Jim, I had the same thoughts myself, and earlier today I decided not to pursue graphical matching any further. My first impulse for a design problem of any complexity is to implement an optimizer. I do this all the time, even for rather straightforward problems. I have a standard local optimizer (Nelder-Mead) and a standard global optimizer (differential evolution). I've cycled through many possibilities for both over the years, mostly for antenna optimization, and I have settled on these two. They are fast, effective, and sure-footed. They are also blind. They find the best performance, according to your criteria of what's good, but they offer no insight on how they did it. The global optimizer can find a design in left field that you would never have thought to check. This is a virtue, but it can leave you wondering: how did it do that? I can now appreciate the attraction of graphical matching network design with a Smith chart since the process is so intuitive and suggestive. But I was reading a technical paper earlier today about constant-Q Smith matching where more sections bought you wider bandwidth and my first thought was: the choice of number of sections could be automatically optimized!
It has been a lot of fun learning about the Smith chart. All I ever really wanted to do was implement impedance renormalization, which just seems magical to me (intentionally misloading a circuit to allow measuring it without a matching network, but then unraveling the misloading later in analysis). The other thing I wanted to implement was the Y21 method, which magically suppresses stray shunt reactance when doing a series-through measurement. It's my idea of something for free. I've looked at a lot of commercial L and C .s2p files. The Y21 method exposes the labs with sloppy fixtures.
Now I've got to get back to coding. I've implemented circles with a precise pixel-by-pixel method instead of using the compiler circle function, which didn't produce high-quality output. Neither did my usual method of drawing line segments between points. How short a line segment is adequate? My homebrew circle generator produces noticeably cleaner constant-R Smith circles. Now I'm about to do it for the constant-X curves, which currently use line segments. I belatedly discovered today that those curves are really arcs of circles.
Brian
