[Cristello]

removed

 

[Dyaems]

slightly OT: in basic terms, what does "attack" mean?
 
cant understand the meaning of that in google or in the glossary xD

 

[atomikn00b]

out of curiosity of where this discussion will lead..
SUBSCRIBED.

 

[vid]

Quote:

For the sake of clarity, would you be able to verify that it was only the driver being measured (and not other possible sources of resonances)? Also, I assume that the driver was measured "free-field". Is this correct?
 
I am testing a theory based on your data and some of my own, but am waiting to post any results until I can confirm their relevance.

 
I've not measured the driver out of its cup, if that's what you meant. I've measured the K 250 with and without a seal (the latter was a microphone capsule hanging freely in the air over the baffle) and gotten the same ringing in the measurements. I also got similar ringing from measuring the K 241, which is possibly the same driver but in a different cup. That specific measurement you quoted was with a seal.
 
I made a post there -> http://www.head-fi.org/t/566929/headphone-csd-waterfall-plots/855#post_8788251 specifically about smoothing out ringing on the K 250. For whatever reason, that convolution filter got rid of the ridges even though the other one didn't. (But the basic problem remains with that filter as well: it's tied to the frequency response you feed it, and without knowing the response at your own eardrum, its applicability is limited. Works nice for the measurement ear, though.)

 

[obobskivich]

^ The spectrum of the convolution recording is flat between 40 Hz or so (lower limit of my recording capability) and about 15 kHz or whatever (where I set the lowpass), as is the original test signal, though without the lowpass. The convolution filter seeks to create a flat response, which it thus has done. The question is, is a flat response an automatic flat CSD (= perfect IR)? If so, why; and if not, why so?

xnor put up a waterfall that showed an *almost* flat CSD, but I didn't follow the posts much further after that to see if he explained how he did it or whatnot (I'm sure he did, I just haven't kept up with the thread).

Perfectly flat FR, interesting - it would have perfect phase response (we're talking DC-inf here, not 20-20), but that wouldn't guarantee no resonance. So I think you'd see a very *good* IR, but CSD I'm less sure on - again, you aren't guaranteeing no resonance, just perfect phase (and I think in this case, what your filtering is doing is essentially guaranteeing perfect phase as well)). But that doesn't eliminate other problems from the chain (in other words, you could still have resonance, and issues with how the ear (or room, if we're talking speakers) mangles the signal, if the original signal is crap, etc.

I think true perfection is impossible (I've seen nothing to indicate this statement as false), but we can certainly improve on "un-processed" (this much has been demonstrated very well with room EQ suites, and I suspect we will see more in the way of headphone EQ hardware before too long, Smyth can't be the only game in town forever).

If you would, please explain how physical modding and EQ are different in this case. Not that I doubt you, just that an explanation is nicer than a true/false statement.

EQ processes the signal before output, physical modding changes the output. They can both be thought of as "filters" if you're going to put the entire playback system into a black box, but it seems overly vague imho. By "upstream" I mean anything that sits ahead of output (which is your transducers and amplifier (we're assuming the amplifier here is normal, not trying to color the sound in some way of its own)).

In other words, I can rip a pair of headphones up and turn them into whatever I like, but the signals from the source (lets say a CPD) all the way through to the output taps are going to be "flat" (generalizing here), and then it's up to the headphones to "filter" the sound (mangle). With EQ that "filtering" is done before the output taps, and we just leave the headphones alone. EQ'ing (using term generally) will help to correct response issues with the output device, but it can only go so far (e.g. electrical damping is of relatively minor importance, but physical damping is a big deal). There's also other things you can do when you start changing the physical nature of the output device, like changing radiation patterns (you can use fairly expensive DSP to try and simulate this, but IME its easier to just move the transducer). The ideal system would rely on both - very good output transducers that get near your ideal performance requirements, and EQ/DSP to take the last step towards "flat" (or wherever you want to go).

The test signal was 'mangled' as you say to produce a desired output (flat response), but you say this can't change the rate at which the driver moves. Below is part of the waveform of (a) the original test signal, (b) the recording off the earbuds of the non-convolved signal, and (c) the recording of the convolved signal off the earbuds.

Interesting. Snipping images out to prevent re-paste.

The rate at which the driver moves looks to have changed. But whether it did or not, it in any case seems that the driver is able to keep up with the test tone transients when fed with a convolved version of the signal, but having big trouble doing so otherwise.

I don't think the driver is moving faster, it is still limited by its motor and other governing factors. What you're doing is creating a filter that basically says "okay, this is screwed up, so if we figure out exactly how it's screwed up, and send our music/whatever through an inverse of that filter, the output should look less screwed up" - it's more along the lines of how ANC works than super-charging the driver. Notice the ringing and other anomolies between A and C - it isn't exactly perfect. I think you're correcting phase anomalies more than transient attack problems, and remember that IEMs/buds generally have better resonance control/damping to begin with - just by nature. So I think in your scenario you're very likely approaching the ideal, but I'm wondering if you were to take a headphone that begins pretty "bad" in terms of resonance (like the Koss MV1) and try your convolving filter, if it would work as well.

I'm thinking that problems of mechanical damping (or lack thereof) will not be completely fixed by this method, just as in room acoustics you can't completely EQ or convolve your problems away - at some point you have to break out the 703 and move speakers around a little bit. But that doesn't mean EQ/filtering doesn't have an advantage - correcting phase anomalies will move you closer to a target flat response within the constraints of your speaker system. But this isn't really improving or even changing the speakers - it improves the system's output quality sure, but that's moreso the result of recognizing and trying to counteract the "flaws" than it is the result of actually changing the conditions that created them.

Where I'm also curious here is that, with most EQ solutions that require substantial filtering, you usually end up boosting a given frequency range at some point - which means increased power and increased excursion. Which means increased THD. Is that happening in this scenario as well?

slightly OT: in basic terms, what does "attack" mean?

cant understand the meaning of that in google or in the glossary xD

Attack is how quickly a driver responds to the "leading edge" of a transient or sound. More here:
http://en.wikipedia.org/wiki/Attack_(music)#ADSR_envelope

Basically how quickly can the driver (or system) go from rest to output, and decay is how quickly it can go back to rest. Perfectly flat CSD would have basically infinitesimally small times for these values, and any sort of resonance or ringing would be from the signal, not the driver (or system).

 

[Dyaems]

Attack is how quickly a driver responds to the "leading edge" of a transient or sound. More here:
http://en.wikipedia.org/wiki/Attack_(music)#ADSR_envelope
Basically how quickly can the driver (or system) go from rest to output, and decay is how quickly it can go back to rest. Perfectly flat CSD would have basically infinitesimally small times for these values, and any sort of resonance or ringing would be from the signal, not the driver (or system).

 
thanks bro

 

[vid]

Quote:

(snip)

 
xnor used convolution, did he not?
 
The transients produced by the driver are as fast as the test signal asks them to be; how much more can be asked for? The raggedness in the c graph I would attribute partly to poor SNR given that it was recorded at low volume. Aligning a and c on top of each other shows that the transients are perfect up and down. I dare you to find an electrostat that can do the same without the input being altered (I also dare you to show rather than just tell). Do this for me and I'll convolve a pair of phones that perform really badly stock.
 
EQ processes the signal before output, physical modding processes it either during or after output. All before it reaches the ear. If the resulting waves at the eardrum are the same, what is the difference?

 

[pb21cyclist]

I think my HE400s (planar) are fast, and when I say that I mean the drivers respond quickly and are well controlled. I think this is because the drivers/diaphragms are very light and easy to move and the magnetic field is very strong. In dynamic headphones the driver is heavier and more difficult to control.
 
Does that make sense?

 

[Cristello]

removed

 

[vid]

1. No idea exactly. Some amount but not massively poor.
2. No need to assume; they're the Yuin PK3. The CSD plot I posted earlier was of the AKG K 250 (somewhat as difficult to drive as the K 240 DF).

 

[warrenpchi]

Quote:

out of curiosity of where this discussion will lead..
SUBSCRIBED.

 
+1

 

[Cristello]

removed

 

[ultrabike]

Some comments

 
Quote:

Perfectly flat FR, interesting - it would have perfect phase response (we're talking DC-inf here, not 20-20), but that wouldn't guarantee no resonance. So I think you'd see a very *good* IR, but CSD I'm less sure on - again, you aren't guaranteeing no resonance, just perfect phase (and I think in this case, what your filtering is doing is essentially guaranteeing perfect phase as well)). But that doesn't eliminate other problems from the chain (in other words, you could still have resonance, and issues with how the ear (or room, if we're talking speakers) mangles the signal, if the original signal is crap, etc.

There is evidence, based on vid's results, that fixing the IR through equalization will fix the CSD, resonance and phase. IME what a linear equalizer cannot fix is non-linear distortion such as THD, IMD, and noise floor.

In other words, I can rip a pair of headphones up and turn them into whatever I like, but the signals from the source (lets say a CPD) all the way through to the output taps are going to be "flat" (generalizing here), and then it's up to the headphones to "filter" the sound (mangle). With EQ that "filtering" is done before the output taps, and we just leave the headphones alone. EQ'ing (using term generally) will help to correct response issues with the output device, but it can only go so far (e.g. electrical damping is of relatively minor importance, but physical damping is a big deal). There's also other things you can do when you start changing the physical nature of the output device, like changing radiation patterns (you can use fairly expensive DSP to try and simulate this, but IME its easier to just move the transducer). The ideal system would rely on both - very good output transducers that get near your ideal performance requirements, and EQ/DSP to take the last step towards "flat" (or wherever you want to go).
Interesting. Snipping images out to prevent re-paste.

As long as the system is linear, applying equalization before or after does not change the outcome. Also, as far as headphones, an equalizer should have a strong impact on acoustic damping. I agree however that a good system should have decent transducers to begin with: low distortion, no notches, and as positional invariant as possible.

I don't think the driver is moving faster, it is still limited by its motor and other governing factors. What you're doing is creating a filter that basically says "okay, this is screwed up, so if we figure out exactly how it's screwed up, and send our music/whatever through an inverse of that filter, the output should look less screwed up" - it's more along the lines of how ANC works than super-charging the driver. Notice the ringing and other anomolies between A and C - it isn't exactly perfect. I think you're correcting phase anomalies more than transient attack problems, and remember that IEMs/buds generally have better resonance control/damping to begin with - just by nature. So I think in your scenario you're very likely approaching the ideal, but I'm wondering if you were to take a headphone that begins pretty "bad" in terms of resonance (like the Koss MV1) and try your convolving filter, if it would work as well.

IME an equalizer should be able to correct for both phase and transient attack problems.

I'm thinking that problems of mechanical damping (or lack thereof) will not be completely fixed by this method, just as in room acoustics you can't completely EQ or convolve your problems away - at some point you have to break out the 703 and move speakers around a little bit. But that doesn't mean EQ/filtering doesn't have an advantage - correcting phase anomalies will move you closer to a target flat response within the constraints of your speaker system. But this isn't really improving or even changing the speakers - it improves the system's output quality sure, but that's moreso the result of recognizing and trying to counteract the "flaws" than it is the result of actually changing the conditions that created them.

My understanding is that mechanical damping will be significantly fixed through equalization (done right.)

Where I'm also curious here is that, with most EQ solutions that require substantial filtering, you usually end up boosting a given frequency range at some point - which means increased power and increased excursion. Which means increased THD. Is that happening in this scenario as well?
Attack is how quickly a driver responds to the "leading edge" of a transient or sound. More here:
http://en.wikipedia.org/wiki/Attack_(music)#ADSR_envelope
Basically how quickly can the driver (or system) go from rest to output, and decay is how quickly it can go back to rest. Perfectly flat CSD would have basically infinitesimally small times for these values, and any sort of resonance or ringing would be from the signal, not the driver (or system).

Equalization can in fact increase THD and noise floor so care most be taken in it's application. There are equalizer "training" approaches that attempt to optimize for both flatness with out having too a detrimental effect on noise and non-linear behavior. That said, an equalizer should improve on the perceived quickness of the driver. Additionally, AFAIK how quick a driver is, in terms of decay and rest, is also function of transducer bandwidth. In terms of bandwidth, the only thing the transducer has to cover is 20kHz worst case with no issues in FR amplitude or phase, and with as low distortion as possible. That is, it doesn't have to stop immediately to sound quick since most of us can't hear past 20kHz.

 
Hopefully this is helpful...