Wanted: the definitive RF equivalent model for a resistor!

[Paul Burridge]

Hi guys,

I've seen a lot of stuff on the net about Spice modelling that takes
into account paractics at RF, but they typically turn out to be....

-----inductor-----resistor------inductor--------
| |
| |
-----cap-----|


Or othertimes, the cap is shown as being across the entire network;
ie., at the outer extrameties of the inductors. The difference in
practice between the two examples is *large* in effect so which is
correct if either, and equally importantly, how can the skin effect be
incorporated into the model to give a more complete picture?

 

[Andrew Tweddle]

Hi guys,

I've seen a lot of stuff on the net about Spice modelling that takes
into account paractics at RF, but they typically turn out to be....

-----inductor-----resistor------inductor--------
| |
| |
-----cap-----|


Or othertimes, the cap is shown as being across the entire network;
ie., at the outer extrameties of the inductors. The difference in
practice between the two examples is *large* in effect so which is
correct if either, and equally importantly, how can the skin effect be
incorporated into the model to give a more complete picture?

Try Puff at http://www.gerstlauer.de/puff/

It's models will take you to 50 Ghz.

Regards Andrew

 

[Fred Bartoli]

Hi guys,

I've seen a lot of stuff on the net about Spice modelling that takes
into account paractics at RF, but they typically turn out to be....

-----inductor-----resistor------inductor--------
| |
| |
-----cap-----|


Or othertimes, the cap is shown as being across the entire network;
ie., at the outer extrameties of the inductors. The difference in
practice between the two examples is *large* in effect so which is
correct if either, and equally importantly, how can the skin effect be
incorporated into the model to give a more complete picture?

How's physically made a resistor ?

Thanks,
Fred.

 

[Paul Burridge]

How's physically made a resistor ?

Metal film in this instance.

 

[Mike]

Hi guys,

I've seen a lot of stuff on the net about Spice modelling that takes
into account paractics at RF, but they typically turn out to be....

-----inductor-----resistor------inductor--------
| |
| |
-----cap-----|


Or othertimes, the cap is shown as being across the entire network;
ie., at the outer extrameties of the inductors. The difference in
practice between the two examples is *large* in effect so which is
correct if either, and equally importantly, how can the skin effect be
incorporated into the model to give a more complete picture?

I suspect models with the cap connected from end to end are better suited
to surface-mount resistors, which have capacitance from end to end and
little lead inductance. If you're modeling a discrete metal-film resistor
there's additional inductance in the leads, so the model you've shown might
work better.

Ken Kundert wrote a paper called "Modeling Skin-Effect in Inductors," that
can be found on his web site http://www.designers-guide.com . Switch the
inductor L and resistor Rs in Kundert's model, and you have the model you
show above (with an additional resistance in series with the capacitor).

Kundert is primarily after a Verilog-A model, and since he can create
complex mathematical models in Verilog-A, he doesn't need to come up with a
circuit representation. He does spend some time talking about the model,
though, and what the model is trying to match.

The difficulty with modeling skin effect with RLC networks is that the skin
effect generally results in the resistance having a sqrt(j*w*f) term, which
has no exact RLC equivalent. If your simulator doesn't have a skin-effect
term in the resistor model, you can model the effect with poles and zeros,
each of which will require additional network elements. The first order
approximation is roughly what you've shown above - it may or may not meet
your requirements.

-- Mike --

 

[Fred Bartoli]

Metal film in this instance.
--

That's not what I meant.
Have a look at your resistor, where are the... wires connected, where
is(are) the resistance(s) distributed, where is(are) the capacitance(s)
hidden, and so on.
Then which effects can be significants, which can be neglected with respect
to others and to your frequency range of interest.

That's as simple (or as complicated) as this.

Try a bit harder Paul, I sure you can sort this out.

For example, take a leaded resistor with 2 looong 1" leads (i.e. 2x25nH lead
inductance) and 1pf parasitics.
What is the resonant freq ?
Is the difference between both models significant at your working freq ?
What happens when you trim the leads ?
What if you use SMDs ?

Thanks,
Fred.

 

[Chaos Master]

Andrew Tweddle ([email protected]) said those last words:

Try Puff at http://www.gerstlauer.de/puff/

The download link does not work for me.

[]s

 

[Fred Bartoli]

Andrew Tweddle ([email protected]) said those last words:

The download link does not work for me.

Be sure to use your email address (or at least a credible one) as the
password an "anonymous" as the username, as noted at the end of the page.

Thanks,
Fred.

 

[Paul Burridge]

Be sure to use your email address (or at least a credible one) as the
password an "anonymous" as the username, as noted at the end of the page.

I get the initial link come up okay but can't download the program;
just get the usual page not found error.

 

[Paul Burridge]

That's not what I meant.
Have a look at your resistor, where are the... wires connected, where
is(are) the resistance(s) distributed, where is(are) the capacitance(s)
hidden, and so on.
Then which effects can be significants, which can be neglected with respect
to others and to your frequency range of interest.

That's as simple (or as complicated) as this.

Try a bit harder Paul, I sure you can sort this out.

For example, take a leaded resistor with 2 looong 1" leads (i.e. 2x25nH lead
inductance) and 1pf parasitics.
What is the resonant freq ?
Is the difference between both models significant at your working freq ?
What happens when you trim the leads ?
What if you use SMDs ?

Thanks, Fred. I take your points. In the modelling I've been doing,
the SRF has come out at 6.5Ghz (assuming much shorter leads with 2n of
inductance and a Cpar of 0.3pF. But I'm going to return to this with a
questionable example from Chris Bowick's book when time permits.

p.

 

[Chaos Master]

Fred Bartoli (fred._canxxxel_this_bartoli@RemoveThatAlso_free.fr_AndThisToo)
said those last words:

Be sure to use your email address (or at least a credible one) as the
password an "anonymous" as the username, as noted at the end of the page.

Now it worked for me. I used the standard Windows FTP client (start > run >
ftp).

[]s

 

[fellow]

Hi guys,

I've seen a lot of stuff on the net about Spice modelling that takes
into account paractics at RF, but they typically turn out to be....

-----inductor-----resistor------inductor--------
| |
| |
-----cap-----|


Or othertimes, the cap is shown as being across the entire network;
ie., at the outer extrameties of the inductors. The difference in
practice between the two examples is *large* in effect so which is
correct if either, and equally importantly, how can the skin effect be
incorporated into the model to give a more complete picture?

Same equivalent circuit in the book"Noise Reduction Techniques", but the two
Ls combined as one L. Personally, I would have thought that there should be
an additional inductance in series with R, both in parallel to cap but I
suspect it's ignored because R and C dominate.

 

[Paul Burridge]

Same equivalent circuit in the book"Noise Reduction Techniques", but the two
Ls combined as one L. Personally, I would have thought that there should be
an additional inductance in series with R, both in parallel to cap but I
suspect it's ignored because R and C dominate.

Absolutely. It's evident from the simulations I've run that having a
generous lead-length doesn't seriously degrade impedance anything like
as much as the size of the parasitic capacitance does. The value of
Cpar is, it appears, of much greater consequence that attempting to
achieve zero lead length, at up to a Gig at any rate.

 

[Mantra]

Hi guys,

I've seen a lot of stuff on the net about Spice modelling that takes
into account paractics at RF, but they typically turn out to be....

-----inductor-----resistor------inductor--------
| |
| |
-----cap-----|

Definitive resistor model... Hmmmm. Sorry I have to preach...

In the far back of your mind do remember that all discrete component
models are themselves approximations of Maxwell's equations based on
the assumption of "lumped equivalence". Lumped equivalence says that
if the physical dimensions of your circuit and/or components are far
smaller the shortest relevant signal wavelength, in a Fourier sense,
("far smaller" is usually sufficiently so if the different is 10x
different), then you *can* use lumped equivalents (aka resistors,
capacitors and inductors, et al) to describe the circuit with
"sufficient" accuracy. Otherwise, all bets are off.

As with any approximation, you lose something in order to gain the
expedience of simplicity that linear circuit theory gives you.
Physics and the real world came first, and man only creates an
approximation to emulate what his limited, feeble brain can imagine.
He calls that feeble vision of reality inductors, resistors and
capacitors, and linear circuit theory, in general. Nothing in the
real world is actually lumped or linear.

Thus using a component model is inherently wrong in a absolute sense
(which is why throwing around the word "definitive" is so dangerous),
but being wrong is OK in engineering if you have not cross the line of
where "just wrong" becomes "too wrong". The trick is knowing where
that line is what usually differentiates a good engineer from a bad
engineer. People say engineering is black-and-white: it's not - it's
definitely grey and fuzzy.

Any model for high frequency effects based on discrete components can
*never* be "definitive" in any absolute sense - its "definitiveness"
will always be dependent on how acceptable the inevitable error due to
*not using Maxwell's equations themselves* will be to your particular
case. In other words: how "definitive" is decision of what is
"sufficiently" accurate. Nearly every application has a slightly
different definition what "sufficiently" means so I have no idea what
a "definitive resistor model" really means without have a very
particular specification of accuracy criteria.

For instance, the claim of using "Puff component models up to 50 GHz"
is highly dubious if you are plugging them into SPICE or are ignoring
the implicit and inherent component/circuit size restrictions in that
component model. Things like distributed effects (component values
varying with jw or with physical dimensions like skin effect) are
exactly the parts of simulation that component models fail at first
and badly.



MM

 

[Andrew Tweddle]

For instance, the claim of using "Puff component models up to 50 GHz"
is highly dubious if you are plugging them into SPICE or are ignoring
the implicit and inherent component/circuit size restrictions in that
component model. Things like distributed effects (component values
varying with jw or with physical dimensions like skin effect) are
exactly the parts of simulation that component models fail at first
and badly.



MM

Not to be impolite, but Puff is a linear simulator that takes board
parameters, surface effects, and other high frequency effects into
account. It is a linear S-parameter circuit simulator ,it is not a SPICE
based simulator. So yes it uses roughly 20K lines of code to basically
do Maxwells equations and deals with all the explicit transmission line
effects that you talk about.

Andrew