Measurement methodology request

Discussion in 'Audio Science' started by Pancakes, Oct 4, 2021.

  1. Pancakes

    Pancakes Friend

    Pyrate Contributor
    Joined:
    Aug 13, 2020
    Likes Received:
    1,510
    Trophy Points:
    93
    Location:
    Atl
    Where appropriate (amps for sure), is it possible to include a graph of the distortion and noise waveform with fundamental notched out? There is potential to correlate shitty vs better sound based on this (despite overall great noise numbers).

    A couple of examples pulled from stereophile of the graph I'm talking about. I won't identify the amps because that's not the point of this post but the second comes from one of the most subjectively enjoyable designers.


    881A9E18-E4CD-4D58-81DD-AE09DB5A3189.jpeg 90A011A9-C9EF-4E15-A3B0-7849A390A0FD.jpeg
     
  2. BarnBurner

    BarnBurner Acquaintance

    Joined:
    May 24, 2021
    Likes Received:
    53
    Trophy Points:
    18
    Location:
    New England, USA
    This is very interesting, Pancakes--but I'm a measurement peasant so I'm easy to impress. Curious how this works. I'll Google it.
     
  3. rlow

    rlow A happy woofer

    Pyrate Contributor
    Joined:
    Jul 18, 2017
    Likes Received:
    8,017
    Trophy Points:
    113
    Location:
    Canada
    Hmm, never paid attention to these graphs before. Is this just an example of class A/B crossover distortion vs. class A distortion?
     
  4. Vtory

    Vtory Audiophileâ„¢

    Pyrate MZR
    Joined:
    Sep 12, 2016
    Likes Received:
    11,096
    Trophy Points:
    113
    Location:
    East Coast
    Isn't it just exactly equivalent to 1khz FFT plots? Just a matter of frequency or time axis presentation. One is perfectly dual to the other AFAIK.

    I also believe waveform can be a little more misleading because human eyes are less familiar with time axis.
     
  5. Beefy

    Beefy Friend

    Pyrate
    Joined:
    Apr 10, 2021
    Likes Received:
    1,887
    Trophy Points:
    93
    Location:
    Canada
    Yeah, it's basically the time domain view of the grass/shrubs/trees on the FFT we are now so used to seeing for SINAD and GONAD.

    I think it could be useful to see, on occasion. The recent review of the Lake People G111 with broadband noise at 1kHz intervals all the way up to and beyond 90kHz, for example. Just how funky does all that distributed noise make the actual waveform look?
     
  6. schiit

    schiit SchiitHead

    Pyrate
    Joined:
    Sep 28, 2015
    Likes Received:
    10,101
    Trophy Points:
    93
    Location:
    Texas
    Home Page:
    This is useful from a limited standpoint.

    Old audio analyzers (see Sound Technology--http://www.stancurtis.com/soundtech.htm) from the 1970s and 1980s were usually configured to output the distortion residual, so an experienced engineer could get a quick visual as to whether or not the distortion was "good" or "bad." (Smooth distortion residual = good, spiky = bad.) Stereophile probably carries this forward because they are familiar with these older kinds of distortion analyzers.

    This, however, was before the days of easily accessible FFT, which show the spectrum of the distortion--is it 2nd, 3rd, 4th, etc? Smooth residual distortion will be low-order--2nd and 3rd. Spiky residual distortion will have lots of components going all the way out--4th, 5th, 6th, 7th, 8th, etc. These days it's easier for an audio engineer to look at an FFT. Modern Audio Precision devices will happily output a distortion residual, but it's not typically used.

    To be clear, I'm not crapping on this idea. If it helps you visualize what's going on, great. it's just a bit old-skool on the audio dev side, even for engineers who specialize in analog stuff.
     
  7. Pancakes

    Pancakes Friend

    Pyrate Contributor
    Joined:
    Aug 13, 2020
    Likes Received:
    1,510
    Trophy Points:
    93
    Location:
    Atl
    I think it's visually intuitive. A minor glance is all you need. Yes, sometimes my brain is lazy.
     
  8. atomicbob

    atomicbob dScope Yoda

    Pyrate BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    19,560
    Trophy Points:
    113
    Location:
    On planet
    There is a substantial piece of information missing from the two graphs. The second trace in each graph must have gain applied to appear on the same y axis as the source signal. Let me demonstrate.

    Source signal 1 KHz at 0 dBu applied to a system. In this case a simple loopback from signal generator to oscilloscope input.
    01 20211004 5243b 1 KHz 0 dBu loopback distortion time domain.png

    If the y axis remains unchanged and the distortion is greater than 40 dB down, it will not be visible on this display.
    02 20211004 5243b 1 KHz 0 dBu loopback distortion time domain -80 dB.png

    The greatest FFT distortion components are observed approximately 80 dB below the fundamental.
    03 20211004 5243b 1 KHz 0 dBu loopback distortion FFT.png

    So was the gain applied to the residual distortion equal in those two graphs?

    If the "bad" distortion displayed were greater than 40 dB lower than the "good" distortion would it be audible? The answer is dependent on many conditions. The FFT spectrum is far more informative than a time domain version with gain applied after notch.
     
    Last edited: Oct 4, 2021
  9. Pancakes

    Pancakes Friend

    Pyrate Contributor
    Joined:
    Aug 13, 2020
    Likes Received:
    1,510
    Trophy Points:
    93
    Location:
    Atl
    As far as I know they simply shift up (and magnify?) the distortion so you can compare with the fundamental. Yes, in this case you're missing the db down level of the distortion itself...but that number is always listed somewhere. I just find it really easy to draw conclusions from a glance - "pretty wave" theoretically means "good", craptastick wave lookalike is "bad". Yes, you would need to also look at the actual level of noise as well.

    I'm a big proponent of simplifying things. No, not the ASR version of simplification but rather simplifying how something is communicated - it allows more people to get the benefit of the information.
     
  10. atomicbob

    atomicbob dScope Yoda

    Pyrate BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    19,560
    Trophy Points:
    113
    Location:
    On planet
    That number is very important when making comparisons. Without incorporating the gain difference visualizations may be quite misleading. Again some examples.

    1 KHz 0 dBu sine stimulus with 2nd harmonic distortion + noise amplified 60 dB on the lower trace:
    05 20211005 5243b 1 KHz dist h2 60 dB gain.png

    1 KHz 0 dBu sine stimulus with possibly mains distortion + noise amplified 90 dB on the lower trace:
    06 20211005 5243b 1 KHz dist H+N 90 dB gain.png
    With 90 dB amplification distortion has been magnified by an additional factor of 31.6 over the first display visualization.

    1 KHz 0 dBu sine stimulus with possibly mains distortion + noise amplified 60 dB on the lower trace:
    07 20211005 5243b 1 KHz dist H+N 60 dB gain.png
    This last visualization is a better comparison to the first as it maintains equal perspective of 60 dB amplification. In this view the mains distortion + noise appears much lower in linear amplitude to the first distortion display.

    Again this is all very easily seen on an FFT with ability to compare directly distortion component amplitudes.
     
    Last edited: Oct 5, 2021
  11. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    91,699
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    It's actually not. A minor glance would more likely than not look like a random waveform. As a result, audiophiles, being audiophiles, are going to stress about "imperfect" or "janky" looking waveforms to time's end.

    Unless you've trained yourself to look at waveforms and can say: oh that's a more odd harmonics and than even and it's slightly increasing with frequency. It's like folks who say that the burst response attack and decay envelopes are redundant because they can see it all in the impulse response. Sure... more credit to you if you can.
     
  12. Pancakes

    Pancakes Friend

    Pyrate Contributor
    Joined:
    Aug 13, 2020
    Likes Received:
    1,510
    Trophy Points:
    93
    Location:
    Atl
    Well, I think it's useful but if the majority here disagree, no biggie.
     
  13. Beefy

    Beefy Friend

    Pyrate
    Joined:
    Apr 10, 2021
    Likes Received:
    1,887
    Trophy Points:
    93
    Location:
    Canada
    Random musing...... I wonder whether shape recognition machine learning could extract any information from those raw waveforms, that is not evident from an FFT.
     

Share This Page