Hi
Thanks to help from all of you, I now get some pretty decent numbers from
my spice simulation regarding next. Fext is still an order of magnitude
higher in spice. Tnt and si9000 also agrees pretty well.
I'm pretty happy with what I've learned so far, and if I think about the
Fext case a bit more I'll probably figure it out.
Best regards
Johan
Den 4 sep 2015 17:33 skrev "Michael Degerstrom" <mike@xxxxxxxxxx>:
Hi Johan,
Always reset the default CSEG and DSEG from 10 to a much larger value,
say 100. This may fix your asymmetry errors. The default settings were
probably set 25 years ago when computers were much slower.
I remember finding some inconsistencies in the TNT output about 20 years
ago but I don't remember the details. I did work with one of the
developers at that time to get results that seemed to makes sense.
You are doing the right things by correlating simulator output to theory.
If you want to extend your correlation efforts you'd have to build coupled
L/C "ladder" networks and make sure your L/C elements have delay
(=sqrt(L/C)) about 1/10 of your rise time. LTSpice may or may not have a
w-element model but you can learn a lot more building these ladder
networks. For example, you'd build a subckt of one and chain them together
in the main netlist. I may have some old PERL scripts that extract TNT
results and build up spice netlists.
Then you'd have to extend on Istvan's suggestions on termination to
provide near and far end termination networks to terminate all modes.
There is some literature out there on equations for crosstalk with
different termination conditions but I think I've found problems with the
FEXT cases.
Regards
Mike Degerstrom
On Sep 2, 2015, at 2:45 AM, Johan Lans <johan.lans@xxxxxxxxxxxxxxx> wrote:
Hishouldn't
Sorry, I should have mentioned that I scaled down the coupling length to
1mm, just to get a more realistic lumped element. Also, I discovered that
TNT was unhappy about the capacitance matrix not being symmetric, so i
added a 75um dielectric layer above the traces, which fixed the problem.
This means somewhat adjusted values:
These are per meter values:
La = L0(1,1) = 3.3022920e-007 H/m
Lv = L0(2,2) = 3.3022920e-007 H/m
Coupling factor, k, between La and Lv.
k = L0(2,1)/sqrt(L0(1,1)*L0(2,2)) = 0.176
Ca = C0(2,1) + C0(1,1) = 1.12412737e-10 F/m
Cv = C0(2,1) + C0(1,1) = 1.12412737e-10 F/m
Cav = -1*C0(2,1) = 2.5099933e-011 F
Ra = R0(1,1) = 11.4286 Ohm/m
Rv = R0(2,2) = 11.4286 Ohm/m
I didn't know about crosstalk coefficients. Are you using the definitions
on this page for the calculations:
http://www.polarinstruments.com/support/si/AP8164.html ?? Anyway, the
coefficients are obviously normalized to length, so my rescaling
matter.coefficient
Using the updated values above I get (using definition from polar page):
Kb = 0.25*(2.5099933e-011/1.12412737e-10 + 0.176) = 0.10
Kf = (Cm/C-Lm/L) = 2.5099933e-011/1.12412737e-10 - 0.176 = 0.047
Is this in line with your calculations?
Also, I played around with an evaluation of Polar SI9000. With the same
stackup I get a very similiar RGLC-matrices, but it also reports this:
Kb(NEXT) Kf FEXT
8,938337E-02 1,962233E-11 1,962233E-04
At least Kb seems to match...
TNT reports this:
Far-End (Forward) Cross Talk:
FXT(Active Signal, Passive Signal)
FXT( ::RectCond1R1 , ::RectCond1R0 )= 2.30877e-007 = -132.73240 dB
Near-End (Backward) Cross Talk:
BXT(Active Signal, Passive Signal)
BXT( ::RectCond1R1 , ::RectCond1R0 )= 1.18904e-005 = -98.49607 dB
Here is a link to my LTSpice model:
https://dl.dropboxusercontent.com/u/4404053/xtalk/xtalk2.asc
Many thanks for the help
/Johan
2015-09-02 4:07 GMT+02:00 Istvan Novak <istvan.novak@xxxxxxxxxxx>:
Hi Johan,
Are you still using the L and C matrix values from your first posting?
With those numbers you can calculate the Kb backward crosstalk
correct.and Kf forward crosstalk coefficient and they come out 0.0625 and 0.025,
respectively. LTSPICE is pretty accurate if the input models are
transmissionE-mail me offline the LTSPICE deck.
Regards,
Istvan
On 9/1/2015 9:14 AM, Johan Lans wrote:
Hi
Well, yes and no. I have simulated propagation delay on the
modelline (one lump LGCR) in LTSpice with parameters from TNT, and they are
spot
on the same as TNT reports them. This makes me think that the Spice
differentis good. However, TNT reports NEXT and FEXT values to be quite
=-58.5dB,than what I'm getting from LTSpice. TNT FEXT -92.7dB=, TNT NEXT
levels. InLTSpice FEXT = -273dB, LTSpice NEXT = -226dB.
A problem is now that I don't know how TNT calculates crosstalk
thoseLTSpice I'm also getting quite a lot of overshoot and ringing in the
victim, which might indicate a problem with accuracy in LTSice (?).
My next step is to get an evaluation copy of Si9000, from Polar. I read
that it too exports LGCR-parameters, so I can then expirement with
work.:parameters aswell. I'm thinking that the Polar tool is well trusted and
can
be used as a reference to both LTSpice and TNT.
Best Regards, Johan
2015-09-01 14:38 GMT+02:00 Istvan Novak <istvan.novak@xxxxxxxxxxx>:
Hi Johan,
I am glad you found it useful.
Does it work now? Do you get the expected answer?
Regards,
Istvan
On 9/1/2015 8:24 AM, Johan Lans wrote:
Great answer, thank you very much.
Johan
Den 31 aug 2015 04:47 skrev "Istvan Novak" <istvan.novak@xxxxxxxxxxx
Johan,
There are a few things that you will need to change to make this
mayTo get near end and far end crosstalk, you need four ports that you
impedancewant to terminate in its average characteristic impedance: the
geometric
mean of even and odd mode impedances). Though you may not be
interested
in
delay and length, the way how the numbers are supplied, refer to a
particular length, so your circuit will be a four port circuit. You
need
to check the field solver, what unit length it assumes. From the L
and C
numbers you quote, it may be one meter. The characteristic
Rvfrom
these L and C numbers appears to be close to 50 ohms, so the Ra and
addnumbers may represent the losses. If you want a lossless equivalent
circuit, you can replace those numbers with zero, but you have to
single-lumpthe
proper termination at all four ports. Finally, this is a
splitequivalent circuit, because you have one lumped capacitance and one
lumped
inductance for the entire length. As a minimum, you may want to
inputthe
capacitance values into two (half values) and place them at the
fieldand
output instead of just being at the output. if you really want a
wideband
model, you will need more lumps.
Hope this will get you going.
Regards,
Istvan Novak
Oracle
On 8/28/2015 3:36 AM, Johan Lans wrote:
Hi
First of all, condolences to everyone who knew Steve Weir. I
referenced
his
work in my masters thesis in physics a few years ago.
I'm trying to understand simulation of crosstalk by doing some
suite,solving of a trace pair using MMTL (the Multilayer Multiconductor
Transmission Line 2-D and 2.5-D electromagnetic modelling tool
LTSpice.http://mmtl.sourceforge.net/). The solver can output an HSpice
w-element
model of the signal pair, and I'm trying to replicate the crosstalk
levels
reported by the solver by setting up an equivalent circuit in
anHowever, I am not quite sure how to translate all the details into
isequivalent circuit, so if I walk through my thought process doing
this,
maybe someone can point if I'm doing anything wrong?
In the end of this email, I'm pasting the solver output from TNT
1.2.2,
as
well as the w-element output.
The Spice model is a simple RLGC-model. One for the aggressor (I'll
call
the components Ra, Ca and so on)and one for the victim (Rv, Cv ..),
with
an
extra capacitor for coupling between the two circuits (Cav). The
inductors
are coupled using a coupling factor. A link to the LTSpice design
interestedhere:
https://dl.dropboxusercontent.com/u/4404053/xtalk/xtalk.asc
Following Eric Bogatins chapter on crosstalk in "Signal and Power
Integrity
Simplified, 2nd ed", I came up with the values like this: From the
w-element file,
La = L0(1,1) = 3.29962e-007 H
Lv = L0(2,2) = 3.29953e-007 H
Coupling factor, k, between La and Lv.
k = L0(2,1)/sqrt(L0(1,1)*L0(2,2)) = 0.175
Ca = C0(2,1) + C0(1,1) = 1.01942e-10 F
Cv = C0(2,1) + C0(1,1) = 1.01942e-10 F
Cav = -1*C0(2,1) = 1.49773e-011 F
Ra = R0(1,1) = 11.4286 Ohm
Rv = R0(2,2) = 11.4286 Ohm
G0 and Gd is zero, and I omitted Rs , since I'm not really
I'min
losses right now.
TNT calculates fare end crosstalk to and near end crosstalk to .
injectednot
sure how to translate this to measurements in LTSpice, but I
signala
1V,
100ns, step into the aggressors resistor and measured how much
difficultwas
crosstalked over at the Cav end of the victim circuit (a bit
LTSpice:to
explain without schematic, but here's a screen capture from
actualhttps://dl.dropboxusercontent.com/u/4404053/xtalk/xtalk.png) The
crosstalked signal has an amplitude of about 18mV, which is about
-55dB. I
interpret this as being near end crosstalk, since there is no
-41dB,length
in the model. The field solver reports near end crosstalk to be
https://dl.dropboxusercontent.com/u/4404053/xtalk/xtalk.hspice-w.rlgcbut
far end crosstalk to be -57dB.
Am I doing anything right here?
Thanks in advance from
Johan
As reference, the field solver files are here:
https://dl.dropboxusercontent.com/u/4404053/xtalk/xtalk.xsctn
https://dl.dropboxusercontent.com/u/4404053/xtalk/xtalk.result
MHz.
2015 08 28 09:12:31 johan.lans NMMTL_2DLF
File = C:/Program Files (x86)/tnt-1.2.2/examples/xtalk
Number of Signal Lines = 2
Number of Ground Planes = 1
Number of Ground Wires = 0
Coupling Length = 1.00000 meters
Rise Time = 100000.0012 picoseconds
Contour (conductor) segments [cseg] = 10
Ground Plane/Dielectric segments [dseg] = 10
Conductivity RectCond1R1 = 5e+007 siemens/meter
Conductivity RectCond1R0 = 5e+007 siemens/meter
Note: minimum frequency for surface current assumptions is 1655
fewMutual and Self Electrostatic Induction:
B(Active Signal , Passive Signal) Farads/Meter
B( ::RectCond1R1 , ::RectCond1R1 )= 1.1691868e-010
B( ::RectCond1R1 , ::RectCond1R0 )= -1.6413763e-011
B( ::RectCond1R0 , ::RectCond1R1 )= -1.4977254e-011
B( ::RectCond1R0 , ::RectCond1R0 )= 1.1813248e-010
Mutual and Self Inductance:
L(Active Signal , Passive Signal) Henrys/Meter
L( ::RectCond1R1 , ::RectCond1R1 )= 3.2996223e-007
L( ::RectCond1R1 , ::RectCond1R0 )= 5.7727959e-008
L( ::RectCond1R0 , ::RectCond1R1 )= 5.7747599e-008
L( ::RectCond1R0 , ::RectCond1R0 )= 3.2995263e-007
Asymmetry Ratios:
Asymmetry ratio for inductance matrix:
0.034021% (max), 0.034021% (average)
**********
Asymmetry ratio for electrostatic induction matrix:
8.751856% (max), 8.751856% (average).
(Note values greater than 1% are a probable indication of too
dBelements.
Try adjusting CSEG and DSEG attributes.)
**********
Characteristic Impedance (Ohms):
For Signal Line ::RectCond1R1= 53.1239
For Signal Line ::RectCond1R0= 52.8495
Characteristic Impedance Odd/Even (Ohms):
odd= 45.186
even= 62.1082
Effective Dielectric Constant:
For Signal Line ::RectCond1R1= 3.36577
For Signal Line ::RectCond1R0= 3.40061
Propagation Velocity (meters/second):
For Signal Line ::RectCond1R1= 1.6340992e+008
For Signal Line ::RectCond1R0= 1.6257061e+008
Propagation Velocity Odd/Even (meters/second):
odd= 1.65982e+008
even= 1.60201e+008
Propagation Delay (seconds/meter):
For Signal Line ::RectCond1R1= 6.1195795e-009
For Signal Line ::RectCond1R0= 6.1511734e-009
Propagation Delay Odd/Even (seconds/meter):
odd= 6.02475e-009
even= 6.24218e-009
Rdc:
Rdc(Active Signal , Passive Signal) Ohms/Meter
Rdc( ::RectCond1R1 , ::RectCond1R1 )= 1.1428571e+001
Rdc( ::RectCond1R1 , ::RectCond1R0 )= 0.0000000e+000
Rdc( ::RectCond1R0 , ::RectCond1R1 )= 0.0000000e+000
Rdc( ::RectCond1R0 , ::RectCond1R0 )= 1.1428571e+001
Far-End (Forward) Cross Talk:
FXT(Active Signal, Passive Signal)
FXT( ::RectCond1R1 , ::RectCond1R0 )= -1.48130e-003 = -56.58711
field
Near-End (Backward) Cross Talk:
BXT(Active Signal, Passive Signal)
BXT( ::RectCond1R1 , ::RectCond1R0 )= 9.25449e-003 = -40.67295 dB
NOTE: Cross talk results assume there are no reflections.
*--------------------------------------------
*
* RLCG parameters for W-Element
* frequency-dependent transmission line
*
*
* MMTL File Name: C:/Program Files
(x86)/tnt-1.2.2/examples/xtalk.result
* Fri Aug 28 09:12:55 W. Europe Daylight Time 2015
*
*--------------------------------------------
*
* | MMTL Simulation data from
* | C:/Program Files (x86)/tnt-1.2.2/examples/xtalk.result
* |
* | 2015 08 28 09:12:31 johan.lans NMMTL_2DLF
* |
* | File = C:/Program Files (x86)/tnt-1.2.2/examples/xtalk
* | Number of Signal Lines = 2
* | Number of Ground Planes = 1
* | Number of Ground Wires = 0
* | Coupling Length = 1.00000 meters
* | Rise Time = 100000.0012 picoseconds
* | Contour (conductor) segments [cseg] = 10
* | Ground Plane/Dielectric segments [dseg] = 10
* | Conductivity RectCond1R1 = 5e+007 siemens/meter
* | Conductivity RectCond1R0 = 5e+007 siemens/meter
* | Note: minimum frequency for surface current assumptions is 1655
MHz.
* |
* | Signal Names (in order):
* | ::RectCond1R1
* | ::RectCond1R0
*
*--------------------------------------------
*
* N = number of signal conductors
*
*--------------------------------------------
2
*--------------------------------------------
*
* L0
*
*--------------------------------------------
3.29962e-007
5.77476e-008 3.29953e-007
*--------------------------------------------
*
* C0
*
*--------------------------------------------
1.16919e-010
-1.49773e-011 1.18132e-010
*--------------------------------------------
*
* R0
*
*--------------------------------------------
11.4286
0 11.4286
*--------------------------------------------
*
* G0
*
*--------------------------------------------
0
0 0
*--------------------------------------------
*
* Rs
*
*--------------------------------------------
0.00119571
0 0.00119571
*--------------------------------------------
*
* Gd
*
*--------------------------------------------
0
0 0
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject
fieldTo unsubscribe from si-list:
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 forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
------------------------------------------------------------------
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject
------------------------------------------------------------------
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 forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
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 forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
------------------------------------------------------------------
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 forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu