[sdiy] Auto Center waveforms

[Jerry Gray-Eskue]

Thinking it through a little further the differentiator stage is very close
to a DC blocking cap with a voltage buffer, if the feed back R is very low
(~0) it is effectively an AC coupled voltage follower. As R increases so
does the time constant you referred to.

Not quite what I was looking for. I suspect my best bet would be to attack
the original source of the DC offset, with that minimized the integration
and negative feed back would not require such significant level changes.

As for the uses, I guess they will just have to be handled on an individual
basis.






-----Original Message-----
From: Eric Brombaugh [mailto:ebrombaugh1 at cox.net]
Sent: Tuesday, July 14, 2009 12:35 PM
To: Jerry Gray-Eskue
Cc: Tom Wiltshire; Synth-diy at dropmix.xs4all.nl
Subject: Re: [sdiy] Auto Center waveforms


Jerry,

Thanks for the clarification. I wasn't being entirely facetious when I
asked why a DC blocking cap wouldn't work, and your elaboration helps to
understand the situation better. Basically it boils down to the fact
that you've got a fair amount of low frequency content that you don't
want to lose, while also having the conflicting requirement of fast
settling time.

As you notice, there are some tradeoffs here: You'd like to eliminate
DC, but you don't want to distort the signal. A big problem is where to
define the boundaries of what you consider to be DC. Your last sentence
below reveals a lot "...these do tend to work the best when the DC
component is fairly constant." DC is by _definition_ constant over a
certain interval. If your DC is fluctuating then you haven't defined the
interval correctly. That means you need to increase the time constant in
your integrators, which has the undesirable effect of making the system
take longer to converge.

This issue is present in all the solutions we've examined so far: The DC
blocking cap time constant is defined by the size of the cap and the
impedance of the source & load it sits between. The integrator /
subtractor circuit Bill described is also limited by the time constant
of the integrator. Your suggestion of a differentiator / integrator
chain again suffers from the time constants of both circuits (which
additionally must be well matched to prevent distortion). Even some sort
of sample/hold circuit would still require some integration time to
acquire the offset.

The only way to get around this is to go non-causal (really only
possible in DSP systems with the possibility to delay signals
perfectly), or use out-of-band or a-priori knowledge of the signal.
Since you don't mention knowing anything else about the waveform I
assume this isn't a possibility.

Eric

Jerry Gray-Eskue wrote:
> Yes it is Audio but has the potential to go into the LFO range. The DC
level
> shift is pronounced and audible when hopping a few octaves. The whole
thing
> has to do with producing traditional analog signals in an unconventional
> way.
>
> I am close to shelving the approach due to this problem.
>
> 0.1 seconds doesn't sound like much but consider that it is 10 cycles of
an
> audible 100hz tone. I actually have tried this method but the DC variation
> is to drastic to control using this method. The integrator generating the
> saw tooth magnifies any duty cycle or offset imbalance. Normally this is
> self correcting using the upper and lower thresholds, but as I said this
is
> not the conventional approach and that mechanism is missing. I have this
> part under control to avoid saturation of the integrator, but I still have
a
> significant issue with DC offsets in the output.
>
> The whole thing got me thinking that there has to be a elegant way to
remove
> DC offsets that is not AC coupled. This would be useful in general for
LFOs
> and Ramp based VCOs asymmetrical wave shapers and etc.
>
> Yes there is a lot to be said for just using a DC blocking cap, as well as
> the integrated DC offset removal, but these do tend to work the best when
> the DC component is fairly constant.
>
>
>
> -----Original Message-----
> From: Tom Wiltshire [mailto:tom at electricdruid.net]
> Sent: Tuesday, July 14, 2009 11:41 AM
> To: Jerry Gray-Eskue
> Cc: Synth-diy at dropmix.xs4all.nl
> Subject: Re: [sdiy] Auto Center waveforms
>
>
> Sorry to repeat Eric's question, but what *is* wrong with a DC
> blocking cap?
>
> I also don't really see a problem with Bill Forbes' suggestion of
> integrating over a number of wavecycles to get the DC level. If you
> integrate (say) 10 cycles, you'll have a delay of 0.1 sec for 100Hz
> before the DC level settles down. 100Hz is pretty low, so consider
> that close to the worst case - and a tenth of a second isn't very
> long. Or am I wrong in assuming these are audio waveforms? Is this an
> LFO?
>
> Why does it matter so much? What have you got following this circuit
> that is so sensitive to DC offsets?