Using/checking models produced by the LMG tool

S

spectrallypure

Guest
Hello all!

I am trying to use the LMG tool to model a simple 100ohm, 30cm FR4 PCB
microstrip line, but I fear am obtaining strange results. For example,
using the tool with (exactly) the following entries...

Model type: Lossy, wideband
Transmission line type: SubLossLine
No. Layers: 2, No. Lines: 1, No. GndPlates: 1
Dielectric constant (er): 4.1 1
Dielectric thickness (d): 800 um
Dielectric loss type: tan=sigma/(w*ep0)
Dielectric loss tangent: 0.018 0.0
Conductor width: 843 um
Conductor thickness: 35 um
Conductor height (h): 800 um
Ground plate thickness: 20 um
Ground plate conductivity: 5.6e7 S/m
Signal line conductivity: 5.6e7 S/m
Conductor length: 30000 um
Fmax: 20 GHz

....I generate a file "my_model.scs" which I later link, in a testbench
schematic, to an instance of the "mtline" component in analoglib
(using the option "use lmg subckt" on the componet's CDF). The
testbench contains just an "vsin" generator feeding the mtline at the
near-end, a termination resistor of 100ohm at the far-end, and a
ground pin. When I perform an AC sweep of this testbench up to 20GHz
and plot the effective impedance seen at the input port of the
transmission line and the attenuation at the far end, I seem to get
strange results:

1. The effective input impedance of the line increases greatly with
frequency (almost linearly after 1GHz), and rises to very high values.
For instance, it reads 763ohm at 10GHz and 1531ohm at 20GHz!

2. The attenuation seems to be too high at high frequencies. For
instances, the -3dB point is 1.318GHz and at 10GHz it is -17.6dB! It
approximately falls with -20dB/dec after 2GHz.

What further makes me unsure about these results is the fact that when
press the button "Generate Macromodel" in LMG, it outputs a message in
the console that reads:

"For substrate loss case, widebandwidth model generation may not be
accurate, you can use mtline to check it".

I really need a wideband model, because I intend to model the
transmission of pulses along this line and not just pure sinusoids.
According to the documentation, the model should be valid from DC to
the specified frequency, 20GHz in this example. I am also clueless
about how to use the .dat LRGC file that is also generated. What are
the differences between using the .scs model and the LRGC data file?
When should I use each one? The examples included in the documentation
seem outdated and talk about components which doesn''t seem to exist
anymore in the analoglib, like "tline3" and "mline" :S

I would be really grateful if anybody could comment on the correctness
of these results, indicate the correct usage of the .scs models and/or
LRGC datafiles, or point out any other resource for transmission line
modeling that I could use for cross-checking the models produced by
the LMG tool.

Thanks in advance for any help! ...and sorry for the long post! :)

Regards,

Jorge.
 
As far as I know FR4 is not a suitable material above 2-3GHz. Its
dielectric loss is high at high frequencies. Instead you should look
at Rogers material. Between 10GHz-20GHz skin effect is also a big
concern. This can cause high losses as you obtained from LMG model.
Below reference can be helpful.

http://www.speedingedge.com/PDF-Files/Materials_RickH2.pdf

Ozgur


spectrallypure wrote:
Hello all!

I am trying to use the LMG tool to model a simple 100ohm, 30cm FR4 PCB
microstrip line, but I fear am obtaining strange results. For example,
using the tool with (exactly) the following entries...

Model type: Lossy, wideband
Transmission line type: SubLossLine
No. Layers: 2, No. Lines: 1, No. GndPlates: 1
Dielectric constant (er): 4.1 1
Dielectric thickness (d): 800 um
Dielectric loss type: tan=sigma/(w*ep0)
Dielectric loss tangent: 0.018 0.0
Conductor width: 843 um
Conductor thickness: 35 um
Conductor height (h): 800 um
Ground plate thickness: 20 um
Ground plate conductivity: 5.6e7 S/m
Signal line conductivity: 5.6e7 S/m
Conductor length: 30000 um
Fmax: 20 GHz

...I generate a file "my_model.scs" which I later link, in a testbench
schematic, to an instance of the "mtline" component in analoglib
(using the option "use lmg subckt" on the componet's CDF). The
testbench contains just an "vsin" generator feeding the mtline at the
near-end, a termination resistor of 100ohm at the far-end, and a
ground pin. When I perform an AC sweep of this testbench up to 20GHz
and plot the effective impedance seen at the input port of the
transmission line and the attenuation at the far end, I seem to get
strange results:

1. The effective input impedance of the line increases greatly with
frequency (almost linearly after 1GHz), and rises to very high values.
For instance, it reads 763ohm at 10GHz and 1531ohm at 20GHz!

2. The attenuation seems to be too high at high frequencies. For
instances, the -3dB point is 1.318GHz and at 10GHz it is -17.6dB! It
approximately falls with -20dB/dec after 2GHz.

What further makes me unsure about these results is the fact that when
press the button "Generate Macromodel" in LMG, it outputs a message in
the console that reads:

"For substrate loss case, widebandwidth model generation may not be
accurate, you can use mtline to check it".

I really need a wideband model, because I intend to model the
transmission of pulses along this line and not just pure sinusoids.
According to the documentation, the model should be valid from DC to
the specified frequency, 20GHz in this example. I am also clueless
about how to use the .dat LRGC file that is also generated. What are
the differences between using the .scs model and the LRGC data file?
When should I use each one? The examples included in the documentation
seem outdated and talk about components which doesn''t seem to exist
anymore in the analoglib, like "tline3" and "mline" :S

I would be really grateful if anybody could comment on the correctness
of these results, indicate the correct usage of the .scs models and/or
LRGC datafiles, or point out any other resource for transmission line
modeling that I could use for cross-checking the models produced by
the LMG tool.

Thanks in advance for any help! ...and sorry for the long post! :)

Regards,

Jorge.
Click to expand...
 
Ozgur, thanks so much for the paper, it is very instructive.

I am interested in FR4 because the spectrum of the pulses I want to
transmit is negligible after 2-3GHz, and thus this material should
suffice. Since I wanted to model the channel in the most realistic
way, I chose to model up to 20GHz, but now I see it is way too much!

What still puzzles me though about my Cadence simulations is the great
increase in the effective impedance of the line. A colleague told me
that if the characteristic impedance of the line is 100ohm and in my
testbench I terminate it with 100ohm, then I should always get an
effective input impedance of 100ohm, irrespective of the frequency. He
told me that the characteristic impedance is only a function of the
geometry of the line, and doesn't vary with frequency....

....but then I argued that Zo = sqrt[(R(w)+jwL)/(G(w)+jwC)] (w=omega,
angular freq.) were R(w) and G(w) are related to the skin effect and
the dieletric losses, respectively, and these quantities are frequency
dependent and behave differently with increasing w. Thus, when
sweeping w the quotient [(R(w)+jwL)/(G(w)+jwC)] changes and Zo should
vary!.... right?

I would be grateful if someone could please spare a bit of light into
this unfortunately still obscure topic for me.

Sorry again for being 90% off-topic! :)

Regards,

Jorge.
 
spectrallypure wrote, on 06/11/08 16:40:
Ozgur, thanks so much for the paper, it is very instructive.

I am interested in FR4 because the spectrum of the pulses I want to
transmit is negligible after 2-3GHz, and thus this material should
suffice. Since I wanted to model the channel in the most realistic
way, I chose to model up to 20GHz, but now I see it is way too much!

What still puzzles me though about my Cadence simulations is the great
increase in the effective impedance of the line. A colleague told me
that if the characteristic impedance of the line is 100ohm and in my
testbench I terminate it with 100ohm, then I should always get an
effective input impedance of 100ohm, irrespective of the frequency. He
told me that the characteristic impedance is only a function of the
geometry of the line, and doesn't vary with frequency....

...but then I argued that Zo = sqrt[(R(w)+jwL)/(G(w)+jwC)] (w=omega,
angular freq.) were R(w) and G(w) are related to the skin effect and
the dieletric losses, respectively, and these quantities are frequency
dependent and behave differently with increasing w. Thus, when
sweeping w the quotient [(R(w)+jwL)/(G(w)+jwC)] changes and Zo should
vary!.... right?

I would be grateful if someone could please spare a bit of light into
this unfortunately still obscure topic for me.

Sorry again for being 90% off-topic! :)

Regards,

Jorge.
Click to expand...
Only a quick answer, because I'm a bit busy to go into the details.

1. "lmg" is really EOL'd - you should instead directly enter the physical
parameters on the instance of "mtline" instead. With recent versions of
MMSIM, the solver is built into the simulator, and so all fixes and
enhancements have gone into the solver in the simulator, not that in
the standalone "lmg" utility. We are restoring the capability to view
the physical parameters from the mtline component.
That said, you should still get broadly reasonable results from "lmg",
but just wanted to point out the "supported" way of doing it.
2. You're better off using RLGC matrices rather than the ".scs" file approach.
The ".scs" file builds a macro-model, and is much harder to make
wideband, and also is for a fixed length. The RLGC matrices are a
per-unit-length set of RLGC matrices, and are handled more directly
by the simulator. If you're simulating in the frequency domain, they can
be used directly; if in the time domain, the simulator builds an appropriate
time domain model. This is effectively what is used when you enter
the physical dimensions on the mtline component - it will build RLGC matrices
within the solver in the simulator, and then simulate using those.

If using an older version, where you only have "lmg" - you can tell the
mtline to use a file of RLGC matrices - you then have to tell the mtline
instance how long your line is.

Hope this slightly vague answer helps a bit!

Andrew.
 
Hi Andrew, thanks so much for your reply. Just a couple of questions:

-On which version of IC the mtline component (in analoglib, right?)
allows to input the physical parameters of the line? I am using
IC5.10.41 and the mtline component only supports the RLGC datafile and
lmg subckt options (besides entering RLGC data manually).

-I tried linking the RLGC datafile to my mtline component, but the
simulation colapsed. I noticed that this happened only with the RLGC
datafiles generated using the "wideband" option in lmg, which seems to
generate RLGC values for a discrete set of frequencies. However, the
"narrowband" models generated with lmg seem to work fine
(unfortunately they are of no use to me since I need a model from DC
to a few GHz, for time domain simulations of non-sinusoidal pulse
transmission). I noticed that for the same frequency, the R and L
values in both wideband and narrowband RLGC datafiles are the same,
but the G and C are different and in the wideband model they get
absurdly small values (like 2.270474e-307!). Do you think there is
any chance of generating manually some sort of wideband RLGC datafile
from the independent, narrowband (i.e., single-frequency) RLGC
datafiles generated by lmg?

Thanks again for any ideas.

Regards,

Jorge.
 
spectrallypure wrote, on 06/13/08 15:08:
Hi Andrew, thanks so much for your reply. Just a couple of questions:

-On which version of IC the mtline component (in analoglib, right?)
allows to input the physical parameters of the line? I am using
IC5.10.41 and the mtline component only supports the RLGC datafile and
lmg subckt options (besides entering RLGC data manually).

-I tried linking the RLGC datafile to my mtline component, but the
simulation colapsed. I noticed that this happened only with the RLGC
datafiles generated using the "wideband" option in lmg, which seems to
generate RLGC values for a discrete set of frequencies. However, the
"narrowband" models generated with lmg seem to work fine
(unfortunately they are of no use to me since I need a model from DC
to a few GHz, for time domain simulations of non-sinusoidal pulse
transmission). I noticed that for the same frequency, the R and L
values in both wideband and narrowband RLGC datafiles are the same,
but the G and C are different and in the wideband model they get
absurdly small values (like 2.270474e-307!). Do you think there is
any chance of generating manually some sort of wideband RLGC datafile
from the independent, narrowband (i.e., single-frequency) RLGC
datafiles generated by lmg?

Thanks again for any ideas.

Regards,

Jorge.
Click to expand...
I did a quick check, and it appears to have been added in IC5141 USR2 - there's
an option "field solver" on the form. You'd probably need to be using spectre
from an MMSIM version.

Something recent would make sense.

I don't think trying to generate a wideband model from a narrowband RLGC file
makes any sense - that's trying to extrapolate from a narrow model, as opposed
to producing a wideband model with a solver. It's hard enough trying to build a
time domain model at the best of times, but doing it from extrapolated data is
asking for trouble.
 
OK Andrew; thanks a lot for the information & advice, as usual.

Regards,

Jorge.
 
You must log in or register to reply here.

Welcome to EDABoard.com

Sponsor

Back
Top