freq. mult.

[Martin Czech]

:::The problem with this method is that the integrator isn't perfect, so at all
:::freq.  I don't get equal amplitude and the wave isn't symmetric with the
:::x-axis, also, the integrator isn't making perfect triangle shapes at all
:::frequencies.

The amplitude could be corrected if the square frequency is 
basically known (control voltage). I needn't be perfect,
a 50% change would be still ok since you basically need
the through zero points, not the amplitude.
And that's already one octave. 4 octaves would mean 
triangle peak dropping to 1/16, and this is still not so bad.

If the square is wobbling very fast all over the place
there's not much you can do. Of course, in this case
it would be difficult to speak of "frequency" at all.

At audio frequency the integrator should look perfect on the scope,
therefore I can not understand what problem you have with shape.
Your "x-axis" problem is DC offset, right?
This is always a problem with open loop integrator circuits,
any error -and be infinitely small"- will finally sum up.
Such circuits are therefore not stable.
Perhaps you should try to close the loop via low pass filtering
the integrator output (very low Fc) and feed that back
into the integrator. If Fc is low enough no shape or phase
margin problem.


:::
:::
:::What about doubling frequency with a PLL?  Good, bad, ?
:::I've also used a method that consists of a negative and positive pulse edge
:::detection circuit.  That doesn't work very well either because after a few
:::generations the pulses all get crowded together where the original pulse
:::edges were.

Pll's are perfect for fixed frequency applications, clock recovery and
the like. They are also ok for slow fm demodulation.  Slow means Fcarrier
>>> Fsignal.

For any other stuff the inability of a servo loop to follow immediately and
precisely will become evident. The PLL is particularly bad in this respect,
because it always contains one integrator: voltage -> frequency converter
(vco). So the control circuit can not use anything with much phase lag,
or the phase margin will be 0. It turns out that the loop can not have
large gain and/or much low pass/integrator characteristic.  But these
are the ingredients for most good servo loops.  And in some cases the
phases detector is no linear device thus giving ripples on the vco control
voltage, preventing you from using much phase lead (differentiator).

One off my professors always said: closed servo loops are only the second best
solution. If you know your system a control solution will be better, because
the servo has to wait for an error signal to react where the control knows
that some parameter changes and will react accordingly before any substantial
error will be visible. So your triangle method is basically the right way
for some application.

Sorry for being academic.


m.c.