[Tyll Hertsens]

Quote:

Originally Posted by MrGreen /img/forum/go_quote.gif I'd hate to complicate things, but what about the effects of rotating the headband forward or backwards?

It would be hard to make this too much more complicated than it already is; no, it's a good point. I did notice that it looked like I was putting on the headphones in that clip witht he bands a little too far rear. The problem is that it was tilted back at a pretty good angle and I couldn't quite tell if things were lining up.

Usually, the headband is slightly forward of vertical from that seem running up and down on the head. Given that the positioning of the headphones don't make as much difference as I thought it might, I kinda doubt rotating the band will do much, but I'll give it a try it though.

 

[rsaavedra]

Quote:

Originally Posted by Tyll Hertsens /img/forum/go_quote.gif It's going to be fun in future to have many measurements of some of the most common headphones to settle some issues once and for all, eh? There's a lot measurements can't tell you, but whether one has more bass than the other surely would be apparent. I do think you have to do a large enough sampling and also know what vintage the headphones were. It would be interesting to see if a model had changed over time. It's a lot of measurements, but we can get there over time.

Definitely, that would very very fun and interesting, Tyll. In case you can't tell, I'm drooling with all these measurements and graphs, and all these possibilities they point to

I strongly support the idea of sampling several headphones of same model to get a more representative response graph per model. That would be a bare minimum statistical recommendation after all. Because of possible manufacturing variability, I even tend to think that measuring several headphones of same model is even more important than measuring one single headphone in 5 positions. (But the latter is a good idea too, so not underplaying it.)

"Same model" not just meaning the market model name or number, but definitely also taking into account what you mention: the vintage version.

 

[Tyll Hertsens]

Big post coming up.


First, I feel like I need to say that I love all the interesting comments and brainstorming, but I do have a mission I'm on, which is being able to measure headphones at CanJam, which is just a few weeks away. So I have to keep moving forward and staying focused on that. If at times I just sort of run by a few things I apologize, it's simply because I'm trying to stay on track. I can tell you that I am carefully reading the comments though and many of them have already been productive.

Before we get too far with this post, I'll point again to this Positioning Tests PDF, I'll be referring to it.

On to some questions:

Quote:

Originally Posted by rsaavedra - Do you know how treble-wise accurately you can reproduce the positioning of the headphones on the dummy head? Or in other words, do you know the standard deviation or variance you can get in your treble measurements when repeating measurements for a specific positioning of the headphones on the dummy head?

No, I don't. But from the test just done on position, it looks like:

Most headphones vary less than I was expecting. In these tests I generally just moved the headphones as far in each direction as I could without being in an obviously incorrect position. The only headphones that really moved very far were the HD800s, which have huge ear cups.

In most cases, as you look at the data, it seems like there is a characteristic cureve, and then amplitude fluctuations in the curve. This can generally be seen in the fact that the left and right channels seem to track each other fairly well.

Also, have a look at the first two K701 tests and the two D5000 tests. In both cases the two tests were taken two days apart. What I see in the data is two fairly strongly repeatable tests. The first D5000 test has quite a bit more deviation in the high frequency information, but when you look at the averaged data curves they look very similar.

Quote:

Originally Posted by rsaaveddra - How exactly do you determine the five different headphone positions you choose to gather FR measurements from?

As shown in the little movie, it's become apparent that simply feeling for the right position is pretty much the only way to do it. At first blush I thought the laser idea might work, but what are you really targeting. It's the same with the square wave thing, I can tell if the seal is changing 'cuz I'll loos bass, and I can tell that things are changing in the highs, but as has been pointed out to me, who the heck know what's right.

Seriously, you should go try putting hradphones on someone elses head, and you'll see that it's fairly obvious when things are right.

I'll also add that how much you can move is a total crap shoot. Some headphones you can move a lot, some not so much. I've come to feel that headphone makers have control over these things (how close the want the cans to fit around the ear, how much space is available) so they are responsible for building a headphone that can be worn in widely differing positions or not.

Given that the method of taking five measurements seems to be repeatable and moderately insensitive to movement, I'm thinking that the "do it by feel five times with moderate movement in each direction away from center" is a good one ... in fact maybe the only practical one.

Quote:

Originally Posted by Roger Strummer /img/forum/go_quote.gif Hey Tyll, I was wondering, since there is difference between channels, when one looks at the graphs in the headroom site (in the page where you can compare various headphones) are you seeing only one (arbitrary) channel, an average of both or what?

It's the left chanel --- the one on the head the seems to measure best.

Quote:

Originally Posted by rsaaverdra I think the concentric shooting target circles around the ear canal, or these laser pointers, or even both strategies together, might really help making placements more accurately repeatable, regardless of headphone (non-IEMs, of course.)

Just wanted to add a point here that I can feel the seam around the ear which allows me to move the earpads fairly reliably to the five different positions.


Quote:

Originally Posted by xnor /img/forum/go_quote.gif The graphs are as usual very nice. It shows that it makes sense to average a couple of re-placements. I wonder how much different the averaged result would look like if you re-did those 5 measurements.

The two D5000 and K701 graphs were done on different days.


Quote:

Originally Posted by xnor Btw, does anyone know how clamping force is defined in IEC 60268-7? Would be nice to get free copies of such standards..

Just gave a quick flash through it, but I couldn't find any spec.

Quote:

Originally Posted by xnor edit: Another thing that just came to my mind is linearity: start measuring the FR at 50 dB and double the voltage (approx. +6 dB) every measurement until you're at ~92 dB. (8 measurements) If you put all the measured curves on top of each other the result should be a single perfectly overlapping curve, shouldn't it?

Quote:

Originally Posted by rsaavedra It would be really interesting to have different FR's for different volume levels per headphone. That would mean, of course, more measurements and more work, I know. I'm just dreaming out loud.

I'll see if I can fit one of these tests in. Currently I'm doing most measurements at 90dBSPL at the ear.

While I don't think we'll see much of a difference in fr based on SPL, I do think well see some differenced in distortion as it gets higher. So I'm currently thinking I will have some Frequency vs THD and hamonic series spectra at different output levels.

My justification for using 90dB is that it's loud, but well within the range where the headphones should be acting normally.

Quote:

Originally Posted by pictureguy /img/forum/go_quote.gif Are you going to build a dummy load for amps? By this I mean NOT just a power resistor and a DVM with signal generator / oscillator. I mean a truly reactive load. I can find some schematics which the loudspeaker folks use. This will show an amps response into a more realistic load. I don't know if it is possible to make a sound transducer of pure resistance. Also, as fun as it is to measure sensitivity and impedance and max spl of a transducer, wouldn't it be also nice to know the PHASE ANGLE of the load? Phase angle is important since increasing angle results in less power being delivered to such load. For example, at a given frequency, if the phase angle is 45degrees, the result is a loss of power of nearly 30%. (cosine of phase angle)

These are both good points. Would you like to help by building some dummy loads for amps? I won't be getting to that for a while yet as the headphone measurements need to be well along the way first, but I'd love the help.

We should be able to get the fphase information out of the impulse response data. I will be persuing that.

Quote:

Originally Posted by rsaavedra /img/forum/go_quote.gif Clamping force is something I've been thinking about as well. Don't know how it is defined in that standard though. Also, in perceptual terms (maybe measurement-wise too,) I think a different level of clamping force might result in bass perceived somewhat differently. PS. Keep in mind I'm mostly thinking about the repeatability/reproducibility of the method and observed measurements.

These quotes are pretty right on. I do see bass warm up a bit as clamping force is increased, and the volume get louder overall.

And in terms of repeatability, I agree, too much external figiting with the cans (like putting a rubber band around the whole thing to improve the seal) is really beginning to over influence the test IMHO.

The manufacturers know (or should know) what the specified head for measurements is, a lot of work has gone into it, I'm of the opinion curerently that we should just place the cans as well as possible on it, and leave it at that. Though I would be willing to adjust the headband a bit (like with Grados, lets say) to make sure the fit as properly as possible on the head. Remember, the heas is slightly smaller than the average male head, so this can be a bit of a problem.

Quote:

Originally Posted by rsaavedra /img/forum/go_quote.gif And that is yet another source of possible variability in the measurements, the volume of the signals being played by the headphones.

At the bigining of each test the volume is servoed to be 90dBSPL at 1000Hz.

There's a bit of a bug in the test somewhere, or the lines would be laying on top of each other better. I thinkk it's servoing at the beginning of the test, but not for each subsequent sweep. I'll be working on fixing this soon.

Is the FR being gathered using pure sine waves or warble tones? And at exactly what level at which frequency? (e.g. 80 dB at 1 kHz?)]

It's a swept sin wave. The IEC spec says that swept sin, integrated pink noise, and calculation from impulse respose are all allowed, but they prefer impulse response as it also deliver phase info. We may go that way befor this is all over.

Quote:

Originally Posted by xnor /img/forum/go_quote.gif As for volume matching I'd use the "standard method" with a 1 kHz sine.

Actooly, I'm thinking I should use something more like 500Hz as most headphone response is pretty well behaved and flat up to that point; above it things start wiggling around some and may result in fluctuation of over level of the sweep. I think it something like this happening as evidenced by the modest but clear level shifts between curves now.

 

[JohnFerrier]

Quote:

Originally Posted by Tyll Hertsens /img/forum/go_quote.gif Actooly, I'm thinking I should use something more like 500Hz as most headphone response is pretty well behaved and flat up to that point; above it things start wiggling around some and may result in fluctuation of over level of the sweep.

Ah . . . noticed, earlier this morning, that Beyerdynamic specs the T1s at 500 Hz.

Nominal SPL . . . . . . . . . . . . . . . . . . 102 dB (1 mW / 500 Hz)
T.H.D. . . . . . . . . . . . . . . . . . . . . . . . < 0.05% (1 mW / 500 Hz)
Max. SPL . . . . . . . . . . . . . . . . . . . . . 126 dB (300 mW / 500 Hz)

T1 Datasheet

 

[Tyll Hertsens]

Thanks John.

 

[rsaavedra]

Excellent clarifications, Tyll! And lots of info as well.

In all of it, the only point you made that I'm kind of least sure of is measuring at 90 dB. It is clearly within the specs of pretty much any headphone, I agree, but I'd think it is also clearly louder than comfortable. Not thinking about distortion levels, but actual reallistic usage of the headphones.

For example, few people would stand a pink noise track playing continuously at 90 dB for hours. It would't be bearable for very long by anyone, I think. In fact, it would start causing hearing damage after not very long. (There are some other tables out there though.) So I wonder if 90 dB is the most appropriate loudness level at which to measure the responses of headphones.

If the responses of headphones aren't that different between let's say 80 dB or 85 dB compared to 90 dB, then we're good. Also, if you decide or eventually find the time to take measurements at < 90 dBs, then we're also good. But in any case, just measuring at 90 dB seems to me louder than it should be. Just my two cents.

 

[xnor]

Thanks for the answers Tyll.

I mentioned this a couple of pages back, have you taken a look at arta yet?

And I'm not sure if you mentioned it somewhere already, but what amp/source do you use for measurements currently? More importantly, is its output impedance low?

 

[rsaavedra]

Quote:

Originally Posted by Tyll Hertsens /img/forum/go_quote.gif At first blush I thought the laser idea might work, but what are you really targeting.

I'm targetting just repeatability.

Let's say someone in Japan decides to take measurements of headphones, wanting to confirm your observations. They get a similar dummy head as yours, same microphones, uses same sine waves, uses a good enough isolation cage to take the measurements from, and also chooses to measure at 90 dB on 1 kHz, just like you. So all seems pretty much equal. (Assuming amp and oscilloscope would make very little difference in the measurements.)

Hence the main variability might come just from the headphone unit itself. But let's say he also chooses, as in your methodology, to take five measurements following your same strategy: center, up, down, back, and forward. Let's say he also chooses to measure at least three headphones per model and vintage, not just one, assuming you had chosen to do the same.

After all of that, the main sources of variability in the measurements would boil down to positioning. Both of the headphones on the head, and also, clamping pressure, or rather, distance between driver and ear canal.

So the aim with the lasers would be to replicate positioning as accurately as possible, at least with respect to the headphones on the dummy head part. (Accurately replicating driver-ear canal distance is a much harder problem, I think.)

If the Japanese guy places his lasers at exactly the same spots in space around his dummy head as your lasers, and you both agree on which spot on the outer surface of each headphone's cup is the center, then for him to replicate your placement of headphones on his dummy heads would be a piece of cake: just place the headphones so that the lasers fall right over the middle dot in the headphone's cup. Then 5 mm up; then 5 mm down; then 5 mm back; then 5 mm forth. (There should have been some agreement on where over the skull the headband would need to be placed though, so that the Japanese guy's entire headphone set ends up placed just exactly like yours.)

This might be overkill, but in terms of repeatability of experiments, and likelyhood of obtaining exactly the same measured results, the most exact possible repeatability of the placement of headphones would be a must.

 

[Tyll Hertsens]

That's funny: when I said targeting, I meant where are you going to point the laser.

Quote:

the most exact possible repeatability of the placement of headphones would be a must.

To be truthful, I don't think this is true. Averaging FR over five placements removes some of the fine detail that happens for a single measurement. I think when the spec calls for five placements, the intention is to rid the measurement of some of this fine detail differences due to small changes in position.

I guess my thought is that your way above you'd set-up some standard that's very hard to replicate.

Also, after doing some more measurements, I'm thinking it's not really needed.

Here's the latest set of measurements: This PDF has three headphones measured a number of times each.

You'll notice the one in the 701 with April in the title; I had my 13 year old daughter put the headphones on the head 5 times (not the up-down-front-back, just had her put them on the dummy five times without any coaching). You can see that she didn't get the seal right because the bass is a bit wonky, but the highs line up fairly well with the other measurements.

One thing to note is the first day I did some of these measurements I had a cable reversed so left and right were swapped making the compensated response inaccurate. So, on some measurements it's better to look at the similarities between raw measurements and ignore the color change.

Bottom line, to my eyes, these measurements show good enough repeatability to move forward.

I'll state for the record now though, that I will be willing to revise this methodology as time goes on and we all learn more.

Next up: Distortion measurements at different volumes.


Edit: But please feel free to make comments on the repeatability observations and issue.

 

[rsaavedra]

Quote:

Originally Posted by Tyll Hertsens /img/forum/go_quote.gif That's funny: when I said targeting, I meant where are you going to point the laser.

Ah, they would point to the ear canals. I had mentioned it, actually:

Quote:

Originally Posted by rsaavedra /img/forum/go_quote.gif two laser pointers somehow fixed symmetrically and appropriately on either side of the surroundings of the dummy head, and each laser aiming directly at each ear canal.

 

[rsaavedra]

Quote:

Originally Posted by Tyll Hertsens /img/forum/go_quote.gif To be truthful, I don't think this is true. Averaging FR over five placements removes some of the fine detail that happens for a single measurement. I think when the spec calls for five placements, the intention is to rid the measurement of some of this fine detail differences due to small changes in position. I guess my thought is that your way above you'd set-up some standard that's very hard to replicate. Also, after doing some more measurements, I'm thinking it's not really needed.

Positioning the lasers at the same coordinates in space around the dummy head, and aiming each one at each ear canal doesn't seem hard to replicate to me. And in general, the purpose of the laser idea (or the concentric circles idea, for that matter) would be to make the positioning as exactly replicatable as possible.

I did admit though, the lasers might be overkill. Headphone manufacturing variability already would call for averaging several measurements and not just on one headphone (I tend to think,) but on a few of the same model, as we discussed already.

The real goal of the accurate positioning is to help replicate the experiments so as to help obtain the same results for a specific headphone, also helping to make comparisons of FR measurements between different headphones more a matter of true apples and apples, not possibly apples and oranges due to different positioning for different headphones -in particular, for headphones of different types, like circumaural vs. supraaural.

 

[xnor]

Anyway, I don't think it's very likely that somebody else would be able to (exactly, within a small margin) reproduce Tyll's results if you look at them as being "absolute". There are too many variables involved.
It makes more sense to look at the differences between two headphones that were measured the same way. If you look at FR, there's no single perfect ideal curve either. Again, too many variables involved as we've seen with stuff like equalization (DF, FF), the vendor provided head correction curve etc. But comparing two FRs makes sense.

Of course, we all want the measurements to be as precise as possible. With averaging the final results are more consistent and easier to compare, even with a bit sloppy positioning I guess.
Tyll doesn't have unlimited time and I also think that the results are accurate enough for now. Time to look at other interesting things!

 

[Feanor]

Tyll, maybe this has been discussed before, but have you thought of normalizing based on these curves?

As far as I understand it, now you are compensating for the ear canals FR, but the psychoacoutics are important too.

 

[Tyll Hertsens]

Quote:

Originally Posted by Feanor /img/forum/go_quote.gif As far as I understand it, now you are compensating for the ear canals FR, but the psychoacoutics are important too.

I answered in the other thread, but will repeat it here:

One hopes that the recording engineers used there ears well and mixed the music for the desired balance. Once they make that determination the job of the electronics is to be flat so that the proportions of bass and treble remain the same until it reaches the ears.

The only usefulness at the point of the Fletcher-Munson curves is if you are reproducing the sound at a different volume than the engineers were mixing it at. If you listen much loer, you should dial in some "loudness curve" to boost the lows and highs a bit.

 

[xnor]

Exactly, it'd be meaningless.
Since Tyll mentioned balance: music has a spectral balance similar to that of pink noise or vice-versa.
And when you play pink noise with your speakers/headphones you most probably want to hear the same noise coming out of the speakers and not something smiley curved.