Products > Crowd Funded Projects

Symbolic circuit analysis. Upcoming Kickstarter

<< < (3/3)

wishboneash:
I like it. It's pretty interesting and the ability to look at the dominant terms makes it quite useful.

MrAl:

--- Quote from: JosefC on February 21, 2022, 07:31:25 am ---Hello,

I want to present you my upcoming kickstarter project:

Sylina, a symbolic circuit analysis software.

Its main target is to derive formulas for transfer functions of analog circuits.
Besides Bode-, Nyquist- and Pole-Zero plots and a circuit editor.

More details on www.sylina.de/sylina-en.html
or direct to the examples on www.sylina.de/tutorial-en.html#examples.

and after that the kickstarter campaign.

JosefC

--- End quote ---

Hi,

That's interesting i have been working on the same since the early 1990's and got it to the point where i can do almost any circuit as long as the assumptions are given.

As to one of the other replies about the 'real world devices' the answer there is to just use different models that's all it comes down to.  The drawback is the resulting matrix will be much much much larger and take more time to calculate, and even fast algebraic software will take time to be able to manipulate the matrix using all symbolic variables.
However, since many analyses are based on determining how the output changes based on just one or two variables, perhaps you can have an input method that allows some variables to be constants while others are true variables and stay that way throughout.

The way i am doing it now is with all variables but i recently considered doing the partial constant and partial variable update.

One of the more complex aspects is modeling things like transistors.  Some of this simply will not have a symbolic solution because in the end the solution must be numerical.  This involves the exponential functions which can get pretty hairy even with the simplest of transistor models.  This would probably limit the functionality to much simpler transistor models such as the T equivalent or something like that where we have just a current controlled current source and some resistors, or a voltage controlled current source and resistors.

I would be interested to hear your ideas about these problems if you  care to share.

MrAl:

--- Quote from: Mechatrommer on June 26, 2022, 06:50:35 am ---
--- Quote from: JosefC on February 21, 2022, 07:31:25 am ---Any comments are welcome.

--- End quote ---
keep going. haven't look closely but i wish something that derives transfer function from actual/empiric device's/circuit's behavior, not from theoretical formula, because that we already have in textbook.

--- End quote ---

I forgot to reply to this post so just now.

I've been working on a symbolic analysis program since the early 1990's and have gotten pretty far with it and along the way i discovered many things about symbolic analysis vs numerical analysis that simulators use.

To answer your wish, i can tell you that if you need to analyze a behavior that is not strictly theoretical but also need to consider parasitic effects for example, all you would need to do is add those parasitics to the schematic yourself, which in most cases would mean creating a model for whatever it is you need to look at.  You could get a lot of info from a spice model to help with this.

The drawback is that the more components you add, parasitic or not, the longer the solution will take because for a symbolic analysis the software has to deal with very very very long expressions.   For example for a buck converter with various parasitics i did i ended up with expressions that would span several regular size book pages.  That's because symbolic expressions for larger circuits have to contain every component in the whole system for every node unless some components get absorbed into others, but that is fairly rare.  Sometimes you can simplify them a little by bulking up some parts of the expressions but it's not that easy to do and it still ends up very long.
The beauty of this though is that you can change any component or time value or whatever and do a calculation, and the calculation once the expression has been obtained completes very fast because computers work well with numbers.  The only problem there is that you may have to go to a numerical precision that is not math coprocessor friendly, which means you have to use big numbers.

As i was saying in the previous post, some symbolic solutions are simply not possible because there is no closed form for some circuit calculations.  An example is a transistor that uses diodes in the model.  The exponential terms can not always be solved symbolically and so we end up back to numerical solutions.  This is even typical with circuits with just diodes in them too.
So it seems that for some things we will always be dependent on numerical solutions.
It doesnt mean that the symbolic analysis is not useful though, because we can go so far with it until we have to jump back to using numerical solutions.  So in a way we get some usefulness even if not the entire solution in one shot.

JosefC:

--- Quote from: MrAl on December 03, 2022, 11:51:50 pm ---
...
However, since many analyses are based on determining how the output changes based on just one or two variables, perhaps you can have an input method that allows some variables to be constants while others are true variables and stay that way throughout.

The way i am doing it now is with all variables but i recently considered doing the partial constant and partial variable update.

One of the more complex aspects is modeling things like transistors.  Some of this simply will not have a symbolic solution because in the end the solution must be numerical.  This involves the exponential functions which can get pretty hairy even with the simplest of transistor models.  This would probably limit the functionality to much simpler transistor models such as the T equivalent or something like that where we have just a current controlled current source and some resistors, or a voltage controlled current source and resistors.
...

--- End quote ---

The kickstarter campaign was not successfully. In the moment I don't work on the project.

In the moment I do an analysis with all variables. To use only some variables is sure an idea for some optimizations.

My analysis is purely linear. The circuit is linearized around an operating point.
I don't know of any symbolic circuit analysis software that works direct with nonlinear
functions.

A possibility would be to do a sequence of linear analysis and move the operating point.
But this is probably very slow.

For me is symbolic analysis no replacement for numerical analysis like SPICE.
It is more to check the principal design of the circuit and to verify the manual
analysis/ideas behind the design.

Nonlinear effects and parasitics must be checked with SPICE, statistics and
lab measurements in my opinion.

Thank you very much

SiliconWizard:
While it's an interesting project technically, not to rain on your parade, but what is the real purpose of symbolic analysis here?

Beyond simple circuits, the symbolic analysis will not really be usable by humans, as the calculations become complex very fast, so one can question the benefit of it. If you use the output of the symbolic analysis to get response values, for instance, what's the benefit compared to simulation? You get the same output. Pretty much.

Does your symbolic analysis allow to solve for some parameters? (For instance, solving for values for resistors and capacitors of some filter to get a given response?) If so, then there would be an added benefit, although you can also do this using simulation and optimization algorithms.

So I'm not too surprised the campaign was not successful. Maybe you should define the purpose of this for real-life use.