optical limiter problem

[Haible Juergen]

A few weeks ago I tried to build a stereo optoelectronic limiter, using
a Vactrol 5C4/2, and I ran into a problem I did not expect.

The idea was:

Vactrols have a (decent) fast time to get low impedance, and a rather
high time
to return to higher impedance. Now that's exactly what is needed for a
peak
limiter. I've built an extra easy one before (thanks Gene for the
idea!): It doesn't even
need a capacitor and just works with the vactrol's inherent time
constants. This
one does some limiting in a BBD delay, and it works great and adds a
nice little
distortion from the remaining regulation voltage ripple. So far, so
good.

But now I wanted to build an ultra-clean stereo limiter to prevent my
DAT from clipping.
I thought that I could simply add a rectifier / capacitor combination
(just like on every
other limiter) to smooth the ripple that was left from the vactrol's own
smoothing.
I really thought this an *easy* project, but it wasn't. A strange
periodic amplitude
modulation occured at the point of limitation.
I think I found the reason for this, but no cure so far.

What I suspect is this:

Because of the very nonlinear R / I characteristic of a vactrol, I have
to use a feedback
rather than a forward compressor topology. So I have a feedback loop,
and inside this
feedback loop one lowpass filter (the inertia of the vactrol). But if I
add an electrical
smoothing as well (i.e. RC-filter), I have a second pole inside the
feedback loop,
and this will cause overshot in the regulation response.

This is hard to optimize:
Decreasing loop gain would lead me away from a rather hard peak
limiting.
Making the electrical time constant faster than the optical results in
almost unchanged
  distortion.
Making the electrical time constant longer than the optical one results
in far too long
  response time.
Making the electrical time constant approximately the same as the
optical one means
  the poles are identical, which will produce the highest overshot.

So far, I have not experimented with a real electrical peak detector in
front of the
LED driver, but I have just connected elkos across the LED of the
Vactrol. So
I have combined an asymmetrical (good!) optical response with a
symmetrical (suboptimal)
electrical respomse. So I am sure there is room for improovement.
Meanwhile, I have put this project aside for some time. But I'd love to
get some hints
from others: Have you been running into similar problems? What did you
do?
How are the commercial optical limiters optimized? (BTW: *are* there
actually
optical hard peak limiters, or do they have limited gain reduction in
the first place ?)

Any ideas ?

JH.