[sdiy] Temperature Compensated Exponential ConverterUsingSSM2164

[Magnus Danielson]

David G. Dixon wrote:
> I've done a little bit more careful analysis over a more realistic
> temperature range of 19 to 31 deg C, and -3V to +3V control voltage, and
> have now determined that, within these ranges, assuming perfectly matched
> 2164 VCAs, it should be possible to achieve 1V/oct conformance to within
> 0.02%.  This means, for example, that if A440 is exactly 440Hz, then A880
> will be within +/- 0.18Hz of 880Hz!  In other words, with proper trimming of
> the tempco voltage V_TC and the feedback resistance R_FB, virtually perfect
> tempco can be achieved.  However, one does depend on the other, and the
> result is VERY sensitive to achieving the correct value of V_TC.
> 
> In case you are curious, here are the relationships (assuming 1V/oct,
> 3300ppm/K, and 33.3mV/dB):
> 
> R_FB = R_CV * log(2)*[1.0033^(1.0033/0.0033)]/1.5 = 0.546
> 
> (i.e., R_FB = 54.6k if R_CV = 100k) and
> 
> V_TC = log(1.0033)*(1.0033/0.0033)/1.5 = 0.290
> 
> The values are somewhat sensitive to the gain coefficient (1/1.5), but
> completely insensitive to the gain temperature coefficient (0.0033).  Here
> are a few sensitivity results based on minimization of the sum of square
> errors to demonstrate the point:
> 
> GC      GTC      R ratio   V_TC
> ----------------------------------------------
> 1/1.5   0.0033   0.54542    0.28945  (baseline)
> 
> 1/1.5   0.0030   0.54544    0.28946
> 1/1.5   0.0036   0.54540    0.28943
> 
> 1/1.45  0.0033   0.56423    0.29943
> 1/1.55  0.0033   0.52782    0.28011
> 
> A change of +/- 10% in the GTC only changes the values at the fifth decimal
> place!  Similar changes to the GC have more significant effects, but these
> changes are far beyond expected tolerance levels.
> 
> I think we have a winner!

The one thing I worry about, is how feasable it is to actually do these 
adjustments, especially in a typical DIY environment (I can use any of 
the climate ovens at work, a luxury which few has) and also how stable 
they are, if they need to be very carefully tuned.

So, how would a practical tuning procedure of such a design work? We can 
play with all the math we want, but it doesn't help the normal guy. A 
more complex synth could use one or more DACs and computer control to do 
auto-tuning, but for sanity reasons I want to see what a manual trimming 
would include. Maybe include heater at the SSM2164 and cut some PCB 
around it so it becomes more termically isolated so that it can be 
forced to exercise different temperatures? I think it would be hard to 
trim this without temperature cycling.

Maybe include a trimmer to the reference current could be necessary 
here, to make the scale and tempco trimmings much more direct as the 
normal offset and scale trimmings now have become offset, scale and 
tempco trimmings, so the traditional offset/scale trimming loop becomes 
troublesome. Adding the reference trimmer allows the unscaled and scaled 
offsets to be trimmed independently of the scale trimmer. Using that 
simplification the scale and tempco balance can be handled easier and 
more direct.

Now that there seems like it is possible to get good cancelation, we 
need to look at the practicality of achieving it.

I think we need to have some heater/tempsensor hooked to it only for the 
purpose of trimming.

Cheers,
Magnus