[sdiy] Temperature Compensated Exponential ConverterUsingSSM2164

[George Hearn]

Slight change of tac but...

The CEM3374 (Oberheim Matrix 12 / Xpander etc.) almost completely
obliterated temperature drift in by providing a 3300ppm/k dependant voltage
on a pin.  If this voltage was used to derive the DAC reference (multiplying
DAC) that drove the VCO then almost perfect temperature compensation is
achieved.
The drawback is of course that all your modulation sources must then be
digital since the input to the multiplying DAC is of course a binary code.
Now, what has stopped us from using this approach in analogue, is that it is
very difficult to make a multiplier that is both very linear and not
temperature dependant (where a multiplying DAC is both very linear and
almost temperature independent).
Set up a simple simulation in SPICE using a classic two transistor expo,
then have a third (identical) transistor with a constant collector current
in the simulation.  Use a multiplier primitive to multiply the input to the
exponential converter by the Vbe of the transistor with a constant collector
current and the result is for all practical purposes *perfect* temperature
compensation (try even 0 to 100degC).
By using the linearization technique with the SSM2164 one can make a linear
multiplier that is almost temperature independent and we know that the
linearity of such a circuit is good over a reasonable range and better as
this range is reduced.
So, why not use the SSM2164 to make an exponential converter temperature
compensation scheme that uses the SSM2164s as the multiplier for the
temperature sensor which is itself a transistor on the same die as the expo
pair (MAT04, LM3046).  The limitation on temperature stability would now
only be the linearity and temperature dependence of the linearized SSM2164
circuit, which is very good.  
If you simulate this with whatever model of the SSM2164 you have, the result
is perfect temperature compensation.  However, mismatching between SSM2164s
will result in some deviation from the ideal, how much means resorting to
breadboard!

Thinking outside the box!  George

-----Original Message-----
From: synth-diy-bounces at dropmix.xs4all.nl
[mailto:synth-diy-bounces at dropmix.xs4all.nl] On Behalf Of David G. Dixon
Sent: 01 September 2009 19:09
To: 'Ian Fritz'; 'Ingo Debus'; 'synth-diy'
Subject: RE: [sdiy] Temperature Compensated Exponential
ConverterUsingSSM2164

This may be stupid, but...

What about a theoretical approach?  We already know that the trim values are
not sensitive to the gain temperature coefficient.  What if we just measure
the gain coefficient of the VCAs, calculate the necessary feedback resistor
value, install the nearest 1% resistor and/or trim to that value (by
measuring the resistance with a DVM), and just tune the VCO with the tempco
voltage trim by eliminating the beats at a two octave interval?

Alternatively, one could simply apply the theoretical resistance of 54.5k
and trim the tempco voltage.  According to my calculations, even if the
inverse gain coefficient is off by +/- 0.05 from the stated value of 1.5
(which is probably severe), the worst-case tempco conformance is still only
0.16% at 54.5k, at a CV of -3V, at 31 deg C.

It seems to me that the whole point of this design is that it compensates
for temperature very effectively.  It should be that much easier to achieve
excellent performance by trimming, not harder.  Why not take advantage of
this fact, using the math we know to be true?


> >>>I think we need to have some heater/tempsensor hooked to it only
> >>>for the
> >>>purpose of trimming.
> >>
> >>How will you know your temp sensor isn't lying?
> >
> >How precise do we need it to be? Maxim, for instance, makes some nice
> >digital temperature sensors with up to +/- 0.5 °C precision. I think
> >the bigger problem is to keep the sensor and the device under test at
> >the same temperature.
> 
> Yeah, I just used the old National whatever deg Kelvin sensor.  Being off
> by a couple of degrees doesn't much matter, especially since you only use
> the *change* in temperature.
> 
> I mount the board in an Al box, which spreads the heat out quite
> evenly.  As an extra, I surround the board with crumpled-up newspaper to
> eliminate drafts.  Oh, and I mounted the T sensor close to or incontact
> with the expo pair. Even if the sensor isn't well coupled it doesn't
> matter
> much, again because we are interested in temperature *differences*.
> 
> We do not need super precise or accurate results here.  But I agree that
> previous experience in doing this kind of work does help.
> 
> 
> Ian
> 
> 
> _______________________________________________
> Synth-diy mailing list
> Synth-diy at dropmix.xs4all.nl
> http://dropmix.xs4all.nl/mailman/listinfo/synth-diy


_______________________________________________
Synth-diy mailing list
Synth-diy at dropmix.xs4all.nl
http://dropmix.xs4all.nl/mailman/listinfo/synth-diy