RE: [SI-LIST] : IBIS models at Gbit speeds

About this list Date view Thread view Subject view Author view

From: Chris Rokusek ([email protected])
Date: Mon Dec 13 1999 - 10:32:47 PST


Laurence,

An mildly-improved linearization technique was added to IBIS2XTK as of V5.01
(released March/99). The man page ($QUADHOME/man/cat1/ibis2xtk) hints at
how it works:

  -deg_vi <val> Maximum Degrees Error for VI linearization. The
                 "degree" method of linearization is the default with
                 V4.92++. Use -old_vil to revert back to the -ae
                 method of linearization. Essentially the new method
                 is the same as prior but works out to be more uniform
                 over the entire range of the curve. A Lower deg_vi
                 is more accurate. A negative value (-1.0) may be
                 used to effectively disable linearization. The
                 tangent of this value is used to limit the ratio
                 of "error distance" to "point to point distance".
                 Note that the current tolerance also affects the
                 number of points in the resultant curve. The
                 allowed range is (-90.0 < val < 90.0) although only
                 -1.0 to 45.0 is practical.

  -deg_vt <val> Max Degrees Error for VT linearization. The
                 "degree" method of linearization is the default with
                 V4.92++. Use -old_vtl to revert back to the -ae
                 method of linearization.
                 <snip>

Before continuing let me point out that changing the default values of these
arguments is likely NEVER TO BE NECESSARY...But of course some engineers
like to tweak so on with the explanation:

Without reverting to ASCII art, consider three sequential points A, B, C at
the elbow of a curve. The algorithm is stepping through all points on the
curve and eventually needs to test point "B" to determine whether it is
useful or redundant in relation to its neighbors. Drawing a triangle ACB
and then drawing a line that intersects the AC line (call this intersection
point X) at 90 degrees creates right triangles AXB and CXB.

Now point B can be removed if either of the following are true:

  1) The important geometric measures in these triangles that are related to
the error of removing point B are the ratio of lengths XB to AX and on the
other side XB to CX. If the tangent of this ratio is less than or equal to
the -deg_* command line arg for this curve then the point can be removed.

  2) If the "Y-axis" component (current or voltage) of the segement XB is
less than the tolerance (either voltage/current using -vt,-it) than it can
be removed. The primary purpose if -vt/-it is to remove measurement
"jitter."

The prior method that uses -ae does not preserve the elbow quite as nicely
and pays a bit too much attention to values near the y-axis.

I have purposely left out details of exactly how the algorithm steps through
all points in the curve.

The reason for reducing the number of points is to decrease simulation time
and decrease file size without any significant loss of accuracy.

Chris Rokusek
Viewlogic Systems

> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]]On Behalf Of Laurence Michaels
> Sent: Monday, December 13, 1999 6:13 AM
> To: [email protected]
> Subject: Re: [SI-LIST] : IBIS models at Gbit speeds
>
>
> > John Lin (�L�·�) wrote:
> >
> > Hi,
> > Indeed, XTK model contains a few number of points compared to IBIS
> > after doing IBIS2XTK. However, plotting both, you will find that they
> > are quite match with each other.
> >
> > XTK will automatically increase points at the truning corner of
> > waveform. The accuracy of simulation is not suffered.
> >
> > John Lin
> > SI engineer at Quanta Computer Inc.
> >
> >
> >
> > -----Original Message-----
> > From: Jim Freeman [mailto:[email protected]]
> > Sent: Saturday, December 11, 1999 2:02 AM
> > To: [email protected]
> > Subject: Re: [SI-LIST] : IBIS models at Gbit speeds
> >
> > Hi,
> > To my knowledge, Quad and Viewlogic Xtk don't directly use IBIS
> > models. Instead, there is a compiler the linearizes the IBIS models
> > and drastically
> >
> > reduces the number of points that are used in simulation.
> >
> > Jim Freeman
>
> I believe this reduction of points can be influenced by using
> IBIS2XTK switches
> -vt <val> voltage tolerance in Volts
> -it <val> current tolerance in Amps
> -ae <val> absolute error (0.0 < val < 1.0)
>
> Exactly how these switches affect the output, and what the default
> values are, does not seem to be documented anywhere. Anyone from
> Viewlogic/Quad care to comment?
>
> Regards,
> --
> Laurence Michaels
> PS, if you have a vacation or "out-of-the-office" auto-response
> program, please make sure it ignores mailing lists. For info,
> contact your local system administrator.
>
> **** To unsubscribe from si-list: send e-mail to
> [email protected]. In the BODY of message put:
> UNSUBSCRIBE si-list, for more help, put HELP.
> si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list
> ****
>
>

**** To unsubscribe from si-list: send e-mail to [email protected]. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP.
si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list
****


About this list Date view Thread view Subject view Author view

This archive was generated by hypermail 2b29 : Tue Feb 29 2000 - 11:39:06 PST