Further to the frequency response compensation:

In a world of ideal R, L and C components, you can insert an attenuator
into a transmission line, matched to the line's characteristic
impedance, to get frequency-independent attenuation while maintaining
good return-loss from both input and output ports. If you then modify
that circuit by shunting a capacitor across the attenuator, and lifting
the ground-return of the attenuator and inserting an inductor in series
to ground, and the inductor and capacitor each have reactances equal in
magnitude to the line's impedance at some frequency F (assuming a
non-reactive, constant impedance line here), the system will still have
good return loss at all frequencies, and will have attenuation equal to
that of the attenuator alone at DC, and no attenuation at frequencies
far enough above F. In the transition region, the maximum slope of the
attenuation will be 6dB/octave. That maximum will not be reached for
low values of attenuation. If you cascade such sections, with each
section having a different frequency F and perhaps differing
attenuations, you can make a tailored response adjustment. Cascading
identical sections of high enough ultimate attenuation results in
maximum slopes in excess of 6dB/octave. Obviously trading the places
of the inductor and capacitor give you the opposite shape: more
attenuation at higher frequencies.

I have a little RFSim99 circuit that illustrates this idea--email if
you'd like the file.

It's going to get really difficult to make this work right at high
frequencies, because of the size of the parts and the parasitics
involved. But with modern parts, you might be able to make it work
well out to a couple GHz.

Cheers,
Tom