I understand the origin, I just don't understand how these parameters relate to those. For instance, what in the world does COM_ Off-Capacitance describe? This surely can't be the small signal cap to ground at that pin because COM is never "off". Could this possibly be Cds of the switch that's off?
It could be COM alone -- they might be able to measure parameters on the die that aren't available once it's pinned out in a package.
However it is they're measuring it, it seems like that's what they're talking about, just the total, versus each at a time.
Fig7 ignores SPDT action, but implies that the other pins are "do not care" during a given test, so the capacitance of a pin includes the capacitance from pin to pin and both pins to ground.
There's also an effect where the capacitance of the channel itself varies (when off, electrons are in fact prohibited from occupying the channel, so its capacitance to substrate is low, whereas, when the channel is "on", it does have capacitance against the substrate, like a regular PN junction), so you should expect C_on > 2 * C_pin. Which seems to be the case.
Here's another thing I don't understand. For a switch that's on (MAX4544), is the small signal cap (to ground) before or after rds?
Yes. Both. A CRC pi filter should give a reasonable model of the frequency response. A more nuanced model might be: C1-R1-C2-R2-C3 (pi topology, extended), plus C4 spanning between input (C1) and output (C3). C1 and C3 are pin capacitances to ground, C2 is channel capacitance to ground, C4 is pin-to-pin capacitance (in the off state). R1 + R2 = Rds(on), which varies with state. In principle, C2 also varies with state (as mentioned), but for most purposes, it should be accurate enough to vary R1 and R2, since that effectively connects C2 to the rest (or not).
Note that it's not especially good (page 5, bottom left). It's already 1% (i.e., -40dB) at 15MHz, and the insertion loss is something like 1.5-4dB (because on resistance is comparable to system resistance -- note the test was done at 50 ohms!). The "off loss" curve will shift up proportionally for higher system resistances: for instance, if you did this for 1kohm audio applications, you'd expect it to cross -40dB at about 2MHz.
The reason why I need to know how to model these for frequency purposes is that I have a constraint on my bandwidth, which is 100kHz and my signal chain has quite a few poles and zeroes, so I have to make sure that my response is nearly flat up to my BW and then I don't really care what it does as long is it maintains a minimum 20dB/dec attenuation.
That should be fine, but alas, frequencies are useless without impedances also! So you'll have to compare those against the switch's impedances.
Tim