[sdiy] linearize OTA for VCF?

[jhaible at t-online.de]

Ian Fritz schrieb:
> So the idea was that if you replace the lower resistors 
> with current 
> sources, then the input structure would be similar to the 
> CA 3280's input 
> structure? Certainly seems like it should work. Was the 
> schematic for this 
> experiment posted anywhere?

When the chip was new, nobody less than Serge T. 
suggested an "improovement" for the 13600 in Electronotes,
replacing the 500R resistors to GND with larger resistors 
to the negative rail, to make them more like current
sources. (The other resistors - input resistor and from
diode anodes to positive rail - aren't exactly current 
sources either, but the 500R's from inputs to GND 
are the worst aberration.)
I turned his verbal description into a circuit
(http://home.t-online.de/home/jhaible/hjmp2.gif).
I never measured the linearity, but Jörgen Bergfors did,
and pointed out that there was a sort of over-compensation
of the diff pair's nonlinearity. I immediately checked it
in Spice, and saw that Jörgen was right. I also replaced
the (now larger) resistors with ideal current sources
(still Spice), and it became worse.
The circuit which Jim just posted will work,
because it reproduces the classic method. It's the
combination of resistors and current sources which does
not work. 
Bottom line: Either use the full set of National's resistors,
or use the classic method (like Jim).
Curious mind wants to know the equations behind National's
method. The explanation in their data sheet is useless
("no big error when current sources replaced with resistors")
and misleading.
*If* the resistor method works as good as the current source
method, National's method is very clever, because it needs
less transistors and maybe is a workaround for a patent. (?)
(Who has/had the patent for the Gilbert cell, btw ?)
I think it's generally agreed that the 3280 is better than the
13700. But I wonder if this is because of the resistor method
(as I used to believe), or simply because of the better specs
of the transistors inside the 3280 (which seems very reasonable
now.)

JH.