[sdiy] Gilbert Cells and VCAs with Predistortion

[Magnus Danielson]

From: Ryan Williams <destrukto at cox.net>
Subject: [sdiy] Gilbert Cells and VCAs with Predistortion
Date: Sat, 02 Jul 2005 16:20:58 -0500
Message-ID: <42C7053A.4040905 at cox.net>

> hi all,

Ryan,

> My internet was down for a few days and I had nothing else to do except 
> study some circuits.

How constructive of you! I had the same situation the other week and I didn't
do much useful.

> I was looking at the data sheet for the HFA3101 (Gilbert Cell) and my
> analysis left me wondering what is the best way to predistort the input so
> that the temperature and tanh effects cancel.  I  saw a circuit and
> description of the multiplier in the book 'Analysis and Design of Analog
> Integrated Circuits' that used two diode connected transistors and a
> differential current source for the predistortion.

Basically, each input is logaritmized to compensate for the exponantiation in
each input resistor, the tanh stuff is just the result of the exp.

> next I applied this predistortion circuit to a simple VCA that needs 
> only 4 matched and thermally connected NPN transistors. the circuit is 
> here (untested and maybe a bit difficult to read):
> http://www.sdiy.org/destrukto/temp/4npn_circuit.png
> http://www.sdiy.org/destrukto/temp/transient.png
> This circuit should handle about 20Vp-p input signals but the voltage to 
> diff. current converter (adapted from MAT-04 data sheet) does not work 
> at DC.

You should really make a frequency analysis on the output and measure the
overtones so that a THD measure can be taken. Do that for various signal
strengths and modulation levels. Now, this is just an idealized situation, so
only real-life evaluation counts in the end.

> next I noticed that the diodes on the LM13700 are the same as the diodes 
> in my circuit and that this would be an easy way to get the same results 
> without an npn array. so the lm13700 and important points of the math 
> are shown here:
> http://www.sdiy.org/destrukto/temp/lm13700_circuit.png

I have been thinking the same thing. The normal way that the LM13700 is not
really using those diodes all to well I guess.

> that last image shows both the tanh and the Vt temperature variable 
> being completely removed.  I assume that I have a flaw somewhere because 
> I've never seen anyone use the LM13700 with a 'real' current input like 
> this. It seems to me that this method is much better than using 
> resistors on the inputs with voltages connected.

The power of datasheets, app-notes and habit is strong.

I want the OTA-experts here to comment on this, because I have seen the same
things (including the book, which I have in my "library") and also wondered if
not that would make it sensible. One comment that I think I've heard was that
the diff current source was component hungry.

> So my questions. Have I made a mistake, is this not as good as I thing 
> it is? also, I want to build these circuits, especially the 4npn VCA 
> because I expect it could turn out to be a good replacement for VCA 
> circuits when all the OTAs are gone.  How could I go about measuring the 
> performance of these circuits (other than looking at the scope and 
> saying 'that looks good'). I want to actually have some numbers that 
> verify my thoughts.

As I was hinting, tossing in a clean sine (low THD) and measure the overtone
spectra would give you a really good idea of the linearity. Do that with
different input signal strengths and different modulation levels.

> btw, the hfa3101 looks like it would make a deccent 4-quadrant 
> multiplier but the price with the rest of the needed circuitry looks 
> like it would be about the same as or more than an AD633.

The usual problem.

Is really nobody doing the DBX VCA anymore? Comments?

Cheers,
Magnus