linear FM : was RE: [sdiy] Temperature Compensated VCO attempt - help?

[Czech Martin]

I may miss the point here but I think the real
problem is in the storage element that decides charge up/down
direction. 

If this is the case, read on.
If not, complain.

Obviously there are situations where
the comparators want up and the sign "toggler" wants
down or vice versa. This can also lead to meta stability
problems since the timing of up/down and toggle
is unknown.

One solution could be as follows:

Imagine a nand type RS FF. It should have two S and two R
inputs (i.e. 3 input nands are needed).
One R and one S input is taken for the usual comparators.
(a nice side effect is that the comparators don't need
any hysteresis). This way the comparators will always
finally win, i.e. no wave form run away out of limits.

The toggle signal is derived from the sign comparator.
The difficulties were described. The toggle signal 
should be a positive pulse, say 300ns long.
This pulse is fed to the remaining R/S inputs via
and gates, that make sure that only the low state
nand of the RS FF will get a pulse.
This feedback from the RS FF state must have a delay
that must be much longer than the toggle pulse
in order to avoid RS FF oscillations. I think
this delay can be made very long, this depends
on the highest possible toggle frequency.
Several us.

All the noise issues will not be cured, also the ambiguity
near the comparator switching points.
Bit as someone said, this is only a source of noise, just
like a noisy level comparator.

However,
I think that the phase will finally run in the "right" direction.
And the triangle wave will not run away.

m.c.



-----Original Message-----
From: Magnus Danielson [mailto:cfmd at swipnet.se]
Sent: Dienstag, 4. Februar 2003 08:55
To: don at till.com
Cc: sbernardi at attbi.com; synth-diy at dropmix.xs4all.nl
Subject: Re: [sdiy] Temperature Compensated VCO attempt - help?


From: Don Tillman <don at till.com>
Subject: Re: [sdiy] Temperature Compensated VCO attempt - help?
Date: Mon, 3 Feb 2003 10:35:07 -0800

Dear Don,

>    > Date: Mon, 03 Feb 2003 10:02:43 +0100 (CET)
>    > From: Magnus Danielson <cfmd at swipnet.se>
>    > 
>    > My bid is:
>    > 
>    > Put a Gilbert-cell on top of an expo-converter. Put the linear
>    > (sign-swapping) modulation on one of the Gilbert-cell input pair,
>    > put the Schmitt-triggered output waveform on the other input pair
>    > of the Gilbert-cell. The current output of the Gilbert-cell is
>    > connected to an integrator and this is followed by a
>    > Schmitt-trigger. The current in the Gilbert-cell is modulated
>    > from the expo-pair. This should be both a compact and pretty
>    > well-behaving oscillator and since there is non abrupt switching
>    > of current but rather linear change in current balance I think
>    > there is a good chance of a nice behaviour.
> 
> Well that's certainly elegant.
> 
> But I think something's missing (or perhaps I'm missing something).
> 
> Say we have it set to a positive frequency, the integrator has just
> hit the positive reference voltage, the Schmitt trigger changes the
> direction to slew down, and then at that point, if we set the control
> voltage to negative, the integrator slews back up to the reference
> voltage, but the Schmitt trigger is still in the positive state so it
> won't notice to change direction again.
> 
> Graceful thru-zero a very difficult problem.

OK, don't say a good nights sleep doesn't help you see things clearer.

Now, what is the problem? Realy?

Well, we have a waveform which charges up and down. We have must swap sign on
the charging to keep the waveform within limits. Just to annoy the designer
someone desided we also is allowed to have both positive and negative
frequencies which is under linear control. When the linear control shifts over
from positive to negative frequency or from negative to positive, the waveform
will change direction. Now, when this happends we will reach the same
switch-point we previously we switched sign to keep within bounds. Naturally,
if we switch sign on the linear frequency control the waveform will switch
direction again, but "back" to the "right" direction. What we learn is that
unless we properly followed the polarity we don't actually know if we hit the
upper or lower limit.

OK, now we know what can happend. Now we ask ourself about actual
implementational issues.

Can we correctly identify the shift in modulation sign?
No, that's very hard. The reason is that if we look at the sign for any signal
we effectively use a comparator, and the offset between the comparator and
actual sign shift hides the sign. Especially if the signal sit tight in the
gap for a sufficiently long time it will make a difference. Also, DC offsets
tend to drift some, so when the static offset have been trimmed away there's
still some around. Eventually it doesn't mean much, since the offset becomes
so small that other effects make it meaningless.

The other way is to handle the situation when we actually hit the limits.
If we have a comparator for the upper and for the lower limit so that these
go "high" when the signal goes above the upper limit or below the lower limit.
We also have an output to the sign shifter (the Gilbert-cell I have been
babbeling about). When we hit either of the limits we must change direction.
That's actually how simple it is. We can acheive this by having a state-machine
such that it's output is the current charge-direction. Whenever either the
upper or lower limit is reached and thus either comparator goes high, the state
machine reacts by changing its sign.

There are different ways to actually design this state machine, but a
D-flip-flop and an or-gate might be all it takes.

So Don, does this seem to bark up the right tree now?
I think so. There is a few details about the detailed state-machine and the
ability to filter noise and blips, but what it should do is clear and I do
think those issues can be dealt with.

Oh, it's snowing again! ;O)
(We had a mild snowstorm yeasterday ;O)

Cheers,
Magnus - likes snow