[sdiy] integrator / capacitor leakage

[JH.]

Hi Michael,

that goes into the same direction I was thinking.
For high frequencies, I there is no doubt.
Maybe it's good to approach that wide-range app from
the "high side", rather than from the "low side" as I did.

JH.


----- Original Message ----- 
From: "Michael Baxter" <mab at cruzio.com>
To: "Magnus Danielson" <cfmd at bredband.net>
Cc: <jhaible at debitel.net>; <synth-diy at dropmix.xs4all.nl>
Sent: Saturday, December 03, 2005 11:09 PM
Subject: Re: [sdiy] integrator / capacitor leakage


This has been an interesting discussion. I'll add my voice, and offer some
of my research in at least a qualitative form. My issue was at the other
end, high-frequency accuracy, and relatedly linearity of the VCO transfer
function.

I was looking at the bandwidth limitations of the opamp as a perfect
integrator. Due to limited bandwidth, the opamp ceases to integrate the
input current as accurately at higher frequencies, which is ultimately a
source of pitch error. It appeared that opamps with MHz of bandwidth would
make this error smaller for audio frequencies, but there was an even
easier way: use less amplitude.

So, it's easier to build an accurate +/- 0.5 Vpp triangle integrator than
to build an accurate +/- 5.0 Vpp integrator with an opamp having the same
bandwidth. So, I discovered that more frequency accuracy at the higher end
of the band could be attained using a smaller amplitude integrator,
followed by a (say) 10X amplifier. Interestingly, the 10X amplifier
provides signal swing for whatever functionality needed (say sine shaper,
or a Serge-like waveshaper circuit...) but at no loss of frequency
accuracy. And the smaller amplitude integrator could work very much more
precisely, because there was less swing to do for the available gain at
bandwidth ... less "work" if you will.

Militating against these advantages are amplifier voltage offset, which
for whatever it is, would be higher against a 1V signal than it would be
against a 10V signal. However, my experience is that all triwave
integrators need some small DC trim for low-frequencies anyway, so that
didn't seem like a serious limitation.

It is a legitimate point that comparator accuracy has an impact on
frequency here, but rather nicely it's DC reference voltage accuracy, not
AC dynamic accuracy (factoring out the finite response time of any likely
comparator that is sufficiently fast for audio signals with small timing
error) as in the case for the integrator.

Wrt the discussion in leakage currents, use of a smaller voltage for the
integration switch-points has no affect really on the range of allowable
input currents ... if your opamp will accept (for example) +/- 1mA of
input current, it would do that for 1V swing as easily as 10V swing. So
the "dynamic range" of currents allowed against the size of leakage
currents is no smaller for a smaller voltage swing.

Capacitor scaling is of course affected. But the smaller swing approach
offers the ability to use smaller capacitors for the required integration
current range. Smaller capacitors with lower leakage are easier to find
than the case for larger-valued capacitors, it would seem.

So, it seemed to me that a VCO with excellent frequency accuracy and
linearity was enhanced by using TWO opamps as the "integrator": one for
integration to smaller voltage limits, and one to amplify the integrated
voltage to larger amplitude.

Best,
Michael

On Sat, 3 Dec 2005, Magnus Danielson wrote:

> From: jhaible at debitel.net
> Subject: [sdiy] integrator / capacitor leakage
> Date: 2 Dec 2005 13:56:22 +0100,Fri,  2 Dec 2005 13:56:22 +0100
> Message-ID: <1133528182.439044762ead5 at www.debitel.net>
>
> Hi Jürgen,
>
> There are many sides to this...
>
> > When I'm building an integrator for a triangle or sine wave oscillator
with
> > an amplitude of 20Vpp, I have two options:
> >
> > a) run the integrator at 20Vpp, or
> > b) run the integrator at a lower voltage, and amplify the signal
> >    with an extra amp.
> >
> > Now, for a certain current into the integrator, the integration
capacitor
> > in case (a) will be much smaller. (Larger voltage to pass in the same
time
> > at th esame current.)
> >
> > What is better, in terms of precision / leakage?
>
> Let me point out that these may not be the same. You are trying to reach
> precision, and leakage is one of several things to battle.
>
> > At first glance, I'd say leakage is mostly leakage _currents_, so
> > it will be the same in both cases.
> > At second glance, I'd say if the leakage currents are not entirely
> > independent with voltage, case (a) will be worse.
> >
> > Is this right? Are there other things to consider?
>
> If we talk leakage, there are two sources, internal and external.
>
> Internal is naturally the leakage trough the dielectrum. Assuming a linear
> low-epsilon material (i.e. good plastic), this leakage is fairly linear
with
> the charge and also recall that we also tend to use these caps in a
fraction of
> their supported voltage range, makes the linear assumption fairly safe.
> Hint: Stay of (most) ceramics.
>
> External sources of leakage includes creep-voltages from power-supplies,
signal
> etc. These are due to the leaking in the PCB and eventually also on the
dirty
> surface. Some of this can be solved by providing a guard-ring on both
sides of
> the mounting-hole, keeping the same voltage as the pin to be guarded. For
an
> integrator this would typically be the ground, which is what the positive
input
> of the op-amp is hooked to. It also helps to make the surfaces of the PCB
and
> cap clean, please apply isopropanol here, and mounting it inside a tight
box
> which keeps dirt from comming (you don't smoke near your system, do you?)
in
> makes it safer yeat.
>
> If you think about it, a higher voltage of the output makes the
drive-current
> for the same frequency lager than for a lower voltage, and then will the
same
> external creep-current from other sources less influential. However, you
will
> suffer if that means that you run closer to saturation of the
expo-transistors,
> and then that will be your main problem, but in a different range of the
> frequency range, since you are already running a higher current, so
leakage is
> more of a low-freq-problem due to current-ratio is smaller.
>
> You also have the leakage from the op-amp.
>
> If you really think you are hurting from leakage, you can compensate for
it by
> bringing a small contra-current to the summing point which compensate for
both
> the voltage (use the output of the integrator) and for the DC current.
> It is fairly simple to do this, but it should be done only after you have
> chosen capacitor, done the guard-ring and concluded that this does not
suffice.
>
> I also assume that you have taken all forms of care to handle things like
> reset-time, non-linearities in expo-transistors (bulk resistance
compensation,
> near saturation compensation, etc. etc.). If you have a "perfect"
temperature
> compensation expo you still need to account the temperature dependence of
other
> parts, especially when they do not drift relative each other so that the
> relation between values change.
>
> I think one does not have to look at the more esoteric non-linearities of
> capacitors if you just avoid the worst dielectrics in this sense, which
you can
> do fairly easilly and at no major cost. Actually, active compensation for
> leakage is probably not needed, but it depends on just how much precission
you
> need.
>
> > Background: VCO which not only will run from 0.03 Hz to 20kHz,
> > but which will also produce very precise waveforms over that
> > whole range.
>
> That is 19,23 octaves... which is possible to acheive, but care needs to
be
> taken. I should clean up my ASM-1 board and re-measure it again. Should
try to
> figure out a good way to measure temperature so I get a good temperature
> dependence measure. I've got some LM-135s lying around somewhere.
>
> > Any ideas / hints welcome.
>
> I hope I have contributed some with my initial comments. I have only
briefly
> seen the rest of the thread, so I wanted to go back to the top and take it
from
> there.
>
> Cheers,
> Magnus
>
>
>