Correct way to compare two microphones with REW

[TheDenix8]

Hello!

I'm trying to calibrate two handheld recorders by comparing their frequency responses to UMIK-1's frequency response using REW. I have made 17 measurements in different places of the room facing different directions and now want to compare them. the measurements were done by putting UMIK-1 on top of recorder so that they are both at almost same location and facing the same direction. Since the handheld recorders are stereo, I decided to measure each mic individually while recorder is pointing in the same direction as UMIK-1. I know I am doing it in imperfect conditions (living room etc), but I want to get rough feeling of frequency response of those handheld recorders.

What I'm confused about is:
- What average should I use when averaging left and right channels of handheld recorder, RMS or dB average?
- What average should I use when making an average of all measurements for final result? (I assume RMS)
- Does smoothing have any effect here, should I use smoothing before averaging?
- What SPL alignment settings are the most optimal to align all measurements before averaging? Currently I use centre 2kHz 3 octaves span
- Should I use Time Align? When I press it nothing seems to change
- When I use Trace Arithmetic and subtract the two measurements, I don't get the actually difference but some curve I cant explain (brown is result of A-B):

Why could that be?

Thanks for any help!

 

[John Mulcahy]

For substitution measurements you should only have one mic in place at a time. All averages should be RMS. Smoothing will help reduce the effects of room reflections on the responses. Typically responses are aligned around 1 kHz. The differences are extracted by dividing the averages. Arithmetic is done on the underlying linear data, not the dB values you see. The result of the division will be the same as if you had subtracted the dB values. The averaging is all magnitude-only so time alignment doesn't matter.

 

[TheDenix8]

For substitution measurements you should only have one mic in place at a time. All averages should be RMS. Smoothing will help reduce the effects of room reflections on the responses. Typically responses are aligned around 1 kHz. The differences are extracted by dividing the averages. Arithmetic is done on the underlying linear data, not the dB values you see. The result of the division will be the same as if you had subtracted the dB values. The averaging is all magnitude-only so time alignment doesn't matter.

Thank you very much! I have been losing my mind over why subtraction isn't working as I expected I went to manually subtract things via JavaScript in the end haha
Which smoothing would be the most sensible in this situation? I have been using Psychoacoustic

 

[John Mulcahy]

Var smoothing is probably good for the comparison, you could also try applying a frequency-dependent window to the measurements before averaging them to reduce reflections.