[sdiy] Interesting gain cells in the CEM3320 (incidentally used in Elka Synthex VCF)

[Aaron Lanterman]

So while we're on the topic of VCFs, I was taking a gander at the Elka Synthex VCF, with an eye towards making a modular version of it (using available parts instead of a CEM). I had looked at this a couple of years ago and recently felt compelled to think about it again. It uses a CEM3320.

Osama Hoshuyama has some analysis of the Synthex filter:

http://houshu.at.webry.info/200706/article_3.html

Here it is through Google Translate:

http://translate.googleusercontent.com/translate_c?hl=en&ie=UTF-8&sl=ja&tl=en&u=http://houshu.at.webry.info/200706/article_3.html&prev=_t&rurl=translate.google.com&twu=1&usg=ALkJrhjf4lpkKFhFr4Ta6vb1dL3Sfu__zg

Tom Wiltshire has a nice unravelling of it here, along with numerous other applications of the CEM3320:

http://www.electricdruid.net/images/cem3320/larger/CEM3320HighpassFilter.gif

-----------------------

I've always assumed that the CEM3320 was a bunch of OTAs with buffers, with the OTAs driven by a unified exponential V-to-I inverter. But when I looked at the circuits on Tom Wiltshire's site, I couldn't make this interpretation make sense; there wasn't any "divide down" resistor set like you usually see with OTAs. In some cases, the gain cell inputs are hooked to one of the power rails through a large resistor, and I couldn't make sense of that either... so I looked at the CEM3320 datasheet:

http://www.synthtech.com/cem/c3320pdf.pdf

and found out that it's really quite a different beast.

There's a "gm" transconductance cell intended for feedback control that feeds back to the 1st input. But the main gain cells for the filter aren't OTAs. Indeed, the gain numbers listed on the datasheet are unitless, which means it must be voltage in-voltage out or current-in current-out. I figure it must be current-input/current-output, since the current output of the "gm" feedback cell feeds right into the first gain cell (although I'm always skeptical of bits of datasheet claiming to show what's actually inside a chip). The output of the gain cells claim high output resistance - which makes sense in a Norton equivalent for a current output - also the output is dumped down a cap like in a SSM2040 OTA style filter. 

My first thought is that this has some interesting implications for trying to do detailed Antti-style DSP models of CEM3320 type filters - the usual OTA tanh function isn't going to work here, probably. 

The datasheet is just two pages and kind of vague. I wonder things like, are the gain cell inputs sitting at virtual ground of some sort? So by putting a voltage through a resistor I can generate a specific current into the cell? That makes sense looking at the various examples of the CEM3320 in use. 

The datasheet refers to a patent, namely 4,004,141, which seems to describe a unique kind of exponential-control VCA. The current inputs described in the patent appear to go into op amps with BJTs in the feedback loop, and those op amps create a virtual ground, supporting my conjecture above about where the current inputs sit. This doesn't look like any other gain cell I am familiar with - it doesn't look like the Blackmer cell used in the THAT chips, or the Figure 23 of the SSM2164 datasheet, or a Gilbert multiplier. 

I'm left wondering what a modern part equivalent to this Curtis gain cell might be. Closest in overall function might be an SSM2164 or a THAT VCA, but the Curtis cell really is different; the SSM2164 outputs need to be held at a virtual ground by a following op amp, whereas clearly the voltage at the output of the Curtis gain cells is going to be going all over the place. I don't know if the THAT chips have a similar requirement, although every example I recall seeing follow the THAT VCA with an op amp holding the VCA output at a virtual ground.

I'm assuming the cells must be inverting, like on a SSM2040, in order for the feedback resistors to work right.

I'm also left wondering about the various circuits in which the inputs are tied to the power rails through resistors - why is that necessary? 

Anyway, food for discussion...

- Aaron

P.S. Interestingly, it looks like the CEM3328 and CEM3350 filter chips have more traditional transconductance "gm" stages.