Non-linear techniques with filters

[Martin Czech]

:::Certain of the Electronotes articles suggest the log-filter-expo could
:::produce interesting results since if H(t) = X(t) * Y(t) then expo H(t) = log
:::X(t) + log Y(t) so the excitation and formant are combined additively.

?

You are using the transform pair convolution 0-0 multiplication.

ie.:  g(t)=s(t)**h(t)  | G(w)=S(w)*H(w)

Now you propose to do:

g1(t)=log(s(t))**h(t)

and

g2(t)=exp(g1(t))

First of all I see problems with negative or zero values for s(t).
But that could be fixed using some other function instead of log/exp.

Next I can not see any transform pair any more, since you are using
intentionally non linear functions.

Or am I missing the point here?
I have to look up Cepstrum methods.


Someone asked me: Hey, what's all this analog fuss all about?
Why don't you do it with some digital algorithm?

I see that all this filter stuff is pretty complicated.
And if it comes to tubes, it gets may be even more complicated.
I can not see if there is enough theory to write down an 
algorithm.

So far any nonlinearities are more or less heuristically
build into digital systems.


Or am I wrong again?


m.c.