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

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

  1. Pancakes

    Pancakes Friend

    Friend
    Joined:
    Aug 13, 2020
    Likes Received:
    878
    Trophy Points:
    93
    Location:
    Atl
    Parallel has an effect as well. It will dampen the impedance of the headphone seen by the amp. It's the same thing as an L pad on a tweeter.
     
  2. 7seven

    7seven Acquaintance

    Joined:
    Oct 28, 2018
    Likes Received:
    33
    Trophy Points:
    18
    Location:
    Ireland
    the input impedance of the analyser is very high. 100k I believe marvey set it for
     
  3. Marvey

    Marvey Super Friend

    Staff Member Friend BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,691
    Trophy Points:
    93
    To follow up on this, I took headphone measurements using Topping L30 and Jotunheim with a two tone 37Hz and 130Hz sine at about -10dBu. First of all, I'd like to establish that despite the worse THD behavior when the Topping L30 is used with a real load at 37Hz, the THD figure is still a very good -91db! It's just not -120db or whatever. Yes, the Jotunheim maintains cosmic excellent at -108db THD at 37Hz even with the load.

    However, when the measurements are taken at the output of the headphone (and again, I agree with Amir 100% on his assertion that this is where it really matters - if you care about measurements), then the results are more or less the same. So similar that I'm not even going to tell you which is which.

    Just know one is from an amp that measures with -108db THD and another -91db THD at 37Hz.

    upload_2021-7-6_14-59-0.png

    upload_2021-7-6_14-59-19.png

    There is still going to be the debate whether back-EMF induced behaviors, as measured at the amp output when the amp is loaded with a headphone, matters or if it's even real or not.

    However, there shouldn't be much debate whether there's any benefit with amps that have better than -80db THD or even -60db when it comes to the lows on most dynamic headphones which are open to allow ambient sounds to enter. The distortion of headphones is simply too high compared to even a "mediocre" SINAD amp. Again, this if you believe that THD / SINAD is a primary indicator of transparency.

    This is the big elephant in the room that ASR chooses to ignore: SINAD of upstream components does not matter unless it gets within a small distance of the distortion of a headphone.
     
    Last edited: Jul 6, 2021
  4. dsavitsk

    dsavitsk Friend

    Friend
    Joined:
    Oct 2, 2015
    Likes Received:
    1,481
    Trophy Points:
    93
    Home Page:
    Those sure don't look the same to me. The part that seems to be missing here is that high harmonics are considerably more damaging than low harmonics. That's why THD is nonsensical. If you look at an ABR (auditory brainstem response), consonant sounds are diminished in the brain stem while dissonant sounds are amplified. The more dissonant the more they are amplified. Consequently, 2nd harmonic distortion, and for the most part 3rd, goes unnoticed. But even small amounts of high harmonics generated by an amplifier, even if they appear to be buried in the noise floor, even if they appear to be swamped by driver distortion, make a big difference. It's a little like adding some amarogentin to your dinner - it does not take much to ruin the soup.
     
  5. Brad358

    Brad358 Rando

    Joined:
    Jan 14, 2020
    Likes Received:
    73
    Trophy Points:
    13
    Location:
    England
    My personal view is that true objective evaluation is to do a level matched double blind test, for the simple reason that if you can't discern any differences between equipment then measured differences are completely irrelevant. If you can discern a difference and have a preference (noting that even if you can discern a difference, the differences may be so inconsequential and/or difficult to discern as to be irrelevant in terms of listening to music) then that will probably be a subjective choice and it is entirely possible you may prefer the worse measuring item. The elephant in the room of measurement of audio gear is that measurement can tell you an awful lot about equipment, but it can't tell you what you like unless you already have a good understanding of your own preferences and how they relate to measured performance. If this makes me sound like I am against measurement then I'm not, I value measurement hugely. My objection is not to measurement but rather to how some are (mis)using measurement.
     
    • Like Like x 1
    • Dislike Dislike x 1
    • Respectively Disagree Respectively Disagree x 1
    • List
  6. Tachikoma

    Tachikoma Almost "Made"

    Contributor
    Joined:
    Oct 4, 2015
    Likes Received:
    320
    Trophy Points:
    63
    As AtomicBob had mentioned before, the results of a double blind test also reflect on the ability of the listener to detect minute differences within the conditions of the test. Differences between upstream components have a tendency to be cumulative - something that does not bother you in 5 minutes might be a dealbreaker in 1 hour. These differences are often very hard to detect in a quick A/B test.
     
  7. Brad358

    Brad358 Rando

    Joined:
    Jan 14, 2020
    Likes Received:
    73
    Trophy Points:
    13
    Location:
    England
    I agree that a double blind test is as much (if not more) about the listening sensitivity and hearing of the listener than it is about the equipment, but to me that is why it is so valuable. If differences are not discernible to you as a listener then they are irrelevant to you. Clearly what is discernible will differ between each of us so a double blind test is quite personal thing but it is interesting to read honest reviews based on an ABX or similar. My experience is that for speakers the speaker choice and set up is vastly more important than DAC and amplifier provided that the amplifier is appropriately rated for the speakers. Similarly for headphones, headphone factors are way more important provided that the amplifier is appropriately rated for the headphone. And really, the most critical part of the chain by far is the source material.
     
  8. Marvey

    Marvey Super Friend

    Staff Member Friend BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,691
    Trophy Points:
    93
    This is a new revision of the test PCB. The last one had higher crosstalk. This test board also replaced the 1/8" phono inputs with RCAs. Here I used a phono to RCA cable from the amp to the board. The cable here is much shorter, about 1 foot. The cables to the analyzer are also one foot.

    PXL_20210707_133111457 (Large).jpg

    Bottom lines are 1kHz. Top lines are 37Hz. This is with the Topping L30.
    upload_2021-7-7_8-39-17.png

    The results are better than prior. The voltage level at which the deviation occurs (presumably from back-EMF) is 100mV higher. I will await the arrival of a phono Y-splitter.
     
  9. Marvey

    Marvey Super Friend

    Staff Member Friend BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,691
    Trophy Points:
    93
    Now this is very strange. I repeated the same test again (top graph). I changed absolutely nothing. The results suddenly got worse, just like last time! I'm going to see if I can replicate this odd behavior with other amps.

    upload_2021-7-7_8-46-52.png

    A third measurement. Different yet again! I didn't touch anything.
    upload_2021-7-7_8-56-1.png

    Maybe the L30 doesn't like this test or is defective. This unit does measure perfectly with Amir-NADS: 1kHz SINAD though via a 300 ohm dummy load. 20kHz bandwith. 2Vrms out. High Performance Sine Analyzer On.

    upload_2021-7-7_9-0-11.png
    upload_2021-7-7_9-1-38.png
     
    Last edited: Jul 7, 2021
  10. atomicbob

    atomicbob dScope Yoda

    Friend BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    16,070
    Trophy Points:
    113
    Location:
    On planet
    I've seen odd behavior on several components. With time consuming effort they often have been traced back to poorly performing or transient malfunctioning external power supplies. Example:

    Dynamic Range varying substantially between measurements, after 2 hours of warmup:
    20210621 gamma2 dynamic range FFT spdif SE animation.gif

    Root cause wall wart having transient ripple issues after 2 hours of warmup:
    20210623 Jameco 150mA load 100mV div 5mS div 1MHz fltr animation.gif

    The behaviors shown above did not become frequent enough to really notice until after 2 hrs of warmup.
    Replacing the external power supply solved the problem.
     
    Last edited: Jul 7, 2021
  11. Marvey

    Marvey Super Friend

    Staff Member Friend BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,691
    Trophy Points:
    93
    I'm going to get rid of the Topping L30 for exploring this avenue. The quality of this unit is too suspect. There are some signs of the one channel being off one moment and then fixing itself again. (We can see this a few posts back in the top graph where I did L+R measurements). There could be a bad connection at the jack, bad solder, or pins on the bottom of the board shorting. I don't have time to track this down.
     
  12. Michael Kelly

    Michael Kelly MOT: Pi 2 Design

    Friend
    Joined:
    Apr 22, 2016
    Likes Received:
    3,469
    Trophy Points:
    113
    Location:
    Rhode Island, USA
    Home Page:
    I would gladly donate to your Patreon in order to allow you to purchase a couple of amps to continue this process. I bet others would as well if you came up with a list and a budget. Run it like a public radio begathon!
     
    • Like Like x 5
    • Epic Epic x 2
    • List
  13. mocenigo

    mocenigo Acquaintance

    Joined:
    Sep 26, 2017
    Likes Received:
    26
    Trophy Points:
    18
    Location:
    Munich
    Marv, I think the differences between the tests are so small that even just how a connector is, well, connected may influence the results. The contact surfaces are not perfect and definitely not perfectly flat and smooth. Therefore the actual contact surface may vary each time you plug and unplug something.
     
  14. Roget

    Roget Acquaintance

    Joined:
    Nov 17, 2018
    Likes Received:
    39
    Trophy Points:
    18
    Thank you marvey for explaining. I know headphones have high distortion. I did not know that 80db SINAD and 120db SINAD amp makes no difference. I thought it was garbage in garbage out so 80db SINAD amp + 40db SINAD headphone = 20db SINAD sound.

    I know SBAF people does not like me because I am objectivist. I learned something today. I do not like how Amir says what happens at headphone sound output matters but ignores what happens at headphone input.
     
  15. Andre Y

    Andre Y Friend

    Friend
    Joined:
    Oct 4, 2015
    Likes Received:
    220
    Trophy Points:
    43
    Location:
    Southern California
    I think it's interesting to look at the input impedances of the headphones being used, especially the phase because that tends to be capacitative near 40 Hz where those anomalies are showing up. Tyll measured headphone impedances a while ago with the results here as many of you know:

    https://www.stereophile.com/content/innerfidelity-headphone-measurements

    For example, the HD650's phase peaks between 30 and 40 Hz. A lot of other dynamic coil headphones behave similarly, and the Focals tend to present a larger reactive load than the Senns which seems to go along with Marv's measurement of higher distortion. Overall, headphones aren't very capacitative, but maybe enough to give opamp drivers some trouble especially when you're dealing with -90dB kinds of levels?
     
  16. Hartono

    Hartono Rando

    Joined:
    Jul 8, 2021
    Likes Received:
    2
    Trophy Points:
    3
    Location:
    Currently in Singapore
    New member here, I was reading topping L30 measurement and this back EMF headphone load issue.

    When it is mentioned balanced and transformer coupled circuit is not affected by this phenomenon, also some other amp have similar problem…. And test using output splitter might not exhibit this…

    I suspect this problem is caused by the shared return ground of single ended headphone connection, the combined ground return wire impedance in parallel with next channel headphone’s transducer creating EMF/reactance , creating less than perfect ground return. The other channel’s driver still connected to amp output trying to maintain zero volt while the driver is not fully resistive, creating a delayed and distorted (worse at high output level) voltage/current response in parallel with the ground return wire.
     
    • Epic Epic x 9
    • Like Like x 2
    • List
  17. Marvey

    Marvey Super Friend

    Staff Member Friend BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,691
    Trophy Points:
    93
    ^ This is would be appear what it seems. A thought experiment: remove amplifier with one channel to see what happens. I ran the tests with both channels running at the same time - which has you said is a double whammy with respect to back-EMF on the ground.

    Now for a history lesson:

    Back in the Headwize days, there was a bit of rage with balanced amps, and from that, a scaled-down solution: active ground. AMB offered designs such as a three channel (one active ground) M3 and I believe the Mini3 portable also implemented an active ground. @Jan Meier also offers active ground amps. NWAVGUY, if you remember him, shit all over AMB's active ground idea, on the grounds that he could not measure benefits to it (in hindsight, of course he couldn't if he just used a 1khz signal).

    The big question:

    Should do we amp tests with real headphones? Most certainly if we want to test the system and see what's going on electrically - more out of curiosity. There's been anecdotal evidence that when it comes to apples-apples (as close to it as possible), balanced output results in better bass quality. For example, the Jotunheim amps do exhibit better bass quality from balanced. I think the same for the Magnius from memory. On the EC and DNA amps which offer both SE and balanced outputs, this didn't seem to matter, possibly because these amps are output transformer coupled. (There were other audible benefits going balanced with these tube amps which can be debated, but I've never heard someone say bass quality was very much better).

    The fact is, even if we see THD or SINAD degradation, it doesn't matter. If the THD degradation is so bad, then we've already pushed the headphone way beyond what it should be doing anyway. For example, the Grado which "made" the amps' THD fall apart, those were being pushed at 0.775Vrms, which was a good 15db louder than the HD650 tested at the same level. In the end, at the output of the headphone driver, it doesn't matter so much if you look at THD.

    The only people afraid of using real headphones as a test load for amps are those with industry interests in maintaining THD / SINAD as the ultimate marketing tool to sell us crappy sounding gear with the lowest quality parts. And besides, because is some anecdotal information or some correlation to what we hear, it doesn't hurt to implement this. Let the people who want to cry about it call me dumb or ignorant or claim such a test results in fake data.

    Those who view THD / SINAD with a healthier understanding will have no issue with such an approach.
     
    • Like Like x 7
    • Epic Epic x 1
    • List
    Last edited: Jul 9, 2021
  18. Beefy

    Beefy Almost "Made"

    Joined:
    Apr 10, 2021
    Likes Received:
    1,086
    Trophy Points:
    93
    Location:
    Canada
    Alternative thought experiment. Can you run left and right channels perfectly out of phase in your testing? Net voltage across wires in the cable is zero (as per Jan's active balanced ground). And if I'm not mistaken, it would mean that the ground theoretically doesn't have to source or sink any current, everything from one driver just 'crosses over' the ground to the other driver.
     
  19. Andre Y

    Andre Y Friend

    Friend
    Joined:
    Oct 4, 2015
    Likes Received:
    220
    Trophy Points:
    43
    Location:
    Southern California
    You don't need to use real headphones if you can build a simulator load that looks like a particular headphones's input impedance. Everything that the amp sees in the headphones can be modeled as a lumped parameter L, R, C circuit, similar to what Stereophile does with their speaker simulator, and it will be simpler because headphones tend to be single driver systems with no crossovers. The load will simulate back EMF effects too.

    I think people don't use real headphones or loads that look like real headphones mostly because they're lazy. I mean it took Stereophile of all people to even make people aware of measuring power amps driving a real speaker load, and even then, they don't measure THD+N into their speaker load.
     
  20. Marvey

    Marvey Super Friend

    Staff Member Friend BWC
    Joined:
    Jun 6, 2021
    Likes Received:
    1,691
    Trophy Points:
    93
    I'm not so sure about that. With RLC circuit (easy enough), we are still assuming a linear system. The issue with headphone drivers is that they are not linear. Their characteristics change depending upon the voice coil's position in the magnetic gap. This is why they distort as we push them louder - and they will distort in unpredictable different ways. RLC networks will not do this, at least not anywhere near the extent of a physical piece of paper or plastic, finely wound coils bathed in a magnet's field of that tapers off toward the ends.

    Planars tend to distort less because their larger surface area does not need as much excursion. I think this lack of excursion is the reason why the effect is much much less. Most of the distortion is probably related to diaphragm deformation rather than motor nonlinearities.

    It's not the impedance curve causing this.
     
    Last edited by a moderator: Jul 8, 2021

Share This Page