Measurement averaging for EQ

[dotnet]

Hi, lurker turned new user here!

I've been using REW for a few years as a tool for measuring room response and optimising acoustic room treatment only. I have recently taken the step into room EQ. After a few attempts (that I would consider somewhat successful if a bit inconsistent) I came across a tutorial titled "Average Measurements with REW". I'd like to ask a few questions to confirm or correct my understanding of what's involved.

1. Aside from averaging measurements taken from different locations around the listener's head (which is what I've done, too), the tutorial talks about separate measurements from both speakers. I took measurements of the combined response from both speakers instead. Should I always measure the speakers individually, is this to keep the filters from trying to compensate for things that can't be EQ'ed away, like speaker interference?
2. The tutorial also stresses that measurements need to be time-aligned before averaging, and shows a screenshot of the control panel on the "All SPL" screen. I'm not seeing the "Time align" button, am I doing something wrong, or has this function been removed (I'm using REW 5.18 on macOS).
3. What is the importance and impact of smoothing before or after averaging? Should measurements be smoothed at all, should the resulting average be Var smoothed before EQ?
4. I'm not able to do any modal analysis or calculate resonances using the average. Is this generally only possible with individual measurements?

My setup, if this matters, consists of a laptop running Audirvana, plugged into a USB DAC, plugged into a power amp with two speaker connected. REW is running on a separate laptop with UMIK-1 as input and the same USB DAC as output. The filters that REW calculates (generic equaliser) are entered manually into the AU N-Band EQ available as plugin in Audirvana.

 

[John Mulcahy]

Aside from averaging measurements taken from different locations around the listener's head (which is what I've done, too), the tutorial talks about separate measurements from both speakers. I took measurements of the combined response from both speakers instead. Should I always measure the speakers individually, is this to keep the filters from trying to compensate for things that can't be EQ'ed away, like speaker interference?

With more than one speaker active there will be comb filtering due to differences in their distance from the mic. Better measuring individually to avoid that.

The tutorial also stresses that measurements need to be time-aligned before averaging, and shows a screenshot of the control panel on the "All SPL" screen. I'm not seeing the "Time align" button, am I doing something wrong, or has this function been removed (I'm using REW 5.18 on macOS).

The time align button is a 5.20 feature. It is only relevant for vector averaging, however, which includes phase. For response EQ purposes magnitude averaging is more appropriate, which is what you get using the "Average the responses" button. Time alignment makes no difference to the magnitude average.

What is the importance and impact of smoothing before or after averaging? Should measurements be smoothed at all, should the resulting average be Var smoothed before EQ?

It doesn't make a lot of difference whether you smooth before or after, but after is a little more flexible. Using var smoothing helps to avoid aggressive filtering at high frequencies if applying EQ across the range, but it is much better to limit EQ to the range below a few hundred Hz. Minimal or no smoothing is best at the lowest frequencies when applying EQ to individual measurements since that allows the best match to modal resonances, but for averaged responses Var smoothing is minimally smoothed at low frequencies anyway.

I'm not able to do any modal analysis or calculate resonances using the average. Is this generally only possible with individual measurements?

Modal analysis needs an impulse response, a magnitude average doesn't have an impulse response. The analysis is really only meaningful for individual measurements, however.

 

[dotnet]

Excellent, thank you very much for this information. I've calculated another set of filters with the correction range capped at 400Hz, and the result sounds nicer to me. The speakers got their "voice" back, and the depth of soundstage is restored (it seemed flattened with full-range EQ). Now, this could all be mind games and familiarity, but I'm happy with the outcome.

Apart from the time align feature in All SPL, is 5.20 the version I should be using, or is 5.18 the recommended stable one?

 

[John Mulcahy]

If you can put up with the odd bug V5.20 has the latest features, otherwise 5.19 was the last full release.