Feasibility of measuring headphones down to 0.01% THD

[Mr. Haelscheir]

Hello. I have been planning an audition for a high-end electrostatic headphone purchase, namely a choice between the Warwick Acoustics Bravura, Audeze CRBN, and Stax SR-X9000, whereby I am hoping to obtain comparative distortion measurements (among others) to bolster my decision, and of course provide the community with this data.

As far as I could tell, to measure down to 0.01% THD (a benchmark level apparently attained by the Sennheiser HE-1), if these headphones actually reach such performance (the DCA Stealth per Jude's measurements on Head-Fi reaches such in the treble), I would need to at the shop provide a portable acoustic enclosure ensuring a noise floor of 20 dB or less while measuring 100 dB signals at the mic. The mic's self-noise would have to be at that level, and its own distortion must not exceed that noise floor at said 100 dB recorded level.

The miniDSP EARS would be the most convenient setup, but does anyone know its self-noise and distortion specs (at least of the supposed UMIK-1 capsules used)? At the minimum per the link in https://audiosciencereview.com/foru...tting-thd-in-the-perspective.9085/post-232599, it attains 1% THD at a venerable 133 dB max SPL. https://www.avsforum.com/threads/the-great-waveguide-shootout.1475674/page-12 mentions issues with second order harmonics. But https://audiosciencereview.com/foru...l-world-performance-review.33918/post-1186487 claims the ability to "measure up to ~120 dB SPL without compression or added harmonic distortion."

If the miniDSP EARS is known to be insufficient for measuring down to 0.01% THD, how feasible would it be to DIY a rig within a $1k budget (where I am not so concerned about ear gain simulation)?

 

[staticV3]

The miniDSP EARS would be the most convenient setup, but does anyone know its self-noise and distortion specs (at least of the supposed UMIK-1 capsules used)?

My own UMIK-1 has an EIN of about 45dB SPL Z-weighted. Not the right tool for the job.

 

[solderdude]

Even if you could reliably measure well below 0.01% (noise floor, wideband) it is a level that is far below any audible thresholds so not very important.
Above 100dB many headphones start to 'compress' a bit already and above 120dB it will already be an audible mess (many many percent).

Averaging techniques can be used to measure below noise levels as well and one can only 'look' in very narrow bands which can also improve things.
For simple measurements (with sweeps) you will indeed run into practical noise floors, acoustical as well as electrical.

 

[Mr. Haelscheir]

0.01% would be the absolute lowest I would want to measure, or the target noise floor. Even if inaudible, I simply want an assurance that measured comparative performance limitations, say, at extremes, aren't being limited by the measurement setup. I also raise these concerns in https://www.audiosciencereview.com/...eeded-for-measuring-speaker-distortion.47099/ in regard to general measurement microphone shopping and the seeming prevalent lack of detailed distortion versus SPL measurements. Basically, how can we know whether the product we are looking at is "the right tool for the job" when the published specifications seem lacking? E.g. I cannot find detailed distortion versus SPL measurements for the Earthworks M23 and would not want to make a blind purchase only to find that my measurements are actually bottlenecked by the mic (e.g. confirming distortion measurements for a KEF Reference 3 while at the shop).

And to confirm, is it correct that if my acoustic enclosure and electronics chain allows for a noise floor of 20 dB to be measured, and that the microphone's own distortion products are below that noise floor for input signals of 100 dB, that I would be able to measure transducer distortion products up to 80 dB below the fundamental (0.01%)?

 

[staticV3]

And to confirm, is it correct that if my acoustic enclosure and electronics chain allows for a noise floor of 20 dB to be measured, and that the microphone's own distortion products are below that noise floor for input signals of 100 dB, that I would be able to measure transducer distortion products up to 80 dB below the fundamental (0.01%)?

Yes.

 

[Blumlein 88]

Usually with condenser microphones, the distortion limits at high levels are from the electronic circuitry in the microphone itself. Various manufacturers pick different distortion levels to decide peak SPL handling. .1%, .5% or 1.0% with some choosing 10%. Some have pads which reduce the signal between the mic diaphragm and the circuitry. So you'll see a 20 db pad gives 20 db louder max SPL at the same distortion.

So I presume the diaphragm does not contribute much distortion, but where that lies at lower levels I'm not sure. You would think for most circuitry distortion will be lower at lower signal levels.

 

[DVDdoug]

I'm not sure what the measurement thresholds are, or what the audibility thresholds are.

Even if inaudible, I simply want an assurance that measured comparative performance limitations, say, at extremes, aren't being limited by the measurement setup. I also raise these concerns in...

I know it's hard but I'd advise you to try to ignore what you can't hear.

This is the second time I've written this today but... I've NEVER heard ANY distortion from ANYTHING that wasn't broken, defective, or over-driven. With the possible exception of vinyl records that were "hard to play", back in the analog days... Maybe the record or cartridge weren't exactly defective, but the cartridge just couldn't track the particular record...

And something else I found fascinating - Dan Clark has a video where he mentions that headphones with more distortion are often described as having "more detail."

 

[Mr. Haelscheir]

I'm not sure what the measurement thresholds are, or what the audibility thresholds are.


I know it's hard but I'd advise you to try to ignore what you can't hear.

This is the second time I've written this today but... I've NEVER heard ANY distortion from ANYTHING that wasn't broken, defective, or over-driven. With the possible exception of vinyl records that were "hard to play", back in the analog days... Maybe the record or cartridge weren't exactly defective, but the cartridge just couldn't track the particular record...

And something else I found fascinating - Dan Clark has a video where he mentions that headphones with more distortion are often described as having "more detail."

Let's say it's more so that I have an insatiable curiosity as to whether these TOTL estats actually have as low distortion as claimed, and I am willing to add a headphone to my small lineup in order to get the opportunity to measure said headphones. This is going up from the HiFiMan Arya Stealth and Meze Elite Tungsten. I also want to investigate the measurable correlates of subjective impressions of sharp transients and their decays (e.g. whether CSD artifacts are audible at least in isolated single-sample impulses). E.g. Why I find the Arya's single-sample transients "amazingly" and "aggressively" sharp compared to the other high-end headphones I auditioned even when EQed relatively similarly and turning up the other headphones louder (the difference between a sparky "pop" versus a hollow stretched plastic tap); one answer could be remaining phase response differences and hence CSD differences, this possibly impossible to match between headphones. I have found that it is indeed possible to EQ headphones to sound extremely similar tonally, but do vie that there are pad size/comfort and driver related factors (like the aforementioned transient qualities; I unfortunately haven't read into any literature regarding this yet) that for me can cause different presentations of that same tonality to feel more preferable or enjoyable, and that is enough to make an estat worth it to me. I know that I can hear amazing things through my EQed Jabra Elite 85h Bluetooth headphones, but also know that large planar presentation with nothing touching my ears is a wonderful experience.

As for sating my current "dissatisfactions" even with the nicest classical recordings I've found, my main solution is rather toward acquiring the Earfish in-ear microphone and binaural headtracking kit and EQing my headphones toward an actual measurably neutral speaker response, this being the next step above my current use of bs2b crossfeed and having fine-tuned a rough Harman or diffuse-field EQ while eliminating peaks. But I'd like the peace of mind that my headphones are actually behaving when blasting a Mahler 5 tutti at 95 to 100 dBA (I swear it was even louder at the concert hall).

So back to the question of what it takes to measure these kinds of distortions or measure accurate CSD plots.

 

[solderdude]

So back to the question of what it takes to measure these kinds of distortions or measure accurate CSD plots.

For one thing it requires not using fake ears and ear canals. You would want to use a setup that can be used to measure driver properties. The problem with industry standard fixtures is that they only conform to a standard. That standard describes how the ear canal + mic should be build. The idea is that such are build resembling an average ear.
An average ear does not exist. There is tons of research showing up to 10dB differences above several kHz between ears so an average ear is going to react in a certain way which depends on many practical circumstances. For this reason most measurements are 'averaged' or one that is thought to be representative is shown.

It is a good thing to have a standard so one can compare measurements but that does not mean they tell the absolute truth nor that your ears comply to that standard.

When you want a CSD to say 'something' you would have to accurately compensate for the changes that are made by the fake ear and ear-canal. As that correction is an 'average' and measurements are 'average' you will not end up with accurate.

Also you would have to ask yourself what you can learn from a CSD. I mean some really good sounding headphones have poor CSD and some have good CSD.
What CSD can be used for is finding some severe cases that point towards excessive ringing. Driver/coupler 'issues' can be made visible.
This you can do with a cheap mic. No 0.001% needed.
As for the audibility of CSD shown issues... Consider that music instruments all have their own decay which, in general, takes a LOT longer than those found in headphones and speakers. Also this is at a very narrow band and harmonics of that instrument would have to be very close to those frequency(cys) and be shorter in decay than the headphone.

So the value of it is limited, so is group delay and phase. Phase differences between L and R may even be more important (in the middle part of the FR range) than some variations.

Maybe looking at bursts (in real time using a scope, not audio card) can say more about performance in some aspects than simple sweeps.

Besides when measuring distortion in electrostats you would also have to take the energizers used in consideration which may not reach the distortion levels the actual headphones can reach (so basically are measuring the energizer).

Not everything is always measured.
Not everything that is measured matters.
Not everything that is measured is measured properly (knowingly or not).
Audibility thresholds vary.
Comfort matters, fit matters, seal matters, conditions matter.

Headphone measurements can be indicative at best. Even when using the latest or greatest standards it still only says something about a measurement to that standard and the choices made by the engineer about what to 'show' and how they are made and why they are made.

Distortion measurements in the acoustical domain are quite limited. You could increase resolution by averaging and thus lower the influence of noise.
Distortion of the mic (+ pre-amps) can skew results even then. One might not be sure what part is the mic and what is the driver.
Even then you only measure distortion of a steady state frequency over a certain time. That may well be a different thing as real life performance with dynamic signals. So such measurements might not tell the whole story of 'distortion' as that would only say something about a certain aspect of a particular type of distortion within its limits.

Moving measurement thresholds usually requires a lot of money and lab conditions.

 

[Mr. Haelscheir]

For one thing it requires not using fake ears and ear canals. You would want to use a setup that can be used to measure driver properties. The problem with industry standard fixtures is that they only conform to a standard. That standard describes how the ear canal + mic should be build. The idea is that such are build resembling an average ear.
An average ear does not exist. There is tons of research showing up to 10dB differences above several kHz between ears so an average ear is going to react in a certain way which depends on many practical circumstances. For this reason most measurements are 'averaged' or one that is thought to be representative is shown.

It is a good thing to have a standard so one can compare measurements but that does not mean they tell the absolute truth nor that your ears comply to that standard.

When you want a CSD to say 'something' you would have to accurately compensate for the changes that are made by the fake ear and ear-canal. As that correction is an 'average' and measurements are 'average' you will not end up with accurate.

Also you would have to ask yourself what you can learn from a CSD. I mean some really good sounding headphones have poor CSD and some have good CSD.
What CSD can be used for is finding some severe cases that point towards excessive ringing. Driver/coupler 'issues' can be made visible.
This you can do with a cheap mic. No 0.001% needed.
As for the audibility of CSD shown issues... Consider that music instruments all have their own decay which, in general, takes a LOT longer than those found in headphones and speakers. Also this is at a very narrow band and harmonics of that instrument would have to be very close to those frequency(cys) and be shorter in decay than the headphone.

So the value of it is limited, so is group delay and phase. Phase differences between L and R may even be more important (in the middle part of the FR range) than some variations.

Maybe looking at bursts (in real time using a scope, not audio card) can say more about performance in some aspects than simple sweeps.

Besides when measuring distortion in electrostats you would also have to take the energizers used in consideration which may not reach the distortion levels the actual headphones can reach (so basically are measuring the energizer).

Not everything is always measured.
Not everything that is measured matters.
Not everything that is measured is measured properly (knowingly or not).
Audibility thresholds vary.
Comfort matters, fit matters, seal matters, conditions matter.

Headphone measurements can be indicative at best. Even when using the latest or greatest standards it still only says something about a measurement to that standard and the choices made by the engineer about what to 'show' and how they are made and why they are made.

Distortion measurements in the acoustical domain are quite limited. You could increase resolution by averaging and thus lower the influence of noise.
Distortion of the mic (+ pre-amps) can skew results even then. One might not be sure what part is the mic and what is the driver.
Even then you only measure distortion of a steady state frequency over a certain time. That may well be a different thing as real life performance with dynamic signals. So such measurements might not tell the whole story of 'distortion' as that would only say something about a certain aspect of a particular type of distortion within its limits.

Moving measurement thresholds usually requires a lot of money and lab conditions.

It does seem sensible to devise a rig that focuses on assessing driver performance while eliminating the effects of a simulated ear or head (i.e. flat plate coupler). Indeed, my intents for this measurement rig are not for use in EQ profile generation as in https://www.audiosciencereview.com/...lding-a-new-flat-plate-measurement-rig.15066/: I intend on using in-ear mics with a real calibrated speaker reference for that. Now, while head simulation and acoustic impedance are most important for "accurate" at-eardrum frequency response predictions, I do not know if they have a significant contribution to distortion or CSD measurements. I suppose we wouldn't expect the physical test rig itself or the in-ear cup reflections to produce distortion components of their own?

I do wish to also conduct a few "dynamic" tests possibly correlated with subjective findings. One test I would like to try is to normalize a few headphones to an identical pink noise or sine sweep magnitude response and volume level, then measure whether the same single-sample Dirac delta impulse input causes some headphones to produce higher or sharper SPL peaks than others, say, due to driver sensitivity or diaphragm lightness. This would have to assume sufficient independence from amp performance. I might be able to compare the Topping EHA5 and Mjolnir Audio Carbon CC in this regard.

Thus, the main points of discussion are:

1. Flat plate test rig materials and damping. Budget: $100
2. How to shop for low-noise, low-distortion microphone capsules known to not contribute their own distortion above the noise floor when measuring at around 100 dB SPL. Target: Be able to measure down to 0.01% THD at 100 dB SPL. Budget: Up to $1,500 for the capsules, required preamps or signal conditioners, and the audio interface.
3. Acoustic isolation enclosure with internal damping. Target: Internal noise floor below 20 dB when closed within a room with a 40 dBA ambient noise. Internal dimensions: Around 1.5' by 1.5' by 1.5'. It must be portable. Budget: $500 if it really comes to that.
   • Are there benefits to using larger enclosures when testing headphone distortion? E.g. Is there a benefit to having allowance for windowing against the remaining internal reflections of the enclosure?

 

[solderdude]

1: the main problem here is distortion of the mic. Most electret mics do show higher 2nd harmonic distortion due to their construction.
You will probably need a very good measurement mic. The ones used for FR calibration won't do as distortion is not important for those measurements.
2: I use a cheap capsule and just know that 2nd harmonics (at 100dB SPL) won't be better than 0.3-0.5%.
3: You will need to make a sound proof box if you want to measure distortion down low. It will not attenuate low frequencies that well though.
Don't make the box too small and use diffraction/absorbtion at the sides of the headphone. estats pass a lot of sounds and you don't want to measure reflections of the emitted sound.

Personally I do not bother as I don't care about the small numbers as my fixture simply cannot measure low enough. It is more than good enough to detect audible issues so enough for me.

It is interesting to do some burst tests for instance or 'dynamic' tests at various frequencies but for that you do not need low distortion. Just a mic that can reach >20kHz.

 

[Mr. Haelscheir]

Is there a general term for the kinds of non-electret low-distortion microphones I should be looking for? I think I've read of piezoelectric implementations. Worst case, I would be looking at supposed coupler clones, whether or not those are even suitable for distortion measurements.

Some impressive things are said about B&K microphones here and on, whether or not it is to be believed: https://audiosciencereview.com/foru...than-speakers-yes-no-maybe.33747/post-1178780

Perhaps clone couplers are the way to go: https://www.audiosciencereview.com/...c-60318-4-clone-coupler-is-it-worth-it.23865/

I don't know if https://www.aliexpress.com/item/1005005948705868.html can be trusted. Alternatively, https://www.aliexpress.com/item/1005005282002642.html plus the couplers recommended in that thread.

 

[amirm]

Not that it helps but I have measured a few headphones to below 0.01% using my Gras fixture and Audio Precision analyze without any chamber. So it can br done.

 

[Mr. Haelscheir]

Not that it helps but I have measured a few headphones to below 0.01% using my Gras fixture and Audio Precision analyze without any chamber. So it can br done.

Interesting that you could attain that SNR without an acoustic test chamber. As for "feasibility", I guess this thread rather regards the feasibility of attaining such measurements on a budget without having to spend for a GRAS 45CC, or on how to identify components that are up to the task. At least the measurements in https://audiosciencereview.com/foru...than-speakers-yes-no-maybe.33747/post-1178780 seem to confirm that the UMIK-1 capsules in the miniDSP EARS will most likely not be able to measure down to 0.01% due to their own distortion.

Given that, I am considering asking the seller of https://www.aliexpress.com/item/1005005948705868.html for details about this product. Now, there are two other shops on AliExpress that seem to be selling the same device with minimal details disclosed.

Otherwise, if anyone knows of any other cost-effective microphones or couplers that could be used for this goal.

 

[Blumlein 88]

Interesting that you could attain that SNR without an acoustic test chamber. As for "feasibility", I guess this thread rather regards the feasibility of attaining such measurements on a budget without having to spend for a GRAS 45CC, or on how to identify components that are up to the task. At least the measurements in https://audiosciencereview.com/foru...than-speakers-yes-no-maybe.33747/post-1178780 seem to confirm that the UMIK-1 capsules in the miniDSP EARS will most likely not be able to measure down to 0.01% due to their own distortion.

Given that, I am considering asking the seller of https://www.aliexpress.com/item/1005005948705868.html for details about this product. Now, there are two other shops on AliExpress that seem to be selling the same device with minimal details disclosed.

Otherwise, if anyone knows of any other cost-effective microphones or couplers that could be used for this goal.

It isn't likely capsule distortion in the umik rather in the electronic section.

 

[Mr. Haelscheir]

It isn't likely capsule distortion in the umik rather in the electronic section.

That would imply one's needing to mod the miniDSP EARS to squeeze more performance out of the existing mics?

 

[Mr. Haelscheir]

If only I had known that https://www.audiosciencereview.com/...sr-009s-electrostatic-headphone-review.47390/ had been posted on the morning that I started this thread. I suppose that yields promising expectations for the X9000 and might save me on measurement gear, though I would still like to test the Warwick Acoustic Bravuras and investigate my theories regarding transient quality.

 

[solderdude]

For the transients (burst or squarewave) you can use the cheapest electret mics out there that reach >20kHz.

I would assume that the Bravuras are at least as good or better sounding than model one.
The only downsides model one had were max SPL (because of the built-in EQ and limitations of energizers) and the price. The Model one sounded excellent.
Why the need to investigate when it is all about enjoyment and tonality ?

 

[Mr. Haelscheir]

Why the need to investigate when it is all about enjoyment and tonality ?

Curiosity with the value proposition of the full Bravura + Sonoma M1 DAC/amp set, especially with its currently being on sale locally, and considering the relative cost of just the headphone versus the DAC/amp. The Bravura set uses the same (if not slightly improved/updated) Sonoma M1 DAC/amp unit as its predecessor, does it not? But yes, maybe one of these headphones will yield a preferred comfort style or presentation.

So far, that SBAF thread on tone bursts is proving to be quite interesting. How FR manifests as differences in tone burst envelope. I remember being skeptical when someone claimed that estats or some headphones not as damped as the DCA Expanse were "exaggerating" the transients, whereby those tone burst measurements seem to be showing real overshoot which could be part of why I am perceiving the Arya's transients as "sharper". Now, I suppose yes, a perfect measured tone burst envelope would imply a perfectly flat transfer function, whereby the human ear's transfer function is bound to modify that envelope anyways. Then there is the question of the relevance of "phase accuracy" in the goal of EQing or applying binaural headtracking to simulate "perfect" free-field speakers, and how that affects transient envelopes.

Thinking upon this brings me back to that whole "frequency response [magnitude and phase] is all that matters" stance, though I guess these kinds of measurements still help with deriving possible subjective transient characteristics from said FR. Perhaps they could reveal phase characteristics that remain in the tone burst envelope even after EQing the headphones to the same magnitude response. I suspect that these time-domain tests also will not reveal anything new (if anything at all) regarding nonlinear distortion performance, unless the same amplitude of tone were to somehow incur more peak limiting while within a burst.

Anyways, I would like to use the same couplers for distortion, multitone (I'm not sure of the benefits of separate two-tone or so IMD measurements), and transient/time-domain measurements.

 

[solderdude]

I don't think there are much phase issues with single driver headphones.
The membrane moves as a whole so there can not be much phase issues, only very small ones near sharp resonances but think the FR aspect of such are much more audible than the possible phase issues.

The Sonoma energizer (it is more than just an amp), just like the HE1 energizer, also has EQ built into it to elevate bass and EQ the headphone itself. As it is only intended for the accompanying headphone this is easy to do.

The bass may sound 'thinner' on estats than on dynamics. The membrane can swing in very quickly for mids and treble but seems to need some time to reach the expected/desired amplitude for lower frequencies. So bass (impulses) may be relatively lower in amplitude than the mids/treble with music (impulses) and thus sound less 'full bodied' compared to dynamics.

That said... the Model1 sounded fine to me and did not lack in bass. Admittedly I am not a Harman speaker alike bass level guy and do not prefer 'flat speaker in a room boosted' bass but prefer nearfield monitor type of lows (sounds tighter to me).

Multitone turns into an unreadable mess in acoustic measurements. I see no value in that other than in electronics.

If you want to go down to the nitty gritty in measurements you should invest heavily. Not in the Chinese clones or EARS.