[SI-LIST] Re: lumped model vs distributed model
- From: "Ingraham, Andrew" <Andrew.Ingraham@xxxxxxxxxx>
- To: <jleung@xxxxxxxxxxxxxxx>
- Date: Mon, 3 Dec 2001 14:08:05 -0500
Jason,
I'd like to suggest a slightly different slant to the discussion about
lumped vs. distributed models.
I believe the question isn't necessarily whether a lumped model or a
distributed model is inherently "better," or even what is the difference
between the two. Maybe it is more correct to simply say that both types
of models are approximations of real world behavior, and neither one is
necessarily better or more accurate than the other. Either one has
plusses and minuses.
=20
It is tempting to say that the distributed model *ought* to be better.
Why? Because real transmission lines themselves are distributed, not
lumped. (Yes, there are fine irregularities if you look closely enough,
due to surface roughness or whatever, but most people usually don't need
to worry about that.)
But when it comes to writing an algorithm and some code, to simulate a
distributed model, you run into difficulties, and are forced to make
compromises and approximations, which limit the accuracy of that
distributed model.
At first glance, a lumped model looks like it would be worse, since it
begins life as an approximation ... by nature of the fact that it
"incorrectly" models real lines which are continuous. But it offers
more flexibility to the user, who can add elements to represent losses,
frequency dependencies, etc. In some cases, you can use that to your
advantage to overcome some of the shortcomings of whatever distributed
models you have at your disposal.
The standard lossless transmission line model built into all SPICE
programs is a distributed model. Within the confines of what it is
trying to model (i.e., a lossless line), its implementation is both
fast, and exact. It works correctly for all line lengths, short or
long. Some people suggest you shouldn't use this model when the line
length is much shorter than the risetime of your signals. This isn't
really true, but it may help SPICE computationally to instead use a
lumped model for very short lines. And of course this model poorly
represents real-world lines that are very long, where the losses become
significant.
The W-line model in HSpice is another distributed model, but it has
frequency dependent losses, and fairly good accuracy in most cases. The
model is also tweaked from release to release.
Lumped models tend to show some ringing ("oscillations") in simulations,
that aren't really there, which gets worse if the sections are too large
compared to the risetimes. Using smaller lumps reduces the ringing, at
the expense of compute time. In the limit with infinitely many,
infinitesimally small lumps, what you really have is a distributed
model, and this problem goes away.
But like distributed models, the inherent accuracy of any lumped model,
depends on the cleverness of the person who constructed it, to
approximate the losses, frequency dependencies, etc. (And they are
approximations, because standard SPICE doesn't have frequency-dependent
resistors or inductors with which to construct a really good lumped
model.) Simply using smaller lumps makes one part of the lumped model's
inaccuracy go away; but it doesn't necessarily make it more accurate
overall.
Regards,
Andy
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List archives are viewable at:
//www.freelists.org/archives/si-list
or at our remote archives:
http://groups.yahoo.com/group/si-list/messages
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
Other related posts: