2012 pb vi trajectory plots for transmission line models evaluation

Copyright 2012 Piero Belforte November 20th 2012
1
V-I trajectory plots for Transmission Line models evaluation
A comparative analysis between an exact distributed model and the
equivalent LC lumped model of a transmission line (TL) is shown. In a
previous paper the comparison is made using Worst Case Eye Diagrams
https://docs.google.com/file/d/0BxZqV10CSiNS1JwTnc1NVIyTWs/edit.
Here a method based on the Voltage/Current (V-I) trajectories at line
ports is presented.
The reference test circuit is the following:
Fig.1 Comparative test circuit

2
The lumped model is built up using 10 LC cells is shown here:
Fig.2 Basic LC cell used for the 10-cell model of the TL
The resulting Impedance √L0/C0 is 50 ohm and the cell delay is √L0*C0=
100ps/cell for a total equivalent delay of 1ns.
The exact model is created using the TL primitive of DWS (Digital Wave
Simulator).
As test stimulus a voltage step of amplitude 1V and 1ps rise time is
applied to both test configurations. The signal propagates trhough the
line and is reflected back by the open termination (R0,R1= 100
Gigaohms). When arrives at generator's end it is inverted and totally
reflected forward by the 0 impedance of the generator. The evolution of
persistent oscillations occurring in both configurations (no dissipative
element is present in the circuits) points out the approximation of the
lumped model versus the exact SWAN model of the TL.
The two models are simulated simultaneously using a sim tstep of 1ps
using the tool SpicySWAN set in SWAN mode.

3
Here the SWAN netlist extracted from the test circuit :
Fig.3 SWAN (DWS) netlist extracted form the circuit of fig.1
Simulation results are shown here as a multiplot:
Fig. 4 Multiplot of sim results coming from circuit of fig. 1

4
Plotting the voltage VOUT_2=V(3) at TL's output port versus the current
flowing into the generator V1= I(V1,1) the following V-I trajectory is
obtained:
Fig. 5 V-I trajectory related to TL SWAN model TSTOP= 50ns tstep=1ps
(20Ksamples plotted, 2.5ps plt step)
Fig. 6 V-I trajectory related to approximated 10-cell LC model, TSTOP=
50ns tstep=1ps (20Ksamples plotted, 2.5ps plt step)

5
Increasing the simulation time window the following trajectory is
obtained:
Fig. 7 V-I trajectory related to approximated 10-cell LC model, TSTOP=
500ns tstep=1ps (.5Megasamples calculated, 20Ksamples plotted, 2.5ps
plt step)

6
Comparing previous results using the same scale the result shown in the
following fig.8 is obtained.
Fig.8 Comparative images showing the V-I trajectories related to circuit
LC_CHAIN_TL plotted with the same I and V scales.
In the previous fig.8 in yellow the inner area of trajectory envelope is
highlighted. This area is perfectly rectangular in the ideal case (SWAN
model of TL). For the 10 LC-cell model this area is no more rectangular
and becomes smaller because waveform distortion increases increasing
the time window of the analysis.

7
Analysis after 5 Million reflections
To stress both models and SWAN algorithms the simulative analysis has
been extended to simulation times in the order of milliseconds with a
time step of 5ps. In particular the following results refer to a time window
of 500 nanoseconds between 5 and 5.0005 msec. 1 BILLION time points
(1 Gigasample/waveform) have been calculated and
100ksamples/waveform are plotted (plotting step: 5ps). The simulation
elapsed time is a few minutes on a standard PC.
Fig. 9 Netlist related to comparative test circuit
In the netlist shown in fig. 9, the GMIN/GMAX options of DWS are set to
values exceeding default values in order to keep power dissipation at
both ends of the line to a minimum. These setting lead to a generator
resistance of 1 nanoohm and termination resistance of 1 Teraohm (1000
Gigaohm). The instantaneous powers entering the generators P(V1,1)
P(V0,10001) are also calculated by DWS.

8
Fig. 10 Simulated waveforms related to netlist of fig 9
The power at generator V1 port has peak values of 19.95mW that means
50uW less than the value exchanged with the line at the first reflections.
After about 5Million reflections 50uW is the power loss due to GMIN and
GMAX settings and dissipated at non ideal (zero and infinite) line
terminations.

9
Fig 11 V-P (Voltage-Power) plots related to LC_CHAIN_TL test circuit
after 5msec from the start (5 Million back and forth reflections).
From V-P plots of fig. 11 it can be pointed out a still near ideal
rectangular diagram for SWAN TL (the 50uW dissipated power is
negligible at this scale). The random oscillations of power exchanged in LC
model have still a behavior similar to that of the beginning of oscillations
because also in this time window the power dissipated at terminations is
still negligible.

10
Lumped LC model with higher number of cells.
A similar comparative analysis can be performed increasing the number
of LC cells of the TL lumped model. Both a 100-cell and a 1000-cell models
have been analyzed in SWAN mode. L and C values are obviously
decreased by a factor 1/10 and 1/100 with respect the 10-cell situation in
order to maintain the same equivalent Z0 and TD of the equivalent TL. To
keep the delay error to a minimum the cell inductance is modeled using a
serial adaptor instead of the standard link model. This model called also
"stub" model is equivalent to the trapezoidal rule of integration of
inductance equations.
Fig. 12 LC cell using a Serial Adaptor stub model of the inductor
equivalent to the trapezoidal rule of integration.
The SERIAL ADAPTOR (AS0) is a SWAN/DWS specific element that places
the net connected at its third node (port) in SERIES between the other
two nodes of the adaptor. Even if this stub model requires two SWAN
circuit elements (adaptor and inductor), it has the advantage of
minimizing the delay error of "LINK" default model of the inductor,
equivalent to a unit-delay (tstep) TL.
This advantage can be particularly useful when several LC cells are
connected in series in a CHAIN as happens for the Spice-like lumped
model of TL.

11
Fig. 13 IV trajectories related to several implementations of the lumped
model Transmission Line (SWAN).

12
As can be pointed out by the differences among the V-I plots of fig. 13,
increasing the number of cascaded cells, the trajectories approach the
ideal line RECTANGULAR shape, but from the 1,000 cell up no significant
advantage is obtained if the number of cells is increased to 2,000 or even
4,000 cells using a simulation time step of 1ps.
To achieve a better result a situation with 9,999 cascaded LC cells has
been simulated with a time step of 10 femtoseconds (SWAN).
Fig.14 Time domain and V-I plots related to a 9999-cells LC ladder
network simulated with a time step of 10 femtoseconds.

13
This last circuit required the calculation of .5 Million points
(Megasample/waveform) for a network containing about 30,000
elements (including series adaptors). As can be pointed out from the plots
of Fig. 13, only in this situation a good agreement with TL distributed
model is obtained at the expense of a huge simulation effort that only
SWAN can afford with simulation times in the order of minutes.
A residual rise time increase of about 500fs and some ringing still affect
the output waveform with respect TL distributed model.
Voltage pulse stimulus
Similar comparative test can be carried out on TL models using a 500ps
(1ps edges) Voltage pulse stimulus in place of step input previously
shown. In this case the IV pattern of the exact SWAN model becomes a
cross instead of a rectangle. The results are shown in the following fig.14.

14
Fig. 14 Comparative model tests using a 500ps Voltage pulse stimulus
Spice models of TL
Spice-derived simulators have several problems dealing with TL models.
Being based on resolution of NA (Nodal Analysis) equations, these
simulators assume no propagation of signals inside the circuit under
analysis. Variable time step control adds further problems in dealing with
fixed delays. To avoid these issues, very often TLs are approximated with
LC networks leading to signal distortions typical of this Kind of models as
previously pointed out. An example is shown in Fig. 15 where the voltage
pulse stimulus (500ps width) is applied to a TL of 50 ohm impedance and
1ns delay on a commercial version of Spice (LT Spice)

15
Fig. 15 Results of LT Spice simulation of TL version of circuit of fig.14
(Yellow: Line input pulse, Violet: signal at the Line end )
In fig.15 the reflected pulses are affected by heavy progressive
distortions typical of a lumped model. The pulse edges slow down
progressively and overshoots/undershoots appear.
Some simulators, like MicroCap, use more accurate TL models and can
can work at fixed time steps, but the simulation times are order of
magnitudes higher than those of SWAN/DWS.
Even the simulation of lossless LC ladder networks are not so easy with
Spice. To keep the error to a minimum a very short fixed time step (10fs-
1ps) should be used also in this case. A comparison between SWAN and
MC10 simulation times are shown here:
http://www.youtube.com/watch?v=pTMBnvUChog
and here: http://www.youtube.com/watch?v=xJnA5ioAwb4
Even for this simple 10-cell test circuit the speedup factor of SWAN vs
MC10 is in the order of 70-100 at same accuracy level (differences less
than 1mV between the waveforms coming from the two simulators,
fig.16). This speedup increases if the number of cascaded cells increases.

16
As shown previously SWAN can simulate a 9,999 cell LC network at 10fs
time step in minutes on a standard PC.
Fig. 16 Comparison of MC10 and SWAN/DWS simulations of a 10-LC cells
ladder circuit
Conclusions
In this paper a new method for evaluating Transmission Line simulative
models is presented. The effectiveness of this method, based on V-I plots,
has been demonstrated comparing TL lumped and distributed models,
both feasible with the SWAN/DWS simulator. Even if DWS has its
maximum effectiveness with distributed (exact) models, its advantages
over conventional NA circuit simulators are also impressive in case of
lumped parameter models. For these last models speedup factors of at
least 2 order of magnitudes have been observed with respect
conventional tools working at the same accuracy level. Large cell number
(up to 10K) models with femtosecond time step are out of reach of NA
tools.

2012 pb vi trajectory plots for transmission line models evaluation

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to 2012 pb vi trajectory plots for transmission line models evaluation (20)

More from Piero Belforte (20)

Recently uploaded (20)

2012 pb vi trajectory plots for transmission line models evaluation