I killed way too many brain cells last century refining side chains for dynamics processing. The majority of my effort was for companding noise reductions mostly used with limited dynamic range audio paths. A BBD delay line or cassette tape could exhibit a marginal to inadequate dynamic range... expanding that 2x with a 2:1 compressor before the limited path, and 1:2 expander after, could remap a -45 dBu noise floor to -90dBu while doubling the head room.
=======TMI about vinyl NR you can skip this====
Vinyl did not escape untouched, CBS records attempted an ill fated 1:2 playback expander, called CX, compressing the recording before cutting to vinyl. I won't list all the mistakes CBS made, but I got approached through Popular Electronics with the offer of a cover article, and free CX license. Hard to quantify the promotional value of a cover article on a 500k circulation magazine, but I did not say no. One perhaps amusing anecdote, I only had a few weeks to turn the design and I was provided pro forma circuit designs in the CX licensee package. Not heavy lifting after all my NR experience. I cranked out my version of CX and corrected a mistake I found in one time constant of the pro forma design. I delivered my article to Popular Electronics on time and notified CBS of the mistake in their design advice. My cover article was scheduled to run in the Christmas- December issue (not too shabby). At the New York AES show that year, literally days away from the issue printing I visited the URIE booth, since they made the 2:1 encoder for CBS. After introducing myself, they said "so you're the guy who found the mistake."
CBS told URIE to change the time constant in their encoder to agree with the mistake because two other licensees had manufactured thousands of decoders with the wrong time constant mistake. Of course CBS did not bother to tell me, so I was about to publish the only "wrong" CX decoder design since they changed the specification over night. I frantically contacted the editor at Popular Electronics and made my design agree with the changed time constant in the nick of time. I still have a stack of CX records CBS gave me as a licensee, but screw them....
======
Of course to compand an audio path well you want to do it as transparently as possible and minimize audible artifacts from the tens of dB gain manipulations.
For a somewhat longer list of side chain features (still incomplete) beginning with the obvious;
-attack time : you want this fast enough to pass sudden transients cleanly, but not so fast that you can hear brief transients modulate the noise floor.
-release time: you want this slow enough that the side chain does not try to release during a single cycle of a LF waveform.
-hold: Ideally you hold the side chain at a stable gain so there is no signal modulation occurring.
-adaptive or signal dependent time constants: Ideally you want faster attack for large signal level changes, and very slow for small level changes. There are too many ways to accomplish this to list.
-HF pre/de-emphasis: widely used in multiple NR systems this takes advantage of the relatively modest amounts of HF content in natural music. Boosting HF before and cutting it again after provides some "free" NR, but just like the real world there is no free lunch, so HF pre/de-emphasis reduces HF headroom.
Within this short list of basic parameters there are subtle psycho-acoustic relationships in human audition that can be capitalized on for transparent NR. For just a couple examples, in human audition loud sounds can mask quiet sounds making them less objectionable unless playing by alone. Another human hearing trick is that our ears sensitivity changes slowly after a loud sound stops, so this provides a brief window where larger gain changes can be concealed. There are other tricks like this.
====TMI about tape NR you can skip this too =====
We are all (some of us are) aware of DBX and Dolby tape NR, I sold two different Tape NR kits and this is my best later design.
This design uses even more tricks than I listed. I have shared this schematic here before so there are more details in the past discussions. To better understand this you might want to look up a NE572 compander chip... that IC contains a fair gain element and two rectifiers per that can be used separately. In fact back in the 80s I used a 572 inside the side chain for a commercial compressor I designed (LOFT-Phoenix Audio Lab). I just did a search and see somebody new is using the "Phoenix" name for audio products.... the new Phoenix like the legend arose from the ashes (but somebody else's ashes. )
JR