Woorkflow of topolo
ogy simuulation in
n automottive com
mmunicatiion
fo
focused on
o frequeency depeendent caable mod
delling prrocedures.
JJürgen Minutth
Esslingen University
U
of A
Applied Sciencces, Göppingeen, Germany
e-mail: juerggen.minuth@hs-esslingen.dee
Katharinaa Feldhues, Sttephan Frei
TU
U Dortmund U
University, Dorrtmund, Germ
many
e--mail: kathariina.feldhues@
@tu-dortmund.de
ABSTRA
ACT
The transsmission of auutomotive com
mmunication- signals in mass production
n is based maiinly on electriical cables
used in toopologies whhich do not meeet technical schoolbook requirements perfectly.
p
Forr evaluating th
heir signal
quality annd their comm
munication ro
obustness timee-domain based simulation
n tools are esssential. One of
o the key
elements hereby is an adjustable caable model. U
Unfortunately real
r cables sh
how frequencyy-dependent behaviours
b
which aree not covered by time-dom
main simulationn tools usually
y. The paper points
p
out som
me summariziing results
of variouus projects which enablee a workflo w beginning
g with (noisy
y) cable meeasurements going
g
via
approxim
mation proceduures up to freq
quency and tim
me domain qu
ualified models.
Keyword
ds: twisted-pair cable, cablee parameter
1. INTRO
ODUCTION
Modern ccars use varioous communiication system
ms in parallell; each of theem shall be im
implemented functional
robust annd cost-efficiennt. During thee early developpment phase several
s
architeecture and toppology variantts compete
against eaach other. Suppporting usagee of simulatioon tools is esseential: e.g. forr getting the syystem costs, the
t overall
weight orr the signal quuality. The pap
per is focused on two aspeccts:
 Behhaviour and modelling of caables as one off the key elem
ments influenciing the comm
munication properties.
 Worrk-flow to inteegrate cable-p
properties intoo the simulatio
on.
The succeeeding exampple shows a top
pology consissting of five CAN-nodes
C
communicating among each other.
o
Figure 1: Top
pology simulaation with fivee CAN nodes at
a 2 Mbit/sec1,
the sim
mulation resullts are superpo
osed to an eyee chart.
perties of the ssignals are deetermined and voted againstt limits:
For topollogy evaluatioon several prop
 Eye chart got by the
t superposittion of all bitss of a selected data stream.
d slew-rate.
 Risiing and fallingg edge timingss like propagaation delay and
 Durration of ringinng.
 etc.
r
the availability
a
off models like transceivers (CAN) or buus-drivers (FlexRay™),
A properr simulation requires
common mode coils, ESD
E
protection
n elements, feerrite beads, teermination circcuits and last but not least cables:
c
me-domain
 A siingle cable-moodel shall guaarantee a practtically perfectt matching bettween measureement and tim
simuulation from <<10
<
kbit/sec up to >10 Mbbit/sec at leastt.
1
supporteed by “CAN flexible
f
data rate”
24 April 20014
ICTON 2014
1-7
Workflow: Cable Modelling and Simulation
ICTON 2014
 The propagation delay of edges shall be covered.
 The damping shall be covered (steady state and edges’ slew-rate).
 Reflections and ringing shall be shown.
The standard cable-model available in most SPICE-compliant tools offers constant losses (constant for all
frequencies). Therefor two main behaviours are covered (simulation and measurement fit):
 Edges’ propagation delay.
 The steady state level of “long bits”.
Further essential properties are not covered, in other words simulation does not fit with measurement:
 Ringing amplitudes and ringing duration.
 Slew rate modifications (when a pulse travels along a cable).
The required new extended cable-model has to meet several requirements, so that simulation and measurement
match practically perfect from engineer’s point of view.
 Common mode and differential mode (because twisted pair cables are used).
 Impedances and delays.
 Frequency dependent losses (which influence slew-rate, damping etc.).
The development of the workflow to get satisfying cable-models was done in two main approaches: a 1st
pragmatic approach (less successful) where measured S-parameter were directly used for time domain simulation
and a 2nd more successful approach: a generic physical cable-model is parameterized according measurements; a
succeeding mathematical approximation enables a time domain simulation.
2. MODELLING OF CABLES
2.1 Measurement and Simulation Results (1st Approach)
When thinking about how to enable frequency dependant behaviour to simulation the obvious answer is:
“Measure the S-parameter and use them as simulation input”: a vector network analyser with a four-terminal-pair
S-parameter measure set-up was used for getting common mode and differential mode S-parameter.
Measure
• standard measurement configurations
• oscilloscope, vector network analyzer
• common/differential mode (CM/DM)
• time/frequency domain
time domain
signal shapes
1st
measurement
2nd
measurement
frequency domain
S-parameter
• frequency domain simulation
• reverse Fourier transformation
1600
logarithmic
steps
1600 linear
steps
1)
∑
3200 steps
time domain
signal shapes
1)
University Esslingen, Automotive
Electronics Lab, Prof. Jürgen Minuth
cross check
from 9 kHz
up to
500 MHz
from 9 kHz
up to
200 MHz
from 9 kHz
up to
500 MHz
A
B
C
1)
Figure 2: Workflow overview 1st approach “measurement  simulation”.
A simulation was done for getting the signal shapes for a single 100 ns bit in a 50 Ω environment. The
succeeding figure shows the differential mode results only:
 Measurement and simulation seem to fit perfect in all three cases A, B and C.
 Unmotivated ringing appears.
100ns
3µs
50Ω
V_Start
V_End
differential mode simulation ≈100 m cable
( S-parameter measurements, reverse Fourier)
15,6V
0V
100Ω
50Ω
Figure 3: Measurement and simulation set-up (100 ns and 3 µs bit).
24 April 2014
2-7
Workflow: Cable Modelling and Simulation
1)
9
ICTON 2014
unmotivated
“ringing”
8
V_Start_lin/log
A
S‐parameter log
B
V_End_lin/log
S‐parameter lin
7
V_Start_log
S‐parameter lin/log C
6
V_End_log
amplitude in V
V_Start_lin
5
V_End_lin
4
3
V_End
2
V_Start
1
0
0
200
400
600
‐1
800
1)
time in ns
1000
1200
University Esslingen, Automotive
Electronics Lab, Prof. Jürgen Minuth
Figure 4: Using S-parameter measurement results in simulation (100 ns bit).
The next example checks the behaviour when applying a “long” 3 µs bit. All three cases are far away from
acceptable results:
 Unmotivated ringing appears.
 Completely faulty signal shapes appear ( linear measure).
 The steady state symmetry enforced by the set-up is not available.
1)
faulty signals
( under-sampling)
15
13
V_Start_lin/log
S‐parameter log
unmotivated
ringing
S‐parameter lin
V_End_lin/log
11
V_Start_log
S‐parameter lin/log C
V_Start
amplitude in V
A
B
V_End_log
9
V_Start_lin
steady state symmetry
7,8V
V_End_lin
7
steady state symmetry
5
steady state
symmetry error
3
V_End
1
‐1 0
1
2
3
4
time in µs
5
6
1)
7
8
University Esslingen, Automotive
Electronics Lab, Prof. Jürgen Minuth
Figure 5: Using S-parameter measurement results in simulation (3 µs bit).
The reasons for this deviation are known, their elimination is difficult or impossible:
 Parasitic components generated by the set-up itself.
 Poor measurement accuracy especially at low frequencies (phase and attenuation).
 Under-sampling at low frequencies in the linear case.
24 April 2014
ICTON 2014
3-7
Workflow: Cable Modelling and Simulation
ICTON 2014
1
frequency domain
S-parameter
time domain
signal shapes
value tables Z, γ
e.g. 100Hz - 600MHz
Approximations (CM/DM)
generic physical model
• dielectric losses
• skin-effect losses
• from …mHz up to …GHz
Approximations (CM/DM)1)
rational functions
• line impedance Z
• propagation coefficient e-γℓ
3
polynomial coefficients
(various lengths)
Z, e-γℓ, edge-speed v0
S-parameter
• frequency domain simulation 2)
• reverse Fourier transformation
time domain
signal shapes
1)
2)
2)
System
topology, stimulus,
CAN, FlexRay™
etc.
2)
2
cross check
simulation
phase
Measure
• standard measurement configurations
• oscilloscope, network analyzer
• common/differential mode (CM/DM)
• time/frequency domain
approximation
phase
2.2 Measurement and Simulation Results (2nd Approach)
The deviation between simulation and measured signal shapes in the 1st approach is not acceptable2. A modified
2nd approach separated into an approximation phase and a simulation phase is necessary:
 Approximation phase
1: Measure signal shapes in time domain and S-parameter in frequency domain.
2: Parameterize a generic physical cable-model3 for best matching (simulation  measurement).
3: Approximate4 the output of the generic model with broken rational functions for several cable lengths.
 Simulation phase:
A dedicated cable-model5 uses the rational functions for user’s simulation of network architectures or
topologies in time domain.
The succeeding figure illustrates the complete work-flow (including the cross-check procedures during the
development process).
time domain
signal shapes
TU Dortmund, On-board Systems Lab, Prof. Stephan Frei
University Esslingen, Automotive Electronics Lab, Prof. Jürgen Minuth
Time-Domain Simulation
Line Model1)
• VHDL-AMS
cable, transceiver,
bus driver, CMC etc.
System evaluation
• timings (delays, edges)
• eye charts
• SI-voting
• ringing
• etc.
Figure 6: Workflow overview 2nd approach “measurement  approximation  simulation”.
When applying the generic physical model to the example according figure 3 practically perfect results are
generated (black lines in the succeeding figure):
 The unmotivated ringing is eliminated.
 The symmetry error is eliminated.
 Self-inductive effect due to cable’s inner inductivity appears.
 Simulation and measurement fits practically perfect (not shown in the figure).
2
Phase and attenuation error especially at low frequencies, parasitics which cannot be eliminated by
recalibration.
3
The model allows adjusting the properties of real cables like DC-resistor, capacitive and inductive loads per
unit length, frequency dependant skin-effect and last but not least frequency dependant dielectric losses
(designed by Prof. Jürgen Minuth, University of Esslingen, Automotive Electronics Lab).
4
designed by TU Dortmund, On-board Systems Lab, Prof. Stephan Frei
5
designed by TU Dortmund, On-board Systems Lab, Prof. Stephan Frei
24 April 2014
4-7
Workflow: Cable Modelling and Simulation
ringing
eliminated
V_Start
amplitude in V
1)
1)
9
8
7
6
5
4
3
2
1
0
‐1 0
ICTON 2014
generic model
V_Start_lin/log
 approach 2
V_End_lin/log
measured S‐parameter C
V_Start_log
 approach 1
A
V_End_log
V_Start_lin
B
V_End_lin
V_start_generic model
V_end_generic model
V_End
100
200
300
400
500
time in ns
600
700
800
900
1000
symmetry error
eliminated
15
13
amplitude in V
11
9
V_Start
7
V_End
5
area below 0 correct
3
1
‐1 0
1)
1
2
3
4
time in µs
5
6
7
8
University Esslingen, Automotive
Electronics Lab, Prof. Jürgen Minuth
Figure 7: Using S-parameter measurements and approximation results (generic model)
in simulation (100 ns and 3 µs bit).
2.2.1 Details to the Approximation Phase Step 3
The generic physical model is able to generate data available from DC up to several GHz. Knowing the
simulation sceneries allows to limit the frequency area which shall be covered by the approximation step 3 to
e.g. a few Hz up to a few 100 MHz. The approximation tool6 fits rational functions best:
 INPUT
o cable impedance Z as value table
o propagation coefficient γ as value table
o required lengths ℓ
 OUTPUT
N
o
polynomial coefficients ak; bk for the cable impedance
Zω  
 a  jω
k
k 0
N
 b  jω
k 0
o
edge-speed
k
k
v0
 c k  jω 
M
o
k
polynomial coefficients ck; dk for the propagation coefficient
e
 γ(ω)

k 0
M
k
 d k  jω 
k
e
j
ω

v0
k 0
o
6
Hints
The coefficients ak, bk, ck and dk are real numbers, the polynomial orders M and N are typically
between 8 and 12.
Cable lengths ℓ can be chosen (length from a few centimetres up to some 10 meters are
typical).
designed by TU Dortmund, On-board Systems Lab, Prof. Stephan Frei
24 April 2014
ICTON 2014
5-7
Workflow: Cable Modelling and Simulation
ICTON 2014
The succeeding example visualizes the approximation quality using a typical automotive twisted pair cable:
 CABLE
o uncoated twisted pair
o specified differential mode cable impedance app. 98 Ω
o generic physical model available from 0 Hz up to 2 GHz
 INPUT
o cable impedance Z as value table from 100 Hz up to 600 MHz
o propagation coefficient γ as value table from 100 Hz up to 600 MHz
o length of the cable 10 m
 OUTPUT (differential mode only here)
o polynomial coefficients for the cable impedance (order 11)
o polynomial coefficients for the propagation coefficient e-γℓ (order 11)
o edge-speed v0 (20 cm/ns)
Inside the approximation area the matching between generic physical model and the rational functions is
practically perfect (see succeeding figures: lines  dotted lines). From automotive communication point of view
the model covers the area from DC up to the speed of FlexRay™ (10 Mbit/sec) perfectly. The deviation of the
cable impedance for low frequency does not influence the results if using cable lengths available in passenger
cars or trucks.
impedance in Ω
4000
2)
1)
generic physical model
3000
Re{Z}
2000
rational function
for time domain simulation
Re{Z}
1000
rational function
deviation for f  0
generic physical model
Im{Z}
2)
rational function
for time domain simulation
Im{Z}
1)
0
frequency in MHz
‐1000
‐2000
‐3000
‐4000
1)
TU Dortmund, On-board
Systems Lab, Prof. Stephan Frei
2)
perfect matching in the definition area 100 Hz … 600 MHz
University Esslingen, Automotive
Electronics Lab, Prof. Jürgen Minuth
Figure 8: Matching cable impedance Z “generic model”  “rational function model”.
2)
generic physical model
Re {exp (‐γℓ)}
2)
generic physical model
Im {exp (‐γℓ)}
1)
1
1
rational function
for time domain simulation
Re {exp (‐γℓ)}
1)
rational function
for time domain simulation
Im {exp (‐γℓ)}
rational function
deviation for f  ∞
0
0,00001
0,0001
0,001
0,01
0,1
0
1
10
100
500
600
700
800
900
1000
1100
1200
1300
frequency in MHz
rational function
deviation for f  ∞
‐1
‐1
1)
perfect matching in the definition area 100 Hz … 600 MHz
TU Dortmund, On-board
Systems Lab, Prof. Stephan Frei
2)
University Esslingen, Automotive
Electronics Lab, Prof. Jürgen Minuth
Figure 9: Matching propagation coefficient e-γℓ “generic model”  “rational function model”.
length ℓ = 10 m; edge speed v0 = 20 cm/ns
24 April 2014
6-7
Workflow: Cable Modelling and Simulation
ICTON 2014
2.3 Simulation in time domain
The On-board Systems Lab team7 designed a cable-model (rational function approach) in VHDL-AMS enabling
time domain simulation. The matching between measurement and simulation is practically perfect from lowspeed CAN via CAN-FD up to FlexRay™. Topology simulations generate results which meet the accuracy
requirements to support system evaluations. Post processes are available e.g. for generating eye charts,
measuring slew-rates or ringing durations.
2.4 Application Example
Often topologies generate ringing when switching bus-interfaces from a driven state to a high-ohmic state. In
case of CAN this corresponds to the change from any dominant bit to any recessive bit; in case of FlexRay™
this corresponds to the change of the EOF-bit to the idle state at the end of each frame. For evaluating the
sampling properties it is essential knowing the ringing duration. The succeeding figure illustrates the ringing
effect of four different cables in the use-case of a passive star with five nodes.
CAN: dominant
FlexRay™: EOF
TxD = 0
1,2
1)
CAN: recessive
FlexRay™: idle
TxD = 1
1
2 x 0,08mm²
2 x 0,14mm²
2 x 0,35mm²
2 x 0,75mm²
0,8
DC
behavior
0,6
0,4
example: sample
threshold
area
voltage in V
0,2
… DC
behavior
0
0
500
1000
1500
2000
2500
‐0,2
3000
3500
time in ns
‐0,4
‐0,6
‐0,8
‐1
≈ 700 ns
1)
≈ 850 ns
University Esslingen, Automotive
Electronics Lab, Prof. Jürgen Minuth
‐1,2
Figure 10: Effect of frequency dependant losses: ringing in CAN and FlexRay™.
The time constant of ringing’s envelop curve is linked strongly with the type of cable, in other words: with their
frequency dependant losses. The complete area from DC op to “high frequencies” with reflections is covered.
The question of a system designer: “How is the earliest possible sample-point influenced by the cable type?” can
be answered. Assuming the blue area shows the threshold variations, we get: “case red/blue: ≈ 700 ns” and
“green/brown: ≈ 850 ns”. The example would support a CAN application up to a few 100 kBit/sec.
3. CONCLUSION
The presented workflow offers a straight forward approach for of a time-domain simulation of topologies; the
cable-models are parameterized based on measurement results. The simulation covers the DC-area up to the
speed of FlexRay™ signals; however the new models can be parameterized for being used with much higher
baud-rates. The matching between measurement and simulation meets engineer’s accuracy requirements. The
achievable simulation speed is high enough to do topology simulations with e.g. 32 nodes where each node may
transmit round robin messages.
ABBREVIATIONS
CM
common mode
CMC
common mode coil
DM
differential mode
EOF
end of frame bit, followed by a high ohmic phase on the cables
ESD
electro static discharge
SI-voting a procedure defined in FlexRay™ for estimating the signal quality
7
TU Dortmund University, Prof. Stephan Frei
24 April 2014
ICTON 2014
7-7