PLL guess

[Martin Czech]

It was supposed to use a  limiter, lowpassfilter (LP, 6dB ?) with rectifier (R) 
and smoothing lowpass filter (SLP)
as initial guess for vco frequency for a pll.

The higher the frequency, the lower the output, the voltage should go down
50% within an octave.
If we had ideal components, this would surely work, the voltage at the output
after some time will give information about incomming frequency.
Indeed, I think I remember the Roland 700 uses the above signal chain 
even without pll.)

But I'm not sure, what happens in the case of step response,
which is the application we want to improve.

I think, that this kind of freq. estimation will introduce the same slowness
or ripple problem through the backdoor again which we were to avoid.
If we want a fast step response, we have to make the SLP wide band open.
But this will lead to more harmonics of R coming through : ripples.
To avoid them, we have to make SLP narrow and steep, and this will cause
slowness of step response. 
And the problem gets worse, if we want to detect very low frequency.

I heard, that Roland 700 has problems with low frequency, too.
But who can sing or whistle with f<<40 Hz ?

Further the signal shape of incomming signal is not well defined in most cases,
even if we use a limiter to make it rectangular,
so this would lead to more errors in estimation.

I think that modern guitar synths like GR50 use very clever algorithms
which are much too complicated to implement them analog, so it's off topic
here ;->.
In case of a guitar, one can make some usefull assumptions about the signal.
That is the maximum input frequency range per string, the waveshape,
the development of the waveshape.
But, playing strange guitar stuff will fool those machines too.

I talked to a swedish manufacturer at the Frankfurt music fair,
they claimed top have the fastes algorithm with a neural network to estimate frequency
out of the very first harmonic relations when the guitar string is plucked,
so they need less than a cycle. But the system has learned to cope with
the E or A string only, for example. If we use a different input signal, or just
swap strings the system will fail, because the assumptions on the input signal
are no more valid.

m.c.