- Joined
- Jul 13, 2018
- Messages
- 23
- Likes
- 16
In another thread, the off-topic topic of headphone measurements came up, where I declared that I think the standard way is problematic. Someone agreed and someone else suggested I start a thread about how I'd do it differently.
In my work, I have reason to do such measurements. I've had a pretty good think about them and done some tinkering on the bench. I started out accepting the standard way, with a microphone at the bottom of an artificial ear canal in an artificial head. I initially reasoned that, since this appears to be a widely accepted standard, at least the results should be meaningful and comparable.
Due to the prominent measurement artifacts and high data variability caused by sensitivity to headphone placement on the head, I soon decided that it was impossible to get a clear idea of what the differences might be in headphone sound based on frequency response data acquired this way, except for a general idea of broad characteristics. Besides, this approach doesn’t make physical sense.
It seems to me that placing a microphone at the bottom of a tube that's supposed to represent the ear canal and surrounding the exit of the tube with an artificial ear is a wrongheaded convention, so to speak.
The reason the tube makes no sense is very simply that the part of the ear that's struck by sound is on the outside. Whatever wiggles are caused by ear canal resonances are disregarded by the brain, so there’s no reason to [attempt to] record them.
The reason an artificial ear makes no sense is that the pinnae are effectively rendered non-functional with headphones -- the in-your-head sound is proof of that. The pinnae are doing something, in that they are causing wiggles in the top octaves of the frequency response, but since those wiggles are not changing dynamically, they are not providing any dynamic sound location information. The brain thus disregards them.
Lastly, recording the wiggles in the upper frequency response caused by cavity resonances inside an over-the-ear headphone makes no sense because those are rejected like the static pinnae effects.
Furthermore, those wiggles are strongly dependent on how the headphones are positioned on the artificial head. In other words, they’re dependent on the measurement apparatus, so your results on your own personal or artificial head will vary, as they will vary between labs.
When more than one lab publishes data acquired this way on the same headphones, they risk provoking kerfuffles, but I don’t care about that. I care about people having access to easily interpreted data and not only that, but also being able to generate data for themselves that’s reasonably comparable from apparatus to apparatus.
I’ve not developed a finished apparatus yet.
What I've been doing so far is merely determining what a credible apparatus and measurement scheme would be, including that the measurements should be immune to the effects of reasonably sloppy placement of the headphones, so anyone can repeat it at low cost and get comparable results.
Broad strokes, the response from, say, 500 Hz up is measured with the can in open air and the mic a short distance from it, perhaps at about the same distance as an ear would be. This is pretty similar to a free field measurement, e.g., nearly anechoic, not be confused with a diffuse field measurement, e.g., with sound coming from all directions equally, which some labs attempt to simulate by equalizing the standard data.
With large area transducers, there will probably be interference artifacts, because the distance to the diaphragm will vary from the center to the edge. Perhaps simply using a larger measurement distance or measuring from some point other than on axis will keep these minimal. On the other hand, at a greater distance, there may be lobing and thus problematic sensitivity to position relative to the transducer central axis. If such problems are insurmountable by way of microphone positioning without equalization, I’ll come up with a way of compensating them based on the transducer diameter and microphone placement.
Below 500 Hz, the can is sealed to a fixture that has a mic located at about the usual ear distance. The two plots would be combined. If it turns out that two different microphone distances work better, then the levels will be adjusted to make their transition coincide.
The particulars of seating the headphones will depend on whether the headphones are over-ear or on-ear. For on-ear, perhaps an artificial ear surrounding the microphone makes sense for creating a typical amount of bass leakage. In that case, a silicone ear that can be bought cheaply through Amazon or Alibaba will be recommended.
I suppose to get the highest cutoff and simulate the ear canal aperture size, a 1/8" microphone is most suitable to the application in a strict sense. On the other hand, these small capsule microphones are very expensive and generally need better electronics because of their small signal strength, so most people aren't going to want to buy one.
To stick to the goal of making this an Everyman's apparatus, I say an inexpensive 1/2" is the way to go, with an expectation that the measurements may be off by broad dips or rises of a dB or two here and there. It also means a response that will not go much over 20 kHz, so an inexpensive or built-in sound card can be used that can only do 48/16, although most will do 196/24 these days. FWIW, a usable 1/2" mic from Dayton (Parts Express) is $60. An Audix TM1 1/4" [6 mm] from Amazon is $200 on sale ($325 list), and will get you to 25 kHz. A 1/4" [6 mm] Earthworks is $700 and will get you to 30 kHz.
Solderdude and mitchco both mentioned within hours of my original posting that condenser microphones tend to have a lot of 2nd harmonic. I'll have to find an affordable microphone that, when compared to my B&K, has very little. If I don't find one, then I can characterize a decent microphone's distortion and this can be subtracted from the measurement data for a particular SPL or small set of SPL's using a spreadsheet or a little software routine.
A 1/8" [3 mm] mic, however, is a specialty item. The GRAS 46DE is an industry standard, and will measure 6 Hz - 70 kHz, but is listed as "call for price" everywhere. I can guess GRAS charges a lot more for this than for its larger mics, which are in the $1600 - $2400 range. To make the most of spending that much money, you'd also be in for the cost of a calibrator or periodic factory calibrations. I think the thing would be quite a responsibility for most people, including me. Anyway, I’d like to keep the basic parts and software list under $300.
FWIW, I've got a Bruel & Kjaer with all the fixin’s that I rarely take out of its case. I just know Murphy dictates that there's a finite number of times I can set it up before it becomes the first mic I've ever broken. For daily work, I've got two identical, reasonably inexpensive 1/2" mics, one of which I've dropped a few times. Each time it gets whacked, I compare it to the other one to check for damage. If I ever do ruin it, it won't be painful to replace.
Back to the original topic. The measurements will include of course the frequency response, which most people can run, have familiarity with, and can use to help them guess what a headset might sound like. For my part, plots with pink noise and fast sweep excitation will both be run and compared, although it may turn out that they are so similar for headphones that one will suffice for publication. I'd also run a distortion plot for the first few harmonics and a waterfall representation of resonant behavior, plus an electrical impedance plot, of course.
Shortly after I posted this, Amir mentioned that power handling is also a nice thing to know about, but rarely reported. An interesting factor is that to attain a given SPL, low sensitivity headphones are driven with more power than those with high sensitivity. I think finding the SPL where a headset falls apart would combine both factors and be an intuitively understandable metric.
To avoid disagreement over what constitutes falling apart, we could agree on a maximum increase in THD and a maximum deviation in frequency response from what they are at moderate levels. An example might be a 3 dB maximum deviation anywhere in the frequency response or 3% THD. We could call it the 3 x 3 sound disintegration proxy.
A problem when testing some headphones this way is that driver damage could occur when dwelling above a power level that still does not exceed the 3 x 3 criteria. A planar magnetic's diaphragm might melt, a dynamic's coil open up, or an electrostatic arc, even when a fast sweep stilmulus is being used, and I don't wish to perform destructive testing on anything but my fellow man.
For this reason, it may be good to set some maximum SPL that is considered worth achieving without falling apart, like 110 dB, and above which the issue of the sound falling apart becomes intertwined with the issue of causing deafness. In cases where a 3 x 3 was never hit, there could be an entry in the results table that means Good at 110 dB.
Put it all together, and I think the data will give a fair clue about the sound, although I don’t expect it to beat taking a listen for yourself or the word of a trusted advisor.
When I’ve got something done, which might not be until Autumn, I’ll post a complete description and maybe a video of me using it. If it doesn’t go up here, it will at least be in the company newsletter.
In my work, I have reason to do such measurements. I've had a pretty good think about them and done some tinkering on the bench. I started out accepting the standard way, with a microphone at the bottom of an artificial ear canal in an artificial head. I initially reasoned that, since this appears to be a widely accepted standard, at least the results should be meaningful and comparable.
Due to the prominent measurement artifacts and high data variability caused by sensitivity to headphone placement on the head, I soon decided that it was impossible to get a clear idea of what the differences might be in headphone sound based on frequency response data acquired this way, except for a general idea of broad characteristics. Besides, this approach doesn’t make physical sense.
It seems to me that placing a microphone at the bottom of a tube that's supposed to represent the ear canal and surrounding the exit of the tube with an artificial ear is a wrongheaded convention, so to speak.
The reason the tube makes no sense is very simply that the part of the ear that's struck by sound is on the outside. Whatever wiggles are caused by ear canal resonances are disregarded by the brain, so there’s no reason to [attempt to] record them.
The reason an artificial ear makes no sense is that the pinnae are effectively rendered non-functional with headphones -- the in-your-head sound is proof of that. The pinnae are doing something, in that they are causing wiggles in the top octaves of the frequency response, but since those wiggles are not changing dynamically, they are not providing any dynamic sound location information. The brain thus disregards them.
Lastly, recording the wiggles in the upper frequency response caused by cavity resonances inside an over-the-ear headphone makes no sense because those are rejected like the static pinnae effects.
Furthermore, those wiggles are strongly dependent on how the headphones are positioned on the artificial head. In other words, they’re dependent on the measurement apparatus, so your results on your own personal or artificial head will vary, as they will vary between labs.
When more than one lab publishes data acquired this way on the same headphones, they risk provoking kerfuffles, but I don’t care about that. I care about people having access to easily interpreted data and not only that, but also being able to generate data for themselves that’s reasonably comparable from apparatus to apparatus.
I’ve not developed a finished apparatus yet.
What I've been doing so far is merely determining what a credible apparatus and measurement scheme would be, including that the measurements should be immune to the effects of reasonably sloppy placement of the headphones, so anyone can repeat it at low cost and get comparable results.
Broad strokes, the response from, say, 500 Hz up is measured with the can in open air and the mic a short distance from it, perhaps at about the same distance as an ear would be. This is pretty similar to a free field measurement, e.g., nearly anechoic, not be confused with a diffuse field measurement, e.g., with sound coming from all directions equally, which some labs attempt to simulate by equalizing the standard data.
With large area transducers, there will probably be interference artifacts, because the distance to the diaphragm will vary from the center to the edge. Perhaps simply using a larger measurement distance or measuring from some point other than on axis will keep these minimal. On the other hand, at a greater distance, there may be lobing and thus problematic sensitivity to position relative to the transducer central axis. If such problems are insurmountable by way of microphone positioning without equalization, I’ll come up with a way of compensating them based on the transducer diameter and microphone placement.
Below 500 Hz, the can is sealed to a fixture that has a mic located at about the usual ear distance. The two plots would be combined. If it turns out that two different microphone distances work better, then the levels will be adjusted to make their transition coincide.
The particulars of seating the headphones will depend on whether the headphones are over-ear or on-ear. For on-ear, perhaps an artificial ear surrounding the microphone makes sense for creating a typical amount of bass leakage. In that case, a silicone ear that can be bought cheaply through Amazon or Alibaba will be recommended.
I suppose to get the highest cutoff and simulate the ear canal aperture size, a 1/8" microphone is most suitable to the application in a strict sense. On the other hand, these small capsule microphones are very expensive and generally need better electronics because of their small signal strength, so most people aren't going to want to buy one.
To stick to the goal of making this an Everyman's apparatus, I say an inexpensive 1/2" is the way to go, with an expectation that the measurements may be off by broad dips or rises of a dB or two here and there. It also means a response that will not go much over 20 kHz, so an inexpensive or built-in sound card can be used that can only do 48/16, although most will do 196/24 these days. FWIW, a usable 1/2" mic from Dayton (Parts Express) is $60. An Audix TM1 1/4" [6 mm] from Amazon is $200 on sale ($325 list), and will get you to 25 kHz. A 1/4" [6 mm] Earthworks is $700 and will get you to 30 kHz.
Solderdude and mitchco both mentioned within hours of my original posting that condenser microphones tend to have a lot of 2nd harmonic. I'll have to find an affordable microphone that, when compared to my B&K, has very little. If I don't find one, then I can characterize a decent microphone's distortion and this can be subtracted from the measurement data for a particular SPL or small set of SPL's using a spreadsheet or a little software routine.
A 1/8" [3 mm] mic, however, is a specialty item. The GRAS 46DE is an industry standard, and will measure 6 Hz - 70 kHz, but is listed as "call for price" everywhere. I can guess GRAS charges a lot more for this than for its larger mics, which are in the $1600 - $2400 range. To make the most of spending that much money, you'd also be in for the cost of a calibrator or periodic factory calibrations. I think the thing would be quite a responsibility for most people, including me. Anyway, I’d like to keep the basic parts and software list under $300.
FWIW, I've got a Bruel & Kjaer with all the fixin’s that I rarely take out of its case. I just know Murphy dictates that there's a finite number of times I can set it up before it becomes the first mic I've ever broken. For daily work, I've got two identical, reasonably inexpensive 1/2" mics, one of which I've dropped a few times. Each time it gets whacked, I compare it to the other one to check for damage. If I ever do ruin it, it won't be painful to replace.
Back to the original topic. The measurements will include of course the frequency response, which most people can run, have familiarity with, and can use to help them guess what a headset might sound like. For my part, plots with pink noise and fast sweep excitation will both be run and compared, although it may turn out that they are so similar for headphones that one will suffice for publication. I'd also run a distortion plot for the first few harmonics and a waterfall representation of resonant behavior, plus an electrical impedance plot, of course.
Shortly after I posted this, Amir mentioned that power handling is also a nice thing to know about, but rarely reported. An interesting factor is that to attain a given SPL, low sensitivity headphones are driven with more power than those with high sensitivity. I think finding the SPL where a headset falls apart would combine both factors and be an intuitively understandable metric.
To avoid disagreement over what constitutes falling apart, we could agree on a maximum increase in THD and a maximum deviation in frequency response from what they are at moderate levels. An example might be a 3 dB maximum deviation anywhere in the frequency response or 3% THD. We could call it the 3 x 3 sound disintegration proxy.
A problem when testing some headphones this way is that driver damage could occur when dwelling above a power level that still does not exceed the 3 x 3 criteria. A planar magnetic's diaphragm might melt, a dynamic's coil open up, or an electrostatic arc, even when a fast sweep stilmulus is being used, and I don't wish to perform destructive testing on anything but my fellow man.
For this reason, it may be good to set some maximum SPL that is considered worth achieving without falling apart, like 110 dB, and above which the issue of the sound falling apart becomes intertwined with the issue of causing deafness. In cases where a 3 x 3 was never hit, there could be an entry in the results table that means Good at 110 dB.
Put it all together, and I think the data will give a fair clue about the sound, although I don’t expect it to beat taking a listen for yourself or the word of a trusted advisor.
When I’ve got something done, which might not be until Autumn, I’ll post a complete description and maybe a video of me using it. If it doesn’t go up here, it will at least be in the company newsletter.
Last edited: