Why all the squiggles > 2,000 Hz in frequency response?

[ruthieandjohn]

Every headphone frequency response curve that I know of starts wobbling frantically above 2,000 Hz.  Witness these four:
 

 

Why is that?  Is it desirable?  If a headphone design could get rid of it, would it be better?  Or is that wobble what gives each headphone its "signature" or unique voice?
 
For a while, I thought that perhaps the wobbles were as a result of the vibration modes of the earphone diaphragm.  I convinced myself that for a typically large 50 mm membrane at the right tension and spring constant, the various mode patterns would cause peaks at funny multiples of a fundamental around 4,000 Hz.
 
But then I saw similar behavior for the frequency response curves of IEMs, which of course have diaphragms that are an order of magnitude smaller.  The green and orange curves above are for IEMs, while the red and blue are for on-ear headphones... both have that wiggle structure in a similar place, and all have a peak at about 9,000 Hz.
 
So then I wondered... is it a result of the head-related transfer function (HRTF), which affects the frequency response measured by microphones inside a surrogate head, and perhaps could impose this pattern?  Here is the HRTF of a particular head used for measurements...
 

 
Sort of looks like that wiggling in the headphone frequency response, but inverted.  Even has a dip at 9,000 Hz where the headphone frequency response has a peak.
 
As I understand it, raw measurements made by microphones in the ears of an artificial head used for headphone measurements are compensated for the HRTF before being plotted.  Is that what is happening?

 

[Claritas]

My understanding is that the 9 kHz spike is caused by sound waves bouncing off the pinna.
 
"Circumaural and supraaural headphones typically crush the pinna to some degree, altering its shape and thereby its acoustical interaction with the headphone" from this Stereophile article about measurement.

 

[bigshot]

I believe it's an attempt to compensate for Fletcher Munson. It would be more useful if they just factored for that and presented a graph of how it sounds rather than how it measures, but the people who make these graphs don't think that way.

 

[bfreedma]

I believe it's an attempt to compensate for Fletcher Munson. It would be more useful if they just factored for that and presented a graph of how it sounds rather than how it measures, but the people who make these graphs don't think that way.

It's a combination of a couple of things:

The graph scale is logarithmic, making the variance in high frequency appear much more steep than the lows

Headphone measurements above 10k are notoriously difficult to capture correctly

Why would we want to alter graphs from the actual measurements? It's not a matter of how those that make the graphs think, the measurements simply represent response. Muddling the actuals with some compensation curve would be incredibly confusing.

 

[bigshot]

Why would we want to alter graphs from the actual measurements?

 
Because the graph uncompensated doesn't represent what people actually hear. The whole purpose of the graph is to get an idea of how headphones sound.

 

[bfreedma]

Why would we want to alter graphs from the actual measurements?

 
Because the graph uncompensated doesn't represent what people actually hear. The whole purpose of the graph is to get an idea of how headphones sound.

Even if I agreed with your concept, which I don't, how would you get the entire industry, hobbyists and professionals included, to agree to and use identical post measurement compensation.

It's hard enough to get consistent raw measurements. Adding in compensation to some measurements would make comparisons impossible without universal adoption and that isn't going to happen.

 

[bigshot]

As if measurement charts are consistent now!

 

[bfreedma]

Measurements are reasonably consistent over time when done by those with the right gear. Innerfidelity, for example. If you believe they aren't accurate, how does adding a compensation curve make them any more useful.

If you prefer to see the measurements with compensation, why not just run the math yourself?

 

[bigshot]

Because I don't want to be an amateur scientist. I just want to know how headphones sound without having to buy them to find out.

 

[bfreedma]

Because I don't want to be an amateur scientist. I just want to know how headphones sound without having to buy them to find out.

Are you really suggesting that we should change the way headphone measurements are presented because you don't want to run some simple calculations? Raw data has been working fine for some decades now - the onus is on you if you want it presented in an alternate format.

Besides, we are discussing frequency response graphs, not measurements tied to your specific hearing (or mine). I really don't see how changing the graph via a constant (Fletcher Munson) in any way improves the value of the data. You either understand how to interpret it or you don't.

 

[Claritas]

Often enough, both sets of measurements are available. Besides, FR isn't the only important measurement, which is why we use several to obtain an understanding of different aspects of what we're hearing (though it's far from simple how they work in concert).

 

[bigshot]

Besides, we are discussing frequency response graphs, not measurements tied to your specific hearing (or mine).

 
Within the core frequencies, human hearing doesn't vary all that much unless you have seriously impaired hearing.