• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). There are daily reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Headphone Measurements - HD650

gabo4au

Member
Joined
Apr 18, 2022
Messages
34
Likes
20
So I had a few questions in my mind about headphones, frequency responses, and corrections.

How consistent are headphone frequency responses, not only among different sets of headphones, but also over the years of production? And also as they age, so not only the question of "does a headphone manufactured 5 years ago measure the same as one made today?," but also the question of "did the 5 year old model change over time?"

If there are meaningful differences in any of this, then it would seem as though the headphone corrections applied through things like sonarworks, etc, are not very good unless you pay for the custom service to measure and correct your specific pair. And even then, if they change over time maybe not so good.

I've seen multiple measurements from some headphones and my understanding is that most of these "correction applications" use an "averaged" response from many units. But if there is a meaningful spread of differences maybe this just means that the average doesn't fit many sets very well.

So to satisfy my curiosity, I decided to try measuring my own Sennheiser HD650 headphones. Instead of spending quite a bit of money buying some sort of testing apparatus, like the miniDSP ears ($249 before shipping/taxes), I decided to try making my own. After all, I already have a seriously good measurement mic, the earthworks M23R, along with a nice laptop, and an Edirol UA-25ex interface for doing room testing. And a fair amount of experience running REW and a few other tools. So why not give it a try.

Somewhere over the years, I recalled seeing a person build a simple wooden device for headphone testing, which he compared to the "EARS" jig and actually said it was as repeatable and for the most part just as good. That was easy enough to build in my shop, so here's what I fashioned together.

Headphone-Jig1.jpg Headphone-Jig2.jpg

I did not expect this jig to be perfect, to be honest I didn't really know what to expect. So when I took some measurements, I was a bit surprised. It actually looks pretty good. I tried it with the mic on either side of the jig, with the mic pushed way in, and with the mic pulled just to the edge. I also cut the hole on the other side a bit larger, and tried all the combinations on that side as well. And I put felt on the side you see in the pics, and there is no felt on the other side. The measurements from all these different positions, angles, with/without felt, we are very consistent. When you pulled the mic out a bit, the curve was the same but the level was not as hot. The side with no felt was basically the same curve but with a few more wobbles above 10khz, but even those tracked pretty close.

Here are the left/right results from my HD650s. Be aware that these are not "calibrated" SPL results, so the SPL scale on the left is not accurate. I may try calibrating them at some point, but that will require me inserting my SPL meter in the hole the same depth as the mic, which might be tricky, but I haven't tried it yet.

1652020094332.png


However, the problem with the graphs, they are indeed quite a bit different from most of the published graphs of HD650s. My graphs show the main "spike" at around 5.2Khz as is shown. But most of the published reviews of these headphones shows the main "spike" to be around 3.5Khz. Here's Amir's review from here on ASR.

Sennheiser HD650 Measurements Frequency Response.png


That chart indeed shows a spike around 5.2K, but it shows a bigger spike at somewhere around 3.5K that mine does not show. Mine also shows a low point around 2.1K that Amir's does not.

So now the questions are:

1. Is there something about my jig that is causing some issue around 2k-3.5K? The response below 1.5K and above 5K is very similar.
2. Or, is the frequency response of these two units actually that different?
3. My HD650s are about 3 years old and have been used significantly, almost every day, in a studio mixing environment. So they are well broken in. Could this be the difference between a "well broken in" pair vs a "relatively new pair?" I don't know how old Amir's were, but I would assume they were newly acquired and sent to him for review.


My original thought was there was some sort of reflection in the 2K-4K region caused by the wood and/or the angle of the mic. That was my initial reason for adding the felt. But that didn't really change things at all. Also if it were some interaction between the mic and the opening, I would at least expect some type of change in response if I moved the mic a bit, in/out, move it to the opposite side, use the larger hole on the other side, etc. But none of that changes the fundamental shape of this curve. Also, doing some math, even at 4khz a 1/4 wavelength is still more than 21mm, which is still far greater than anything in that chamber. So I can't see any sort of resonance/reflection being in play.

The only thing I can come up with is that my headphones simply do not have the same response from about 2k-3.5k. But maybe I'm missing something.

What say all the experts here?

Thanks, gabo
 

MayaTlab

Addicted to Fun and Learning
Joined
Aug 15, 2020
Messages
541
Likes
745
1. Is there something about my jig that is causing some issue around 2k-3.5K? The response below 1.5K and above 5K is very similar.

I think that this post from Oratory may provide some pointers :
https://www.reddit.com/r/DIEMs/comments/75pfaq/_/do7wka1
Basically, for a start you're lacking the ear canal and the pinna. This affects the absolute values you get (ie at frequency Xa headphones A are Y dB louder than at frequency Xb), but it can also affect the relative values between headphones (ie at frequency X headphones A are Y dB louder than headphones B). How the absolute and relative values are affected is a complex subject for which I am not superbly qualified to comment on :D.

It could also be the case that the rest of the rig affects the response, such as the materials used, the assembly, the width and orientation of the plates, etc - but that's a secondary issue here in regards to your own concerns.
 
OP
G

gabo4au

Member
Joined
Apr 18, 2022
Messages
34
Likes
20
I think that this post from Oratory may provide some pointers :
https://www.reddit.com/r/DIEMs/comments/75pfaq/_/do7wka1
Basically, for a start you're lacking the ear canal and the pinna. This affects the absolute values you get (ie at frequency Xa headphones A are Y dB louder than at frequency Xb), but it can also affect the relative values between headphones (ie at frequency X headphones A are Y dB louder than headphones B). How the absolute and relative values are affected is a complex subject for which I am not superbly qualified to comment on :D.

It could also be the case that the rest of the rig affects the response, such as the materials used, the assembly, the width and orientation of the plates, etc - but that's a secondary issue here in regards to your own concerns.

Correct, clearly the "big boys" have a lot more realistic rig. However, what I've failed to find thus far are what the effects of the things you list, realistic ear, ear canal, impedance of the ear drum, etc. make?

Does a realistic ear shape, canal, and material effect all frequencies? Probably not, so what frequencies and how much? If the results are +-1db in a few frequencies, then it's not nearly as big as the errors from my specific headphones to the averaged curves. In that case, I would still be better of using my tested curves and equalizing to that vs equalizing to a published average curve.

However, if the differences are +-10db and/or variable and all over the place, then maybe it's better to use the published average curves and correct for that.

From my tests and just from the science, it seems that below some frequency, maybe 1.5K, my tests are pretty accurate. In my case, above 10K is probably not worth doing any EQ because I'm an old guy and have limited hearing up there anyway. Also from my tests, it looks like from some frequency, maybe 4K or 5K up to 10K it looks reasonably good. So if I can gain some level of confidence in the 1.5K to 4K range, or tweak some things to make those measurements more accurate, then maybe I have a method to EQ my headphones that is better than "off the shelf" EQ based on average curves. And if not, I learn some things! :)
 

MayaTlab

Addicted to Fun and Learning
Joined
Aug 15, 2020
Messages
541
Likes
745
From my tests and just from the science, it seems that below some frequency, maybe 1.5K, my tests are pretty accurate.

The effect of the ear canal may start sooner than that. So for absolute values I'd be a little bit less optimistic than that.

That illustration is circulating around the web, but it might be an over-simplification :
1631040777885.png


For relative values (ie the difference between headphones, "headphones A are 2.7dB louder at 3250Hz than headphones B"), I've spent quite a bit of time measuring headphones in situ, on my own head, with in-ear mics, and I think I can get repeated divergences between blocked and open ear canal measurements starting around 1.5kHz, but so far I have seen no indication that the spread of the "different difference", while quite likely to be above minimum thresholds of audible difference, is huge, at least for the 1.5-7kHz range (maybe around 2dB of potential error between two headphones).
Below 1.5kHz however I have seen no divergence of any significance.

In regards to the influence of the pinna, as I'd rather not chop off my own I can't isolate its effect.
 
OP
G

gabo4au

Member
Joined
Apr 18, 2022
Messages
34
Likes
20
The effect of the ear canal may start sooner than that. So for absolute values I'd be a little bit less optimistic than that.

That illustration is circulating around the web, but it might be an over-simplification :
View attachment 205448

For relative values (ie the difference between headphones, "headphones A are 2.7dB louder at 3250Hz than headphones B"), I've spent quite a bit of time measuring headphones in situ, on my own head, with in-ear mics, and I think I can get repeated divergences between blocked and open ear canal measurements starting around 1.5kHz, but so far I have seen no indication that the spread of the "different difference", while quite likely to be above minimum thresholds of audible difference, is huge, at least for the 1.5-7kHz range (maybe around 2dB of potential error between two headphones).
Below 1.5kHz however I have seen no divergence of any significance.

In regards to the influence of the pinna, as I'd rather not chop off my own I can't isolate its effect.

Good points, thanks for that! For headphone's, can you ignore the torso/neck and spherical head influence? Seems like the headphones, being clamped against your head (talking over ear headphones), removes those influences.

I was on a similar thought line and it turns out much of the difference is in the impedance of the ear canal and other mechanisms, which for headphone testing has been defined in IEC711. If you search around, you'll find that an IEC711 has an internal volume of about 2 cubic centimeters. So basically all that works much like a port on a speaker.

So being the inquisitive DIY hack that I am :) I looked around the shop and came up with a piece of pvc tubing, pretty much a similar feel to human skin, cut to proper length to produce about a 2 cubic cm volume. So here's a pic of that modification to my jig. You can see the clear tubing with the mic inserted to the proper depth to create the correct volume.

Headphone-Jig4.jpg

And that produces much better results.

1652036656533.png



So that gives me the peaks right about the spots I would expect, although they are a bit rounded. But much progress.

From your excellent graph, that looks like it did a reasonable job of simulating the ear canal and ear drum. Also from your graph, it looks like I need to simulate the "concha" a bit to boost that 5.5K region.

Of course all this is one way to do it, but another way might be to go back to my original graphs and add the curves from your graph as a correction. That might get a more repeatable and better simulation.

gabo
 
Last edited:

MayaTlab

Addicted to Fun and Learning
Joined
Aug 15, 2020
Messages
541
Likes
745
That's interesting to note. Do you have a wider range of headphones to test on that fixture ?

I was on a similar thought line and it turns out much of the difference is in the impedance of the ear canal and other mechanisms, which for headphone testing has been defined in IEC711. If you search around, you'll find that an IEC711 has an internal volume of about 2 cubic centimeters. So basically all that works much like a port on a speaker.

This is best left to the experts (not me here), but my understanding is that 711 couplers are an attempt quite a bit more complex than just a mere question of volume size to approximate the ear canal.
 
OP
G

gabo4au

Member
Joined
Apr 18, 2022
Messages
34
Likes
20
That's interesting to note. Do you have a wider range of headphones to test on that fixture ?



This is best left to the experts (not me here), but my understanding is that 711 couplers are an attempt quite a bit more complex than just a mere question of volume size to approximate the ear canal.

Yes, but your friendly DIY hacker can only try one simple thing at a time :) I suspect a more detailed design would potentially sharpen up that boost and not make it so rounded. That's why maybe just adding in the boosts you've shown in your graph might actually be more accurate, not sure. I do know that with the jig before the added piece of tubing was very repeatable. With the tube, it's a bit fussy if you don't get the mic in just the right depth and the tube sealed correctly on the jig. My next attempt might be fiddling with adding in the graph results. I can do that in REW by adding it into the soundcard calibration curve.

I do have several other headphones I can test, a pair of Beyerdynamic DT990s, some Sony MDRV6's, and some Senn HD280s.

gabo


UPDATE: Tried a quick and dirty EQ add in rePhase to see if adding in the responses from your graph worked. But for some reason that didn't give me a graph that looked right. Although I'll work on it some more, maybe I have something not right.
 
Last edited:

oratory1990

Member
Reviewer
Joined
Mar 1, 2020
Messages
38
Likes
141
That's interesting to note. Do you have a wider range of headphones to test on that fixture ?



This is best left to the experts (not me here), but my understanding is that 711 couplers are an attempt quite a bit more complex than just a mere question of volume size to approximate the ear canal.
That's correct - the volume of air inside the 60318-4 ("711") coupler is frequency dependent, as it's divided into three parts, a main tube and two side-volumes that are connected with thin slits, which will gradually close off towards higher frequencies.
The idea being to simulate the frequency-dependency of the acoustic impedance of human ear canals.

There are other couplers that have a constant volume (most notably the 2cc coupler specified in IEC60318-5), but these do not simulate the acoustic impedance as accurately, obviously.

To add something to the thread:
Here's a graph showing the average of all HD650s that I measured so far, including 90% confidence intervals.
Sennheiser HD650.png
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
11,682
Likes
26,004
Location
The Neitherlands
Here's a graph showing the average of all HD650s that I measured so far, including 90% confidence intervals.

Can I ask how many HD650's these were and were they new or older (pad condition I mean) and, roughly, between which years of manufacture ?
 
OP
G

gabo4au

Member
Joined
Apr 18, 2022
Messages
34
Likes
20
That's correct - the volume of air inside the 60318-4 ("711") coupler is frequency dependent, as it's divided into three parts, a main tube and two side-volumes that are connected with thin slits, which will gradually close off towards higher frequencies.
The idea being to simulate the frequency-dependency of the acoustic impedance of human ear canals.

There are other couplers that have a constant volume (most notably the 2cc coupler specified in IEC60318-5), but these do not simulate the acoustic impedance as accurately, obviously.

To add something to the thread:
Here's a graph showing the average of all HD650s that I measured so far, including 90% confidence intervals.View attachment 205535

Thanks for that, and just FYI, I've been using an IR based on YOUR measurements in my DAW for the HD650, so thanks for all your work on these and many other headphones. I'm actually amazed at how close my measurement is, given that I clearly am missing a lot in my model. I think it would fit right into that grouping of measurements. I thought, based on comparing with Amir's measurement, that I was missing something in the 5Khz region maybe due to lack of a concha model. But looking at y our graph, I'm not sure if that's true. Still could be, but it's not as clear. Also, my boost at 3.2K is about 14db above the low area down around 700hz, which would put it in the top range of your measurements. So that rise may be a bit too high. I need to put my other headphones on there today and compare.

Solderdude had a good question as well, however, I'm not as concerned with pad condition. It's easy to put pressure on the pads and understand the results of poor fit or worn pads. More pressure/better seal results in a fairly linear rise in the low frequencies from about 300hz down. Which of course you can also hear if you press on your headphones. But understanding how many you have tested, their age and manufacturing info would be interesting. Or even just your observations of how those things affect the results.

I wouldn't mind purchasing an IEC711 device, but when I've looked at them it's difficult for me to understand what I'm purchasing. The only place I've found to buy them is aliexpress and they range in price from $14 to $168. It isn't very clear on aliexpress what you're actually buying some times. It seems like a small plastic piece would be relatively cheap, I don't need a stainless steel machined piece that is very expensive. It's all just a fun thing to play with for me, not a career! I'm sure someone with a 3D printer could make one relatively easily.
 

oratory1990

Member
Reviewer
Joined
Mar 1, 2020
Messages
38
Likes
141
Can I ask how many HD650's these were and were they new or older (pad condition I mean) and, roughly, between which years of manufacture ?
20 earcups in total (11 headphones, as one two of them only had one earcup working).
I did not keep track of manufacturing years / serial number, bot anything from "very worn earpads" to "fresh out of the box".
I wouldn't mind purchasing an IEC711 device, but when I've looked at them it's difficult for me to understand what I'm purchasing. The only place I've found to buy them is aliexpress and they range in price from $14 to $168. It isn't very clear on aliexpress what you're actually buying some times. It seems like a small plastic piece would be relatively cheap, I don't need a stainless steel machined piece that is very expensive. It's all just a fun thing to play with for me, not a career! I'm sure someone with a 3D printer could make one relatively easily.
Read up on IEC60318-4 if you want to learn more about the ear simulator.
The aliexpress knockoff couplers are... questionable. Some fall roughly within IEC specifications, others are blatantly out of spec. There is no way to know, as there is no quality control by the manufacturer, typically. (You're not getting a measurement chart proving that the acoustic impedance of your specific unit falls within specs, as is normally common when buying measurement equipment)

One thing is for certain though, if they're made from plastic then you can trust them about as far as you can throw them. I don't see any way these could be produced precise enough without using precision-milled aluminium to be honest. You can't really 3D print these, the tolerances would be much too large.
Sadly this also inherently means they'll be rather expensive. You're quickly entering low 4-digits for a coupler from GRAS or B&K.
 
OP
G

gabo4au

Member
Joined
Apr 18, 2022
Messages
34
Likes
20
20 earcups in total (11 headphones, as one two of them only had one earcup working).
I did not keep track of manufacturing years / serial number, bot anything from "very worn earpads" to "fresh out of the box".

Read up on IEC60318-4 if you want to learn more about the ear simulator.
The aliexpress knockoff couplers are... questionable. Some fall roughly within IEC specifications, others are blatantly out of spec. There is no way to know, as there is no quality control by the manufacturer, typically. (You're not getting a measurement chart proving that the acoustic impedance of your specific unit falls within specs, as is normally common when buying measurement equipment)

One thing is for certain though, if they're made from plastic then you can trust them about as far as you can throw them. I don't see any way these could be produced precise enough without using precision-milled aluminium to be honest. You can't really 3D print these, the tolerances would be much too large.
Sadly this also inherently means they'll be rather expensive. You're quickly entering low 4-digits for a coupler from GRAS or B&K.

Thanks for that info.

Hmm, since the ear is not a precision-milled piece of aluminum, and indeed every one is different, I'm not convinced that something representative couldn't be made from something relatively inexpensive. Certainly if you want to create measurements across the industry that are the same and correlate exactly, then everyone needs to be using the "exact" same piece of gear. Which is what you're doing so the precision is much more important for you, down to the 0.1db level I would guess. But that's all just thinking logically and speaking without any experience or knowledge working with these things.

But I totally agree, and suspected, that many of the low cost things off aliexpress are way out of spec. So I'll stay away from them as what I have now may actually be as good as most of those. I've done DIY molds to produce custom molded IEMs myself, so I need to look around and see what pieces I might have from my own ears. After all, I'm more interested in modeling "MY" ears anyway. Not sure I can get anything that is within even +-2db for my own ears based on what I have learned. But if I could, it would probably be better than using the averaged results as the cone of error in your graph looks to be about +-2db in the critical mid-frequencies.
 

MayaTlab

Addicted to Fun and Learning
Joined
Aug 15, 2020
Messages
541
Likes
745
I'm more interested in modeling "MY" ears anyway.

If so, while measuring headphones in situ with in ear mics is not straightforward past around 1kHz or so, I still think that it might be easier to reach actionable results with that approach rather than trying to model your own ears on a DIY ear simulator.
 
OP
G

gabo4au

Member
Joined
Apr 18, 2022
Messages
34
Likes
20
If so, while measuring headphones in situ with in ear mics is not straightforward past around 1kHz or so, I still think that it might be easier to reach actionable results with that approach rather than trying to model your own ears on a DIY ear simulator.

I have no idea how that is done or what is required. Can you point me to some information about that?
 

MayaTlab

Addicted to Fun and Learning
Joined
Aug 15, 2020
Messages
541
Likes
745
I have no idea how that is done or what is required. Can you point me to some information about that?

That is an excellent question to which I have been trying to find an answer for a bit less than a year :D. I don't think that I am in any position to be prescriptive about what to do or make any recommandation other than that you'll have to experiment and find out what works best for you.

That said you may want to start by reading a few articles from Hammershøi and Møller - a lot of them are available without a paywall on the web, such as this one : https://vbn.aau.dk/ws/portalfiles/portal/227875204/1995_M_ller_et_al_AES_Journal_b.pdf

As far as the hardware is concerned, this is a bit of an over-simplification, but I think that you can categorise the different sort of in-ear microphones as follow :
- in concha microphones, with the ear canal either open or blocked (with ear plugs)
- closed ear canal entrance microphones
- open ear canal entrance microphones
- probe tubes (inserted inside the ear canal)

In order of, IMO, easiest to approach to hardest (the latter in particular). All of them with pluses and minuses in my opinion. What increased my confidence in the results I obtain is to use them all and compare their results, and I'll often use one type in particular depending on what exactly I want to test.

If like me you're not a soldering expert, Sound Professional makes a large variety of "binaural" microphones models that you can then mod to suit your objectives, but there are countless other binaural mics options available. If you're handy with soldering and cabling you can buy for much cheaper the individual parts.

In threads such as these you'll find several examples of various DIY solutions :
As this is an old practice you'll find plenty of other threads on the subject.

Feel free to PM me.
 
Top Bottom