2013 belforte caniggia_cst_coax_cable_si_ final_240312
DESCRIPTION
A comparative study on modeling techniques of coaxial cables.TRANSCRIPT
1
CST coaxial cable models for
SI simulations: a comparative
study
Piero Belforte, Spartaco Caniggia
March 24th 2013
2
Outline
• Introduction
• S parameters in frequency domain
• S parameters in time domain
• Comparison between measurements and
simulations
• Ultra Wide Band (UWB) source
• Proposal for efficient and accurate simulation of
lossy cables
• Conclusion
3
Introduction
• The task of this report is to show that some important
signal integrity (SI) problems arise when Cable Studio
(CS) is used to simulate high-speed digital signal
transmission with lossy lines (cables or traces in PCB)
[1]
• An 1.83-m RG58 coaxial cable is modeled by CS and
commercial programs: MC10 [2] and DWS , based on
Digital Wave Network equivalent of the electrical network
[3].
• Simulations are compared with measurements
• It is shown that CS doesn’t provide good results
• A method is proposed to solve the SI problems with
CST Cable and Design studio.
4
S parameters in frequency domain
5
S parameter computation
• Cable: RG58
• Length: 5cm
• Frequency range: 0-10GHz
• Characteristic Impedance Z0: 49.94Ω
• Only ohmic losses are taken into account because dielectric losses with tanδ=0.0002 at 100MHz (Polyethylene) doesn’t give significant changes.
• SPICE simulation performed by MicroCap10 (MC10) because of good TL models [2]
• DWS (Digital Wave Simulator) analysis because of speed (50X MC10), accuracy and time-domain scattering parameters.
• Comparison among CST Cable Studio, CST MWS, MC10 and DWS (Digital Wave Simulator)
6
Equivalent circuit used by MC10 (SPICE) for theoretic S11
& S21 computation (analytic approach)
rs
ts
εr
2rw
Coaxial cable
geometry
50Ω
50Ω
For details, see [1, clause 11.2.3]
5-cm RG58: Z0=49.94Ω
File:S_LOSSYTL_ANALYTICAL_10GHZ.CIR (MC10)
Permittivity=2.3, Loss
angle tanδ=0.0002
Insulator outside: thickness=0.5mm,
permittivity=3, Loss angle tanδ=0.02 Solid shield screen type
7
CST cable studio for S11 & S21
computation
Equivalent circuit to compute S
parameters by CST DESIGN
STUDIO
50 Ω 50 Ω
RG58: length=5 cm,
Z0=49.94Ω
File: Ex_coax_S_5cm.cst
8
3D RG58 model by MWS
Meshcells=41,515 Waveguide port
Time domain solver: adaptive mesh refinement was used
9
S11
• S11 computed by Cable Studio 2010 &
2012 provide the same results
• S11 computed by MWS and MC10 provide
similar results and about some dB lower
•Level differences are due to impedance
mismatching
• Resonance frequencies are slightly higher
for MWS (lower cable delay)
Cable Studio (CS) 2012
Cable Studio (CS) 2010
MWS Studio 2012
MC10 2012
10
S21
Cable Studio (CS) 2012
Cable Studio (CS) 2010
MWS Studio 2012
MC10 2012,DWS 8.4
RL-TL model
• S21 computed by MC10 is the lowest curve
(more losses)
• S21 computed by CST 2012 is too higher
than CST 2010 (less losses)
• S21 computed by MWS is in the middle
between MC10/DWS and Cable studio 2012
and close to cable studio 2010
MWS
MC10,DWS
CS
11
Comments on computation of S
parameters
• S11 computed by MWS and MC10/DWS provide similar values both in time domain and frequency domain
• S11 computed by Cable Studio 2010 & 2012 are about 15dB higher than MWS and MC10/DWS due to characteristic impedance mismatching
• S21 computed by Cable Studio 2012 provides much less losses than those computed by Cable Studio 2010
• S21 computed by Cable Studio 2010 is close to MWS
• CST should investigate the last two items
12
S parameters in time domain
13
Lossy line matched at both ends
Typical source and load voltage waveforms for an interconnect matched
at both ends: lossless TL (dashed line), frequency-dependent lossy TL
(solid line) [1, Fig.7.3]
When TL has characteristic impedance different from the loads, distortions occur
Definitions of S
parameters in time
domain:
•S11=VS-1
•S21=VL
14
Voltage computations in time
domain
• Cable: RG58
• Length: 1.83m
• Line terminations: 50Ω
• Source: step waveform with rise time tr=0.1ns
• Frequency range: 0-10GHz
• Characteristic Impedance Z0: 49.94Ω
• SPICE simulation performed by MC10 [2]
• DWS simulations performed by DWS 8.4 [4]
• Comparison between CST & SPICE results
• DWS results are the same of MC10
15
Coaxial cable structure
50 Ω
50 Ω
Z0=49.94 Ω Length:1.83m
V1 V2
Vsource=2 V
trise=0.1 ns
Ramp
Source
16
Circuit and model used in MC10 and DWS (RL-TL
approach)
Coaxial cable matched at both ends and modeled as a
cascade of 610 3-mm RL-TL cells including the skin effect,]
V1=VS V2=VL
Step
signal
Remark: the cascade of RL-TL cells provides the same S11 and S21 in
frequency domain computed by the analytic approach used in the previous
section, see Fig.7.22 of [1]
RL-TL model: RL parameters
were computed by vector
fitting technique starting from
analytic expressions for ohmic
losses, see [1, clause 7.2.1.3]
17
Circuit and cable model used in CST
RG58 model with
length 1.83 m
Skin effect only
10GHz
Vinit: 0.0
Vpulse: 2.0
Tdelay: 1e-9
Trise: 0.1e-9
Thold: 100e-9
Tfall: 0.1e-9
Ttotal: 200e-9
File: Ex_coax_S_1_83_10GHz.cst
18
Voltages V1 & V2 (cst 2010)
MC10 (SPICE) CST
V1 V2
V1 V2
V1 V2
V1 V2
ns ns
ns ns
Samples 1001 in
transient1 task
Samples 5001 in
transient1 task
? ?
MC10 and CS have the same losses except the oscillations provided by
CST 2010 that should not occur
19
Voltages V1 & V2 (cst 2012)
MC10 (SPICE) CST
• CST cable studio 2012 provides
less losses than MC10 and CS 2010,
as evidenced by frequency
computation of S parameters.
• Oscillations remain
• Using normal or very high accuracy
the results do not change
20
With 1-GHz model computed by CST 2012
Oscillations
increase!
21
DWS 37-cell model vs CST MWS: S11
•It can be noted that MWS
computes about half
losses than DWS.
•S11 of MWS was
obtained calculating the
integral of the reflected
wave (o1,1) as response
to a step source.
DWS
MWS
22
Comments on computation of V voltages
• V1: the voltages at source end computed by MC10 (SPICE)/DWS and CST 2010 are in good agreement.
• V2: the voltages at load end computed by SPICE/DWS and CST 2010 are in good agreement except for the oscillations in CST waveform.
• V1 and V2 computed by CST CS 2012 are not in agreement with MC10/DWS, less losses are computed by CST 2012 and unrealistic oscillations on V2 remain.
• CST should investigate these two last items
• Time domain S11 from CST MWS is lower (about half) of that from RL-TL model simulated with DWS as already noticed in return loss vs frquency
23
Comparison between
measurements and simulations
24
Comparison between
measurements and simulations
The measurements performed on 1.83-m RG58 cables are compared with three simulation methods:
1. CST cable studio.
2. MC10, based on SPICE [2] and using a cascade of 610 3-mm RT-TL unit cells.
3. DWS models using both 366 X 5mm RL-TL chain of cells and a 3660 X .5mm RL-TL chain inserted in actual CSA803 measurement setup.
25
50-Ω RG58 model with length
1.83 m (very high accuracy,
ohmic losses in CS)
Vinit: 0.0
Vpulse: 2.0
Tdelay: 1e-9
Trise: 0.1e-9
Thold: 100e-9
Tfall: 0.1e-9
Ttotal: 200e-9
CST model (Step source)
Open
•V1 (or VP1) voltage at the input of the cable was computed and measured
•Dielectric losses are neglected for SPICE (MC10) and CS (Cable Studio 2012)
26
DWS (4) cable cell on Spicy SWAN (5)
(Due to DWS sim speed, even a .5mm cell has been tried)
27
Example of Spicy SWAN (DWS) circuit for S-parameter
cable characterization using a chain of cells
(Due to DWS sim speed, even a chain of 3660 X .5mm RL TL cells has been
utilized, getting practically the same results of the 366X 5mm cell model)
28
RG58 CU (TASKER) specs
29
Measurement set-up (CSA803)
30
Measurements with cable open at far-end voltage
The measurements were
performed by Piero Belforte
on two commercial 1.83-m
RG58 cables: Tasker and
GBC.
Comparison of the
reflected edge of the
two cables: very little
differences.
V1
V1
ns
ns
-1
0
1.2
0
1.2
4
Reflected edge
Tasker GBC
50
31
VP1:voltage at cable input
V1
ns
V1
ns
CST 2012
Measurement
MC10
DWS (including TDR
setup)
32
• There is good agreement on reflected edge among RL-TL
model using both MC10 and DWS simulators (DWS is
50X faster than MC10) and measurements. Note that
dielectric losses were neglected in the RL_TL model and
actual cables have stranded conductors (not solid)
• CS reflected edge is affected by not acceptable
oscillations
VP1 voltage details
V1
ns
CST 2012
Measurement
MC10
DWS
33
S-parameters measurements and comparison with
366 RL_TL model in the actual setup (DWS)
S21 S21
S11 S11
34
Actual S-parameter measurements:
considerations
• Actual cable (stranded conductors) shows significant
distributed impedance discontinuities
• S11(S22) in time domain shows larger values than
model
• Actual S11 and S22 are not identical (not symmetrical)
due to impedance discontinuities
• S21(S12) edge is slightly slower from 0 to 50% due
probably to dielectric losses
• S21(S12) edge is slightly faster from 50% to 100% due
probably to stranded conductors (lower skin effect losses
at high frequency)
35
DWS BTM (Behavioral Time Model) of
1.83m cable using Spicy SWAN
366 cells
of RL-TL
1 cells
S from
measurements
BTM
RL-TL
50 ns
12 ns 50 ns
1 V
1 V 0.035
BTM
RL-TL
36
Comments on measurements and
simulations
• MC10 (SPICE) and DWS open cable and S21 are in good agreement with measurements despite the stranded (not solid) conductors of actual cable.
• S11 of measurements takes into account slight distributed impedance mismatching along the cable therefore more accurate models should be needed for a high level of accuracy.
• Dielectric losses are much less important than ohmic losses and can be neglected for most applications
• CST cable studio provides not realistic oscillations (distorted waveforms) as verified by measurements
37
Ultra Wide Band (UWB) source
38
Coaxial cable with source an UWB
signal
• The same coaxial cable of previous
example was tested by using as a source
an ultra wide band (UWB) signal instead of
a step waveform.
• The signal is introduced into design studio
as imported file.
39
MC9 model (UWB source)
Coaxial cable matched at both ends and modeled as a cascade of 610 cells
including the skin effect: comparison between measured (dashed line) and
computed (solid line) waveforms [1, chapter7]
Validation
Model
40
CST model (UWB source)
File: Ex_coax_UWB.cst
Imported file:
New_uwb_input_by2.txt
Ohmic losses
RG58
41
Comments on coaxial cable with
UWB source
• SPICE (MC10) runs in some minutes and
gives waveform on 50-Ω load in good
agreement with measurement
• CST runs with very long time and the
simulation was aborted.
42
Proposal for Signal Integrity of
lossy cable
43
Method • Define the cable by its geometrical and electrical parameters
• Choose between two unit-cell models: 1. RL-TL: the unit cell should be electrically short for the frequencies of
interest. It is modeled as a network of resistances and inductances to take into account the ohmic and electric losses (analytic expression in frequency domain) computed by vector fitting technique in series with an ideal transmission line (TL) as reported in chapter 7 of [1]. Simulator: SPICE with good TL model [2], DWS (50X faster) [3].
2. S-parameter: the unit cell should have a length to satisfy the rule that the rise-time excitation should be less than 1/10 the unit-cell delay. It is modeled by using S-parameters in time domain (2D or 3D computation) as defined in [1,3]. Simulator: DWS only [3]
• Model the line by a cascade of unit cells.
• Perform simulations in time domain by using SPICE [2] or DWS (more accurate and 50X faster) [3] to get the voltage or current waveforms.
Remark: the method can also be used for interconnections in PCB such
as microstrip and stripline traces
44
Flow chart
Define the cable
RL-TL Model
(SPICE, DWS) S-parameter Model (DWS)
Cascade of unit cells
Results obtained by SPICE or DWS
time domain simulations
Which
solution ?
2D/3D S-parameter
computation
Vector fitting
to set RL
network
Define an unit-cell cable
S1
1 S2
1
TL RL
RL
unit cell unit cell
45
Conclusion • The 2D (TL) modeling in CST CABLE STUDIO should be revised
because it provides unexpected oscillations on signals when the source is a step waveform.
• CST Cable Studio 2012 provides less losses than CST 2010.
• CST Cable Studio results are not in agreement with MWS, SPICE and DWS simulations and measurements.
• There are instability problems in CST when the source is an ultra wide band signal imported as external file.
• We suggest to use the method presented at the end of this document that consists of a cascade of unit-cable cells simulated by SPICE or DWS (50X faster).
• DWS supports fast simulations of both time domain s-parameter and RL-TL chain of cells.
• BTM (Behavioral Time Model) method supported by DWS is the fastest and most accurate if unit-cell S-parameters are taken from actual measurements.
46
References
[1] S. Caniggia, Francesca Maradei, “Signal Integrity and Radiated
Emission”, John Wiley & Sons, 2008
[2] www.spectrum-soft.com
[3] P.Belforte “Time domain simulation of lossy interconnections using
wave digital networks” ISCAS 1982 Rome
[4] DWS (Digital Wave Simulator) user manual
http://www.slideshare.net/PieroBelforte1/dws-84-
manualfinal27012013
[5 ] Spicy SWAN : www.ischematics.com
http://www.slideshare.net/PieroBelforte1/spicy-swan-concepts-
16663767
[6] DWS and SWAN, ( Simulation by Wave Analysis) are trademarks of
Piero Belforte http://www.linkedin.com/in/pierobelforte