Re: [SI-LIST] : Frequency dependence and all that jazz

About this list Date view Thread view Subject view Author view

From: Steve Corey (steve@tdasystems.com)
Date: Thu Dec 23 1999 - 13:50:02 PST


Nilesh -- please see comments below.

"Shah, Nilesh N" wrote:

> Steve:
> Thanks for your reply.
> Actually, it's not so much the rlc's I'm worried about and building
> a frequency dependent model.
> My focus is more on the simulation itself.
> I gather from your answer that if I have a simple inverter driving
> into the rlc network , the output waveform of the inverter is just a
> bunch of voltage versus time points

You're right, and a basic time-domain simulator handles time-varying signals by keeping track of x(t) and its derivative(s). A frequency-domain simulator
handles time-varying signals by tracking of the attenuations (or gains) of each of their frequency components at every point in a circuit. This is why
advanced techniques (e.g., harmonic balance) are required to simulate a nonlinear system in the frequency domain, since conversion of energy from one
frequency to another can cause frequency components to appear at the output that were not present at the input.

Reading (or just as likely misreading) between the lines, I would say be very careful, whether using a time-domain or frequency-domain simulator, about
just converting the output of the inverter into a set of v vs. t or V vs. f points and performing a linear simulation, since the driver model that you have
takes the nonlinear properties of the driver into account when you do a nonlinear simulation. Such a linearized simulation may give reasonably accurate
results under light loading conditions, but in any other situation could do you more damage than good. In my opinion, nonlinear simulation is a must for
accurate prediction of digital system behavior, since it takes loading into account.

> . So if I had a frequency dependent
> rlc network and I fed my inverter output to it, it's as good as having
> a frequency independent model extracted at the fundamental frequency.

If I understand your question, it's actually better. An accurate frequency-dependent model represents device behavior across the range of frequencies for
which it is valid. A frequency-independent model, by definition, pretends that your device behaves the same at every frequency, from DC to daylight, as
they say. It is accurate at some frequencies, but not all.

Sometimes when people talk about a frequency-independent model, they are referring to an RLGC ladder (quasi-static) representation for a transmission line,
in which the R,L,G, and C matrices are not functions of frequency. If this is what you mean by frequency-independent, then please let me know and I can
address that question as well.

  -- Steve

>
>
> Thanks again for throwing some light.
> BTW, I'd like to know more about your extraction tool.
>
> Nilesh
>
> Nilesh N Shah,
> Package Electrical Design Engineer,
> PCG, Folsom, CA.
> Intel Corporation.
> tel:916-356 1129.
>
> -----Original Message-----
> From: Steve Corey [mailto:steve@tdasystems.com]
> Sent: Wednesday, December 22, 1999 4:04 PM
> To: si-list@silab.eng.sun.com
> Subject: Re: [SI-LIST] : Frequency dependence and all that jazz
>
> Nilesh -- SPICE-like simulators doing time-domain analyses solve their
> differential equations in the time domain, based on the charge/flux stored
> in
> capacitors/inductors and the derivatives of the voltage/current variables at
> each node/branch. A time-domain simulator has no concept of frequency.
>
> It is quite possible to get frequency dependence included in a model for a
> time-domain simulator, since L's and C's are frequency-dependent
> impedances/admittances and can be shown, when used in conjunction with R's
> (and inductance coupling coeffficients) to accurately represent any
> well-behaved
> function if used in the correct topology and with the correct values.
> However, for certain structures (e.g., long cables used at high frequencies)
> a large
> number of lumped elements is required to represent the delay.
>
> A full wave solver is one of many approaches to getting a frequency
> dependent model. Other approaches, such as 2-D solvers, may be quite
> sufficient
> without going to such great lengths -- you have to evaluate what
> frequency-dependent effects you are trying to model. /* begin brazen
> advertising */ We
> offer a tool which extracts models from measured data and handles some
> effects, such as non-uniform impedance and next-door coupling, quite well.
> /* end
> brazen advertising */ General full-wave modeling from measurement is on the
> horizon, and is a current research topic in academia, but we do not offer it
> at
> this time.
>
> Hope this helps,
>
> -- Steve
>
> -------------------------------------------
> Steven D. Corey, Ph.D.
> Time Domain Analysis Systems, Inc.
> "The Interconnect Modeling Company."
> http://www.tdasystems.com
>
> email: steve@tdasystems.com
> phone/fax: (206) 417-3439
> -------------------------------------------
>
> "Shah, Nilesh N" wrote:
>
> > Hi:
> > I have a question:
> > It's all well and fine to build frequency dependent MODELS,
> > but what about the simulator(ex Hspice) itself?
> > For example, if I feed the output of a simple inverter to an
> > L,C, R network, (and I'm doing a simple time domain transient analysis)
> >
> > INVERTER------LRC NETWORK----OUTPUT
> >
> > does the simulator decompose the output waveform of the inverter
> > into it's frequency components before feeding it into the LC network?(i
> > don't think so)
> > For example, if the output of my buffer looks like a repeating square
> pulse
> > running at 400Mhz 50% duty cycle with a finite rise time, does the LRC
> > network "see"
> > the input wave as
> > Asin(w1t)+Bsin(w2t)+..... etc
> > where w1=400Mhz and w2,w3 etc are the harmonics?Or does one have to
> > explicitly
> > take the output of the waveform, decompose it and then feed it
> artificially
> > to the LRC network?
> >
> > or does the LRC network "see" as it's input,a bunch of dc points for
> > which the simulator solves the nodal matrix using difference or
> differential
> > equations? i.e. does the LRC network "see" only 1 frequency coming in
> which
> > is the fundamental?
> >
> > How does this work?How is the frequency content of the output waveform of
> > the buffer conveyed to the LCR network?
> > I was planning to investigate this but I thought some of
> > you experts out there might have the answer at the top of your head...
> >
> > Also, tools like Ansoft Q3d extract models only at 1 frequency(I think),
> so
> > unless you use HFSS, how do you get a true frequency dependent model?
> > I've generated R,L versus frequency and other parameter versus frequency
> > curves, but this is with 2 d models using TEM assumptions.what about 3-d
> > models?
> > is ther ANY way at all except for using a full wave solver like HFSS to
> > generate true frequency
> > dependent models and simulate in the time domain?
> >
> > Thanks for your ideas in advance.
> > Nilesh
> >
> >
> > Nilesh N Shah,
> > Package Electrical Design Engineer,
> > PCG, Folsom, CA.
> > Intel Corporation.
> > tel:916-356 1129.
> >
> > **** To unsubscribe from si-list: send e-mail to
> majordomo@silab.eng.sun.com. 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
> majordomo@silab.eng.sun.com. 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 majordomo@silab.eng.sun.com. 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
> ****

--
-------------------------------------------
Steven D. Corey, Ph.D.
Time Domain Analysis Systems, Inc.
"The Interconnect Modeling Company."
http://www.tdasystems.com

email: steve@tdasystems.com phone/fax: (206) 417-3439 -------------------------------------------

**** To unsubscribe from si-list: send e-mail to majordomo@silab.eng.sun.com. 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 - 11:39:15 PST