envelope follower

[Grant Richter]

Has anyone messed with the "Cepstrum" concept. i.e. take the Log to the
signal, filter it then take the anti-Log? This converts the frequency axis
to time (according to Bernie Hutchins) and makes the envelope components
(long time) more easily filterable from the signal components (short time).

----------
> From: Don Tillman <don at till.com>
> To: synth-diy at horus.sara.nl
> Subject: Re: envelope follower
> Date: Wednesday, October 01, 1997 1:22 PM
> 
>    Date: Thu, 30 May 1996 14:52:27 +0200
>    From: Rene Schmitz <uzs159 at ibm.rhrz.uni-bonn.de>
> 
>    Nononono! The thing is you measure an amplitude of a signal whose
>    frequency you *know*, and you design the circuit for exactly and
>    only that frequency, and it has to be a sinewave. (So it is more an
>    AM demodulator; fixed carrier; variing amplitude) This was what was
>    assumed in the earlyer post. (Divided audio bands...)  Ok, I should
>    have said this more clearly, but it can be seen from the maths!!
>    Erm, thats what I meant by *discrete* frequency.
> 
> I too was confused by the initial description... Splitting an audio
> signal into seperate bands immediately made me think that the
> post-rectifier filters were going to be optimized for each band, which
> by itself would work pretty well, but that's not what was being
> proposed.  (Although, the proposed circuit's performance improvement
> should probably be compared to that solution.)
> 
> The squared-90-degree approach is very cool, and you can implement
> that approach without splitting the signal into frequency bands by
> using a 90-degree phase difference network (like on frequency
> shifters).  It doesn't have to be nearly as accurate as on a frequency
> shifter so the parts count wouldn't be high.
> 
> Of course if you're building a Vocoder, you'll be splitting the signal
> into bands anyway.
> 
> Here's a parts saving variation on the proposed idea: Depending on the
> band-pass filters you'll be using, you can get a 90-degree phase
> difference by tapping off at different points in the filter.  
> 
>   -- Don