I built a measurement rig with a UMIK-1 and some foam. Here are a few of my first runs from cross spectrum labs, using the 0 degree calibration file for the mic. Any tips/advice/comments are appreciated. Edit: Saw a picture was required: Left to Right: Alessandro MS2i, Beyerdynamic DT250-80, AKG Q701, Sennheiser HD800, Mr Speakers Aeon Sony MDR7506 on the bottom. Aeon w/ Filters inserted: Channel matching (this plot doesn't use the Mic's calibration file, so it's more rolled off than it should be in the top end: ------------------------------------------------------------------------------------------------------------------------------------------------------------ HD800: ---------------------------------------------------------------------------------------------------------------------------------------------------------- Q701: I've added a lot more hashmarks (vertical and horizontal) since this picture, used some polyfill around the mic to try and minimize the effects of the hole (mic hole is close to the diamater of the mic, the polyfill is mostly glued to the surface of the outer foam layer).
Measurements look pretty good but they have a pretty significant but gradual roll off like they are tilted. The CSDs look more accurate. See how the peak on the HD800 CSD is above 200Hz but the frequency response it is way below the bass and mids. The calibration won't show up in the first two tabs of REQ Wizard, only the distortion tab with frequency response implements the calibration file. You can compare switching back and forth but try posting the Frequency response and distortion tab and see if it looks closer to the CSD results.
Here is the link to the raw data if anyone cares to look at it in REW: https://www.dropbox.com/s/tap8l66lze1xs5l/AeonHD800Q701.mdat?dl=1 Added distortion charts so the Mic calibration file could be included in some frequency response measurements. I didn't know that about REW!
Looks like your CSDs at zero time (presuming that's zero at the back wall) are different from your FR profiles. You might be having the issue with your waterfall plots that was mentioned in this thread. If you're using a long (1M) sweep, try a short one (128 k) to see if that fixes the problem. You can look at your impulse response results to see if they change with sweep time. Have a look at this thread for measurement settings, particularly at posts by @ultrabike , @sorrodje , @Serious and @philipmorgan . Serious mentions the sweep timing/CSD issue (how I first became aware of how to fix it) and philipmorgan has a link to a description of his full measurement routine. The rest of the thread has good descriptions and photos of different rigs, too.
Yeah, I tried to use the CSD guidlines in philipmorgan's .pdf . I'm sure you know but it'll look different from the FR at 0 time in part because the the CSD stops at 200hz vs 20hz for the FR. I was using the 1M sweep though, I wasn't aware of the issues with that, philip recommends 256k in his guide (I thought it was just because 1M requires a lot of computing power that lower ones were recommended, but looks like there are more reasons). Will try again with 128 and see what happens. Here's a SS of the settings I used (data from a 1M sweep).
Amplitudes are different at the same frequencies - eg the Q701 is 90 at 1kHz on the FR plot, but 84 on the CSD. A long sweep is recommended for distortion measurements, but not necessary for FR & CSD. Do check your impulse response at different sweep times; the USB interface that gave me problems was ok at 128 and 256 but at 1M gave a huge pre-ring and a polarity reversal.
Here are the results when I tried to compare 128, 256, and 1M sweeps for the waterfall (All left channel only). If the point is to get the initial peak as close to possible as 0, then yeah, 128 is a little better! I need to do a lot of reading about impulse response, still have a ton to learn. Thanks for the help, I really appreciate it.
Distortion measurements look closer to the CSDs. I would say that you have a good baseline for modding and comparing to other headphones.
128 and 256 impulses and CSDs both look pretty good. 1M impulse isn't as bad as I'd imagined - here's one of mine... [/IMG]
Alessandro MS2i. This thing was a major FotM boner about 10 years ago on head-fi. Pretty much an SR-325 I think. Sony MDR-7506 Beyerdynamic DT250-80 ohm (20 years old) Channel Matching was pretty bad on that one I think; I've read that beyer kind of had a bad reputation for that:
Is it funny how Beyer was more or less doing things right back then with a laid back agreeable frequency response, and now they turned that sad face frequency response upside down.
My wife wears those DT250's about 12 hours per day now. She could use any of the ones I have, but those are by far her favorite. They're definitely dark tilted.
Not sure, but even if I drop the floor of the waterfall another 12db it looks about the same above 2k. Left channel looks the same too. I tried listening to some music with them. Don't like them, they sound very veiled. although when I was 16 I thought they were great, maybe my hearing was more sensitive back then. I carried them around with a headroom total airhead and a sony discman. I prefer the MDR-7506 to them for a cheap beater closed HP. Always wanted to try the HD280 in that category.
Can anyone explain more about what the "window" setting in REW waterfall plots does? In Philip Morgan's guide, it says to use 10ms. I notices that lowering this number down to say, 5 ms has a huge cleaning effect on those big, low level ripples below 1k. Are these broad ripples in the -20db range useful information or extraneous? The way I understand it, in CSD mode, it's a sort of high pass filter, except instead of reducing SPL below the freq. cutoff it reduces the amount of time measured. Is 10ms a good standard? Example of a 10ms vs 5ms:
I redesigned my coupling surface to use an old CD covered with felt instead of just creatology foam based on something I saw Purr1n write somewhere. This seems to have counteracted some of the high end rolloff I was experiencing with the calibrated UMIK-1, raising the response above 5k by a few dB. I also picked up a pair of new HD650 just to have since they're being discontinued. Here is my measurement of them: HD650 (new 2018):
Re. your question on window size, have you seen this thread? http://www.superbestaudiofriends.org/index.php?threads/csds-and-what-they-tell-us.65/ Seems it's above 1k where CSDs are most informative. I gather from his posts that Marv tweaks his settings to maximize the information in that part of the range. It looks from your plots above that the 10ms window shows more resolution in the higher frequency peaks (eg 6 & 7.5k) than the 5ms does. I'm interested too; hopefully someone more knowledgable will chime in.
