Headphone Measurement Proceedures - Introduction and Equipment

Headphone Objective Measurements Overview
There are two camps in the world of audio: objectivists and subjectivists. In an oversimplified summary, objectivists believe audio performance is measurable; subjectivists believe audio performance can only be evaluated as an experience. There are, of course, those with a more sophisticated and balanced perspective who believe both have merit; I fall into that camp. I can look at a set of measurements and can tell you roughly what to expect when you put the cans on, but there’s no way to really tell what the experience is going to be like until you listen. None the less, having a good database of measurements will let you know if one headphone has more bass than another; will allow you to rapidly find headphones of similar characteristics; and will let you rule out poorly performing headphones without having to listen.

Unlike speakers, which are more complicated with multiple drivers and crossovers, and get put in rooms with reflections from walls, headphones are usually a single driver directly coupled to your ears. As a result, my experience leads me to believe the relationship is closer between headphone measurements and what you hear, than speaker measurements and what you hear. Similarly, audio electronics gear these day is quite good, and measures of their performance is rather more like splitting hairs --- the difference between 0.001% and 0.0001% distortion is not easily audible. The imperfections of headphones are orders of magnitude larger than electronics gear, so, while I believe it is important to measure electronic gear, I also feel it tends to be less informative regarding the experienced sound quality than headphone measurements.

What I’m trying to get at here is that I’ve found the correlation between headphone measurements and the listening experience quite a bit stronger than with speaker or electronics measurements. Though this may simply be a result of my intimate familiarity with headphone measurements. At any rate, I think you’ll find this journey into objective headphone evaluation interesting and illuminating.

Headphone Measurement Equipment Overview

The Head Acoustics HMS II.3 Head Simulator
The heart of a headphone measurement system is a microphone in the shape of a head. I use a Head Acoustics HMSII.3 head acoustics simulator. At around $20,000, this noggin ain’t cheap, but it’s got a lot of very specialized characteristics and careful engineering that make it a very important tool.


The Head Acoustics HMSII.3 head acoustics simulator.

The head is designed to act exactly like the average human listening system. Inside the head at very much the same position of your eardrums at the end of your ear canal, are two very special microphones (one in each ear) that mimic the exact acoustic impedance characteristics of your eardrum. The shape of the ears is also very precisely defined (IEC specification 60268-7) so as to act just like the average human ear. The goal of this device is to provide a measurement of exactly what is heard at the eardrum for sound arriving at the head. Because headphones are coupled directly to the head, it is only by using a device such as this that headphone measurements can be made that relate strongly to what we hear. There are other headphone measurement couplers, which are not head shaped, that can achieve similar results, but the head allows me to measure both ears at the same time, and allows me to measure the amount of isolation from outside noise accurately.

The Test Chamber

In order to achieve repeatable measurements, I need to have a way to isolate the head from random outside noise that might appear in the measurements. Also, because one of the measurements needed is the amount of isolation from outside noise, I need an environment in which I can repeatably create a controlled amount of noise. To do this I’ve built a custom chamber in which to place the head during tests. It’s about 4’x2’x2’ and is constructed of 1” furniture grade plywood with a layer of drywall very thoroughly glued to the inside walls. The very heavy and stiff walls provide significant acoustic resistance; I currently get about 15dB of isolation from outside noise. This chamber is mounted on air bladders (wheel barrow innertubes) in order to isolate it from mechanically born noise from the rest of my home. Cables are routed through a hole in the chamber that is filled in with modeling clay.


My home-brewed acoustic isolation chamber for testing headphones.

The chamber has a speaker in it that is used to generate pink noise for measuring isolation. The interior of the chamber has some noise damping foam, but also has reflective surfaces designed to evenly distribute noise around the chamber interior so that the sound approaches the headphones from many directions when measuring isolation.

The Audio Precision System 2 Cascade


The Audio Precision System Two Cascade, model SYS2522 dual-domain audio tester.

Once we have a way to accurately simulate the acoustics of headphone listening, we need a piece of gear that is able to generate and analyze the audio signals needed to tease out meaningful measurements. I use an Audio Precision System 2 Cascade dual-domain audio tester. This piece of gear can generate and analyze both analog and digital audio signals with state-of-the-art precision and can be programmed to automatically run the device under test through a broad battery of standard and customized tests. I think if the AP had a soul, I’d marry it. Unfortunately it doesn’t, so I settle for a healthy dose of gear lust whenever I get to play with it.

The Computer
A fairly old PC controls the Audio Precision (AP); nothing special here. The PC is loaded with a software application that controls the AP, and receives measured data, which it inserts into preconfigured Excel spreadsheet templates. The AP software has control panels, which can be configured into the various tests (frequency response, %THD+noise, impulse response, etc), and then called by a script written in an AP extended basic programming language. It is possible to “drive” the AP manually from these pre-configure panel set-ups to do custom experiments.

Other Gear


The Tektronics model TDS 210 digital oscilloscope used to monitor signals while positioning headphones.

A small Tektronix TDS 210 oscilloscope is mounted above the door of the chamber and is fed left and right channel signals from the monitor output of the AP. I use this o’scope to monitor the signal from the head in order to properly position the headphones on the head prior to test. Positioning of the headphones is critical, and the o’scope provides important feedback on whether or not the headphones are positioned and sealing properly.

The microphones in the head have phantom power voltage and pre-amplification provided by a G.R.A.S. Type 12AA low-noise pre-amplifier mounted next to the chamber. The speaker in the chamber is driven by an NHT C20 power amplifier.

Continue on with Measuring Headphone Frequency Response.

Share | |
COMMENTS
LFF's picture

You posted this article before I even had a chance to ask you to post something like this! Thanks Tyll!

xnor's picture

"Measuring headphones is a long and involved process which includes elaborate costuming, waterboarding cats, and just the right wand.
:)

No ... not really."

awwwwww :(

Nice article though and love the vid(s), your enthusiasm is contagious. :D

Tyll Hertsens's picture
I think both of you will like the next post: the actual measuring process. But for different reasons. Be sure to watch the next video, LFF. :)
outerspace's picture

It would be better if you show %THD+noise in LINEAR scale instead of logarithmic. :)

Or just add additional counts to THD scale and additional horizontal lines.

13mh13's picture

It's about time someone posted such long-neglected (= useful) info. I hope these "instructions" start a much-anticipated trend in head-gear reviewing: toward more a more balanced style (= objective and tradit. subjective criteria).

Not a freebie by any means ... I see you custom-built your test chamber. That's time/$ investment. Also ... instruments cost $$ ... learning to use them is not straightfwd (and an evolving science in itself) ... and then interpreting the data, number-crunching, graphing, benchmarking, publishing ... all pioneering efforts in many ways.

Again great work creating this page ... and using science to improve our favorite transducer type!

ultrabike's picture

Hi Tyll,

Why do you use Pink noise for isolation tests? I understand that Pink noise has a 1/f power characteristic (not flat as white noise), and that people like it a bit more. May be useful for listening test as you pointed out your other articles.

However, for characterization of LTI systems (or close to), typically white noise (usually a high order MLS sequence which would also yield your impulse response after self correlation) should be used...? Isn't the isolation channel an LTI channel with its way cool impulse response as well? Do you calibrate your results to account for HH* = 1/f then? Wazup?

Tyll Hertsens's picture
In audio, pink noise is heard as flat---equal energy per octave.

For the isolation test, it really doesn't matter much what signal I use (as long as it contains all frequencies) as I subtract the response heard by the head without headphones on. So it's a relative measure of the difference heard by the head with and without the headphones on.

Didn't quite understand your second paragraph.

ultrabike's picture

I agree. As long as the mic has sufficient sensitivity to properly capture the response of all the intended frequencies, including the least exited one (i.e. the highest frequency), the implied calibration should take care of the measurement. Since you seem to have fairly high end gear, my best guess is that you are golden.

Sorry for my second paragraph ramblings. It really means that I think you should be able to use white noise like waveforms to characterize isolation and may benefit from the extra bits of resolution in the higher frequencies. Furthermore, if properly calibrated white noise is used (mics show white noise with out the HP), you could just put the headphones on the head dummy and call the result the isolation response with out having to subtract the result of the headphones off, and maybe improve your measurement accuracy... that's all. I don't know, I don't think its a big deal.

ultrabike's picture

Besides the notion that we hear pink noise as flat, you are PROBABLY using pink noise because your measurements are x-axis log scale. In general, when doing a linear scale sweep white noise is used, if doing a log scale sweep pink noise is used (probably has to do with resolution bandwidths of the measurements). But I really don't know what your equipment does, and the best I can do with out clarification is jump into conclusions, unless I start asking for manuals, which IMO is going too far. The answer could just as well, be: NOPE, it really is just that we use pink noise in measurements because it is heard flat... so what do I know.

I may not have posed my questions correctly, and may have annoyingly jumped into conclusions. For that I apologize. I'll try to stay clear for now. It was obviously very exciting learning a little more about the audio field from your articles, and may do some research on my own as a hobby. I have much more experience in digital communications (and a few revenue making designs of my own), and can relate to some of the things you are describing, but this is not really my field, and probably approached this site with a little too much enthusiasm. :) I guess I kind of came and went out of nowhere, but then again... who doesn't?

Tyll Hertsens's picture
... I've got no problems with your questions in the least. The AP tester allows me to use whatever signal I'd like, so I could have used white noise as easily as pink and plotted it on any scale I chose. I just choose pink noise 'cuz I'm comfortable using it for a variety of measurements. Again, it really doesn't matter as long as it contains the whole audio spectrum.
IgorC's picture

I was looking at THD+N of my beloved HD800 http://www.innerfidelity.com/images/SennheiserHD800.pdf
and comparing them to Stax 009.

At first sight my conclusion was that Stax have less distortion in LF while for middles and highs both phones have more or less the same amount of distortion and do an excellent job.

Yes, but the ear is significantly less precise in LF (<100 Hz) and HF (>16kHz) ranges.
A-weighting (dBA) is more appropriate for the measurements of audio equipments.

Applying the A-weighting to data sheets of HD800 and any other phones the THD at frequencies <100 Hz have gone down to 0.1% or so.
It makes sense as I don't hear any issue with the bass of 800's neither I've seen any reviews about distorted bass for these cans.

It will be great to see datasheets with extra data (THD+N in dbA).

And Thank You for informative and useful sources.

i_like_macs's picture

Hello,

First, thank you Tyll for creating such a huge database of meaningful headphone measurements!

I can't remember where I read this, but I think an Audio Engineering Society (AES) paper mentioned that headphones should be measured away from walls because reflections could affect the measurements. I see that the inner walls to the sides of the headphones in the acoustic isolation chamber aren't lined with acoustic wedges. Do the reflections from these walls cause problems?

Thank you very much,

i_like_macs

X
Enter your InnerFidelity username.
Enter the password that accompanies your username.
Loading