envelope follower

[Rene Schmitz]

At 15:12 25.09.1997 +0200, you wrote:
>Design idea for narrow bandwith envelope follower:
>
>Usually, envelope followers detect the amplitude of a signal
>via rectifying and lowpass filtering.
>
>Problem:
>
>To convert the rectified output to "dc" some lowpass filter is
>required. For proper filtering at low input frequencys high steepness/
>high order and low cutoff frequency are necessary.
>
>On the other hand, for good transient tracking the cutoff should be as
>high as possible, with no overshoot. This means also low steepness.
>
>It is not possible to get both with a simple recitifier/lp
>arrangement.

Yep, but via a peak detector and a s&h that is triggered in the zero crossing,
you can get a very good approximation of the envelope, this approach does
not have
the phase problems like diode detector circuits.
The output is a stair-case envelope with every stair the peak amplitude of
the signal.
A low order filter might be used to "round" the stairs.
Since the circuit uses a zero cross detection it is independant of the
signal frequency.
But I must say that this might be too complex to do in a multi channel vocoder.


>I think a better way to handle this is to seperate the audio signal
>into severall bands and to process each band seperately like it is done
>in compander systems.
>
>Any experience about that out there?  Do you know any follower that
>utilises this kind of idea ?
>For very narrow bandwith systems, like vocoder channels, the signal can
>be assumed as sinusoid, fixed frequency. I remember the usuall ac/dc
>converter scheme for 3-phase-power-distribution with 6 diodes giving
>rather good dc with little ripple. Now, this can be used for the above
>situation:  I used three precision rectifiers, the first "normal", the
>second with simple highpass filter before the input, the third with
>simple lowpass filter before it's input.  The filters are set up in
>such a way, that +-60 deg phase shiftet signals get into the
>rectifiers. The rectifiers outputs are summed such that the amplitude
>loss for the filtered branches is compensated.
>
>Result :
>The circuit output gives pretty good "dc", the peak to peak ac part is
>only 13% (100% with standard rectifier).  Furthermore the main ac
>component has a 6 times higher frequency than the audio input (only 2
>times higher in standard case). This makes it much easier to find a
>proper lowpass filter suitable for all applications.
>
>
>Example :
>
>Lowpass is 4-th order Bessel (best choice for no overshoot) with fc such
>that it needs one audio signal cycle to settle.
>
>"3-phase" system yields -70dB ac suppresion
>
>"standard" rectifier -31dB ac suppresion
>
>In case of 10% frequency or phase shifter mismatch
>
>"3-phase" system yields -60dB ac suppresion
>
>"standard" rectifier -34dB ac suppresion
>
>This means 30dB improvement on the expense of tripple rectifiers
>and two resistors and two low tollerance caps. The filter fc
>could have been a bit lower in the example, so that -80dB are possible
>for a reasonable follower.

Cool, I like the idea!

This reminded me of an other method for amplitude detection i found in a book:

The method is derrived from the formula

 Uo^2 * sin^2 wt + Uo^2 * cos^2 wt = Uo^2 * (sin^2 wt + cos^2 wt) = Uo^2 
 (since sin^2 x + cos^2 x = 1)
 
To detect the amplitude of the signal u~ =Uo*sin wt
you need a squaring circuit, that produces Uo^2 sin^2 wt,
you phase shift the input signal 90° by an integrator( RC=1/w ),
so that you get Uo * cos wt
a second squarer gives Uo^2 * cos^2 wt, you sum the both voltages,
and use a root circuit at whose output you get the Amplitude as a constant
voltage.

The advantage of this circuit is that you get the amplitude (esp. at low
frequencies )
* even before * the signal has finished a complete cycle!!!

The disadvantage is the cost of this: Two ringmodulators, a root extractor,
an integrator...
And this would only be suitable for discrete frequencies, like in a vocoder.