REW trace arithmetic - wrong?

[Londek]

I'm trying to use Trace arithmetic functionality of REW in order to calculate on-axis response of my condenser mic. YES, yes I know, 34mm true condenser capsule is "nowhere comparable" to 10mm electret capsule (of my UMIK), I'm just trying to get rough measurements, for fun. However, for some reason division function doesn't work correctly (as of my understanding), up to 1khz responses literally overlap each other, but like in result of the function I can see random 1db peaks/dips, again, visually responses LITERALLY overlap each other. Also trebles are off by ~1db too. Can somebody explain to me why does it work like that?
Also, what is Regularisation parameter.
Thanks

 

[Blumlein 88]

I don't know, it seems to work for me though I don't get a perfectly flat line even running everything the same twice. It gets pretty close that way. I'm thinking environmental noises could be the reason. Also, I've found you must be in the same spot. Your body being in a different place will slightly alter things like you describe. Plus you might try some smoothing if you are using an unsmoothed plot. The mic stand must be rigid as well.

I don't think I've seen Regularisation parameter. Where are you seeing that?

If there are published curves for your microphone, I've found those are usually about correct. Using web plot digitizer you can turn the graphical FR of your microphone into a .csv calibration file for your microphone.

 

[Londek]

It's in trace arithmetic window itself, looks like it's division exclusive. REW V5.20.13

I think I found out why I get such weird results, trace arithmetic doesn't take current smoothing into account, meaning it also calculated graph for +/- 5db peaks/dips caused by reflections. Would be really nice if devs could make it run with "what can we see on screen" (with checkbox field), as the primary reason I smoothed graphs was to average out reflections.
I fixed the issue by exporting and then reimporting smoothed graphs and doing operations on them.

 

[Blumlein 88]

Using v5.20 I don't have an issue. Here is one with 1/6th smoothing and no smoothing. As soon as I change smoothing in the controls menu and then apply smoothing, it changes the smoothing on the result of the traces.

This was two microphones of the same type. I think most differences are noise in the low end and slightly different positioning at high frequencies. Still pretty close for non pair matched microphones.

 

[Londek]

As soon as I change smoothing in the controls menu and then apply smoothing, it changes the smoothing on the result of the traces.

My theory is that arithmetic is done on no smoothing and then the result is being smoothed. It's a discreet difference, but it's there (particularly in aggressive peaks/dips), somewhat comparable to averaging averages vs averaging elements of all averages
Maybe it was introduced as a bug in >V5.20

 

[Blumlein 88]

My theory is that arithmetic is done on no smoothing and then the result is being smoothed. It's a discreet difference, but it's there (particularly in aggressive peaks/dips), somewhat comparable to averaging averages vs averaging elements of all averages
Maybe it was introduced as a bug in >V5.20

What would you want it to do? Makes sense it works off of no smoothing and then applies smoothing to that result. I'd think if you smoothed it and then did the math it would be less accurate even if more comforting to see.

 

[Londek]

What would you want it to do? Makes sense it works off of no smoothing and then applies smoothing to that result. I'd think if you smoothed it and then did the math it would be less accurate even if more comforting to see.

Because random 3db peaks in low range, again, visually responses LITERALLY overlap each other (on 1/6 smoothing), 3db is really huge difference. While the "Regularisation" helps taming low-end it also drops >2khz response by 3db. With my fix implemented all the issues are eliminated + there's no need for regularisation, it just works as it should.
If I recall correctly UMIK-1 calibration are 1/3 smoothed, my point is that virtually all microphones are flat(ish) with mostly low-q variations

 

[JohnPM]

No, it is not a bug, that is how it is supposed to work. The help explains trace arithmetic. If the measurements have impulse responses the operations are carried out on those impulse responses, with their respective IR windows applied. Fractional octave smoothing has no effect on the impulse response, it only affects the displayed frequency response. You could reduce the effects of reflection higher up in the frequency range by applying a frequency-dependent window to the measurements.

 

[Londek]

No, it is not a bug, that is how it is supposed to work. The help explains trace arithmetic. If the measurements have impulse responses the operations are carried out on those impulse responses, with their respective IR windows applied. Fractional octave smoothing has no effect on the impulse response, it only affects the displayed frequency response. You could reduce the effects of reflection higher up in the frequency range by applying a frequency-dependent window to the measurements.

Thank you for response