[Qucs-devel] Re: ASCO

[Stefan J. <st...@gr...>]

Am Mo, 15.05.2006, 11:03, schrieb asco developer:

> Hello Stefan,

Hi!

> I have quickly looked at the files you have just sent to me.
>
> ...
>> Attached the "auto-fied" version of the 0.4.1 asco code.  I added
>>   autogen.sh
>>   configure.ac
>>   lots of Makefile.am's
>
> Well, lots of new files. Why do I need to have Makefile.* in the doc/
> and examples/ directories? Is it really necessary?

Yep.

>> These should go into CVS.  When you have these files from CVS just type
>>
>>  $  sh autogen.sh
>
> So far I do not have CVS.
>
>>
>> and everything should be created
>>
>>  $ ./configure --enable-maintainer-mode
>>
>> You should use the above command line.  You get a ready to distribute
>> package by:
>>
>>  $ make distcheck
>
>
> I have tried
>
> $ sh autogen.sh
> $ ./configure --enable-maintainer-mode
> $ make distcheck
>
> and I get the following error.
>
> cp: cannot create regular file
> `../../asco-0.4.2/tools/alter/../auxfunc/auxfunc.c': No such file or
> directory
>
> I will investigate what is really happening.

Any error messages during 'sh autogen.sh'??

>> Also I've incorporated some patches to the source files so you can
>> cross-
>> compile asco to mingw32.  So you can use it on Win32 natively.  It
>> compiles
>> but we'll see if it also really works...
>
> I will patch my files according to your changes so that future
> releases include your work.

Ok, thanks.

>> Also attached some examples (schematics+netlists).  We don't have a
>> BSIM3
>> but a simple MOSFET model.  I tried to setup the inverter example using
>> the
>> simple model.  Just have a look.  The other examples should be clear.
>> If
>> not, just ask.  One note about the "sp_fit" examples.  I think it should
>> be possible to use asco to find the R,L,C values for the elements in the
>> bfg425w surrounding network to fit the measured data to the simulated.
>> What do you think?
>
> The example look well and once I have them documented, I will include
> them in ASCO.
>
>
> Concerning your question about RLC curve fitting using ASCO, I see it
> as very much possible. Once you define how who want to evaluate how
> close the simulations are close to the measurements, this is not a
> problem. I have helped others doing similar thing. Either fitting
> about 17 variables in a RLC circuit to a measured frequency response
> (magS21) or full matching (Yxx) 32 variables of a power high-frequency
> transistor.
>
> The procedure is to find at every or some selected frequency(ies) the
> error between simulation and measurements. Add them together to make
> one cost. Then, ASCO (or any other optimizer) just has to minimize the
> error. It looks more difficult than what really is! :-)

That is why I already added the equations.  The task would be to
minimize those *_rms_* values shown in the sp_fit_dpl.ps at the bottom...

Thanks, Stefan.