[Tyll Hertsens]

I'm sick and tired of getting errors or not having the cans I want on our measurement site.

I know, however, that the fix is quite a bit of work, and if we're going to do quite a bit of work on our measurements, I'd rather spend developing a major upgrade than cleaning up what's there. So we've started on the road toward developing a revised measurement capability. I thought this time, at the beginning of the process, was really the best time to bring you guys into the mix.

A couple of rules: My inquiry of your opinion in no way implies that we don't know what we'd like to do, or that we're going to utilize all suggestions. It does mean that we know we're not any smarter than the average (technically competent) bear, and realize that many good ideas come from a dialog between interested parties.

I don't want to have a shouting contest between folks saying measurements are worthless and those saying measurement are the only thing of worth---both positions are just plain wrong. Objective measurement and subjective observation are both completely valid views within their own domain. What I would like us to do is discuss the relationships, if any, and how they might be best described. Anybody taking an excessively hard position on either side will be asked for a full and meaningful justification of thier position. Should one be: not forthcoming, irrational, or just plain stupid, IMH---but moderator privileged---O, your post will be summarily axed.

If someone has a particularly good idea and wants monetary compensation for it don't post it! You'd be surprised at how many things we've already thought of and just because you say it here first, doesn't mean that it hasn't been thought of before. If however, I like your thoughts and use them (or even if I really just like them), I might award you a gift to say thanks for being here and contributing to the dialog.

I am scaling back my general posting, but I also intensely desire to contribute to the community. This thread will make up the largest part of my postings here for a while, but I'm not going to chit chat and burn time. So please, let's keep this one on topic, or significantly entertaining. I will be moderating this thread so that folks don't have to wade through 10 posts for every on on headphone measurement.

On With The Show!

I reckon these threads are going to be loosely structured something like this:

Headphone Measurement Thread 1, Prolog: A time to generally express the kinds of things you'd like to learn in the course of the thread, and a time to throw measurement revision request noodles against the wall to see what might stick.

Headphone Measurement Thread 2, Measurement Introductions: This will be a tutorial on audio measurements in general, with a special emphasis on measurements done on speakers and sound that travels to your ears; how they're taken, what they mean, and the variety available.

Headphone Measurement Thread 3, Headphone Measurements: Once we understand basics of sound measurement, we'll explore the special case of headphone measurements: how they are similar to speaker measurements, how they are different, and the equipment used to acquire.

Headphone Measurement Thread 4, Exploring the Measured Data: Here we will dialog about actual measurements, what they might mean, how they compare to the subjective impressions.

Headphone Measurement Thread 5, The New Rev: Here's we'll discuss the merits of the various measurements, which should be included in our presentation, how to filter and present the data so that it is useful to the widest number of people.

Headphone Measurement Thread 6, a little decompression: what have we learned, what was surprising, what headphones should be included in the archive.

I reckon we'll spend about a month on each.

Just to get things started, I'll post some clicks to download some XL spreadsheets that have all the measurement data the we collected on our first pass at looking what might be done. We've picke three traditional cans, and three IEMs to look at: AKG K701, Senn HD650, Beyer DT880 (new), Ety ER4S, Shure SE310, Shure SE530.

OK folks, the floor is yours. What do you want to learn? What is the relevance of measurements to you? What measurements might be relevant for a wide base of users? Stray related comments?

 

[PiccoloNamek]

Something that I have always wanted to see is what the graphs look like prior to having the dummy head's effects subtracted from the measurement. This would give us a much better look at what a headphone's frequency response might actually look like at our eardrums, after the pinna and ear canal have shaped the sound. Of course, it wouldn't be exact because our ears and canals don't match the dummy's exactly, but I still think it would be helpful.

Also, IMD measurements, of course.

 

[Tyll Hertsens]

Hi Piccolo,

You'll find that the spreadsheets include pre-compensated data, so you can have a look at that. It's unlikely that the final product on the web will include such data as it's not really useful for most folks, but I agree that discussing and identifying all the artifacts the occur between the free-field audio and how it changes by the time it gets to the ear drum is a great topic for Thread 3 in the conversation.

As a brief teaser check out page seven of this word document, it shows all the things that go into the modified signal arriving at the ear canal. The compensation curve in our Head Acoustics HATS (head and torso simulator) is essentially a special case of those generic curves.

 

[PiccoloNamek]

Cool, I didn't realize those documents contained the uncompensated responses. Wow, the raw response for the HD650 sure is interesting. It certainly doesn't sound like that! This should make for excellent discussion later.

 

[Tyll Hertsens]

Quote:

Originally Posted by PiccoloNamek /img/forum/go_quote.gif Wow, the raw response for the HD650 sure is interesting. It certainly doesn't sound like that!

Yup. That's why you need the compensation.

Quote:

Originally Posted by Piccolo This should make for excellent discussion later.

I think so, too. I'm looking forward to the whole thing.

 

[Kees]

I find it very difficult (impossible actually...) to get any meaning out of absolute measurements (like The frequency respons curve of a particular pair of phones).
I think I could get a lot more info / meaning out of relative measurements.
If I interpret the differences I get a better view of the general behaviour of the components.
For instance:
If I can compare the FRC's of HD650s with five different (types of) amps I get a better view on what kind of effect a certain amp has on HD650s; i.e. how the HD650 behaves under different conditions.
If I have the FRCs of 5 other heaphones with the same amps I can see how they react differently to the same amp.
At the same time I can see how they are different from each other under different circumstances.
It sort of makes a moving picture instead of a static one.
Hope I made myself somewhat clear, I find it difficult to explain...

 

[thrice]

Quote:

Originally Posted by Tyll Hertsens /img/forum/go_quote.gif What do you want to learn? What is the relevance of measurements to you? What measurements might be relevant for a wide base of users? Stray related comments?

Tyll,

Thanks for posting this thread. I've referred to your site's measurements frequently and I would love to know more about measurements and their implications. Ultimately I think I'd like to get a better understanding of what each measurement means and how I can use them as tools to help me determine which headphone I might prefer over others (to help make an educated choice).

I wonder if people fell the FR graphs are accurate. I recently compared the HD 600 and 650 in my system to see which I would prefer. I find the the graphs mirror what I heard:

i.e.

 

[Tyll Hertsens]

Quote:

Originally Posted by Piccolo Also, IMD measurements, of course.

We tried a bit of that; the data looks very noisy. I'll see if we can't work up a coupla data sets to post so y'all can have a look. Remember that IMD measurements were really invented to detect non-linearities induced by the rather long delays in global feedback. It may simply not be relevant in speaker transducer measurements.

Quote:

Originally Posted by thrice /img/forum/go_quote.gif Ultimately I think I'd like to get a better understanding of what each measurement means and how I can use them as tools to help me determine which headphone I might prefer over others (to help make an educated choice).

Yeah, that's the goal alright. The thing that's difficult is that I've never seen a Tektronics 'scope that has a "headphone I might prefer" knob. Sigh.

Well we'll take a whack at it here over the next few months.

Quote:

Originally Posted by thrice I find the the graphs mirror what I heard: i.e.

Meaning that they sound pretty much the same?

 

[PiccoloNamek]

Quote:

Yup. That's why you need the compensation.

But wouldn't the raw response be more indicative of what the dummy itself is actually "hearing"? I mean, they certainly don't sound like the compensated response either, with two large notches cut out of the upper frequencies.

 

[HFat]

Quote:

Originally Posted by Kees /img/forum/go_quote.gif I find it very difficult (impossible actually...) to get any meaning out of absolute measurements (like The frequency respons curve of a particular pair of phones). I think I could get a lot more info / meaning out of relative measurements. If I interpret the differences I get a better view of the general behaviour of the components.

I agree but eyeballing a graph only takes you so far as the thrice/Tyll exchange over those Senns illustrates.

I think it would be more informative to actually listen to the differences revealed by these measurements. This is why I think Headroom should consider providing convolver impulse files as well as graphs. For example, one would be able to equalize a HD600 to have essentially the same FR as the HD650... if the measurements are good enough, that is. More than a tool for prospective buyers (does the HD650 really have too much bass?), I think it would be most useful as an educational tool to understand what part the FR plays in the sound of a headphone (in this case, you would need to have both Senns on hand as well as the impulse files). This might also be an elegant way to assess the quality of the measurements.

We don't need Headroom to provide impulse files actually because they can be derived from the graphs (how well though?). Head-fi member sound_man posted matlab scripts to do just that.

 

[thrice]

Quote:

Originally Posted by Tyll Hertsens /img/forum/go_quote.gif Meaning that they sound pretty much the same?

No they sound different to me. I hear deeper and more quantity of bass with the HD650 and while the graph may not represent the difference as dramatically, it's there. If you take the impedance graph and the FR graph together you see:

Impedance:

FR:

So while the FR indicate that they ought to sound almost the same, wouldn't the impedance graph suggest otherwise and add to the story.

It's this type of discussion that I'm hoping to read and contribute to

Cheers,
thrice

 

[gtp]

I've always wondered about how the FR graphs were normalized? When the curves have different mean values, it makes it more difficult to compare.

 

[Tyll Hertsens]

Quote:

Originally Posted by gtp /img/forum/go_quote.gif I've always wondered about how the FR graphs were normalized?

Prior to running the FR sweep a 1kHz tone drives the can and the volume is adjusted until the 1k tone hit's 90dB, then the sweep is done at that level. Zero on the graph scale is 90dB, all data sets are at zero dB at 1Khz.

 

[gtp]

Quote:

Originally Posted by Tyll Hertsens /img/forum/go_quote.gif Prior to running the FR sweep a 1kHz tone drives the can and the volume is adjusted until the 1k tone hit's 90dB, then the sweep is done at that level. Zero on the graph scale is 90dB, all data sets are at zero dB at 1Khz.

OK, but if 1kHz is a peak or valley, it moves the whole FR curve up or down relative to another headphone's FR curve.

Wouldn't it make more sense to set 1kHz @ 90dB as you say, do the sweep, but then normalize so Sum(FR(freq))=1?

 

[Tyll Hertsens]

Quote:

Originally Posted by gtp /img/forum/go_quote.gif OK, but if 1kHz is a peak or valley, it moves the whole FR curve up or down relative to another headphone's FR curve. Wouldn't it make more sense to set 1kHz @ 90dB as you say, do the sweep, but then normalize so Sum(FR(freq))=1?

Possibly. Hold that thought and when we get to Thread Five we may try to produce the data and show it in that form. Good suggestion.