[sdiy] Speaking of OTAs....

[jh.]

> >I have tried 100K/220 No Id and 100K/510 with 15K Id and could hardly
tell
> >the difference.
>
> I'd agree also. The diode will only help with linearity before clipping.
It
> also lowers the input impedance. Try making those input
> resistors a lot smaller... maybe 10K and 51 ohms if you can stand it.
> That will help.

Harry is on the right track, but 10k / 51R is not the solution either.

100k/510 *and* Id makes no sense.
When Id is applied, the diodes take over the part of the shunt resistor
(only that they are nonlinar to pre-distort the input voltage), so their
resistance must be small compared to the outer resistors (to both of
them!) If you look at the data sheet, something like 10k / 500
is suggested with Id = 1mA. The diodes, in conjunction with the 10k
resistor, will divide the input voltage down to the required millivolts.
The 500R's to GND are just for biasing the diodes. If you make them
too small, they will "bypass" the diodes -> less predistortion -> bad!
The data sheet shows the bias resistors as "current sources" in a simplified
diagram, and this gives an idea of what their purpose is.
But even this is misleading, and it did even mislead a genious as Serge
who wrote the performance could be improoved with a closer
approximation of real current sources. (And I swallowed it, until Jörgen
Bergfors put me right - it's all in the archives.)

The LM13700 is a good building block for not so critical OTA applications
(as in musical synthesizers (;->) ). Its main problem is the offset voltage,
not the noise, IMO. Anyway, for slightly more demanding VCA applications
(and when you don't want a specialized DBX or SSM chip), a matched pair
of low noise transistors is always a good choice. Why would one run
an audio signal thru current mirrors that are not necessary in most
applications ?

JH.