[sdiy] what is the amplitude envelope of a signal

[Czech Martin]

Yes, that is what I also tryed.
The peaks of a rectified signal seems easy,
but if you define them as local maxima 
things get messy when noise comes into play.
Connecting such peaks will not look like the envelope, but
more like the rectified signal.
So, only the larger peaks should be used, but what does
that mean in a formal way?

The filtering approach must guarantee to have not too much
lag, and especially an unsymmetrical lag does not make sense,
since the incomming signal could start slow and end very suddenly.

But I will dig out this EN paper.
Thanx!

As I said: what seemed to be so easy turns out to be very difficult!



m.c.

> -----Original Message-----
> From: owner-synth-diy at dropmix.xs4all.nl
> [mailto:owner-synth-diy at dropmix.xs4all.nl]On Behalf Of Scott Bernardi
> Sent: Montag, 29. September 2003 15:03
> To: Synth-DIY list
> Subject: Re: [sdiy] what is the amplitude envelope of a signal
> 
> 
> How about an interpolation between the peaks of a rectified signal?  
> Interpolation means that if you take the peak of the upper half of a 
> bipolar waveform (that's half wave rectification), and 
> connect them with 
> lines, the resulting shape would be your envelope. In the analog 
> implementation the interpolation is usually accomplished  by 
> filtering, 
> and the rectification is usually made full wave. 
> The trick in making a decent envelope follower is choosing the time 
> constant for the filter. It needs to be fast enough to be be able to 
> follow transients on the envelope, but too fast and it will 
> follow the 
> peaks and valleys of the waveform cycles themselves 
> (appearing as ripple 
> in the envelope).  This is a tradeoff.
> By the way, check out an all analog implementation of an envelope 
> follower that uses three cascaded FWR's with a "dc restorer" 
> circuit to 
> get 3x improved ripple rejection while retaining transient 
> response at 
> http://home.comcast.net/~sbernardi/elec/og2/og2_envfollower.ht
ml. This 
is an Electronotes design.
In the digital domain, I guess you would want to take a moving average 
of the samples on the full wave rectified waveform (which you get by 
taking the absolute value of the sample voltages) where the averages are 
taken at a rate of say one half the lowest signal frequency you want to 
process.
I'm sure this has been thought out by someone before that knows more of 
what they are talking about than I do.

Czech Martin wrote:

>Following the (long, excuse me) e-mail about my FS experiments
>you can see that I have some little signal processing tools.
>But one: envelope follower.
>
>I thought that programming this was an easy thing, everbody
>knows what the amplitude envelope of a signal is, and the
>methods for analog circuits are well known (rectifier, filter,
>bleeder).
>
>Unfortunately it turns out that extracting the amplitude envelope
>of an audio signal is not so easy.
>I used speech as sample signal, and all my experiments showed
>that considerable feedthrough of the original speech into
>the envelope signal was present.
>
>Which led me to think more about these envelpes.
>If I think about them, I feel I know them.
>If I try to explain I do not know them (stolen from Augustinus).
>
>What the heck is the amplitude envelope of an audio signal?
>(please give a formal description, i.e. something so precise
>that programming would be easy).
>
>m.c.
>
>
>  
>

-- 
Scott Bernardi
sbernardi at comcast.net