[sdiy] Compensating PTC and NTC resistors

[René Schmitz]

Barry Klein schrieb:
> Just saw this.  Maybe it puts more wood to the fire of the ongoing
> "discussion"....

Well, I this thought had actually crossed my mind as well, but as with 
such things I didn't persue that.

Like neural networks, genetic algorithms are suited to a wide variety of 
optimisation problems. But shooting them at a problem in 3 or 4 
variables, seemed a bit overkill to me. The higher the dimension of your 
search space gets the more you are benefiting (or even haveing to resort 
to) such algorithms.
(My first real contact with GAs was a 2D-Image recognition technique 
where hand gestures would be recognized, I think the hand model had 18 
or 19 variables, luckily I did just have to do a seminar paper, and not 
implement it.)

Here the problem has 6 variables, and might already be at the point 
where you get a benefit. If you have to often design such networks.

But then you can as well write a program that (basically *) brute 
forcely tests all possible combinations of E96 values, and delivers its 
possible candidates also within seconds. No trouble with convergence at 
all, and you'll really find an optimum solution...

* Well you'd make some guesses on the ranges, to speed it up a little... 
Or use a binary search strategy in to narrow down possible candidate 
ranges, which you then search exhaustively. Makeing use of steadiness of 
the functions involved.

Cheers,
  René