Headphone CSD waterfall plots

Purrin et al., these graphs are great - I should have commented earlier to thank you for the work. Now, I've got a couple of technical questions, if you don't mind spending a bit of time explaining to someone too lazy to read a textbook, such as myself : )
 
I've got no experience using FFTs for audio data, but have a pretty good idea of the method for engineering purposes. So, here's a few questions - what does 'front window' and 'back window' mean? Are you using a Blackman-Harris window for half of your 512 samples, and then a rectangular window for the other half, meaning that this half all has equal weighting until the 512th sample, after which there's an abrupt cut-off? If so, what's the advantage of this over using a full window rather than a half and half? If this is correct, then which half is the front and which is the back?
 
With such a long window of 10.67ms, there much be a very large overlap between each one - quickly casting my eye over the graphs, it looks like each window is shifted 2 samples along from the previous. Did you need a window this long to obtain sufficient frequency resolution, or was there another reason?
 
If I've completely missed the point of what you're doing, and you're struggling to politely tell me I'm asking stupid questions, feel free to point me towards a textbook or paper that explains it (we have electronic subscriptions to most engineering and mathematics journals, so that should be fine).
 
Thanks, and keep up the great work!
 
p.s. that HD800 appears to be an incredible piece of kit - I should give one an audition sometime, though I might get some strange looks if I fire up MATLAB during my listening session...

Quote:

arleus said:

So, here's a few questions - what does 'front window' and 'back window' mean? Are you using a Blackman-Harris window for half of your 512 samples, and then a rectangular window for the other half, meaning that this half all has equal weighting until the 512th sample, after which there's an abrupt cut-off? If so, what's the advantage of this over using a full window rather than a half and half? If this is correct, then which half is the front and which is the back?
 
With such a long window of 10.67ms, there much be a very large overlap between each one - quickly casting my eye over the graphs, it looks like each window is shifted 2 samples along from the previous. Did you need a window this long to obtain sufficient frequency resolution, or was there another reason?

Click to expand...

 
The Blackman-Harris function is applied only for the first 0.25ms on the front edge (that should clarify a lot!) It's only there to smooth out artifacts from sudden deletion of data when the window is moved right. The size of the window is maintained at X samples, but as I move the window to forward in time, I replace the back edge values with zeros. I found this approach preferable to actually decreasing the size of the window. FFTs applied on decreasing windows sizes would result in aliasing if I wasn't careful in lining up each successive calculation. And I wasn't in the mood to figure out or code an anti-aliasing function - I'm sure you can understand why!

The back-end function was intended to be applied to the last half. I never implemented any function (hence it's basically rectangular throughout) because it didn't seem necessary, i.e. the visual results did not improve noticeably.
 
Just a note. I padded 0.25ms of zeros to the the front of the window so the initial pulse wouldn't get garbled.
 
This is what's done for calculating each successive time slice.
 

Ah! I see what you've done now - that little ASCII drawing helped a lot. Thanks for taking the time to explain. I thought this was simply a case of having many more than 512 samples, then taking successive short windows (with maybe 50% overlap) along the length of the data, but I suppose if you wanted to do that, you would have to have a very high sample rate to get a good enough frequency resolution, as the number of samples in the moving window would be quite small. I'm heading off for a week and a bit soon, but do you mind if I have a play around with your raw data when I get back? THAT SOUNDS INAPPROPRIATE.

Quote:

arleus said:

p.s. that HD800 appears to be an incredible piece of kit - I should give one an audition sometime, though I might get some strange looks if I fire up MATLAB during my listening session...

Click to expand...

 
The HD800 measures well from a CSD point of view. The sample measured here was modified with pieces of foam placed at strategic places inside the earcup. The treble energy from 5kHz onward has been reduced about 3db compared to a stock sample.
 
The other headphones, the ESP950 and SR009 measure just as well, but in other ways. The hash and minor ringing on these two are at such low levels and duration where they are not audible.
 
I am debating whether to go back to the -30db scale, or expanding the dark blue color area further up the db ladder. I am also considering going back to the rectangular window or using another window instead of the Blackman-Harris window. It's a balance between capturing dynamic range or resolution.

Quote:

arleus said:

but do you mind if I have a play around with your raw data when I get back? THAT SOUNDS INAPPROPRIATE.

Click to expand...

PM me your contact info. I'll send you a few samples or more.
 

I'll give you a buzz when I get back - should be an interesting side project.
 
The SR009 is so ridiculously pricey, I don't think I'll ever consider buying one in my lifetime, but the 950 is actually rather tempting - the measurements show it should be very competitive for price, but I think my next purchase should really be a T50rp. It's about time I manned up and bought some brightly coloured modelling clay. Maybe I can get some from the Early Learning Centre. Hmm.

@Arleus:
 
Modified T50RPs sound good and measure well: http://www.head-fi.org/t/452404/just-listened-to-some-fostex-t50rps-today-wow/2760#post_7774288

Anetode was kind enough to ship me a bunch of headphones. Another set of measurements will begin!
 
Grrrrrrr. Now I only need those helicopters, rice-boys, and rednecks to turn off off their motors. Oh great. Garbage truck and gardeners just pulled in.

Another batch of measurements...

 
Before I measured each of these headphones, I listened to them first and jotted down some impressions. The listening rig is set up as follows: PC--ethernet-->Bridge-PWD--balanced-->BA(EML 300B, Brimar 6SN7)--single-ended-cable-->headphone.
 
I will post results in the order I tested them.
 

AKG K601

Thin. Little bit of ringing in top treble. Fatiguing. Harsh mids - glare. No Body. Ouch. Doesn't pick up ambience. No inner detail. Yuck. How can it sound so bad of the BA which is set up to sound thick? Like photoshop sharpening filter applied incorrectly. Inefficient - turned up volume knob, even more than HD800s. Turned it off. I can't listen to these. There's ringing in at least two spots. Two spots dominate  (one in midrange and the other in treble.)
 
OK, let's see what the plots say:
 
 
I thought it would look worse. There is a broad hill a little above 2kHz that takes a while to decay. Since these plots only go down to 200Hz, they don't necessarily give you a good idea of the overall sound balance (the bass starts to drop off starting at 200Hz.) I don't know if that sharp knife-like edge in the top-most octave is responsible for what I'm hearing. Hmm, looking at the plot more closely, I see the same ringing behavior carried at three octaves (~4k, ~8k, ~16k). It's more evident on the right channel. That probably explains it.
 

HD650

Wow these sounds good. Not as dead, thick, or bassy sounding as I remembered (from other setups.) Bass a little slow - not as tight as I would like. Actually bass is not bad. Not bad at all. Good air and treble extension. Surprising ability to extract low-level information (I mean real information, not treble boosted pseudo information.) No appreciable ringing - maybe one small spot. The HD650/600 should be the go to phone for anyone relatively new to the hobby - before they start spending big bucks. These would be even better if converted to balanced on the BA. FR is smooth with a boost in bass, and a touch of boost in treble. An excellent implementation of pleasant and forgiving withOUT decimating musical information. Why couldn't Sennheiser make an HD800 with a tonal balance like this? LOL. Having a hard time taking these off for measurements. These headphones move me: the musical information, the sense of "it is there." Hmm, just switched to some different material, there's some midrange boost. The EML 300Bs are a little too thick and bloomy in the lower mids for these, KR PX4 may be better.
 
OK, here we go:
 
 
Everything after 1ms past 1kHz is dark blue (less than -30db). This is fairly fast decay.

^ Looks very nice and clean...
 
Gimme some more food , COME ON!

Beyer T1

Just as I remembered them. An advanced DT880 with darker tonal balance. Similar tone to HD650. A tad more clarity. Treble is peaky. Definitely a spot that rings up in the treble. Not horrible ringing though - only activates on occasion like LCD2's ringing, but higher in frequency. Although more clear and brighter sounding than HD650, less able to extract low-level information compared to HD650. That treble peak is starting to bug me. It's worse than I initially thought. FR is wonky - not smooth from midrange to treble. Had enough. Taking these off, going back to HD650s.
 
Plots here:
 
 
It took me a while to get back because I re-measured these a few times to make sure there wasn't a cup placement or ear-pad thickness issue. The left channel is just off. I would expect better driver matching than this. Way to go Beyer with your $1000+++ headphones! 

I have to agree, the T1 just never impressed me that much; nor has any other full-size beyer I've heard. The DT990 was fun before it induced listening fatigue. Not so much fatigue with the T1 (surprising actually, given your findings) but at the same time there's not too much "there" there.
 
edit: I still bought them, unseen & unheard, on a presale discount. I've learned my lesson, now I keep careful track of new phones. Like the 009
 
The HD650 is definitely the stable workhorse of hifi phones. For a while I thought I heard more detail in brighter phones, however after comparisons it is there, if a tad bit subdued. Also one of the two or three headphones that I like over tubes more than solid state. I love the bass bloom and warmness they get from a Little Dot.

I hope I'm not upsetting you with my impressions.    They are just my opinions. I can be pretty direct and honest at times - this also applies to stuff I own. I'm also very sensitive to ringing - this does appear to vary from person to person.
 
As far as the word "detail," I try not to use that word because of its certain connotations. I prefer to use "clarity" and "ability to resolve low-level musical information". I feel these are more precise terms and don't relate to brightness induced "false" detail.
 
The low-level information is there with the HD650, although it can be hard to make out because HD650 are not as clear sounding as other headphones. My setup happens to be an excellent match with the HD650s. The HD650s performed much better than I expected.

Oh come on now, I've lived with these phones for a while, I'm aware of their flaws
After owning the K701 and comparing it with the K601 I chose to discard it for the lower-numbered model because it was even more harsh, not too mention uncomfortable due to those ridges digging into your head. The K601, while very sparsely used nowadays, gets most of its head time during gaming sessions. For me the shrill highs actually work with simulated-positional audio of explosions and whizzing bullets, they help keep you on edge.
 
The T1 channel differences are quite a surprise. Inner Fidelity also shows two notably incongruent sets of measurements for the pair of T1s, so there may be batch variations. The early ones, like the one I sent you, may be the runts of the litter.