FabFilter

Hi,

After watching multiple times FabFilters's video on Upward Compression (link below), I'm still very confused about some of the transfer curves when doing upward compression with ratios greater than 2.2:1 and the range set at +30db.

As shown in the video, the curves go up, then get a sharp dip around the threshold (the curves suddenly go down), and then go up again. This is referred by Dan Worrall as "over-compression".

That means the overall order of loudness is not preserved across the input and output signals. In mathematical terms:
I(A) <= I(B) <= I(C) does not imply O(A) <= O(B) <= O(C)
with I meaning I(x) being the loudness of input signal x, and O(x) being the loudness of output signal x.
In some cases, you'll have O(A) > O(B) and O(B) < O(C).

To me that seems like a very unexpected behavior because:

1. none of the other types of compression in Pro-MB have a similar behavior (ie a curve that goes down at some point).

2. nothing mentions that in the Pro-MB manual. The curve next to upward compression shows a regular a curve with a section that is flatter but still goes up around the threshold.

3. the ration value at which this happens is dependent on the range value. The only way to know about is to use PluginDoctor.

I have a few questions about this:

Q1. When is this sort of curves with a dip are actually musical?
Why would this sort of curves would ever be preferred to a transfer curve that preserves the overall ordering of signal loudness across the input and output? Are there resources online other than Dan's video that talk about this (and ideally refer to this as over-compression as well)?

Q2. How come none of the documentation talks about this? How are we supposed to know that this type of over-compression is actually happening if it depends on a varying combination of ratio and range?

Thank you very much for your clarifications!
Julien

Video ink: youtu.be/8z4X8WZv33s?t=243