[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: ScanTesla program



Original poster: Terry Fritz <teslalist@xxxxxxxxxxxxxxxxxxxxxxx>

Hi Antonio,

At 01:46 PM 5/15/2005, you wrote:
Tesla list wrote:

Original poster: "Jim Lux" <jimlux@xxxxxxxxxxxxx>

Matlab wouldn't find closed form solutions.  You build the matrix up from
the component values, then solve it for the voltages and currents (which is
what SPICE does for the "steady state" AC solutions).  Nothing keeping you
from changing the matrix as a function of time.

In this case we are not interested in the sinusoidal steady state
solution. The input is a square wave, and the transient is important.
The simulator that I have implemented in the sstcd program uses
state equations, solved by numerical integration with a trapezoidal rule. It´s simple and efficient, for a fixed structure.

That is what I am "trying" to do too! I have a 3GHz computer with 1G of RAM and the speed is stunning!! Given a +-340V square wave input and say a 5pF+220kOhm load. The program and computer should have the power to scan just about "all" the reasonable parameters to find the very "best" ONE in a reasonable amount of time. I think C is the way to go, so I will not try to do it first in "Microsoft Quick Basic" this time :o))) I really like Lcc even though it has FAR more power than I know how to use... If I ran Linux all the time, it would be Gcc which is even more powerful... My C++ book had so much dust on it I could not see the writing on the front cover ;o))) I spent two hours figure out what "nan" meant ;-))


I, too, thought that.  But once you've got a half dozen or so components in
the circuit, it gets easier to write a simple routine to crank up the
admittance matrix (you write a "connect component between node x and node y"
routine, so your circuit design is just a series of calls to that routine).
Especially if you make the routine something like:
addcomponent (complex impedance, node 1, node 2)
and then do
Ymatrix = zeroes(20);        % makes a 20x20 matrix of zeros
addcomponent (resistance, a, b,Ymatrix)
or
addcomponent (1/(1j * omega * capacitance), a, b, Ymatrix)

Jim is probably right ;-)) But I think the C libraries "already have" the functions we need already(!) but my programming skills are poor so I have to struggle to find out how to use them. If anyone has any examples of how that stuff works it would help "me" a lot ;-)) I need to figure out the complex number + - X / and possibly the complex matrix manipulation functions. I guess you have to define number "structures" for the complex numbers. I am pretty lost there 0:-) But I can figure it out eventually...


It's possible to write a nodal system that solves for the time-domain solution at a small time interval ahead. A bit more complicated that sinusoidal steady-state analysis, and somewhat slower than state equations. But equally precise solutions are possible. If everything is linear, that solution can be very fast.

:-))) That is what I am thinking too. The advantage is that odd things like streamer loading or streamer load varying over time is very easy to add to the program if it is time domain.


I would not consider using Matlab, Excel, or similar huge packages for this, quite simple, analysis, that has to be done fast. Everything that is needed is a routine to solve a real system of linear equations.

:-)))))) It is also easier for others to adapt and go forward from. There are millions of super good C programmers... All of them can take a simple C program and go forward with it... If you start to use "not so popular" things, then the number of experts is vastly less and the program looses value. I think someone had E-Tesla6 running on a palm at one point ;-))) I especially like the "free and easy to change" part since so many others can "enjoy and play" with us too!! With E-Tesla, everyone could easily handle the "program", but I forget how many times I had to explain how to make the pretty charts with Excel or MathCad.... MicroSim is the same in that it needs a little help to get started... Simple C programs just ask for the basic stuff and give the answers without torturing the users to figure out a fancy GUI program with they don't care about in this case already. Antonio seems to be able to make nice little GUI programs that are real easy to figure out though!!!


I wonder if a program is given say a +-340 volt input into a 5pF +220K load that the program would be a "one time" thing...

After 100+ years of fiddling with Tesla coils, it might just tell use that the secondary should be (making the numbers up) 34.5878564378056mH, the toroid should be 21.232352525pF, the primary 56.565791823561278567uH and the Cp 42.34645485468nF.... and then it would be all "done"... I think there very well be , given the model I mentioned, a single "ultimate BEST answer".... So if you want the best 5 foot streamers, use this Cp, Lp, k, Ls, Ct and you have the "best" coil... All done... Of course, there is still the challenge of making the physical machine to match the magic ;-)) But then we will have Paul's models and such to push for further advances till the changes really "just don't matter anymore..."

I should mention a few things,

As far as I can tell, the 1pF/foot + 220k ohm streamer model is holding up very well!!! Just small variations from it give "bad" results... I have no problem using it as the "perfect model" until someone can prove something better. Steve Conner suggested an "on time" for it, but the streamer grows mostly inside the on-time so "I" don't think that matters.

I think it is VERY important in include Rp and Rs!!! They have a dramatic effect!! Although it is "easier" to work in the loss-less case, the real world seems to be very affected by them. Loss-less cases are nice for some theoretical stuff, but TC losses, even in the DRSSTC case, are very REAL!!! Without them, the DRSSTC could easily put out millions of volts. Losses "restrict" the DRSSTC voltages very dramatically!!

For a model of a Tesla coil, I think this is the "best" there is now (first figure):

http://drsstc.com/~terrell/modeling/ScanTesla.pdf

Other factors just don't seem to matter much and this models predicts things very well. Changes to the parameters from what is known to be true, tend to give wrong results... I have always been happily surprised at how well the streamer load model works. It have never failed us yet!!!

So Antonio, if you (or anyone else) can write up the best program to explore this stuff, PLEASE DO!!!!!! I am pretty lost at the moment and you probably have a better understanding and ability to use the tools than anyone. If not, I will do it eventually for sure!! Please let me know if you do so I will not waist time trying too ;-))

I think this is all sort of a major turning point!!! The DRSSTC is a very stable and predictable machine. We have the ability to measure just about everything about them. We have the computer tools to study them... It is very surprising how well the present models predict them!! ;-)))* It really is just a matter of "doing it" now...

*Here is a very simple, but very effective DRSSTC model:
http://drsstc.com/~terrell/modeling/Model-DRSSCT.gif
http://drsstc.com/~terrell/modeling/

Cheers,

        Terry


Antonio Carlos M. de Queiroz