[SI-LIST] Re: N-port model limitations in simulators

  • From: Ray Anderson <Raymond.Anderson@xxxxxxx>
  • To: si-list@xxxxxxxxxxxxx
  • Date: Mon, 28 Apr 2003 11:33:09 -0700 (PDT)

Since my original posting on the topic several days ago an amazing amount of
really interesting and valuable technical discussion has taken place. Thanks
to all who have chosen to respond.

Before I get to the point of my present posting, I just wanted to comment
that the amount of back and forth discussion that has taken place in the past
week between learned colleagues across the globe would have probably taken at
least a year to transact perhaps as recently as just 10-12 years ago. It
would have started with a letter to the editor of some technical journal.
There would have been months delay prior to publication. Someone would have
responded, more time would have passed before the response was published and
so on and so on ...... Despite all it's shortcomings, the internet has proven
to be a fantastic medium for technical interchange.

Getting back to the original topic, I'd like to submit that all the questions
and discussion regarding n-ports and simulators that are the current topic of
this string are relevant to the way n-port models were utilized by the
"microwave" people prior to today's SI engineers taking notice that circuits
characterized by s-parameters are a powerful tool.

Prior to getting involved in the SI realm I used to do microwave circuit
design for satcom applications. The simulators then available were pretty
crude by today's standards (Compact and Touchstone). I primarily used
s-parameters to model amplifiers, filters, and various linear networks in the
frequency domain. Today we are utilizing similar circuit theory concepts to
model high-speed digital links in the time domain. While we are generally
concerned with concepts like return loss and gain and loss in relation to
microwave circuits, today in the digital realm we are generally interested in
characterizing a complete communication's channel end-to-end and quite often
neglect to recognize that the traditional concepts of return loss, loss and
gain are are alive and well in each and every link simulation we perform even
though in our quest for the eye-diagram (and it's info on jitter and
amplitude) at the end of the link we may not recognize the implicit 
contributions of these basic mechanisms.

Also, we are concerned with certain circuit configurations and concepts in
the SI world that were not often considered in microwave circuits.  A couple
examples that come to mind are the modeling of power planes of a size that is
large wrt to lambda and the ability to stack planes. I would submit that these
two examples can place special requirements on the n-ports characterized by
s-parameters that might not be so important in some other applications.

In the case of power plane modeling a couple of issues immediately come to
mind. One is that planes are usually large in comparison to wavelength.  When
one measures s-parameters of a plane they will probe the plane at the two
ports of interest. There will be a local reference at each probe location.
Now let's utilize the captured s-parameters in a circuit simulation. If you
utilize an n+1 2-port model in a simulator are you really doing the right
thing? I don't think so. Since there is but a single reference node you will
be misinterpreting any simulated voltage since one (or both) of the ref nodes
has lost their true physical relationship.  In order to measure a voltage it
must be in relationship to some reference which needs to be physically
located such that it is in the same relative place in time with respect to
the signal node (co-located near vertically). Failure to do this will result
in an undefined result where the signal node is at some point in time other
than that of the ref. node.  What does measuring a voltage across time mean

Another aspect of plane modeling stems from the ability to model stacked
planes. I won't get into the details except to say that it is possible to
stack planes defined by a t-line matrix or two-ports if there are separate
references. when using models with a single reference it isn't apparent to me
how to properly stack the planes.

For the applications that dictate the use of 2*N terminal models as opposed
to N+1 models the N+1 model may be converted to a 2*N terminal model by the
addition of perfect 1:1 transformers on the input and output of the n-port.
the following snipped of hspice code illustrates the method:

* ref1 is reference for left side port
* ref2 is reference for right side port
* ref_int is reference terminal for S-element

Ein n1 ref_int  TRANSFORMER IN  ref1  1
S1 n1 n2 ref_int mname=smodel
.model smodel s tstonefile=nport.s2p
Eout OUT ref2  TRANSFORMER n2 ref_int 1

Thanks to all who have entered the discussion. It is truly proving to be
an education.

-Ray Anderson

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:

For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field

List archives are viewable at:     
or at our remote archives:
Old (prior to June 6, 2001) list archives are viewable at:

Other related posts: