disagree x 1
    • List
  17. YMO

    YMO it's not drinking alone if you're on Zoom

    Friend
    Joined:
    Apr 1, 2018
    Likes Received:
    7,089
    Trophy Points:
    113
    Location:
    DUVALLL-904
    Guys....lets not go too deep in this...it is making us seem like we are a bunch of virgins.

    We could be listening to music....right?

    As usual...Fuck MQA. Fuck proprietary formats that locks you down. Fuck this and fuck that. The audiophile is mostly a fool who monies are soon parted because they think it will lead them to permanent masturbation in their ears.

    I love the music hobby...but I believe a good bit of audiophiles are cancer who are ruining the hobby for some.

    *Rant Over*
     
    • Like / Agree Like / Agree x 6
    • Epic Epic x 1
    • List
  18. crenca

    crenca Friend

    Friend
    Joined:
    May 26, 2017
    Likes Received:
    2,402
    Trophy Points:
    93
    Location:
    Southern New Mexico
    This is as good a summary as any as far as I am concerned. However like @SSL says no one thought this very low level "temporal blurring" was anything more than an minor implementation headache before Bob Stuart came along and convinced John Atkinson and the like that it was a real problem to be solved.

    Thanks @Josh83 for reaching out to Miska, I stand corrected. I may be thinking of "mansr" who did much of the early analysis over at CA, or maybe I am misremembering entirely :rolleyes:

    In the end however I think @purr1n is right in that MQA is really just a confidence game and play, and that we should be resisting (to the extant that it is not futile) on this level.

    Fuck smooth talking British con men and their wares!!!

    MQA show us the math!!!

    :D
     
    • Like / Agree Like / Agree x 4
    • List
  19. Biodegraded

    Biodegraded Friend

    Friend
    Joined:
    May 28, 2017
    Likes Received:
    5,402
    Trophy Points:
    113
    Location:
    Vancouver BC
    You fuckers all need to take a step back and look at the basics.

    1. MQA reduces 16 bit files to 14 bits. That's loss.
    2. MQA takes ultrasonics and shits them down into the audible band. That's a reduction in 'Quality'.
    3. MQA streams you pay for on Tidal might not be the original master - and you won't know - so that cannot be 'Master Authenticated'.

    If you like the way it sounds, fine. If you want original files to play bit-perfect and let your own systems impose their own sound on them, you should be annoyed that MQA is trying to get a monopoly on 'hi-res' streaming while keeping its encoding process secret, proprietary, and with 100% penetration and you should be thinking in anti-trust terms.

    Please note, @iFi audio : I love my Nano BL, but MQA makes me want to take a huge shit on it. Please consider that in the spirit of user feedback (or food-back).
     
    • Like / Agree Like / Agree x 15
    • Epic Epic x 2
    • List
  20. Merrick

    Merrick A lidless ear

    Friend
    Joined:
    Jan 6, 2016
    Likes Received:
    7,985
    Trophy Points:
    113
    Location:
    Portland, OR

    I would put money on this never having happened, and not ever happening. First, it’s likely utter bullshit and thus MQA have a vested interest in making sure that no such files are ever produced, and second, it’s not worth anyone’s time or money.
     
    • Like / Agree Like / Agree x 2
    • List

Share This Page