[sdiy] Looking for OB-8/CEM3320 filter gurus

[ChristianH]

Hi again,

while synthdiy was down on thursday, I did some research, et voila - I
found the full 3320 data sheet. I bought it 20 years ago, from a certain
Dieter D... :-/
Furthermore, I found an Elektor article from the same era, which is more
or less a literal translation of the data sheet. And today you are more
or less confirming what I assumed from the various sources and
measurements. So I seem to be heading in the right direction now...

What I didn't understand at all was the DC level of the buffer outputs.
Your explanation and the data sheet makes this much clearer now.
Apparently, this is where the problem is. The sick filter has different
DC values, compared to a good filter. But now I finally know how to
interpret this.

When I redrew the OB-8 filter in a more readable fashion, I did see that
the first stage has its cap removed in 2p mode, and I did recognize that
there are several feedback loops - but I can't believe that it took me so
long to realize that it's a good old state variable filter :-)
It happended only when I read the data sheet with the SVF example. 'Slap'
said the forehead to the hand... Without really thinking, I always
assumed 2 pole operation to simply remove 2 cells - although I wondered
why that much CMOS switching would be necessary for that.
And it's not _that_ surprising to find a SVF, considering Oberheim's filter
traditions before the OB-Xa/8 models.

And: of course the negative bias resistors _are_ present in the data
sheet. I just didn't notice them at first. And now I even know why there
is none for the first cell.

Thanks for replying, JH - somehow I knew it would be you. I guess, the
magic OB-8 keyword simply was too tempting to stay silent ;->

regards
Christian


On Fri, 26 Jul 2002 10:56:30 +0000 (UTC) jhaible wrote:

> 
> >I finally started working on that dead OB-8 filter.
> >Although the schematic looks a bit weird at first sight, it's starting
> >to become clearer, once all the 2 pole circuitry is omitted, leaving the
> >basic 4 pole part.
> 
> In 2pole mode, one gm cell is used as an opamp (reducing the
> filter cap, but not to zero, because you need a phase compensation).
> It's remarkable that this is stable, because this "opamp" gets the whole
> range of bias currents for its input stage (just like the stages that are
> used as integrators).
> 
> >However, I don't quite understand the way the gain cells are working.
> >Is there a more complete data sheet than the 2 pages found at synthech.com?
> 
> Look for the Curtis patents. All the details are there. It's basically the
> same OTA + cap + buffer topology as the SSM2040, but a very different
> circuit design. Linearized, a current input that needs to be biased for
> a slightly positive voltage.
> 
> 
> >To me each gain cell/buffer combo looks like an inverting op amp (plus
> >filtering of course), with a summing node, Rc for input, and Rf for
> >feedback.
> 
> Well yes, a simple opamp is nothing else than an OTA, a compensation cap,
> and a unity gain buffer. But here you have a rather high cap, and a variable 
> transconductance in the input stage. (And splitting hairs, it's not a
> transconductance, but a "current gain" in Curtis chips.)
> If you want to understand it better, get the data sheet for the "missing link"
> between OTA and opamp, LM346. (www.national.com)
> 
> 
> >But in the OB-8 circuit there are additional resistors (RX44, 59, 58)
> >that appear to add a negative DC offset current into the summing nodes.
> >These resistors are not present in the data sheet sample applications, so
> >they don't seem to be required by the 3320.
> >What's the use of this? Is it to improve the chip's operation in some way,
> >or is it to intentionally cause asymmetric clipping in each stage, in
> >order to spice up the filter's sound?
> 
> This is to bias the input stage properly. What you want is an output voltage
> between positive supply and GND (no negative swing, here!), and for that
> you need a bias current at the input. Part of it comes from the feedback
> resistor of the filter stage's own output: If output is at 7.5V and input is
> at 0.6V (typical conditions), there is a bias current flowing.
> Sometimes you have an additional bias current from a previous
> filter stage (DC coupled, antother resistor from approx. 7.5V to 0.6V).
> Then you must compensate this with a resistor to th enegative supply
> voltage, in order to keep that filter stage in the intended output voltage
> range. For various filter conigurations, and for various negative supply
> voltages,
> these resistor values must be different.
> 
> Speaking of negative supply: Curtis filters are typically working between
> positive supply and GND for audio. (Complicated to calculate, but
> nice to interface with 4000 series CMOS switches!) A negative supply is
> only needed for internal CV biasing. It's limited to a small value by an
> on-chip
> zener diode (I think the process they used was only 24V, so no +/-15V supply
> without extra means of voltage limiting!). Application note circuits mostly
> use this internal zener, but the OB-8 uses an external -5V supply instead,
> effectively bypassing the on-chip zener and giving more precision.
> 
> There was a *lot* of design work behind the OB-8 (and a lot of experience
> from the OB-Xa), and the switchable 2pole / 4pole filter is simply amazing.
> This is neither a 4pole filter that is tapped after the 2ns stage, like so
> many
> others do, nor two SV filters as in the Jupiter 6 (also easy to design), but
> a complete re-configuration from 4-pole polygon-filter to 2pole state variable
> filter. Real pole stages become integrators, one stage tunrns into an opamp,
> and all that with the DC bias current issues everywhere. 
> 
> Enough of OB-8 praise for now. (;->)
> 
> When you have fixed yourOB-8, how about doing some modifications?
> http://home.debitel.net/user/jhaible/hjob8mod.html
> and
> http://home.debitel.net/user/jhaible/hj_pmob8.html
> 
> JH.