MQA Review - Technical Analysis

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

  1. purr1n

    purr1n Building Magnis part time because it's peaceful.

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    75,112
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    I am kind of wondering is MQA is in fact vaporware, particularly with the "second unfold". We never heard much about this second unfold that supposedly can only be done in hardware (DACs) early on with MQA - it is only recently that we are hearing about it (I could be wrong, but it does seem the narrative has changed quite a bit since I last kept up with MQA).

    I'm betting the second unfold is in essence a minimum phase filter at a sufficiently high sampling rate to apply appropriate parameters to minimize post impulse ringing. Here's the kicker: Meridian (Bob Stuart) had been one of the earliest proponents of minimum phase filters.

    --

    With respect to John Stuart MQA calling PCM linear phase filters blurry hence lossy (I don't remember the exact words), minimum phase isn't any better. It's just a tradeoff when all things are equal:
    https://www.superbestaudiofriends.o...d-post-effin-ringing-and-shit-like-that.2627/

    Note only that, @ultrabike shows us how minimum phase filters wreck the original waveform: https://www.superbestaudiofriends.o...n-ringing-and-shit-like-that.2627/#post-70767

    MQA smells so much of bullshit.
     
    • Like / Agree Like / Agree x 10
    • List
  2. GoldenOne

    GoldenOne Almost "Made"

    Joined:
    Jul 17, 2020
    Likes Received:
    297
    Trophy Points:
    53
    Location:
    England
    That's exactly what it is.
    The second/third unfold are simply a fixed upsampling filter, or rather, a collection of upsampling filters, which MQA selects one of depending on what product it goes into.
    [​IMG]
    Mans Rullgard reverse engineered it from some bluesound devices and posted SoX configs to replicate it (also provided Archimago with the various filter info for this post: http://archimago.blogspot.com/2017/07/measurements-audioquest-dragonfly-black.html )

    He has also created a tool to add MQA flagging to any audio file to trick a DAC into using it's MQA renderer filter, allowing for easy testing.

    This is why almost any device is able to have 2nd/3rd unfold 'rendering' capability (cause its basically just a filter swap), but the first unfold can only be done in software or in devices with some DSP capability as there is seemingly some adaptive stuff going on there.
     
    • Epic Epic x 8
    • Like / Agree Like / Agree x 4
    • List
  3. crenca

    crenca Friend

    Friend
    Joined:
    May 26, 2017
    Likes Received:
    2,344
    Trophy Points:
    93
    Location:
    Southern New Mexico
    MQA was and is vaporware in the sense that is was a grand idea - an "end to end" multifaceted "solution" that was going to fix several alleged problems at once. When MQA 1.0 ran up against the reality of trying to penetrate the consumer DAC market on the one hand, the streaming market on the other, we got MQA 2.0 which is where we are now. MQA as an actual sytem in the studio and the attendant fix for ADC "blurring" was laid to one side or just given up on completely. The implementation was split up because all ready existing and in the pipeline DAC's did not have the processing power to implement the core decoding functions, so they agreed to allow software decoding within Tidal and Roon and have DAC's do the rendering (aka second unfold which is just further upsampling and filtering) when they can. I suspect this is when MQA's DRM functionality (which is a mix of design, IP, and possibly more traditional encryption) was layed aside because of the need to get market penetration first. Future MQA 3.0 could be the switching on of existing latent DRM (or the updating of) through mandatory software update. MQA 4.0 could be something that compels artists/studios/mixers into actually implementing MQA on their end.

    It's like Mike Moffat of Schiit says, it's a kind of Dolby "end to end" monopolistic scheme. Fortunately it is, for now at least, if not stillborn then regulated to the eccentric "audiophile" backwaters of consumer audio
     
    • Like / Agree Like / Agree x 4
    • List
  4. purr1n

    purr1n Building Magnis part time because it's peaceful.

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    75,112
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Interesting. Given the lack of ringing after the impulse, MQA is definitely upsampling like mad to allow room to shove the passband down without rolling off the highs. At least it's something of an approach - but kind of cheesy since any DAC manufacturer could do similar, especially ESS parts which provide all sorts of parameters to tweak.
     
  5. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    10,035
    Trophy Points:
    113
    Location:
    Irvine CA
    If the original signal is 44.1 kHz then upsampling before the DAC will need a pretty sharp interpolation filter to remove ultrasonics.

    Those POS filters will fail miserably.

    On top of that, minimum phase filters are shit when it comes to audio reproduction.

    How much pre-cursor "ringing" the filter exhibits determines only filter transient response duration at the beginning. If the pre-cursor is N number of sample before the peak, that means signal will come out after N samples of transient response.

    How much post-cursor "ringing" the filter exhibits determines only how much transient response there is after signal reproduction finished completely. That is, at the end of the song there will be some small transient.

    During the song, there is no issue if the filter is linear phase and symmetrical. A minimum phase filter is not linear phase and it is not symmetrical. It will fuck up your time domain.

    In fact, in many systems a minimum phase filter will be catastrophic. Awesome that our ears can resolve these issues to the extent they do.

    MQA is bullshit. So sad if manufacturers and developers yield to these clowns.
     
    • Like / Agree Like / Agree x 5
    • List
  6. Hrodulf

    Hrodulf Prohibited from acting as an MOT until year 2050

    Friend
    Joined:
    Sep 28, 2015
    Likes Received:
    3,763
    Trophy Points:
    113
    It's only because the consumers have forced them to do so.

    The media circus around MQA is a thing to behold. They spared no expense to pamper the journos, but as we saw with Beats by Dre, good PR has an expiration date.
     
    • Like / Agree Like / Agree x 1
    • List
  7. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    10,035
    Trophy Points:
    113
    Location:
    Irvine CA
    This is the perfect example of what can go wrong when marketing goes full retard mode.

    Ok guys.

    What does linear phase mean?
    It means all frequencies through a filter will be delayed by the same amount. They are phase coherent.

    What does minimum phase mean?
    It means that for a given frequency magnitude response, the filter output will be minimally delayed. However, all frequencies will be delayed differently. They are not phase coherent.

    What does "ringing" do?
    In this context and with a linear phase filter and proper bandlimited signal, not a damned thing other than adding latency to your output.

    Am I going to hear this filter "ringing"?
    No.
    It will not affect steady state response. What you are looking there is transient response, not steady state. If you are an EE I hope you know WTF I mean.

    Is my audio output going to be affected by a minimum phase filter?
    Well I don't know if you are going to hear it. But you better believe it will affect the signal, and it will be evident on a scope.

    Are all these MQA quality and filtering claims bullshit?
    Yes.

    Why are some companies following suit then?
    Because some people with cash like to drink a lot of Kool-Aid. Don't be that guy.
     
    • Like / Agree Like / Agree x 14
    • Epic Epic x 4
    • List
    Last edited: Apr 26, 2021
  8. crenca

    crenca Friend

    Friend
    Joined:
    May 26, 2017
    Likes Received:
    2,344
    Trophy Points:
    93
    Location:
    Southern New Mexico
    Your the only one, at least that I can recall, that puts it like this @ultrabike. Everyone/thing else implies implicitly (though only occasionally explicitly) that this ringing is an effect throughout the entire signal such that it does effect "steady state response", albeit at a low level. They assume that it effects transients most of all. Further they argue that minimum phase is the better trade off because it puts said ringing into the "shadow" of transient where it is masked by the loudness of the entire signal at this point even though it is still distorting the signal. The time domain fuckery of the min phase filter, if it is acknowledged at all, is said to be the lessor of two evils.

    I say all this just to note how the usual story of "ringing" goes.
     
    • Like / Agree Like / Agree x 1
    • List
  9. purr1n

    purr1n Building Magnis part time because it's peaceful.

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    75,112
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    What @ultrabike is trying to say is that the ringing is insignificant, even when it comes to real signals with transients. Minimum phase is great only if you take a myopic view and stare at the impulse response. However, minimum phase wrecks a real audio signal.

    Note how Bob Stuart doesn't show below. If linear phase causes "blurring" (it doesn't), then minimum phase... idk WTF it does.

    Linear phase:
    [​IMG]


    Minimum phase:
    [​IMG]
     
    • Like / Agree Like / Agree x 9
    • List
  10. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    10,035
    Trophy Points:
    113
    Location:
    Irvine CA
    The transient will affect signal on the transition and stop band of the filter. Minimal in the passband as the plots show above (Assuming a linear phase flat low pass filter).

    How much distortion you get on the passband might depend on the passband ripple.

    Minimum phase will give you way more issues in the passband for a given filter length. Shorter the filter, the worse the performance with the sole exception of the No-Filter, which arguably is both minimal and linear phase.

    To further discuss

    An impulse is not a bandlimited signal. In continuous time, it is an infinite power, infinite bandwidth, instantaneous signal. It does not exist in the real world. In discrete time, it is a trivial signal with only one sample that contains all the frequencies below half the sampling rate. In other words, all signals in the real world are band-limited. Furthermore, the audio band is relatively narrow (and that includes the sweet sound of artillery).

    If an audio band-limited signal hits a linear phase low pass filter whose flat passband is larger than the BW of the audio signal, the output will be essentially the same as the input with a delay and transients in the start and end of the signal. The "ringing" will have minimal effect in the steady state (as in the fart of a mosquito fart). Unless of course the filter has a screwed up passband.

    Now. In a DAC, in most cases you can expect the original signal to fully fit inside the filter BW (including the audible sweet sound of artillery). Only ultrasonic oversampling images would pop up, and it would be the job of the (low pass interpolation) filter's to remove those images w/o affecting the low pass original signal.

    Now if the signal BW is larger than the filter BW, then yes I expect the transient response will impose some characteristics to signal out of band... which includes attenuation and stop band behavior (hopefully in the ultrasonic range). Passband issues, probably a function of the passband ripple and deviations from perfect flatness, and of course phase response.

    I expect a minimum phase filter to perform worse than it's linear phase counter part.

    Considerations

    Honestly, feel free to disagree. If I wrote something that is not correct, feel free to point it out.
     
    • Like / Agree Like / Agree x 4
    • List
    Last edited: Apr 27, 2021
  11. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    10,035
    Trophy Points:
    113
    Location:
    Irvine CA
    Shut... Err...

    Well ok guys.

    So the "ringing" you see in the impulse response, that is the transient. Sounds like MQA guys are saying that you mask the transient with the transient?

    "pre-ringing" is the turn on transient.
    "post-ringing" is the turn off transient.

    Also, Minimum phase filters don't trade off "ringing" placed into the "shadow" of the loudness at the expense of time domain fuckery. They trade off turn on transient for more turn off transient if you want to just look at that.

    And again, "ringing" is not masked by the "shadow" of the transient. "Ringing" is the transient.
     
    • Like / Agree Like / Agree x 4
    • List
  12. purr1n

    purr1n Building Magnis part time because it's peaceful.

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    75,112
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Yup. With minimum phase (right graph below), simply a tradeoff with more fuck'ry after the impulse vs linear phase which has less fuck'ry before and after - all things (stopband and passband parameters) being equal. The laws of the universe have a way of there's no free lunch.

    [​IMG]
     
    • Like / Agree Like / Agree x 6
    • List
  13. crenca

    crenca Friend

    Friend
    Joined:
    May 26, 2017
    Likes Received:
    2,344
    Trophy Points:
    93
    Location:
    Southern New Mexico
    You guys are nailing something I never understood in this argument for min phase filtering that the MQA guys make as well as others who promote min phase like the late Charles Hanson of Ayer Acoustics (though he was a fierce critic of MQA before he died): why is it better to simply shift this low level distortion (aka ringing/ripple) when just taken on its own, let alone with the added effect of the negative time domain effects of min phase? Talk about 1 step foward and 2 steps back. If it were not for the (technically ignorant) promotion of such things by Stereophile/TAD and their webzine imitators, would it be anything at all in the audio world...who knows.
     
  14. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    10,035
    Trophy Points:
    113
    Location:
    Irvine CA
    Because digital signal processing is not intuitive all the time and folks take advantage of that to sell stuff.

    It is very difficult to figure out what the "ringing" is going to do to the passband just by eye balling it. You need a frequency response magnitude and phase to visualize how the signals of interest (audio bandwidth) are being processed.

    What many folks qualify as "ringing" in the impulse response can be very hard to read when it comes to steady state. Two linear phase filters with arguably similar "ringing" will behave very different: different stopband, transition band, stopband ripple, passband ripple, rejection, phase response, and so forth.

    If you want a better visualization in the time domain, and impulse will cause the filter to start up, process, and die out in a certain way. It will "ring" to the abrupt signal. But bandlimited signals, relative to fs/2, don't have impulses. After start up transient and before die out transient, a smooth and continuous signal can be tracked almost perfectly in steady state by a well designed filter with shit loads of alleged "ringing". In steady state you would need to look at the steady state output of the filter, not the impulse response of the filter alone.

    In a way, the frequency response does that for you. It is a plot showing how different signals in different bands are affected in steady state. If you were to pass N frequencies to your filter and plot the magnitude and phase of the output relative to the input, that is a frequency response plot. It is sort of brute force.

    It is possible to do the same by passing a bandlimited and smooth noise signal, covering the entire audio band, and see what it does. That's essentially what I did above. To qualify performance you could take a difference magnitude square average of the before and after given some delay compensation as well... all in the time domain. And I feel confident a linear phase filters will outperform all these minimum phase filters.

    With all this in mind, it should be apparent that there is no frequency domain fuckery at the expense of time domain fuckery. Of time domain fuckery at the expense of frequency domain fuckery. Time and frequency domain are duals describing the same system in different ways. Fuck with one, and you fuck with the other.
     
    • Like / Agree Like / Agree x 5
    • List
    Last edited: Apr 28, 2021
  15. crenca

    crenca Friend

    Friend
    Joined:
    May 26, 2017
    Likes Received:
    2,344
    Trophy Points:
    93
    Location:
    Southern New Mexico
    I have wondered if this truth is not a way to educate the typical "audiophile" and help counter MQA/min phase promotion. It is a simplistic reduction to think of time domain without frequency domain, or vice versa - most audiophiles don't understand that they are two aspects of the same thing, and if you change one you change the other. Upstream someone links a video of an interview with Mike and Jason of Schiit and towards the end Mike says something like (going from memory) "we use a frequency domain filter, and then apply a time domain filter...". Well, that may make some sense in a certain context but in reality there is really no such thing as I understand it - every filter is both at the same time...
     
    • Like / Agree Like / Agree x 1
    • List
  16. purr1n

    purr1n Building Magnis part time because it's peaceful.

    Staff Member Friend BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    75,112
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    There is no absolute truth. There is only relative truth: what one believes matters or what one hears. My ears tell me that I prefer the sound of linear phase filters with steeper knees for most DACs; but on occasion, I have preferred minimum phase filters.
     
    • Like / Agree Like / Agree x 2
    • List
  17. mitochondrium

    mitochondrium Friend

    Friend
    Joined:
    Sep 1, 2017
    Likes Received:
    865
    Trophy Points:
    93
    Location:
    A Cell
    Makes my imagination run wild naughty little airwaves

    Now I never was in any doubt that BS and Amir are wankers but I think even if he hasn’t borked as many measurements as Amir JA has definitely joined the club (maybe some time ago but it is quite apparent now).
    Point which only confirms my point of view that there are very few Audio journos who aren’t industry brown nosers.
     
    • Like / Agree Like / Agree x 1
    • List
  18. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    10,035
    Trophy Points:
    113
    Location:
    Irvine CA
    There are many filter classifications.

    I don't know exactly what Mike meant by that, but a filter is a convolution operation. Several way to implement that though. When a filter is sufficiently short, a straight up convolution can be used. Many implementations out there including lattices and stuff.

    When a filter is very large, it is best to do filtering in the frequency domain. There are two methods: overlap-and-save and overlap-and-add. You can look it up. You make use of the fact that convolution in the time domain, is multiplication in the frequency domain. But you have to avoid circular convolution (see mentioned methods), and you have to be careful about quantization noise specially if the filter is too long and requires a large FFT/IFFT pair.

    You could split a long filter into two by factoring the transfer functions. Then have a large one implemented in the frequency domain, and a short one implemented in the time domain.

    Now... that's what I know that resembles what is described. I have no clue if that's what Mike meant.
     
    • Like / Agree Like / Agree x 2
    • List
  19. yotacowboy

    yotacowboy McRibs Kind of Guy

    Friend
    Joined:
    Feb 23, 2016
    Likes Received:
    5,487
    Trophy Points:
    113
    Location:
    NOVA
    Home Page:
    Humble thanks from the popcorn gallery: Your last several posts have elucidated the differences in filtering immensely, for me at least. Makes it much easier to understand the original "problem" MQA was intending to solve (beyond the filesize/streaming non-issue given 4G/5G). If I could buy you a beer, I would!
     
    • Like / Agree Like / Agree x 5
    • List
  20. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Friend MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    10,035
    Trophy Points:
    113
    Location:
    Irvine CA
    :sail:
     

Share This Page