Difference between curves?

[Rick Auricchio]

It would be nice to be able to display a pair of curves in the Measurement screen, then click a "difference" curve. Similar to the "Separate the Traces" option.

[salvasol]

Rick Auricchio wrote:

It would be nice to be able to display a pair of curves in the Measurement screen, then click a "difference" curve. Similar to the "Separate the Traces" option.

What do you mean??? ... to compare measurements???

I know that you can load any graph from previous measurements and compare them (they show in the window where you separate the traces)

[Rick Auricchio]

Exactly. It would be nice to see the difference, to compare two curves. I would be comparing responses of two speaker systems (especially two of the same model).

I know I can view multiple curves at once and visually compare them. But if the program could do the calculation and draw a third curve---the difference between the two---it would be great.

The difference-curve would display at its absolute dB value. Perfectly matching curves result in a difference that's a horizontal line at 0dB. Any variation between the original curves (call them A and B) shows up as bumps above/below the 0dB line.

We might need to be able to choose whether the difference is "A vs baseline B" or "B vs baseline A." This allows me to say "Our reference speaker is B, and here you see how A differs."

With due respect to the author, the algorithm is:

for each point along the measurement

if (baseline == curve A) then
diff = curve A dB - curve B dB
else
diff = curve B dB - curve A dB
endif

plot new Y point at diff
next

[Wayne A. Pflughaupt]

I know I can view multiple curves at once and visually compare them. But if the program could do the calculation and draw a third curve---the difference between the two---it would be great.

The difference-curve would display at its absolute dB value. Perfectly matching curves result in a difference that's a horizontal line at 0dB. Any variation between the original curves (call them A and B) shows up as bumps above/below the 0dB line.

It's not jumping at me what the usefulness of such a function would be...

Regards,
Wayne

[Rick Auricchio]

Well, I admit that one can visually see the difference between two similar curves.

But having the program compute the difference curve makes it easier to see variation. This is especially true if the original curves "cross." For example, curve A has some areas that bump above curve B, and some areas below.

Having a difference curve "A vs. baseline B" lets you immediately see the bumps above and below the 0dB line.

I actually created a difference curve, by exporting curves A and B to text files. Then a pass through Excel to create the difference text file, then importing it back into REW.

In the diagram below:
- Red curve is the baseline (our "standard"), against which we're comparing the green.
- Blue is the difference: How green differs from red.

Since the green is deficient from about 780Hz upward, it shows as a few dB below 0 (negative). Where the green level exceeds red, we see positive values.

[Wayne A. Pflughaupt]

But having the program compute the difference curve makes it easier to see variation.

Sure, I get that. I just don't see what it's useful for. What would you do with that information?

Regards,
Wayne

[DrWho]

To create a difference curve with REW, you need to first take a measurement and then save it is a calibration file.

Then, you need to take the measurement again using the calibration file you just made.

The only downside is that you must create the calibration file and load it before you take the second measurement. I don't know of any way to do it after the fact.

Having an extra feature put into REW would be real nice though. Being able to measure differences probably wouldn't be very helpful for most of the REW users, but it is very useful for determining allowable tolerances when doing xover or driver design/OEM, etc...

[Rick Auricchio]

DrWho wrote:

To create a difference curve with REW, you need to first take a measurement and then save it is a calibration file. Then, you need to take the measurement again using the calibration file you just made.

Ooh...very nice insight. Thanks.

Having an extra feature put into REW would be real nice though. Being able to measure differences probably wouldn't be very helpful for most of the REW users, but it is very useful for determining allowable tolerances when doing xover or driver design/OEM, etc...

Agreed. It's nice when doing A/B tests. I had a better (simpler) example, where I tested the same speaker using a different power amp. One of the amps was much flatter below 35Hz. There's where the difference curve would be good.

But with tab- and comma-separated export, we can use outboard processing (e.g. Excel or perl) to grind out the difference. Having TSV and CSV options may be a handy, easy feature.