> The bands should be ∗in∗ phase with each other, no? Is there something
good
> that happens if you invert alternate bands at the output or something...?
They are ∗never∗ in phase. With inversion you decide for a 180deg shift,
i.e. weather you rather like your phase difference between 0 and 180deg
or 180 and 360deg. It's not like choosing "minimum difference" either. You
have to make extensive calculations (with complex numbers, i.e. amplitude
and phase) and see if you like the overall result. It's mostly a choice
between slight peaks or slight notches (or heavy peaks and deep notches,
if you started the wrong way) - don't even think of "flat" response.
OTOH, these irregularities will give a certain character to your filter bank
or vocoder - which is what we want with analogue designs - so my
_guess_ about that W.C. topic would be that she just liked more notches
for the vocoder than for the filter bank (or the other way round).
But this is really just speculation.
JH.