There are two main points to bear in mind while making measurements of a
compressor's "Amplitude Transfer Function" curve.
1) It's NOT a transfer function in the same way an EE normally thinks of
one. Rather, the axes are normally labelled in dB, not volts, in order to
dispense with conceptual issues like "lower part of a sinusoid". Think "level"
not "instantaneous value of a voltage wave", and you're on the right
track.
It also means the lower left-quadrant is NOT symmetrical with quadrant #1.
Minus infinity (in dB) resides in the lower left quadrant, whereas the knee,
say, in an above threshold compressor, manifests (e.g., appears) ONLY in one
place on the resulting DUT's characteristic curve.
In contrast, the operation of a crude, Plain Jane "clipper" is often described
by plotting its transfer function against axes marked with linear volts, in
which case the negative half of the sinusoid function gives rise indeed to a
knee in BOTH quadrants, not just in quadrant 1.
2) The original poster's question does indeed relate to measuring the
steady state response of the DUT to the amplitude level of an input signal, as
another reply has already indicated. I'll merely add that a well designed
measurement system, like the "Audio Precision" company's products (System
ONE; System TWO; etc.) will wait for the DUT output to settle prior to storing
the measured output as a data point (making life easy).
If you roll your own measurement setup, just wait for the DUT's output to
settle, say, once having bumped the input amplitude by some amount (typically 1
or 2 dB), prior to recording your observation as a data point. Equally, in an
automated measurement setup, build this logic into the process, much as Audio
Precision does.
As to the answer to the OP's question: be sure to use a steady state
sinusoid of a mid-band frequency. Then be prepared to await the DUT settling
time after each step in stimulus amplitude (i.e., a stepped-swept amplitude
stimulus). Otherwise, if using a continuously swept amplitude stimulus, be sure
to use a sufficiently slow sweep rate, namely: one that is slower than the time
constant of the DUT (i.e., be aware of the DUT's attack and release times,
and their definitions).
Get a copy of Bob Metzler's Handbook of Audio Measurements from Audio
Precision's web site library section. He's the man! These guys were
the first to suggest and use the unit: "dBFS", and got its definition adopted by
the IEEE. Read the story in a back issue of their newsletter: "Audio.tst".