AnsweredAssumed Answered

Model Extraction

Question asked by JERRY_MARKETOS on Jan 11, 2005
Latest reply on Jan 11, 2005 by JERRY_MARKETOS
Model Extraction

(Note: If the length of this post exceeds the forum text limit, remainder will be posted in first reply)

The term "Model Extraction" means (in this instance) harnessing the power of Genesys software
to take a datafile (such as S-parameter data) of a given component or network and help the user to
obtain schematic component values which produce RF performance that matches the datafile.
The most obvious benefit of model extraction is that it provides a means to pursue an equivalent circuit
model using semi automated means and the associated time savings.  A less obvious benefit of model
extraction can be deeper (sometimes unanticipated) insight into effective circuit parasitics and their structures.

The general setup for model extraction is as follows:  (assuming 2-port circuit)

Equations are used to create internal variables (for each of the four S-parameters)
which represent the difference between the data in the datafile and the data from
a schematic within the workspace.  Given such a setup, when the difference between
the measured data (datafile) and the circuit schematic equals zero (or very nearly zero),
it can be concluded that the schematic is producing the same performance as the
datafile, and thus a model has been "extracted".

Genesys contains a built in template which is set up for model extraction, accessible from
main menu File, New From Template, Model Extract.wsp
As is often the case, there are multiple approaches that can be taken to reach a common
objective.  In the Model Extract template, the datafile is accessed by a "link to datafile"
and the equations are implemented in matrix form.  This is a perfectly valid approach, but
sometimes the link to datafile is not as well understood for obtaining data, and the matrix
equations, although simpler in text than the corresponding long form, are perceived as cryptic.

In this example we'll do exactly the same thing as the Model Extract template, just with a slightly different approach:
1) A 2-port block will be used on the schematic to access the datafile;
this is often perceived as being more easily accessible and changeable.

2) The equations will be written in long form with comments.  This often helps in understanding the origin of the data.

See the attached workspace.  (A copy of this text will also be placed in the notes page of the workspace for distribution).
Note that schematic SCH1 contains the 2-port block which points to the datafile and schematic SCH2 contains the circuit.

We begin by creating variable names on the equations page, one variable for each of the four S-parameters.
Here, I'll choose to call them  vars11  vars21  vars12  vars22
Please note there's NO requirement that the variable name contain the desired measurement- that was simply
done to make the variables more intuitive.  We could just as easily have called the variables  Tom, John, Sue and Donna

Note this setup contains a linear analysis called LINEAR1, dealing with two schematics SCH1 and SCH2
Thus we write an equation subtracting the corresponding measurements for each of the four S-Parameters as follows:
(these will be placed in the workspace equations page)

'take the magnitude of S11 from schematic 1 and subtract from it the S11 from schematic 2.
VARS11=LINEAR1.SCH1.DB[S11]-LINEAR1.SCH2.DB[S11]
'take the magnitude of S21 from schematic 1 and subtract from it the S21 from schematic 2.
VARS21=LINEAR1.SCH1.DB[S21]-LINEAR1.SCH2.DB[S21]
'etc
VARS12=LINEAR1.SCH1.DB[S12]-LINEAR1.SCH2.DB[S12]
'etc
VARS22=LINEAR1.SCH1.DB[S22]-LINEAR1.SCH2.DB[S22]

Note that once we have the variables created, there's no specific need to graph them.
The only need that follows is to set up optimization targets within Genesys such that we can direct
the Genesys optimizer to optimize in search of zero difference between the data from the datafile and data from schematic.

[file]model_extract_ex_v407_jm1.wsp[/file]
[file]08cs391.s2p[/file]

Outcomes