Oh shit! SINAD can suck my ****s

Discussion in 'Measurement Techniques Discussion' started by Marvey, Jul 2, 2021.

  1. GoldenOne

    GoldenOne Friend

    Pyrate
    Joined:
    Jul 17, 2020
    Likes Received:
    1,049
    Trophy Points:
    93
    Location:
    England
    It'd definitely be a pretty fantastic way to do things!

    Would be great to have proper measurements available from various people using a few common headphone model simulations. Much more realistic information for consumers to use
     
  2. nishan99

    nishan99 Friend

    Pyrate
    Joined:
    Dec 1, 2019
    Likes Received:
    1,617
    Trophy Points:
    93
    Location:
    Layla
    And more easy and useful for others trying to replicate and compare such measurements using the headphones instead of building headphone simulators.
    The latter maybe useful for you guys trying to test the EMF hypothesis but I don't think it's more useful for others than using the actual headphones.
     
  3. Marvey

    Marvey Super Friend

    Staff Member Pyrate BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,712
    Trophy Points:
    93
    Different methods and approaches, for us to discover relative truths within their own contexts. A lot of people can't handle this because they want absolute truths. Those who want absolute truths that can be distilled into betterer and bestest, I'll leave that to the TV pastors and religious zealots.
     
  4. ergopower

    ergopower Friend

    Pyrate
    Joined:
    Mar 21, 2018
    Likes Received:
    815
    Trophy Points:
    93
    Location:
    South Central PA
    I'm scratching my head at the above. My understanding is back EMF is the result of mechanical motion of the voice coil (or conductive planar array) relative to the magnet inducing voltage & current that is seen at the amp output. How would a lumped element simulate that?
    Note: @Marvey has already commented on the non-linear effect of the moving coil
     
  5. JohnM

    JohnM Author of REW - Rando

    Joined:
    Dec 13, 2015
    Likes Received:
    134
    Trophy Points:
    33
    That's more or less it, but the electrical impedance above the bass range isn't well modelled without some extra components. This is a slightly better model:

    [​IMG]

    If the impedance is measured with REW (or loaded into REW) the Thiele-Small tool will spit out a set of values in its "Simplified Model Parameters" box, just select the measurement as the "Free air measurement", enter a reasonably close DC resistance value (or guess one from the minimum of the impedance measurement) and click Calculate Parameters (don't need anything in the added mass measurement boxes).
     
  6. GoldenOne

    GoldenOne Friend

    Pyrate
    Joined:
    Jul 17, 2020
    Likes Received:
    1,049
    Trophy Points:
    93
    Location:
    England
    That's true. But then using the actual headphones introduces so many situational/setup/unit to unit factors that make relative comparisons fairly impossible.

    As @Marvey said, as much as it would be nice if there were a single truth or number and single all-encompassing test setup we could point to that'd be great. But unfortunately there isn't and so all we can do is find the balance between thorough whilst not testing so many variables that we drown in data and can't get anything useful or comparative.

    Having a fixed test rig such as this is useful because even if it doesn't exactly replicate the impedance curve of a headphone model, it's fixed, will not change based on placement, the headphone unit, or various other characteristics, and can be exactly replicated by anyone who has the same load or is able to construct the same circuit.
    Its not that this is 'the ideal'. There is no true ideal. Even if we had a real headphone that was fixed to a rig and constructed in such a way with such tolerances that all the units were identical and never changed, and was immune from external factors, that then wouldn't resemble other headphones.

    The reason we're doing this is because testing amps with loads that actually resemble and exhibit similar characteristics to headphones is Important. A purely resistive load could potentially be something which a manufacturer optimises for even at the potential expense of real world performance. And even if the manufacturer has not done that, products could simply happen to do better or differently with reactive loads.

    For example opamps in some configurations will produce excellent results with a resistive load. But end up being less stable with a reactive one. And can exhibit issues such as overshoot, ringing, and oscillation. https://www.analog.com/ru/analog-di...-to-avoid-instability-capacitive-loading.html
     
  7. Biodegraded

    Biodegraded Friend

    Pyrate Contributor
    Joined:
    May 28, 2017
    Likes Received:
    7,985
    Trophy Points:
    113
    Location:
    Vancouver BC
    3 load simulators, Senn HD6--, Focal Elex/Elear, and Grado SR, would seem to be enough to be broadly representative. Scanning through the Innerfidelity compilation, sure there are outliers but many other headphones have similar impedance profiles close enough to one of these or (more commonly) to flat. Add a couple of representative dummy single-number values to represent planars, and you'd be done.

    BA & hybrid IEMs are another story though - they're all over the place. Not sure if it's worth even trying to go there.
     
  8. Marvey

    Marvey Super Friend

    Staff Member Pyrate BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,712
    Trophy Points:
    93
    Indeed, if testing amps using real loads is too "flakey", that is we cannot get consistent results because things such as cables or connectors can cause wild fluctuations (consistency) or say I can get consistent results but others cannot replicate (repeatability), then the next best thing would be a static impedance simulator.

    This doesn't mean that if I can wring out consistent results using real headphones as loads that I will not use them. But it does mean that I should also use the impedance curve simulator load so that others can attempt to duplicate the results.

    Testing real loads, real life, where non-linearities rule, has never been easy. Think how an auto manufacturer would test for NVH? Testing on the roads in CC TX would be much different than the roads in Dallas. Different drivers could also result in different results even if using the same roads.
     
    Last edited by a moderator: Jul 9, 2021
  9. Marvey

    Marvey Super Friend

    Staff Member Pyrate BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,712
    Trophy Points:
    93
    Would be interesting to use them as real loads. :) Lots to discover and learn. We do not have all the answers. Learning stops when we think we have all the answers and assume we already know everything.
     
  10. atomicbob

    atomicbob dScope Yoda

    Pyrate BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    18,653
    Trophy Points:
    113
    Location:
    On planet
    There is a strong tendency to rank audio components based on a few measured metrics, the most common lately being THD+N and SINAD. Such information reduction discounts a tremendous amount of data while also promoting the winner / loser model of selection. Easy to understand this method is far too simplistic and ignores the subject of listener preference.

    Imagine a system where substantial amount of audio component measurement data could be mapped into a three dimensional characteristic profile. In a like manner listener preferences would be mapped into the same 3D profile. Listener satisfaction is enhanced where the audio system performance profile achieved a high degree of correlation to the listener preference profile. Determining a listener preference profile has largely been ignored, though there are some who have figured out systems which work for their own benefit. Creating a standard for this mapping is not trivial and will require considerable effort. But for the sake of this visualization assume such a standard method is possible.

    The following graphs have been created to assist in visualization. It should be noted the units in this example are completely arbitrary. The three dimensions include:

    1) SPL – level at which performance is measured or auditioned
    2) Dynamic Response – a mapping of many measurements into an approximation of the dynamic behavior of the system
    3) NFB vs Distortion Ratio – not just distortion but how the system design achieved that distortion. 2nd and 3rd harmonics are removed, 4+HD+N measured
    Example: System A measures 0.001% using 100 dB of NFB, System B measures 0.01% using 2 dB of NFB
    Ratio for A would be 100,000, while ratio for B is 200. There may be good reason to desire lower NFB Distortion ratio to a point.

    01 System performance profile 1 small.png
    System A performance profile displayed as a surface in a 3D


    02 Listener preference profile small.png
    A specific listener preference profile also mapped as a 3D surface


    03 profile mismatch small.gif
    Observe the disparity between preference and performance for this particular listener. Another system might be a better choice.


    04 System performance profile 2 small.png
    System B performance displayed as a surface in 3D


    05 profile match small.gif
    System B performance profile has better correlation to this listener’s personal preference profile


    06 System performance SINAD small.png
    Even without listener specific preference profiles consider the amount of information present when a single metric is mapped to a 3D performance profile. SINAD at a single level with a single stationary measurement is displayed for a particular system. The information reduction is obvious. Too much information is lost.


    07 Multiple system SINAD small.png
    Now consider using SINAD as a ranking method. Rather sparse when compared to the amount of information previously displayed.

    Even without a 3D mapping method, there is much information currently available when standard measurement suites are performed in a consistent way such that valid comparisons may be made. Information reductions to one or two metrics inadequately describe system performance and should be avoided.
     
    • Epic Epic x 31
    • Like Like x 5
    • List
  11. Marvey

    Marvey Super Friend

    Staff Member Pyrate BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,712
    Trophy Points:
    93
    OK, so here's an update. The simple Y-splitter came in today. Here are the results. They behavior still exists, but the point which the back-EMF rears it's ugly head upon the measurements comes at a higher voltage level. I suspect there are some interactions going on with the break-out board, maybe with the switches, which is amplifying the back-EMF effect. I will continue to investigate the cause of this.

    This is a voltage sweep with a 300-ohm load 1kHz compared to HD650 as the load at 1kHz and 37Hz.
    The Magni 3+ was used.
    upload_2021-7-9_22-28-50.png

    Again, Magni 3+, full spectrum view at 0.775Vrms. 300-ohm load (bottom) and HD650 as the load.
    upload_2021-7-9_22-44-59.png

    I was able to reproduce this result consistently several times over the course of the evening. The effect is still there, but less pronounced.

    That's it for tonight. Note that the Grado falls apart at a much lower voltage level. This could be that it's much more sensitive. I will demonstrate this another time. It may be a good idea to normalize them to SPL rather than use the same voltage level range.
     
  12. atomicbob

    atomicbob dScope Yoda

    Pyrate BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    18,653
    Trophy Points:
    113
    Location:
    On planet
    HD650 LCR data posted here
     
  13. Marvey

    Marvey Super Friend

    Staff Member Pyrate BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,712
    Trophy Points:
    93
    The next step I wanted to take, now after using a simple Y-splitter to be able to replicate Yang's results consistently, was to proceed with the same headphones measured prior. Note that I chucked the Topping L30. A precision device such as the AP needs an amplifier that isn't flakey, cheaply made with bad connectors, prone to static electricity killing headphones, etc. Here I'm going to use the IEMagni.

    One concern I had about tests before is that I used the same voltage level (0.775Vrms) or range (1m to 2Vrms). This skews the results because the Focal Elex and SR325 are much more sensitive than the HD650. I decided to set the voltage levels for each headphone to produce 100db SPL at 1kHz.

    HD650 0.864Vrms
    Elex 0.295Vrms
    SR325X 0.295Vrms (this is wrong, I need to redo this, but you should get the idea)

    Here are the results with the cheap Y-splitter. They are not as overblown as before with the test PCB with the switches, but we definitely still see the effect of the back-EMF not hitting a "stiff" active return on the outputs of the amplifiers when there is a real (traditional dynamic) headphone load.

    upload_2021-7-11_14-12-19.png

    What is interesting is that after normalizing volume levels, the Grado still exhibits a stronger effect than the Elex. And the Elex to the HD650. The next step would be to try different amplifiers.

    Indeed. Here I heeded John Yang's advanced knowledge and expertise of precision testing methodology (see cheap Y-splitter below) and was able to get very similar behavior (replicate his results) with the HD650. The outcome does change, where the back-EMF effect gets pushed to higher voltage levels. The effect is hardly noticeable with the HD650. However still quite pronounced with the Elex and Grado.
    DSC00517(2).JPG

    You know, I did mention several posts ago that I ordered this part. :rolleyes: Sometimes it's not good to go in with all guns blazing - some reading comprehension, deliberation of the meaning of words, and patience helps too.
     
    Last edited: Jul 11, 2021
  14. Andre Y

    Andre Y Friend

    Pyrate
    Joined:
    Oct 4, 2015
    Likes Received:
    220
    Trophy Points:
    43
    Location:
    Southern California
    There are lots of interesting questions your measurements have raised. One is how the response of the amp driving a simulated load compares to the real thing. My money is still on the increased capacitance around 30-40Hz making opamp drivers uncomfortable, but who knows? But maybe it's the non-linearities of the driver reflecting back onto the amp, or both?

    The mechanical parameters of the drivers can be converted into electrical equivalents in terms of impedances (like the Thiele-Small model mentioned above), so stuff like the driver's motion causing back EMF can all be modeled with a lumped circuit. The TS paper explains this more, and this is an alternate way of deriving it:

    https://www.har-bal.com/wp-content/uploads/Speaker_theory.pdf
     
  15. Marvey

    Marvey Super Friend

    Staff Member Pyrate BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,712
    Trophy Points:
    93
    You realize that paper has the exact same schematic that I whipped up on a paper towel, took a picture of, and posted two days ago? The only difference is that the paper goes to lengths to prove this via equations.
    https://www.superbestaudiofriends.o...-sinad-can-suck-my-s.11175/page-7#post-353173

    The problem is that the nonlinear distortion of the driver is actually being reflected back. Or at least the low frequency portions of dynamic drivers - the higher the Q of the driver, the greater the reflection (this part is conjecture). RL-RLC circuits cannot mimic nonlinear distortion of specific headphones like this.

    HD650
    upload_2021-7-11_17-44-26.png

    Elex
    upload_2021-7-11_17-47-16.png

    SR325X
    upload_2021-7-11_17-51-1.png
    upload_2021-7-11_17-52-1.png

    [​IMG]
     
    Last edited: Jul 11, 2021
  16. drgumbybrain

    drgumbybrain Science Nut

    Pyrate
    Joined:
    Jan 9, 2017
    Likes Received:
    2,219
    Trophy Points:
    93
    Location:
    Lives in Fortaleza, Heart in Girona
    unnamed.jpg
     
  17. BarryT

    BarryT MOT: Austin Audio Works

    Pyrate
    Joined:
    Feb 8, 2021
    Likes Received:
    425
    Trophy Points:
    53
    Location:
    Austin, Texas
    Home Page:
    WOW, on the recommendation of a friend I tuned into this thread, it is tumultuous. So I guess it is fair to throw in my two-bits worth based on having been active participant in it for over 60 years, having two product designs currently on SBAF tour, loving it and learning a hell of a lot from it.

    Part 1

    The fundamental issue is that there is and has been an ongoing unsustainable search going on in the Audio Community. It originated with talking movies and peaked in the Golden Age of Audio (1970's) and now it's being replayed because:

    a) all the original players are dead or retired so there is new gang at the wheel
    b) no one today studies history so limited critical thinking skills and reinvention of the past.

    This all is a very personal search for truth, more exactly a cathartic moment of truth.

    What is sought is the physics to explain a singularly human metaphysical experience.

    We want a number (or set thereof) to repeatedly tell us how good an illusion will occur in one's head.

    Before you get your knickers in a wad please remember that intrinsic to reproduced audio listening is an illusion, a hallucination .

    We who design audio stuff are illusionists, makers of 'eargams'.

    If listening to our work gets you off, then we did a good job If not, then our work, our 'art'. lacks the right anthropomorphic information alteration process needed to invoke altered reality in you.

    YOU, not them. All this exists in YOUR HEAD. Because we are built of the same stuff and have similar acculturation we can share the rudiments of a experience - but your experience only happens to you.

    In this we are typically humanly selfish, which is fine, even good, But we have to confront the reality that here is no single answer for all in any aspect of the human experience, so why should Audio be any different?

    Part 2

    I have been down this road, know it well, owned hardware production, have over 50 products on the market from 1968 on, 34 patents, physicist and cultural anthropologist from 8 years in college (no Nam for me, 2S all the way), know the history, made some of it,

    I also know you don't care about any of that, only about what you want, which seems to part of human nature. The goal is a simple answer that makes you secure.

    I can tell you lots of stories, show hardware, done experiments but the fact is what is most important is that I have the wisdom to know a 'religion' when I see it, and the "number" is a religion, defined as a based on your personal faith.

    If I can direct your studies or contribute to your wisdom then ask, otherwise remember this always -

    Data Points collect to become Information
    Information
    aggregates to become Knowledge
    Knowledge
    congeals into Wisdom
    Wisdom
    lets you understand what are the next Data Points

    Part 3

    A reminder.

    We all love music, all mankind that is; it is common to all groups of humans greater than zero.

    Why?

    Simple, the very first sensory events in your life were your mother's heart beat and the chemicals in her blood stream. The heart beat was a massive sensual pulse felt by all your nerve endings at one time (no - you didn't' hear anything, while in your mother, there were non-compressible liquids on both sides of you undeveloped auditory system and your nerve system would take months after your birth to get the right data, in the right form, to the right place in the brain for the mind to figure out what is sounds. You had no idea if mom was listening to any thing, all you knew was that mom was putting off good feeling endorphins and you probably wanted more of them. Her heartbeat was where you got learned rhythm, the program at the bottom of the stack and where music starts.

    My regards to your search,

    Barry Thornton
     
    • Like Like x 20
    • Dislike Dislike x 2
    • List
    Last edited: Jul 12, 2021
  18. bboris77

    bboris77 Friend

    Pyrate
    Joined:
    Dec 12, 2015
    Likes Received:
    778
    Trophy Points:
    93
    Hey @purr1n, here is something I received directly from Solderdude. He has been following this discussion and thought it may be helpful.

    upload_2021-7-13_16-16-58.png
     
    • Like Like x 8
    • Epic Epic x 1
    • List
  19. atomicbob

    atomicbob dScope Yoda

    Pyrate BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    18,653
    Trophy Points:
    113
    Location:
    On planet
    simulation of Solderdude's HD650 load simulator:
    HD650 simulator.png

    Parts ordered.
     
    • Like Like x 3
    • Epic Epic x 3
    • List
  20. atomicbob

    atomicbob dScope Yoda

    Pyrate BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    18,653
    Trophy Points:
    113
    Location:
    On planet
    With a few value tweaks this is the latest:
    HD650 simulator v2.png

    Comparison between tweaked value simulation and HD650 LCR Z measurements:
    HD650 LCR Z vs Sim Z.png
     

Share This Page