I've been worried about the high supertweeter level (+15dB) actually causing some problem in it's own right. Normally SPL levels above 20kHz are very low. But you could have some ultrasonic noises, and even at -15dB such noises could cause huge signals in the supertweeter drive.
Something like that seemed to be happening when I played the song Atom Heart Mother Suite by Pink Floyd. The supertweeter channels appeared to be clipping as the red lights were lighting on the Behringer.
So I spent the next 12 hours working on adjustments to fix this, only to later find myself unable to reproduce the original problem to see if the adjustments really help. The original problem might have been caused by some kind of gremlin (like loose connection, or maybe I mistakenly turned off LR48 crossover for supertweets). But I still think the adjustment is worthwhile and makes for better sound, so I keep it anyway. (I had been thinking I would try dynamic EQ instead, but since I can't reproduce the original problem anymore, I'm sticking with the new changes.)
Mainly what I did was add a low pass parametric EQ at 20kHz in the supertweeter drive on the Behringer. I choose 12dB/octave and -15dB level (I'm not sure why Behringer requires you to set levels for a low pass; I think it actually makes it like a shelf rather than a low pass.)
So I have both lowpass and highpass at 20kHz. And I think this *is* the right way to set up a supertweeter: with both highpass and lowpass. (The highpass is a Linkwitz-Riley at 48dB/octave at 20kHz).
With just the LR48 highpass, the level at 20kHz is actually 6dB lower than the maximum (which it approaches asymptotically above 20khz, being pretty much there at 25kHz, just as could be predicted). Now what good is that? Perhaps the supertweeter needs the extra 6dB boost above 20k to maintain flatness, though I would think not. But whether it needs it or not, that is just the way using LR48 works out.
And it's highly undesirable to have the 20-40kHz range be boosted that high, it's just asking for trouble from ultrasonics, etc. What we'd really like is response that keeps on rising until 20kHz, and then doesn't rise anymore, maybe even falling back slightly.
Just to get a sanity check on the voltage sent to the supertweeter, I put my Fluke 8060 and 804 meters on it, and drove my system using my Sound Technology ST-1400B distortion analyzer/generator. I set the level for about 1V output at Acoustat speaker terminals. At 20kHz, the Elac drive wire was about 2.5 volts, and it peaked around 30kHz at 5 volts (!), 14dB higher than Acoustat drive level.
After adding new 20k lowpass, the peak occurs near 20kHz at about 1.8V, then declines gradually. I also changed crossover point for the LR48 high pass to 18kHz. Although I had recently changed levels down to +10 on the Behringer, I am back to +15 with these adjustements (which lowered response quite a bit by themselves).
Not only is this "safer", I think the supertweeter sounds better not being driven so loudly as it was. That's exactly when it makes the "metallic" sound. The new highpass/lowpass combo seems to prevent that.