GIC String Filter - first experiments (was: [sdiy] FilterBanks& CSound)

[René Schmitz]

Hi Harry and List,

> no. I have a whole TUBE of LM3900 and only a few BBD.  I just began to understand the
> true nature of the 3900 as current controlled voltage source... a real useful block.
> Imagine
> my surprise to hear Bob Pease slam the chip as a piece of sh!t (my original opinion
> btw).

The principle is great. Only the implementation is mediocre (to say the 
least). For example why didn't they use a Wilson mirror at the input. 
And the limited slewrate and openloop bandwidth...
But there is a way to get that behaviour using more modern parts. Say a 
cheap integrated current mirror, an extra transistor to make it a Wilson 
mirror, some Emitter degeneration, and an Opamp of your choice for the 
actual amplifier...

The real kick is that current can be passed from one input to the other. 
(If we were merely after a current differencing thing, we could build a 
diffamp and add two transimpedance amplifers at the input. But most of 
the circuits for the 3900 wouldn't work then.)

I guess it never really caught on, because most people didn't understand 
it. So the semiconductor industry probably never bothered to develop 
this much further.

But hey, I don't buy the "this is a piece of crap". How much "precision" 
does a normal NPN transistor have? Lousy specified gain, high offsets, 
production spread, bad thermal behaviour. Qualifies as "low quality". 
But you can find these in almost any gear. Think about that! Simply 
because something isn't highly spec'ed doesn't mean its useless.

> I guess the principle and practice of the chip do not match well... and it is indeed a
> crappy
> "opamp" if you call it that.  The ARP filter is really brilliant, imho.

I don't know if everybody is aware that the principal structure of that 
ARP transconductance cell, is the same that is used in the SSM2040. (A 
long tail pair working on a current mirror, only that it stands on the 
head compared to the 2040.) The only difference is that the current does 
run into a virtual ground node, instead of a cap to GND.

Cheers,
  René