Building a Headphone Measurement Lab
May 6, 2010 at 10:51 AM Post #152 of 355

Tyll Hertsens

Garmentus Vulgaris & Headphoneus Supremus
Member of the Trade: Innerfidelity
Joined
Jun 22, 2001
Posts
4,102
Likes
1,185
Okay, I spent time over the past few days measuring total harmonic distortion. Actooly, THD+Noise vs. Frequency.
 
In these measurements a tone is swept through he frequency range. After it comes back from the head, the fundamental tone is filtered out; the remaining audio will be the distortion components (2nd, 3rd, 4th harmonics, etc) and any noise in the system.
 
There's a couple of things to be aware of as you view the graphs:
  1. Each curve on the graph uses a tone of a different amplitude (10dB apart). Typically the higher the volume, the more distortion. But remember that each curve represents the percentage of distortion relative to the reference tone level. As the reference tone goes down in level, the noise in the system goes up as a proportion of the reference tone. Therefor, you'll notice that in the higher frequencies of the graphs, where noise dominates over harmonics, the curves cross, as the noise is an ever larger percentage of the reference tone as it gets lower in level.
  2. Because we are measuring the amount of energy contained in the harmonics of the fundamental tone as it is swept, once we get beyond about 5kHz the measurements become ever more meaningless. At 5kHz for example, the second harmonic is at 10kHz, third at 15kHz, and the fourth at 20kHz. Because the system begins to loose response above 20kHz, the amount of harmonic distortion is being artificially reduced by the system bandwidth. This can be seen in the higher reference levels where the distortions are extraordinarily low in the high frequencies. Also at 7kHz the AP automatically limits the measurement to 2nd and 3rd harmonics only.
 
Sennheiser HD600
 


Sennheiser HD800
 

 

AKG K702
 

 

The data rather surprises me. The lowest curve in these graphs is 90dB in two cases; 85dB in the other; that's still a fairly strong volume level. These curves show that good headphones have fairly low levels of distortion, unless being driven to loud levels. That the K702 behaves poorly at 110dBSPL is really not too much of a problem because that's a ridiculously high level to be blasting your ears with. In fact, your ears are probably distorting worse than the headphones at that point as your acoustic reflex will have kicked in and the dominant noise source will be the muscles in your ears tensely telling you to TURN IT DOWN!!!
 
Another thing about these graphs: driving a headphone amp into, or near, clipping in order to measure it is one thing; driving headphones into clipping to measure them is Russian Roulette with the headphones. As I measured the K702 at 110dB I though for sure it was going to blow up. So the highest two curves of the graphs (110dB and 105dB) I would be very hesitant to impose on headphones on a regular basis.
 
Looking at the graphs we can see they rise as they go lower in frequency. At the moment I'm assuming this distortion is the driver diaphragm approaching its excursion limits and going non-linear. Since all the headphones above showed some characteristic distortion at 100dB, I figure we can use that as an upper limit for the moment.
 
Since these pretty good headphones were behaving quite well by the time they reached 90dB (the curves are fairly flat meaning distortion products are basically at or below the measurement noise floor) there's not much reason to test at lower volumes.
 
So making the measurements at 90dB and 100dB seems a good start. Let's see if it yields some good information.
 
Here's a graph of a bunch of different headphones measured at 90dB and 100dB. (You'll notice I didn't test above 7kHz as it becomes somewhat unmeaningful as mentioned previously.
 

 
 
Things I see (this is a lot of guesswork):
  1. The HD600 measures best of all the headphones in terms of low distortion. I'll go out on a limb here and say that the perception of the "Sennheiser Veil" may, at least in part, be due to folks not being used to hearing the smoothness of such low levels of distortion.
  2. The at the low end 650 and 800 do very well, the D5000 is pretty good followed shortly thereafter by the DT880; in the high frequencies they perform similarly, with the remarkable exception of a significant distortion peak centered just above 2kHz, possibly cone break-up of some kind.
  3. The 800 has a markedly different shaped curve, which might be indicitive of it's toroidal diaphragm design that it so fundamentally different than other more traditional drivers of this group.
  4. The small supra-aural PX200 had a VERY hard time putting out sound at these levels. It struggled mightily even at 90dB.
  5. The cyan (aqua) colored trace of the cheap HD201 headphones show a sharp distortion peak at about 205Hz. Again, I assume this is a resonant cone break-up.

At this point I switched over to looking at distortion spectra. I was curious if I could see a marked change in distortion spectra between, say, 170Hz and where the distortion peak occurred at 205Hz on the HD201.
 
In the graph below a single tone at 90dB is played at 170Hz (yellow) and 205Hz (green). The series of peaks to the right of the fundamental are the byproducts of non-linear distortions and occur at multiples of the fundamental tone. if you look carefully, you can see that the even order harmonics (2nd 4th 6th) harmonics of the 170Hz (yellow) are much lower in level than the even harmonics of the 205Hz signal. This indicates to me that the spike in the HD201 THD vs F data is an exitation of an even order vibrational mode in the driver diaphragm.
 

 
 
I'm really happy with how much information I got out of the THD vs F graphs, and I think I'd like to include them in the final test sequence. OTOH, I had planned to include distortion spectra data, but after seeing the significant difference on the 201 between 170Hz and 205Hz I'm beginning to wonder if it's a fair measure. Since a single frequency would be chosen to make the test, and because the harmonic series may be markedly different at different frequencies there is no way to make the test indicative of the whole character of the cans.  A single (or even a few) distortion spectra measurements might be misleading.
 
However, because all the cans show marked distortion at the low end, I did run some tests displaying distortion spectra at 50Hz. I did the test at both 90dB and 100dB for the cans that I've been playing with. I'm don't think it delivered any where near the rich amount of info the THD v F graphs did, IMHO. See what you think:
 
50Hz @ 90dB
 

 
50Hz @ 100dB
 

 
 
The intention at the moment is to construct a standardized series of tests to characterize headphones for public display. It seems to me that it would be impossible to select one frequency and amplitude to produce distortion spectra that would be clearly characteristic of the headphones under test. I feel the THD+noise vs. F do a much better job of that.
 
Once we get past the first hurdle of developing a standardized test and have tested the bulk of the headphones needed, there will certainly be room and time to make closer and individualized measurements of headphones to draw out some finer detail on a case by case basis.
 
At the moment I'm thinking that a 90dB THD+noise Vs. F is definitely needed. I'd like to do the measurements at 100dB, but I worry about some headphones not possibly becoming damaged, and it's certainly not an "approved" listening level.  OTOH, it's nice to test the limits of things; seeing the curves at both 90dB and 100dB give a very good indication of power handling capability.
 
Looking forward to your comments on distortion measurements.
 
wave.gif

 
May 6, 2010 at 11:20 AM Post #153 of 355

MrGreen

Headphoneus Supremus
Joined
Oct 9, 2009
Posts
2,389
Likes
67
The HD600 is a surprising result for THD since it sounds less clean to my ears than the K701. Probably a product of other factors involved with the K701, though since the THD for both is quite low at the volumes I listen
 
May 6, 2010 at 12:04 PM Post #154 of 355

Jamey Warren

Member of the Trade: HeadRoom
Joined
Sep 23, 2003
Posts
739
Likes
69
Nice work Tyll.
 
My thoughts about the 100 dB measurement.
 
Even though it's not a recommended listening level, live music often reaches this level though at times it's not sustained.  With a group like an orchestra it may only be reached at the end of the performance.  With a good recording there's plenty of dynamic range available to require the ability to reproduce sounds at or above 100 dB.  How well a headphone can do that would seem pretty important as to it's ability to resolve music naturally.
 
Just a quick thought...
 
May 6, 2010 at 12:23 PM Post #155 of 355

Tyll Hertsens

Garmentus Vulgaris & Headphoneus Supremus
Member of the Trade: Innerfidelity
Joined
Jun 22, 2001
Posts
4,102
Likes
1,185


Quote:
Nice work Tyll.
 
Thanks mate.
 
My thoughts about the 100 dB measurement.
 
Even though it's not a recommended listening level, live music often reaches this level though at times it's not sustained.  With a group like an orchestra it may only be reached at the end of the performance.  With a good recording there's plenty of dynamic range available to require the ability to reproduce sounds at or above 100 dB.  How well a headphone can do that would seem pretty important as to it's ability to resolve music naturally.
 
Just a quick thought...


I think I agree with you, Jamey. I thought the contras between the two measurements (at 90dB and 100dB) showed a good contrast in performance over a tough region. I worry though, that some of the earbuds and smaller earpad designs (and cheap crappy cans of whatever size)  might struggle so much that they would be damaged or be so distorted that the measurement goes pretty wonky. OTOH, that's something I'd want to know ... cans should work at that level, as you point out.
 
May 6, 2010 at 2:04 PM Post #156 of 355

xnor

Headphoneus Supremus
Joined
May 28, 2009
Posts
4,092
Likes
221
I'm sorry for posting inside the quotes, but this new editor imho is a PITA. 
tongue.gif
 (can't be reverted with an option, right?)

 
That's some very nice and informative data you gathered there. 
Quote:
Another thing about these graphs: driving a headphone amp into, or near, clipping in order to measure it is one thing; driving headphones into clipping to measure them is Russian Roulette with the headphones. As I measured the K702 at 110dB I though for sure it was going to blow up.
 
Hehe, I'd love to see a picture of you standing in front of the dummy head that is about to explode with one of those bordering on (scientific) madness facial expressions. 
k701smile.gif

 
 
 
I'm really happy with how much information I got out of the THD vs F graphs, and I think I'd like to include them in the final test sequence. OTOH, I had planned to include distortion spectra data, but after seeing the significant difference on the 201 between 170Hz and 205Hz I'm beginning to wonder if it's a fair measure.
 
That's cool and I also think it's more useful, that's why I suggested it. 
 
 
Once we get past the first hurdle of developing a standardized test and have tested the bulk of the headphones needed, there will certainly be room and time to make closer and individualized measurements of headphones to draw out some finer detail on a case by case basis.
 
All excited about what you'll be coming up with next.


 
Quote:
cans should work at that level, as you point out.

I also think they should be able to produce 100 dB with f around 1 kHz but I'm sure you'd blow one or the other 'phone by pushing them to produce such levels at low frequencies.
 
May 6, 2010 at 2:27 PM Post #157 of 355

xnor

Headphoneus Supremus
Joined
May 28, 2009
Posts
4,092
Likes
221
Comparing your K701 meas. with those of a jap. guy
 

 

 
You can see that 2nd and 3rd harmonic is down about -46 and -52 dB respectively. That's about ~0.55% (2nd & 3rd harmonic only!). So it looks like he's also measuring above 90 dB but not 100 dB.
It's nice to see that he also measured the 1.3-ish kHz peak. 
wink.gif

And the CSD also is nice to look at, hehe.
 
May 6, 2010 at 3:44 PM Post #158 of 355

Tyll Hertsens

Garmentus Vulgaris & Headphoneus Supremus
Member of the Trade: Innerfidelity
Joined
Jun 22, 2001
Posts
4,102
Likes
1,185
It's nice to see that he also measured the 1.3-ish kHz peak. 
wink.gif

And the CSD also is nice to look at, hehe.


Sweet.  What do you think you can see in the CSD data?
 
xnor said:
All excited about what you'll be coming up with next.
 
(Wha!?  Oh, my.   
frown.gif
  Anybody know how to build a quote from scratch?)
 
Aaaaanyway .... yeah what is next? 
 
What I'm actually doing here is getting the program and spreadsheets all in place to do the measurements at CanJam. The ones that are already in place are the ones on this sheet.
 
The phase measurement is problematic as the distance between the driver and ear should be compensated out one each headphone, and it doesn't do that at the moment.  However, some math on the impulse responce will get you that, and probably much better accuracy.
 
50Hz and 500Hz square waves are in there, as xnor mentioned before, it might be interesting to have a look at lower frequencies for both. Maybe 30 and 300.
 
The new one I'd like to try is some intermodulation distortion measurements.  The AP has some very cool multi-tone measuring capabilities; I'd like to look into that a bit.
 
 
Suggestions?
 
 
May 6, 2010 at 10:15 PM Post #161 of 355

rsaavedra

Headphoneus Supremus
Joined
Jan 20, 2002
Posts
5,819
Likes
21
Awesome work Tyll!!! 
atsmile.gif

 
Interesting that you find the HD600's distortion lower than the HD650's. I remember, back when the 650 was released, it was measured to have even slightly lower distortion than the HD600.
 
In any case, I'd only like to mention that your measurements are just making me even happier about my recently acquired HD600. Enjoying its sound immensely btw. What a great sounding headphone it is. 
 

 
Quote:
  1. The HD600 measures best of all the headphones in terms of low distortion. I'll go out on a limb here and say that the perception of the "Sennheiser Veil" may, at least in part, be due to folks not being used to hearing the smoothness of such low levels of distortion.
 
Looking forward to your comments on distortion measurements.
 

 
May 6, 2010 at 10:34 PM Post #162 of 355

Tyll Hertsens

Garmentus Vulgaris & Headphoneus Supremus
Member of the Trade: Innerfidelity
Joined
Jun 22, 2001
Posts
4,102
Likes
1,185


Quote:
Awesome work Tyll!!! 
atsmile.gif

 
Interesting that you find the HD600's distortion lower than the HD650's.

I didn't measure the 650 in this round, the 600 was thew lowest of the batch I maesured.  I think the 650 will have less, but we'll see sooner or later.
 
But I do like the sound of both. Yes, good cans.
 
 
May 6, 2010 at 10:41 PM Post #163 of 355

rsaavedra

Headphoneus Supremus
Joined
Jan 20, 2002
Posts
5,819
Likes
21

Oops my bad, sorry for my confusion. Read that you had found the 600 to have the lowest, and assumed the HD650 was among the others. Replied too quickly. Looking forward to your 650 measurement.
 
(Well, in fact, looking forward to ALL the measurements from CanJam!!! 
atsmile.gif
)
 
Quote:
I didn't measure the 650 in this round, the 600 was thew lowest of the batch I maesured.  I think the 650 will have less, but we'll see sooner or later.

 
May 7, 2010 at 8:35 PM Post #164 of 355

xnor

Headphoneus Supremus
Joined
May 28, 2009
Posts
4,092
Likes
221
Quote:
Sweet.  What do you think you can see in the CSD data?

 
I think that the lower end's just artifacts from FFT processing.
Looking at mids and treble shows:
- ridges at the same frequencies where the FR shows peaks (1.2 kHz)
- that 6 ms is not enough for some resonances to fall by 30 dB 
k701smile.gif

- looks like there are also some delayed resonances (those that just appear)
- lot of stuff going on around that 1.2 kHz peak, maybe noise/distortion?
 
 
Quote:
The phase measurement is problematic as the distance between the driver and ear should be compensated out one each headphone, and it doesn't do that at the moment. However, some math on the impulse responce will get you that, and probably much better accuracy.


As are square waves, imho.
 
 
Quote:
50Hz and 500Hz square waves are in there, as xnor mentioned before, it might be interesting to have a look at lower frequencies for both. Maybe 30 and 300.


I was just picking 30 Hz back then because that's were phase shift seemed to be lower for the (at 50 Hz square wave worse looking) headphone according to the image I posted above.
Wanted to see if that would "turn the tables" resp. reduce the skewing of the square wave. 
wink.gif

 
 
Quote:
The new one I'd like to try is some intermodulation distortion measurements. The AP has some very cool multi-tone measuring capabilities; I'd like to look into that a bit.

 
Great! 
 
edit: Changed quotes, thanks MrGreen.
 
May 8, 2010 at 3:35 AM Post #165 of 355

MrGreen

Headphoneus Supremus
Joined
Oct 9, 2009
Posts
2,389
Likes
67
 
 
Quote:
Originally Posted by Tyll Hertsens /img/forum/go_quote.gif
xnor said:
All excited about what you'll be coming up with next.
 
(Wha!?  Oh, my.   
frown.gif
  Anybody know how to build a quote from scratch?)
 


Next to the smiley button is a " button. That'll make you the code box and you can write in it.
 
Its a really bad system for quoting. Haven't worked out another way. They should add a bbcode for it.
Quote:
like this

 

Users who are viewing this thread

Top