Burst Response! HD800, SR-207, HD650

I think that's a bit of cherry picking. If this is the case, the 4-5th cycle (red) on the 10kHz burst should be heavily elevated because of the 6kHz spike of the HD800, and also there should be a huge depression around the 5th and 6th cycles (purple). But we don't see these things.


There is some correlation here between SR and FR, but it gets rather weak at the lower frequencies farther away from 10khz and also to the 10kHz burst.

 

I dunno, that it seems pretty weak. You could make the argument there should be a huge dip in cycles 5,6,7 of the burst because of the HD800's midrange dip. But it isn't there.

 

The IR argument makes sense. I was being a retard before and forgot that the IR does not just consist of the impulse, but everything that comes after that for at least few seconds, at least for the lengths that I used to obtain them. In other words, there is temporal information contained in the IR, quite a bit of it, that can be used to reconstruct the bursts.

So using the IR to process a burst and getting very similar results to the recorded burst was not surprising to me in the end. The remaining question is how close to reality is it, or next best thing, the actual recording of such a burst.

FR is obtainable from the IR (and SR if the transform is inversible). However a burst pattern such as these cannot be obtained from the FR because all time domain has been lost. This is not to say there may be some occasional associated elements between FR and the burst. But FR isn't a reliable determiner of the behavior of any burst.

 

Makes sense. 10kHz is highly directional and probably bounces all over the place, especially inside the HD800 cups. There's probably a combination of direct arrival with a series of indirect and bounced off arrivals over time, with maybe some cancelations.

 

I think we can get the IR from FR (magnitude and phase) in our discrete time cases here. If one only has FR magnitude, one can still make some assumptions about the phase and get something proly very close in time domain.

That said, these are different ways of looking at things which gives us different insight as to what is going on.

 

Yeah, as long as the phase and all the fine granular data points at the FFT bins are there, we can get the IR back.

 

OCD attack there. Sorry.

I'm Ok now.

 

So adding some comparos using some files @Marvey sent my way.

Note that there are a couple of non-idealities here:

1) The measured files were sampled at 192 kHz, while the IRs were 44.1 kHz. So I upsampled by 48 and decimated by 11 to get the 44.1 kHz to 192.44 kHz which means things will slip a bit as time goes by.

2) There may be some variation in the measurements for IR and bursts. Specially with Stax stuff that require heroic efforts to properly seal with the coupler.

3) This time I used Octave, and I have very little trust in it's upsampling, downsampling capabilities. Octave is good cuz it's free. But IME Matlab kicks it's butt.

So here is

SR207 full



SR207 start



Atticus full



Atticus start



Anyhow, tired. Nite, nite.

 

I was hoping the results would be closer, at least for the Atticus where I had the headphone on the rig and didn't move it for both measurements.

I wonder if maybe non linear behaviors in the headphone (and microphone) are also responsible for the differences. Reality is never linear.

I'll take another set for Eikon and send you the data and process it myself if I have time.

 

I'll see if I can do some of these tests later with my cans. Will report.

 

BTW @JohnM, while I do know impulse response fully captures the linear behavior of a system, I greatly appreciate your comments which go a long way in encouraging us into thinking about these things more carefully. Maybe later we will review wavelets in the context of audio here as well.

I also know you've been an engineer for a while, seen your REW work, and like how you handle yourself. There are a couple of things at hand. I try to avoid using certainty terms unless I really have to. Folks in here are quite smart and I'm sure we will digest the concepts, as we have done continuously in the past and currently. Hopefully we will provide new insight and intuition to problems in the process.

As you know, I also like REW quite a bit, and I'm very happy to see you participate here. LOL! I would like to see more of you around here actually.

 

Don't worry guys. It should work. There were a few non-ideal things.

For one, having different sampling rates complicates things, specially when some are multiples of 44.1 kHz and the others are multiples of 48 kHz. What this does is that we no longer have a 1 kHz burst, but a 1.something kHz burst (equivalent) in some of the tests. I did the best I could to keep things close. It is also very important that the IR measurement and the burst measurements are done the same way (same coupler, same mic position, same everything).

I should have done these tests yesterday, but had a lot of stuff to do (like school parent's meeting, honey do this, honey do that...) We'll get to the bottom of this.

 

Here are the processed files vs. the measured files:

HD800 Coupler 1




HD800 Head



I can see noise in the burst measurements which was not present in the IR measurements. That said, the channel characteristic capture in both the burst and IR measurements seem fairly consistent.