Originally Posted by Arleus
So, I took a page from Twinster's book (you cunning fox, Twinster), and built a budget measurement rig from a cheapo SPL meter I got from amazon (probably too cheap - it's not great, and I'll go into why a little later). I was lazy though, and instead of mounting it in a box, I just duct taped a CD to the aperture, and covered the CD in paxmate. I've been measuring a couple of mods, and some of my other headphones over the past few evenings, and I thought I'd share, in case anyone's interested.
Forty test tones were produced in MATLAB, logarithmically spaced between 20Hz and 20kHz (code in the spoiler section below - in the interests in openess, I'm going to post my code in case anyone wants to use it to generate tones in Octave or MATLAB. I know nobody will, but I feel better posting it anyway). I played each tone in turn, with the 485Hz tone set to 90dB, and recorded the SPL measured on the meter, set to 'slow' speed, and C weighting. I then switched sides and went through the same process for the other cup.
Warning: Spoiler! (Click to show)
Fs = 44100; % Sample rate in Hz
maxt = 10; % Maximum length of the tone in seconds
notones = 40; % Number of tones
amp = 0.9; % Amplitude of the tone
% Generate frequency vector
f = logspace(log10(20), log10(20000), notones);
% Generate time vector
t = [0:1/Fs:maxt];
% Loop through each frequency
for n = 1:length(f)
y = amp*sin(2*pi*f(n).*t);
if f(n) < 100
name = ['000' num2str(f(n)) '_tone.wav'];
elseif f(n) < 1000
name = ['00' num2str(f(n)) '_tone.wav'];
elseif f(n) < 10000
name = ['0' num2str(f(n)) '_tone.wav'];
name = [num2str(f(n)) '_tone.wav'];
wavwrite(y, Fs, name);
After this point, I 'correct' the data by setting 485Hz to zero, and removing the C weighting curve - this is one of the dodgiest bits I do, and if someone would like to tell me that I'm wrong to do so, please let me know. It is my understanding, however, that the meter will 'artificially' reduce the dB value at low and high frequencies to account for human hearing at those ranges. I wanted the raw numbers, not the weighted ones. I got the formula for the weighting curve from Wikipedia, and subtracted it from the recorded numbers.
I then put a cubic spline through the points (using MATLAB's spline function, upsampling by a factor of 10) to interpolate between the rather coarse 40 frequencies I recorded. My SPL meter does have an analogue output on it, so in future I might see if I can chav (read: steal) an oscilloscope, work out how the devil it works, and record a continuous sine sweep so I don't have such a coarse set of data points... but if I'm going that far, I might as well get a better meter too.
Then it's just a case of plotting. For reference, here are my results for my K702 and RS1i to give you some idea of how my graphs look. It's generally not a good idea to compare frequency responses from one person's measurement equipment to another person's results... especially if the equipment is home brew.
These look, I have to say, surprisingly good I think. Okay, so first thing to note - I don't think you can trust my results below 50Hz or so. All of the headphones I've measured show a very similar slope of roll-off below 50Hz, and I suspect that's the meter that's at fault (considering I've 'corrected' for the C-weighting). Compared to Tyll's results for the RS1, the RS1i looks incredibly flat through most of the midrange, and I don't see any 100Hz bump that he has for the older model headphone. At this point, I don't know if it's really there and I can't detect it, or if the RS1i doesn't have one. Upper midrange looks like classic Grado to me though, so that gives me some confidence. That channel imbalance at 2k? I didn't realise I had one, but it looks pretty big to me - I did double check it.
The K702 response again looks pretty reasonable, though a slightly different shape to Tyll's. Maybe I could chalk it up to the HRTF he uses, and my complete lack of one, but frankly it's probably because I bought cheap equipment and built it from gaffa tape and dreams.
Given those two results as a baseline, what do the T50RPs look like that I've modded? Well, I have to admit at this point that I don't have a stock pair to measure, so take these results with an ever growing pinch of salt.
Plasticine on back and walls of cup (I'm too cheap for dynamat), paxmate covering it all, including the centre screw post. Stock bass port felt in place, 3 slots covered by electrical tape. A square of floppy craft felt, maybe 2mm thick, resting on the back of the cup (not glued in place). Stock white fabric on the cup side of the driver square, 3 sausages of tungsten putty (18g per cup) surrounding 3 sides of the driver housing. Plasticine in the seal between the baffle and the cup. On the ear side, the felt over the driver has been removed, and a paxmate lifter added under the rear-side of the pads (no other paxmate or putty on this side). Shure 840 pads with the centre foam removed.
Holy channel imbalance Batman! I would describe the sound as a bit confused, quite dark, somewhat congested in the upper bass / low midrange. The 1k resonance that's been plaguing my builds is also clearly visible, and quite annoying on certain songs, especially some female vocals that hit this note. At least the sound is fairly open, and the midrange, though imbalanced and a bit bumpy, has that amazing liquid quality that I can only assume is what everyone raves about when they talk about orthos (I don't own any other orthos). To my ears, the bass lacks extension - remember that you can't trust my graphs down there though.
No plasticine in the cups. Paxmate on the bottom of the cups. Bass port felt removed entirely, and the ports left uncovered by paxmate on the inside of the cup. On the outside, electrical tape covers 3 slots. Half of a UK sized cotton ball in each cup, teased into a square the size of the bottom of the cup. The same tungsten putty surrounds the driver housing as in Mod 1, still got the plasticine seal between the baffle and the cup. Ear-side is identical.
Oh man, much much better. Firstly, I sorted the channel imbalance - that makes a big difference to the cohesiveness of the sound, I think. The mods have also really smoothed over the midrange, making it fairly flat now. The sound is tighter, significantly faster too, partly due to there being less mid-bass bloat, and partly due to the lifted high frequency response (this one is only down at -10dB at 5kHz, whereas Mod 1 was down at -20dB. There's a dip immediately after 5k on this mod, but it doesn't go lower than Mod 1 does). I've not noticed a sibilance problem with the 8k peak yet, but I think I'm fairly tolerant to sibilance compared to some people on these forums. The bass sounds to be better extended too - if I play my test tones, I would say things are pretty good down to about 30Hz, then take a bit of a dive after that, but nevermind.
In conclusion, I'm pretty happy with Mod 2 (most of the ideas for which came from bmf of course). If I do any more mods, I'll make sure to measure them, and if I think the mod is worth something, I'll post my measurements here. Whilst I would love to take a systematic approach, and measure each mod individually, I frankly don't have the time to be able to do this, plus I doubt that the mods are linear, in that if you stick two mods together, I doubt that the results will simply be the same as the sum of the individual parts, if that makes sense. If anyone else is toying with the idea, I encourage you to go and get your own SPL meter and give it a go - it's easy, and it seems to work surprisingly well. Was a great idea Twinster!