zyxwvutsrqponmlkjihg

zyxw
558 IEEE TRANSACTIONS ON EDUCATION, VOL. 39, NO. 4, NOVEMBER 1996

Feinan zyxwvut
ents on a Coupled Oscil
N.Yan, S. T. Yip,

Abstract-The theory of a symmetric and asymmetric coupled

and H. K. Wong

oscillator circuit with an arbitrary initial condition was verified

by a novel experimental technique.

MmMn
I. INTRODUCTION L L

C OUPLED oscillators (Fig. 1) are often used in mechanics

to explain normal modes [I] and in materials science
to model molecule adsorption on crystal surfaces [a]. Each
normal mode corresponds to a vibration of the system with
only one (resonant) frequency. An arbitrary vibration can be
described as a mixed state of these normal modes. An electric
oscillator containing capacitors and inductors has analogous
essentials, with the inductance, capacitance, voltage, current, Fig. 1. A coupled oscillator circuit.
and resistance corresponding to the mass, compliance, force,
velocity, and friction of a mechanical system, respectively. In
addition, an electric coupled oscillator is more nearly ideal
equations by Kirchhoff‘ s rules
than most mechanical oscillators and easy to work with. Some
experiments [3J using electric oscillators have been designed
to demonstrate normal modes, but they appear to be too
simplified. In fact, the circuits used in these experiments
are forced and thus not identical to that shown in Fig. 1.
The analysis of these forced oscillation circuits is actually where

zyxwvu
quite complicated [4]. To avoid such confusion, we have
developed a new method by which the circuit shown in Fig. 1 w: = 1/LC1, wz” = 1/LCa, x = l/LCO. (2)
can be studied theoretically and experimentally without any
modification. In this paper, we will explain the method and
The total charge on the upper part of the circuit is
then use it to study the normal modes and the effect of
initial conditions for both symmetric [I], [31 (C, = CZ)
and asymmetric [4] (C, # C2) coupled oscillators. The Q Q1 + Q 2 + Qo (3)
experiments are straightforward and easy to follow. The results
match the theory very well and thus facilitate the learning
of the concept of normal modes as well as the transient
characteristics of electric oscillators. The same approach can
be extended to nonlinear coupled oscillators that have found
many recent applications in neural networks [5] and chaos
[a ~71.

11. THEORY
zyxwvu
which depends on the initial condition of the oscillation and,
in general, is a nonzero constant. (Both [3] and [4] assume
Q = 0.) To simplify (I), we define a voltage V such that

By also defining

The theory of coupled oscillators for an arbitrary initial

condition can be found in some works [SI. To facilitate
the interpretation of our experimental results, we found it Equation (1) can be rewritten as
necessary to outline the theory first. For the coupled oscillator

zyxwv
circuit shown in Fig. 1, we may write down the two loop

zyxwvutsrqponm
Manuscript received February 10, 1994; revised June 25, 1996.
The authors are with the Physics Department, The Chinese University of
Hong Kong, Shatin, Hong Kong.
Publisher Item Identifier S 0018-9359(96)08741-9.
Equation (5) can be simplified in terms of the normal mode
coordinates x1 and x2, which are linear combinations of 41
0018-9359/96$05.00 0 1996 IEEE
and q2
zyxwvutsrqp
zyxwvutsrqponmlkjih
zyxwvutsrqponmlk
YAN et al.: EXPERIMENTS ON A COUPLED OSCILLATOR

Thus, from (9) and (10)

559

frequencies w&
zyxw
zyxwvutsrqpon
zyxwvutsr
By definition, the normal mode coordinates oscillate with

XI = 2
-w+x1
similarly, for the mode with frequency w-, l:lo = 0
and the initial condition must satisfy

2
x2 = -w-x2.

These require [from (5)] In this mode

v, = (VI0 - V )cos w-t +v

and v,= ( V & - V ) c o s w - t + V .
2x
tan24 = (7) For a symmetrically coupled oscillator, C1 = and
w; - wz”’
~

(72
t a n $ = 1. From (6), (ll), and (13), we get
We now consider the effect of initial conditions. The capacitors
C1 and C2 are first charged at dc voltages VIO and Vm, W: = w: + 2X when VZO= VIO,
respectively, i.e., at t = 0, (14)
w- = w1 when V ~=O-Vlo.

The expression for w+ clearly indicates that the coupling

strength is stronger for smaller C.
In practice, the circuit unavoidably carries a nonzero resis-
tance R in each loop. All previous equations for Q1, Qz, V I ,
and Vz must be modified by the insertion of a damping factor
ePRtIzL; e.g., (12) should be rewritten as

The solutions that satisfy this initial condition (41 = 42 = 0)

111. EXPERIMENT

zyxwvutsrqp
are
The circuit used in our experiments is illustrated in Fig. 2(a).
x1 = x10 cosw+t ana 5 2 = 2 2 0 cos w-t. For clarity we highlighted the coupled oscillator configuration
as that shown in Fig. 1. With the help of five switches, we can
The charges on the capacitors are thus given by set up any initial condition required for a particular oscillation.

zyxwvuts
These switches are controlled by a square wave generator
&I = 210 cos 4 cos w+t - 520 sin $ cos w-t + C1V (operated at a frequency of 1-3 kHz). Their positions aae listed
QZ = 2 1 0 sin 4 cos w+t + 520 cos 4 cos w-t + C,V (9) in Fig. 2(b). State 1 is for setting up the initial conditions.
Qo = Q

520
- Q1 - QZ

where, as required by the initial condition

q o =+C1(V1o-V)cos4+C2(V2o-V)sin+
= -CI (VI, - V )sin 4 + Cz (Vz, - V )cos 4.
For the normal mode with frequency w+, 5 2 0 = 0, which
requires V10 and V20 (the initial condition) to satisfy
(10)
In state 2, the circuit becomes exactly the same as shown

zyx
in Fig. 1. In other words, the same circuit is used in both
the theoretical analysis and the experimental study. This will
eliminate the confusion that can arise in previous approaches.
In order to reduce the damping of the oscillations, 5‘4 and
S5 must have very small contact resistances. Initial efforts
using relay switches were not successful because of their slow
speed, high contact resistance, and bouncing contact problems.
The integrated-circuit industry has developed a number of
solid-state switches for high-speed operations, and these so-
called analog switches have become commercially available.
where t a n 4 is determined by (7). We have tested many analog switches and found that only
560 IEEE TRANSACTIONS ON EDUCATION, VOL. 39, NO. 4, NOVEMBER 1996 zy

Yo zyxwvutsrqpo "20

Fig 2 zyxwvutsrqp
zyxwvutsrqp (b)
(a) Schematic diagram of the circuit for the experiments. The square wave generator was operated at 1-3 kHz. We used DG180BA for switches

zyxwvuts
S y , S4,and Ss , DG411D.I for S
SI, z, and 7400S1 for the inverters. (b) Positions of switches

DG180BA [9], which has less than 10 R resistance when Energy Transfer: In this expeiiment we set V ~=O5 V and
closed, can fulfill the requirements. Identical inductors with Vzo = 0. For this initial condition
L = 860 Ifr 2 pH and dc resistance less than 1.2 R were
selected for these experiments. The values of L and C were
measured with a universal bridge. The voltages across C1 and
C2 were measured with a low-cost digital storage oscilloscope fV
that was interfaced to a PC for data storage and analysis.
The oscillation data VI and V2 were fitted to the derived
equations using a commercial software package called PEAK-
FIT [lo].
Fig. 5 shows the oscillations of VI and Vz demonstrating the
IV. RESULTS energy transfer phenomena. The effect is more conspicuous for
large values of CO.The time required for energy to transfer
back and forth is
A. Symmetric Coupled Oscillator
Normal Modes: Identical capacitors with C1 = 6 2 = CO= 2T
to = -. (16)
39.2 k 0.2 nF were selected for this experiment. Fig. 3 shows w+ - w-

zyxwvuts
some typical results. Fig. 3(a) and (b) illustrates the two pure
normal mode oscillations. Fig. 3(c) is a typical graph for other
initial conditions. These data can be fitted with (15) and
the fitting parameters give us the normal mode frequencies
(fh= w*/27r), the damping time constant ( 2 L / R ) ,and the
value of V as given by (8).
Coupling Strength: In this experiment, we measured w* as
a function of CO.The result is shown in Fig. 4. The lines on
the graph were drawn according to (14).
We measured t o as a function of COand the result is plotted in
Fig. 6. The curve on the graph was drawn according to (16).
B. Asymmetric Coupled Oscillator
Normal Modes: As we discussed in the theory section, the
initial conditions required to observe normal modes in an
asymmetric coupled oscillator [expressed by (11) and (13)]
are not as simple as in a symmetric coupled oscillator [(14)].
 ~

zyxwvutsrqp
zyxwvutsrqponmlkjih
zyxwvutsrqponmlk
YAN et al.: EXPERIMENTS ON A COUPLED OSCILLATOR 561

:zyxwvu
tzyx
zyxwvuts t

zyxw
4-

0 I I I I 2-
-
c.
h
6 -

4 h
(b)
->P- 0-

-2 -
2 c
8
-
-4t
n
- 0
0

>"
-2

-4
m
f . "F------.
4

-0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

I I I I

zyxwvutsrqp
-0.1 0 0.1 0.2 0.3 0.4
t 1"

zyxwvutsrqponm t (ms>
Fig. 3. Oscillation graphs for the symmetric coupled oscillator. The curves
are simulations with fitting parameters f- = 27.4 kHz, f+ = 47.5 kHz, and

zyxwvuts
R = 11.4 0 : (a) VIO = VZO= 5 V. (b)VIo = -VZO= 5 V, (c) VIO = 5
v, V 2 0 = 7 v.
1 -
Fig. 5. Demonstration of energy transfer Initial condition here was V ~ =
v, V2/20 =

0'7
0.6
0.
O5

7
f'

500L I
0 0.2 0.4 0.6 0.8 1 1.2 Fig. 6. Time of energy transfer t o as a function of CO.The curve was
drawn according to (16).
c, /CO
Fig. 4. Normal mode frequencies for the symmetric coupled oscillator. The
lines were drawn according to (14).
in Fig. 3(a) and (b)]. The results for various combinations of
C1 and CZ are plotted in Fig. 7, which shows that the initial
conditions for normal mode oscillations can be predicted very
For each combination of C1 and C Z ,we need to adjust the
well by (11) and (13).
initial conditions (e.g., vary V ~with
O V& fixed) to achieve w* Normal Mode Frequencies: The normal mode frequencies
normal mode oscillations [i.e., exhibiting behavior like that are given in (6). To verify these complicated expressions, we
562 zyxwvutsrqponm
zyxwvutsrqponm
1.2, I
IEEE TRANSACTIONS ON EDUCATION, VOL. 39, NO. 4, NOVEMBER 1996

very low contact resistance when closed. With the help of

these novel switches, one can set up any initial condition
and place the circuit in a configuration identical to the one
studied in theory, allowing direct comparison between theory
and experiment. This new approach can be easily extended to
other more complicated circuits, such as anharmonic coupled
oscillators [6], [7], coupled oscillating neural networks [5],
,/ and lumped LC transmission lines [ 111, etc. The technique is
especially useful in studying the transient states of an electrical
system and chaos in a nonlinear coupled oscillator, both of
which depend on the initial conditions [6], [7].

ACKNOWLEDGMENT

zyxwvutsrqpon
zyxwvu
The authors appreciate the helpful discussions with C. F. Lo

zyxwvutsrqpon
zyxwvutsrqpo
zyxwvu
binations of C1 and Cz, where X +

(11) and (13).

0
tan +
Fig. 7. The initial conditions of normal mode oscillations for various com-
c,(v,,-v),X- E -cl(vlo-v)
c1( VlO - V ) cz(vzo,-v)’
and tan 4 was calculated using (7). The line shown was drawn according to
and the technical assistance of W. K. Chao. They also thank C.
C. Cheung and H. K. Won for their preliminary experiments.

REFERENCES

F. S. Crawford, Jr., Waves. New York: McGraw-Hill, 1968, pp. 27-28.

F. M. Hoffmann, “Infrared reflection-absorption spectroscopy of ab-
sorbed molecules,” Suq! Sci. Rep., vol. 3, pp. 107-192, 1983.
“I A. M. Portis, Laboratory Physics. New York: McGraw-Hill, pt. B,
1965, pp. 39-45.
N. W. McLachlan, Theory of Vibration. New York: Dover, 1951, pp.
62-64.
J. G. Taylor and C. L. T. Mannion, Eds., Coupled Oscillating Neurons.
London: Springer-Verlag, 1992.

zyxwvu
C. Hayashi, Nonlinear Oscillations in Physical Systems. Princeton, NJ:
Princeton University Press, Part 111, 1985.
T. Yamada, K. Fukushima, and T. Yazaki, Phase Transitions, vol. 29,
pp. 15-20, 1990.
K. R. Symon, Mechanics, 3rd ed. Reading, MA: Addison-Wesley,
1971, pp. 191-197.
DG18OBA and DG411DJ are supplied by Siliconix Inc., 2201 G
Laurelwood Rd., Santa Clara, CA 95054 USA.

0’
zyxwvut
1 1.5 2 2.5 3 3.5
1
4
Product of Jandel Scientific, 2591 Kerner Blvd., San Rafael, CA 94901,
USA.
R. C. Cross, Amer. J. Phys., vol. 53, pp. 563-567, 1985.

zyxwvuts
0.5

1/C, + 1/C, (0.lnF”) Feinan N. Yan received the B.Sc. and M.Sc. degrees in physics from the
University of Science & Technology, ROC, in 1984 and 1987, respectively.
Fig 8 Normal mode frequencies as a function of &+& with She is a Teacher in the Fundamental Physics Center, University of Science
6, i2
_ _ _ = A(#-’ The lines were drawn according to (6). & Technology. In 1993 she visited the Chinese University of Hong Kong,
Hang Kong and performed the experiments presented in this paper.

varied C1 and Cz such that 2

c1 & &
=~ (nF)-l (i.e., w l - w i
was kept constant). Following the procedures described in
Section IV-B, we obtained the values of w* and plotted them S. T. Yip received the B.Sc. and M.Ph. degrees in physics from the Chinese
as a function of &+& (see Fig. 8). The straight lines in the
University of Hang Kong, Hang Kong, in 1982 and 1984, respectively.
He became the Technician at the Chinese University of Hong Kong, and is
plot were drawn according to (6). now the Laboratory Safety Officer and Radiation Protection Officer.

V. CONCLUSION
We have presented the theory for a symmetric and asym-
metric coupled oscillator with arbitrary initial conditions. All
theoretical results were verified by experiments. The key for
the success of these experiments is the use of the analog
switches. These new electronic devices are very fast and have
H. K. Wong received the Ph D degree in physics from Northwestern
University, Evanston, IL, in 1985
After a period as a Visiting Scientist at the IBM Research Laboratory, San
Jose, CA, he joined the Physics Department of the Chinese University of
Hang Kong, Hang Kong He is currently Associate Professor His research
concentrates on thin film physics
IEEE TRANSACTIONS ON EDUCATION, VOL. 39, NO. 4, NOVEMBER 1996 zyxwvutsrqp 563

Time Domain Solutions to Partial

Differential Equations Using SPICE
zyxwvut
William R. Zimmerman, Member, ZEEE

Abstract-In this paper a method for using a time domain

circuit code to solve partial differential equations is described.
Rather than following the usual approach of developing a
lumped-element circuit model, the partial differential equation
is finite differenced in space and written in state variable form.
The resulting system of coupled ordinary differential equations is
then modeled by an array of coupled voltage dependent current
sources connected to a string of capacitors. A preprocessor is
used to write the network list in a form usable by the SPICE
circuit analysis code. Examples for advection, diffusion, and
electromagnetic propagation in one spatial dimension are given.

I. INTRODUCTION

C IRCUIT ANALYSIS programs such as SPICE [1] have

been successfully used to analyze a wide variety of
linear and nonlinear electrical networks. These codes are
characterized by automatic time step adjustment algorithms
and sparse matrix techniques and can address large electrical
networks. Through the use of lumped-element network analogs
[2]-[4] circuit analysis programs have also been used to solve
other systems of differential equations such as transmission
line equations [5] and Laplaces’ equation.
Since much effort has been expended in developing SPICE
and other circuit analysis codes, it would be useful to extend
the use of these codes to general systems of partial differ- Fig. 1. Method for solving a system of linear ordinary differential equations
ential equations. In principle, this should be straightforward using SPICE. (a) System of linear ODE’S. (b) Equivalent circuit equations.
(c) Equivalent circuit.
as removing the spatial derivatives in a system of partial
differential equations by an arbitrary finite difference grid
from a millifarad to a nanofarad gave identical results. The

zyxwvutsrqpo
results in a system of ordinary differential equations (ODE’S).
parallel resistor R shown in Fig. 1 is required to insure a finite
As demonstrated by Hamill [6] systems of ordinary differential
admittance element in the nodal matrix and is simply given a
equations can be integrated by using an array of capacitors
large value so that the RC time constant is much largl-r than
interconnected by voltage controlled current sources. The
the largest time constant of the ODE of interest.
method is illustrated in Fig. 1 where the system of ordinary
As a demonstration the system of equations
differential equations

zyxwvutsrq
is replaced by a simple network of capacitors driven by voltage
controlled current sources. The state variables y j are replace
by node voltage variables e3 and the voltage controlled current
source multipliers are defined by a,k = C * A3k. The value
chosen for the capacitor is in principle unimportant although
values too large or too small may result in numerical errors.
For the SPICE calculations presented here values of C ranging
used by Gear [7] to test stiff ODE solvers was integrated
using the SPICE circuit analysis program [l]. The SPICE
input deck is given in Fig. 14 and can be divided into three
sections. The first section tells the SPICE program which
voltage nodes should be saved for plotting. The second :section
gives the program time step and accuracy control information.
The third section defines the various circuit elements, which

zyxwvut
nodes they are connected to, their values, and for the case
Manuscript received January 1, 1994; revised July 15, 1996.
The author is with Mission Research Corp., Albuquerque, NM 87106 USA. of the voltage controlled current sources the locations of the
Publisher Item Identifier S 001 8-9359(96)08743-2. controlling voltage nodes. The interested reader shoulld refer
0018-9359/96$05.00 0 1996 IEEE
564 zy
zyxwvu
zyxwvu
IEEE TRANSACTIONS ON EDUCATION, VOL. 39, NO. 4, NOVEMBER 1996

2.0

1.5
zyxwvutsrqpo
zyxwvutsrqpon
Gear's P r o b l e m
Comparison of Analytic Result with SPICE
in Fig. 3 where

(a-1)
r
ar = r3
dY
at .
=Yj

g=-
- r3-1

( a - 1)
rj
Yj
resulting in

zyxwvutsrqp
1 .o

x 0.5 y l , numerical

-0.5

-I
0.0

n
,;-dl
zyxwvut
' "';0-32
y2,analytic
y2.numerical

zyxwvutsrqponmlkjihgfe
10-22 10-1 2
P
s

100 2
ULI
101
where y3 is collocated with
equations is written
rj. The resulting system of
time(sec)

Flg. 2. Comparlson of analytical and numerical (SPICE) solutions of (1). (7)

to the SPICE users documentation [l] and work though some where
example problems in order to better understand the input
listing.
The results of the SPICE solution of (2) are compared to
the analytical solution

in Fig. 2. As can be seen the agreement is quite good.
Based on the above discussion it was decided to select some
typical partial differential equations (PDE's) and integrated
them in time using the above approach. The finite difference
schemes chosen are typical but not unique as there are in
principle an arbitrarily large number of finite difference rep-
resentations of a partial differential equation. No attempt was
made to control the time step other than to insure there were
enough points for plotting purposes.
when 'U = 0 Burger's equation reduces to the diffusion
equation.
The first problem examined was thermal diffusion where
one boundary is set to an arbitrary value T3=1= y3=1 = 1
and the rest of the TJ's are set to zero at time t = 0. The
temperature time history as predicted by SPICE at selected
locations is shown in Figs. 4 and 5 for planar and spherical
diffusion, respectively. For the planar problem the temperature
of the left-hand boundary of a 1 = 10-m-thick slab was set to
unity. The analytic solution to this problem is given by [lo]

11. NUMERICALSOLUTION OF (9)

PARTIALDIFFERENTIALEQUATIONS
The first PDE examined was Burger's equation [8]
- + V . (yV) = DV2y
at
which for one spatial dimension can be written in the form
where a = 1 , 2 , 3 for Cartesian, cylindrical, and spherical
geometries and the magnitude of v is constant. Burger's
equation [9] is used to model transport phenomena where the
divergence term in equation [4] gives the rate of change in
y due to spatial variations in flux y'u and the V2 operator in
equation [4] gives the rate of change in y due to diffusion. The
spatial derivatives in equation [5] can be finite differenced on
a uniform spatial mesh T I , . . . , ~ ~ - r1J,,r J + l , . . . , r,, shown
For the spherical problem the temperature at the surface of an
a = 1 m cavity cut into a 10-m sphere was set to unity. The
(4)
analytic solution to this problem is given by [ l l ]
T
(10)
For both problems the grid spacing was set to Ar = 0.1 m and
the diffusion coefficient D was set to unity. As can be seen
in Figs. 4 and 5 the agreement between the numerical results
and the analytic solutions is good. The SPICE input decks for
these examples are given in Figs. 15 and 16.
Next the velocity in (5) was set to 'U = 1.0 m/s and
the advection problem was run. As it is known that the
differencing scheme used here is dispersive and gives rise to
oscillations the diffusion coefficient D was not set to zero
but rather to D = 0.01 m2/s. The propagation of a jump
discontinuity y3=l(t = 0) = 1.0 at rJ=l = 0 through
ZIMMERMAN: TIME DOMAIN SOLUTIONS TO PARTIAL DIFFERENTIAL EQUATIONS USING SPICE zyx
zyxwvut 565

zyxwvutsrqponml
Fig. 3. Method for representing a partial differential equation with a single independent variable on a 1-D grid

Planor Diffusion
Temperature BGUndOry Condition, SPICE
Spherical Diffusion
Temperature Boundary COndltlGn, SPICE

0.9 0.5
0.8

0.7 0.4

0.6
0.3
0.5 +
>
z
k
0.4

zyxwvutsrqponm
0.3 Numerical, x = l m
Numerical. x=2m
0.2 Analyticol, x = l m J
Analytical, x = 2 r r
0.1

0.0, I
"I:/,
0.0 , , , , , , , , : , , , , , , , ,
d

Fig. 4. Comparison of analytical and numerical (SPICE) solutions for the

planar thermal diffusion problem with a T = To temperature boundary
condition at the left boundary.

zyxwvutsrq
B u r g e r s ' Equation
the mesh is shown in Fig. 6 and the SPICE input deck is v = l -0, D=0.01

shown in Fig. 17. Note that a transmitting boundary condition

was incorporated at y3=ns. This was accomplished by means 1 1 -
, \
1 0 - '-
of a voltage controlled voltage source forcing yjyj,nz to be -
I

09
equal to ynz-l. The problem of modeling a cylindrical or 0 8 - l

spherical compression by setting the right-hand boundary to 0 7 -

unity at t = 0 was also examined. The cylindrical case 06 -
x
performed satisfactorily and showed the expected result that y 05 -
,
was proportional to 1/r. The spherical case ran satisfactorily
to t = 6 s and then became unstable as the discontinuity
04
0 3 -
-

zyxwvutsrqponmlkji
,

zyxwvutsr
I
02-
converged on the origin with a l/r2 dependence on radius. I

Many physical problems of interest are described by two

or more partial differential equations. For example thermal
diffusion is often modeled by the continuity equation t(sec)

8T Fig. 6. Propatation of a discontinuity at the left-hand boundary of ;1 slab

pcv-
at
+v .Q = s
thermal flux is related to the temperature gradient by Ficks'
where T(deg) is the temperature, Q(w/m2) is the thermal law
flux, S(w/m3) is the volumetric energy source, p(gm/m3) is
the density, and Cv(Joules/gm-deg) is the specific heat. The Q = -xVT (12)
566

Fig 7
zyxwvutsrqponmlkjih
<
=I-1

I <
zyxw
zy zyx
zyxwvutsrqp
zyxwvuts
R=1Ohm

zyxwvutsr
zyxwvut
R a2
1.

=j

Ahr
IEEE TRANSACTIONS ON EDUCATION, VOL. 39, NO. 4, NOVEMBER 1996

Method for representing the thermal diffusion problem with two independent variables TJ,Q3 on a 1-D grid.

Planar Thermal Diffusion Cylindrical Thermal Diffusion

= I
I_

=j+1
R=lOhm

zyxwvutsrq
Impulse Response, Analytic vs SPICE Impulse Response, Analytic v s SPICE

zyxwvuts
Numerical, r = l .Gm

R
e Analytical, r=l .Om
* Analytic, x = l m Numerical, r=4.0m
Numericol, x=l m Analytical, r=4.0m
Analytic, x=4m
Numerical, x=4m

0.00c- Le
0 10 20 30 40 50 0 2 4 6 8 10 12 14 16 18 20
t(sec) t(sec)

Fig. 8. Comparison of analytical and numerical (SPICE) solutions for the Fig. 9. Comparison of analytical and numerical (SPICE) solutions for the
planar thermal diffusion problem with a unit source at z = 0. cylindrical thermal diffusion problem with a unit impulse source at T = 0.

where X(w/m-deg) is the thermal conductivity. This problem The flux elements are given by ( j odd)
can be finite differenced on a one-dimensional (1-D) grid as
shown in Fig. 7 where y j = Q j for j = 1 , 3 , 5 , . . ., ( j odd) flux, (watts/m2) = Q, , Cartesian (154
and y j = Tj for j = 2 , 4 , 6 , . . , ( j even). By recognizing that flux, (wattdm) = 27rr, Q, , cylindrical (15b)
the finite volume elements bounded by rj+1 and rj-1 have flux, (watts) = 47rr,2Qg, spherical. (15c)
a mass Mj ( j even) and that the divergence operator 0 . Q
represents the difference between the thermal energy flux into

zyxw
The source term is given by ( j even) Sp(w/m2), Sc(w/m),
less the flux out of one of these finite volumes, (11) can be or S, (w) for Cartesian, cylindrical, and spherical coordinate
rewritten (12) ( j even) systems. The resulting system of equations is written ( j even)
,+I
$3 = Xjkyk + s, (16)
k=,-1
where for A = l / p C v , A,,, = 0
The mass elements are given by ( j even) r 1

M,(gm/m 2 ) = p(rg+l - ~ , - 1 ) , Cartesian (144

M,(gm/m) = 7rp(rj2+1-.j”-1), cylindrical (14b)
MJ(gm) = (4?r/3)p(r?+,- ~ j ” - ~ ) spherical.
, (14c) L “
ZIMMERMAN: TIME DOMAIN SOLUTIONS TO PARTIAL DIFFERENTIAL EQUATIONS USING SPICE

j (even)
zyxwvut
zyxwvut
zyx J (odd)
567

- I zyxwvutsrqponmlkjihgfedcbaZYXWV
/-

zyxwvutsrqpo
zyxwvut
.-

and
I.<
zyxwvu
zyxwvut Ar II
Fig. 10. Method for representing the electromagnetic propagation problem with two independent variables E 3 ,H3 on a 1-D grid

P!cnar Electromagnetic Pulse

SPICE

where B = 1,1/71-, and 3/47i for a! = 1 , 2 , or 3. Equation (12)

can be represented by the simple finite difference equation ( j
odd)

Therefore for odd values of j

1 0 1 ' ' ' ' ' ' ' ' ' ' '
OEO 1E-8 2E-8 3E-8 4E-8 5E-8
t(sec)
where (Xj,j = 0) Fig. 11. Electric field propagating to the right of a unit current pulse

where p = Cv = D = 1.0. As can be seen the agreement is

zyxw
good. The SPICE input listings are shown in Figs. 18 and 19.
Equation (20) does not involve a time derivative and so a
As a final example consider Maxwell's equations [51
capacitor is not required in the SPICE model. This is shown in
Fig. 7 where a 1-R resistor is used to force equality between aE
the current and the voltage in the odd-numbered nodes. The E- =VXH- U E - J (234
at
time-dependent response to an impulsive source is shown aH
in Figs. 8 and 9 for planar and cylindrical geometries. The p- = -UXE
at
impulsive source was approximated with a short pulse of du-
ration St. The magnitude of the corresponding SPICE current where E(V/m), H(A/m) are the electric and magnetic fields,
source is given by I(SP1CE) = C(S,, S,, S,)/AVSt where E(F/m), p(H/m), u(S/m) are the permittivity, permeability, and
S,(Joules/m2), S,(Joules/m), S,(Joules) is the magnitude of conductivity of the media, and J(A/m2) is the volumetric
the impulse and AV = A r , r r i , or ( 4 r / 3 ) r r : for Cartesian, current density. By considering the transverse electromagnetic

cylindrical, or spherical problems. The results are compared
with the analytical solution for an impulse at the origin given
by H11, U31
zyxwv
mode (TEM), propagation along the radial direction and
assuming 8 = 90 degrees for the spherical case Maxwells
equations can be written

(22) e-BEZ - BHY ~ - a E Z - J Z , Cartesian

at dX
568 zyxw
zyxwvutsrqponmlkjihgfe F!anar Electromagnetlc Pulse
IEEE TRANSACTIONS ON EDUCATION, VOL 39, NO 4, NOVEMBER 1996

Radiatlng Dipole Field

SPICE SPICE

0.02 r 3r

-0.12'
@E0
1

zyxwvutsrq
' '
1E-8
'
2E-8
' '

t(sec)
'
3E-8
' '
4E-8
' '
5E-8
-31 I
@EO
'
1E-8
' ' '
2E-8
'

t(sec)
'
3E-8
' '
4E-8 5E-8

zyxwv zyxwvutsr
Fig. 12. Magnetic field propagation to the right side of a unit current pulse. Fig. 13. Electric field propagating radially outward from an oscillating
current source at the origin.

6- -
- H4
~
+ aH4
~- aEZ - J Z , cylindrical (24b)
dt r dr

and
6-

P--%-

aH4 -
at

aHY - dEZ

aH4
-

EO dEO
- -_ -
P-dt 7- dr
-
-H_ aH4
' - __

, Cartesian

, spherical.
-
dX
E
dr
r

,cylindrical
~
dr

z -

zyxwvut
a E Z - JO, spherical (24c)

For a uniform grid r,+1 - rJ-l = Ar. Equation (26) can be

represented by(16), j odd, where (AJ,, = - - C / E )

This problem can be differenced on a 1-D grid as shown in

Fig. 10, where yj = E? for j = 1 , 3 , 5 , .+ .,(j odd) and y, =
,(j even). This choice of alternating grid
points for E and H is common in numerical electromagnetics
and can be related back to elementary electromagnetic texts
that relate the time derivative of the magnetic field normal
to an elemental area to the line integral of the electric field
around the elemental area. The electric field grid points are
chosen to be the odd values of j since conductive boundaries
are easily introduced by setting El and E, max to zero.
The electric and magnetic field equations can be finite
differenced on the spatial grid shown in Fig. 10, i.e., ( j odd)
H&+i - H&-1
EEZ, =
r,+1 - rj-1
and ( j even)
p H T = EZ,+l - EZj-1 , Cartesian
r3+1- rj-l
The SPICE current source is given by I(SP1CE) =
CJ(A/m2)/t. Equation (27) can be represented by (7), j
even, where ( X j , J = 0)
Xijp1 -(-
1 1
=
P-
- =3
- aEZ, - J Z , , Cartesian (26a)
As an example a current density pulse with peak magnitude
J = 1 A/m2 and a duration of 10 ns was injected into the
midpoint of a planar grid with cell thickness z = 0.2 m. The
observed electric and magnetic fields propagating to the right
at a distance of x = 1 m and 3 m are shown in Figs. 11 and
12. The ratio and signs of the fields are correct for a pulse
traveling to the right with an impedance Z = E / H = 377
0. The magnitude of the fields can be seen to be correct by
noting that a current density of 1 A/m2 in a cell of thickness
r = 0.2 m results in a current sheet J , = 0.2 A/m.
From Amperes law the magnetic field on each side of this
sheet is J 3 / 2 = 0.1 A/m as shown in Fig. 12. The SPICE
 zyxwvuts
zyx
ZIMMERMAN TIME DOMAIN SOLUTIONS TO PARTIAL DIFFERENTIAL EQUATIONS USING SPICE 569

STATElb

zyxwvutsrqp
*GEARS STIFF PROBLEM
* S PICE-NET
*ALIAS V ( 1)=Y1
*ALIAS V (2) =Y2
.PRINT TRAN V(1) V(2)
.PLOT TRAN V(1) V(2)
.OPTIONS LIMPTS=1000
* PRINT INCRIMENT,TSTOP,TSTART,TMAX,INIT CONDITION
.TRANS 0.01s 2s 0.0s 0.1s U I C
C1 1 0 IMF IC=1
R1 1 0 1E6
c2 2 0 1MF IC=O

zyxwvutsrqponm
R2 2 0 1E6
G1 0 1 POLY(2) 1 0 2 0 0 +0.998 +1.998
G2 0 2 POLY(2) 1 0 2 0 0 -0.999 -1.999
.END
Fig 14 (Also run with INCRIMENT = 0.0001, TSTOP = 0.025, TMAX = 0.001 S to generate points).

STATE2
*BURGERS EQUATION
*SPICE-NET
*ALIAS V(11)=Y1
*ALIAS V(21)=Y2
.PRINT TRAN V(11) V(21)
.PLOT TRAN V(11) V(21)
.OPTIONS LIMPTS=1000 RELTOL=O.O1
*PRINT INCRIMENT,TSTOP,TSTART,TMAX,INITCONDITION
.TRANS 0.1s 10s 0 . 0 s 0.1s UIC
c1 1 0 1MF IC=O

zyxwvuts
R1 1 0 1E9
c2 2 0 1MF IC=O
R2 2 0 1E9
G2 0 2 POLY(3) 1 0 2 0 3 0 0 0.100000 - 0 . 2 0 0 0 0 0 0 . ~ 0 0 0 0 0
c 3 3 0 1MF IC=O
R3 3 0 1E9
G3 0 3 POLY(3) 2 0 3 0 4 0 0 0.100000 -0.200000 0.100000

GlOO 0 100 POLY(3) 99 0 100 0 101 0 0 0.100000 -0.200000 0.100000

ClOl 101 0 1MF IC=O
RlOl 101 0 1E9
V201 1 0 DC 1
R201 1 0 1E3
v202 101 0 E€ 0
R202 101 0 1E3
.
END
Fig. 15.

zyxwvutsrqpon
input deck is shown in Fig. 20. Note that a free space boundary
condition at the left and right sides of the Cartesian mess was

zyxwvutsrqp
incorporated by means of a voltage controlled voltage source
forcing yJZnZ to be equal to 377~,,,,-~ and y1 to be equal
111. CONCLUSION
The above results demonstrate that circuit analysis codes
such as SPICE can produce accurate solutions to systems of
partial differential equations in one spacial dimension. This

zyxwvutsrq
to 377y2. approach should be useful in both an educational setting and

zyxwvutsrq
As a final example a sinusoidal current source of unit
magnitude was placed at the r3=3 grid point of a spherical
grid with Ar = 0.2 m as before. This results in a volumetric
current source Id1 given by 5(47~/3)(r:- r $ ) = 0.109 A-m.
Given an oscillatory frequency f = 150 MHz the expected
radiated electric field at T = 4 and 8 m is found from [5] to
be E, = (IdZ)f/2r = 2.57 and 1.29 V/m, respectively. The
SPICE prediction shown in Fig. 13 is in good agreement with
these values. The SPICE input deck is shown in Fig 21.
for combined problems such as electromagnetic propagation
from and to electrical circuits and thermal effects on circuits.
The input data list can be written by a text editor and
then adapted to the problem of interest. The author used
a simple preprocessor to write the SPICE data list. This
procedure is recommended as it lowers the possibility of a
typographical error in the input data list. In addition once
written a preprocessor can be easily adapted to other circuits
codes as desired.
570 zyxwvutsrqponmlk
zy IEEE TRANSACTIONS ON EDUCATION, VOL. 39, NO. 4, NOVEMBER 1996

STATE2

zyxwvuts
*BURGERS EQUATION
*SPICE-NET
*ALIAS V(11)=Y1

zyxwvuts
*ALIAS V (21) =Y2
.PRINT TRAN V(l1) V(21)

zyxwvutsrq
.PLOT TRAN V(11) V(21)
.OPTIONS LIMPTS=1000 RELTOL=O.Ol
*PRINT INCRIMENT,TSTOP,TSTART,TMAX,INITCONDITION
.TRANS 0.1s 10s 0.0s 0.1s UIC
c 1 1 0 1MF IC=O
R1 1 0 1E9
c2 2 0 1MF IC=O
R2 2 0 1E9
G2 0 2 POLY(3) 1 0 2 0 3 0 0 0.090909 -0.200000 0.109091
c3 3 0 1MF IC=O
R3 3 0 1E9
G3 0 3 POLY(3) 2 0 3 0 4 0 0 0.091667 -0.200000 0.108333

GlOO 0 100 POLY(3) 99 0 100 0 101 0 0 0.099083 -0.200000 0 ,100917

ClOl 101 0 1MF IC=O
R101 101 0 1E9
VlOOl 1 0 DC 1
RlOOl 1 0 1E3
V1002 101 0 DC 0
R1002 101 0 1E3
.Em
Fig. 16.

STATE2
*BURGERS EQUATION
*SPICE-NET
*ALIAS V (21)=Y1
*ALIAS V ( 41) =Y2
*ALIAS V(61)=Y3
*ALIAS V (81)=Y4
.PRINT TRAN V(21) V(41) V(61) V(81)
.PLOT TRAN V(21) V(41) V(61) V(81)

zyxwvutsr
.OPTIONS LIMPTS=1000 RELTOL=O.Ol
*PRINT INCRI"T,TSTOP,TSTART,TMAX,INIT CONDITION
.TRANS 0.1s 10s 0.0s 0.1s UIC
C1 1 0 1MF IC=O
R1 1 0 1E9
c2 2 0 1MF IC=O
R2 2 0 1E9
G2 0 2 POLY(3) 1 0 2 0 3 0 0 0.006000 -0.002000 -0.004000
c3 3 0 1MF IC=O
R3 3 0 1E9
G3 0 3 POLY(3) 2 0 3 0 4 0 0 0.006000 -0.0'02000 -0.004000

GlOO 0 100 POLY(3) 99 0 100 0 101 0 0 0.006000 -0.002000 -0.004000

ClOl 101 0 1MF IC=O
RIO1 101 0 1E9
V201 1 0 DC 1
R201 1 0 1E6
E202 0 101 POLY(1) 0 100 0 1.0
R202 1 0 1E5
.END
Fig. 17.
 zyxwvuts
zyxwvutsr
ZIMMERMAN: TIME DOMAIN SOLUTIONS TO PARTIAL DIFFERENTIAL EQUATIONS USING SPICE zyx 571

STATE2
*THERMAL EQUATION
*SPICE-NET
*ALIAS V(210)=Y1

zyxwvuts
*ALIAS V(220)=Y2
*ALIAS V(230)=Y3
*ALIAS V (240) =Y4
.PRINT T W V(210) V(2ZO) V(230) V(240)
.PLOT TRAN V(210) V(220) V(230) V(240)
.OPTIONS LIMPTF=2000 RELTOL=O.Ol
*PRINT INCRIMENT,TSTOP,TSTART,TMAX,INITCONDITION
.TRANS 0.1s 50s 0.0s 0.1s UIC
R1 1 0 1E6
C2 2 0 1MF IC=O
R2 2 0 1E6
G2 0 2 POLY(3) 1 0 2 0 3 0 0 0.005000 0.000000 -0.005000
R3 3 0 1.0
G3 0 3 POLY(3) 2 0 3 0 4 0 0 5.000000 0.000000 -5.000000
c4 4 0 1MF IC=O
R4 4 0 1E6
G4 0 4 POLY(3) 3 0 4 0 5 0 0 0.005000 0.000000 -0.005000
R5 5 0 1.0
G5 0 5 POLY(3) 4 0 5 0 6 0 0 5 . 0 0 0 0 0 0 0 . 0 0 0 0 0 0 - 5 . 0 0 0 0 0 0

G400 0 400 POLY(3) 399 0 400 0 401 0 0 0.005000 0.000000 -0.005000

R401 401 0 1E6
RlOOl 1 0 1E6
R1002 401 0 1E6
I1 0 200 PULSE 0.0 0.05 0.0 0.01 0.01 0.09
.
END
Fig. 18.

STATE2
*THERMAL EQUATION
*SPICE-NET
*ALIAS V (10)=Y1
*ALIAS V ( 2 0) =Y2
*ALIAS V (30) =Y3
*ALIAS V(40)=Y4
.PRINT TRAN V(10) V(20) V(30) V(40)
.PLOT TRAN V(10) V(20) V(30) V(40)
.OPTIONS LIMPTS=2000 RELTOL=O.Ol
*PRINT INCRIMENT,TSTOP,TSTART,TMAX,INITCONDITION
.TRANS 0.1s 50s 0 . 0 s 0.1s UIC
R1 1 0 1E6
c2 2 0 lMJ? IC=O
R2 2 0 1E6
G2 0 2 POLY(3) 1 0 2 0 3 0 0 0.000000 0.000000 -0.010000
R3 3 0 1.0
G3 0 3 POLY(3) 2 0 3 0 4 0 0 5.000000 0.000000 -5.000000
c4 4 0 1MF IC=O
R4 4 0 1E6 -
G4 0 4 POLY(3) 3 0 4 0 5 0 0 0.003333 0 . 0 0 0 0 0 0 -0.006667
R5 5 0 1.0
G5 0 5 POLY(3) 4 0 5 0 6 0 0 5 . 0 0 0 0 0 0 0.000000 -5.000000

G399 0 399 POLY(3) 398 0 399 0 400 0 0 5 . 0 0 0 0 0 0 0 . 0 0 0 0 0 0 -5.000000

C400 400 0 1MF IC=O
R400 400 0 1E6
G400 0 400 POLY(3) 399 0 400 0 401 0 0 0.004988 0.000000 -0.005013
R401 401 0 1E6
RlOOl 1 0 1E6
R1002 401 0 1E6
I1 0 2 PULSE 0.0 0.07958 0.0 0 .01 0.01 0.09
.
END
Fig. 19.
 zyxwvutsrqponmlkj
zy
512

*SPICE-NET
zyxwvutsr
*ALIAS V ( 61)=Y1
IEEE TRANSACTIONS ON EDUCATION, VOL. 39, NO. 4, NOVEMBER 1996

zyxw
*ALIAS V ( 62) =Y2
*ALIAS V ( 81)=Y3
*ALIAS V (82) =Y4
.PRINT TRAN V(61) V(62) V(81) V(82)
.PLOT TRAN V(61) V(62) V(81) V(82)
.OPTIONS LIMPTS=2000 RELTOL=O.Ol
"PRINT INCRIMENT,TSTOP,TSTART,TMAX, INIT CONDITION
.TRANS O.1NS SONS O.ONS O.1NS UIC
R1 1 0 1E6
c2 2 0 1MF IC=O
R2 2 0 1E6
G2 0 2 POLY(3) 1 0 2 0 3 0 0 -3968.254150 0 . 0 0 0 0 0 0 3968.254150
c3 3 0 1 m
R3 3 0 1E6
G3 0 3 POLY(3) 2 0 3 0 4 0 0 -564716544.000000 -0.000000 564716544.000000
c4 4 0 1MF IC=O
R4 4 0 1E6
G4 0 4 POLY(3) 3 0 4 0 5 0 0 -3968.254150 0.000000 3968.254150
C5 5 0 1MF
R5 5 0 1E6
G5 0 5 POLY(3) 4 0 5 0 6 0 0 -564716544.000000 -0.000000 564716544.000000
,
GlOO 0 100 POLY(3) 99 0 100 0 101 0 0 -3968.254150 0.000000 3968.254150
RlOl 101 0 1E6
R1001 1 0 1.OE+9

E1002 101 0 POLY(1) 100 0 0 -377.0

I1 0 51 P'JLSE 0.0 1.1293+08 0.0 2.5NS 2.5NS 5NS
~ zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
END
Fig. 20.

*SPICE-NET
*ALIAS V(21)=Y1
*ALIAS V(41)=Y2
*ALIAS V(61)=Y3
*ALIAS V (81)=Y4
.PRINT TRAN V(21) V(41) V(61) V(81)
.PLOT TRAN V(21) V(41) V(61) V(81)
.OPTIONS LIMPTS=2000 RELTOL=O.Ol
*PRINT INCRIMENT,TSTOP,TSTART,TMAX,INIT CONDITION

zyxwvutsrq
.TRANS O.1NS 50NS O.ONS O.1NS UIC
R1 1 0 1E6
c2 2 0 1MF IC=O
R2 2 0 1E6
G2 0 2 POLY(3) 1 0 2 0 3 0 0 0.000000 0.000000 -7936.508301
c3 3 0 1MF
R3 3 0 1E6
G 3 0 3 POLY(3) 2 0 3 0 4 0 0 282358272.000000 -0.000000 -847074816.000000
c4 4 0 1MF I C = O
R4 4 0 1E6
G4 0 4 POLY(3) 3 0 4 0 5 0 0 2645.502930 0.000000 -5291.005859
c5 5 0 1MF
R5 5 0 1E6
G5 0 5 POLY(3) 4 0 5 0 6 0 0 423537408.000000 -0.000000 -705895616.000000

G99 0 99 POLY(3) 98 0 99 0 100 0 0 558954112.000000 -0.000000 -570478912.000000

ClOO 100 0 1MF IC=O
RlOO 100 0 1E6
GlOO 0 100 POLY(3) 99 0 100 0 101 0 0 3928.170898 0.000000 -4008.337402
RlOl 101 0 1E6
RIO01 1 0 1.OE+9
R1002 101 0 1.OE+9
El001 1 0 POLY(1) 3 0 0 1.0
E1002 101 0 POLY(1) 100 0 0 377.0
I1 0 3 SIN 0.0 1.1293+08 1.5E+08
.END
Fig 21
 zyxwvu
zyxw
zyxw
ACKNOWLEDGMENT
zyxwvutsr
zyxwvutsrq
ZIMMERMAN: TIME DOMAIN SOLUTIONS TO PARTIAL DIFFERENTIAL EQUATIONS USING SPICE

The SPICE runs were carried out on version 386.1 of uc

Berkeley SPICE 2G.6 from intusoft [ 11. The preprocessor was
written using Borland Turbo C++ Version 3.0. The plots were
prepared using the AXUM program from TriMetrix, Inc.
513

[6] D. C. Hamill, “Learning about chaotic circuits with SPICE,” IEEE Trans.
Educ., vol. 36, no. 1, pp. 28-35, Feb. 1993.
[7] C. W. Gear, Numerical Initial Value Problems in Ordinary Differential
Equations. Englewood Cliffs, NJ: Prentice-Hall, 197 1.
[8] J. H. Ferziger, Numerical Methods for Engineering Application. New
Wiley, 1981,
[9] D. Potter, Computational Physics. London: Wiley, 1973.
-1101- D. R. Pitts and L. E. Sissom. Heat Transfer. New York: McGraw-Hill.

zyxw
1977.
[ 111 M. N. Ozisik, Boundary Layer Problems of Heat Conduction. New
REFERENCES York: Dover, 1968.
[ 121 W. R. Zimmerman, “Finite hydraulic conductivity effects on optimal
[l] L. G. Meares and C. E. Hymowitz, Simulating With Spice, intusoft, 1988. groundwater pumping rates,” Water Resour. Res., vol. 26, no. 12, pp.
[2] G. Kron, “Numerical solution of ordinary and partial differential equa- 2861-2864, Dec. 1990.
tions by means of equivalent circuits,” J. APPl. PhYS., vol. 16, PP. [13] H. S. Carslaw and J. C. Jaeger, Conduction of Heat in Sollds, 2nd ed.
172-186, Mar. 1945. Oxford: Clarendon, 1959.
[3] W. J. Karplus, Analog Simulation. New York: McGraw-Hill, 1958.
[4] W. R. Zimmerman, “Network analogs of Maxwell’s field equations in
one and two dimensions,” IEEE Trans Educ., vol. E-25, no. 1, pp. 4-9,
Feb. 1982.
[5] E. C. Jordan and K. G. Balmain, Electromagnetic Waves and Radiating William R. Zimmerman (M’79), biography not available at the time of
Systems, 2nd ed. Englewood Cliffs, NJ: Prentice-Hall, 1968. publication.