[sdiy] OTA woes

[Harry Bissell Jr]

I'd like to add...

you can use matched diodes, or a diode connected
dual transistor, or an array like the CA3046

There is an app note that shows this for the CA3080
iirc.

H^) harry


--- Ian Fritz <ijfritz at earthlink.net> wrote:
> Hi group --
> 
> Just in the past week I tried making a VCO core
> using a 13600 as a switched 
> current source combined with a high-performance
> opamp integrator and an 
> LM311-based hysteretic switch.  My goal was to use
> the other half of the 
> 13600 as a Sin shaper.
> 
> This core had extremely poor tracking.  I tried
> different input drive 
> levels to the OTA as well as using/not using the
> linearizing 
> diodes.  Replacing the 13600 with a 3080 immediately
> fixed the problem.
> 
> This problem seemed similar to some problems I had
> many years ago using the 
> 13600 as 2Q or 4Q multipliers.  So yesterday I took
> some time to look more 
> carefully for the nonlinearity.
> 
> What I did was to look at the OTA output (through
> either an opamp I/V 
> converter or a buffered resistor to ground) as a
> function of programming 
> current (Iabc), with fixed input voltage.  I used
> Iabc values from 0 to 
> 0.5mA.  As a rough characterization of the
> nonlinearity, I compare below 
> the output for Iabc of .25 mA and .5 mA.
> 
> For the 13600 the results are as follows.  With an
> input voltage of 17 mV 
> and an opamp I/V converter at the output, the
> midpoint error is 8%.  Using 
> a buffered resistor to ground increases this error
> to 9%.  Increasing the 
> input voltage to 170 mV lowers the error slightly to
> about 7%.  In all 
> cases the midpoint output is too high compared to
> the endpoint (downward 
> curvature).
> 
> For the 3080 the error is much smaller, about 2%. 
> Also, the error is in 
> the opposite direction (upward curvature).  In
> contrast to the 13600, the 
> error increases with input drive voltage.
> 
> Linearity vs Iabc doesn't seem to be spec'ed for
> either device, but the 
> results I got seem to be at odds with the curves in
> the data sheet.  (Hard 
> to tell on a log plot, though.)
> 
> The bottom line is that the 13600 seems to be a
> pretty crummy 
> multiplier.  It *cannot* replace the 3080 in
> applications requiring good 
> linearity.  That's a good reason for everyone to
> grab some 3080's while 
> they are still available.  If you need the
> linearized input, you can simply 
> add a couple of diodes.  (The 13600 input simply has
> diodes connected to 
> the inputs at one end and to each other at the
> other.)
> 
>    Ian
> 
>