Tutorial - Amplifier Distortion vs Amplitude - A Visual Guide

Discussion in 'Measurement Setups, Systems, and Standards' started by atomicbob, May 11, 2022.

  1. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    15,919
    Trophy Points:
    113
    Location:
    On planet
    The following is a visual guide to amplifier distortion vs amplitude with a schiit Vali 2+ and Ultron 7DJ8 tube as the example.

    00 202200510-31 vali 2+ THD+N  4+HD+N vs amp 300R - vol set for 0 dB gain +20 dBu.png
    Above is a typical tube or hybrid amp 1 KHz distortion spectrum. This graph will look unusual due to the dual axis delineations. On the y-axis blue is linear in units dBu while yellow is logarithmic in % distortion. Only the log scale is showing. The x-axis also has dual delineations with blue logarithmic in units Hz while yellow is linear in units dBu. This graph has been initialized for a measurement sweep of amplifier % distortion vs level in dBu.

    Amplifier gain has been adjusted for 0 dB (unity) and load is 300 Ω. In this measurement the signal generator is sending +20 dBu to the amplifier.

    01 202200510-41 vali 2+ THD+N  4+HD+N vs amp 300R - vol set for 0 dB gain +20 dBu annotated.png
    There are five meters displayed in the graph above.
    Upper left is a continuous reading of amplifier output level in dBu.
    The right column consists of four FFT derived readings:
    1) THD+N: a measure of the entire spectrum with the fundamental (1 KHz) removed.
    2) D2: 2nd harmonic distortion only
    3) D3: 3rd harmonic distortion only
    4) 4+HD+N: the distortion that remains after fundamental, 2nd and 3rd harmonics has been removed

    4+HD+N is what I call the Crap Factor™ (CF). It includes AC mains noise, higher order harmonic distortion, intermodulation distortion, inharmonic distortion, SMPS noise, system noise, etc.

    This spectrum is from a Vali 2+ with 7DJ8 / PCC88 tube at +20 dBu. At this level a typical dynamic headphone will produce between 120 and 125 dB SPL. Not a very likely use case if the listener wishes to avoid a rapid decline to deafness.

    THD+N and D2 are almost the same value indicating THD+N is dominated by D2.
    D3 is considerably lower by a factor of approximately 5x.
    4+HD+N is observed to be much lower than THD+N, a considerable disparity. In this measurement CF is dominated by the higher order harmonics.


    02 202200510-45 vali 2+ THD+N  4+HD+N vs amp 300R - vol set for 0 dB gain 0 dBu annotated.png
    Reducing the signal generator to 0 dBu (Vali 2+ remains set for unity gain) observe a substantial reduction in THD+N, D2, D3. Higher harmonics have nearly disappeared. Also observe how the 60 Hz mains noise remains constant. At this level D3 and CF distortion readings are nearly identical.

    At this level typical dynamic headphones will produce between 100 and 105 dB SPL.

    Harmonic distortions in a descending order and a CF of 0.0054% are consistent with a amplifiers that typically sound pleasing if accomplished with a low amount of negative feedback.


    202200510-30 vali 2+ THD+N  4+HD+N vs amp 300R - vol set for 0 dB gain - small.gif
    Performing the sweep from +20 dBu to -20 dBu observe the following
    1. THD+N and D2 are nearly identical above -10 dBu
    2. Higher harmonics begin to disappear below 0 dBu
    3. CF will hit a minimum between +5 and 0 dBu then begin to rise as AC mains noise become dominant in the measurement
    4. CF will rise and cross over D2 at -15 dBu

    Watch 60Hz, and to a lesser degree, 300 and 420 Hz remain relatively constant throughout

    Pay particular attention to the spectrum from which the sweeps are derived based on the specific measurement parameters. While the sweep graphs are useful, watching the spectrum over the level changes is considerably more informative.


    04 202200510-39 vali 2+ THD+N  4+HD+N vs amp 300R - vol set for 0 dB gain -20 dBu.png
    The final result of a sweep from +20 dBu to -20 dBu at 5 dBu per step
    Above -15 dBu THD+N is primarily D2.
    At -15 dBu, D2 and CF are approximately the same value.
    Below -15 dBu level 60 Hz AC mains noise dominates both CF and THD+N.


    05 202200510-22 vali 2+ THD+N  4+HD+N vs amp 300R - vol set for 0 dB gain.png
    A refined measurement sweep from +20 dBu to -20 dBu at 2 dBu per step

    THD+N sweep alone over-simplifies the underlying complex behavior exhibited by this amplifier and is true for nearly every amplifier.

    The behavior with a 1 KHz sweep over various amplitudes has been observed. Sweeps at several other select frequencies would be useful to determine the amplifier behavioral consistency.


    16 202200511 vali 2+ 1 KHz dist vs amp - 0 dB gain 4+HD+N 400Hz-20KHz - 300R - no FFT annotated.png
    In the above and subsequent measurements bandwidth for 4+HD+N will be altered
    from 22 Hz ~ 22 KHz (normal)
    to 400 Hz ~ 22 KHZ (AC mains noise reduction)

    Vali 2+ noise below 400 Hz is primarily a -90 dBu 60 Hz AC mains noise. By removing it from 4+HD+N result the amount of information reduction that occurs observing THD+N alone will be highlighted.

    In the graph above THD+N primarily reflects the contribution of D2 above -15 dBu. Below THD+N is comprised of mostly 60 Hz AC mains noise.

    The modified 4+HD+N now reflects system noise below 0 dBu. Above +2 dBu, 4+HD+N is dominated by higher harmonics.


    07 202200511-1 vali 2+ 100 Hz distortion vs amp - 0 dB gain 4+HD+N 400Hz-20KHz - 300R.png
    100 Hz distortion sweep for Vali 2+ with 7DJ8 from +20 to -20 dBu


    08 202200511-2 vali 2+ 1000 Hz distortion vs amp - 0 dB gain 4+HD+N 400Hz-20KHz - 300R.png
    1000 Hz distortion sweep for Vali 2+ with 7DJ8 from +20 to -20 dBu


    09 202200511-3 vali 2+ 5000 Hz distortion vs amp - 0 dB gain 4+HD+N 400Hz-20KHz - 300R.png
    5000 Hz distortion sweep for Vali 2+ with 7DJ8 from +20 to -20 dBu

    This amplifier demonstrates relatively consistent behavior at the various frequencies used to test amplitude vs level.

    Compare and contrast the foregoing extensive amount of data obtained from measurement sweeps against a single observation such as:

    0.46% 1KHz THD+N at 2V.

    Not only is the level ridiculous as being far above a normal use case, such a simple, single point observation represents a profound reduction of information to evaluate amplifier performance for this attribute.
     
    • Epic Epic x 18
    • Like Like x 9
    • List
  2. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    15,919
    Trophy Points:
    113
    Location:
    On planet
    In this section we take a more detailed inspection of analyzer settings to obtain specific numeric distortion observations.

    01 20220515 1 KHz distortion vs amp analyzer settings THD+N 3rd octave.png
    Top FFT reading in the column of four FFT displays is THD+N. Detector parameters used for the reading are shown in the large box in the trace window. Note the default band reject bandwidth uses a 1/3rd octave notch filter. The yellow trace depicts the default filter response. All components under the yellow trace are incorporated into the THD+N observation.

    01a 20220515 1 KHz distortion vs amp analyzer settings THD+N 3rd octave annotated.png
    THD+N reading is slightly lower than D2. The 1/3rd octave filter is attenuating the 2nd harmonic incorporated into the calculation by a small amount. Any noise closer to the fundamental will also be attenuated as shown by the filter response.

    02 20220515 1 KHz distortion vs amp analyzer settings THD+N window notch.png
    Changing the bandwidth of the notch filter to a window notch takes advantage of the FFT but is no longer the classical THD+N. Window notch doesn’t simply remove the FFT bin associated with the stimulus fundamental frequency. The window notch accounts for the FFT window type, processing bins on either side of the fundamental accordingly.

    Note THD+N and D2 are nearly identical given D2 is the dominant distortion for this THD+N measurement.


    03 20220515 1 KHz distortion vs amp analyzer settings D2 window notch.png
    Second FFT reading in the column is D2 using a window notch bandpass filter depicted by the green trace. Everything below the green trace is incorporated into the D2 observation.

    04 20220515 1 KHz distortion vs amp analyzer settings D3 window notch.png
    Third FFT reading in the column is D3 using a window notch bandpass filter depicted by the brown trace. Everything below the brown trace is incorporated into the D3 observation.

    05 20220515 1 KHz distortion vs amp analyzer settings 4+HD+N window notch.png
    Fourth FFT reading in the column is 4+HD+N, Crap Factor™. This filter version has a high pass filter at 400 Hz, 3 window notch band reject filters and a 22 KHz low pass filter as depicted by the red trace. Everything below the red trace is incorporated into the CF observation. I have two variants of CF. The wideband version has the high pass filter set to 22 Hz and is the observation reported unless specifically noted as shown.

    06 20220515 1 KHz distortion vs amp analyzer settings 60 Hz noise manipulated.png
    In this tutorial I wanted a prominent 60 Hz AC mains component present for purpose of demonstrating differences between THD+N and the constituent component distortions. The astute will note 60 Hz noise appears much higher than shown in the Vali 2+ Technical Measurements. I manipulated the analyzer to create this artificially high value.

    Vali 2+ is connected to the analyzer generator unbalanced BNC outputs. Setting the output to Balanced normal inserts 25 Ω between the BNC shield and chassis ground. Vali 2+ will be more susceptible to radiated AC mains.

    Further consideration. It is relatively easy to make a mistake, or more insidiously, purposely manipulate the analyzer to make a specific component measure worse than its full potential.


    07 20220515 1 KHz distortion vs amp analyzer settings 60 Hz noise correct.png
    Properly configured Vali 2+ 60 Hz measures 16.8 dB lower and is now consistent with those presented in the Vali 2+ Technical Measurements thread found here
     

Share This Page