Lengthy but here goes ...
At a former company, rf designers were wrestling
with at times a 3dB difference between one side
of the board and the other. The company was a
CATV place. Waveforms out to the customer were
controlled by the FCC. This concern was crucial.
I picked up on this because I thought I "saw"
something with the problem. I ended up writing
a paper internal to the company whereby I
mathematically modelled by hand a simple trace
over a ground plane between an active source and
passive load with FR-4. What I ended up with was
a frequency response that was flat out to about
800 MHz where a resonance occurred then the
response fell off at about -40 dB/decade. This
roughly correlated with what I expected since
the basic model of the trace involved two energy
storage devices acting as a lo pass filter.
Namely the inductance and capacitance per length
There were of course a lot of variables to that
analysis so that's not to be taken as a hard
fast general rule.
This was a totally theoretical discussion between
me and the designers since they were moving onto
10 GHz ECL design using FR-4. I concluded that
just with the edge rates they were considering
(I forget exactly what they were) they had to
find other material and drop their beloved FR-4.
Besides that, mfring variability of the FR-4 for
the tighter constraints at the fab house was of
concern.
Here's why -
The board must be capable of maintaining a particular
bandwidth in order to support a particular edge rate.