On distortion measurements - an experiment

Discussion in 'Measurement Techniques Discussion' started by Serious, Apr 30, 2016.

  1. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,589
    Trophy Points:
    113
    Location:
    near Munich, Germany
    Continuing this discussion from the One Day Ticket for Stax Galaxy - The STAX Compendium thread.

    An excerpt from post #42:
    To see if this was really the case, I decided to conduct an experiment. First, I measured the HD800 at a very loud level (about 100db). To see how much distortion is caused by the headphones and how much is caused by my mic + preamp + adc combo I then put foam between the headphone and the coupler plate, to decrease the level at the microphone. I used 1 very porous foam material (like a sponge) to break the seal and thus lower the level. Finally, I also used the thicker material that my coupler is made of (described in this post).
    I calibrated the volume level by using the specced mic sensitivity (15mV/Pa) and line in sensitivity (2Vrms) in ARTA. (Remember that my setup has no gain between the mic and the input) I don't think this calibration is very accurate at all, but I simply do not have the means to calibrate it more accurately (also added this to the other post).

    Here are the results in frequency response:
    1 - distortion FR.png
    As you can see, breaking the seal dramatically lowers the bass response. This was to be expected.

    Because the FR is changed so dramatically, we now absolutely have to check the "use harmonic frequency as reference" box. Without this the below results would be useless.

    Here are the distortion plots for each measurement:
    2- On coupler dist.png
    Yuck! ZOMG! WTF IS UP WITH SENNHEISER? 1% DISTORTION THROUGHOUT THE MIDRANGE!?

    3 - 1 foam.png
    o_O "Why did it get lower?" But wait ... there's more

    4 - 2 foam.png
    Note how the D3 distortion in the bass stays mostly constant. This is actually the headphone.

    5 - 1 thick foam.png
    This one is higher - look at the FR. This is because this is actually louder up to 800Hz.

    6 - 1 thick + 1 thin foam.png
    Now the noise seriously cripples the measurements. The only thing that we can say for sure is that the HD800 probably produces less than 0.3% of distortion at 100db - and very likely much less than this.

    Conclusion:
    Know the limits of your measurement rigs. Mine should never be used to measure distortion at 100db. I even have my doubts if it is accurate at 90db (it probably isn't) and most plots would most certainly be limited by microphone noise. I'm sure the measurement rigs used by the other folks have lower distortion and lower noise than mine, but I'm still not certain that the results for super low distortion headphones, like the STAX electrostatics, are accurate. I simply don't believe that the SR009 has 0.2% D2 at 90db throughout the midrange. It is probably lower than 0.05% (Even the HD650 has less than 0.2% distortion. Those plots look more like less than 0.05% and limited by noise). We would have to drastically lower the noise of our measurement systems (The noise floor of my Schiit Vali is easily audible to my ears, yet my measurement system doesn't pick it up). The first step here would be to use larger microphone capsules, 1/2 inch or even 1 inch (those are maybe too rolled off up top), but those cost hundreds of dollars alone. Looking at the specs I really like the ACO 7012, B&K 4191 (and maybe G.R.A.S. 40AC and Gefell MK 202) capsules.

    This basically means that we should look at the distortion measurements with a grain of salt. When multiple different headphones all measure way too close in terms of distortion (every headphone in the STAX galaxy thread measures at a nearly perfect 0.2% D2 at 90db), it is very likely that these plots aren't limited by the headphones.
     
    Last edited: Apr 30, 2016
  2. sorrodje

    sorrodje Carla Bruni's other lover - Friend

    Staff Member Pyrate MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    2,812
    Trophy Points:
    113
    Location:
    Dijon / Burgundy / France / EU
    I hear your but why Stat headphones distorsion should be that different for you .

    Do you Suggets those Distorsion plots of different headphones are not different enough ? !

    HE60 DIsto L.jpg

    Dharma_ Distorsion_L.jpg

    HE400S_Distorsion.jpg
     
  3. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,589
    Trophy Points:
    113
    Location:
    near Munich, Germany
    I just think that STAX have less than 0.2% 2nd order distortion at 90db. I think there are graphs that prove this. The SR-Omega pic I linked in the other thread does show much less distortion.
    Measurements for headphones with more than 0.2% distortion will be accurate on your coupler (Dharma, ...).
     
  4. sorrodje

    sorrodje Carla Bruni's other lover - Friend

    Staff Member Pyrate MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    2,812
    Trophy Points:
    113
    Location:
    Dijon / Burgundy / France / EU
    So we can conclude that all Stax and probably all Stats exhibit Stellar Distorsion performance than our rig can't really do justice to? All those demonstrations for that somewhat obvious Statement ??
     
  5. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,589
    Trophy Points:
    113
    Location:
    near Munich, Germany
    Yes, essentially this:
    I just don't want that noobs get confused by the different results.
     
  6. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,959
    Trophy Points:
    113
    Location:
    Irvine CA
    I'm not 100% sure I understand the experiment. But it looks like you are over driving something. Not sure if it's the DAC or the ADC. Probably the ADC.

    One nice thing about the 2i2 is that it sort of visually tells you if it's clipping. I can do some experiments on with levels also. Note that IMO seal helps the headphone driver behave linearly when pushed hard. But I'm no expert in headphone/speaker drivers.
     
    Last edited: Apr 30, 2016
  7. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,589
    Trophy Points:
    113
    Location:
    near Munich, Germany
    As far as I know it's that the phantom supply isn't an ideal load for the mic and this causes distortion(?). I'm not sure if I understood that correctly tho and the link I linked in the other thread explains this much better than I ever could.
    The thing is that I seem to notice these issues (D2 distortion higher than the headphone alone) on most measurement rigs, just to a varying degree. Your measurements show less D2 than sorrodje's. Marv's IEM rig shows more distortion than Rin-Choi's, ...
    This is of course only if the volume is calibrated accurately.

    EDIT: Yes, I agree. Frankly, I also expected distortion to rise without a good seal but this wasn't the case. I am using the harmonic frequency as a reference, which is super important when comparing measurements of the same transducer with different frequency responses. D3 was super consistent which shows that these are indeed comparable, even if the FR is so different.
     
    Last edited: Apr 30, 2016
  8. JohnM

    JohnM Author of REW - Rando

    Joined:
    Dec 13, 2015
    Likes Received:
    134
    Trophy Points:
    33
    A couple of things to note when measuring distortion with log sweeps:
    • It is important to use a long sweep setting (1M for REW) to maximise the separation of the distortion harmonic images in the IR, shorter sweeps may suffer from overlap between the harmonics giving misleading results
    • Log sweeps struggle increasingly to separate the distortion from the noise floor at higher frequencies and for higher harmonics. Tends to be more of an issue for measuring electronics than for transducers though.

    In REW V5.19 beta 5 I added a stepped sine distortion measurement feature. It is of course agonisingly slow, but can resolve distortion much further below the noise floor than the log sweep and also captures traces for noise and non-harmonic distortion.
     
  9. Priidik

    Priidik MOT: Estelon

    Pyrate BWC
    Joined:
    Sep 27, 2015
    Likes Received:
    2,148
    Trophy Points:
    113
    Location:
    Estonia
    Sennheiser spec for HD800 says 0.02% at 1 kHz, 1 V (102 dB). I have always wondered if they are full of shit with that claim.
    It starts to look like they aren't.
     
  10. Biodegraded

    Biodegraded Friend

    Pyrate Contributor
    Joined:
    May 28, 2017
    Likes Received:
    7,952
    Trophy Points:
    113
    Location:
    Vancouver BC
    A recent blunder has possibly illustrated limits for my particular rig. I'd be interested in opinions from those more knowledgeable.

    I never had great faith in my system’s resolution, but I’d been wondering why the distortion I measured on the HD598Cses was so stupidly high cf. the manufacturer’s spec (< 0.2% at 1 kHz at 100 dB) and given the better measurements I’d made on other phones. Then I recalled that what I’d changed to allow for the higher sensitivity of the 598s was the ‘input volume’ in the REW preferences (reduced from 1.0 to 0.6).

    Here are the results at 0.6 using a 1M sweep as recommended by @JohnM; SPL is nominally 90 but my calibration method is pretty hokey:

    [​IMG]

    [​IMG]

    Returning input volume to 1 and getting the SPL down entirely by reducing the volume at the output gave this result (same regardless of different permutations of amp pot, Windows, and REW ‘output volume’):

    [​IMG]

    [​IMG]

    My conclusions from this (by all means correct me, I’m here to learn) are:
    • my mic/input interface distortion is least at unity input gain but at the expense of more noise
    • the practical lower THD limit of my rig at (approx.) 90 dB is no better than c. 0.2%
    • (tentatively): noise floor through the midrange at unity input gain is about 15 dB? (see 4th plot).
    Must download the new beta and see what how the stepped sine-wave method breaks out the noise in the different cases...
     
  11. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,623
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Could be something happening with your microphone. D2 shouldn't track the FR that like, especially with dynamic type drivers. I discovered a simliar issue with my IEM microphone a while back.
     
  12. Biodegraded

    Biodegraded Friend

    Pyrate Contributor
    Joined:
    May 28, 2017
    Likes Received:
    7,952
    Trophy Points:
    113
    Location:
    Vancouver BC
    Hmm, I see a similar pattern with other 'phones I've measured...

    Could also be the USB 'soundcard' dongle I'm using; I'll do some checking; thanks.
     
  13. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,623
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    In your last plot, you are getting the second harmonic 40db down. That's 0.05%. 0.05% is still very good, so the bottlenecks can be anywhere: amp, microphone, microphone pre, etc.

    BTW, I use a Magni 2 and Modi 2 to drive the headphones because of the MM stack's low distortion. I may probably switch to Jotunheim as the amp since it measures even better. The Dayton USB microphone also seems to have low distortion characteristics.
     
  14. JohnM

    JohnM Author of REW - Rando

    Joined:
    Dec 13, 2015
    Likes Received:
    134
    Trophy Points:
    33
    40 dB down would be 1%.
     
  15. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,623
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Brainfart. Yeah, it's 1%. In that case, check the microphone and microphone gain, although it doesn't look like it's a gain / noise issue since D3 and above look good.
     
  16. Biodegraded

    Biodegraded Friend

    Pyrate Contributor
    Joined:
    May 28, 2017
    Likes Received:
    7,952
    Trophy Points:
    113
    Location:
    Vancouver BC
    I did wonder where that extra zero came from, but presumed it was something I really wasn't understanding :D

    I'll try different DAC & amp combos, but given the lack of difference when changing output software & hardware settings vs the big difference made by the input setting change, it does seem more likely the problem is on the input side. A hopefully better mic interface is incoming; if it makes a difference, I'll report back.
     
  17. JohnM

    JohnM Author of REW - Rando

    Joined:
    Dec 13, 2015
    Likes Received:
    134
    Trophy Points:
    33
    I've made a few updates in V5.19 beta 6 to help with distinguishing noise from distortion and recognise where the measurement chain's noise floor is intruding. The noise floor is captured before the measurement starts (sweep or stepped) and there is an option (which is on by default) to mask harmonics that lie beneath the noise floor.

    There's more info about the new features of the distortion graph at http://www.roomeqwizard.com/help/help_en-GB/html/graph_distortion.html
     
    Last edited: Jun 24, 2017
  18. Dr. S

    Dr. S New

    Joined:
    Aug 28, 2019
    Likes Received:
    5
    Trophy Points:
    3
    Location:
    Germany
    Another necro, please forgive me!

    Has anyone seen distortion graphs for the reference pressure microphones used in the dummy heads by B&K, GRAS or head Acoustics? I wonder how low the THD will be in the 90 to 110 db region.

    Regards
    Dr. S
     
  19. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,589
    Trophy Points:
    113
    Location:
    near Munich, Germany
    What's more interesting (to me) is finding out how much distortion typical consumer electret mics have. For example my UMIK-1 seems to get 0.2% distortion at 90dB and around 0.5-0.6% at 100dB! Looking at other people's measurements this seems to be common. Sorrodje gets pretty much the same distortion. My current WM-61A (with Solderdude's circuit) seems to get slightly more, but that is not the norm and you can cherry-pick them to get significantly lower distortion than that*. Hands seems to get similar distortion, though.
    With a Primo EM172 plugged into the same circuit I get around 0.15% at 100dB, which is a lot more usable. Should be 0.05% or less at 90dB. I bet using a better circuit or an EM173 you can get even less distortion than that. (Or perhaps by cherry-picking microphones. But at more than $10 per microphone instead of 50 cents like the WM-61As ordering many and cherry-picking makes less sense.)

    At 90dB noise also plays a significant role IME. WM61A may be around 22dBA. EM172 is specified around 14dBA (I measure a tiny bit more, but let's not argue about 1 or 2 dB). Actually 14dBA is not bad, but headphones usually have very low distortion at 90dB in the midrange. You can easily hit the noise floor at 90dB and then you need to run very long sweeps or even long FFTs to get a reading (which I think is super annoying). I measured -83dB at 96dB for the HD800 here: https://www.superbestaudiofriends.o...stortion-cancellation-ideas.8682/#post-282433

    At 100dB my current EM172 rig (which is not very optimized, yet) still has more distortion than the HD800 or HD600. I measure D2 at around -75dB at 1kHz for both the HD800 and HD600, but while the HD800s D3 is more than 10dB down, HD600 has similar amounts of D3 and D2 driving up the THD. For comparison: the HE 1 is specified at -80dB THD at 1kHz at 100dB, while the HD800 is specified at -74dB THD at 102dB at 1kHz, very close to what I get. SR-Omega is at -70dB D2 according to an ancient STAX measurement.**


    As far as real measurement equipment goes it's gonna be very, very low, to the point where it's going to be very hard to measure, I suspect. I think they use special resonance tubes with a very high Q when trying to measure microphone distortion.

    On page 121 here: https://www.bksv.com/media/doc/ba5105.pdf
    You can see that the B&K 4191 1/2" free field microphone has slightly less than 0.1% distortion at 135 dB, decreasing/increasing by 10x every 20dB. That'd give us 0.001% at 95dB, way below the noise floor (maybe with insanely long FFTs :p). Others are similar. Smaller, less sensitive microphones supposedly have less distortion (less diaphragm excursion).

    Page 19 from here on: https://www.bksv.com/media/doc/bv0048.pdf is also interesting.
    There's a part about distortion in the B&K microphone handbook here: https://www.bksv.com/media/doc/be1447.pdf

    *You can see an example of that here: https://www.superbestaudiofriends.org/index.php?threads/hd800-measurements.2720/#post-80296
    ** Both of the HD800 and HD600 have D2 distortion peaks in the upper midrange with significantly higher distortion than that, around -55dB at 100dB.
     
  20. Dr. S

    Dr. S New

    Joined:
    Aug 28, 2019
    Likes Received:
    5
    Trophy Points:
    3
    Location:
    Germany
    That is some interesting stuff, thanks for the links!

    That's what I speculated on the past few days, comparing data sheets, reviews and measurement reports.
    And so far, I didn't come to much pleasing results.

    Even when handpicking capsules, the mechanical properties of cheapo microphones are still way too random (frequency, impulse and polar response, also self noise versus overall sensitivity). You might improve the performance with some tweaks like a 3-wire or charge-amp circuit (see mic builders community for documentations) or run longer sweeps with sophisticated filtering. Still, without a reference system to verify your findings with a specific sample it will always be kind of vague.

    Some references:

    GRAS 40AG: 20 dB(A), +5 dB(A) with preamp
    B&K 4944: 48 dB(A)
    B&K 4938: 30 dB(A)
    HEAD Acoustics HMS IV: 16 dB(A)

    I don't think that the ~15 dB(A) of the Primo EM172 or PUI 5024HD (similar performance but much cheaper in smaller quantities btw.) perform that bad in terms of noise. :)

    But yet again, you probably have to filter a lot on the lower scale ...

    That looks very promising, indeed!
    But I'm not sure if you can interpolate that down to the 90 dB region. I suppose the curve would flatten to a certain baseline when lowering levels. Until you finally hit the noise floor.

    Edit: I have read that as a rule of thumb the THD triples with every 10 dB of input. This does match the observations with the B&K specifications.

    Viele Grüße,
    Dr. S
     
    Last edited: Apr 19, 2020

Share This Page