REW Beta Release Pls assist on headphone calibration is trace arithmetic broken?

[Longfellow78]

Hi, I have built a DIY headphone frequency testing setup using a capsule mic. My "head" is a head sized block of wood and covered in paxmate.
Shockingly it actually works pretty well! I have been runnin sine sweeps with various headphones at 90db @1khz normalization and the curves look similar in shape to reference curves that I downloaded from the AutoEQ project github. I am using an AKGk371 and Philips shp9500 as reference.

I am at the stage where I want to calibrate my mic and am running into some problems. I understand the theory - Measure my headphones, add in the reference curve, subtract one from the other, use this as a txt file to act as a calibration filter on the mic when testing, all good.

My problem is that the "trace arithmetic" function in REW doesn't seem to output a curve that is a correct representation of the difference between the two curves. If I try it one way round and put the reference as A and my measure as B and try A-B - one way round (I can't remember which) the graph producted is simply identical to the reference curve, the other way round, it goes haywire and makes some gibberish curve. Applying either of these as calibration filters doesn't work.

I know in principle the calibration filter works when doing this because I manually created a curve of one minus the other by exporting them to google sheets and doing a load of calcualtions. This took about half a day because one file had 55000 entries and the other had 1000 and I had to match them up. Eventually I did it, and it really did work, but I have since made tweaks to my measurement methodology and need to redo it.

Trace arithmetic should just work for my purpsoes and it doesn't I don't know what I'm doing wrong. Is it bugged? would an older version work? Is it something to do with the settings in the popout box, like what "speaker" I should select, I think it defaults to "subwoofer". It's driving me nuts trying to work this out. Or is there another way of cleverly creating a compensation filter that doesn't use the (for me) broken A-B function, like doing an average or one of the other functions.

Any help very gratefully received thank you.

p.s. I'm only measuring one channel at a time. I have tried doing this with left, right, and an average of left and right. Thanks

 

[John Mulcahy]

Not sure I follow completely, but you would usually be dividing the curves rather than subtracting them.

 

[Longfellow78]

Hi. Thanks for response. My understanding is I need to subtract them, so say my curve is 10db above the reference curve at at frequency a, the same at frequency b, and 10db below at frequency c, taking one away from the other would create a sort of - 10, 0, +10 offset to apply to my calibration filter, so it should compensate for the inaccuracies in my mic, and better reflect the reference curve. I could be totally wrong, am a noob so maybe is a better way. As I said when I did it manually in excel it seemed to work as it should, so I assume this is in theory correct (chatgpt told me what to do!) but the inbuilt a-b function won't product an actual a-b curve. Unless I have some wrong setting. Everything else like make an average works. But this doesnt and it's giving me big headache. Thanks

 

[Longfellow78]

When you say divide them? You mean A/B rather than A-B? Would that work?. Maybe the calibration filter doesn't work like I thought in my message above?

I asked chatgpt and it seems to be able to answer a lot of questions about REW and it said it was A-B. I see however are you the author of the software? It's great BTW, so any guidance on my issue would be useful thanks.

Lol. I asked chatgpt again based on you suggesting dividing the, and now it changed its mind as said to divide the measured response by the reference. Haha. Bloody thing. I will try that thanks.

 

[Longfellow78]

Thanks you were right. My £5 rig is now essentially measuring close to a professional setup for headphones. I used the k371 sweep for calibration, then compared my shp9500 using the k371 filter, and my measure was half way between two professional grade shp9500 curves made on a Gras rig. Lol. Thanks

 

[John Mulcahy]

But are you sure you got value for money?