Search All of the Math Forum:

Views expressed in these public forums are not endorsed by NCTM or The Math Forum.

Notice: We are no longer accepting new posts, but the forums will continue to be readable.

Topic: Optimization with external simulator (SPICE)?
Replies: 5   Last Post: Jul 14, 2013 3:52 AM

 Messages: [ Previous | Next ]
 Marcus M. Edvall Posts: 356 Registered: 12/7/04
Re: Optimization with external simulator (SPICE)?
Posted: Nov 1, 2005 6:02 PM

Hi Jens,

TOMLAB /CGO is specifically designed for costly global black-box
optimization. What's the time for one objective function evaluation?

<http://tomlab.biz/products/cgo/>
<http://tomlab.biz/products/cgo/solvers/rbfSolve.php>
<http://tomlab.biz/products/cgo/solvers/ego.php>

Best wishes, Marcus
Tomlab Optimization Inc.
<http://tomlab.biz>

Jens Benecke wrote:
>
>
> Hello everybody,
>
> I have a circuit simulator (SPICE) that runs a high frequency model
> of an
> electrical device. I want to tune the parameters of this model to
> fit my
> measurements. I am considering using the Optimization toolbox for
> this.
> Basically I have two sets of frequency dependant variables m(f) and
> s(f),
> and I need to tune about 20 variables (values of resistors,
> capacitors,
> etc) within quite restrictive constraints (factor 10 at most) so
> that
>
> || m(f) - s(f) ||² == min
>
> All circuit elements are linear, within certain limits. Just R, L,
> C.
>
> Normally this would be simple (using \), however s(f) can only be
> provided
> with an external circuit simulator (SPICE in this case), it is too
> complicated to be formulated mathematically.
>
> So I need to find a way to make the Matlab optimizer decide on the
> values
> for the next iteration, but perform the actual calculation of s(f)
> by an
> external program (and then read the external program's output
> file).
>
> The idea is:
> Matlab decides on initial values for all variables
> Matlab writes SPICE netlist with those values
> Matlab calls SPICE simulator
> ... loop until max iterations or required min. difference reached
>
> My ultimate goal is to do this with m(f) and s(f) being matrices
> vectors, considering multiple "versions" of the circuit with
> different m(f)
> and parameter constraints for each one (e.g. one circuit with
> filter, one
> without filter), and running seperate external simulations for each
> one.
>
>
> Any ideas? Is this possible with the opt. toolbox? Is it feasible?
> CPU power
> and RAM is not the problem (I can order a new 8GB Dual Opteron if
> that's
> what it takes). But I would like to have an idea whether Matlab is
> the
> right tool for this at all beforehand. :-)
>
>
> Thank you!
>
>
> --
> Jens Benecke
>

Date Subject Author
11/1/05 Jens Benecke
11/1/05 Marcelo Marazzi
11/2/05 Jens Benecke
11/1/05 Marcus M. Edvall
11/2/05 Jens Benecke
7/14/13 Sandeep Sulakhe