Help with mic calibration effort...

[TNT]

Hallo!

I'm trying to calibrate 2 microphones that I built to one commercial good microphone. So I have three measurements made in the exact same location with the same sweep stimuli. Absolute linearity is not important - I regard Ref as linear

My problems:

- There is quite a big gain difference so I need to "move" the level of the ref say -30 dB (or vvv). When I do this it looks like the shape of the Ref mic trace changes shape a bit - why is this? It should remain but only change level.
- When I try to calculate the difference between the ref and one of the mics using Filters A magn (Ref) over B magn, (Mac L) I get too high values - what might I do wrong here. They don't end up around 0 and the differences are too big I think. It seems like REW uses the original level of Ref somehow as the result ends up around 30dB and not around zero.

I attach the .mdat file with the 3 measurement to aid my question...

I would ideally like to compensate for bot FR and Phase for both mics individually compared to Ref. I do use rePhase...

My aim is to use the diff files for L and R to convolute my recordings with the correction .wav files using CamillaDSP at 384 ksps / 24bit.


//

 

[John Mulcahy]

I don't see any change of shape when adding an offset to the REF mic measurement. Remember to use "Add offset to data" after you move it.

All the measurements have abrupt high frequency cut offs, the REF cuts off at 10 kHz and the other two at 13 kHz. That's probably due to failing to capture the end of the sweep due to replay latency, use the acoustic timing reference to fix that.

 

[TNT]

Thanks for the answer - I managed to priced the files. There are many ways to do it it seems..

The below video is was my experience and caused the question about changing shape...

//

 

[John Mulcahy]

Changing the gain of one channel in an alignment sum isn't the same as adding an SPL offset to a measurement. One of the measurements rolls off at 10 kHz, the other at 13 kHz, so their sum is going to look odd if you are changing the contribution of the one that stops at 10 kHz.

 

[TNT]

I clearly don't know what I'm doing. Where does it say that it's a "summing" action? Didn't get that.

Anyways.. I should use the "Align SPL..." tool in this case I suppose?

And "Remember to use "Add offset to data" after you move it."!

//

 

[TNT]

Is it at all possible to preserve some phase data through this process. It seems that the generated difference traces dont contain such data - maybe its not possible... my math fails me again

//

 

[John Mulcahy]

If you divide magnitudes, there will not be any phase.

 

[TNT]

OK - so I need to redo the measuremnts with the FR compensations in place and then do phase corrections on these... right!?

//

 

[John Mulcahy]

The chances of you getting meaningful phase compensation data are close to zero. You definitely cannot do it with measurements at the listening position, that is also true of the magnitude compensation. It is hard to do it even with carefully collected nearfield measurements from a suitably wideband source far from boundaries or in a test chamber. If you feel you must have phase data for the corrections you can create a minimum phase response from the compensation file, but if the compensations are invalid due to the measurement conditions the phase data will be worse than useless. Mics have negligible phase contributions anyway, other than small shifts at their lower and upper roll-offs and around any resonances.

 

[TNT]

I do my calibration trials with substitution method. I place the ref 40 cm from the source and mark the tip of the mic very careful with a thin thread with a knot hanging from the roof. Do a sweep and record with the reference mic. Then place the mic to be calibrated in the exact same position and rerun the sweep. These are the traces in the attached file.

Should this not get usable input for a calibration? At least for FR. Maybe phase is much harder... and not necessary.
Both types mic are omni.
Is the stimuli really critical as long as it is reproducible with a high degree of sameness? A bit of reflection even if it is the same for all measurements?

//

 

[John Mulcahy]

You mean the file you attached in the first post? Those measurements were incomplete (stimulus not fully captured), all the measurements look to have gone through resampling (was the sample rate for the interface in Audio Midi Setup the same as the rate in REW) and they have large close reflections, the L mic and R mic particularly. Perhaps the reflections are partly from whatever you are using to hold the mics. Mics differ little in their LF response, you may get better measurements by placing the mic near field to a tweeter.

 

[TNT]

I hope I'm correct in that: FR trace linearity is not important as the "krocket" stimuli is the same for all measurements.

The stimuli has gone through signal processing as it is excited by my DSP active 2-way stereo system - using only one channel at the time for measurements!

But, should it matter really??

By doing A magn (Ref mic) over B magn (Mic) - the difference is extracted whatever stimuli - no?

I actually did the full go-around... took the error, convolved it with the mic - deducted the ref mic FR and got to more or less a flat trace. I.e. got my mic to behave like the ref. All this in REW - remains to be verified in reality.,..

So it "worked" as I anticipated on the bigger picture.

//