Harmony Central Forums
Announcement Announcement Module
Collapse
No announcement yet.

Excellent vid debunking digital 'stairstepping'

Page Title Module
Move Remove Collapse









X
Conversation Detail Module
Collapse
  • Time
  • Show
Clear All
new posts

  • Excellent vid debunking digital 'stairstepping'

    Glenn Bucci posted this over at GS. It's a very nicely done 'investigation' (cum tutorial) of claims that digital audio is 'stairstepped' and cannot accurately reproduce 'smooth' analog wave signals. From Wired...


    http://www.wired.com/gadgetlab/2013/02/sound-smart-watch-this-excellent-primer-on-digital-audio/


    ‚Äč
    music and social stuff

  • #2

    Why to people STILL have to debunk this? If you follow the law, it works perfectly.

    I did watch the video and I thought it was pretty good, but I"m  not sure that skeptics will get the point that digitizing hardware works because it assures that, regardless of what goes into the input jack, the input signal is band-limited before it gets sampled. If it wasn't (which would have been a good demonstration), while you still don't get stair steps, you get something out that doesn't resemble what went in.

    An important thing that he didn't say about dither is that he ony spoke about dither going into the quantizer. Regardless of how the arithmetic is performed inside a digital box, the output stream is word length limited. If you're quantizing to 24 bits and have to squeeze that through a 16-bit pipe (like a CD), you have to truncate the word. Doing this without dither at the output will cause a different kind of distortion, not simply added random noise. By dithering, we trade the distortion for noise.

    Any manipulation of the signal while it's digital, like changing the level, either with a knob or a compressor plug-in, will produce an output that, if not truncated, has more bits with actual data in them than what went in. Back when everything in the digital pipeline was 16 bits, if dithering wasn't performed after every word-lengthing operation, truncation would be occurring in several places between the input and output. With processors that today use 32 bit floating point arithmetic, you can leave those extra bits floating around inside the box, and not get rid of them until it's time to spit out a true 16-bit word. This is why it's advantageous to send a 24-bit master that's intended to go on to a CD to the mastering house. After they do their final tweaks to your 24-bit file, the'll do the final dithering so it'll fit on to a CD.

    In general, I aggree with his premise that 192 kHz 24-bit downloads don't make much sense, but people with high end playback systems are hearing some improvement over "CD quality" from 96 kHz 24-bit sources. Part of the reason for that is technical - some processes work better at sample rates well above the minimum required to produce the normal audible range. Another reason is that in some cases, there's simply more care taken in preparing the master.

     

    --
    "Today's production equipment is IT-based and cannot be operated without a passing knowledge of computing, although it seems that it can be operated without a passing knowledge of audio." - John Watkinson, Resolution Magazine, October 2006
    Drop by http://mikeriversaudio.wordpress.com now and then

    Comment


    • blue2blue
      blue2blue commented
      Editing a comment

      MikeRivers wrote:

      Why to people STILL have to debunk this? [...]


       


       


      In general, I aggree with his premise that 192 kHz 24-bit downloads don't make much sense, but people with high end playback systems are hearing some improvement over "CD quality" from 96 kHz 24-bit sources. Part of the reason for that is technical - some processes work better at sample rates well above the minimum required to produce the normal audible range. Another reason is that in some cases, there's simply more care taken in preparing the master.


       




      To the plaintive lead question, I suppose just because they're people and many people are amazingly resistant to logic when they've vested themselves in a contrary belief. I guess.


      But it does keep coming up -- and not just in consumer/audiophile milieus. As I know you know -- since I occasionally see you over at the GS place where there are almost always one or two science vs. magic ear threads raging in the So Much Gear forum.


       


      With regard to distribution formats (which was what started his enagement with this 'debate'), he of course points out in his original article that longer sample word lengths are important during the production phase.


      With regard to better operation of specific units at higher SR's, certainly this has often been reported in various contexts by various folks. In the days before oversampling, it wasn't hard to see how higher sample rates could provide extra headroom for antialias filtering.


      But modern oversampling designs, as I understand the issues, should be able to deliver the same level of performance within the narrower band limits dictated by lower SRs. Should, of course, being a qualifier implying an ideal world. Which this ain't. So, as always, mileage tends to vary.  wink.gif 


       


      I did think he missed some opportunities with his discussion of square waves and sampling though. I'm not sure that all in the presumed audience would necessarily come to the discussion with a full understanding of the fundamental connection between frequency and rise time. God love 'em.


       


      PS... speaking of free/cheap video software as folks were in another thread, I couldn't help but notice at the end that this vid -- which admittedly will not make Steve Speilberg quake in fear but is a pretty nicely done little piece -- was created using all free, Open Source software!


       http://xiph.org/video/vid2.shtml (No info on the software on that page but there's a list of the software they used at the very end of the vid. The vid is also downloadable in a number of formats in keeping with the Open Source ethos.)

      Attached Files

    • ermghoti II
      ermghoti II commented
      Editing a comment

      MikeRivers wrote:

      Why to people STILL have to debunk this?

       


      This is why.


  • #3

    This is such a fail (No offense Blue).  The screen resolution of analog oscilloscopes has never been adequate to show any stairstepping if there were any.  I'm not arguing the point that there is or is not stairstepping, but only that these tools would not allow us to see it at these frequencies if there were.  The equipment visually rounds it off no matter what it might really look like if we could see it.  Thus there's no debunking of anything here.  If it were only as easy as this presentation tries to make it.  Stairstepping is just one of the ways we try to conceptualize the digital problem.  The bottom line will always be there's something people can hear.  Technical speculation may fail to pinpoint the issue, but that doesn't mean its not there.  Once again, ears win when it comes to sound. 

     

    (Yeah, you can point the people at GS to this post here since I don't post there.)

    "Everybody loves you when you're six foot in the ground."
    ~John Lennon

    Comment


    • MikeRivers
      MikeRivers commented
      Editing a comment

      Beck wrote:

      The screen resolution of analog oscilloscopes has never been adequate to show any stairstepping if there were any.  I'm not arguing the point that there is or is not stairstepping, but only that these tools would not allow us to see it at these frequencies if there were. 

      It depends on the word length. If you were to sample 1 kHz at, say 4 bit resolution, you could certainly see stair steps on an analog scope, but only if there were any, which there aren't, as long as you look at the output of the reconstruction filter, which assures a band-limited output. A sine wave with real stair steps exceeds the bandwidth of any sampling system since each step has practacally a zero rise time.

      I don't know what the presenter in the video was looking at when he showed that waveform with the stair steps. Surely not what he showed that he hooked up.


    • blue2blue
      blue2blue commented
      Editing a comment

      Beck wrote:

      This is such a fail (No offense Blue).  The screen resolution of analog oscilloscopes has never been adequate to show any stairstepping if there were any.  I'm not arguing the point that there is or is not stairstepping, but only that these tools would not allow us to see it at these frequencies if there were.  The equipment visually rounds it off no matter what it might really look like if we could see it.  Thus there's no debunking of anything here.  If it were only as easy as this presentation tries to make it.  Stairstepping is just one of the ways we try to conceptualize the digital problem.  The bottom line will always be there's something people can hear.  Technical speculation may fail to pinpoint the issue, but that doesn't mean its not there.  Once again, ears win when it comes to sound. 


       


      (Yeah, you can point the people at GS to this post here since I don't post there.)




      Perhaps you were distracted when he addressed that concern at some length.


      Visualizing stairstepping with regard to sample values as a way to 'conceptualize' the 'digital problem' is like visualizing a  freeway filled with cows in order to visualize 'the freeway problem' -- it might be tempting but it is inaccurate. There are no cows and cows do not behave the same as humans in cars.


      And your apparent notion that the vagaries and inconsistencies of human hearing make it better for objective measure of audio accuracy seems to demonstrate you have a less than complete understanding of the human auditory system and human perception.  


      The human auditory system is capable of truly amazing things... just watch a blind person navigate by echolocation. It apparently evolved amazing capabilities for giving important audible cues about the environment,  everything from changes in weather to the approach of a large predator.


      But it didn't evolve in such a way as to make it good at objective measurement -- which over a century of perceptual testing has been underlined again and again.


      I don't hear things the same way you do. But I also don't hear things the same way I did yesterday or even two hours ago.


      As everyday experience and easy, real life experiments can demonstrate, and thousands of scientifc experiments and studies confirm, our hearing can change from minute to minute. We all know how hard it can be to hear when we have a cold or other congestion in the Eustachian turbes. Or how much what we hear can change after simple acts like yawning or even swallowing.


      But many folks don't seem to realize that, independent of changes within the body, very tiny environmental changes, from moved objects to very slight shifts in listening position/angle, even barometric pressure, can make subtle and surprisingly often not-so-subtle differences in how one hears the sonic events in that environment. For instance, it's been demonstrated that in a poorly treated listening room a shift in position of as little as 4" can mean a diffence of as much as 30-40 dB at a given modal frequency. 


      All of the above combine with various cognitive distortions to which all humans are variously subject to suggest that, if you want objective measure of audio, you need to use carefully designed tools and procedures.


       


      Now, none of that is to minimize the amazing nature of our hearing. 


      I just think it's important to understand its capabilities in a realistic, down-to-earth manner. Otherwise, one's own preconsceptions, fears, and other cognitive distortions may lead one down the primrose path of the self-agrandizing fantasy of magic ears.


       


       


    • learjeff
      learjeff commented
      Editing a comment

      Beck wrote:

      This is such a fail (No offense Blue).  The screen resolution of analog oscilloscopes has never been adequate to show any stairstepping if there were any.

      I don't think you're very familiar with what oscilloscopes are capable of.  When I worked at Ford Scientific Research Labs, they had oscilloscopes that could show microvolt signals with nanosecond-level responses, and that was back around 1980.  That's way good enough to show stair-stepping on a small signal (e.g., -86dBFS @ 16 bits) in the audio range.

      In any case, it's ludicrous to make generalizations about "analog vs. digital".  We have to discuss specifics to make any reasonable comparisons.  Some digital is better (more accurate) than some analog, and vice versa.  The only generalization that makes sense is that quantization error noise is nastier sounding than most common kinds of analog noise (but not all, e.g., 60-cycle buzz/hum).



Working...
X