Want to see a W2002 response plot?

[halcyon]

Tomcat,

thanks (re: HRTF). Fixed.

gerG,

Sources to read:

D. Hammershøi and H. Møller, "Sound transmission to and within the human ear canal." J. Acoust. Soc. Am, Vol. 100. No. 1. 408-427. p. (1996)
- Very good paper on various measurement ways and how they affect the results. Summary says that blocked ear canal has lowest standard deviation from subject to subject, but then again that's not taking into account the full canal resonance.

Henrik Møller, Clemen Boje Jensen, Dorte Hammershøi and Michael Friis Sørensen, "Design criteria for headphones," J. Audio Eng. Soc., 43(4), 218-232(1995).
- Another Møller classic that shows the various open/closed diffuse/free field equalized headphone design differences.

Henrik Møller, Dorte Hammershøi, Clemen Boje Jensen and Michael Friis Sørensen, "Tranfer characteristics of headphones measured on human ears," J. Audio Eng. Soc., 43(4), 203-217(1995).
- I don't have this yet myself. Considered essential by many.

Loudspeaker and Headphone Handbook, John Borwick (ed.), 3rd. edition, Focal Press, 2001
- The chapter on Headphones (chap 14 by C.A. Poldy) is very good, although a bit dense (at least for me). I haven't distillied nearly all of it yet. Good nevertheless

Why head-related transfer functions are so difficult to measure?, http://www.acoustics.hut.fi/teaching/S-89.144/kar.pdf
- Teaching material by Klaus Rieder on difficulty on getting good HRTF in general. He's doing his doctoral dissertation on HRTFs, I believe. Klaus also has two good AES papers, which I can't find right now.

Headphone listening test methods, masters thesis by Toni Hirvonen, http://www.acoustics.hut.fi/publicat...rvonen_mst.pdf
- Easy to follow basic stuff that blends theory to measurements and shows graph results. Not everything is covered, of course.

I'm probably forgetting something really important and I've only covered the surface myself, but those come to my mind right now.

Quote:

the ear canal will have a resonance effect, which is why I wear an earplug when I am doing these tests. Mic placement is

Yes, blocking the canal (should be silicon - all the way to eardum I believe) can lessen the subject to subject variance in measurements. This has it's downsides (see later on).

Quote:

critical, but only above a certain frequency level. That frequency seems to be a simple function of the cavity dimensions.

For the canal yes, AFAIK. The pinna has also distorting effects, although they are linear as well (and intersubject variations can be large). These _mainly_ affect HF though, AFAIK again.

Quote:

I believe that this can be done, and is done, but at a basic level. For example a headphone with a 3" id pad and a driver [.... SNIP]

An interesting point and this is probably what headphone designers do.

The problem of course is finding the generic mean head based on which the damping should be calculated.

If one removes all the resonances, there go some of the first wavefront (non-reflected) sound localisation cues as well.

So how do we design a good headphone, when we should damp the cavity resonances, but not destroy the listener specific HRTF localisation cues?

Quote:

effects. Overall I do believe that getting rid of these reflections is a good thing.

I'd agree on this. However, it is tricky not to kill the initial transducer/enclosure performance when doing the damping of reflected sound, when the damping is part of the enclosure...

It becomes an endless loop of correcting the corrections of corrections of ... ad nauseaum.

Still, that's how headphones (at least the best closed ones) should be designed, IMHO.

But I'm not an expert on headphone design, so don't take my word for it

Quote:

As far as a simulation model, it would seem feasible to build an ANSYS model to simulate the whole system right down to the eardrum, but only for a specific individual. It would be a big job, and I would not have a clue how to interpret the results. otoh it would be very interesting.

Indeed, a big job and for one individual.

However, if one wanted to kill the signals and not the associated transducer performance, I think one would have to have a DSP system which would look into the the future and past of signal to be played and adaptively cancel most of the effects of the cavity.

However, even a slight change in position of the headphones could change the extent to which this cancellation would work, as the transfer function of listener head/ear + headphones needs to be measured at some location. Of couse, one could use a statistical mean (possibly) and just use that, so that the cancellation would work in most cases, but not fully.

Again, a lot of work and I'm not sure it's feasible in practise.


Quote:

I think that many of your observations deal with how we percieve directional cues, and how careful measurements must be taken to capture those effects.

Well I wasn't actually aiming for that.

In hindsight I must say that much of my 'criticism' against your measurements was unfounded, because you *did* use head coupling, blocking ear canal, etc.

However, my comments were still related to the various (non HF related) problems with your and other measurements as well:

- individual ear changes make measurement data unpredictable in the 3-5 kHz region (ear canal resonance at presence region)

- the proper way to position headphones and the mic (not to mention how to block the canal, at concha or at ear canal level) again alter the measurement data

As such the measurement data can be somewhat or even very misleading for other people than you. Also, they might not be even very accurate for you, because the ear canal resonance has been eliminated.

This of course makes them less listener dependent results, but less accurate for you.

Also, for supra-aural headphones the position will change the measurements quite a lot.

Quote:

My perspective is that headphones do not even approach creating the sort of detail required to simulate those effects.

Spatial cues? Well with stereophonic ordinary recordings I fully agree.

With personal head (your own head) binaural recordings I think this changes quite a lot, but I agree that there are still plenty of advancement to be made. Even some generic dummy head recordings can be quite remarkable, even better than many stereo recordings played back on loudspeakers, IMHO.

Again, the recording should be fitted to the transducer device as there is no one single accurate way to reacord an acoustic event (using current equipment).

As you know, sound as a pressure even has both velocity and direction component of the wavefront, which the current method of recording (pressure sensitive microphone) doesn't fully record. It's a crude approximation of a wave event.

Quote:

Again, it is just spectral distribution of energy at my reference plane. If I measure 2 cans that have identical curves (it could happen) they will not sound identical, but they will have a very similar tonal character.

Yep, if they are both of the same type. I think the risk of getting misleading results increases quite a lot when you compare open to closed cans as this is when the resonance of the ear canal is even more important. With a (partially) blocked canal you are not measuring the canal resonance effect, which should be much greater for the closed phone than the open one. However, you will hear this resonance, even if you don't measure it.

Also, if you compare across supra-aural and circumaural (with tight fitting) you might get similar results, even when the sound characteristics are remarkably different.

In either case measurements might not correlate very well with listening impressions, IMHO.

Further, when writing my article on hearing last week it finally dawned on me that human hearing is in many ways not a very good absolute frequency analyser (even within the limitations of the critical bands and the linear errors of the ear).

It could be labelled more accurately as a combination of relative phase, intensity, frequency and duration detector.

As such, pure amplitude response measurements may not tell the whole story of what we actually hear.

For example. do you think that low Q or high Q resonances are more detrimental to sound? How about how quickly they born and die?

It's not a straightforward thing to categorise, but people almost always look for high Q peaks, when in fact the low Q resonance (with a wider frequency band and a longer duration) are usually the ones that are more audible (see F. Toole).

Quote:

I would like to continue to share my observations, but I don't want to mislead people or fuel arguments. I do want to fuel discussion, and I appreciate the feedback. I promise to keep an open mind if everybody else does

Again, I think you are already doing fine, now that I know your methodology a bit more. Of course, one can always improve upon it, but even with the 'state of the art' methodology the problem is that measurements do not always correlate with the listening impressions of even a wide panel of listeners.

This is where we must understand (and especially those of us who only see the measurements, but have not heard the phones) that map is not terrain. It is a crude approximation that might work for some purposes, but fail for others.

This was the gist of my initial message, even though it came across a bit badly and I mischaraterised your measurements. Sorry about that.

I for one would like you to offer your contributions, and I will keep my mouth shut when I don't have anything worthy of contributing myself

It's always good to see people actually doing experiments, rather than being just armchair scientists. Ahem ... :-D

I'll have to get back you on that low freq measurements if I have something to contribute. It's now almost two and my brain has started a self-initiated shutdown sequence...

Cheers,
Halcyon

 

[gerG]

Halcyon, you are a gentleman and a scholar, and I really mean that!

Thanks for the references. Time to start the search! The handbook by Borwick looks especially interesting, since I am also intrigued by speaker design.

You highlighted a very critical point:

Quote:

pure amplitude response measurements may not tell the whole story of what we actually hear.

This cannot be emphasized enough. To use even highly accurate spectral traces as the only measure of goodness is a mistake. For myself, I use them to highlight objectionable "hot spots" or odd tilts to the curves. I try to listen to phones for a while before I do the measurements, and take notes on what I am hearing. I do the response last to avoid setting mental filters. Funny device, the brain (and mine is probably funnier than most). If I see something in the trace that I didn't catch by ear, I go back and see if it is really a problem. So far things have lined up pretty well. btw, I stay away from supra-aural phones. I can't stand them!

As for designers tuning the standing waves, I discovered something interesting when I modified some Beyerdynamic DT250 the other night. I was trying to improve isolation, so I sealed up a gap between the housing and the earpads (I will document and post details). I did change the isolation quite a bit (I have measurements, of course

). Unfortunately I created a dip in the response that was not there before, and altered the midrange much more than I would have expected. Conclusion: these things are touchy! One could spend an awful lot of time just tuning these effects. If a can come up with solid response variables it might be worth setting up a DOE. Then again I might just buy some cans that isolate better to begin with. wth, I can always find room for more headphones.

I would like to read the article that you wrote. Is it published anywhere that I can get to it?


Thanks

gerG

 

[halcyon]

Thanks for the compliment.

I don't consider myself a scholar, but perhaps an acolyte

I try to learn these things, but some of them are hard.

About the article I wrote....

I'm afraid the article is in Finnish and it's quite basic in level. Namely it covers outer ear, middle ear and inner ear, plus some mid brain, thalamus and cortex basic functions (the latter parts being relatively unknown even to the real research scientists). It's not very deep as it is aimed for the layman.

It's an article in the Finnish Hifi-magazine for a seriers on the Psychology of Hearing. My co-author (who knows much more about philosophy and psychology than I do) and I want to bring to people's attention that hearing is not a known machine, it is often upredictable, it distorts in a non-linear fashion and that it can can be trained to become more responsive to minute details. And also that measurements are no subsitute for listening

Many people who are finely trained in engineering have not been taught these things (although some know much more than I could ever hope to!), so we try to balance the usually very engineering lead view in Hifi circles by point out that hearing is a perceptual and experiental phenomenon, still very much removed from measurement sciences in many ways.

If you want a good introduction to hearing, the chapter on Goldstein's 'Sensation and Perception' (latest edition, is it 6th now?) is pretty decent.

Other good sources are:

Acoustics and Psychoacoustics, 2nd. ed., David M. Howard, James Angus, Musical Technology series / Focal Press, 2001

and a bit more advanced text is:

An introduction to the Psychology of Hearing, 4th edi., Brian C.J. Moore, Academic Press, 1997.

I have some other titles on order also, but haven't received them yet. If any of them are exceptional, I'll mention them as well.

Luckily the web has also many knowledgeable sources (I'll dig up some later on, when I have time... I don't have links arranged in this field, I'm afraid).

Your headphone experiments sound very intriguing indeed. Maybe I should get a decent pair also for modifying and start measuring

One can learn only so much by reading books. Experimentation is crucial and you are doing it!

Best regards,
Halcyon

 

[gerG]

Hi halcyon

Quote:

...the article is in Finnish...

I was afraid that you were going to say that. Dang.

You are doing interesting work, and I now understand your perspective much better. I think that it is important to keep the engineers honest, to themselves especially. In my 20 years as an engineer I have been continuously amazed at how a highly qualified technical individual can interpret data in a manner that will support a preconcieved conclusion, or to tell a rosey picture so that the deadline is met. It bothers me to see this, especially when I catch myself falling into that trap.

As for the experiments, I tend to look at devices from a functional perspective. The experiments help me to understand what part does what, and why it was configured that way. It also seems to be the fastest way for me to learn. Besides that, it is jolly good fun!


Take care

gerG

 

[halcyon]

Let's get back to bass response and the original curves...

I looked around the curves you posted and what are on the Headroom site.

Some possible reasons for the difference in roll-off characteristics of the bass:

1) Head coupling. The better fit you have to the head, the stronger the bass response usually is. Also, the better the fit, the more accurately the intentional leakage of the headphone (designed into it) works as intended. Both with open headphones (on which the circumaural design matters, even though the design is open otherwise) and closed headphones (all of which have leakage designed into them)

2) Mic positioning, calibration and sensitivity (I don't know the what's the difference between the gear you've used). While the sound field in a closed headphone is pretty uniform inside the headphone cavity (for frequencies below c. 1 kHz), the mic itself can register more or less of it (depending on figure, sensitivity, calibration, etc).

3) I can't find data on whethere the Headroom figures are raw, corrected, smoothed or weighted. Perhaps some difference there? They look pretty raw, but it's hard to be sure.

4) What voltage level where you feeding the phones? I don't know what headroom used either. This really shouldn't matter that much, AFAIK, but it would be something to look into if other things are eliminated from the list of possible culprits.

As for the damping. I was under the impression that the amount of material needs to be at least 1/10th of the wavelength before the damping starts to work and 1/4th of the wavelength before it has significant effect for the first wavefront.

Clearely this would be impossible achieve in a headphone design for even lower mids.

Perhaps with a resonator some frequencies could be attenuated, but I think the only way to do it for bass is to make leaks or to use the natural roll off characteristics of the drive.

Some headphones use clamping force to control the acoustic impedance of the loaded chamber and hence the whole working impedance load of the headphone.

This is probably nothing new to you, but at least it helped me to gather my thoughts for myself.

Cheers,
Halcyon

 

[j-curve]

Go gerG! The stuff Head-Fiers are getting into lately is fantastic, sticking SPL meters into their 'phones to check sound levels or using their heads as "dummies" and generating their own response plots using capsule mic's. My experience with the latter showed up exactly the same problems that you're having, namely resonances in the top end and rolloff at the low end due to the microphone I used. One popular Philips capsule is rated down to 20 Hz but no response tolerance is quoted so the specification doesn't actually mean anything. I think condenser mic's roll off because they are ported to protect the membrane from pressure build-up inside the capsule due to weather, thermal effects, plane travel etc. You could take the risk and gum up the vent at the back of the capsule but I'm not sure this would improve the bass because the mic then becomes a closed design. A better approach is to mount the capsule in a long tube like the ones used as sensor microphones for automatic equalisation of speaker & PA systems, but this approach may not be practical for our current purposes...

Regarding the method of measurement, I agree that a head is necessary. Using a biological head, we need not even worry about the "average listener's head" if we each limit ourselves to the selfish pursuit of what sounds best to me, myself and I(!)

One point I don't agree with (as yet) is the idea of blocking the ear canal. I think it is more important to keep the total volume of air as close as possible to that of the normal listening situation, in order to load the transducer correctly. Secondly, since headphones are specifically designed to interact with ears, allowances (may) have been made in the headphone design to complement the natural function of the ear. Thirdly, if the ear canal resonated in a similar way when wearing headphones as it does without, then designers could presumably ignore the canal resonance effects. I think it's exactly because the ear canal behaves differently when wearing headphones that the 4kHz/12kHz boosts and 8kHz dip have to be recreated. The graphs of "ideal" headphones measured this way shouldn't be flat, they should look like HRTF's, right?

My experiments with binaural recordings have influenced my views on this, because I've noticed that my voice doesn't sound natural on a recording made with my ear canals blocked, whereas with only a partial blockage (due to the mic capsules in my ears but not fully sealed) my voice sounds normal on a recording. Perhaps there is another reason for this though, related to vocal sounds travelling up the Eustachian tubes.

Quote:

halcyon: With a (partially) blocked canal you are not measuring the canal resonance effect, which should be much greater for the closed phone than the open one.

It would be nice to be able to treat whatever's behind the transducer as separate from whatever's in front, influencing each other only via their effects on the transducer diaphragm, thereby allowing us to think of the diaphragm as a barrier of a kind. In practice though I'm sure you're right. [I can't hear any difference in the amount or quality of ambient sounds entering my HD580's before and after plugging them into an (idle) amp, so the diaphragm is obviously a flimsy barrier indeed]. Taking your argument to the extreme case, when you plug an Ety into your ear, what was an open ear canal becomes closed and the fundamental resonance (of the canal, not the transducer) should roughly double in fequency (from a wavelength 4X canal length (about 4kHz) to one only twice the length (about 8kHz)).

Quote:

gerG: I have concluded that AT cans do not appeal to my ears. The W2002 have a very emphasized upper midrange. The AD10 have an emphasized upper mid, with an audible peak around 4000.

What you hear agrees with what I heard, but note that these are both Type A driver designs. The W100 and the W10/W11 series phones use a different driver design which doesn't suffer from that excessively forward presentation (although they are still quite bright on the highs).

 

[gerG]

Hi j-curve, and welcome to team EG (Extreme Geek)!

As luck would have it (or maybe halve it) I have a new mic on the way, and it exactly the sort of calibration mic that you described. More fun to be had!

As far as the ear canal blocking goes, I did check for differences, and I found them. Problem is that I am using pink noise for the test, and that is a lot of energy on my poor old TM. I have sensitive ears in that I can't handle even moderately loud noises for very long. I am sure that this is partly psychological from being so careful to protect my ears all of these years, but I just did not like that racket in my ear. I got a bit worried about S/N when I lowered the din to reasonable levels, so I thought that the plug was a good compromise. I intend to check for differences when I get a better handle on my measurement system.

As far as the ATs go, I think that a very large part of the problem was driver positioning. When I swapped out the pads on the AD10s so that the drivers no longer touched my ears things changed dramatically, and they actually became pleasant to listen to. I can't help but think that the design intent and the final product differ substantially. I have seen this happen before due to a supplier change, or even to a last minute decision made by someone not aware of the specific design sensitivities.

A question (AT guys close your eyes and cover your wallets) do you know if it is possible to buy the different drivers on a piece part basis? It could be fun to swap them out and see what changes!

On the headroom plots, I am still puzzling over that one. I will chase through my system to see what I turn up. I told Tyll that if it weren't winter I would already be on my way up there just to see that great test setup that he has put together. The suspicious thing was the curve that I got when I subtracted my data from his. Very close to the A weighting curve used in noise analysis. It is an important issue, since his graphs show headphones reaching all the way down to 20 hz, while mine are showing levels that are down 20 db or more at that limit. fwiw my mic is optimized for bass response, but I am not using it with the intended electronics. This may be part of the problem. If it is, I will track down the error.

Any thoughts on the rolloff slope that one would expect in headphones at the lower limit? My first thought on an open circum-aural design is a second order slope, since it is a single driver with a single cavity (although the ear is inside the cavity). Closed cans could be almost anything, depending on what sort of resonance tuning is used (e.g. I could tune the inner cavity with a port to ambient, same with outer cavity, and maybe one between the cavities just for good measure). I would expect higher order rolloffs, although that could get clouded by a tuning which gives a milder slope before the cliff is reached.


Later

gerG

 

[gerG]

I just checked your driver link. Good job!

I am suspicious that the AD10 may have had type C drivers. I base this on my measurement of the voice coil, which came out very close to 20mm. I didn't get in there with calipers (pointy objects near delicate parts makes me cringe) but they were well under 25mm. Most interesting.

gerG

 

[j-curve]

Quote:

gerG: I have sensitive ears in that I can't handle even moderately loud noises for very long.

...and pink noise is a killer, I can vouch for that! If the S/N ratio is a problem then I guess the open ear canal is a bit reckless.
Quote:

gerG: do you know if it is possible to buy the different drivers on a piece part basis?

I suspect only A-T service centres could do that, but I'll look into it. Expect a delay of a couple of weeks...
Quote:

gerG: I told Tyll that if it weren't winter I would already be on my way up there just to see that great test setup that he has put together.

Call me a cynic but I was wondering if they merely hired all that expensive gear, since there are a few glaring omissions and there doesn't seem to be ongoing testing.
Quote:

gerG: I am suspicious that the AD10 may have had type C drivers. I base this on my measurement of the voice coil, which came out very close to 20mm.

Damn, I hope you are wrong about that! One possibility is a scale error in my calculations, since I was basing my estimates on a common 53mm total diaphragm dimension. It's possible that my voice coil numbers are all too large but hopefully still consistent within the Types A, B & C. Worth looking into...

 

[gerG]

Hi J.

I am anxious to see what you find out on the drivers. The OD of the driver was less than 53mm, but I was measuring to the edge of the diaphragm, while most official driver dimensions probably include the mounting flange.

Trust me, Tyll is really getting into the measurement thing. He bought all of that gear, and he is coming up to speed on all of the functions. Expect more discussion over in the headroom lab area. He is also looking at transient response. He has already posted some square waves at 100 hz which tell a lot about diaphragm damping. He is also exploring the effects of amplifier impedence. Cool stuff.

It is interesting to imagine a sufficiently large number of well informed consumers in this hobby. We could actually have some impact on the development process by manufacturers. Maybe we will get there.


gerG

 

[Number9]

I find it very interesting that all of the graphs are simaler to the diffuse field equalization curve I found at Headwize. (found mid-way on the page).

http://www.headwize.com/articles/judging_art.htm

This must have something to do with explaining their shape.

 

[gerG]

Oooo, thanks. That was a nice article. cmoy continues to impress me. There is a lot of fuel for discussion in there.

I am curious about the equalization of headphones now. I think that I will take some speaker responses with my dummy head in the circuit. That should give me the HRTF that I am looking for.

You are right about the similarity of the curves. Not sure what it means, but worth pondering.


Think think think think...
t
h
u
n
k
!


gerG

 

[j-curve]

gerG, I just read your AD10 review in pdf format. Top stuff! I was very interested to see the two different curves according to output impedance of the amplifier. One suggestion for the graphs - how about matching levels on the basis of "average output between 100Hz-2kHz" instead of the 100Hz level? I realise that adds a wishy-washy factor to your graphs but when you compare two sets of cans you start by matching the volume level which is kind of the same. I think that would show both the bass hump due to higher output impedance and the relative brightness of the Beyer.

Also, re: the A-T voice-coil diameters... You've got 0.9" which is 23mm. Also you give the diaphragm diameter as 1.8" which is 46mm. If A-T's diagrams correspond to this and the total diameter is 46mm, not 53mm as I assumed, then my estimates are all about 13% high(!!)

So Type A (was 25mm) should be about 22mm (which is close enough to your 23mm).
Type B (was 29mm) should be 25mm.
Type C (was 19mm) should be 16mm.
I think I'll edit the Drive Units thread. Thanks for the info.

 

[gerG]

J, did I really say .9"? I thought it was .8, which means that I can't trust my memory anymore. I will try to get more accurate measurements. It is a bit tough sighting through that perf plate crosseyed trying to avoid paralax. This may require getting a LASER or 2 involved

As far as the graphs go, the traces can be shifted for a relative level change without redoing the test (within a reasonable variation). I can mail you the spreadsheet if you want to play. The trouble is picking what to match. I think that everybody sets levels a little differently. As for the differences due to impedance, I am a bit skeptical about those values. There are a couple of extra potential sources for error with those big DT831 cans, and I want to do a more controlled experiment to chase them out. I am also puzzled by the slope change in the Beyers. More riddles to unravel!

btw, where did you get all of that info on the AT cans? Very nice!

gerG

 

[j-curve]

Hi gerG - just a couple of things:-

1. I think I know why your bass response is dropping off so sharply. Did you use the microphone input to your soundcard? I just found out (the hard way) that my soundcard mic input is only mono and has a pretty savage high-pass filter in it, chopping everything below 300Hz. Obviously intended only for voice. My line input, on the other hand, is stereo without the filter.

2. Retailers are telling me that you can't buy A-T drivers separately, even for repairs. You have to send broken 'phones in for service.