[TowerTalk] OT ?? - Manual tuners

[jimlux]

David Gilbert wrote:
> I don't have the schematic for that tuner handy, but assuming it is a 
> T-Network the most efficient setting will probably be the one where 
> either the input or the output capacitance is closest to maximum (which 
> capacitor depends upon whether the load impedance is less than or 
> greater than the source impedance (rig, coax).  In general, the most 
> efficient setting for a T-Network tuner is the one that makes it looks 
> as close to an L-Network as possible.  You can prove this to yourself by 
> checking out the online T-Network virtual tuner by W9CF at 
> http://fermi.la.asu.edu/w9cf/tuner/tuner.html .... you can set the 
> values manually yourself, but if you use the "autotune" function it will 
> seek the most efficient match.  It will even tell you the efficiency for 
> various settings.   It's a very cool application.
> 
> 73,
> Dave   AB7E
> 

I've been working on genetic algorithms to set stepped L/C networks for 
a variety of purposes (phased arrays and tuners), and have looked at 
W9CFs modeling and adjustment code.  I don't think his code actually 
solves for an optimum, nor does he claim it does.  He uses a good rule 
of thumb strategy.   It just finds a "good enough" setting:

"The autotune algorithm tries to minimize the inductance used by first 
finding a starting match by setting one of the capacitors to its maximum 
value and if that fails it tries each of the components at its current 
value. If no initial match is found, it gives up and a "Tune Failed" 
will be written in the message panel. If a starting match is found, a 
binary chop is done between the starting inductance value and zero to 
find the smallest inductance consistent with a match. Note it is 
straightforward to program a search algorithm that finds the minimum 
loss, but since our tuners don't come with a loss meter, a rule like 
tune for minimum inductance is much more useful. In any case, the Q of 
real components changes as they are adjusted."

"Note, if you are looking for design software for optimal networks with 
well characterized components, please look elsewhere."

---

that said, I think the idea of "minimize inductance" is a good rule of 
thumb, because the L is the lossier component, so minimizing the current 
through it seems like a good idea.

The XLZIZL excel spreadsheet *does* use an optimizer to solve for best L 
and C for lowest loss, but I'm not sure that the Excel Solver (which 
uses Nelder-Mead) is the best approach.

There's also some papers out there on provably optimum solutions, but I 
think they're of not much practical use (any more than W9CF's rule of 
thumb approach):  "real" tuners with "real" components aren't likely to 
follow the model assumptions.  For instance, Q is not constant for most 
variable components.  that's why I've been fooling with genetic 
algorithms (aside from the fact that they deal with the discrete relay 
switching quite nicely)... I can have an arbitrarily complex underlying 
model (or even better, measured data), and let it churn until it works 
the best.

Even slicker would be a "hardware in the loop" technique (e.g. measure 
net power to the antenna and "tune for maximum smoke")