Theory of the Symmetrical Super-Condensed
Node for the TLM Method
Vladica Trenkic, Christos Christopoulos,
Abstract | This paper describes a novel time-domain node
for the TLM method. It has the unique feature of mod-
elling arbitrary inhomogeneous media on a generally graded
rectangular TLM mesh without using stubs. Variations in
material properties and arbitrary aspect ratios of mesh cells
are modelled by allowing dierent characteristic impedances
in a cell, maintaining impulse synchronism throughout the
mesh. The complete theory of the new node is given and its
implementation on a general TLM mesh is discussed. Nu-
merical results for a canonical resonator loaded with dielec-
tric layers are presented for dierent grading cases. Sub-
stantial savings in computer storage and run-time as well
as improved accuracy compared to the uniform mesh are
achieved when an appropriate non-uniform grading of the
TLM mesh is used.
I. Introduction
N CONVENTIONAL TLM schemes 1], 2], the charac-
I teristic impedance of interconnected transmission lines
(also called link lines) is kept constant and equal to Z0 ,
the intrinsic impedance of the background medium. This
ensures that the propagation delay remains constant and
synchronism is maintained throughout the mesh. Varia-
tions in material properties and di erent cell aspect ratios
are accounted for by introducing stubs 2] to slow down the
cients and
propagation of pulses in the mesh.
In order to overcome di
culties imposed by the very un- implementation of the new SSCN in a general rectangular
favourable time steps required in a mesh using a graded
symmetrical condensed node (SCN) 2], the hybrid sym-
metrical condensed node (HSCN) was introduced 3]. In
the HSCN only capacitive stubs are required, since all in-
ductance is modelled in the link lines resulting in three
di erent values of characteristic impedance.
A number of papers discussing the use of di erent char-
acteristic impedances of link lines inside a single cell were
published recently. Hoefer and Sautier 4] have shown that
a 2D-TLM network with the cells of arbitrary aspect ratio
can be modelled by altering the characteristic impedances
of transmission lines. In this scheme, di erent dielectric
layers can be modelled by adjusting the mesh aspect ra-
tio. An extension of this model to the 3D expanded (dis-
tributed) node was shown by Zhang and Hoefer 5]. Al-
though these nodes are capable of modelling a variety of
structures, it is not shown how di erent gradings can be
used over a single dielectric layer and a uniformly graded
mesh over di erent layers.
Manuscript received August 12, 1994 revised December 7, 1994.
This work was supported by the Engineering and Physical Sciences
Research Council, UK.
The authors are with the Department of Electrical and Electronic
Engineering, University of Nottingham, NG7 2RD Nottingham, UK.
IEEE Log Number 9410715.
Member, IEEE, and Trevor M. Benson
In a recently introduced frequency-domain TLM (FD-
TLM) node 6] the characteristic impedance of link lines is
also constant but variations in electrical properties are ac-
counted for by modifying the phase constants of the lines.
However, dispersion in the FD-TLM may be reduced by us-
ing a `distributed' node with transmission lines of di erent
characteristic impedance 7].
The main obstacle in formulating an SCN consisting of
di erent line characteristic impedances is the complexity
of deriving a scattering matrix using the standard ap-
proach of demanding that this matrix is unitary. In 8]
a simple and elegant way of obtaining the scattering ma-
trix was described and this approach has been applied to
the derivation of a Symmetrical Super-Condensed Node
(SSCN) where there are no stubs but six di erent link line
characteristic impedances are necessary. The basic prop-
erties of the SSCN together with information on run-time,
storage, accuracy and propagation were presented in 9] for
a uniform mesh and in 10] for a generally graded mesh.
In the present paper, the complete theory of a general
SSCN for TLM will be given including a compact denition
for the line parameters, an elegant form of the scattering
equations, the optimization of scattering coe
the evaluation of maximum permissible time step. The
TLM mesh containing di erent regions of arbitrary grad-
ings and material properties will be described, including
the connection procedure, modelling of external bound-
aries, calculation of the output for all eld components
and excitation in the mesh. The SSCN TLM algorithm
will be validated by performing the analysis of a canonical
resonator loaded with layers of dielectric.
II. Theoretical development of the SSCN
A symmetrical condensed node is shown in Fig. 1 with
the notation of voltages in the form Visj where i denotes
direction and j polarization of the appropriate transmission
line (i j 2 fx y z g and i 6= j ), while s 2 fn pg indicates
the position of the port on the negative (n) or positive (p)
side assuming the origin of coordinates at the centre of the
node.
The development that follows refers to a cuboid cell of
space with arbitrary dimensions x, y, z and mate-
rial properties " = "0 "r ,  = 0r . The characteris-
tic impedances Zij , distributed capacitances Cij and dis-
tributed inductances Lij of the transmission lines in a node
are marked so that the rst index i shows the direction and
the second index j the polarization of the line (i 6= j ). The
lines with the same direction and polarization at a node
have identical characteristics on either side of the node cen-
 Vypz
L yx
Vypx µ = µ0 µr
L xy ∆y
Vzny
∆y Iz
Vznx

Vxny ∆z
Vxpz ∆x y
x
Vxnz ∆x z

Vxpy
Fig. 3. Modelling inductance in z-direction
Vzpx

Vzpy where the left-hand side of (1) shows the total capacitance
in the y direction contributed by the appropriate trans-
y

x
Vynx mission lines, while the right-hand side shows the total ca-
z Vynz pacitance in the same direction of the block of material
modelled by the TLM cell. Similar equations can be ob-
Fig. 1. Symmetrical condensed node tained for the other two directions and summarized in a
general form for the total capacitance Ck in the k direction
Cik i + Cjk j = " ik j
C zy
(2)
C xy ∆y

ε = ε0 εr
∆z
∆x
∆x
Fig. 2. Modelling capacitance in y-direction
tre. This means that six di erent transmission line char-
acteristic impedances are allowed in a 12-port node.
In the derivation of a TLM node capable of modelling
a 3D space of arbitrary material properties on a gener-
ally graded mesh without using stubs, two basic conditions
must be fullled:
1. The transmission lines in a node fully model the elec-
tromagnetic properties of the block of space repre-
sented by that node.
2. The transmission lines in a node allow the same prop-
agation delay, t, in all directions.
A. Formulation of the System of Equations
The development starts by equating the electrical pa-
rameters of a cell to the transmission line parameters of
the node. The total capacitance in, for example, the y-
direction is illustrated in Fig. 2.
It can be seen from Fig. 2 that the cell's total capacitance
Cy in the y-direction is represented by the distributed ca-
pacitance of two y-polarized transmission lines of the node.
Thus, it follows that
where i j k 2 fx y z g and i 6= j 6= k
The total inductance in, for example, the z -direction is
illustrated in Fig. 3.
∆z It can be seen from Fig. 3 that the cell's total inductance
Lz in the z direction is represented by the distributed in-
y
ductance of two transmission lines contributing to the mag-
z
x
netic eld in the z direction. Thus, it follows that
Lxy x + Lyx y =  xz y (3)
where the left-hand side of (3) shows the total inductance in
the z direction contributed by the appropriate transmission
lines, while the right-hand side shows the total inductance
in the same direction of the block of material modelled by
the TLM cell. Similar equations can be obtained for the
other two directions and summarized in a general form for
the total inductance Lk in the k direction
Lij i + Lji j =  ik j (4)
where i j k 2 fx y z g and i 6= j 6= k
In order to maintain time synchronism in the TLM mesh,
the propagation delay t must be kept constant on all lines
at every node in the mesh. The pulse velocity on the line
ij is
i = p 1
vij =  t (5)
Cij Lij
and the propagation delay t can be expressed as
p
t = i Cij Lij (6)

Cxy x + Czy z = " z y x (1) where indices i j can take all possible combination of x y z
giving a set of six equations.
 The three equations obtained from (2), the three equa- Using (10), (11) and (14), the general equation (6) can
tions obtained from (4) and the six equations obtained also be rewritten in terms of the equivalent cubic cell and
from (6) form a system of twelve equations with the dis- normalized link line parameters as
tributed capacitance and inductance of the six transmission q
lines as the twelve unknowns. This is a non-linear system l = 2i C^ij L^ ij (17)
of equations, but an analytical solution can be found as
shown in the next subsection. We now solve a system of 12 equations (15){(17) with 12
B. Derivation of Link Line Parameters unknowns C^ij and L^ ij .
To express (2) and (4) in a more elegant form, we intro- Taking L^ ij from (17) for all possible combinations of
duce normalized parameters for the link lines. For the pulse i j 2 fx y z g and substituting in (16) we obtain three
polarized in the j direction and propagating in the i direc- equations of the general form
tion, the medium distributed capacitance Cijm , the medium (l)2 + (l)2 = 1
distributed inductance Lmij and the medium characteristic (18)
impedance Zijm are 4C^ij (i)2 4C^ji(j )2

Cijm = Cj = " k (7) Substituting C^ji in (18) using an instance of (15) and
i j dividing (18) by (l)2=4 we obtain
j
Lmij = Lki =   (8) 1 + 1
s m r k
^ ^ = (4l)2 (19)
L Cij (i) (1 ; Cki)(j )
2 2
Zijm = Cijm = "  j
k (9)
ij Substituting triplets (i j k) in (19) with (x y z ), (y z x)
It is convenient to normalize the parameters of the link and (z x y) respectively, we obtain a system of three equa-
line to those of the block of medium, so we dene the nor- tions with three unknowns C^xy , C^yz , C^zx , which has a so-
malized distributed capacitance C^ij , the normalized dis- lution of the form
tributed inductance L^ ij and the normalized characteristic j )2 (k)2 + B
impedance Z^ij of the i-directed j -polarized link line, re- C^ij = 2(i)2(
2 (k)2 (2j=l)2 ; 1] (20)
spectively, as
where p
C^ij = CCmij = C"ij  j
k (10) B = A  A2 ; (xyz l)2 (21)
ij
L^ ij = LLmij = Lij  k
j (11) with
!
ij
s A = (xyz )2 4 ; X 1 (22)
^
Z^ij = ZZijm = L^ij (12) (l)2 s=x y z (s)2
ij Cij
We also introduce the equivalent cubic cell parameter, The other three normalized capacitances, namely C^xz , C^yx
l, which actually represents the dimension of a cubic cell and C^zy , can be found using (15). Note from (21) that
having the same propagation delay t as an arbitrarily there are two sets of solutions for the link line parameters
graded TLM cell 4]. The velocity v of a pulse propagating of the SSCN. Both solutions are physical and either can be
along a transmission line in the SCN is twice the plane used in the denition of the SSCN. The preliminary testing
wave velocity in the medium with properties "  2] and is of the two sets of parameters in the SSCN TLM simulations
given by showed that they are equivalent within the frequency band
v=  l = 2 (13) of interest.
t p" The normalized characteristic impedance Z^ij and the
Thus, l is given by actual characteristic impedance Zij are obtained us-
ing (17),(12) and (9) as
l = p2"t (14)
Z^ij = l^ (23)
2iCij
Multiplying (2) by k=("ij ) and using (10) we obtain
r
C^ik + C^jk = 1 (15) Zij = Z^ij Zijm = ij k ^l " (24)
Multiplying (4) by k=(ij ) and using (11) we ob- 2Cij
tain For the benet of later derivations we also solve the sys-
L^ ij + L^ ji = 1 (16) tem of equations (15){(17) directly in terms of L^ ij .
 Taking C^ij from (17) for all possible combinations of
i j 2 fx y z g and substituting in (15) we obtain three
equations of the general form
(l)2 + (l)2 = 1 (25)
4L^ ik (i)2 4L^ jk (j )2
Substituting L^ jk in (25) using an instance of (16) and
dividing (25) by (l)2=4 we obtain
1 + 1 = 4 (26)
L^ ik (i)2 (1 ; L^ kj )(j )2 (l)2 Fig. 4. Maximum time step for dierent nodes
The system of equations (26) has a similar form to (19).
Moreover, if we introduce the mapping C^ij $ L^ ik for all D. Scattering Properties of the SSCN
possible combinations of i j k 2 fx y z g the two systems
of equations become equivalent, which leads to an impor- Scattering in the symmetrical super-condensed node can
tant general relation for the symmetrical super-condensed be derived from the principles explained in 8], without
node: the need to explicitly enforce the unitary condition on the
C^ij = L^ ik (27) scattering matrix. The reected voltages are described in
which can be used for the optimization of scattering coef- terms of incident voltages and appropriate nodal voltages
cients, as will be shown latter in this section. and loop currents. Using the notation formulated at the
beginning of this section and bearing in mind that the pa-
C. Determination of the Maximum Permissible Time Step rameters of a transmission line are identical on either side
So far, we have obtained the parameters of the link lines of the node centre, the scattering equations can be written
in terms of the linear dimensions of a node and the prop- in the form
erties of the medium, but we have not yet determined the r = Vj + Ik Zij ; V i
Vinj (31)
time step, or propagation delay, t, and accordingly the ipj
equivalent cubic cell parameter l which we have used r
Vipj = Vj ; Ik Zij ; Vinj i (32)
in (20).
Analysing (20){(22) and demanding that the character- where the node voltage Vj is
istic impedance of a link line must be positive when mod-
elling passive media 11], the following condition is imposed Vj = Zkj (Vinj i + V i )+ Zij (V i + V i ) (33)
ipj Z + Z knj kpj
2 2
Z ij + Z kj ij kj
A ; (xyz l) > 0 (28)
and the loop current Ik is
This condition can be used to determine the maximumper-
missible time step at a node. A cubic inequality has to be i ;Vi +Vi ;Vi
Vipj inj jni jpi
solved for (l) , giving one physical solution
2 I k = Zij + Zji (34)
l < lmax =  2  (29) Indices i j k take all possible combinations of x y z giving
C cos 13 arccos(D=C 3 ) a set of 12 scattering equations which completely describe
where the scattering in the SSCN.
s X Using (24) and (17), the scattering coe
cients appear-
C = 43 1 D = 8
ing in formulae (31){(34) can be expressed in terms of the
s=x y z (s) xyz
2
normalized parameters of link lines
The maximum time step is then calculated using (14) Zkj = C^ Zij = L^
l p " Zij + Zkj ij Zij + Zji ij
tmax = max2 (30)
Therefore the node voltage Vj and the loop current Ik can
A comparison of the maximum permissible time step in be expressed in a more compact and physically meaningful
graded meshes for three di erent nodes, namely the stub- form
loaded SCN 2], the HSCN 3] and the SSCN, is made in
Fig. 4. The node spacing in the y- and z -directions is xed Vj = C^ij (Vinji + V i ) + C^kj (V i + V i )
ipj knj kpj (35)
at y = z = l0 , while x varies from l0 to 10l0.
The maximum time step t is given relative to the time ^ i
step t0 used in a uniform mesh with node spacing l0 . Ik = LZij (Vipj ; Vinji +Vi ;Vi )
jni jpi (36)
It is clear from Fig. 4 that the time step allowed in the ij
graded SSCN is consistently higher than for the HSCN and Replacing these expressions in the scattering equations (31)
the stub-loaded SCN. and (32) and using identities L^ ij = C^ik and C^ij = 1 ; C^kj ,
the scattering coe
cients used in the derivation of the orig- Ik
inal 12-port SCN scattering matrix 2] can be conveniently 2 Vij (i)
+ +
2 Vij (i+1)
written as -
Vj
-

aij = 1 ; bij ; dij bij = C^kj Z ij (i) Z ij (i+1)

cij = dij ; bij dij = C^ik
Indices i j in coe
cients a ; d denote the port voltages on Fig. 5. Equivalent Thevenin circuit for nding voltage and current
the j -polarized i-directed line on either side of the node.
The fact that both Vinj and Vipj have the same scattering
coe
cients reects the symmetrical property of the SSCN. C. External Boundaries in the SSCN TLM Mesh
It can be readily checked that the scattering matrix S External boundaries of arbitrary reection coe
cient w
formed of coe
cients described above satises the unitary may be modelled by terminating the link lines at the edge
condition given as STYS = Y where Y is a 12  12 diago- of the problem space with an appropriate load. The equiv-
nal matrix with elements of the characteristic admittances alent link line reection coe
cient, ij , generally di ers
of the 12 link lines. This also proves that the approach for from w and can be found as follows.
obtaining scattering coe
cients for a general TLM node For a pulse polarized in the j direction travelling towards
given in 8] does conserve the energy in nodal scattering, a boundary in the i direction, the characteristic impedance
even though this is not explicitly enforced by initial condi- of the medium is Zijm , and the resistance needed to termi-
tions. nate it in order to give reection coe
cient w is
III. Implementation of the SSCN in a TLM Mesh
R = Zijm 11 + w (37)
We dene the term unique node region as the region of ; w
TLM nodes describing the same material properties ("r r ) The link line reection coe
cient, ij , is found by termi-
on a part of the mesh where the cells are of constant linear nating the link line, of characteristic impedance Zij , with
dimensions (x y z ). this same resistance
A. Time-Step in the SSCN TLM Mesh
ij = R ; Zij (1 + w ) ; Z^ij (1 ; w )
In an actual implementation of the SSCN on a TLM R + Zij = (1 + w ) + Z^ij (1 ; w ) (38)
mesh containing di erent dielectric layers and variable
gradings, the time step in the mesh must be chosen as the where the expression Zij = Z^ij Zijm is used.
smallest of all maximum permissible time steps calculated D. Output in the SSCN TLM Mesh
for separate node regions using formulae (29) and (30). For
example, when modelling two di erent materials, one with Electric and magnetic eld components can be found
background properties ("0 0) and another with the prop- from appropriate values of voltages and currents using
erties ("r "0 , r 0 ) using the uniform mesh (x = y =
z = l0), then we obtain from (29) identical maximum Ej = ; Vjj Hk = Ikk
equivalent cubic cell dimensions equal to l0 , as expected.
However, it can be shown from (30) that the maximumtime Field values at a node can be readily obtained using formu-
steps for the two node regions di er by a factor p"r r and lae which dene nodal voltages (35) and loop currents (36)
the smaller one, corresponding to the background medium, already used in the scattering procedure.
must be used. Accordingly, the equivalent cubic mesh pa- In obtaining the output on link lines, the di erent char-
rameter for the material with properties "r "0 , r 0 be- acteristic impedances on interfaces between two node re-
comes smaller, i.e. l = l0 =p"r r . This shows that gions must be taken into account. Voltages and currents
in the TLM region modelling the medium whose electro- are found by solving the equivalent Thevenin circuit shown
magnetic properties are higher than the background ones, in Fig. 5
pulses travel more slowly, as expected.
Vj = 2 Vij (i)YijY (i()i)++VijY (i(+i +1)1)Yij (i + 1)] (39)
B. Connection in the SSCN TLM Mesh ij ij
Inside a unique node region, the connection proceeds as
in the usual SCN mesh, by simply swapping the pulses Ik = 2ZVij((ii) ++ 1)Z ;(iV+ij (1)i)] (40)
ij ij
between adjacent ports of two neighbouring nodes. This is If the output is required on a link line completely inside
valid since the characteristic impedance of a link line is the the unique node region, i.e. Zij (i) = Zij (i + 1), formu-
same on either side of the node. lae (39) and (40) are simplied to
Care must be taken at the interface between two di erent
node regions where the characteristic impedance changes. Vj = Vij (i) + Vij (i + 1) (41)
A simple two-port junction must be modelled by calculat-
ing appropriate reection and transmission coe
cients, as Ik = Vij (i + Z1) ; Vij (i) (42)
explained in 4]. ij
E. Excitation in the SSCN TLM Mesh a/4

If the excitation is specied in terms of eld quantities, εr

then the equivalent voltage pulses must be found. When a/2
exciting a particular eld component, care must be taken
not to excite others. This can be done if the electric charge a
a/4
and the magnetic ux are applied symmetrically to the
node centre. εr
In exciting the electric eld component Ej , pulses must y a/2

be injected into the mesh to give a total voltage Vj = z

10 a

;Ej j . From expression (35) and the charge symmetry

x

conditions, the following ports need to be excited Fig. 6. Resonator loaded with dielectric layers of "r =4 (a=7.112mm)

Vinj = Vipj = V^j Vknj = Vkpj = V^j a) 40 x 4 b) 20 x 8

4Cij 4Ckj
From (35) and (36) it can be conrmed that all other nodal
voltages and loop currents are zero, therefore only the de-
sired eld component is excited. y

Similarly, for exciting the magnetic eld component Hk , x c) 10 x 16 d) 10 x 12

pulses must be injected into the mesh to give a loop current

Ik = Hk k. Applying symmetry conditions for the mag-
netic ux, it can be easily shown that the following voltage
pulses need to be injected
Vipj = ;Vinj = Vjni = ;Vjpi = Ik^Zij = Ik^Zij
4Lij 4Cik
For exciting eld components on a link line, formu-
lae (39) and (40) with conditions, respectively, Ik = 0 and
Vj = 0 can be used to obtain the values of voltages for the
electric eld Ej excitation
Vij (i) = Vij (i + 1) = V2j (43)
and the magnetic eld Hk excitation
Vij (i) = ; Ik Z2ij (i) (44)
Vij (i + 1) = Ik Zij 2(i + 1) (45)
F. Implementation of Other Features in the SSCN TLM
A number of other features, e.g. internal boundaries,
electrical and magnetic losses, thin features (wires, lms,
apertures), voltage and current sources, etc., can be imple-
mented in the SSCN TLM. Procedures similar to the ones
normally used in the SCN mesh can also be used in the
SSCN mesh. Attention must be paid to the fact that the
characteristic impedances of link lines in SSCN TLM di er
in di erent node regions and this should be dealt with in
the implementation in a manner similar to the one used in
the denition of outputs and excitation in the SSCN TLM
mesh, described earlier in this section.
IV. Numerical Results
To validate the proposed TLM SSCN algorithm, we have
modelled a canonical resonator loaded with dielectric, as
Fig. 7. Schematic of the dierent gradings of the TLM mesh
illustrated in Fig. 6. Dielectric layers with relative per-
mittivity "r = 4 are placed at the top and bottom of the
cavity which is otherwise lled with free space ("r0 = 1).
Dimension a of the resonator is a = 7:112mm.
The benchmark resonant frequency for the TE110 mode,
obtained by using two di erent TLM methods on a very
ne uniform mesh, namely the SCN 2] and the SSCN,
is f0 = 16:595GHz. Using the SSCN TLM method on
a ne uniform mesh of 400  40  1 cells we have ob-
tained f = 16:584GHz, which is 0.07% lower value than f0 .
Note that appropriate excitation and short-circuit bound-
ary conditions were used to reduce the number of cells in
the z -direction, since for this particular mode there is no
propagation in the z -direction.
We then used a coarse uniform mesh with 40  4  1
cells, part of which is illustrated in Fig. 7a and obtained
a frequency of 15:480GHz, underestimating f0 by 6:72%.
This inaccuracy occurred mainly because only one node
per dielectric layer was used in the y-direction. The time
step used in this simulation was t0 = 2:965ps.
Retaining the same number of cells, by increasing the
node dimension in the x-direction by a factor of 2 and de-
creasing it by the same factor in the y-direction, as illus-
trated in Fig. 7b, a uniformly graded mesh of 20  8  1 cells
was used and the resonant frequency obtained was 16:322
GHz, i.e. 1:65% below the correct one. The time step was
slightly decreased (0:899t0), but the number of cells and
memory requirements remained the same and a more ac-
curate result was achieved than in the case of the cubic cell
mesh.
Further increase in aspect ratio, illustrated in Fig. 7c
produced a mesh of dimension 10  16  1 cells and the
 TABLE I SSCN. The standard SCN properties may be obtained as
Calculation of TE110 frequency for different grading cases a special case of the SSCN described here, by substitut-
(f0 = 16:595GHz) ing C^ij = 0:5 and Z^ij = 1 for all possible combinations
of indices i j 2 fx y z g. The scattering matrix for the
Mesh Freq.GHz] Error %] Time-Step per t0
SSCN may be represented by only three quantities, e.g.
U: 400 40 16.584 0.07 0.100 C^xy C^yz C^zx, by exploiting equation (15). The output,
U: 40 4 15.480 6.72 1.000 excitation and boundary conditions in the SSCN TLM
G: 20 8 16.322 1.65 0.899 must be treated especially to account for the fact that line
G: 12 12 16.476 0.72 0.645 impedances are not always the same.
G: 10 16 16.530 0.39 0.491
N: 10 12 16.584 0.07 0.938 The SSCN may be operated at a considerably higher
N: 20 24 16.594 0.01 0.469 time step than other nodes. The numerical example for a
U { uniform mesh, G { uniformly graded mesh, N { nonuniformly
graded mesh
canonical resonator lled with dielectric layers reected the
versatility of the SSCN TLM scheme and showed its superi-
ority in terms of computational requirements and accuracy.
Further comparisons of this new implementation with other
simulated resonant frequency of 16:530GHz was just 0.39% conventional implementations of the TLM method will be
below the correct one. However, even though the same described in a future publication.
number of cells was used, the time step was decreased to
0:491t0.
So far, we have used only uniformly graded meshes to References
achieve the improvement in the modelled resonant fre- 1] S. Akhtarzad and P. B. Johns, \The solution of Maxwell's equa-
quency. Consider now a grading case when the grading tions in three space dimensions and time by the TLM method
of numerical analysis", Proc. Inst. Elec. Eng., vol. 122, no. 12,
in the y-direction within the dielectric is higher than in pp. 1344{1348, Dec. 1975.
the region of freepspace. Since the propagation velocity in 2] P. B. Johns, \A symmetrical condensed node for the TLM
the dielectric is "r = 2 times lower than in free space, method", IEEE Trans., vol. MTT-35, no. 4, pp. 370{377, Apr.
we choose to grade that part of the mesh with twice as 3] 1987. R. A. Scaramuzza and A. J. Lowery, \Hybrid symmetrical con-
many cells. However, to maintain a reasonable number densed node for TLM method", Electronics Letters, vol. 26, no.
of cells, we increase node spacing in the free-space region, 4] 23, W.
pp. 1947{1949, Nov. 1990.
J. R. Hoefer and P. Sautier, \Characteristics of the general
thus forming a mesh of 10  12  1 cells, part of which is rectangular 2-D TLM network", Int. Journ. of Num. Modelling,
shown in Fig. 7d. The number of cells for this mesh was vol. 7, no. 2, pp. 127{139, March 1994.
25% smaller then in the previous case, the allowable time 5] Q. Zhang and W. J. R. Hoefer, \Characteristics of 3-D dis-
step was almost twice as high, namely 0:938t0, and the tributed node TLM mesh with cells of arbitrary aspect ratio",
in IEEE Intl. Microwave Symp. Dig., San Diego, CA, USA, pp.
modelled frequency obtained was f = 16:584GHz. This ac- 369{372, May 23-27 1994.
curacy was the same as that achieved with a ne uniform 6] D. P. Johns, A. J. Wlodarczyk, A. Mallik, and C. Christopou-
mesh (400  40  1 cells). However the number of TLM los, \New TLM technique for steady-state
eld solution in three
dimensions", Electronics Letters, vol. 28, no. 18, pp. 1692{1694,
cells used in the non-uniformly graded SSCN mesh was August 1992.
over hundred times smaller and the time step was almost 7] D. P. Johns, \An improved node for frequency-domain TLM {
10 times higher than in the uniform case. By increasing the `Distributed Node'", Electronics Letters, vol. 30, no. 6, pp.
the resolution of the non-uniformly graded SSCN mesh, 8] 500{502, March 1994.
V. Trenkic, C. Christopoulos, and T. M. Benson, \Simple and
by a factor of 2, i.e. using 20  24  1 cells, we obtained elegant formulation of scattering in TLM nodes", Electronics
f = 16:594GHz, an accuracy of better then 10;4. Letters, vol. 29, no. 18, pp. 1651{1652, Sept. 1993.
The summary of the results obtained for the di erent 9] V.Trenkic, C. Christopoulos, and T. M. Benson, \New symmet-
rical super-condensed node for the TLM method", Electronics
mesh congurations used is shown in Table I. It is clear Letters, vol. 30, no. 4, pp. 329{330, Feb. 1994.
from Table I that a uniformly graded mesh o ers better re- 10] V.Trenkic, C. Christopoulos, and T. M. Benson, \Gener-
sults with less memory requirements than a uniform mesh, ally graded TLM mesh using the symmetrical super-condensed
node", Electronics Letters, vol. 30, no. 10, pp. 795{797, May
but that a non-uniformly graded mesh chosen in an appro- 1994.
priate manner shows even better characteristics. 11] D. A. Al-Mukhtar and J. E. Sitch, \Transmission-line matrix
method with irregularly graded space", IEE Proc., part H, vol.
V. Conclusion 128, no. 6, pp. 299{305, December 1981.
A substantial development to the TLM method has been
described, based on the symmetricalsuper-condensed node.
The parameters and scattering properties of the node, its
implementation including the maximum permissible time
step, connection, boundaries, excitation and output, were
described.
The use of normalized quantities presented in this paper
enhances physical understanding and simplies the deriva-
tion of line parameters and scattering properties of the
 Vladica Trenkic was born in Aleksinac, Yu-
goslavia, in 1968. He received the Dipl. Ing.
rst-class honours degree in physics and the
degree in electrical engineering with computer
science from the University of Nis, Yugoslavia,
in 1992, with an average mark 9.94 out of 10.
Since then he has been a Research Assistant
Trevor M. Benson, born in 1958, received
a
Clark Prize in Experimental Physics from the
University of Sheeld in 1979 and a Ph.D.
from the same university in 1982. After spend-
ing over six years as a lecturer at University
in the Department of Electrical and Electronic College, Cardi, Dr Benson joined the Univer-
Engineering at the University of Nottingham, sity of Nottingham as a Senior Lecturer in Elec-
England, where he is currently reading for a trical and Electronic Engineering in 1989 and
Ph.D. His research interests include numeri- was promoted to the post of Reader in 1994.
cal modelling using the transmission line mod- His current research interests include experi-
elling method and its implementation to electromagnetic compatibil- mental and numerical studies of electromagnetic
elds and waves with
ity and microwave heating problems. particular emphasis on propagation in optical waveguides and elec-
tromagnetic compatibility.

Christos Christopoulos (M'92) was born

in Patras, Greece, in 1946. He received the
Diploma in electrical and mechanical engineer-
ing from the National Technical University of
Athens in 1969, and the M.Sc. and D.Phil.
from the University of Sussex in 1970 and 1975,
respectively. In 1974, he joined the Arc Re-
search Project of the University of Liverpool
and spent two years working on vacuum arcs
and breakdown while on attachment to the
UKAEA Culham Laboratories. In 1976, he
joined the University of Durham as a Senior Demonstrator in Electri-
cal Engineering Science. In October 1978, he joined the Department
of Electrical and Electronic Engineering, University of Nottingham,
where he is now Professor of Electrical Engineering. His research in-
terests are in electrical discharges and plasmas, electromagnetic com-
patibility, electromagnetics, and protection and simulation of power
networks.