[inarc]

Quote:

Originally Posted by Tyll Hertsens /img/forum/go_quote.gif That the DF curve that comes with the head is likely not the best background subtraction to make because ... well because there's so much darn controversy about it. However, taking the data on a head is probably a good place to start since so much of what these curves look like are due to the very specific shape of an ear in the earpiece.

Have you considered using the independent-of-direction equalization?
I like where this thread is going!

 

[Tyll Hertsens]

I have. Theoretically it wouldn't be the right thing, I'm thinking, but since theoretically no one is on the same page it's probably worth a try.

Thanks for reminding me. I was actually going to load up those curve a week ago, but I only have them on a 5" floppy. Gotta go find someone with a floppy drive to extract the files.

No need for offers, I'm sure HeadRoom has an old machine I can do it on if I don't.

 

[inarc]

If the equalization model is really such a controversial topic, I'd just provide graphs with all three of them (FF, DF, ID). Once you have the raw data and the equalization curves, applying them isn't much of a hassle, is it?

 

[Tyll Hertsens]

Nope.

People might get a little frustrated over the next few months as the important bit is really the issue of gathering the raw data.

We're going to have a lot of fun doing background subtractions of various sorts; heck, by the end of all this you may be able to upload your personal guess as to what flat is and the everything displays according to your ears.

Maybe by the end of all this I won't be measuring headphones at CanJam, I'll be measuring ear canals so that you'll have a personal measured response.

I do have a capillary mic to stick up your ear canal. Don't worry ... I'll be gentle.

But as fun as those things are to think about, at the moment it's all about developing the proper raw data set and test procedure.

At this red hot moment, I'm running a bunch of previously measured cans through the HeadRoom measurement procedure so that we can verify that the HeadRoom can measure new headphones that come in. (I still contract headphone measurement to them as a consultant, so to get that up and running is important.)

I'll post the PDFs of the measurements over the weekend. I'll be measuring: 7000, 5000, 650, 600, 800, 701, 702, 880 32 and 250 ohm, and T1.

The next Hooby Lab developement will be measuring maybe 4 different headphones each with five different positionings on the head, and then averaging the measurements. I'll be comparing that to a single measurement with headphones very carefully positioned then heavy smoothed. I've got a method for positioning the headphones using the oscilloscope view of square waves playing through the headphones.

Nailing down the best frequency response data gathering technique is critical.

 

[xnor]

Quote:

Originally Posted by inarc /img/forum/go_quote.gif If the equalization model is really such a controversial topic, I'd just provide graphs with all three of them (FF, DF, ID). Once you have the raw data and the equalization curves, applying them isn't much of a hassle, is it?

And ignore the artificial head and ears?

Quote:

Originally Posted by Tyll Hertsens /img/forum/go_quote.gif I do have a capillary mic to stick up your ear canal. Don't worry ... I'll be gentle.

Tyll, do you have access to an anechoic chamber? To make the most out of the data I don't think you'll get around measuring that head's response (for FF you'd just need a single "flat" speaker 3m in front of it, right?) and calculate the DF curve from there.

edit:
Still, I'm wondering why not just take the head into a sufficiently treated (to get rid of greater room reflections, deep nulls ..) room and measure the impulse response / FR with "flat" monitors in an equilateral triangle - a typical speaker / listening setup.
Of course this brings up loads of other problems, like different responses depending on the roome etc. but..
Speakers are what music is made for and you can literally see that (stereo imaging). So I think it should be worth a try to do this, without worrying about stuff like FF/DF/What.

 

[inarc]

Quote:

Originally Posted by Tyll Hertsens /img/forum/go_quote.gif I'll post the PDFs of the measurements over the weekend. I'll be measuring: 7000, 5000, 650, 600, 800, 701, 702, 880 32 and 250 ohm, and T1.

That'll be great! Could also please test the AKG K601? I ask because the K601 appears to have the flattest (DF) frequency response of all the HeadRoom headphones.

Quote:

Originally Posted by xnor /img/forum/go_quote.gif And ignore the artificial head and ears?

I don't get what you mean by this. These equalization curves are provided by Head Acoustics, the manufacturer of the dummy head. If the dummy head is properly calibrated, which seems to be the case, the equalization models will also fit.

 

[Tyll Hertsens]

Quote:

Originally Posted by xnor /img/forum/go_quote.gif So I think it should be worth a try to do this, without worrying about stuff like FF/DF/What.

Like I said, at the moment I have to stick to basics and gather data.


Your idea has merit though, I've though the same thing myself before.
I'd love to get my hands on the 30deg off axis compensation curve.

One step at a time though.

Told ya.

 

[Tyll Hertsens]

Quote:

Originally Posted by inarc /img/forum/go_quote.gif That'll be great! Could also please test the AKG K601? I ask because the K601 appears to have the flattest (DF) frequency response of all the HeadRoom headphones.

Wow, I've never noticed that. Thanks for pointing that out.

Unfortunately, that's all I have in the box that I borrowed from HeadRoom.

Once again,

.

 

[SP Wild]

This is gonna be a cool project.

Looking at the HD800 measured data - a few members and I seem to agree more with the uncompensated data than the dummey head compensation. What I hear with the HD800 is a smooth and creamy midrange, a dip in bass response and an extended treble range - whereas the compensated data suggests otherwise. I would love to see the uncompensated data of other popular phones.

Even if for arguments sake my ears do not follow a typical compensation curve and are infact different to the majority and in fact boosts all treble response - then this would also manifest itself in everything I hear and my reality is a trebly one. I assure you that I feel that is not the case. My subjective conclusion then creates a negative response from HD800 owners that are happy with it's sound. Any way to correlate a headphone to reality more so than an engineered representaion of reality?

Perhaps one could 1. record reality samples 2. measure the recorded response - then 3. play through the headphones 4. record the output of the headphone 5. measure the recorded headphone output.

1 should equal 4 for a perfect headphone (dependent on perfect recording techniques - another can of worms) as evidenced by the equal value of 2 and 5 - of course this will never happen.
So one applies compensation lines to make 5 equal to 2 - This "compensation line" becomes the primary focus as it tells us how far removed this phone is from reality.

This is also problematic, I know - but am I way off-line?

 

[xnor]

I've created a little overlay of the head DF response (blue) and the response you measured with the speaker in The Box (yellow).

*removed*

I know that The Box isn't a perfect DF, maybe it's far from it (especially on the low end), but the overlay shows at least some similarities.

edit: Removed image since the curve wasn't aligned properly.

 

[xnor]

Quote:

Originally Posted by SP Wild /img/forum/go_quote.gif Perhaps one could 1. record reality samples 2. measure the recorded response - then 3. play through the headphones 4. record the output of the headphone 5. measure the recorded headphone output.

That wouldn't be different from what I wrote, would it?
It doesn't matter which sample you take, usually it's a sine sweep or pink noise that covers 20 - 20 kHz. It also shouldn't matter if the source is a speaker or an instrument.
What matters is the room, the position of the head and source ... => FF/DF standards define this.

 

[Tyll Hertsens]

The problem with trying to get too accurate, IMHO, is that as you get a more accurate and detailed measurement from the head, you also get more data that has to do with the individualized response of the head that may be substantially different than the your individual response. So the detail is more about the head than the general characteristics of the headphone.

What I really want in the end is a fairly simple looking curve which accurately lets you see whether it's got more or less energy in fairly wide bands --- More high treble, less mid-bass, etc. I don't think the FR data needs to be more useful than that because I don't think the FR data can be more useful than that.

Kind of rambling here, the good news is that a careful and plodding approach should let us just follow the data where it leads us.

I'm gonna get those cans measured today, then I'm gonna start working on getting those series of measurements with 5 replacements of the headphones. I'm also going to get the ID background curve for the head into the system. My goal is to get a raw data measurement that's an average of the 5 placements (which should average out some noise in the highs) and then a look at the DF and ID compensation results with various amounts of smoothing.

That'll take a few days cuz I have to construct a pretty complicated test sequence that pokes all the data into spreadsheets to do the math. Prolly by mid next week.

 

[SP Wild]

Quote:

Originally Posted by xnor /img/forum/go_quote.gif That wouldn't be different from what I wrote, would it? It doesn't matter which sample you take, usually it's a sine sweep or pink noise that covers 20 - 20 kHz. It also shouldn't matter if the source is a speaker or an instrument. What matters is the room, the position of the head and source ... => FF/DF standards define this.

With the help of Tylls measurements for dummies courses I hope I too will be able to understand and contribute to this thread without looking like a tool!

 

[xnor]

I'm sorry, I just noticed that the head's curve in the overlay is totally off. No real similarities after all. :/
I will remove the image to avoid confusion.


Quote:

Originally Posted by Tyll Hertsens /img/forum/go_quote.gif The problem with trying to get too accurate, IMHO, is that as you get a more accurate and detailed measurement from the head, you also get more data that has to do with the individualized response of the head that may be substantially different than the your individual response. So the detail is more about the head than the general characteristics of the headphone.

I'm not so sure about this. I assume that the head is built with an average characteristic, probably from doing tests with hundreds of people, in mind.
Correcting the raw measurement data coming from the head with the smoothed correction curves from the manufacturer would just decrease accuracy of the results, imho.

Quote:

What I really want in the end is a fairly simple looking curve which accurately lets you see whether it's got more or less energy in fairly wide bands --- More high treble, less mid-bass, etc. I don't think the FR data needs to be more useful than that because I don't think the FR data can be more useful than that.

Noooo, (too much) smoothing is evil!

Imo, not only does it get rid of the spikes that could be accounted for driver resonances (admittedly CSD would be more useful here) or construction imperfections ... but the smoothing could make any headphone look good. (and maybe sell well?

That's not what this should be about, right?)

Looking forward to those measurements. Thank you for doing this.

 

[Tyll Hertsens]

Let's just say that I agree that intemperate smoothing is evil.

I'm hoping that just doing 5 measurements with slight movements in between and averaging the five (which, after all, is in the IEC spec) will provide some smoothing all on it's own.

Also, looking at the "independent of direction" curve might get rid of the large hump in the lows that are seen on the measurements seen on the FRs after compensation.

I guess I'll say again that this slow plodding approach will hopefully wring things out bit by bit so that we don't have to bite too big a chunk off at any one time.

Anyway, next step is doing multiple position measurements with an average, and we'll see what that looks like.