Gyraf Audio Gyratec XXI (G21)
The G21 is somewhat of an elaboration of the G14 parallel-passive EQ - the parallel topology enables multi-band control without running into the traditional problems in multi-banding (i.e. the problem in phase at recombining).
Use carefully - this is in no way a fire-and-forget gadget: Multiple types of interacting nonlinearities are at large, even a few in macroscopic time-domain because of components heating and temperature coefficients. Changing something in one band will result in something else showing up somewhere else.
A classic problem is that multi-banding gives inadequate results because of filter phase shift messing up the recombining of frequencies. the XXI deals with this by NOT band-splitting, but instead dealing with all frequencies in a parallel fashion, only opening frequency-selective channels into the clipping circuitry. A byproduct of this is that transient/peak phase is altered first, only to be real-clipped if pushing the phase won't get the total signal below threshold. Kinda like an audio file hair-remover (work title of the G21 was "The AudioShaver")
As we know, clipping generates harmonics - there's simply no way around that. But what you do with those harmonics will make a lot of difference. The beauty is here, that as the clipping is done at the remote end of a frequency selective channel, the clipping byproducts (harmonics, at frequencies multiple times the clipped) have a hard time getting back out and into the audio, because they are now no more in the right frequency range.
And other than the three bands of clipping, we have an overall clipper that can be set for linear (white), pink -3dB/oct or red -5dB/oct emphasis - enabling gradually lower "threshold" at higher frequencies.
Then there's a fine-trim for output level and a 18dB/oct lowcut at the input - mainly because most of our test pilots were mastering engineers :-)
But why passive?
In essence, we want complete control and predictability in the clipping process - and we want to preserve signal integrity as far as possible. In passive domain, we are more or less restricted to what could be described as 'first order' non-linearities - we do not convolute clipping on top of clipping (i.e. distortion on the distortion products) like it would most often happen in an active circuit under feedback control. It's hard enough to navigate the nonlinear processes already - we do not want to add unnecessarily to the mess.