New way to present distortion

Discussion in 'Measurement Techniques Discussion' started by purr1n, Dec 11, 2018.

  1. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,621
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    A while back @JohnM (creator of REW) mentioned an alternative way to present distortion. I believe he linked an AES paper. If someone can find that post, I'd appreciate it.

    At that time, I sort of discounted that approach because I felt normal distortion presentations were sufficient and already good enough - it was already kind of hard to get people to understand them anyway. Another kind of presentation would probably just cause confusion.

    Over time, I'm come to realize that distortion plots are still difficult to understand, so I've been playing around a bit. I don't know if I recall that AES paper correctly, but here is an approach I think we should consider. Normally, harmonic distortion (per order) is plotted at the fundamental with a different color. Second, third, fourth, etc. harmonics of a fundamental frequency, say 100Hz is plotted at the 100Hz point. It's up to the viewer to understand the gist of what second order distortion or third order distortion plots mean - that is crap at 200Hz and 300Hz that shouldn't be there when playing a 100Hz signal.

    This technique shifts the second order plot right to the second harmonic and the third order plot to the third harmonic. What this does is better illustrate the frequency of where the distortion actually manifests. The twist here is that the distortion signal is then taken relative to the amplitude (from the frequency response) at the frequency of where that distortion lies instead of the fundamental. For example, the amplitude of third order harmonic distortion of 100Hz is not measured relative to the 100Hz signal, but relative to the amplitude of the FR at 300Hz.

    With headphones, there are some good reasons for using this novel approach. For many normal distortion plots, we will see distortion spikes that appear an octave below FR spikes. These spikes really don't make much sense because the distortion probably isn't "real" because the distortion at those points is exacerbated by resonances inside the cup and/or cavity between the ear and the driver. By shifting the plot right (according to order) and also plotting against the amplitude at that point on the FR, we may get a more understandable or realistic presentation. If anything, distortion spikes, if they are indeed greater than what would be expected to be caused by internal resonances, should align at the points of these resonances from a visual presentation point of view.
     
  2. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,621
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Now let's take a look. The top distortion plot is what we've traditionally been seeing. The bottom one uses the above-suggested approach. This is the Dharma D1000, a headphone with great potential, but totally undone by the massive third order distortion (purple) of its dynamic driver.

    HD800dist.png

    If anything, the bottom plot emphasizes the harm of third order bass distortion and how it "pollutes" farther into the mids. It's kind of a "pollution" plot that illustrates how much crap that is there at that frequency that shouldn't be there.
     
    Last edited: Dec 11, 2018
  3. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,621
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Here is the good ol' standby, the HD800.

    HD800dist.png
     
  4. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,621
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Sony MDR-Z1R

    z1rdist.png
     
  5. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,621
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Focal Elex

    Elexdist.png
     
  6. monacelli

    monacelli Friend

    Pyrate
    Joined:
    Jul 23, 2017
    Likes Received:
    1,649
    Trophy Points:
    93
    Location:
    The Land of Enchantment
    Cool idea, @purr1n! I think you might be referring to [this post] from 2016 that references the AES paper: "How to graph distortion measurements" (Temme, 1993). For convenience, [here is the link] to the pdf that @JohnM originally posted. The thread "On distortion measurements - an experiment" started by @Serious last year also covers similar territory. In fact, the page linked in [this post] mentions the same AES paper.
     
  7. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,826
    Trophy Points:
    113
    Location:
    East Coast
    Adding one more data point.

    upload_2018-12-12_2-26-17.png

    upload_2018-12-12_2-26-28.png

    Yes, the newer way seems much more informational.
     
  8. Taverius

    Taverius Smells like sausages

    Pyrate
    Joined:
    Dec 27, 2017
    Likes Received:
    3,026
    Trophy Points:
    113
    Location:
    Rapallo, Italy
    Just needs another axis with % opposite the db axis.
     
  9. Ash1412

    Ash1412 Friend

    Pyrate
    Joined:
    Sep 27, 2015
    Likes Received:
    667
    Trophy Points:
    93
    If we're already shifting the xth order plot to the right frequency, why not add them up?
     
    Last edited: Dec 12, 2018
  10. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,621
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Is there an option to normalize the distortion to the actual frequency of the excitation point?
     
  11. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,621
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Was thinking about that and I'm not sure it makes sense to combine them.

    Let's say we look at distortion excitation at 1kHz. The 2nd order plot would be the distortion from the fundamental at 500Hz, the 3rd order line would be the fundamental at 333.33Hz and D4 from 250Hz.

    I guess you could make an overall assumption that music is complex and that all three points from 250Hz to 500Hz are going to be active at the same time or very close together in a time interval.

    I'd like to hear thoughts about this.
     
  12. Vtory

    Vtory Audiophile™

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    10,826
    Trophy Points:
    113
    Location:
    East Coast
    Sadly no -- at least with the current REW v5.2.0 beta 1. Of course we can somehow do it in dirty ways (manually).. For example, my graph was done with excel and exported (from REW) text files.

    I am not a fan -- indeed kinda hate -- of the idea of "total" (harmonic) distribution for several reasons..
    • In most cases, thd (say, D2+D3+D4+D5) closely follows D2 (H2).
    • If this is not a case (e.g. D3 > D2 for some freqs), summation is just misleading -- particularly for average hobby people including myself.
    • Like you said, harmonics D_N at (excitation) f can be viewed in multiple ways: (1) excitation at a f or (2) changing timber when the fundamental is at f/N.
    And for the same reasons, I don't think it makes sense to adding up all distortions.
     
  13. JohnM

    JohnM Author of REW - Rando

    Joined:
    Dec 13, 2015
    Likes Received:
    134
    Trophy Points:
    33
    REW allows the reference for normalisation to be either the excitation frequency or the harmonic frequency, per the distortion graph control option. Normalising to the response at the harmonic frequency was added to account for responses that are not flat, per Temme's paper.

    [​IMG]

    Plotting harmonics at the frequency they occur rather than at the excitation frequency that generates them does make life tricky for the graph legend - cursor values at 1 kHz (say) would be showing the 2nd harmonic level of 500 Hz, 3rd harmonic of 333.3 Hz etc. Could be a bit confusing.
     
  14. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,589
    Trophy Points:
    113
    Location:
    near Munich, Germany
    @Ash1412 I don't think combining them makes sense. There's a different sound to the different harmonics and adding that up would remove that information.

    I'm also not sure if moving the harmonics makes a ton of sense. The 3rd harmonic that generates a tone at 300Hz only gets excited by a tone at 100Hz. We lose that information from shifting them up. However I like the presentation using the harmonic frequency as reference in REW. It makes more sense to me subjectively. This will make bass distortion in dynamic driver headphones look a bit more concentrated around the midbass instead of rising upwards into the subbass. The less linear the frequency response, the more it makes sense to me. For speaker measurements with room modes (peaks and dips all over the place) it's absolutely necessary to get a legible graph at all. For headphones it's more debatable.
     
  15. Ash1412

    Ash1412 Friend

    Pyrate
    Joined:
    Sep 27, 2015
    Likes Received:
    667
    Trophy Points:
    93
    Yeah I think its a showing cause vs effect problem. The traditional way would show what excited frequency causes what distortion. The new way would show what frequencies are affected by that distortion when the excitation occurs. I was curious about adding them all up because I wanted to see what the distortion would look like in higher frequencies where the notoriously bad high-order distortion is, and modern drivers usually show deceptively low distortion in high frequencies when using the traditional method because 2nd order for 10k is already at 20k. Adding them up would show stuff thats usually not visible like 5th order distortion from 50hz bass doing weird things to the mids, Focal drivers clipping with shaker tests, etc...
     
  16. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,621
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    ^ yup on cause and effect.

    The traditional graph is the distortion where it arises from. The Temme method is distortion where it happens.

    I am tending toward the latter because of my experiences explaining distortion graphs to people who are not well versed in them.

    It also demonstrates how much suckier high 3rd order is in the lows. There is a reason why I picked the Z1R and D1000 has examples.
     
  17. Taverius

    Taverius Smells like sausages

    Pyrate
    Joined:
    Dec 27, 2017
    Likes Received:
    3,026
    Trophy Points:
    113
    Location:
    Rapallo, Italy
    I really like the graphs where you have both, the only problem: how to put distortion % numbers on the right axis for both at once.
     
  18. JohnM

    JohnM Author of REW - Rando

    Joined:
    Dec 13, 2015
    Likes Received:
    134
    Trophy Points:
    33
    Just labelling an axis doesn't work, for percentage figures you need normalised plots. REW gets around that by offering an option for the graph legend values to show distortion percentages while the graph itself still shows the absolute dB plots.
     
  19. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,621
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    I've moving away from % distortion figures. It's too hard for people to understand the difference between 1% to 3% and 0.05% to 0.15% distortion because % should be plotted logarithmically to make sense. This goes way back to reviewer peeps who have posted % distortion graphs linearly to say "Looky, everything below 1% on the 0% to 100% scale - can't see anything - no distortion at all. Awesome!"

    I am beginning to think that posting distortion graphs in db is better. People have a better sense of what db means compared to %. db expresses a ratio based on log scale already, which is how we hear things.

    dbdistplot.png

    It's easier and makes much more sense for people to say, that that 3rd harmonic is "-40db below" rather than "1% distortion" because 2% distortion happens to be -34db. The percentages don't make sense from a human perspective unless you look at these graphs all day and can automatically visualize / process the % scales in log.
     
    Last edited: Dec 15, 2018
  20. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,621
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    The relatability of db to human experience is crucial, especially as we near 2019, and measurement equipment detects stuff down to -140 / 150db. The downside to super measurement gear is that people get super neurotic and weird and start chasing measurements for the sake of measurements, and as a result, stop listening to gear. At the end of the day, it's still a subjective hobby.

    I don't read Amir's site, but I had heard that its denizens congratulated themselves on pushing @schiit to make the Modi 3 measure excellently. I hope to God not - for two reasons: 1) the prior Modi 2U DAC measured good enough (and depending upon who did the measurements, better than the much heralded Topping D30); 2) @schiit may have ended up unnecessarily spending more time tweaking the PCB layout rather than listening to the unit and making tweaks based on listening. There is far more to how humans hear than can be neatly encapsulated into a handful of measurements, i.e. %THD and Amir-bits.

    To illustrate where I am going, let's think of a distortion figure, say 1%. Ugh. That's horrible right? Especially when we are used to 0.001% in 2019. Well, translated into decibels, 1% is -40db.

    I'll admit, -40db isn't that great, but let's take examine things in proper perspective. Say we are playing music in the house at 75db SPL, which is about as loud as a vaccum cleaner or busy street in NY. If distortion is -40db below that, we are talking about 35db SPL. 35db SPL is like a quiet room at night or in the library. I know for sure 35db is much quieter than is typical for inside my house. If we sit still and listen intently to background noise, we start to hear all sorts of stuff: the neighbors coughing, planes in the sky, computer fans, light ballasts, a guy three streets over flooring his Bugatti Veyron, etc.

    I've been asked before what I think is a good distortion figure. My answer is always 0.1% or 0.05%, if we really want to be sure, 0.01%. That's -60db, -66db, and -80db respectively.

    If you listen to music at 80db SPL, you can't hear below the threshold of hearing at 0db SPL. People can't hear stuff 55 to 65db below anyway - effects of masking, etc. and those figures are being generous.

    This is why I strongly feel that manufacturers and "sciency" reviewers of the "measurbator school" should shove their AP-555s up their ass. I don't know. Maybe these guys do listen to music at volumes equivalent to the USS Iowa shooting all her big guns at once.
     
    Last edited: Dec 15, 2018

Share This Page