AW: Re: envelope follower

[Haible Juergen]

	>As is, it is only useable for a single frequency.  HOWEVER, it
seems that
	>you could build a 90 degree constant phase shift network, where
the input is
	>run through two allpass networks that produce 2 signals that
are 90 degrees
	>(+- 0.5 degree) apart for a given audio range.  This technique
is used in
	>the construction of frequency shifters; see the Electronotes
frequency
	>shifter in the Electronotes Preferred Circuit Collection for an
example of a
	>constant 90 degree phase differencing network. Not a simple
circuit, but it
	>seems like it could be used in the construction of a very
accurate envelope
	>follower, in addition to frequency shifting.
	>
	>Anyone on the list who actually knows what I'm talking about (I
really don't
	>- I think my idea would work, but I'm not explaining it too
well), please
	>feel free to comment, clarify, castigate, etc.

I just saw this thread yesterday, so I am sure I missed most of it.
So please tell me if I got it right.
So we want a precicion envelope follower with minimum ripple and
fast response at the same time.
I know that you can build a precicion envelope regulation for
sin/cos oscillators, by using Asin(x)**2 + Acos(x)**2 = A**2.

So did I get that right that this would also work with a hilbert
transformator (90degree phase shift network), as they are used
in frequency shifters ? This would be handy, because I recently
built such a network, and an envelope follower is something still 
missing in my JH-3 modular. So I just have to square the normal
and quadrature output of the filter (for the whole frequency range!),
and get the squared amplitude without much filtering ?

Sounds kind of reasonable, though I haven't done the maths.
Two questions remaining:

(1) Signals with identical spectra can have very different waveforms,
      and thus different amplitudes, depending on the phase relation of
      harmonics. Would the proposed method work nevertheless?
      (I can "feel" it could work, but I'd love to see a good argument
...)

(2) More serious objection: Unlike the sin/cos oscillator application,
     where you can get almost lag-free envelope detection, the phase
     shifter network has delay times necessarily built in. Remember
     the Hilbert transform is non-causal (is that the word?), i.e. it
needs
     to "look into the future" which is of course impossible in the real
world.
     So any approximation needs some delay, and the lower the
     frequency covered, the longer the delay. 
     Sounds too familiar ... 
     So maybe you won't get any better than with rectifier / integrator
     combinations, after all ?

Just a few thoughts - please comment!

JH.