Head-Fi.org › Forums › Equipment Forums › Headphones (full-size) › Headphone CSD waterfall plots
New Posts  All Forums:Forum Nav:

Headphone CSD waterfall plots - Page 54

post #796 of 937

The effects on the AKG K 340 Elec-Dyna of velour/pleather pads and cotton/no cotton inside the cup.

 
Also, ray-traced CSD graphs. Improved depth perception vs. 2d graphs I think, but less clear relative amplitudes due to me not color-coding the plot.
 
Pleather pads, cotton in cup:

 

Velour pads, cotton in cup:

 

Pleather pads, no cotton in cup:

 

I preferred velour pads vs. pleather on this K 340; treble sounded better.
 
It's curious that having cotton in the cups causes ringing in the treble.
 
(There's also an optional Mountain View - good for Grados, etc.

)

 


Edited by vid - 10/4/12 at 7:22pm
post #797 of 937

The mountain view is great.  I prefer the original purple graphs when in overhead view, though.

post #798 of 937
Quote:
Originally Posted by vid View Post

The effects on the AKG K 340 Elec-Dyna of velour/pleather pads and cotton/no cotton inside the cup.

 
Also, ray-traced CSD graphs. Improved depth perception vs. 2d graphs I think, but less clear relative amplitudes due to me not color-coding the plot.
 
Pleather pads, cotton in cup:

 

Velour pads, cotton in cup:

 

Pleather pads, no cotton in cup:

 

I preferred velour pads vs. pleather on this K 340; treble sounded better.
 
It's curious that having cotton in the cups causes ringing in the treble.
 
(There's also an optional Mountain View - good for Grados, etc.

)

 

That is some fantastic rendering!

 

Maybe try the mountain view but from a "higher altitude", so to speak?

post #799 of 937

The three graphs basically are the mountain view but from a higher altitude (minus the sky, though sky light still contributes to the model). The other difference is that the mountain view shows the plot from a slight angle on the x axis, which is how most CSD graphs seem to be (i.e. ridges move progressively down and left). I could try rotating the view a bit to the left; the issue then might be that big peaks in the middle (like the 4 kHz one) would possibly obscure some data below them. And of course the mountain view has noticeable direct lighting - but it looks somewhat ugly in the lower frequencies of the graph due to the plot being logarithmic and thus having poorer resolution the lower you go (FFT's fault) and thus the model being more blocky to the left. I could smooth it out, but I'd be making up data that isn't there...

 

The back of the plot is maybe a bit too far from the camera and the treble decay portion is somewhat pushed back then. I could try rotating the graph a bit more on the up axis and/or zooming in a bit.

 

I prefer the rendered plot to the 2d one, though... I'll implement some color-coding for dB and rotate the graph a bit and see how that turns out. The reflective plate on the left of the graph isn't really needed, it increases rendering time and possibly makes the plot harder to read, but I'll see about that as well.


Edited by vid - 10/5/12 at 6:29am
post #800 of 937

Tinkered with the CSD plot generation a bit. Plots are still ray-traced, but now come with some extra 2d visual cues. CSD calculations were modified a little as well. The color scheme I might change - it doesn't strike me as very good seeing it on the forum now.

 

AKG K 240 DF (again, for reference)

 

AKG K 240 DF modded

 

AKG K 240 DF (non-modded) with an FFT size of 2048 rather than the 512 used for the above.

 

The larger FFT size gives better resolution, but I'm not sure how much of that is really useful (whether you really need to know about a ridge peaking at -30 dB) and also whether it would simply magnify inaccuracies in the measurement rather than actual detail.

 

Also, the plots cut off at 500 Hz or so as I'm not sure if knowing about ringing below that is a useful thing or not.


Edited by vid - 10/9/12 at 7:55pm
post #801 of 937
Quote:
Originally Posted by vid View Post

Tinkered with the CSD plot generation a bit. Plots are still ray-traced, but now come with some extra 2d visual cues. CSD calculations were modified a little as well. The color scheme I might change - it doesn't strike me as very good seeing it on the forum now.

 

AKG K 240 DF (again, for reference)

 

AKG K 240 DF modded

 

AKG K 240 DF (non-modded) with an FFT size of 2048 rather than the 512 used for the above.

 

The larger FFT size gives better resolution, but I'm not sure how much of that is really useful (whether you really need to know about a ridge peaking at -30 dB) and also whether it would simply magnify inaccuracies in the measurement rather than actual detail.

 

That is really nice, nice 3D angle giving maximum information and details, and a sense of depth so ridges and peaks can be gauged more readily. Not a fan of the colour scheme though, your earlier ones were much nicer coloured.

post #802 of 937

I prefer purrin's "energy coloring". If you're examining a certain point of a particular frequency on the z-axel, you don't get a fully accurate sense where it stands on the y-axel. One can approximate, but integrating something like that would help I think. 

post #803 of 937
Quote:
Originally Posted by electropop View Post

I prefer purrin's "energy coloring". If you're examining a certain point of a particular frequency on the z-axel, you don't get a fully accurate sense where it stands on the y-axel. One can approximate, but integrating something like that would help I think. 

 

Y is up, I assume? How do you mean energy coloring? Purrin's coloring has a higher contrast of values, which may have something to do with it. I've intentionally tried to avoid colors that clash too much, but of course that means less easy to read.... It needs some balancing between those two I think.

 

Another possible coloring might be to color not based on 'global' dB but on relative amplitude change for that particular frequency band. As purrin often notes, it needs to be considered where the decay starts at, not just where it ends. But I'll experiment.

post #804 of 937
Quote:
Originally Posted by vid View Post

 

Y is up, I assume? How do you mean energy coloring? Purrin's coloring has a higher contrast of values, which may have something to do with it. I've intentionally tried to avoid colors that clash too much, but of course that means less easy to read.... It needs some balancing between those two I think.

 

Another possible coloring might be to color not based on 'global' dB but on relative amplitude change for that particular frequency band. As purrin often notes, it needs to be considered where the decay starts at, not just where it ends. But I'll experiment.

 

Yes, y is up/down :) I see these as xyz-coordinates. It would be interesting to see coloring based on relative amplitude change and how that would work. You and purrin have a point that it indeed is to be considered where it starts, though if the start deviates too much from the ideal point (relatively flat FR), there might be other issues afoot. 

 

I used the term "energy coloring" because I couldn't come up with a better term, heh. But yes, if you for instance have ringing at 4kHz and examine the ridge at 2ms, if there's no color to show amplitude it's more difficult to pin point where it is on the y-axel (amplitude). The high contrast coloring scheme also allows 2D charts as illustrated by purrin. 

 

Anyway, good looking regardless. Hope to see more from you! 

post #805 of 937

I'm not sure who's responsible for it being so, but y for up depends on the specific xyz coordinate system - Blender for instance will insist that z is up. I'd agree that y is (naturally) up, but some people very much disagree.

 

The graphs I posted last time make use of this energy coloring, but in a less clear way than how purrin does it. Finding a decent (yet readable) color scheme that isn't the usual blue, yellow, red is somewhat of a challenge...


Edited by vid - 10/10/12 at 7:42pm
post #806 of 937
Thread Starter 

Another way is changing the color as we go forward with each timeslice. It comes down to what you want to see and how you want to see it.

post #807 of 937

Not to forget that (headphone) CSD plots are inherently poised to mislead anyway. Differences between plots are very small in time and amplitude, and a clean-looking plot can become very messy just by lowering the noise floor by, say, 3 dB - which can be done covertly simply by manipulating the positioning of the response on the graph while the graph's stated noise floor remains the same.

post #808 of 937

I'm still hoping to see more CSD graphs of impulse response convolved with their minimum phase inverse filter, such as posted by xnor.  As has been pointed out before much of the "ringing" we see is just a consequence of the frequency magnitude response.  A CSD graph of the impulse response with the magnitude corrected to flat would let us separate actual non-minimum phase ringing from the FR-induced "ringing" that can just be EQed out.

post #809 of 937

^ Make a detailed guide on how to do that and I'll try. (Though I can't guarantee visually clear graphs.)

post #810 of 937

Some tweaks again to the plots. The perspective is closer to orthographic, there's an additional top-down view that extends down to 7 ms, bigger ridges are individually marked for frequency again, and the color scheme was changed a little, though not much.

 

AKG K 250 modded with a nylon disc in the cup (not a big effect) and another nylon disc in front of the driver (a bigger effect):

 

Original K 250:

 

It's a good driver with a few problems. The 4.5 kHz peak is interesting in that its frequency isn't affected by the ear's anatomy - it's asking for some EQ later on.

 

AKG K 240 DF again, for reference in the new design:

New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Headphones (full-size)
Head-Fi.org › Forums › Equipment Forums › Headphones (full-size) › Headphone CSD waterfall plots