Message written at 27 Jan 1999 11:27:33 +0000

At teh weekend I indicated that I had doubts about this opcode.  I
think that i shoudl say more, and i have been thinking about it.

At present want it does is to initialise teh output value to zero, and
also the max and min values to zero.  It then reads the number of
samples specified in the period, and it outputs the remembered output
value, and resents max and min if they are greater/less than before.
At the end of each period it calculates the next outout value to the
average of max and -min.

So problems:
a) there is a period delay, and all answers refer to previous period
-- which is kind of expected with direct processing. 
b) If the signal is a constant +1 the answer will be a constant 0.5.

I think that an envelope follower should be idempotent to an envelope,
at least in the constant case.  I think it would be better to use the
largest absolute value rather than this average, and it should be on
the last period samples rather than in sections.  I think I know how
to code that but I do not want to change teh functionality pof old
code if it is being used.  hence my question as to whether anyone used
follow, or had attempted to use it.