Power compression is interesting too. Higher sensitivity alone doesn't guarantee its absence, either but IMO still really helps.
Our ears/brains are supposedly most sensitive to compression in that 1khz-3khz range. This is right where most tweeters tend to operate, especially in most better-designed speakers with smooth off-axis response.
You can start with a reasonably sensitive tweeter, but thermal compression isn't the
only kind of power compression. The reality is that BL product is pretty notable as the biggest source of power compression in most drivers, and they might compress before even hitting their linear xmax. How much can you expect out of a 3/4" or 1" driver?
Even the arguably best (documented) dome tweeter in the world, a scanspeak 6600, has about 0.2mm of xmax and 7cm^2 of surface area - 0.14cm^2 of volume displacement.
@ 2.2khz, to get 112db @ 1m, or 100db @ 12ft, you need 0.6 cm^2 of displacement. If that's your crossover point, you can probably reduce that by 6db to 0.3 cm^2 of volume displacement. Which is still double the spec for an aircirc. By spec, the
$300+ aircirc is volume limited to 100db @ 1m, which is about
89db @ 12ft. Yes it depends what you listen to, but the point is that power compression is more than just thermal compression.
Compression drivers help, as they tend to have more throw, and notably larger diaphrams, but they get loaded to mediocre horns that have their own issues which surface at high SPLs. The Audiokinesis / Geddes stuff that use optimal waveguides sure ain't cheap. The real issue with compression drivers is that you're using a large diaphram, so now your driver's not operating pistonically, unless it's exotic material IE TAD or JBL's best drivers, and there's really no speaker existing out there that has both a great horn and a great C.D. The closest would be the Gedlee and Audiokinesis stuff, which respectively use some good tweeters - B&C De250 tweeter and BMS 4550 Tweeters in good Oblate Spheroid constant directivity waveguides.
FWIW, though, CSS just came out with an interesting XBL^2 dome tweeter.
http://creativesound.ca/pdf/LD25XFEB11.pdf
So progress is still being made. It states its xmax at a whopping 1.85mm
BTW, so we're not comparing apples to apples. Some companies define xmax differently - based on what I've seen, XBL^2 drivers xmax specs x ~0.6 = ~scanspeak Xmax - so if high end Scanspeak's stated xmax is 0.2, you can probably guess the CSS xmax to be about 1.1mm though there's no discrete math behind that... just a guess.
The CSS' volume displacement is thus at least .58cm^3 which puts your max SPL at 112db @ 1m - or ~101db @ 12ft @ 2.2khz! Max rated thermal power handling is 100 watts nominal, 150w peak.
A burst of 100 to 150w input into the "mere" 89db driver puts you near 109db.
A burst of 90 to 150w input into the 2.5db more sensitive aircirc, on the other hand, would put you closer to 113db, but it can't actually produce that because of its volume displacement (which is 100db based on the spec sheet).
I'd say the following:
Avoid a scenario where you're trying to drive over 130w into your speaker. This much power will give 21db of amp gain, with minimal compression. Once you double that, you only gain 3db of amp, and so forth. The thermal compression rises significantly. So figure out the sensitivity / distance and intended peak SPL (normally ~97-100db in the midbass, and progressively less as you go up in frequency) and set that as a bare minimum. Remember that just because it calculates nicely still doesn't mean it behaves nicely.
Sensitivity isn't everything. Thermal power handling isn't everything. BL(x)/KMS(X)/Le(x) curves aren't everything. The biggest mistake is to look at one spec and judge a book by its cover. Everything as a whole, is everything. But speakers with uneven polars aren't worth your time. Narrow down the speakers with even polar response, and decent sensitivity, and an impedance chart your amp can drive, and a listening window response that isn't a roller coaster and then audition them. The other stuff, you can listen for, because 999/1000 times, it ain't available for you.
