[sdiy] Why do we need the buffers? - gm-C filter question

[René Schmitz]

Hi Aaron and all,

Aaron Lanterman wrote:
> It might be fun to try building a filter with a four-CMOS-transistor OTA 
> instead of the usual four-BJT-transistor OTA. I've never looked into 
> whether MOSFETS come in matched pairs, or for that matter how tightly 
> matched things on say the 4007 are. The CMOS would give you a different 
> type of distortion instead of a tanh type of BJT distortion, which could 
> be interesting.

Thats true. But transconductance cells based on FETs have their 
peculiarities.

First of all, the FET law makes only for a very mild variation on the 
transconductance of the drain current. IIRC gm varies only with sqrt of 
the current or something. So to get sweep ranges of 1:10000 or higher 
we'd need to use either tiny currents, or end up at the 10Amps range at 
the high end pretty soon. (Ok, no problem with Power MOSFETs, but not 
good for the environment. :-))

Second the linear operating range of a CMOS gm cell also varies with the 
standing current. So as you diminish your current, so does your linear 
range, but since your V+ - V- is still the same, you end up with 
distortion that _increases_ as you go _lower_ in Icontrol. Pretty funny 
actually.

> I've heard there's this guy named Rene that's looked into doing weird 
> things with CMOS... ;)

Yep, thats me. :-)

> Or, you could mix and match maybe? Use CMOS for the diff pair and BJT 
> for the current mirror, that would give you high-Z on your OTA input.

IMO a good combination is to use CMOS for the buffers, just like in the 
famous WASP filter. Hi Z input, and ok-ish drive capabilities.

> You could try either (4-CMOS or 2-BJT/2-MOS) as a current switcher in a 
> triangle core OTA...

Indeed you can do this, but then its probably better to use CMOS switches.

> Things with high input impedance are just so much easier to analyze...

Thats right.

Cheers,
  René

-- 
uzs159 at uni-bonn.de
http://www.uni-bonn.de/~uzs159