stupid envelope follower idea

[Don Tillman]

   Date: Wed, 22 Nov 2000 08:40:50 -0500
   From: Terry Michaels <104065.2340 at compuserve.com>

   Don:
   :::Imagine splitting the input signal into three bands with three
   :::envelope detectors:
   :::  20 to 80 Hz, slow envelope detector
   :::  80 to 320 Hz, medium envelope detector
   :::  320 Hz and up, really fast envelope detector
   :::
   :::And summing the results.

   One problem with this approach is the various band filters are bound to
   have different group delay times, so waveform peaks passing through the
   filters may not line up timewise afterward.  Since harmonics of the input
   signal will not coincide after passing through the band filters, they will
   not add up algebraically, and you will then detect an envelope that will
   have a different shape than the original signal had.  This wouldn't happen
   with sine waves, but it likely will with harmonically complex signals.

What would be the worst case input signal for this situation?  Maybe a
rectangular tone burst?  How bad would it be?

Envelope detection is not an exact science.  There are all sorts of
interesting approximations we can use; peak detection, RMS detection,
average each half cycle, separate rise and fall times, pre-detection
filtering, post-detection filtering, etc.  

Since we're in a musical context: Whatever works best.

My own creative musical instrument philosophy would be: Okay, give me
an envelope follower with multiple outputs, each with a different
detection scheme.  If you use each output for a separate function
(oscillator, VCF, whatever) your input source can find expressive ways
to control the differences between them.  Your input source can do one
thing for more peak detection and do another for more RMS detection.

  -- Don

-- 
Don Tillman
Palo Alto, California, USA
don at till.com
http://www.till.com