Using genetic algorithms to optimize pentode parameters

[Bitrex]

I've been trying to get back into doing some computer programming after not doing much for many years. Back in college I took a course in "AI" programming techniques, and was introduced to the concept of "genetic programming" to solve optimization problems; i.e. you start with some random possible solutions to the problem, evaluate their "fitness", and then keep crossbreeding/mutating the possible solutions until one that fits your criteria is found. I was thinking the other day that this might be a way to find the optimal component values for a pentode amplification stage at a given supply voltage and desired operating current from the SPICE model instead of resorting to empirical methods since all the parameters interact with each other in a complex way that's a nightmare to find an exact closed form solution from the equations. You could say you had a supply of X volts and wanted to run the tube at a certain quiescent current and optimize it for lowest distortion or highest gain. Or maybe find the "best" operating point for a given tube for a certain purpose. There's no question in this post, really - just thinking out loud.

[Robert McLean]

What you describe sounds like neural net programming. But in any event, I dont think you need anything that exotic. Just use parameter stepping in SPICE to vary as many components you like, plot results, and pick the best combination. This is standard Spice. Some Spices such as MicroCap, and presumeably many others, also have Monte Carlo simulation which could be used to do what you want.

[diy4]

For those who are unfamiliar with the concept, here's a link:

http://www.damninteresting.com/?p=870#more-870