06d

15. S. D.
Gedney, An anisotropic perfectly matched layer absorb-

ing medium for the truncation of FDTD lattices, IEEE Trans.
Anten. Propag. 44:16301639 (1996).
16. O. M. Ramahi, The concurrent complementary operators
method for FDTD mesh truncation, IEEE Trans. Anten. Prop-
ag. 46:14751482 (1998).
17. M. Krumpholz and L. P. B. Katehi, MRTD: New time-domain
schemes based on multiresolution analysis, IEEE Trans.
Microwave Theory Tech. 44:555572 (1996).
18. Q. H. Liu, The pseudospectral time-domain (PSTD) method: A
new algorithm for solutions of Maxwells equations, Proc.
IEEE Antennas & Propagation Society Int. Symp., 1997,
Vol. 1, pp. 122125.
19. D. F. Kelley and R. J. Luebbers, Piecewise linear recursive
convolution for dispersive media using FDTD, IEEE Trans.
Anten. Propag. 44:792797 (1996).
20. R. M. Joseph, S. C. Hagness, and A. Taflove, Direct time in-
tegration of Maxwells equations in linear dispersive media
with absorption for scattering and propagation of femtosecond
electromagnetic pulses, Opt. Lett. 16: 14121414 (1991).
21. D. M. Sullivan, Z transform theory and the FDTD method,
IEEE Trans. Anten. Propag. 44:2834 (1996).
22. R. M. Joseph and A. Taflove, FDTD Maxwells equations mod-
els for nonlinear electrodynamics and optics, IEEE Trans.
Anten. Propag. 45:364374 (1997).
23. Y. Liu, Fourier analysis of numerical algorithms for the Max-
wells equations, J. Comput. Phys. 124:396416 (1996).
24. T. Namiki, A new FDTD algorithm based on alternating-
direction implicit method, IEEE Trans. Microwave Theory
Tech. 47:20032007 (1999).
25. F. Zheng, Z. Chen, and J. Zhang, Toward the development of a
three-dimensional unconditionally stable nite-difference
time-domain method, IEEE Trans. Microwave Theory
Tech. 48:15501558 (2000).
26. H. A. De Raedt, K. F. L. Michielsen, J. S. Kole, and M. T. Fig-
ge, One-step algorithm to solve the time-dependent Maxwell
equations, Phys. Rev. E 67:056706 (2003).
27. H. A. De Raedt, K. F. L. Michielsen, J. S. Kole, and M. T. Fig-
ge, Solving the Maxwell equations by the Chebyshev method:
A one-step nite-difference time-domain algorithm, IEEE
Trans. Anten. Propag. 51:31503155 (2003).
28. S. Gonzalez Garcia, T. W. Lee, and S. C. Hagness, On the ac-
curacy of the ADI-FDTD method, IEEE Anten. Wireless Prop-
ag. Lett. 1:3134 (2002).
29. R. Holland and J. Williams, Total-eld versus scattered-eld
nite-difference, IEEE Trans. Nucl. Sci. 30:45834587 (1983).
30. D. S. Katz, E. T. Thiele, and A. Taflove, Validation and ex-
tension to three-dimensions of the Berenger PML absorbing
boundary condition for FDTD meshes, IEEE Microwave
Guided Wave Lett. 4:268270 (1994).
31. J. W. Nehrbass, J. F. Lee, and R. Lee, Stability analysis for
perfectly matched layered absorbers, Electromagnetics
16:385389 (1996).
32. S. Abarbanel and D. Gottlieb, A mathematical analysis of the
PML method, J. Comput. Phys. 134:357363 (1997).
33. G. Waldschmidt and A. Taflove, The determination of the ef-
fective radius of a lamentary source in the FDTD mesh,
IEEE Microwave Guided Wave Lett. 10:217219 (2000).
34. C. L. Wagner and J. B. Schneider, Divergent elds, charge,
and capacitance in FDTD simulations, IEEE Trans. Micro-
wave Theory Tech. 46: 21312136 (1998).
35. R. Holland, Threde: A free-eld EMP coupling and scattering
code, IEEE Trans. Nucl. Sci. 24:24162421 (1977).
36. R. Holland, L. Simpson, and K. S. Kunz, Finite-difference
analysis of EMP coupling to lossy dielectric structures, IEEE
Trans. Electromagn. Compat. 22:203209 (1980).
37. C. A. Balanis, Advanced Engineering Electromagnetics,
Wiley, New York, 1989.
38. K. R. Umashankar and A. Taflove, A novel method to analyze
electromagnetic scattering of complex objects, IEEE Trans.
Electromagn. Compat. 24:397405 (1982).
39. R. J. Luebbers, K. S. Kunz, M. Schneider, and F. Hunsberger,
A nite-difference time-domain near zone to far zone trans-
formation, IEEE Trans. Anten. Propag. 39:429433 (1991).
40. J. J. Simpson and A. Taflove, Three-dimensional FDTD mod-
eling of impulsive ELF propagation about the Earth-sphere,
IEEE Trans. Anten. Propag. 52:443451 (2004); also (online)
http://www.ece.northwestern.edu/ecefaculty/taflove/
3Dmovietext@gif.avi
41. Modeling example provided by M. J. Piket-May, University of
ColoradoBoulder, Boulder, CO.
42. Modeling example provided by J. G. Maloney, Georgia Tech
Research Institute, Atlanta, GA.
43. E. J. Bond, X. Li, S. C. Hagness, and B. D. Van Veen, Micro-
wave imaging via space-time beamforming for early detection
of breast cancer, IEEE Trans. Anten. Propag. 51:16901705
(2003).
44. Graphic provided by S.-T. Ho, Northwestern University,
Evanston, IL.
45. S. C. Hagness, D. Razadeh, S. T. Ho, and A. Taflove, FDTD
microcavity simulations: Design and experimental realization
of waveguide-coupled single-mode ring and whispering-gal-
lery-mode disk resonators, J. Lightwave Technol. 15:2154
2165 (1997).
46. O. Painter, R. K. Lee, A. Scherer, A. Yariv, J. D. OBrien,
P. D. Dapkus, and I. Kim, Two-dimensional photonic band-gap
defect mode laser, Science 284:18191821 (1999).
47. S.-H. Chang, FDTD Modeling of Light Propagation and Cor-
relations in Active Random Media, Ph.D. dissertation, North-
western Univ., Evanston, IL, 2004.
FINITE ELEMENT ANALYSIS
JIAN-MING JIN
MATTHYS M. BOTHA
University of Illinois at
UrbanaChampaign
Urbana, Illinois
1. INTRODUCTION
The nite-element method is a numerical procedure used
to obtain approximate solutions to boundary value prob-
lems of mathematical physics. The method was rst pro-
posed by Courant in 1943 to solve variational problems in
potential theory [1]. Thereafter, the method was developed
and applied extensively to problems of structural analysis
and increasingly to problems in other elds. Today, the -
nite-element method is recognized as a general, preemi-
nent method applicable to a wide variety of engineering
and mathematical problems, including those in RF and
microwave engineering.
FINITE ELEMENT ANALYSIS 1589
Previous Page
The rst application of the nite-element method to
microwave engineering and electromagnetics appeared in
1969, when Silvester employed it to analyze wave propa-
gation in a hollow waveguide [2]. The importance of the
method was quickly recognized, and successful applica-
tions were achieved for the analysis of electrostatic, mag-
netostatic, and dielectric-loaded waveguide problems. In
1974, Mei developed a technique that combined the nite-
element method with eigenfunction expansion to deal with
open-region electromagnetic problems such as antenna
and scattering analysis [3]. In 1982, Marin developed an
alternative method to deal with open-region scattering
problems, which combined the nite-element method and
a boundary integral equation [4].
An important breakthrough in the nite-element anal-
ysis of vector eld problems occurred in the 1980s with the
development of edge-based vector elements [57]. These
new elements accurately model the nature of the electric
and magnetic elds and eliminate much of the problems
associated with traditional node-based scalar elements.
Since the development of the vector elements, the nite-
element method has become a very powerful numerical
technique in computational electromagnetics. Today, the
method is used as a major design tool for RF and micro-
wave devices. Its basic principle and a variety of applica-
tions have been described in many books, such as those by
Silvester and Ferrari [8], Jin [9], and Volakis et al. [10].
2. FINITE-ELEMENT ANALYSIS OF SCALAR
POTENTIAL PROBLEMS
To illustrate the basic principle of the nite-element meth-
od, consider the problem of calculating the static electric
potential f due to electric charge density r distributed in
domain O. The domain can be either two- or three-dimen-
sional and is lled with a medium having a permittivity e.
On the basis of electromagnetic theory, the problem is to
solve the following Poisson equation
r
.
erf r on O 1
subject to given boundary conditions. Typical boundary
conditions include the Dirichlet type, which species the
value of the potential on the boundary, and the Neumann
type, which prescribes the normal derivative of the poten-
tial. To illustrate the treatment of both boundary condi-
tions, the following boundary conditions are assumed for
this example
ff
D
on G
D
2
^ nn
.
erf k
N
on G
N
3
where f
D
denotes the specied value of the potential on
the Dirichlet boundary G
D
and k
N
denotes the prescribed
value for the normal derivative of the potential on the
Neumann boundary G
N
. The entire boundary of domain O
consists of G
D
and G
N
and is denoted by G.
The boundary value problem expressed by Eqs. (1)(3)
is usually too complex to allow a closed-form solution. This
is particularly true when O has an irregular shape and the
medium, characterized by e, is inhomogeneous. In this
case, a numerical solution is the only alternative. Among
numerical methods, the nite-element method is very
powerful because of its ability to deal with arbitrary
shapes and inhomogeneous media. The basic principle of
the nite-element method is to divide the entire domain
into many small subdomains, called nite-elements, and
then seek an approximate solution in each of the subdo-
mains. The commonly used subdomains are triangular el-
ements in two dimensions and tetrahedral elements in
three dimensions (Fig. 1) because of their exibility in
modeling complex shapes, although other elements can
also be used for special problems.
Once the domain O is divided into small elements, the
potential in each element can be approximated as a simple
function, such as linear, quadratic, and cubic functions.
This approximation can be obtained by interpolating the
potential values at a set of discrete points on the element.
For example, the potential in a triangular element (Fig. 2)
can be approximated as
f
e
x; y N
e
1
x; yf
e
1
N
e
2
x; yf
e
2
N
e
3
x; yf
e
3
4
where f
e
1
, f
e
2
and f
e
3
denote the values of the potential
at nodes 1, 2, and 3, respectively, and N
e
1
x; y, N
e
2
x; y,
and N
e
3
x; y are the corresponding interpolation functions.
These interpolation functions are completely determined
by the coordinates of the three nodes. The interpolation
functions, also known as basis functions or expansion
functions, determine the order of the elements. An impor-
tant requirement for the interpolation functions is
that they guarantee the continuity of the interpolated
potential.
When the potential in each element is interpolated us-
ing its values at discrete points, the potential in the entire
Figure 1. Finite-element mesh with (a) triangular and (b) tetra-
hedral elements.
1590 FINITE ELEMENT ANALYSIS
domain can be expressed as
f
X
N
i 1
N
i
f
i
X
N
D
i 1
N
D
i
f
D
i
5
where N denotes the total number of nodes, at which the
potential is unknown, and N
D
denotes the number of
nodes on G
D
, where the potential is as given by (2). Fur-
thermore, f
i
denotes the value of the potential at node i
and N
i
is the corresponding interpolation or basis function
(the superscript D denotes the quantities on G
D
). This
interpolation function consists of the corresponding inter-
polation functions in all the elements that are directly
connected to the associated node. Figure 3 shows the lin-
ear interpolation function in a two-dimensional triangular
mesh.
Expression (5) can now be substituted into (1) to deter-
mine the unknown potentials f
i
, using the method of
weighted residuals. It should be noted that the method
of weighted residuals is only one of the methods for the
nite-element formulation. Another commonly used meth-
od is the so-called variational method, which starts from
the variational representation of the boundary value prob-
lem dened in (1)(3). The method of weighted residuals is
used here for its simplicity. In this method, a suitable
testing or weighting function w is applied to (1) and the
resulting equation is integrated over the problem domain,
yielding
Z
O
r
.
erf wdV
Z
O
rwdV 6
Using the well-known vector identity r
.
werf
r
.
erfwerf
.
rw and Gauss theorem
Z
O
r
.
f dO
I
G
b nn
.
f dG 7
(6) can be written as
Z
O
erf
.
rwdO
Z
O
rwdO
I
G
b nn
.
erfwdG 8
where one of the del operators is now transferred from f to
w. Substituting (3) into (8), one obtains
Z
O
erf
.
rw dO
Z
O
rwdO
Z
G
D
^ nn
.
erf wdG
Z
G
N
k
N
w dG
9
This equation is called the weak-form representation of
the boundary value problem dened by (1)(3). The corre-
sponding solution is called the weak-form solution, which
satises (1) in the weighted average sense.
Next, a suitable weighting function needs to be chosen
for w. A popular choice is
wN
i
i 1; 2; :::; N 10
where N
i
is the interpolation function associated with un-
known f
i
. The resulting formulation is known as Galer-
kins formulation. Doing so, one obtains
X
N
j 1
f
j
Z
O
erN
i
.
rN
j
dO
Z
O
rN
i
dO
Z
G
N
k
N
N
i
dG
X
N
D
j 1
f
D
j
Z
O
erN
i
.
rN
D
j
dO
11
and it is noted that the integral over G
D
in (9) disappears
here since N
i
vanishes on G
D
. Equation (11) can be written
more compactly as
X
N
j 1
K
ij
f
j
b
i
i 1; 2; . . . ; N 12
where
K
ij
Z
O
erN
i
.
rN
j
dO 13
b
i
Z
O
rN
i
dO
Z
G
N
k
N
N
i
dG
X
N
D
j 1
f
D
j
Z
O
erN
i
.
rN
D
j
dO
14
1
node i
Figure 3. Basis function N
i
for linear triangular elements. (After
Jin [9], r2002 Wiley.)
1
2
3
e
Figure 2. Linear triangular element.
Equation (12) represents a set of linear equations, which
can also be written in matrix form as
Kffg fbg 15
where [K] is an NN square and symmetric matrix, ffg is
an N1 column vector containing the unknown potential
values at all the nodes except those on G
D
, and fbg is the
N1 known vector contributed by the known charge den-
sity and the Dirichlet and Neumann boundary conditions.
The matrix equation (15) can be solved using one of the
many standard matrix solvers, and its solution yields the
potential values at all the nodes. The potential elsewhere
can then be obtained by interpolation in (5).
Avery important property of the nite-element method
is that its matrix [K] is extremely sparse. This becomes
obvious from (13), which is nonzero only when N
i
and N
j
overlap with each other. Since N
i
is nonzero only within
the elements directly connected to node i, it overlaps with
N
j
only when nodes i and j belong to the same element.
Hence, in each row of matrix [K], there are only a few
nonzero elements no matter how large the dimension of
the matrix is. Therefore, the memory required to store [K]
is proportional to O(N), and the solution can also be made
very efcient by using special solvers that exploit the
property of sparsity. Such solvers are often referred to as
sparse solvers. As a result, the nite-element method is
very suitable for large-scale applications, where one has to
deal with a very large number of unknowns.
In the actual implementation of the nite-element
method described above, it might prove difcult to nd
the explicit expressions of N
i
and N
j
to evaluate K
ij
, since a
node may be connected to a different number of elements,
each having a different shape. To alleviate this difculty,
(13) is rewritten as
K
ij
X
M
e 1
Z
O
e
erN
i
.
rN
j
dO 16
where O
e
denotes the domain of element e and M denotes
the total number of elements in O. Using (16), one can
process elements one by one and calculate its contribution
to [K]. This process, called assembly, is described in detail
in Ref. 9 using a specific example.
3. FINITE-ELEMENT ANALYSIS OF VECTOR
FIELD PROBLEMS
The nite-element method described in Section 2 can be
extended to deal with problems involving vector elds. For
this, consider the problem of calculating the electric eld
intensity E due to electric current density J in a domain O
characterized by relative permittivity e
r
and relative per-
meability m
r
. Maxwells equations for this case can be con-
verted into the vector Helmholtz equation
r
1
m
r
rE

k
2
0
e
r
E jk
0
Z
0
J on O 17
where k
0
denotes the free-space wavenumber and Z
0
de-
notes the free-space intrinsic impedance. Typical bound-
ary conditions for the electric eld include the
homogeneous Dirichlet condition on a perfectly conduct-
ing surface and the mixed boundary condition on an im-
pedance surface. To illustrate the treatment of these two
conditions, the following are assumed for this example
^ nnE0 on G
D
18
^ nn
1
m
r
rE

jk
0
Z
^ nn ^ nnE K
N
on G
N
19
where Z denotes the normalized surface impedance on G
N
and K
N
is a known function introduced to make (19) more
general.
Similar to the scalar problem discussed in Section 2,
instead of solving the boundary value problem (17)(19)
directly, one can seek its weak-form solution by multiply-
ing (17) by an appropriate weighting function W and
integrating over the problem domain, which yields
Z
O
W
.
r
1
m
r
rE

k
2
0
e
r
E
!
dO
jk
0
Z
0
Z
O
W
.
JdO
20
By invoking the vector identity
W
.
r
1
m
r
rE
!
r
.
W
1
m
r
rE
!
1
m
r
rW
.
rE
21
and Gauss theorem (7), one obtains
Z
O
1
m
r
rW
.
rE k
2
0
e
r
E
.
W
!
dO
Z
G
D
1
m
r
rW
.
^ nnE dG
Z
G
N
jk
0
Z
^ nnW
.
^ nnE K
N
.
W
!
dG
jk
0
Z
0
Z
O
J
.
WdO
22
after the boundary condition (19) is applied.
To nd a numerical solution of (22) using the nite-
element method, the entire domain O is rst divided into
small elements. Within each small element, E can be in-
terpolated using a set of discrete values. One approach is
to assign E at a few points on the element and then in-
terpolate E elsewhere using a set of scalar interpolation
functions. This approach turns out to be very problematic
because of the difculty in applying correct boundary con-
ditions to the interpolated eld E. A better approach is to
assign the tangential component of E at each edge of the
element and then interpolate E elsewhere using a set of
vector basis functions. For example, the eld in a trian-
gular element can be interpolated as
E
e
x; y N
e
1
x; yE
e
1
N
e
2
x; yE
e
2
N
e
3
x; yE
e
3
23
where E
e
1
, E
e
2
, and E
e
3
denote the tangential component of
E at edges 1, 2, and 3, respectively, and N
e
1
x; y, N
e
2
x; y,
and N
e
3
x; y are the corresponding interpolation or basis
functions. Different from those in (4), these functions are
vector functions and the corresponding element is called
the vector element or edge element, in contrast to the pre-
vious scalar element or nodal element. Figure 4 shows the
vector basis functions for a triangular element. Clearly,
such basis functions ensure the tangential continuity of
the interpolated eld while allowing the normal compo-
nent to be discontinuous. Hence, they accurately model
the nature of the vector eld E.
When the eld E is interpolated in each element using
its tangential values at the edges of the element, the eld
E in the entire domain O can be expressed as
E
X
N
e
i 1
N
i
E
i
24
where N
e
denotes the total number of edges excluding
those on G
D
, E
i
denotes the tangential component of E at
the ith edge, and N
i
is the corresponding basis function.
Obviously, for an edge inside O, N
i
spans over two neigh-
boring elements that share the common edge, and for an
edge on G
N
, N
i
spans over only one element. Figure 5
shows N
i
for an interior edge.
By substituting (24) into (22) and using the same N
i
as
the weighting function W, one obtains
X
N
e
j 1
K
ij
E
j
b
i
i 1; 2; . . . ; N
e
25
where
K
ij
Z
O
1
m
r
rN
i
.
rN
j
k
2
0
e
r
N
i
.
N
j
!
dO
jk
0
Z
G
N
1
Z
^ nnN
i
.
^ nnN
j
!
dG
26
b
i
jk
0
Z
0
Z
O
J
.
N
i
dO
Z
G
N
K
N
.
N
i
dG 27
Note that the integral over G
D
in (22) disappears here
since ^ nnN
i
0 on G
D
. Equation (25) can be written com-
pactly as
KfEg fbg 28
which can be solved for fEg. Similar to the scalar case, [K]
is a sparse and symmetric matrix, which can be solved ef-
ciently using a sparse solver. Once fEg is obtained, the
eld everywhere in O can be calculated using (24).
4. COMPUTATIONAL ASPECTS
When solving a problem using the nite-element method,
one rst sets up the mesh, then lls the matrix equation
based on the mesh together with the boundary conditions,
and nally solves the resulting system of linear equations.
The computational issues of mesh generation and matrix
solvers are discussed in this section.
4.1. Mesh Generation
The nite-element method assumes the availability of a
mesh on the structure under consideration. However, ob-
taining this mesh is not trivial. Finite-element meshes can
be split into two main categories: structured and unstruc-
tured meshes. Structured meshes have the property that
all elements have the same number of neighbors, while
unstructured meshed have no such constraint. When us-
ing rectangular or brick elements, structured meshes usu-
ally result, while unstructured meshes most often consist
of triangles or tetrahedrons. Unstructured meshing with
Figure 4. Vector basis functions N
e
i
for a lin-
ear triangular element. (After Jin [9], r2002
Wiley.)
Figure 5. Vector basis function N
i
for linear triangular elements.
triangles and tetrahedrons is much more general, since
any polygonal/polyhedral domain may be represented by
such a mesh.
The quality of an unstructured mesh is determined by
the shape regularity of its elements, which is dened on an
elemental basis as the ratio between the radius of the
largest inscribed circle/sphere and the maximum dimen-
sion of the element. Among the most widely used ap-
proaches to unstructured meshing are Octree,
Delaunay, and Advancing front. The rst approach
breaks the structure into blocks until a desired resolution
is obtained, and then the blocks are meshed. The second
approach actually refers to a criterion for ensuring shape
regularity when creating a mesh from a given set of nodes.
The third approach relies on an already meshed boundary,
to which elements are progressively added toward the in-
terior.
As various requirements are placed on the meshing al-
gorithm, such as continuity between different regions and
variations in element size, the process becomes increas-
ingly complicated. See [11,12] and the references therein
for further discussions on mesh generation.
4.2. Matrix Solvers
An efcient solution of the nite-element matrix equation
is very important, since in the course of solving a problem
with nite-elements most computer resources (memory
and time) are usually expended on this step. The impor-
tant issues are matrix storage schemes, matrix solvers
(direct or iterative), and matrix preconditioners (in the it-
erative case).
The matrices produced by the nite-element method
are sparse with only a very small percentage of nonzero
elements. By storing only the nonzero entries, the matrix
storage requirement is reduced from O(N
2
) to O(N). Pop-
ular approaches to sparse storage are compressed row and
compressed column storage.
The choice of matrix solver can have a significant
impact on computational efciency, so it is important
to choose a solver that best suits the properties of the ma-
trix. There are two types of matrix solvers. The rst type is
direct solvers based on Gaussian elimination or LU
decomposition. These solvers are commonly used for full
matrices, but are also applicable to sparse matrices stored
in a band format, or even a fully sparse format in the
case of the frontal and multifrontal methods [13,14].
The alternative to direct solvers are iterative solvers,
where very little extra memory is required, since
they are based on calculating successive matrixvector
products according to an iterative algorithm to converge to
the solution [15]. The main drawback of such algorithms is
that they might require a prohibitive amount of iterations
to converge, depending mainly on the locations of the
eigenvalues of the matrix; if the eigenvalues are all locat-
ed close to unity, convergence is usually rapid. This prob-
lem is addressed by using a preconditioner to move the
eigenvalues closer to unity and thereby reduce the itera-
tion count. Construction of a preconditioner can be based
on physical insight into the problem or on the structure of
the original matrix.
4.3. Fast Frequency Sweep
Many RF and microwave engineering applications require
computation of frequency responses over a broad band
rather than at one or a few isolated frequencies. Such cal-
culations can be very time-consuming when a traditional
frequency-domain numerical method is used because a set
of algebraic equations must be solved repeatedly at many
frequencies. The number of calculations is proportional to
the electrical size of the problem and can be large for most
applications. Therefore, there is a need to nd approximate
solution techniques that can efciently simulate a frequen-
cy response over a broad band. This can be accomplished by
the method of asymptotic waveform evaluation [16]. In this
method, the unknown solution vector, the right-hand vec-
tor, and the system matrix are rst expanded into Taylor
series at a chosen frequency. The expansion coefcients of
the solution vector are then determined by moment match-
ing. The Taylor series of the solution vector is then con-
verted into a Pade rational function to broaden the radius
of convergence. With this approach, one obtains a solution
that is accurate at frequencies near the point of expansion.
The accuracy of the solution decreases when the frequency
moves away from the point of expansion. In many practical
applications, one is often required to nd the solution over a
specied frequency band, where one point of expansion
may not be sufcient, and multiple points of expansion be-
come necessary. These points can be selected automatically
using a simple binary search algorithm [9].
5. APPLICATIONS OF THE FINITE-ELEMENT METHOD
The nite-element method has been widely applied to the
analysis of problems in RF and microwave engineering. A
few typical applications are discussed here to illustrate
the treatment of various problems commonly encountered
in the nite-element analysis.
5.1. Cavity Analysis
Typical cavity analysis involves nding the resonant fre-
quencies and modal distributions of a given cavity. Math-
ematically, it translates to solving the homogeneous vector
Helmholtz equation
r
1
m
r
rE

k
2
0
e
r
E0 on O 29
for a set of discrete values of k
2
0
that yield nontrivial so-
lutions for E. Application of the nite-element analysis
described in Section 3 to (29) yields a generalized eigen-
value problem
AfEg k
2
0
BfEg 30
where
A
ij
Z
O
1
m
r
rN
i
.
rN
j
dO 31
B
ij
Z
O
e
r
N
i
.
N
j
dO 32
The eigenvalues solved for in (30) yield the resonant
frequencies of the cavity modes, and the corresponding
eigenvectors fEg provide the modal distributions.
5.2. Waveguide Analysis
Typical waveguide analysis involves nding the propaga-
tion constants and eld distributions of the propagation
modes in an innitely long waveguide. Since the wave-
guide is assumed to be uniform in its longitudinal direc-
tion, say, the z direction, it is sufcient to consider (29)
over the cross section of the waveguide.
For all the propagating modes, the electric eld can be
written as
Ex; y; z
1
b
e
t
x; y j ^ zze
z
x; y
!
e
jbz
33
where b represents the propagation constant, e
t
repre-
sents the transverse component, and e
z
represents the
longitudinal component. The coefcients of e
t
and e
z
are
introduced to cast the nal result in a more desirable form
[17]. By introducing the weighting function
Wx; y; z
1
b
w
t
x; y j ^ zzw
z
x; y
!
e
jbz
34
the weak form of (29) can be found as
Z
O
1
m
r
r
t
e
t
.
r
t
w
t
k
2
0
e
r
e
t
.
w
t
&
b
2
1
m
r
r
t
e
z
e
t
.
r
t
w
z
w
t
k
2
0
e
r
e
z
w
z
!'
dO0
35
where r
t
denotes the transverse del operator.
When edge elements are used for e
t
and nodal elements
are employed for e
z
, (35) can be converted into a general-
ized eigenvalue problem
A
tt
0
0 0
" #
e
t
e
z
( )
b
2
B
tt
B
tz
B
zt
B
zz
" #
e
t
e
z
( )
36
in which
A
ttij
Z
O
1
m
r
r
t
N
i
.
r
t
N
j
k
2
0
e
r
N
i
.
N
j
!
dO 37
B
ttij
Z
O
1
m
r
N
i
.
N
j
dO 38
B
tzij
Z
O
1
m
r
N
i
.
r
t
N
j
dO 39
B
ztij
Z
O
1
m
r
r
t
N
i
.
N
j
dO 40
B
zzij
Z
O
1
m
r
r
t
N
i
.
r
t
N
j
k
2
0
e
r
N
i
N
j
!
dO 41
where the indices of N
i
and N
j
run from 1 to N
e
and those
of N
i
and N
j
run from 1 to N.
From (36), one can solve for a set of eigenvalues b
2
and
the corresponding eigenvectors fe
t
e
z
g
T
for a given value of
k
0
. These eigenvalues and eigenvectors give the propaga-
tion constant and eld distribution of each waveguide
mode. Figure 6 shows the dispersion curves for an insu-
lated image guide, calculated using the method described
here.
5.3. Microwave Device Analysis
A microwave device has typically one or more ports, and it
is usually characterized by its s parameters. To perform
the nite-element analysis of a microwave device, it is
necessary to terminate each port to limit the volume of
analysis. A boundary condition is then required at each
port to uniquely dene the boundary value problem.
To derive such a boundary condition, one needs to know
the propagation constants and eld distributions of the
waveguide modes. For most waveguides such as rectan-
gular, circular, and coaxial waveguides, the propagation
constants and modal elds can be obtained analytically.
For other waveguides such as partially lled waveguides
and microstrip lines, they can be calculated using the
nite-element method described in Section 5.2. Once
the waveguide modes are known, the total electric eld

a
b
d
w
h

r
s
0 0.5 1 1.5 2 2.5 3
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
k
0
h

/

k
0
Figure 6. Dispersion characteristics of an insulated image guide
(w/h2.25, d/h0.5, a/h13.5, b/h8.0, e
r
3:8, and e
s
1:5).
(After Jin [9], r2002 Wiley.)
at a port can be written as
Ex; y; z E
inc
x; y; z
X
1
n1
a
n
1
b
n
e
tn
x; y j ^ zze
zn
x; y
!
e
jb
n
z
42
where n denotes the mode number, (x,y,z) are the local co-
ordinates (where z denotes the axis of the waveguide), and
E
inc
represents the incident eld at the port. Using the
orthonormality property of the modes, one nds
a
n
b
n
e
jb
n
z
0
Z
G
A
e
tn
x; y
.
EE
inc
z z
0
dG 43
where G
A
denotes the aperture of the port and z
0
denotes
its position. From (42), it can be shown that
^ nn
1
m
r
rE

PE U
inc
at z z
0
44
where
PE
j
m
r
X
1
n1
b
n
e
tn
r
t
e
zn

Z
G
A
e
tn
.
EdG 45
U
inc
b nn
1
m
r
rE
inc

PE
inc
46
Equation (44) can be considered as a generalized mixed
boundary condition and can be treated in a manner sim-
ilar to that for (19).
Once the eld inside the device and at its ports has
been obtained, the s parameters can be calculated in a
straightforward manner. Figure 7 shows the transmission
coefcient jS
12
j for a circular cavity resonator. The input
and output ports are WR75 waveguides, coupled to the
internal resonator through two rectangular slots.
5.4. Antenna Analysis
Most antennas are used to radiate electromagnetic waves
into an open space. An antenna is typically characterized
by its radiation pattern and input impedance. For multiple
antennas such as an antenna array, the mutual coupling
is also important. The main difculty in the nite-element
analysis of antennas is how to deal with the surrounding
open space. A common approach is to introduce an arti-
cial surface to truncate the innite solution domain into a
nite one. A boundary condition is then required at this
surface for a unique solution of the electromagnetic elds.
This boundary condition should be transparent to the ra-
diated eld; that is, it should permit the radiated eld to
pass through without any reection. Unfortunately, there
is no simple boundary condition that meets this require-
ment. Over the years, a variety of approximate boundary
conditions, also called absorbing boundary conditions,
have been developed to approximately satisfy this require-
ment. The simplest absorbing boundary condition is the
well-known Sommerfeld radiation boundary condition,
which states that if the truncation surface is sufciently
far away from the radiation source, the radiated eld sat-
ises
b nn
1
m
r
rE

jk
0
b nnb nnE % 0 47
This equation is simply the homogeneous version of the
mixed condition (19), the treatment of which is discussed
in Section 3.
Another issue in the nite-element analysis of anten-
nas is the modeling of the antenna feeds. If the antenna
feeds can be modeled approximately as electric current
sources, they can be treated as J in (17). If a more accurate
modeling is desired, they can be modeled as waveguide
ports, the treatment of which is described in Section 5.3.
It must be noted again that (47) is approximate; hence,
the analysis result is also approximate. Although the ac-
curacy can be improved by placing the truncation surface
farther away from the antenna, one then has to deal with
a very large computational domain, which makes the
analysis inefcient. A better alternative is to use an exact
boundary integral equation as a boundary condition at the
truncation surface. This allows the truncation surface to
10 11 12 13 14 15
35
30
25
20
15
10
5
0
Frequency / GHz
|
S
1
2
|

(
d
B
)
Measured
FEM
19.05 mm
9.525 mm
9.7 mm
3 mm
1 mm
100 mm
12 mm
Incident Wave
Figure 7. Transmission coefcient S
12
of a cylindrical cavity res-
onator. (After Liu et al. [18], r2002 Wiley.)
be placed as close to the antenna as possible. This ap-
proach is briefly described in Section 6.3.
Figure 8 shows the input impedance of a microstrip
patch antenna, which consists of a 5.0 3.4-cm rectangu-
lar patch residing on a substrate having thickness t
0.08779 cm, relative permittivity e
r
2:17, and a loss tan-
gent of 0.0015. The antenna is excited by a current probe
applied at (x
f
1:22, y
f
0.85 cm) and has a 50-O resistor
placed at (x
L
2.2cm, y
f
1.5 cm), assuming that the
patch is centered at the origin.
5.5. Scattering Analysis
The nite-element analysis of electromagnetic wave scat-
tering by an object is similar to the analysis of antennas.
An articial surface is rst introduced to enclose the en-
tire object and to truncate the innite solution domain to a
nite one. An absorbing boundary condition is then
employed to permit the scattered eld to leave the trun-
cation surface without significant reection. If the So-
mmerfeld radiation condition is used as the absorbing
boundary condition, it can be written as
b nn
1
m
r
rE
sc

jk
0
b nn b nnE
sc
% 0 48
where E
sc
denotes the scattered eld, which is the differ-
ence between the total and the incident elds
(E
sc
EE
inc
).
Since the absorbing boundary condition applies only to
the scattered eld, the formulation for scattering analysis
is slightly different from that for the antenna analysis for-
mulation. There are two approaches to formulating the
scattering problem. One approach uses the total eld that
satises (29). The absorbing boundary condition (48) is
then written in terms of the total eld as
b nn
1
m
r
rE

jk
0
b nn b nnE U
inc
49
where
U
inc
b nn
1
m
r
rE
inc

jk
0
b nnb nnE
inc
50
Another approach is to work on the scattered eld using
(48) directly. In this case, (29) can be written for the scat-
tered eld as
r
1
m
r
rE
sc

jk
0
e
r
E
sc
F
inc
51
where
F
inc
r
1
m
r
rE
inc

jk
0
e
r
E
inc
52
and the boundary conditions on conducting surfaces be-
come
b nnE
sc
b nnE
inc
53
Both approaches are mathematically equivalent. They
differ in the numerical implementation, specifically, in the
implementation of the excitation. In the rst approach,
the excitation is introduced as a boundary source on the
truncation surface; in the second approach, it is intro-
duced on the conducting surface and over the object. Nu-
merical experiments show that the second approach is
slightly more accurate since the wave travels a shorter
distance in the nite-element mesh, resulting in a smaller
dispersion error.
Figure 9 shows the VV-polarized monostatic radar
cross section of a metallic double ogive at 9GHz, which
is formed by joining two different half-ogives. The top
piece has a half-length of 12.7 cm, a maximum radius of
2.54 cm, and a half-angle of 22.621 at the tip, and the bot-
tom has a half-length of 6.35cm, a maximum radius of
1 1.5 2 2.5 3 3.5 4
0
5
10
15
20
25
30
Frequency (GHz)
R

(
o
h
m
s
)
FEM
Measured
1 1.5 2 2.5 3 3.5 4
10
5
0
5
10
15
20
25
30
Frequency (GHz)
X

(
o
h
m
s
)
FEM
Measured
(a)
(b)
Figure 8. Input impedance of a loaded microstrip patch antenna:
(a) resistance; (b) reactance. (After Jin and Volakis [19], r1991
IEEE.)
2.54 cm, and a half-angle of 46.41 at the tip. The calculated
results are compared with measured data.
6. ADVANCED FINITE-ELEMENT ANALYSIS
There are many advanced methods that build on and com-
plement the basic nite-element method framework as il-
lustrated in Sections 2 and 3. A few of the most prominent
of these methods are discussed here.
6.1. Higher-Order Elements
Most nite-element methods use basis functions that vary
linearly within every element (see Figs. 3 and 5). However,
it is also possible, and often very advantageous, to consid-
er basis functions of higher polynomial orders. Such basis
functions are dened for scalar nodal elements in Refs. 8
and 9 and for vector elements in Refs. 21 and 22.
Higher-order basis functions can be categorized into
two classes: interpolatory and hierarchical. Interpolatory
basis functions relate to a set of elemental points, such
that every basis function is of the same order, and is equal
to unity at one point and zero at all others. On the other
hand, hierarchical basis functions are formed by adding
new higher-order basis functions to the lower-order ones;
thus the elemental solution is expanded in terms of basis
functions of differing polynomial orders. Both approaches
have the same accuracy, but one may be favored over the
other, depending on the application. Interpolatory basis
functions generally lead to better conditioned matrices,
while hierarchical basis functions permit the use of dif-
ferent elemental orders in a single nite-element solution.
Since higher-order basis functions interpolate the so-
lution eld much more accurately, the nite-element
method may be expected to yield much more accurate re-
sults, as the elemental order is increased. Specifically, for
smooth functions it can be shown that if p is the order of
the basis functions, h denotes the elemental size divided
by the order of the element, and l denotes the excitation
wavelength, the interpolation error is of the order
Oh=l
p1
. Therefore, in the case of smooth solutions,
the nite-element solution error may be reduced accord-
ingly by using higher-order basis functions. If the true so-
lution to the nite-element analysis contains a singularity,
this interpolation error estimate does not hold anymore
and it becomes more advantageous to use smaller ele-
ments of low polynomial order around the singularity.
When the nite-element method is applied to the Helm-
holtz equations (scalar as well as vector cases), the simu-
lated wave propagates at a speed slightly different from
the exact value. Consequently, there is a numerical error
in the phase of the numerical solution, which is called
dispersion error. Analysis based on the nite-element disc-
retization of plane wave propagation in a uniform medium
shows that the dispersion error per wavelength is propor-
tional to Oh=l
2p
. From this result follows the very im-
portant conclusion that phase errors may be decreased
exponentially by increasing the order of the elements.
Therefore, higher-order elements are especially suitable
for simulating large-scale wave propagation problems.
6.2. Curvilinear Elements
In the same way that elemental basis functions can be
dened to arbitrary polynomial order, the geometric rep-
resentation of the elements can also be dened to arbi-
trary polynomial order. The elements shown in Figs. 3 and
5 are of linear geometric order since all edges and faces are
straight and at, and can thus be described by linear
functions of position. Such elements are termed rectilin-
ear. Elements of higher geometric polynomial order may
be dened by modeling elemental edges and faces with
higher-order polynomial functions of position. For exam-
ple, one may require the edge of a triangular element to
pass through two of its vertex nodes together with an ad-
ditional node that may not lie on the straight line con-
necting the vertices. In this case, a second-order geometric
representation would sufce. Such elemental representa-
tions are very useful when modeling curved boundaries
and are termed curvilinear. When the order of geometric
representation is the same as the order of basis functions,
the element is called isoparametric. Otherwise, it is called
either subparametric (when the order of geometric mod-
eling is lower) or superparametric (when the order of geo-
metric modeling is higher).
6.3. Finite-Element/Boundary Integral Method
As mentioned in Section 5.4, instead of using an absorbing
boundary condition to approximate an innite exterior
volume to the nite-element domain, one may use a
boundary integral equation as a boundary condition
such that the innite exterior region is modeled exactly
through the Green function [9]. This approach, called the
nite-element/boundary integral method, allows complete
freedom in choosing the truncation surface enclosing the
region of interest.
For example, consider the case of solving (17) on a
truncated mesh with innite exterior free space. The trun-
cation surface, denoted by S, is closed, and given the elec-
tric and magnetic elds on such a surface, one can
calculate the electric and magnetic elds to its exterior
using the electric and magnetic eld integral equations,
FEM
Measured
90 60 30 0 30 60 90
60
50
40
30
20
10
Elevation Angle (degrees)
M
o
n
o
s
t
a
t
i
c

R
C
S

(
d
B
s
m
)
Figure 9. VV-polarized radar cross section of a metallic double
ogive at 9 GHz. (After Greenwood and Jin [20], r1999 IEEE.)
which are dened as follows
EE
inc
Z
0
LS; J
S
KS; E b nn 54
Z
0
HZ
0
H
inc
Z
0
KS; J
S
LS; E b nn 55
where E
inc
and H
inc
represent the exterior incident elds
and
LS; v jk
0
I
S
vr
0
G
0
r; r
0

1
k
2
0
r
0
.
vr
0
rG
0
r; r
0
!
dS
0
56
KS; v
I
S
vr
0
rG
0
r; r
0
dS
0
57
in which G
0
r; r
0
denotes the scalar, free-space Green
function [9,23] and J
S
b nnH on S. For the nite-element
region one may then use the simple boundary condition
b nn
1
m
r
rE

jk
0
Z
0
J
S
onS 58
and solve it together with a numerical system obtained by
applying Galerkins method to either (54) or (55) (or linear
combinations of these), in terms of both E and J
S
. This
approach has been widely used, with various modica-
tions to improve robustness and efciency [9,24].
The nite-element/boundary integral method generally
leads to sparse matrices with fully populated subblocks,
corresponding to the unknowns associated with the exte-
rior boundary. This aspect implies increased computation-
al resource requirements, but on the other hand, the
method enables one to reduce the computational domains
size to a minimum. More importantly, the method is ac-
curate and reliable since its formulation involves no ap-
proximation other than numerical discretization.
6.4. Time-Domain Finite-Element Method
The nite-element method for electromagnetic elds may
be formulated in the time domain, rather than the fre-
quency domain, so that transient responses may be calcu-
lated directly. This is especially useful when data over a
wide frequency band are required or when nonlinear de-
vices are considered. In the time domain, the vector wave
equation (17) takes the following form:
r
1
m
rE

e
@
2
E
@t
2
s
@E
@t

@J
@t
on O 59
In order to solve (59) with the nite-element method,
both spatial and temporal discretization is required. The
spatial discretization follows the procedure described in
Section 3 by using vector elements to represent the elds,
which yields the following system in terms of the time
variable:
T
d
2
fEg
dt
2
R
dfEg
dt
SfEg ff g f0g 60
This equation represents a set of second-order ordinary
differential equations in terms of the time-dependent
spatial unknowns fEg. This equation is solved numerical-
ly by rst dividing the time axis into discrete intervals of
length Dt. Next, the time derivatives are approximated
such that fEg can be calculated at successive timesteps,
t nDt, n1,2,3,y. For this purpose, various timestep-
ping schemes are available, such as central differencing
and the Newmark b method [9,25]. The Newmark b meth-
od, when applied to (60), takes the following form:
1
Dt
2
T
1
2Dt
R bS
& '
fEg
n1
2
Dt
2
T 1 2bS
& '
fEg
n
1
Dt
2
T
1
2Dt
R bS
& '
fEg
n1
bff g
n1
1 2bff g
n
bff g
n1

61
This method has a very important characteristic that
when the parameter b is set to b ! 0:25, it becomes
unconditionally stable, which means that it can be used
with a timestep independent of spatial discretization size.
Usually the stability of a timestepping scheme is tied to
the minimum size of the spatial discretization, given a
xed Dt.
When considering open-region problems, the time-do-
main equivalents of the absorbing boundary condition and
the nite-element/boundary integral method can be
employed [9].
6.5. Adaptive Finite-Element Method
Associated with any nite-element solution is an error due
to the approximate representation of the solution on the
mesh. Quantifying this error is an important extension to
the nite-element method, since it opens up the possibility
of adaptive nite-element analysis, which can enhance the
efciency of the nite-element method significantly.
The goal of an adaptive nite-element analysis is to
optimize the mesh size (designated by h) and element or-
ders (designated by p) with which a given accuracy can be
obtained with the minimum number of unknowns. This is
accomplished in the following way. Choose an initial disc-
retization and solve the problem. On the basis of the so-
lution, estimate the error distribution and the global
solution error, as measured in a relevant quantity such
as energy, impedance or scattering cross section. On the
basis of the error distribution, rene the solution repre-
sentation selectively with respect to h and/or p. Resolve
the problem using the new discretization and repeat the
process. Iterate in this way until the global error measure
drops below a required level.
The core component of this process is the error estima-
tion, which is termed a posteriori error estimation, since
an existing solution is required. Many different approach-
es to a posteriori error estimation exist [26,27]. Usually
the error estimator is derived mathematically, based on
the nite-element formulation at hand and the output
quantity of interest, but sometimes one can construct it on
the basis of physical insight alone. For example, for vector
eld problems one can use the residues of (17) and (19) to
estimate the error over each element. The eld continuity
condition across each element edge or face can also be in-
corporated into the error estimation.
7. CONCLUSION
This article has described the basic principle, the numer-
ical formulation, and some applications of the nite-ele-
ment method for analysis of electromagnetic problems in
RFand microwave engineering. It also has discussed brief-
ly some computational aspects and advanced topics asso-
ciated with the method. The nite-element method is a
powerful analysis and design tool because it can model
very complex geometries and can deal with a variety of
media, including inhomogeneous and anisotropic cases.
Moreover, the method yields a very sparse matrix that can
be generated, stored, and solved efciently.
The nite-element method in RF and microwave engi-
neering, and electromagnetics in general, has undergone
several important developments. For example, the devel-
opment of vector elements enabled the correct modeling of
the electric and magnetic vector elds for the rst time.
The development of higher-order versions of the vector el-
ements permitted accurate and efcient modeling of large-
scale eld problems, especially wave propagation prob-
lems, because of the reduced interpolation and dispersion
errors. The development of absorbing boundary conditions
and perfectly matched layers [28,29] allowed fast, approx-
imate analysis of wave radiation, scattering, and propa-
gation problems. The development of the hybrid nite-
element/boundary integral method provided a more reli-
able and robust approach to deal with these problems. It
also created a mechanism to incorporate fast integral solv-
ers into the nite-element simulations, such as FFT-based
methods and the fast multipole method [30]. The hybrid-
ization with asymptotic methods [31] made it possible to
analyze electrically very large scatterers with small and
complex features. The development of adaptive methods
[32] further enhanced the efciency of the nite-element
method and provided an error-controllable solution. The
development of the time-domain nite-element method
[33] enabled numerical modeling of nonlinear devices
and materials and made efcient broadband simulation
possible.
The material covered in this article represents only a
fraction of the theory and application of the nite-element
method in electromagnetics. With its continuous advance-
ment and the rapid development of computer technology,
RF and microwave devices will in future be designed rou-
tinely on computers in the same way as is done today with
electronic circuit designs.
BIBLIOGRAPHY
1. R. L. Courant, Variational methods for the solution of prob-
lems of equilibrium and vibration, Bull. Am. Math. Soc. 49:
123 (1943).
2. P. Silvester, Finite-element solution of homogeneous wave-
guide problems, Alta Freq. 38:313317 (May 1969).
3. K. K. Mei, Unimoment method of solving antenna and scat-
tering problems, IEEE Trans. Anten. Propag. 22:760766
(Nov. 1974).
4. S. P. Marin, Computing scattering amplitudes for arbitrary
cylinders under incident plane waves, IEEE Trans. Anten.
Propag. 30:10451049 (Nov. 1982).
5. J. C. Nedelec, Mixed nite elements in R
3
, Numerische Math-
ematik 35:315341 (1980).
6. A. Bossavit and J. C. Verite, A mixed FEM-BIEM method to
solve 3-D eddy current problems, IEEE Trans. Magn. 18:431
435 (March 1982).
7. M. L. Barton and Z. J. Cendes, New vector nite elements for
three-dimensional magnetic eld computation, J. Appl. Phys.
61:39193921 (April 1987).
8. P. P. Silvester and R. L. Ferrari, Finite Elements for Electrical
Engineers, 3rd ed., Cambridge Univ. Press, Cambridge; UK,
1996.
9. J.-M. Jin, The Finite Element Method in Electromagnetics,
2nd ed., John Wiley, New York; 2002.
10. J. Volakis, A. Chatterjee, and L. Kempel, Finite Element
Method for Electromagnetics: Antennas, Microwave Cicuits
and Scattering Applications, Oxford Univ. Press, Oxford and
IEEE Press, New York, 1998.
11. J. -F. Lee and R. Dyczij-Edlinger, Automatic mesh generation
using a modied Delaunay tesselation, IEEE Anten. Propag.
Mag. 39:3445 (Feb. 1997).
12. D. N. Shenton and Z. J. Cendes, Three-dimensional nite el-
ement mesh generation using Delaunay tesselation, IEEE
Trans. Magn. 21:25352538 (Nov. 1985).
13. B. M. Irons, A frontal method solution program for nite el-
ement analysis, Int. J. Numer. Meth. Eng. 2:532 (1970).
14. J. W. H. Liu, The multifrontal method for sparse matrix so-
lution: Theory and practice, SIAM Rev. 34:82109 (March
1992).
15. R. Barrett, M. Berry, T. F. Chan, J. Demmel, J. Donato,
J. Dongarra, V. Eijkhout, R. Pozo, C. Romine, and H. V. der
Vorst, Templates for the Solution of Linear Systems: Building
Blocks for Iterative Methods, 2nd ed., SIAM, Philadelphia,
1994.
16. L. T. Pillage and R. A. Rohrer, Asymptotic waveform evalu-
ation for timing analysis, IEEE Trans. Comput.-Aided Design
9:352366 (April 1990).
17. J. -F. Lee, D. -K. Sun, and Z. J. Cendes, Full-wave analysis of
dielectric waveguides using tangential vector nite-elements,
IEEE Trans. Microwave Theory Tech. 39:12621271 (Aug.
1991).
18. J. Liu, J. -M. Jin, E. K. N. Yung, and R. S. Chen, A fast three-
dimensional higher-order nite-element analysis of micro-
wave waveguide devices, Microwave Opt. Technol. Lett.
32:344352 (March 2002).
19. J. -M. Jin and J. L. Volakis, A hybrid nite-element method
for scattering and radiation by microstrip patch antennas and
arrays residing in a cavity, IEEE Trans. Anten. Propag.
39:15981604 (Nov. 1991).
20. A. D. Greenwood and J. M. Jin, A novel efcient algorithm for
scattering from a complex BOR using vector FEM and cylin-
drical PML, IEEE Trans. Anten. Propag. 47:620629 (April
1999).
21. R. D. Graglia, D. R. Wilton, and A. F. Peterson, Higher order
interpolatory vector bases for computational electromagnet-
ics, IEEE Trans. Anten. Propag. 45:329342 (March 1997).
22. J. P. Webb, Hierarchical vector basis functions of arbitrary
order for triangular and tetrahedral nite elements, IEEE
Trans. Anten. Propag. 47:12441253 (Aug. 1999).
23. C. T. Tai, Dyadic Green Functions in Electromagnetic Theory,
2nd ed., IEEE Press, New York, 1994.
24. X.-Q. Sheng, J.-M. Jin, J. Song, C.-C. Lu, and W. C. Chew, On
the formulation of hybrid nite-element and boundary-inte-
gral methods for 3-D scattering, IEEE Trans. Anten. Propag.
46:303311 (March 1998).
25. N. M. Newmark, A method of computation for structural dy-
namics, J. Eng. Mech. Div. Proc. Am. Soc. Civil Eng. 85:6794
(July 1959).
26. M. Ainsworth and J. T. Oden, A posteriori error estimation in
nite-element analysis, Comput. Meth. Appl. Mech. Eng.
142:188 (1997).
27. J. T. Oden, L. Demkowicz, W. Rachowicz, and T. A. Westerm-
ann, Toward a universal h-p adaptive nite-element strategy,
Part 2. A posteriori error estimation, Comput. Meth. Appl.
Mech. Eng. 77:113180 (1989).
28. J. Berenger, A perfectly matched layer for the absorption of
electromagnetic waves, J. Comput. Phys. 144:185200 (Oct.
1994).
29. D. Jiao, J. -M. Jin, E. Michielssen, and D. Riley, Time-domain
nite-element simulation of three-dimensional scattering and
radiation problems using perfectly matched layers, IEEE
Trans. Anten. Propag. 51:296305 (Feb. 2003).
30. R. Coifman, V. Rohklin, and S. Wandzura, The fast multipole
method for the wave equation: A pedestrian prescription,
IEEE Anten. Propag. Mag. 35:712 (June 1993).
31. J. M. Jin, S. Ni, and S. W. Lee, Hybridization of SBR and FEM
for scattering by large bodies with cracks and cavities, IEEE
Trans. Anten. Propag. 43:11301139 (Oct. 1995).
32. M. Salazar-Palma, T. K. Sarkar, L.-E. Garc a-Castillo, T. Roy,
and A. Djordjevic, Iterative and Self-Adaptive Finite-Ele-
ments in Electromagnetic Modeling, Artech House, Boston,
1998.
33. J. -F. Lee, R. Lee, and A. C. Cangellaris, Time-domain nite-
element methods, IEEE Trans. Anten. Propag. 45:430442
(March 1997).
FINLINE COMPONENTS
WOLFGANG S. MENZEL
University of Ulm
Ulm, Germany
1. INTRODUCTION
The nline has been intensively investigated since 1970 as
a planar transmission-line medium for millimeter-wave
components. The electromagnetic wave is guided by slots
in a metallization printed on a thin dielectric substrate.
All passive circuit functions are realized by these slotline
structures, partly in combination with other types of pla-
nar transmission lines, such as microstrips or coplanar
lines printed on the same substrate. These circuits are
correctly known as E-plane circuits, but the terms nline
and E-plane circuits are often used interchangeably. Semi-
conductor devices, preferably beam-lead devices, can eas-
ily be soldered, glued, or bonded across the slot. The
complete transmission line is held and shielded by a met-
al housing that most often has the (inner) dimension of the
respective metal waveguide (Fig. 1). The transition from
metal waveguide to nline can be achieved by tapering the
nline slot to the waveguide height as shown in Fig. 1.
Therefore, nline components can easily be combined with
waveguide circuits. Because the circuit functions are de-
termined by the planar structure, the tolerance require-
ments for this housing are considerably relaxed; thus this
housing may be fabricated by low-cost methods such as
metal casting or even plastic injection molding and elec-
troplating.
Because of the relatively large transmission-line
cross section, nline losses typically are lower than
for microstrip or coplanar lines, but higher than those
for metal waveguide. Typical nline types, cross sections,
and propagation properties are described in detail in
Ref. 1.
2. PASSIVE FINLINE COMPONENTS
2.1. Filters
Finline lters, especially band-pass and low-pass lters
are based on transmission-line structures [2]. Bandpass
lters typically are realized using either side-coupled (Fig.
2a) or end-coupled (Fig. 2b) resonators. To reduce losses,
end-coupled lters may degenerate to pure E-plane wave-
guide lters as shown in Fig. 2c. In this case, the slot
width is equal to the waveguide height, and no taper to
waveguide is necessary; even the dielectric substrate ma-
terial may be omitted, resulting in simple metal insert l-
ters. For better stopband attenuation, bilateral nline or
several side-by-side metal inserts may be used. These ar-
rangements strongly suppress higher-order mode cou-
pling and therefore improve the stopband behavior.
Detailed descriptions of these kinds of lters, together
with diplexers, are presented by Shih et al. [3] and Dittloff
and Arndt [4].
Lowpass lters typically are based on low- and high-
impedance line sections (Fig. 2d) or stub structures (Fig.
2e). High impedances are easily realized with wide slots.
For uniplanar nline, the lowest impedance value is de-
termined by the minimum slot width that can be techno-
logically realized; for bilateral or antipodal nline,
overlapping of the metallization may further reduce the
characteristic impedance of the respective line segment.
The discontinuities involved have to be included in the
design procedure, especially for large steps in impedance.
Figure 1. Cross section of a (unilateral) nline and metallization
pattern for a transition from waveguide to nline.
FINLINE COMPONENTS 1601
2.2. Directional Couplers
The design of a directional coupler in conjunction with
nline was rst reported by Meier [8] based on printed
metal probes between two adjacent metal waveguides,
placed in between two thin substrate layers in the E
plane. Finline couplers, however, are designed mostly on
the basis of coupled slots. Such a coupled transmission-
line structure supports two modes; one is similar to the
normal nline mode with the electric elds in the same
direction, and the other one is a coplanar-type mode with
zero cutoff frequency. The coupler design shown in Fig. 3a
relies on a quarter-wave section of coupled lines. The fun-
damental principles of operation are comparable to those
of respective couplers in the microstrip or microstrip/slot-
line technique. The center strip may alternatively be re-
placed by a bond wire, or the metallization section
containing the strip may be placed on the backside of
the substrate (antipodal nline), thus achieving an in-
creased design exibility. Another coupler principle is
based solely on the different phase velocity of the two
modes on the coupled transmission line (Fig. 3b). Feeding
a wave to one of the ports, an increasing part of the elec-
tromagnetic eld couples from one slot to the other with
increasing propagation distance; that is, the amount of
coupling depends on the length of the coupled-line section.
This length will be several wavelengths, but in the milli-
meter-wave range, this problem is reduced because of the
short wavelength. In contrast to the coupler design shown
in Fig. 3a, a smooth transition from the feedline to the
coupled-line section with minimum reections is manda-
tory. Furthermore, the coupling of the transition region
contributes to the coupler performance. With both types of
couplers, broadband performance over a complete wave-
guide band is achieved [57].
2.3. Nonreciprocal Components
Some effort has be made to develop circulators and isola-
tors in nline or by a compatible E-plane technique. A
ferrite sphere or a small cylinder is placed in the center of
a nline Y junction to realize circulators [9,10]. For a eld
displacement type of isolator, an additional layer of ferrite
substrate together with an absorbing sheet is added to the
nline on a standard substrate [11]; optimization is done
by full-wave calculations of the composite structure. Both
approaches, however, lead to rather complex and therefore
mechanically sensitive structures, so these are rarely used
in practical applications.
A more successful approach is the realization of wave-
guide-type circulators in a metal waveguide, Y junction in
the E plane [12]. A ferrite disk is placed into the junction.
The diameter of the disk is equal to the inner diameter of
the waveguide Y junction. The center frequency is deter-
mined by the height of the disk. A permanent magnet is
placed into a hole in the waveguide mount close to the
ferrite disk. This type of circulator, therefore, is fabricated
easily, and is integrated into the E-plane split block to-
gether with single or integrated nline components, as
(a)
(b)
(c)
(d)
(e)
Figure 2. Typical metallization patterns of different nline l-
ters: (a) side-coupled bandpass lter; (b) end-coupled bandpass
lter; (c) end-coupled bandpass lter (the slot width has been in-
creased to the waveguide height, losses are lower, and no taper to
waveguide is required); (d) highlow impedance lowpass lter; (e)
stub-type lowpass lter.
z/4
(a)
(b)
Figure 3. Metallization patterns of two different nline couplers.
1602 FINLINE COMPONENTS
shown later. Circulators of this type are even built up to
the 140GHz frequency range.
2.4. Finline Antennas
Extending the nline substrate out of the waveguide
mount allows the realization of slotline or Vivaldi-type
endre antennas (Fig. 4). Beamform, sidelobe level, and
bandwidth can be determined, to some extent, by the form
of the tapered metallization structure and by a proper
choice of the substrate thickness and its dielectric con-
stant. Beamwidth, however, remains relatively wide asso-
ciated with low gain. Applications of nline antennas are
found in the integration of several antennas as an antenna
array or as a feed cluster to form multiple beams in con-
junction with lens or reector antennas [13].
3. DIODE-BASED FINLINE COMPONENTS
Schottky diodes, pin diodes, and varactor diodes have been
fabricated as beam-lead devices for many years and are
ideally suited for integration into nline circuits. Because
of their low parasitic capacitances and inductances, high-
performance components with wide bandwidth can be
built.
3.1. Detectors
For a nline detector, a low-barrier Schottky diode is
placed across the nline slot. One of the nline metal-
lizations has to be isolated from the metal mount by a thin
dielectric sheet (1020mm). Alternatively, an oxidized alu-
minum mount is used. For broadband operation, some ab-
sorber material is placed behind the diode. Figure 5 is a
sketch of the detector layout. Detectors of this type are
capable of zero-bias operation over one or two waveguide
bands with sensitivities of several hundred millivolts (mV/
mW) per milliwatt, and are even designed for frequencies
up to 200 GHz and above [7,14].
3.2. pin Diode Attenuators and Switches
In the same way as with the detector, one or more beam-
lead pin diodes can easily be placed across a nline slot.
With bias applied to the pin diodes, these exhibit a low
series resistance together with some parasitic inductance.
Thus they basically short-circuit the nline slot. Without
or with reverse bias, pin diodes load the nline slot only
with their fairly low capacitance. Basically, two types of
attenuator or switch have been developed. The rst one
uses diodes in the shunt conguration shown in Fig. 6a.
Placing two or more diodes approximately a quarter-wave-
length apart results in some compensation of the diode
capacitance in the ON state of the switch, that is, without
biasing the diodes or with reverse bias. Attenuators (the
attenuation is controlled by the amount of diode bias) or
switches of this type typically have an insertion loss of
0.5 dB in the Ka band (2640 GHz) up to 1.5dB in the W
Waveguide
mount
Finline and antenna
metallisation
Substrate
Figure 4. Waveguide mount and metallization pattern of a n-
line antenna. The planar substrate with the nline metallization
protrudes out of the waveguide mount and acts as a slotline-type
antenna.
Low barrier Schottky diode Absorber
Figure 5. Basic setup of a nline detector. One of the nline
metallizations is DC-isolated from the waveguide mount to allow
extraction of the rectied output signal.
PIN diodes (a)
(b)
(c)
Figure 6. Metallization patterns of nline shunt-type (a) and se-
ries-type (b) attenuator or switch, and single-pole double-throw
(SPDT) switch with shunt diode conguration (c).
band (75110 GHz) over a nearly complete waveguide
band. Maximum attenuation amounts to 15 dB per diode.
For attenuation values higher than B50 dB, parasitic
power leakage along the DC isolation between the metal-
lization pattern on the substrate and the waveguide
mount is a concern, and absorbing isolating foils must be
used to prevent this problem. For attenuators operating
over even two waveguide bands, the nline can be placed
in a ridged waveguide mount.
The second type of pin diode switch is based on a series
conguration of the diode (Fig. 6b). A diode is placed
across a lateral slot in the metallization. This slot is
short-circuited for RF signals by the wall of the wave-
guide mount. Together with the diode capacitance, this
stub forms a parallel resonator in the OFF position of the
attenuator, whereas the stub is short-circuited in the ON
position. Accordingly, this attenuator has a more narrow-
band performance, but with increased attenuation for a
single diode. Furthermore, it presents some advantages
with respect to DC isolation from other components in in-
tegrated nline circuits. In the case of a single diode, the
metallization at one side of the stub has to be DC-isolated
from the waveguide mount. With two cascaded structures
of this type, only the metallization between the stubs
needs to be DC isolated (see also Section 5).
Two or three switches as described above, preferably
those with the diodes in a shunt conguration, can also be
arranged for single-pole double-throw (SPDT) or single-
pole triple-throw (SPTT) switches, as shown in Fig. 6c.
This results in performance similar to the single switches
except for slightly increased losses due to the additional
transmission line(s) loading. Further details and numer-
ous results are described by Callsen et al. [15].
3.3. Finline Mixers
Although single-ended mixers are easily realized placing a
beam-lead Schottky diode across the nline slot [8], much
more effort has been put into the design of a balanced
mixer based on a combination of nline and coplanar line
as shown in Fig. 7. This structure gives an ideal combi-
nation for forming an inherently broadband, 1801, hybrid
junction. The Schottky diodes are placed across the junc-
tion, in series with respect to the nline and antiparallel
with respect to the coplanar line. The RF signal is fed to
the diodes via the nline and the local oscillator (LO) sig-
nal via the coplanar line. Then the intermediate frequency
(IF) is also extracted from the coplanar line. Therefore,
some diplexing circuit must be added to the coplanar line.
This single balanced mixer design has inherent high mu-
tual isolation between RF and LO ports (and RF and IF
ports, respectively), but in practice this is limited by asym-
metries in the circuit structure or the diodes.
Many mixers based on this conguration have been de-
veloped. Basic differences were found mainly in the LO/IF
feeding arrangement. Some mixers use a transition from
nline to coplanar line via an antipodal nline, as
sketched in Fig. 8, possibly together with an end-coupled
band-pass lter for the LO signal and a microstrip lowpass
lter for the IF. A typical Ka-band nline lter of this type
was presented by Ball and Bui [16] and is shown in the
photograph in Fig. 9. Another design [7,17] uses a transi-
tion from nline to coplanar line based on magnetic cou-
pling, as shown in Fig. 10. Very broadband performance
and operating frequencies up to 150GHz are reported [18]
with this type of mixer. A photograph of a 140150GHz
mixer is shown in Fig. 11.
Because sufcient LO power in the millimeter-wave
frequency range is sometimes difcult or too expensive to
generate, efforts have been undertaken to design subhar-
monically pumped mixers. For mixing with even harmon-
ics of the LO frequency, two diodes are used in an
antiparallel conguration. For odd harmonics, a circuit
and diode conguration as used for the balanced mixer
and shown in Fig. 7 can be employed. Together with suit-
able lter congurations (Fig. 12), mixers with a LO fre-
quency at about half the RF frequency [19,20] and at much
lower frequencies, for example, f
LO
Ef
RF
/8 [21], have been
realized.
Until the late 1990s, a major problem in the design of
nline mixers was determination of the actual nonlinear
RF
Finline
Coplanar
line
I.O. IF
Figure 7. Basic principle of a balanced
nline mixer and its equivalent circuit.
The combination of nline and coplanar
line forms an inherently broadband
1801 hybrid junction.
Back side metallization
Front side metallization
Figure 8. Layout of a transition from nline to coplanar line via
an antipodal nline and a microstrip line.
diode impedances for signal, LO, and IF frequency. The
problem is even more challenging as the diodes are placed
in a major discontinuity (Fig. 7) where parasitic capaci-
tances and inductances affect the impedances seen by in-
put and output transmission lines. This problem now has
been solved applying a full-wave nite difference in time-
domain (FTDT) eld solver, including lumped nonlinear
elements such as the mixer diodes [22]. The mixer diodes
are represented by a suitable equivalent circuit (valid up
to the respective frequency of operation) and embedded
into the FTDTcomputation. In the millimeter-wave range,
the actual size of the beam-lead diodes can be accounted
for by distributing the equivalent circuit over several cells
of the FTDT grid. The nonlinear response of the diodes to
the LO power is calculated by a harmonic excitation of the
diodes, including the actual power level. Following this, a
Figure 9. Photograph of a Ka-band balanced
nline mixer [16] (courtesy of Academic Press).
LO
IF RF
Backshort
Backside
metallisation
Figure 10. Basic layout of a balanced nline mixer with probe-
type transition to coplanar line [18].
Figure 11. Photograph of a 140-GHz nline mixer [18]. The
waveguide mount is opened, and the quartz substrate with the
mixer circuitry can be clearly seen.
small-signal, modulated Gaussian pulse is fed to the sig-
nal port of the mixer (with the LO signal still present).
From the reected signals, the return loss at the respec-
tive ports is computed, and the conversion loss is extracted
from the input signal and the IF output signal. On the
basis of these calculations, a matching and an optimiza-
tion of (harmonic) mixers are possible. An example of a
harmonic mixer with the LO frequency at about one-third
of the signal frequency is shown in Fig. 13. A conversion
loss of down to 15 dB for a 60-GHz third-harmonic mixer is
achieved using low-cost silicon Schottky diodes (Fig. 14).
3.4. Phase Shifters
Finline phase shifters are based on two different princi-
ples. The rst one uses a 3-dB coupler described earlier in
this article together with pin diode switches to modify the
two output port loads [6]. The other principle uses a
conguration very similar to that of the balanced mixer
(see Figs. 7 and 10 and (Refs. 23 and 24). For the phase
shifter, either pin or Schottky diodes are alternatively
switched to the ON or OFF state via a bias applied to the
coplanar line. With an appropriate matching of the cir-
cuitry, signal transmission between the nline port (RF
port in the mixer case) and the coplanar port (LO port of
the mixer) undergoes a phase differences of 1801 depend-
ing on the sign of the control voltage, that is, depending on
which diode is in the ON (biased) state.
4. FINLINE OSCILLATORS AND AMPLIFIERS
A number of efforts have been made to realize oscillators
by the integrated nline technique. A basic problem is al-
ways heat removal from the active elements. Therefore,
Gunn elements are usually used screwed directly into the
broadside of the waveguide mount. Then the diode cap is
soldered directly to one side of the nline metallization,
which contains the necessary resonance and matching
circuits (Fig. 15). Final tuning is typically done with a
backshort in the metal waveguide behind the diode. A
more detailed review of nline oscillators is given by Hoe-
fer [25]. Although some oscillators are used in integrated
nline front ends (see next section), they have not proved
suitable for high-volume and low-cost applications.
Some authors published designs of nline ampliers
[25,26]. Their arrangements, however, basically are mi-
crostrip FET ampliers with transitions to nline, and
RF
RF
LO
LO
IF
IF
High-
pass
RF
High-
pass
RF
Band-
pass
LO
Band-
pass
LO
Low-
pass
IF
Low-
pass
IF
Figure 12. Basic block diagrams of harmonic mixers. For even
harmonic mixing products, an antiparallel diode conguration is
used; for odd mixing products, a parallel/antiparallel diode ar-
rangement is preferred.
RF
input
Front side
metallization
Back side
metallization
Mixer
diodes
Figure 13. Layout of the key section of a third-harmonic 60-GHz
mixer that has been optimized using a FDTD calculation. The
steps in widths of slot and strips as well as the backside metal-
lization under the LO/IF line provide the impedance matching for
the different ports.
10
15
20
25
30
1 2 3 4 5 6
IF frequency (GHz)
Measurement
FDTD C
o
n
v
e
r
s
i
o
n

l
o
s
s

(
d
B
)
RF frequency: 55 to 60 GHz
Figure 14. Conversion loss of the third-harmonic 60-GHz mixer.
As a result of the full-wave calculation, even resonances can be
predicted and shifted to uncritical frequencies.
Resonator and
matching circuit
Waveguide
backshort Gunn element
Figure 15. Basic setup of a nline Gunn oscillator. The Gunn
element is screwed directly into the waveguide mount; resonance
and matching structures are integrated into the planar structure.
they use the waveguide mount as some kind of package.
Furthermore, because nline does not lend itself easily to
a monolithic integration, amplier design in nline was
not pursued further.
5. INTEGRATED FINLINE COMPONENTS AND
FRONT ENDS
The major advantages of planar components are exploited
by integrating several components to supercomponents
or complete front ends. With respect to nline, this can be
done in two different ways. Because nline circuits consist
of planar circuits and waveguide mounts, a rst step is to
integrate several nline circuits in a single waveguide
mount, although each circuit still has its own transition(s)
to waveguide. This concept allows optimizing and testing
all nline circuits separately and combining them with
waveguide circuits like oscillators, waveguide lters, or
circulators realized in a waveguide mount split in the E
plane. The complete waveguide mount can be fabricated
using computer-controlled milling or, for high-production
quantities, using plastic injection molding and electroplat-
ing. Some disadvantages of this arrangement are in-
creased losses due to the repeated tapers from waveguide
to nline and possibly some problems with interaction of
components over greater distances causing an increased
ripple in transmission and return loss behavior. Integrated
front ends of this type have been realized for communica-
tion equipment in the Ka band [27], for military surveil-
lance receivers [28], and for radar front ends for military
seeker heads [29] or automotive radars for intelligent
cruise control [30]. For example, Barnes et al. [27], de-
scribe a transmitter front end for a 29-GHz communication
system that includes an E-plane waveguide circulator, two
couplers, a pin diode biphase modulator, and a pin diode
attenuator. The power is generated by an external Gunn
oscillator. An E-plane, metal insert, bandpass lter is add-
ed to this unit with an additional waveguide mount.
An even higher degree of integration is achieved by in-
tegrating several components on a single substrate with
direct nline interconnects, that is, without transitions
from nline to waveguide and back to nline. This tech-
nique results much smaller components with reduced loss-
es. Furthermore, the single components are placed close
together, thus avoiding strong phase variations from mu-
tual interactions as a function of frequency. On the other
hand, the single components in such a supercomponent
can no longer be tested separately.
A number of integrated front ends of this type have
been realized by the author and his group, starting with
the integration of balanced mixer and pin diode attenua-
tors acting as sensitivity time control (STC) for radar ap-
plications [31] or as Dicke switch for a radiometer [32].
Figure 16 shows the opened waveguide mount with a
quartz substrate containing a 94-GHz mixer and a SPDT
switch. Two diodes in each of the two input arms of the
circuit are mounted with opposite polarities. Thus only a
single control input for the Dicke switch is necessary. Ef-
forts were also made to integrate Gunn oscillators as
shown previously (Fig. 15). In Ref. 32, the integration of
a mixer with a two-diode series type of pin diode attenu-
ator (Fig. 6b) and a Gunn oscillator as LO is demonstrated
(see photograph of Fig. 17). In this example, the series
type of pin diode attenuator is used favorably for bias iso-
lation; only the nline metallization between the two
stubs loaded with the diodes must be DC-isolated.
Even the integration of a Ka-band pulse radar sensor on
a single substrate including pulse oscillator and LO, an
SPDTswitch as a transmit/receive switch, a series type pin
diode STC, and a balanced mixer were demonstrated [31].
Figure 16. Photograph of a 94-GHz integrated nline circuit in-
cluding a balanced mixer and a pin diode SPDT. The waveguide
split block is opened to observe the quartz substrate with the
planar circuit.
Figure 17. Photograph of a 60-GHz integrated nline receiver
consisting of a Gunn oscillator as LO, a balanced mixer, and a
series-type pin diode attenuator. The Gunn element is screwed
into a copper heatsink in the waveguide mount; the series pin
diode attenuator allows a DC isolation with respect to the other
components.
For a 94-GHz dual-polarization radar, an integrated
receiver front end was realized using both an integrated
nline circuit (2 mixers, 2 pin diode STCs, and a power
divider on a single substrate) and waveguide components
(E-plane circulator, E-plane metal insert lters, and a
turnstile coupler) [33]. Figure 18 shows a block diagram
of the front end. Figure 19 shows the two receiver conver-
sion loss curves. They include conversion loss of the mix-
ers and the insertion losses of pin diode attenuators, the
E-plane lters, and the E-plane circulator. The overall
system noise gure amounts to 12 dB.
BIBLIOGRAPHY
1. W. J. R. Hoefer, FINLINES (this encyclopedia).
2. R. Vahldieck, Quasi-planar lters for millimeter-wave appli-
cations, IEEE Trans. Microwave Theory Tech. MTT-37:324
334 (1989).
3. Y. C. Shih, T. Itoh, and L. Q. Bui, Computer-aided design of
millimeter-wave E-plane lters, IEEE Trans. Microwave The-
ory Tech. MTT-31:135141 (1983).
4. J. Dittloff and F. Arndt, Rigorous eld theory design of mil-
limeter-wave E-plane integrated circuit multiplexers, IEEE
Trans. Microwave Theory Tech. MTT-37:340350 (1989).
5. H. Callsen and L.-P. Schmidt, Quasiplanar 3-dB hybrid for
mm-wave integrated circuits, Electron. Lett. 18:161163
(1982).
6. E. Kpodzo, K. Schuenemann, and G. Begemann, A quadri-
phase n-line modulator, IEEE Trans. Microwave Theory
Tech. MTT-28:747752 (1980).
7. B. Adelseck, H. Callsen, H. H. Meinel, W. Menzel, and K.
Solbach, A survey of planar integrated mm-wave components,
Radio Electron. Eng. 52:4650 (1982).
8. P. J. Meier, Millimeter integrated circuits suspended in the E-
plane of rectangular waveguide, IEEE Trans. Microwave The-
ory Tech. MTT-26:726733 (1978).
9. A. Beyer and I. Wolff, A n line ferrite isolator and circulator
for the R-band, Proc. 11th European Microwave Conf., Am-
sterdam, The Netherlands, 1981, pp. 321326.
10. M. Braas and C. Schieblich, E-type circulator for n lines,
Electron. Lett. 17:701702 (1981).
11. A. Beyer and K. Solbach, Fin line ferrite isolator for integrat-
ed millimeterwave circuits, Proc. IEEE Int. Microwave Symp.
MTT-S, 1981, pp. 296298.
12. K. Solbach, E-plane circulators aid millimeter-wave design,
Microwaves RF. 7379 (Dec. 1983).
13. K. S. Yngvesson, T. L. Korzeniowski, Y.-S. Kim, E. L.
Kollberg, and J. F. Johansson, The tapered slot antennaa
new integrated element for millimeter-wave applications,
IEEE Trans. Microwave Theory Tech. MTT-37:365374
(1989).
14. H. H. Meinel and H. Callsen, Fin-line detectors for frequen-
cies up to 260GHz, Proc 7th Conf. Infrared and Millimeter-
Waves, Marseilles, France, 1983, Book of Abstracts, p. 34.
15. H. Callsen, H. H. Meinel, and W. J. P. Hoefer, p-i-n diode con-
trol devices in E-plane techniques, IEEE Trans. Microwave
Theory Tech. MTT-37:307316 (1989).
16. D. W. Ball and L. Q. Bui, Wideband mm-wave mixers for EW
applications, Microwave J. 25:6576 (June 1982).
17. W. Menzel, Integrated n-line components for communica-
tion, radar, and radiometer applications, in K. J. Button ed.,
Infrared and Millimeter Waves, Vol. 13, Millimeter Compo-
nents and Techniques, Part. IV, Academic Press, 1985,
pp. 77121.
Turnstile
junction
E-plane
circulator
PIN-diode
STC
Balanced
mixer
Integrated on a single finline substrate
E-plane
metal insert
filter
Figure 18. Block diagram of an integrated dual-channel 94-GHz
radar receiver front end. All components (except for the turnstile
junction) are combined in a single E-plane split-block mount. Bal-
anced mixers, pin diode STCs, and the LO power divider are in-
tegrated on a single substrate.
10
11
12
13
14
15
16
17
18
19
20
93 93.5 94 94.5 95
O
v
e
r
a
l
l

c
o
n
v
e
r
s
i
o
n

l
o
s
s

(
d
B
)
Frequency (GHz)
Figure 19. Overall conversion loss of the two channels of the 94-
GHz radar receiver (see Fig. 18).
18. W. Menzel, A 140GHz balanced mixer for nline integrated
circuits, Proc. 13th European Microwave Conf., Nu rnberg,
Germany, 1983, pp. 179182.
19. P. J. Meier, Wideband subharmonically pumped W-band mix-
er in single ridge nline, Proc. IEEE Int. Microwave Symp.,
1982, pp. 201203.
20. U. Guettig, K. M. Strohm, and F. Schaefer, D-band subhar-
monic mixer with silicon planar doped barrier diodes, IEEE
21. L.-P. Schmidt and W. Menzel, Planar integrated harmonic
mixers for use up to 150GHz, Proc. German Conference on
Microwaves and Optoelectronics (MIOP), Sindelngen, Ger-
many, May 1987, Vol. I, p. 4A4 .
22. W. Thiel and W. Menzel, Full-wave design and optimization of
mm-wave diode-based circuits in nline technique, IEEE
23. W. Thorpe and J. D. Gilliland, 29GHz E-plane biphase mod-
ulator, Electron. Lett. 19:107109 (1983).
24. B. Adelseck and B. Rembold, A 1801 phase modulator in n-
line technique (in German), Wiss. Berichte AEG-Telefunken,
54(4/5):238240 (1981).
25. W. J. R. Hoefer, Oscillators and ampliers in integrated
E-phase techniques, IEEE Trans. Microwave Theory Tech.
MTT-37:351364 (1989).
26. J. Ruxton and W. J. R. Hoefer, A quasi-planar FET amplier
in integrated nline and microstrip techniques, IEEE Trans.
Microwave Theory Tech. MTT-37:429432 (1989).
27. B. C. Barnes, R. N. Bates, and I. M. Clarke, An integrated
E-plane mm-wave 29GHz transceiver, Proc. 14th European
Microwave Conf., Lie`ge, Belgium, 1984, pp. 537542.
28. B. H. Newton, A. F. Dodds, D. M. Nugent, and D. H. Body, A
sensitive K-band receiver for ESM application, Milit. Micro-
wave Conf. (Brighton, UK) (June 1986).
29. M. B. Williams and I. M. Clarke, Radar transceiver for smart-
munition applications, Microwave System News & Commu-
nication Technologies, 48 (Dec. 1987).
30. A. G. Stove, Automotive radar at 8090GHz, Proc. IEEE Int.
Microwave Symp. MTT-S, 1992, pp. 613616.
31. W. Menzel, H. Callsen, K. Solbach, and H. Meinel, Integrated
Ka-band radar increases target recognition, Defense Electron.
9599 (Oct. 1982).
32. W. Menzel and H. Callsen, Integrated n-line components
and subsystems at 60 and 94 GHz, IEEE Trans. Microwave
Theory Tech. MTT-31:142146 (1983).
33. W. Menzel, H. Callsen, and K. Solbach, A integrated receiver
frontend for a 94GHz dual polarization radar, Proc. 13th Euro-
pean Microwave Conf., 1983, Nu rnberg, Germany, pp. 142147.
FINLINES
WOLFGANG J. R. HOEFER
University of Victoria
Victoria, British Columbia
Canada
RUEDIGER VAHLDIECK
Institut fu r Feldtheorie und
Hochstfrequenztechnik
Finlines are guiding structures for electromagnetic waves
that consist of one or several metallic ns suspended in
the E plane of a waveguide enclosure. They are the basic
transmission media for the so-called integrated E-plane
circuit technology. The ns can either be suspended freely
(metallic E-plane lines) or supported by dielectric sub-
strates of low permittivity. The general term nline usu-
ally applies to the latter type, and there exist many
variations that are classied according to the manner in
which the metallization is placed on the substrate, and
how the planar structure is mounted in the enclosure.
Finlines are applied predominantly in the frequency
range between 10 and 100GHz; but some applications at
frequencies up to 170GHz have been realized. Having
losses of typically 0.1 dB per wavelength, nlines are not
suitable for long-distance transmission. However, they ex-
cel as a circuit medium for millimeter-wave components
and systems due to their low manufacturing cost, wide
single-mode bandwidth, compatibility with discrete active
and passive devices, and suitability for integration with
other hybrid circuit techniques as well as with standard
rectangular waveguides. Manufacturing costs are moder-
ate for two reasons: (1) the planar insert can be fabricated
using printed-circuit batch processing; (2) since the circuit
characteristics are dominated by the topology of the print-
ed insert, dimensional tolerances for the waveguide en-
closure can be greatly relaxed (typically by a factor of 5
compared with standard waveguide). The enclosure also
prevents radiation losses, reduces susceptibility to outside
interference, and may act as a heatsink for active devices.
Unfortunately, there is a price to pay for the technological
advantages of nlines; their analysis and design are dif-
cult because of the metallic edges that lead to eld singu-
larities, the presence of two or more different dielectrics in
the structure giving rise to hybrid modes of propagation,
and to structural complications necessary for mounting
the substrate and for biasing of devices.
1. A BRIEF HISTORY
Since the early 1940s it has been known that one or two
metallic ridges in the E plane of a rectangular waveguide
increased its single-mode bandwidth and lowered its char-
acteristic impedance. In particular, Cohn [1] pointed out
the potential of this feature for realizing impedance trans-
formers, tapers, lters, slow-wave structures, and broad-
band matched loads by simply changing the ridge depth;
Hopfer [2] published extensive and accurate data for the
cutoff frequencies, characteristic impedances and power-
handling capability of ridged waveguides. While the ridge
waveguide is not truly quasiplanar, its early theoretical
treatment and the manner in which circuit functions were
realized have laid the groundwork for the design and re-
alization of nline circuits. In 1955, Robertson [3] built the
rst truly planar nline component. It was a circular-to-
rectangular waveguide coupler tted with a pair of thin
metal ns. Robertson recognized that this technique was
suitable for very broadband hybrids, directional and po-
larization-sensitive couplers. Sixteen years later, Konishi
and Hoshino [4] reported a 100-GHz converter featuring a
planar metallic circuit in a waveguide. However, the major
development in the evolution of E-plane technology was
FINLINES 1609
Meiers invention of the integrated nline in 1972 [57],
for which he received the Microwave Application Award of
the IEEE-MTT society in 1984. By supporting the planar
conducting ns with a thin, low-permittivity substrate,
Meier extended thin-lm fabrication techniques to E-
plane circuitry and gained additional advantages such
as DC isolation of the ns for bias purposes, intermediate-
frequency (IF) and modulation connections, and mechan-
ical support for discrete devices mounted in parallel or in
series with the ns.
Meiers invention spurred intense international re-
search activities and product development. First commer-
cial nline products were announced in the late 1970s.
They included mixers, broadband switches, attenuators,
detectors, and complete integrated radar front ends for
frequencies up to 140GHz [812].
2. CLASSIFICATION OF FINLINES
Finlines can be viewed as derivatives of the four generic
structures shown in Fig. 1. In fact, they share some of
their characteristics and can be reduced to any of them. A
nline is thus created either by placing a slotline into the
E plane of a rectangular waveguide, by partially loading a
ridge waveguide with dielectric, or by partially metallizing
the dielectric in a slab-loaded waveguide. Four basic types
of nlines may result; their simplied cross sections are
shown in Figs. 2a2d. Insulated, unilateral, bilateral, and
antipodal nlines can be distinguished by the way in
which their ns are arranged on the substrate. Further
variations of each type are obtained by moving the sub-
strate or the slot away from the symmetric position, or by
adding further planar inserts, strips, or slots. For exam-
ple, Figs. 2e and 2f show cross sections of asymmetric n-
lines.
3. CONSTRUCTION OF FINLINES
A nline consists of printed metallic ns on a substrate
(the planar insert) and a metallic housing that holds the
insert and connes the electromagnetic elds. Both soft
and hard substrate materials may be used. Fiberglass-re-
inforced Teflon (RT Duroid, e
r
2.2) is the most widely
used soft substrate material. It is easily processed and cut,
costs little, and has low dielectric permittivity and losses.
It is employed mostly at lower millimetric frequencies; a
thickness of 0.01 in. (0.254 mm) is suitable for components
up to 40 GHz, and 0.005 in (0.127mm) is used for frequen-
cies up to 90 GHz. Finline components and circuits of
highest quality and for frequencies beyond 90 GHz call
for quartz substrates (e
r
3.75) which are more expensive,
fragile, and difcult to process, but provide better me-
chanical support for delicate semiconductor devices. The
housing is usually machined from brass or aluminum in
the form of a split block. Metallized Plexiglas has also been
used with mixed results because of adhesion problems.
Three housing congurations (Fig. 3) can be found in
practice.
The rst type has been proposed by Meier [6]. Two II-
shaped metal pieces clamp the planar insert (Fig. 3a) held
in place by nylon screws. Since the substrate extends be-
yond the connes of the inner waveguide channel, this
conguration is partially open. To minimize RF leakage,
the clamping region is made l
d
/4 wide, where l
d
is the
wavelength in the dielectric substrate at midband fre-
quency. Furthermore, to suppress the excitation of TEM-
type modes in the slot region, longitudinal current ow
must be impeded by serrating the substrate metallization
in the slot region [42]. On the other hand, since one of the
ns can be isolated from the housing with a dielectric in-
sert, bias is easily supplied to embedded semiconductor
devices.
The second type has been proposed by Adelseck et al.
[8]. Here, the two halves of the split block touch each oth-
er, thus completely shielding the nline (Fig. 3b). A shal-
low groove is machined into one of the blocks to
accommodate the substrate. However, stringent mechan-
ical tolerances (typically 710 mm) must be maintained
o
(a)
o

o
r
(c)
o
(b)
Metal
o
o
r
(d)
Figure 1. The four generic waveguides from which nlines can
be derived: (a) rectangular waveguide; (b) double-ridged wave-
guide; (c) slab-loaded waveguide; (d) slotline.
o
(c)
o
(d)
o
(e)
o
(f)
o
(a)
r
(b)
Fins
Substrate Housing
Figure 2. The four basic nline types and some of their variants:
(a) insulated nline; (b) unilateral nline; (c) bilateral nline; (d)
antipodal nline; (e) asymmetric bilateral nline; (f) asymmetric
antipodal nline.
1610 FINLINES
here. If the groove is too wide, poor contact between the
ns and the housing can introduce considerable attenua-
tion. If the groove is too narrow, hard substrates may
break, and soft substrates may be compressed and pushed
out into the waveguide channel, causing bending and pos-
sibly rupture of delicate leads of devices. Bias can be fed
locally to the clamping region through the metal block,
and ns can be insulated from the housing by inserting a
thin Mylar sheet into the groove.
The third type combines the features of both approach-
es (Fig. 3c). It is suitable for quartz substrates at high
frequencies (90 GHz and above), where machining toler-
ances become prohibitive because of the small size of the
groove. As in type 1, the walls of the waveguide channel
are made l
d
/4 wide to form a choke, but the substrate
protrudes on either side into a secondary channel that
contains RF absorbing material. In this way the nline is
sealed. Fins can be biased by inserting a thin insulating
sheet between ns and housing, and by contacting the ns
through a feed connection.
All three techniques have been used successfully in the
realization of nline components. Acceleration tests have
shown that components of types 2 and 3 with hard sub-
strates can survive up to 30,000g for 2 ms, while compo-
nents with soft substrates withstand 2000 g for 0.5ms or
shocks of up to 4000 g [9].
4. BASIC ELECTROMAGNETIC PROPERTIES OF FINLINES
The most important properties are the propagation con-
stant and the characteristic impedance associated with
the dominant mode of propagation. These depend on the
geometry of the cross section, the substrate permittivity,
and the frequency of operation. Furthermore, the single-
mode bandwidth is proportional to the separation between
the cutoff frequencies of the dominant and the rst higher-
order mode. If nline discontinuities are to be analyzed
using mode matching in the longitudinal direction, a sig-
nificant number of higher-order modes must be charac-
terized as well. While the generic topologies in Fig. 2 yield
the characteristics of real nlines quite accurately at low-
er frequencies, the inuence of the geometric details
shown in Fig. 3, such as the substrate mounting grooves
and the nite thickness of the metal ns, must be included
in the analysis and design of practical components, par-
ticularly at higher frequencies where the size of these fea-
tures is not negligible vis-a` -vis the dimensions of the
waveguide channel and the substrate thickness.
Similar to regular waveguides, nlines are dispersive;
that is, their guided wavelength l
g
changes nonlinearly
with frequency. In waveguides without partial dielectric
lling, the guided wavelength l
g
(or propagation constant
b) at any frequency above cutoff can be calculated from the
knowledge of the cutoff wavelength l
c
with the well-known
formula
l
g
o
b

l
1 l=l
c
2
q 1
However, as soon as the waveguide is partially lled with
a dielectric material, this formula no longer applies. With
increasing frequency, more of the wave energy tends
to concentrate inside the material with the higher permit-
tivity, slowing down the propagation velocity. This effect
is not included in Eq. (1). To accurately determine the
guided wavelength in a nline at any frequency above
cutoff, the eld problem must be solved separately for each
frequency.
o
Fins
o
Fins
Fins
Planar insert Insulating gasket
Nylon
screws
Split block
housing
Planar insert
RF absorber
Metal gasket
(a)
(b)
(c)
d
/4
<<
d
d
/4
Figure 3. Three types of nline construction: (a) nline mount
proposed by Meier; (b) nline mount proposed by Adelseck et al.
(c) combination of (a) and (b) for hard substrates and high fre-
quencies. (After Meier [6], copyright r1974 IEEE, reprinted with
permission.)
FINLINES 1611
As in all waveguides carrying non-TEM waves, voltage
and characteristic impedance are not uniquely dened.
While the voltage is usually obtained by integrating the
electric eld in the slot along the shortest path on the
substrate surface, impedance definitions vary with the
application, and the most appropriate definition must be
established on a case-by-case basis. Meinel and Rembold
[10] found that the voltagecurrent definition Z
v,i
(volt-
age across the slot)/(current in the ns) was most appro-
priate for predicting the interaction of nlines with
discrete devices, while Knorr and Shayda [14] reported
best results with the ridged waveguide definition and
Willing and Spielmann [15] preferred the voltagepower
definition Z
v,p
(voltage across the slot)/(2 average
power transmitted).
There exist a variety of techniques for the computation
of nline parameters, ranging from approximate analyti-
cal expressions to sophisticated numerical procedures.
They will be outlined briefly in the following. For detailed
information the reader is invited to consult to the appro-
priate references given in the bibliography.
4.1. Approximate Methods of Finline Analysis
Since most nlines are made with low-permittivity sub-
strates such as RT Duroid (e
r
2.2) or quartz (e
r
3.75)
that occupies only a small fraction of the waveguide cross
section, their dispersion characteristics can be approxi-
mated by the behavior of commensurate waveguides that
are lled uniformly with a ctitious frequency-indepen-
dent dielectric of permittivity e
0
e
re
(Fig. 4). (The permit-
tivity of this ctitious dielectric e
re
is called the equivalent
dielectric constant to distinguish it from the so-called
effective dielectric constant e
eff
(l/l
g
)
2
of a nline.) The
uniformly lled ridge waveguide obeys a rather simple
dispersion formula for the guided wavelength l
g
and the
characteristic impedance Z
0
:
l
g
e
re
l=l
c
2
q ; Z
0
Z
1
e
re
l=l
c
2
q 2
According to Meier [6], these expressions agree with mea-
surements within 72% when l
c
and Z
N
are the cutoff
wavelength and characteristic impedance at innite fre-
quency of an air-lled ridge waveguide of identical dimen-
sions (available from Cohns [1] or Hopfers [2] paper, or
from other published data), and e
re
has been determined
by measuring the guided wavelength of a test structure at
a single frequency in mid-band [6].
More accurate results can be obtained by the trans-
verse resonance approach that has previously been ap-
plied by Cohn [16] and other researchers to analyze ridge
waveguides, slab-loaded waveguides, and slotlines. In the
simplest formulation, the dominant mode in the nline is
considered to be a TE mode in the transverse direction
that resonates between the two sidewalls of the enclosure.
In transverse direction, the nline thus appears as a chain
of piecewise homogeneous waveguide sections of width
l
g
/2 (l
g
is the wavelength in longitudinal direction of the
nline and corresponds to the cutoff wavelength of the
transverse waveguide sections). The ns form a capacitive
transverse discontinuity. The characteristic admittances
and propagation constants of the subsections are those of
homogeneous waveguides. The corresponding equivalent
circuit is shown in Fig. 5; its transverse resonance condi-
tion is obtained by setting the total admittance in the
plane of the ns equal to zero
B
Y
a
cot b
a
h
Y
d
Y
a
cot b
d
d
1
b
d
tan
1
Y
d
Y
a
tan b
a
a d h
! & '
0
3
where
b
a
2p
l
1 l=l
g
2
q
;
b
d
2p
l
e
r
l=l
g
2
q
and
Y
d
Y
a
b
d
b
a
By xing the guided wavelength l
g
and solving Eq. (3) for
l, one obtains the frequencies at which the dominant and
higher-order modes propagate down the nline with that
wavelength l
g
. The accuracy of the solution depends es-
sentially on the accuracy of the capacitive susceptance B,
and the various transverse resonance approaches report-
ed in the literature differ in the way in which this
susceptance is determined. They range from quasistatic
approximations to variational expressions and mode-
matching solutions [1720]. The group of approximate
methods also comprises the empirical closed-form equa-
tions that are essentially formulas tted to more accurate
numerical results [21,22]. Their advantage resides in their
simplicity and their suitability for synthesis, but they are
limited to certain combinations of parameters and have
lost some of their attractiveness due to the proliferation of
high-performance computers that yield more rigorous
numerical solutions almost instantly.
o
(a)
re
(b)
r
Figure 4. Insulated nline and its equivalent ridged waveguide
with identical cross section and uniform dielectric lling of per-
mittivity e
0
e
re
.
r
(a)
d
w
o
(b)
h
a
b
Y
a
a

d

a
Y
a
jB
Y
d
Figure 5. (a) Unilateral nline; (b) and its equivalent transverse
network.
1612 FINLINES
4.2. Rigorous Methods of Finline Analysis
Rigorous methods yield eld solutions by solving Max-
wells equations subject to the boundary conditions im-
posed by the nline geometry. Several are based on the
method of moments, from the rst formulation in the
space domain by Hofmann [23] to the spectral domain ap-
proaches described by Itoh [24], Knorr and Shayda [14],
and Schmidt and Itoh [25]. A mode-matching solution was
proposed by Beyer and Wolff [61], and a modied mode-
matching approach was presented by Vahldieck and
Bornemann [26,27]. A singular integral equation formu-
lation was proposed by Omar and Schuenemann [28].
In all these approaches the elds in the dielectric and
air-lled subregions of the nline are expressed in terms
of rectangular waveguide TE and TM modes resonating in
transverse direction, also called longitudinal section elec-
tric (LSE) and longitudinal section magnetic (LSM)
modes. In fact, each of these modes must satisfy a condi-
tion similar to Eq. (3). The modal coefcients are deter-
mined such that their superposition satises all boundary
and interface conditions, in particular, the edge condition
at the metal ns where most of the energy in the structure
is concentrated. Approaches differ in the way in which the
relationship between slot eld and n current distribu-
tions is formulated. In the space domain formulation, this
relationship takes the form of integral equations. Their
kernels are the Green functions of the housing and sub-
strate without the metal ns. These integral equations are
then solved with the method of moments and Galerkins
procedure (identical basis and testing functions are used).
By performing a spatial Fourier transformation of all
elds and currents with respect to the coordinates parallel
to the substrate plane, one can transform the integral
equations into a set of algebraic equations. The resulting
spectral-domain formulation leads to a very efcient algo-
rithm for the dominant and rst higher-order nline
eigensolutions and eigenvectors, especially when the slot
elds or n currents are developed into physically realistic
basis functions that individually satisfy all the boundary
and edge conditions in the plane of the ns. In fact, only a
single constant basis function for the slot eld is sufcient
to determine the dominant wavenumber within 23%.
However, the spectral-domain method is restricted to ide-
alized nline geometries (no mounting grooves, zero met-
allization thickness). It is also less suitable than the
singular integral formulation for determining eigensolu-
tions of high order because the set of required basis func-
tions for the slot eld or n current grows quickly in size
and complexity.
The general mode matching approach that enforces
conservation of tangential eld components or complex
power across all boundaries between subregions can take
these realistic features into account more readily, as can
the well-known family of space discrete solution methods,
such as the nite-difference, nite-element, and TLM
methods of analysis, in both their frequency and time-
domain formulations, or the method of lines. A detailed
description of these methods with extensive bibliographies
can be found in Refs. 29 and 30. However, the salient fea-
tures of the most frequently employed rigorous methods of
nline analysis, namely, the spectral-domain and the
mode-matching methods, will be summarized in the fol-
lowing, together with some typical results.
4.2.1. Spectral-Domain Analysis of Finlines. In the spec-
tral-domain formulation the Fourier transform of the
Green functions
~
GG
ij
in the nline (expressed in terms of
LSE and LSM modes) relates the transforms of the electric
eld in the slot with those of the current density on the
ns via the equation
~
GG
11
a
n
; b; k
0
~
GG
12
a
n
; b; k
0
~
GG
21
a
n
; b; k
0
~
GG
22
a
n
; b; k
0
2
6
4
3
7
5
~
EE
x
a
n
~
EE
z
a
n
2
6
4
3
7
5
~
JJ
x
a
n
~
JJ
z
a
n
2
6
4
3
7
5
~
GG
11
a
2
n
Z
LSE
b
2
n
Z
LSM
=a
2
n
b
2
n
~
GG
12
~
GG
21
a
n
b
n
Z
LSE
Z
LSM
=a
2
n
b
2
n
~
GG
22
a
2
n
Z
LSE
b
2
n
Z
LSM
=a
2
n
b
2
n
4
where Z
LSE
and Z
LSM
are the impedances seen from the
nodes of the current source, looking into the respective
equivalent circuits of Fig. 6b and 6c. a
n
is the Fourier
variable, b2p/l
g
is the propagation constant in the axial
direction, and k
0
is the free-space wavenumber;
~
EE
x
,
~
EE
z
,
~
JJ
x
,
and
~
JJ
z
are the electric elds in the slot and the current
densities on the ns, respectively. This relationship is
most readily formulated using Itohs [24] immittance ap-
proach. Figure 6 shows the cross section of a unilateral
nline, together with the equivalent transverse modal cir-
cuits for the immittance formulation of the Green func-
tions. The slot elds and n currents are expressed in
terms of known basis functions with unknown coefcients.
Applying Galerkins procedure and Parcevals theorem re-
sults in a set of algebraic equations. The roots of the char-
acteristic equation yield the eigenvalues of the nline.
The method is particularly efcient when basis func-
tions are selected that satisfy already the edge condition
at the ns. Thus, only a few basis functions are necessary
to accurately describe the eld in the slot or the current
density on the ns. Once the coefcients of the basis
(a)
d
w
y
(b)
h
a
b
Y
LSE, 2
2,
Y
LSE, 2
2,
Y
LSE, 1
1, J
~
v
(h)
2
1
2
(c)
Y
LSM, 2
2,
Y
LSM, 2
2,
Y
LSM, 1
1, J
~
u
(h)
Figure 6. Cross section of a unilateral nline (a) and its equiv-
alent transverse modal circuits for LSE (b) and LSM (c) modes.
FINLINES 1613
functions are determined, the eld functions inside the
structure can be derived, and quantities such as slot volt-
age, n current and characteristic impedances can be de-
termined by appropriate integration over the eld
functions. Figure 7 shows the effective dielectric constant
and voltagepower impedance of the unilateral nline in
Fig. 6, computed with a single constant basis function for E
x
in the slot. The enclosure is a standard WR-28 rectangular
waveguide.
As mentioned earlier, geometric details such as the
substrate mounting grooves and nite metallization thick-
ness must be accounted for in practical realizations, par-
ticularly at higher frequencies. These can easily be
included in the mode-matching analysis. Representative
nline characteristics obtained with this approach will be
presented in the following as well.
Two typical nline structures with realistic geometric
details are presented in Fig. 8. Depending on the applica-
tion, structures that support quasi-TEM modes (Fig. 8a)
or congurations with pure slotline layout (Fig. 8b) are
utilized. The principal difference between the two types is
that in the rst the dominant mode of propagation is the
quasi-TEM mode without a lower cutoff frequency, while
in the slotline structure the dominant mode is limited by a
cutoff frequency that is determined mainly by the width of
the slot(s).
Both types of nlines carry hybrid modes due to the
presence of the dielectric substrate that supports the ns.
Hybrid modes are neither TE nor TM but a combination of
both, denoted as either HE or EH modes with all six eld
components present; HE modes are those with a predom-
inant H
z
eld component, while EH modes exhibit a
stronger E
z
eld component. Only at cutoff are these
modes purely of type TE or TM. These modes often carry
the same indices as TE and TM modes, such as HE
nm
or
EH
nm
, indicating that the latter two originate from the
corresponding TE or TM modes at cutoff. At frequencies
above cutoff, the eld conguration of the individual hy-
brid modes can differ significantly from the TE and TM
modes. Therefore, hybrid modes are equally often num-
bered according to the sequence at which they occur, such
as HE
n
and EH
m
.
In the mode-matching approach, the nite metallizat-
ion thickness and substrate mounting grooves are taken
into account from the outset. It yields a somewhat longer
guided wavelength than spectral domain analysis since a
higher percentage of the electric eld travels in air within
the slot region. The hybrid modes are derived from the z
components of the magnetic and electric vector potentials
~
ff
h
and
~
yy
e
, respectively
~
EEr~jj
hz
1
joe
rr
~
yy
ez
~
HHr
~
yy
ez

1
joe
rr~jj
hz
5
where
j
hz
X
n
Q
h
n
xf
n
ye
jb
n
z
y
ez
X
n
P
e
n
xg
n
ye
jb
n
z
6
Q, P, f and g denote transverse eigenfunctions dened in
each subregion of the nline cross section (i.e., regions 1,
2a, 2b, 3, 4 in Fig. 8a). The electric and magnetic elds
tangential to the interfaces between the subregions are
proportional to the vector potentials and their derivatives
E
y
/
dQ
h
dx
P
h
; E
z
/ P
e
; H
z
/ Q
h
; H
y
/
dP
e
dx
Q
e
7
Matching the tangential eld components at each inter-
face and utilizing the orthogonality of modes leads to one
coupling matrix (Cm for m1,2a,2b,3,4) per interface. The
coupling matrices can be linked successively by transfer-
ring the electric and magnetic elds from one boundary of
a subregion to the opposite one. These transfer matrices
(T
m
for m 1,2a,2b,3,4) are the generalized (fundamental
and higher-order modes) transmission-line equations in
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
15 GHz
0 0.2 0.4
w/b
0.6 0.8 1
15 GHz
20 GHz
20 GHz
25 GHz
25 GHz
30 GHz
30 GHz
40 GHz
40 GHz
600
500
400
300
200
100
0
V
o
l
t
a
g
e
p
o
w
e
r

i
m
p
e
d
a
n
c
e

i
n

e
f
f

=

(

/

g
)
2
eff
Z
L
/
Figure 7. Effective dielectric constant and voltagepower imped-
ance of the unilateral nline shown in Fig. 6, computed with the
spectraldomain method using a single constant basis function for
the slot eld E
x
. The dimensions are: a7.112mm; b 3.556mm;
h3.429mm; d0.254mm; e
r
2.2.
(a) (b)
Housing
2a
2b
3
4
Substrate
1 b
a
e
t
Figure 8. Two basic nline structures with nite metallization
and substrate mounting grooves (t nite metallization thick-
ness, e depth of mounting grooves, awaveguide width, b
waveguide height): (a) edge-coupled unilateral nline; (b) asym-
metric bilateral nline.
1614 FINLINES
each subregion. The resulting matrix equation yields
P
h
0
P
e
0
Q
h
Q
e
2
6
6
6
6
6
4
3
7
7
7
7
7
5
x 0
Y
n3
n1
T
n
C
n
T
4
|{z}
G
P
h
0
P
e
0
Q
h
Q
e
2
6
6
6
6
6
4
3
7
7
7
7
7
5
x a
8
From the resonance condition in Eq. (8) the characteristic
equation G
12
is obtained. Solving for det(G
12
) 0 yields
the propagation constant b or guided wavelength l
g
. Typ-
ical dispersion characteristics obtained in this way for the
rst six eigenmodes in a unilateral nline are shown in
Fig. 9. The modes that are not excited by an incident TE
10
mode are dashed. The rst important higher-order mode is
thus HE
3
; the single-mode bandwidth of the dominant
mode is situated between the cutoff frequencies of HE
1
and
HE
3
(frequencies at which the curves intersect the fre-
quency axis). Note that these dispersion curves change
with both the groove depth e and the metallization thick-
ness t as shown in the following.
4.2.2. Effect of Mounting Grooves. Mounting grooves
are required to hold the substrate in the waveguide hous-
ing. A detailed analysis of the effect of the groove depth on
the modes is given in Refs. 26 and 31. It was found that for
unilateral nlines the cutoff frequencies of higher-order
modes react very sensitively to the groove depth (e in
Fig. 8), which for one-fth of the waveguide height reduces
the cutoff frequency of higher-order modes to such an ex-
tent that the nline is useless for most practical purposes.
This is due to modes that originate from the TE
20
(HE
3
)
and TE
01
(HE
2
) modes of the empty waveguide. Both
modes appear to be coupled, are highly concentrated be-
tween the metallization within the substrate, and are thus
very sensitive to changes in the groove depth. This effect is
particularly pronounced in Fig. 10. The asymmetric bilat-
eral nline carries an additional metallic strip inside the
gap on the right-hand side of the substrate. For centered
n-lines this interaction between modes is greatly reduced
since some of them will not be excited. This is not the case
for unilateral nlines. However, since the metallization is
only on one side of the substrate, the problem is not as
pronounced.
4.2.3. Effect of Finite Metallization Thickness. The effect
of nite metallization on the propagation constant and the
characteristic impedance has been investigated by several
researchers [26,3133,62]. A typical behavior of the dom-
inant mode in a unilateral nline with three different val-
ues of metallization thickness is shown in Fig. 11. The
effect of mounting grooves is not included here. The char-
acteristic impedance is reduced throughout the operating
bandwidth when the metallization becomes thicker. This
is consistent with the behavior of the ridge waveguide
[1,2], where the capacitive loading of the waveguide in-
creases when the ridge becomes wider. However, the prop-
agation constant is not affected in the same way at lower
and at higher frequencies. While the cutoff frequency is
reduced by thicker metallization for the same reason as
the characteristic impedance, the effect is reversed at
higher frequencies; because of the wider airgap between
thicker ns, more of the eld energy travels in the airgap
and less is concentrated in the dielectric, leading to a
higher phase velocity. As a result, there is a crossover of
the dispersion curves at midband. In conclusion, the effect
2
1
0
HE
1
HE
3
HE
4
HE
6
50 100
F/GHz
150 200
g o
/
a
e
HE
2
HE
5
r
b
Figure 9. Dispersion characteristics of the rst six hybrid eigen-
modes in a unilateral nline. The dashed curves indicate modes
that are not excited by the incident TE
10
mode. The dimensions
are a1.65mm; b a/2, e 0.5 mm; t 5 mm, e
r
3.75. (After Va-
hldiec [26], copyright r 1984 IEEE, reprinted with permission.)
3
2
1
0
HE
1
EH
1
HE
3
HE
4
EH
0
10 0 20 30 40 50
f/GHz
g
a
e
t
HE
2
HE
5
b
Figure 10. Dispersion characteristics of an asymmetric bilateral
nline with an additional metallic strip in the gap on the right-
hand side of the substrate. The modes appear to be strongly cou-
pled and are very sensitive to changes in the groove depth. The
dimensions are a7.112 mm; b 3.556mm; e 0.7 mm; t
71mm, e
r
10. (After Vahldieck and Bornemann [27], copyright
r 1985 IEEE, reprinted with permission.)
FINLINES 1615
of nite metallization on the propagation constant is min-
imal around the center of the operating bandwidth.
5. FINLINE DISCONTINUITIES
Discontinuities in the nline geometry form the building
blocks of nline circuits. Given the many degrees of free-
dom in the design of nlines, the possibilities are endless,
and it would be impossible to give a complete treatment of
this subject. Nevertheless, one can identify some generic
types of nline discontinuities and provide some basic
characteristics. Some of the most important are inductive
strips, step changes in gap width, notches, dents, and se-
ries stubs. The shapes of typical discontinuities are shown
in Fig. 12, together with their equivalent lumped-element
networks. Note that, in contrast to similar waveguide dis-
continuities, the circuit elements are determined solely by
the metallization pattern of the planar insert, while the
geometry of the enclosure remains unchanged. This re-
sults in a considerable reduction in the manufacturing
cost, as the discontinuities are realized using printed-
circuit techniques rather than machining.
The effect of discontinuities on the guided wave in the
nline can be represented either by generalized scattering
parameters (S parameters) or by equivalent lumped net-
works. These are not determined solely by the local chang-
es of the dominant waveguiding parameters, but include
the effects of mode conversion, energy storage, and possi-
bly multimode interactions between neighboring disconti-
nuities. Within the single-mode propagation range,
dominant-mode S parameters are sufcient to describe
the behavior of discontinuities, provided that they have
been determined at reference planes beyond the reach of
evanescent higher-order modes.
The theoretical characterization of nline discontinu-
ities is a challenge for the same reasons as the computa-
tion of nline characteristics. Inhomogeneous dielectric
lling and edge singularities call for special approaches,
ranging from experimental [6,3436] over variational [37]
and spectral domain techniques [35,38,39] to mode match-
ing in longitudinal [4042] as well as transversal [43] di-
rections.
5.1. Inductive Strips
Inductive strips are important elements of bandpass l-
ters. Meier [6] has determined the equivalent shunt in-
ductance of strips from insertion loss measurements and
from the resonant frequency of a one-section lter. Pic and
Hoefer [34] have characterized strips and impedance steps
by measuring the shift they caused in the resonant fre-
quencies of an oversized cavity. The resulting empirical
formulas for the strip inductance were conrmed by spec-
tral-domain evaluations performed by Koster and Jansen
1.5
1.0
0.5
0.0
500
400
300
200
t = 0
t = 35 m
t = 100 m
Z/
10 20 30 40
Z
50 60 70 80
f/GHz

g
w r
Figure 11. Effect of metallization thickness on the dispersion

characteristics and the characteristic impedance of a unilateral
nline. The dimensions are a4.776mm; b2.388mm; d
0.127mm; w0.2b; e
r
3.8. (After Kitazawa and Mittra [32],
copyright r 1984 IEEE, reprinted with permission.)
(a)
B A B
L
A
l l
(c)
B A B A
C
p
C
p
L
s
(b)
1 : n
2
A A
L
A
(d)
B A B A
C
p
L
s
L
s
(e)
Metallization Substrate
B A
B A
C
p
Z
e
I
e
C
p
Figure 12. Prole of some important nline discontinuities and
their equivalent lumped-element circuits: (a) inductive strip; (b)
impedance step (step change in gap width); (c) inductive notch; (d)
capacitive strip; (e) series stub. Note that the distance between
planes A and B is short compared to the wavelength.
1616 FINLINES
[39]. Further methodologies for the characterization of in-
ductive strips and related discontinuities can be found in
papers by Knorr and Deal [35], Zhang and Itoh [38], Saad
and Schuenemann [44], and Omar and Schuenemann
[45]. To achieve maximum Q factors in bandpass lters
formed by inductive strips, the gap width of the resonant
sections is usually maximized to the full height b of the
enclosure. Such full-height strips and their interactions
have been analyzed extensively with mode-matching tech-
niques by Arndt et al. [46] and Shih et al. [47], leading to
highly accurate CAD programs for E-plane bandpass l-
ters.
5.2. Single and Interacting Impedance Steps
These discontinuities are formed by sudden changes in the
gap width and come in many shapes and forms (Figs. 12b
12d). They serve as impedance transformers, tuning and
matching elements, mounting platforms for semiconduc-
tor devices, or constitutive elements of lters. Besides em-
pirical characterization [34], most systematic analytical
approaches employ the representation of elds near the
discontinuities by eigenmodes of either the adjacent uni-
form nline subsections (mode matching in longitudinal
direction) or of transverse homogeneous waveguide sub-
sections. As mentioned earlier, the former approach re-
quires the determination of the eigensolutions of the
nline subsections. It has been pointed out by Omar and
Schuenemann [45] that the spectrum of nline modes can
contain pairs of complex eigensolutions, and that their
omission in the modal decomposition of the elds at the
discontinuity can lead to significant errors. Various ana-
lyses and properties of impedance steps have also been
described by El Hennawy and Schuenemann [40,41], He-
lard et al. [42], Sorrentino and Itoh [43], and Beyer [48].
5.3. Narrow Series Slots
The series slot (Fig. 12e) is a popular tuning element in
the design of oscillators and lowpass lters. It can be rep-
resented by a series-connected transmission line of effec-
tive length l
e
and two stray capacitances that account for
the eld energy stored in the stray eld at the junction.
The effective length includes the stray effect at the end of
the slot. Burton and Hoofer [49] have characterized such
slots experimentally and have given closed-form expres-
sions for the elements of the equivalent circuit in Fig. 2e.
These expressions are valid over an entire waveguide
band and are independent of the dimensions of the enclo-
sure. This indicates that narrow slots in the nline met-
allization behave essentially like slotline elements [16].
6. TRANSITIONS TO OTHER TRANSMISSION MEDIA
Most practical nline circuits include one or several types
of transitions, either between different types of nlines or
between nlines and other transmission media such as
rectangular waveguides, coaxial line, or microstrip. Of
particular importance and interest is the transition be-
tween nline and the commensurate rectangular wave-
guide due to the need to interface nline circuits with
measurement systems and conventional components.
Transitions between different nline types and also be-
tween nlines and other planar circuits are needed to
combine them and to take advantage of their respective
properties in an integrated design. The topologies and
properties of such transitions will be summarized in the
following.
6.1. Transitions Between Finline and Waveguide
Transitions between nline and the commensurate rect-
angular waveguide are realized by varying in some way
the slot between the full height of the waveguide and the
nal slot width. The bandwidth and the VSWR of the
transition depend entirely on the prole of the n contour
and the length of the transition. The geometrically sim-
plest transition is a linear taper extending over at least
three guided wavelengths, as used by Cohen and Meier
[13]. More sophisticated designs that are shorter than a
wavelength employ single or double exponential proles,
cosine proles, parabolic and circular proles, as well as
other proles derived by nonuniform transmission-line
theory [5053]. Figure 13 shows some typical taper pro-
les employed for the realization of nline-to-wave-guide
transitions.
The problem of taper optimization amounts to realizing
the lowest possible insertion loss over a given frequency
band with a taper of minimum length. A rigorous treat-
ment of the tapered transition leads to a nonlinear differ-
Metallization Substrate
(d)
(c)
(b)
(a)
Figure 13. Prole of typical nline tapers and matching ele-
ments for the reduction of reections from the dielectric slab in-
terface: (a) linear taper with quarter-wave substrate protrusion:
(b) exponential taper with quarter-wave substrate notch; (c) ex-
ponential taper with printer quarter-wave section; (d) double
circular prole taper with triangular substrate protrusion.
FINLINES 1617
ential equation of the Riccati type that must be solved
numerically. Approximate solutions can be found, howev-
er, yielding the so-called DolphChebyshev taper (51).
This taper has impedance discontinuities at its extremi-
ties that introduce parasitic reactances. However, since it
is impossible to uniquely dene the local characteristic
impedance of a tapered nline, one might as well select a
prole that approximates the optimal contour. Indeed,
most researchers who have measured the characteristics
of such tapers agree that the prole is not very critical,
especially when the taper is about a wavelength long or
more.
Detailed formulas and instructions for the synthesis of
nline tapers have been reported by Hinken [50], Schie-
blich et al. [51], Beyer and Wolff [52], and Pramanick and
Bhartia [53]. A transition comprising a short nline taper
with a circular contour and matching capacitive disconti-
nuities has been proposed by de Ronde [54], who claims
that its performance is similar to that of an optimized
taper, but its length does not exceed
3
8
ths of l
max
. Typical
nline tapers have a return loss better than 20 dB over a
standard waveguide band.
One feature that cannot be neglected in most applica-
tions is the discontinuity introduced by the nline sub-
strate itself.
When the slotwidth is equal to the waveguide height,
the nline becomes a dielectric slab-loaded waveguide.
The reection due to the substrate alone deteriorates the
return loss of a subsequent taper typically by 5 dB. Cohen
and Meier [13] reduced these spurious reections by cut-
ting a quarter-wave notch into the dielectric substrate
with dimensions determined experimentally. Verver and
Hoefer [55] later developed a closed-formed expression for
the design of such notches and their dual, dielectric pro-
trusions. Drawings by Piotrowski and Wojtasiak [56] and
Beyer et al. [36] show transitions with triangular protru-
sions of the substrate, but no design information is given.
For hard substrates that are difcult to cut (quartz), a
printed transformer structure consisting of one or even
two quarterwave sections in front of the taper is more ap-
propriate [57]. Figure 13 shows how such matching ele-
ments can be incorporated in the design of a nline taper.
6.2. Transitions between Finlines and Other Planar Media
Many transitions between various types of nlines [58] as
well as between nlines and other planar media such as
microstrip, coplanar waveguide, and suspended and in-
verted microstrip have been developed. A good survey of
such transitions can be found in by Bhat and Koul [11].
The topologies are rather difcult to analyze and thus
have been optimized empirically. Care must be taken not
to excite unwanted modes and to avoid resonances that
can considerably affect the transfer function of such tran-
sitions [59].
7. CONCLUSION
The principal advantages of nline technology at milli-
metric frequencies are wide single-mode bandwidth, mod-
erate insertion loss, freedom from radiation and
interference, suitability for batch processing and low-
cost production in both small and large quantities, poten-
tial for circuit integration, and compatibility with wave-
guide instrumentation. When properly constructed,
nline components are rugged and reliable, withstanding
considerable thermal and mechanical stress. Practically
all imaginable circuit functions have been realized thanks
to the many advances in the eld-theoretic modeling of
quasiplanar circuits. While early design efforts relied
mostly on approximate and empirical characterization of
nlines, rigorous formulations of the eld problem and
their numerical implementations have made it possible to
design, optimize, and fabricate nline components entire-
ly by computer.
The major drawback of the nline is its limited com-
patibility with transistors. On the other hand, active and
passive diodes are easily embedded into a nline; oscilla-
tors, frequency converters, switches, phase shifters and
attenuators with excellent performance have thus been
realized. Entire integrated millimeter-wave radar front
ends for frequencies up to 140GHz have been built by the
nline technique, demonstrating the potential and versa-
tility of this technology that favorably combines the high
electrical quality of the waveguide with the advantages of
planar circuitry. When paired with the available comput-
er-based design tools, nline stands out as an important
and versatile technology in the arsenal of the modern mi-
crowave and millimeter-wave engineer.
8. LITERATURE
The literature on nlines is quite extensive; only a small
number of key references could be cited here. The reader
in search for more detailed information is referred to the
comprehensive survey papers by Solbach [9] and Meier
[60]. Bhartia and Pramanick [12] have assembled a col-
lection of reprints of key papers on E-plane integrated
circuits. Finally, in their book on the analysis, design, and
applications of nlines, Bhat and Koul [11] have gathered
and organized a wealth of design information, methods,
and references on the subject. Further details on numer-
ical techniques for the analysis of microwave circuits and
electromagnetic wave problems can be found in two vol-
umes edited by Itoh [29] and Yamashita [30].
BIBLIOGRAPHY
1. S. B. Cohn, Properties of ridge waveguide, Proc. IRE 35:783
788 (1947).
2. S. Hopfer, The design of ridge waveguides, IRE Trans. Micro-
wave Theory Tech. MTT-3:2029 (1955).
3. S. D. Robertson, The ultra-bandwidth nline coupler, Proc.
IRE 43:739788 (1955).
4. Y. Konishi and N. Hoshino, 100-GHz-band low-noise mixer,
Inst. Electron. Commun. Eng. Jpn. MMW-71 (July 1971).
5. P. J. Meier, Two new integrated-circuit media with special
advantages at millimeter wavelengths, 1972 IEEE MTT Int.
Microwave Symp. Digest, Arlington Heights, IL, 1972, pp.
221223.
1618 FINLINES
6. P. J. Meier, Integrated nline millimeter components, IEEE
7. P. J. Meier, Microwave Transmission Line, U.S. Patent
3,825,863 (1974).
8. B. Adelseck et al., A survey of planar integrated mm-wave
components, Radio Electron. Eng. 45:4650 (1982).
9. K. Solbach, The status of printed millimeter-wave E-plane
circuits, IEEE Trans. Microwave Theory Tech. MTT-31:107
121 (1983).
10. H. Meinel and B. Rembold, New millimeter-wave n-line at-
tenuators and switches, 1979 IEEE MTT-S Int. Microwave
Symp. Digest, Orlando, FL, 1979, pp. 249252.
11. B. Bhat and S. K. Koul, Analysis, Design and Applications of
Fin Lines, Artech House, Norwood, MA, 1987.
12. P. Pramanick and P. Bhartia, E-Plane Integrated Circuits,
Artech House, Norwood, MA, 1987.
13. L. D. Cohen and P. J. Meier, Advances in E-plane printed
millimeter-wave circuits, IEEE MTT-S Int. Microwave Symp.
Digest, 1978, pp. 2729.
14. J. Knorr and P. Shayda, Millimeter-wave n-line character-
istics, IEEE Trans. Microwave Theory Tech. MTT-28:737743
(1983).
15. H. A. Willing and B. E. Spielman, Experimental assessment
of bilateral nline impedance for device matching, 1981 IEEE
MTT-S Int. Microwave Symp. Digest, Los Angeles, CA, 1981,
pp. 105107.
16. S. B. Cohn, Slot-line on a dielectric substrate, IEEE Trans.
17. A. M. K. Saad and G. Begemann, Electrical performance of
n-lines of various congurations, IEE Proc. Microwaves Opt.
Acoust. H 1:8188 (1977).
18. W. J. R. Hoefer, Finline design made easy, 1978 IEEE MTT
Int. Microwave Symp. Digest, Ottawa, ON, 1978, p. 471.
19. R. N. Simons, Analysis of millimetre-wave integrated, n line.
IEE Proc. Microwaves Opt. Acoust. H 130:166169 (1983).
20. J. K. Piotrowski, Accurate and simple formulas for dispersion
in nlines, 1984 IEEE MTT Int. Microwave Symp. Digest, San
Francisco, CA, 1984, pp. 333336.
21. A. K. Sharma and W. J. R. Hoofer, Empirical expressions for
nline design, IEEE Trans. Microwave Theory Tech. MTT-
31:350356 (1983).
22. P. Pramanick and P. Bhartia, Computer-aided design models
for millimeter-wave nlines and suspended substrate micro-
strip lines, IEEE Trans. Microwave Theory Tech. MTT-
35:14291435 (1985).
23. H. Hofmann, Dispersion of planar waveguides for millimeter-
wave applications, Arch. Elek. U
bertragung 31:4044 (1977).

24. T. Itoh, Spectral domain immittance approach for dispersion
characteristics of generalized printed transmission lines,
(1980).
25. L.-P. Schmidt and T. Itoh, Spectral domain analysis of dom-
inant and higher order modes in n-lines, IEEE Trans.
26. R. Vahldieck, Accurate hybrid mode analysis of various n-
line congurations including multilayered dielectrics, nite
metallization thickness and substrate holding grooves, IEEE
27. R. Vahldieck and J. Bornemann, A modied mode-matching
technique and its application to a class of quasi-planar trans-
mission lines, IEEE Trans. Microwave Theory Tech. MTT-
33:916926 (1985).
28. A. Omar and K. Schuenemann, Formulation of the singular
integral equation technique for planar transmission lines,
(1985).
29. T. Itoh, Numerical Techniques for Microwave and Millimeter-
Wave Passive Structures, Wiley, New York, 1989.
30. E. Yamashita, Analysis Methods for Electromagnetic Wave
Problems, Artech House, Norwood, MA, 1990.
31. R. Vahldieck and W. J. R. Hoefer, The inuence of metallizat-
ion thickness and mounting grooves on the characteristics of
n lines, IEEE MTT Int. Microwave Symp. Digest, St. Louis,
MO, 1985, pp. 143144.
32. T. Kitazawa and R. Mittra, Analysis of nline with nite met-
allization thickness IEEE Trans. Microwave Theory Tech.
MTT-32:14841487 (1984).
33. J. K. Piotrowski, Efcient analysis of nline with nite met-
allization thickness, IEEE MTT-S Int. Microwave Symp.
Digest, 1986, pp. 213216.
34. E. Pic and W. J. R. Hoefer, Experimental characterization of
n line discontinuities using resonant techniques, IEEE MTT
Int. Microwave Symp. Digest, Los Angeles, CA, 1981.
35. J. B. Knorr and J. C. Deal, Scattering coefcients of an in-
ductive strip in a n-line: theory and experiment, IEEE
36. A. Beyer, D. Kother, and I. Wolff, A combined theoretical and
experimental characterization of discontinuities in unilateral
nlines, IEEE MTT Int. Microwave Symp. Digest, Baltimore,
1986, pp. 127130.
37. K. J. Webb and R. Mittra, Avariational solution of the nline
discontinuity problem, 15th Eur. Microwave Conf. Digest,
1985, pp. 311316.
38. Q. Zhang and T. Itoh, Spectral domain analysis of scattering
from E-plane circuit elements, IEEE Trans. Microwave The-
ory Tech. MTT-35:138150 (1987).
39. N. H. L. Koster and R. H. Jansen, Some new results on the
equivalent circuit parameters of the inductive strip disconti-
nuity in unilateral n-lines, Arch. Elek. U
bertragung 35:
497499 (1981).
40. H. El Hennawy and K. Schuenemann, Analysis of n-line
discontinuities, IEE Proc. 129:342350 (1982).
41. H. El Hennawy and K. Schuenemann, Impedance transfor-
mation in n lines, 9th Eur. Microwave Conf. Digest, 1979,
pp. 448452.
42. M. Helard et al., Solution of n-line discontinuities through
the identication of its rst four higher order modes, IEEE
MTT Int. Microwave Symp. Digest, 1983, pp. 387389.
43. R. Sorrentino and T. Itoh, Transverse resonance analysis of
nline discontinuities, IEEE MTT-S Int. Microwave Symp.
Digest, 1984, pp. 414415.
44. A. M. K. Saad and K. Schuenemann, A rectangular wave-
guide equivalent for bilateral and unilateral nlines, Arch.
Elek. U
bertragung 35:287292 (1981).

45. A. S. Omar and K. Schuenemann, Transmission-matrix rep-
resentation of n-line discontinuities, IEEE MTT-S Int.
Microwave Symp. Digest, 1984, pp. 339341.
46. F. Arndt et al., Theory and design of low-insertion loss n-line
lters, IEEE Trans. Microwave Theory Tech. MTT-30:
155163 (1982).
47. Y. C. Shih, T. Itoh, and L. Q. Bui, Computer-aided design of
millimeter-wave e-plane lters, IEEE Trans. Microwave
Theory Tech. MTT-31:135142 (1983).
48. A. Beyer, Calculation of discontinuities in grounded n lines
taking into account the metallization thickness and the
FINLINES 1619
inuence of the mount-slits, 12th Eur. Microwave Conf. Di-
gest, 1982, pp. 681686.
49. M. Burton and W. J. R. Hoefer, An improved model for short-
and open-circuited series stubs in n lines, IEEE MTT-S Int.
Microwave Symp. Digest, 1984, pp. 330332.
50. J. H. Hinken, Simplied analysis and synthesis of nline
tapers, Arch. Elek. U
bertragung 37:375380 (1983).

51. C. Schieblich, J. K. Piotrowski, and J. H. Hinken, Synthesis of
optimum nline tapers using dispersion formulas for arbi-
trary slot widths and locations, IEEE Trans. Microwave The-
ory Tech. MTT-32:16381645 (1984).
52. A. Beyer and I. Wolff, Finline taper design made easy, IEEE
MTT-S Int. Microwave Symp. Digest, 1985, pp. 493496.
53. P. Pramanick and P. Bhartia, Analysis and synthesis of ta-
pered nlines, Microwave RF 26:111114 (1987).
54. F. C. de Ronde, Miniaturization in E-plane technology, 15th
Eur. Microwave Conf. Digest, 1985, pp. 329333.
55. C. J. Verver and W. J. R. Hoefer, Quarter wave transformers
for matching transitions between waveguides and n lines,
IEEE MTT-S Int. Microwave Symp. Digest, 1984, pp. 417421.
56. J. K. Piotrowski and W. Wojtasiak, Awaveguide-to-microstrip
transition, 16th European Microwave Conf. Digest, 1986, pp.
505510.
57. W. J. R. Hoefer and C. J. Verver, Optimal waveguide to E-
plane circuit transitions with binomial and Chebyshev trans-
formers, 15th Eur. Microwave Conf. Digest, 1985, pp. 305310.
58. H. El Hennawy and K. Schuenemann, Hybrid nline match-
ing structures, IEEE Trans. Microwave Theory Tech. MTT-
30:21322138 (1982).
59. K. Solbach, H. Callsen, and W. Menzel, Spurious resonances
in asymmetrical nline junctions, IEEE Trans. Microwave
Theory Tech. MTT-29:11931195 (1981).
60. P. J. Meier, Integrated nline, the second decade, Microwave
J. 28(12):3048 (1985).
61. A. Beyer and I. Wolff, A solution of the earthed n line with
nite metallization thickness, 1980 IEEE MTT-S Int. Micro-
wave Symp. Digest, Washington DC, pp. 258260.
62. A. Beyer, Analysis of the characteristics of an earthed n line,
(1981).
FRACTAL-SHAPED ANTENNAS: A REVIEW
JAUME ANGUERA
CARLES PUENTE
CARMEN BORJA
JORDI SOLER
Fractus S.A.
Barcelona, Spain
1. INTRODUCTION
Among the different telecommunications systems, mobile
communication systems have experimented the fastest
relative growth. A wireless system requires an antenna
to radiate and receive electromagnetic waves. The anten-
na is therefore an essential component. Antenna perfor-
mance is usually very sensitive to the antennas geometric
form and sizewavelength ratio. This has two important
consequences:
1. Once the antenna has been designed to operate at a
particular wavelength, it will rarely be useful at
other wavelengths. Since telecommunication sys-
tems operate at separate frequency bands to avoid
mutual interference, one usually must account for
an individual antenna for every single system.
2. Given a particular service band, the antenna cannot
be made arbitrarily small owing to the same con-
straint.
In this sense, the present article deals with the investiga-
tion of antennas presenting small properties and multi-
functionality features. The aim is to describe fractal
techniques and their applications to design miniature,
multifrequency, and high-directivity antennas. Fractal
techniques have been chosen as their geometric proper-
ties are useful in designing miniature, multifrequency,
and high-directivity antennas.
An overview of fractal-shaped antennas summarizes
most of the work carried out since the early 1990s.
2. FRACTAL GEOMETRY
Since the early days of mathematics and Euclidian geom-
etry, objects have been classied as one-, two-, or three-
dimensional (1D, 2D, or 3D). The full stop we use to sep-
arate sentences in a written paper is 1D (i.e., point/line);
the paper we use to write on can be considered approxi-
mately 2D (i.e., a plane/an area); while the ballpen we use
to write these words is a 3D object (i.e., a cube/volume).
However, we are surrounded by other objects (clouds, cost
lines, arm trees, etc.) that are best described not by using
Euclidian geometry but by using fractal geometry [1].
Among the several properties that characterize frac-
tals, two are of interest in terms of antenna design: self-
similarity/self-afnity and space-lling/fractal dimension
properties. We will briefly dene these properties and in
the next section we will relate such geometric properties
to the antenna eld.
Some fractals are self-similar, which roughly implies
that there are copies of the whole structure within the
structure at different scales. For example, the futuristic
tree of Fig. 1 can be understood in a loose sense to be self-
similar; if we cut one branch, we will see that it is a re-
duced simplied copy of the whole structure. If such a
scale factor is different for several directions, the structure
is no longer self-similar but self-afne.
A characteristic feature of fractals is their space-lling
property. To characterize such space-lling properties, the
concept of fractal dimension is used. There are several
definitions of fractal dimension. What really matters here
is that a fractal object can have a fractional dimension or a
larger dimension than its topological one [2], as it is the
case of the Hilbert curve used to obtain a small monopole
antenna presented in the next section.
Finally, two more definitions are presented here: mass
fractals and boundary fractals. Mass fractals implies that
1620 FRACTAL-SHAPED ANTENNAS: A REVIEW
the internal structure is fractal, as in the Sierpinski gas-
ket presented next. Boundary fractals implies that the
perimeter is fractal-like, as in the Koch island patch pre-
sented in the next section.
3. FRACTAL-SHAPED ANTENNAS
3.1. The Concept
Having seen the geometric properties of fractal geometry,
it is interesting to explain what benets are derived when
such geometry is applied to the antenna eld.
Fractals are abstract objects that cannot be physically
implemented. Nevertheless, some related geometries can
be used to approach an ideal fractal that are useful in
constructing antennas. Usually, these geometries are
called prefractals or truncated fractals. In other cases,
other geometries such as multitriangular or multilevel
congurations [3,4] can be used to build antennas that
might approach fractal shapes and extract some of the
advantages that can theoretically be obtained from the
mathematical abstractions. In general, the term fractal
antenna technology is used to describe those antenna en-
gineering techniques that are based on such mathematical
concepts that enable one to obtain a new generation of
antennas with some features that were often thought im-
possible in the mid-1980s.
After all the work carried out thus far, one can summa-
rize the benets of fractal technology in the following way:
*
Self-Similarity. This property is useful in designing
multifrequency antennas, as, for instance, in the ex-
amples based on the Sierpinski gasket [510]. Anoth-
er interesting application of the self-similar property
has been applied in designing multiband arrays
[11,12].
*
Fractal Dimension. It is useful to design electrically
small antennas, such as the Hilbert and Koch mono-
poles, fractal loops, and microstrip antennas [1317].
*
Mass Fractals and Boundary Fractals. These frac-
tals are useful in obtaining high-directivity elements
[1820] and undersampled arrays [2123]. They have
also been useful for designing low-sidelobe arrays
[8,11,24,25].
The next section explores fractal techniques used to obtain
multifrequency antennas and arrays; miniature, high-di-
rectivity antennas; and undersampled and low-sidelobe
arrays.
3.2. Applications of the Self-Similar Property:
Multifrequency Antennas and Arrays
3.2.1. Multifrequency Antenna Elements. The rst re-
ported multifrequency fractal-shaped antenna was de-
signed by C. Puente at the Universitat Polite`cnica de
Catalunya (UPC) in the early 1990s and was named the
Sierpinski monopole (Fig. 2) [5,7,26]. This monopole is a
ve-iteration Sierpinski gasket that features ve operat-
ing bands (as many bands as fractal iterations) spaced
logperiodically by a factor of 2 (the same scale factor used
to generate the fractal). At the ve operating bands, this
antenna presents a good level of return loss (around
10 dB) (Fig. 3) and similar radiation patterns across
the bands (Fig. 4).
The antenna depicted in Fig. 2 was extensively studied
[26] and rst patented in May 1995 [27].
Besides the experimental work, numerical computation
using the nite-difference time-domain (FDTD) and meth-
od-of moments (MoM) codes were also used to fully un-
derstand the antenna behavior. Perhaps the most
interesting analysis is that related to the current distri-
bution on the conductor surface. Figure 5 shows the elec-
trical current distribution at each of the ve operating
bands, where it can be observed that the current is con-
centrated in several regions called active regions. At the
lowest operating frequency, the active region is extended
Figure 1. Futuristic vision of a tree. Every arm can be consid-
ered a scaled-down copy of the whole tree. (This gure is available
in full color at http://www.mrw.interscience.wiley.com/erfme.)
Figure 2. The Sierpinski monopole on a square ground plane.
FRACTAL-SHAPED ANTENNAS: A REVIEW 1621
throughout the whole structure; on the other hand, for the
highest operating frequency, the active region is concen-
trated in the smallest iteration number near the feed-
point.
Figure 6 shows an iterative method that has been used
to predict the return loss of the Sierpinski monopole and
Sierpinski patch antennas. Such a method as well as the
segmentation method is interesting because it enables
parametric analysis to be performed in a very short time
and allows us to understand what occurs if the scale factor
is changed. The S matrix is the S parameter of intercon-
nected transmission lines, where one can factor in the
radiation loss mechanism by considering transmission
line with losses. The present method as well as the seg-
mentation method have been used to predict input pa-
rameters of the Sierpinski monopole as well as other
fractal-shaped antennas [2831].
Another analytical method called the multiperiodic
traveling-wave vee model has also been used to predict
the radiation patterns as depicted in Fig. 7 [26]. Such a
model is able to predict the radiation pattern of a Sier-
pinski monopole if the current along the edge is known.
0
10
20
30
2 3 4 5 6 7 8 9 10 1
Frequency (GHz)
Experimental
FDTD
DOTIG4

i
n

(
d
B
)
Figure 3. Measured and computed return loss
for the Sierpinski monopole of Fig. 2. Numerical
simulations agree fairly well with measure-
ments. The antenna is matched at 10dB re-
turn loss at ve bands, as many as the
iterations (ve). (This gure is available in
full color at http://www.mrw.interscience.wiley.
com/erfme.)
Figure 4. Measured radiation cuts at the four upper bands for
f901 plane for the Sierpinski monopole of Fig. 2.
Zoom 1
Zoom 2
Zoom 4
Zoom 8
B
a
n
d

#
2
B
a
n
d

#
3
B
a
n
d

#
4
B
a
n
d

#
5
3025 20 15 10 5 0
dB
Figure 5. FDTD electrical current simulations on the conductor
surfaces. An active region is observed (circled in the left column).
(This gure is available in full color at http://www.mrw.
interscience.wiley.com/erfme.)
The model has been extended not only to better under-
stand the physical radiation of the classic Sierpinski
monopole [32,33] but also to obtain the radiation pattern
of Mod-P Sierpinski monopoles [34,35].
For a fractal-shaped antenna operating at the highest
frequency, the antenna is electrically very large.
If the MoM is used for example, this requires a small disc-
retization elements resulting in a large number of
unknowns (N). If N is large enough, the matrix imped-
ance Z of MoM does not t in the computer memory (N
2
)
and moreover, large computation time is needed (time
is proportional to N
3
). Therefore, advanced numerical
methods and other strategies are required to correctly
model fractal-shaped antennas such as the multilevel
matrix decomposition algorithm (MLMDA) and multilev-
el fast multipole method (MLFMM). MLFMM is very ef-
cient for 3D problems, while MLMDA is very efcient
for planar antennas, as is the case with microstrip
patches or dipoles. Other methods such as adaptive mesh-
es can be used (Fig. 8). For example, one can use a ne
mesh for the active zone related to the highest frequency,
while coarse meshes can be used to the lowest frequencies
[36].
Genetic algorithms have also been used to optimize the
fractal antenna geometry [37,38], as is the case of a dual-
band dipole antenna having a SWR (standing-wave ratio)
below 2 at 1.225 and 1.575 GHz and small monopole an-
tennas [39].
For the numerical experiments, we have shown that
an active zone is associated with the fractal Sierpinski
monopole antennas (Fig. 5). One can observe the same
effect using experimental procedures [40]. Figure 9
photographically illustrates the temperature elevation
on the antenna surface. A magnetron operating
at 2.45 GHz was used. In order to observe the active
region, several Sierpinski prototypes printed in ber-
glass have been built. The losses on the berglass
allow one to acquire thermic images. The antennas
have been scaled to obtain the active region at
2.45 GHz.
Following those results, one can conclude that at each
operating frequency, only some part of the structure is re-
sponsible for radiation. As these parts present approxi-
mately the same electrical height and shape, the antenna
radiates similar at each band (Fig. 4).
After the successful result described above, research
on the Sierpinski monopole continued and variation
of the are angle and geometric perturbation based on
self-afne transformations were explored to study how
one could control the input impedance and frequency
allocation [4144]. Those experiments were really
interesting because they claried the design of practical
monopole antennas for base station systems such as
GSM 900, GSM 1800, and UMTS [4547]. Another inter-
esting modication when changing the are angle is
to obtain a dual-band antenna with bidirectional
pattern [48].
The Sierpinski gasket was also used to design a dipole
antenna (Fig. 10), and it was used by Romeu and Rahmat-
Samii [4951] to implement a frequency-selective surface
[So] [Si]
i=0
(a)
(c)
(e)
(b)
(d)
i>0
T [Si]
[Si+1]
[S2] [ S1]
[ S1]
[ S1]
[So] [So]
[So]
[So]
[ S1]
2"
1"
3'
1'
2'
3"
1
0
5
10
15
20
2
3
10 20 30 40 50 60 70 80 90 100 110
band#1 band#2 band#3
f (GHz)
i
n
Figure 6. (a) Iterative method to calculate the
input impedances of fractal-shaped antennas;
(bd) cascading the S parametersin this case,
the model is useful for predicting the evolution
of the operating bands with the number of it-
erations; (e) result for a three-fractal-iterated
Sierpinski antenna.
(FSS) operating at two bands. Other fractal shapes have
also been studied to create FSS [52,53].
Novel geometries based on the Sierpinski triangle
are investigated in Solers thesis [54] to control the num-
ber of bands and also to control the frequency separation
(Fig. 11). These antennas are based on Pascals triangle
[34,35]. By using the Mod-P Sierpinski structures, the fre-
quency separation between bands can be larger than that
obtained for the classic Sierpinski gasket [55].
In addition, fractal-related antenna structures can be
top-loaded as shown in Fig. 12 in order to control the
spacing between bands and the input resonances at each
resonance [56,57].
In general, depending on the feeding scheme used for
the antenna, the ground plane becomes a useful contrib-
utor in the radiating process [58]. The fractal geometry
has been also used not only to shape the antenna but also
to design ground planes inspired on the Sierpinski fractal
shape [59] (Fig. 13).
It has been shown that by means of Sierpinski fractal
and other related shapes, it is possible to obtain several
matched bands with a similar radiation pattern [6062].
To improve the radiation pattern, a 3D arrangement has
been proposed [63] and is shown in Fig. 14.
The Sierpinski gasket was reexplored as a microstrip
patch as well [64,65]. In Borjas thesis [66,67], an in-depth
study is devoted to analysis of the Sierpinski patch in or-
der to obtain several bands with broadside radiation pat-
terns. For the patch version, it is more difcult to obtain a
multifrequency behavior, although several proposals have
been already made [66,68]. Figure 15 shows a particular
Figure 7. Multiperiodic traveling-wave V (vee) model is useful for predicting the radiation pat-
terns of fractal-shaped antennas such as the Sierpinski monopole.
3rd band
Active
region
4th band
Figure 8. A fractal-shaped antenna becomes and electrically
large structure at the highest operating bands. Advanced meth-
ods are required to model these complex structures. Adaptive
meshing can be used to mesh properly the active regions.
solution where some junctions are broken in order to elim-
inate undesired higher modes that degrade the radiation
patterns.
By modifying the scale factor and fractal geometry, a
dual-band dual-polarized microstrip antenna was de-
signed to operate at GSM 900 and 1800 [69]. Moreover,
such an element was the basic radiating element of a pro-
totype for a dual-band dual-polarized array for mobile
communications (Fig. 16). The feeding scheme is based on
an etched capacitor designed to cancel the inductance ef-
fect of the coaxial probe [70].
In this sense, research continues with the aim to go a
further step: investigate multifrequency microstrip anten-
nas using fractal-based geometries (Fig. 17), while intro-
ducing a broadband behavior at each frequency band
employing bandwidth enhancement techniques [6,71].
The current distributions of the multistacked antenna
are calculated at the central operating frequencies as
shown in Fig. 18. It is important to observe that when
the antenna is excited at f 1.612 GHz, the bottom para-
sitic patch remains nonexcited; that is, it seems to be
transparent to the stated frequency, and only the top par-
asitic patch is operating. By contrast, when the antenna is
excited at f 2.697 GHz, the bottom patch is excited while
the top patch is almost totally disconnected. Therefore, for
the rst operating frequency, the top parasitic patch is re-
sponsible for enhancing the bandwidth of the active patch
while the bottom parasitic patch is responsible for broad-
ening the second band. The measured prototype shows a
BW4.7% and BW6.8% (SWR2) for the rst and sec-
ond operating bands, respectively, with similar radiation
patterns for the entire bandwidth and similar for both
bands, conrming the dual-broadband performance.
Figure 9. Temperature elevation on the an-
tenna surface measured in operating bands 24
using an infrared camera. (This gure is
available in full color at http://www.mrw.
interscience.wiley.com/erfme.)
Figure 10. The Sierpinski gasket was also used to design a di-
pole antenna.
Sierpinski mod-3 Sierpinski mod-5
Figure 11. The Sierpinski monopole mod-3 and mod-5. Figure 12. A top-loaded modied Sierpinski monopole.
3.2.2. Antenna Arrays. The self-similar property of
some fractals has been useful in designing multiband ar-
rays [11,72]. This can be achieved if the array factor is a
self-similar curve such as, for example, the Koch curve
(Fig. 19). In Fig. 19, an array factor following the self-sim-
ilar curve is shown. If the progressive phase b kd, the
same shape is obtained at f, f/3, f/9 and so on. Although the
same radiation pattern is maintained at several frequen-
cies, the radiation parameters such as gain, radiation re-
sistance, and efciency are not maintained because the
pattern magnitude is reduced when the frequency increas-
es.
Other functions as the Weierstrass function can be
used as well to synthesize fractal arrays with self-similar
properties [73].
Werner takes advantage profit of the self-similar prop-
erty of fractal-based arrays to develop fast algorithms for
calculating the impedance matrix of such arrays [74]. The
self-similar property is also useful for developing rapid
beamforming algorithms [75,76].
Having reviewed some of the work on multifrequency
antennas and arrays using fractals, in the next section we
Figure 14. 3D antenna solution used to tailor the antenna pat-
terns at different frequency bands. The antenna is built with two
planar structures based on the Sierpinski fractal.
Figure 15. The Sierpinski patch with the top
junction broken in order to obtain a dual-band
antenna with broadside radiation patterns.
Figure 16. A dual-band microstrip array for GSM 900 and 1800
using a modied version of the Sierpinski geometry [3,4].
Figure 13. Straight monopole above a fractal-shaped ground
plane derived from four mod-2 Sierpinski gaskets.
present some of the research related to electrically small
antennas.
3.3. Applications of the Fractal Dimension Property:
Electrically Small Antennas
Antenna size is a critical parameter because antenna be-
havior depends on antenna dimensions in terms of wave-
length (l). In many applications, space is a constraint
factor; therefore an antenna cannot be comparable to the
wavelength but smaller (i.e., a small antenna). An anten-
na is said to be small when its larger dimension is less
than twice the radius of the radian sphere; its radius is
l/2p. Wheeler and Chu were the rst who investigated the
fundamental limitations of such antennas [7779]. The
goal of this section is to present the miniature features of
the Hilbert monopole [14,80,81] and compare it with the
Koch monopole studied previously [13,26,27]. Some new
advancements are also presented. It should be pointed out
that at the limit, the Koch and Hilbert curves feature
fractal dimensions of 1.23 and 2, respectively.
Figure 20 shows a l/4 monopole and the rst ve iter-
ations of the Hilbert monopole where the total height for
all antennas is h7 cm. The antennas are etched on a
1-mm FR4 substrate acting merely as an antenna support.
The copper used is 0.4mm wide. As the total length of the
Hilbert monopole increases as iteration increases, the res-
onant frequency will decrease. Thus, the antenna becomes
electrically smaller as the iteration number increases.
For the fth iteration, the length is 65 times h. One
might think that the resonant frequency for a fth-itera-
tion Hilbert monopole would be 65 times less than the
resonant frequency of a linear monopole with height h,
which will be an extraordinary frequency reduction. Ob-
viously, this is not true since coupling between turns pro-
vides a shorter path for currents owing from one tip to
the other. However, even with the coupling effect, the res-
onant frequency reduction or compression factor (CF),
which is dened as the ratio between the resonant fre-
quency for the vertical monopole and the Hilbert resonant
frequency, can achieve values up to 10, taking into account
the negligible effect of the dielectric support. Large CFare
interesting for designing miniature antennas where space
is a constraint factor such as, for instance, integrated low-
frequency antennas in automotive applications [82].
Finally, a new concept called compression efciency
(CE) is introduced. CE is dened as the ratio between
the rst resonant frequency of the equivalent vertical
linear monopole with a height equal to the total length
of a Hilbert (Koch) monopole and the rst resonant fre-
quency of the Hilbert (Koch) antenna. CE is an important
parameter for comparing different space-lling geometries
to determine which structure can decrease the resonant
frequency with less wirelength.
Figure 21 shows CE as a function of the fractal ite-
ration, showing that CE decreases as the iteration
number increases. In the same gure, the CE for the
Koch monopole is shown [14]. It is clearly seen that the
CE decreases more rapidly for the Koch monopole. For ex-
ample, for the same height (h) of the vertical monopole,
the Hilbert 1 and the Koch 5 monopoles have approxi-
mately the same length; however, CE is 70% for the
Hilbert monopole and only 40% for the Koch monopole,
indicating that the Hilbert monopole (with a larger fractal
dimension) achieves a larger frequency reduction with the
same wirelength.
Parasitic patch 2
Parasitic patch 1
Coaxial feed
Active patch
Groundplane
Substrate(c
r3
,H
P2
)
Substrate(c
r2
,H
P1
)
Substrate(c
r1
,H
a
)
c
r
=1
c
r
=1
h
gap1
Y
X
Z
h
gap2
Figure 17. A multistacked microstrip patch an-
tenna based on the modied Sierpinski fractal.
f=2.697GHz
f=1.613GHz
Active Parasitic 1 Parasitic 2
Figure 18. Simulated current distribution. It can be observed
that only one parasitic patch contributes significantly to radiation
depending on the operating frequency. (This gure is available in
full color at http://www.mrw.interscience.wiley.com/erfme.)
Analytical models have been proposed by Vinoy et al. to
predict the resonant frequency of the Hilbert-based an-
tenna [83]. Modications of the feeding point have been
investigated in order to match the antenna [84]. Methods
for fast calculation of the input impedance of a Koch
fractal dipole have also been proposed [85].
The Hilbert curve have also been useful in designing
small microstrip patch antennas. In Ref. 86, a microstrip
patch is loaded with a transmission line to reduce the res-
onant frequency of the fundamental mode of the unloaded
patch [82]. The Hilbert curve has also been used to con-
struct metamaterials, in particular to design articial
magnetic conductors as proposed by McVay, Engheta,
and Hoorfar [87].
The classic miniature technique of short-circuited
monopoles has been combined with fractal technology, as
is the case of a shorted monopole inspired in the Sierpinski
fractal in Songs thesis [88,89].
Another advantage of miniature antennas is that they
tend to reduce mutual coupling in an antenna array as
studied in Gianvittorios thesis [90]. By reducing the mu-
tual coupling, the scan-blindness effect can be reduced to
shift the blind angle toward larger angles near the horizon
[16,17,91].
In order to overcome the problem of miniature micro-
strip antennas, that is, small bandwidth and radiation ef-
ciency, parasitic techniques have been combined with
fractal techniques to obtain miniature and wideband an-
tennas with improved efciency [68,92,93]. Figure 22
shows a modied Sierpinski-based microstrip antenna
consisting of an active patch and a parasitic patch. Using
such a geometry, the resonant frequency of the antenna is
1.26 GHz while it is 2GHz for the lled version. By adding
the parasitic patch, the bandwidth with respect to the
single active element is increased by a factor of 15, result-
ing in a bandwidth of BW2.7% at SWR2; the radia-
tion efciency for this antenna is 84% [68]. Similar
approaches of stacking elements can be found in Ref. 94.
3.4. Application of Mass and Boundary Fractals: High-
Directivity Antennas and Undersampled and Low-
Sidelobe Arrays
3.4.1. High-Directivity Antennas. Another interesting
feature of fractal-based microstrip antennas is the exis-
tence of localized modes called fractons and fractinos
1
0.8
0.6
0.4
0.2
0
[=kd progressive phase
[+kd [-kd
A
r
r
a
y

f
a
c
t
o
r
Koch Pattern
1 2 3 4 5 6 0
Visible range at f/27

Visible range at f
Figure 19. The Koch array factor. The curve re-
tains its similarity at six different scales. By add-
ing a progressive phase b kd, the visible range
is always centered at a secondary lobe that has
the same shape as the total pattern. The frequen-
cy change by a factor
1
3
reduces the visible range
around this similar subpattern [11].
Figure 20. (a) Vertical monopole; (b) Hilbert 1; (c) Hilbert 2; (d)
Hilbert 3; (e) Hilbert 4; (f) Hilbert 5. The antennas are etched on a
FR4 substrate acting as a support.
depending on whether the structure is based on mass
fractals or boundary fractals [66,68,9598]. Figure 23
shows a Koch-island patch where the current distribution
for the fundamental mode and fractino mode are shown
[18,99]. It is worth noting that for the fractino mode there
are zones of high current density (localized mode). More-
over, such zones are coherent; that is, they radiate in
phase, and thus the radiation pattern is broadside. This
feature is rarely obtained with classical Euclidian geom-
etries such as squares, circles, and triangles. Other geom-
etries presenting localized modes have been investigated
[66,68].
The fracton-mode behavior of the Sierpinski bowtie
patch has been investigated [100]. The fracton-mode prop-
erties are studied as a function of the iteration number.
Besides fracton and fractino modes, microstrip anten-
nas supporting several fracton modes have also been stud-
ied. Figure 24 shows an example of the Sierpinski bowtie
fractal-based antenna, which supports as many fracton
modes as iterations [101103]. It is interesting to note that
the directivity of the fracton mode increases with the mode
order.
3.4.2. Undersampled Arrays. Operating in such a frac-
ton or fractino mode, one can obtain a larger directivity
than that for a fundamental mode. In this sense, Anguera
[68] has investigated structures supporting fracton modes
for microstrip arrays (Fig. 25); this type of array is called
an undersampled array [22,23,104,105]. Such a novel
method presents an advantage over conventional designs
in that one can obtain in the same electrical area of a con-
ventional array using Euclidian-based elements, with the
same directivity but using fewer elements. Other advan-
tages are that the feeding network complexity decreases
and there is more available space to integrate other mi-
crowaves devices such as, for example, ampliers, lters,
and matching networks. This reduces the antenna cost,
volume and weight, which can be an enormous advantage
for satellite antenna applications [21,106].
Figure 26 shows the radiation pattern in the upper
space for the Euclidian and fractal arrays of Fig. 25. Both
C
o
m
p
r
e
s
s
i
o
n

e
f
f
i
c
i
e
n
c
y

(
%
)
100
90
80
70
60
50
40
30
20
Koch 5
Koch 4
Hilbert 5
Hilbert 4
Hilbert 3
Hilbert 2
Hilbert 1
Koch 3
Koch 2
Koch 1
Linear Monopole
10
0
1
Normalised total length (Total length/Linear Monopole length (h))
2 3 4 5 6 7 8 910 20 30 40 50 60
Figure 21. Compression efciency for the
Hilbert and Koch monopoles.
Active
z
y
SWR=2
f=1.26GHz x
Parasitic
Figure 22. A miniature microstrip antenna inspired in the Si-
erpinski fractal. Current distributions on the patch surfaces are
shown. The parasitic patch enhances the bandwidth when the
input impedance loop is correctly centered at the center of the
Smith chart.
f=3.7 GHz 0 dB = 4.2 (A/m)
0 dB
1.5 dB
3 dB
4.5 dB
6 dB
7.5 dB
9 dB
10.5 dB
12 dB
13.5 dB
15 dB
16.5 dB
18 dB
19.5 dB
21 dB
22.5 dB
24 dB
25.5 dB
27 dB
28.5 dB
30 dB
Fundamental mode of the Koch2 patch
f=1.24 GHz
Figure 23. Current distribution for the fundamental and localized modes of a Koch microstrip
patch. The localized mode presents high current density on the boundary with coherent radiation;
a broadside radiation pattern with a directivity larger than that of the fundamental mode can be
obtained. (This gure is available in full color at http://www.mrw.interscience.wiley.com/erfme.)
25
30
25
20
15
15
10
10
10
10
10
10
20
10
5
5
0
0 0
0
0
0
5
20
15
10
5
0
20
15
10
10
10
10
10
10
10
10
15
5
0
5
10
15
20
25
10
10
20
25
15
5
5
5 0
0
0
10
5
0
0
5
20
10
10
0
0
10
10
20
25
20
20
15
10
20
f=0.4GHz
y
y
y
x
x
x
x
x
y
y
0dB
15dB
30dB
E
l

(
d
B
)
E
l

(
d
B
)
E
l

(
d
B
)
E
l

(
d
B
)
E
l

(
d
B
)
f=1.55GHz
f=6.33GHz
f=13GHz
f=3.04GHz
Figure 24. 3D-simulated radiation pat-
terns for the Sierpinski bowtie 4: the fun-
damental and four fracton modes are
shown. (This gure is available in full
color at http://www.mrw.interscience.
wiley.com/erfme.)
arrays present the same electrical area; however, for the
fractal array operating in a fracton mode, the number of
elements is 2.8 times less (14 13 for the Euclidian and
only 8 8 for the fractal one).
3.4.3. Low-Sidelobe Arrays. The fractal spatial arrange-
ment of array elements have been useful for designing
low-sidelobe arrays with equally weighted current ele-
ments, which is advantageous current element for practi-
cal purposes because the feeding network is easier than
those used when different current amplitudes are required
in order to reduce sidelobes.
Kim and Jaggard rst proposed a nonuniform random
fractal spacing for improving the sidelobe level of random
arrays in 1986 [24]. The Cantor array has also been
DX DX
dx dx
dy
dy
DY
DY
Figure 25. The classical microstrip array with circular patches
and the novel scheme with fractal-shaped microstrip antennas.
Circular array Bowtie array E
total
(dB)
0dB
5dB
10dB
15dB
20dB
25dB
30dB
Y
X
0=0
0=90
Figure 26. Normalized simulated radiation
patterns for the arrays shown in Fig. 25. (This
gure is available in full color at http://
www.mrw.interscience.wiley.com/erfme.)
Cantor array, o=3, 64 elements
Cantor array, o =1.1, 64 elements
1
0.5
0
200 150 100 50 0
Z(wavelength)
Z(wavelength)
50 100 150 200
R
e
l
a
t
i
v
e

c
u
r
r
e
n
t

d
i
s
t
r
i
b
u
t
i
o
n
1
0.8
0.6
0.4
0.2
0
200 150 100 50 0 50 100 150 200
R
e
l
a
t
i
v
e

c
u
r
r
e
n
t

d
i
s
t
r
i
b
u
t
i
o
n
Figure 27. Cantor array constructed with 64 array elements from a 2d generator and a logperiod
d 3 (top) and d 1.1 (bottom).
analyzed [11,107,108]. Mathematically, the generalized
Cantor distribution can be obtained by convoluting two
delta function as
cz
Y
1
n1
ff z
.
d
n
g 1
where f(z) is a 2d function and d is the scaling factor.
Figure 27 shows two cantor arrays formed by 64 array
elements and scale factors of d 3 and d 1.1, respective-
ly. For the Cantor array having a logperiod of d 1.1, the
corresponding pattern looks very similar to a binomial
distribution: sidelobe level below 65 dB (Fig. 28). The
main advantage is that although the sidelobe level is close
to the binomial distribution, the current excitation is uni-
form with drastically simplies the feeding network.
An array with a low sidelobe level has been investigat-
ed by small position offsets of fractal elements [109].
4. CONCLUSIONS
Research and applications of fractal-shaped antennas is
increasing [110,111]. So far, we can divide the applications
into six main groups:
1. Multifrequency Antennas, FSS, and Arrays. The
early work was dedicated to the Sierpinski dipole.
Nowadays, fractal technology is also being applied to
the design of multifrequency microstrip patches.
Moreover, multifrequency FSS and arrays can be
designed using the fractal technology.
2. Electrically Small Antennas. Koch and Hilbert
curves have been used so far and are inspiring
many researchers to search for new space-lling
curves to optimize antenna performance.
3. High-Directivity Elements. Fractal-based antennas
supports fracton and fractino modes. As we have
shown, these modes present a broadside radiation
pattern with a directivity larger than the fundamen-
tal mode.
4. Low Sidelobe Array. Using fractal arrangements,
arrays with very low sidelobes can be achieved
with a uniform excitation. This is an advantage com-
pared with nonuniform excitations, which required
a complex feeding network.
5. Undersampled Arrays. With this novel application,
microstrip arrays use a fewer elements than in the
classic approach using Euclidian radiators. This re-
duction of the number of elements potentially re-
duces the feeding network complexity and might
become a very interesting application for satellite
antennas, where space, cost, and volume are con-
strain factors.
6. Fast Computational Methods. Taking advantage of
the fractal geometry, rapid algorithms to calculate
antenna parameters and beamforming are being de-
veloped.
Acknowledgment
This work has been nanced by FRACTUS, S.A.
The authors would like to thank M. Navarro, J. Romeu,
A. Cardama, J. M. Rius, F. Moyano, and the Electromag-
netics and Photonics Engineering Group of the Univ-
ersitat Polite`cnica de Catalunya and J. M. Gonza lez for
the thermograms of the Sierpinski gasket.
BIBLIOGRAPHY
1. B. B. Mandelbrot, The Fractal Geometry of Nature, Freeman,
1983.
2. H. O. Peitgen, H. Ju rgens, and D. Saupe, Chaos and Fractals,
New Frontiers of Science Series, Springer-Verlag, 1990.
3. C. Puente, C. Borja, J. Anguera, and J. Soler, Multilevel An-
tennae, Patent WO0122528.
4. C. Puente, C. Borja, M. Navarro, and J. Anguera, Multi-tri-
angular Antennas for Cellular Telephony GSM and DCS,
Patent App. WO9957784.
5. C. Puente, J. Romeu, R. Pous, and A. Cardama, On the be-
havior of the sierpinski multiband antenna, IEEE Trans. An-
ten. Propag., 46(4):517524 (April 1998).
6. J. Anguera, E. Mart nez, C. Puente, C. Borja, and J. Soler,
Broadband dual-frequency microstrip patch antenna with
modied sierpinski fractal geometry, IEEE Trans. Anten.
Propag., 52(1):6673 (2004).
7. C. Puente, J. Romeu, R. Pous, and X. Garc a, Fractal multi-
band antenna based on the sierpinski gasket, IEE Electron.
Lett., 32(1):12 (Jan. 1996).
8. D. L. Jaggard and A. D. Jaggard. Cantor ring arrays, Micro-
wave Opt. Technol. Lett., 19:121125 (1998).
9. N. Cohen, Are fractals naturally frequency inavariant/inde-
pendent? 15th Annual Review of Progress in Applied Compu-
tational Electromagnetics, Naval Postgraduate School,
Monterey, CA, March 1999, Vol. I, pp. 101106.
10. N. Cohen, Fractal and shaped dipoles, Commun. Quart. 2536
(spring 1996).
11. C. Puente and R. Pous, Fractal design of multiband and low
side-lobe arrays, IEEE Trans. Anten. Propag., 44(5):730739
(May 1996).
12. D. H. Werner, R. L. Haupt, and P. L. Werner, Fractal antenna
engineering: The theory and design of fractal antenna arrays,
Anten. Propag. Mag., 41(5):3758 (Oct. 1999).
20
0
40
A
r
r
a
y

f
a
c
t
o
r

(
d
B
)
60
80
100
120
3 2 1 0 1 2
3
Cantor
Binomial
o=1.1
Figure 28. Array pattern of a binomial distribution and a Cantor

distribution with a logperiod of d 1.1, both having the same
number of array elements: 64.
13. C. Puente, J. Romeu, R. Pous, J. Ramis, and H. Hijazo, Small
but long koch fractal monopole, IEE Electron. Lett., 34(1):
910 (Jan. 1998).
14. J. Anguera, E. Mart nez, C. Puente, and E. Rozan, The fractal
hilbert monopole: A two-dimensional wire, Microwave Opt.
Technol. Lett., 36(2):102104 (Jan. 2003).
15. N. Cohen and R. G. Hohlfeld, Fractal loops and the small loop
approximation, Commun. Quart. 7781 (winter 1996).
16. J. P. Gianvittorio and Y. Rahmat-Samii, Fractal element an-
tennas: A compilation of congurations with novel character-
istics, IEEE Int. Symp. Antennas and Propagation Digest,
Salt Lake City, July 2002, Vol. 3, pp. 16881691.
17. J. P. Gianvittorio and Y. Rahmat-Samii, Fractal patch anten-
nas: Miniaturizing resonant patches, IEEE Antennas and
Propagation Society Int. Symp. URSI Digest, Boston, July
2001, p. 298.
18. C. Borja, G. Font, S. Blanch, and J. Romeu, High directivity
fractal boundary microstrip patch antenna, IEE Electron.
Lett., 36(9):778779 (2000).
19. J. Anguera, C. Puente, C. Borja, R. Montero, and J. Soler,
Small and high directivity bowtie patch antenna based on the
sierpinski fractal, Microwave Opt. Technol. Lett., 31(3):
239241 (Nov. 2001).
20. J. Romeu, A. Aguasca, S. Blanch, and J. Girona, Observation
of localized modes in the koch waveguide, Proc. IEEE Anten-
nas and Propagation Society Int. Symp., Boston, July 2001,
Vol. 3, pp. 644647.
21. J. Anguera, G. Montesinos, C. Puente, C. Borja, and
J. Soler, High-directivity microstrip array using high-direc-
tivity fractal-based elements, Proc. 25th European Space
Agency (ESA) Antenna Workshop on Satellite Antenna Tech-
nology, Sept. 2002 ESTEC, Noordwijk, The Netherlands,
2002.
22. J. Anguera, G. Montesinos, C. Puente, C. Borja, and J. Soler,
An under-sampled high directivity microstrip patch array
with a reduced number of radiating elements inspired on
the sierpinski fractal, Microwave Opt. Technol. Lett.,
37(2):100103 (April 2003).
23. J. Anguera, C. Puente, and C. Borja, Undersampled Micro-
strip Array Using Multilevel and Space-Filling Shaped Ele-
ments, Patent PCT/EP02/07835.
24. Y. Kim and D. L. Jaggard, The fractal random array, Proc.
IEEE, 74(9):12781280 (Sept. 1986).
25. D. L. Jaggard and A. D. Jaggard, Cantor ring arrays, IEEE
Int. Symp. Antennas and Propagation Digest, Atlanta, June
1998, Vol. 2, pp. 866869.
26. C. Puente, Fractal Antennas, Ph.D. dissertation, Dept. Signal
Theory and Communications, Univ. Polite`cnica de Catalunya,
1997.
27. C. Puente, R. Pous, J. Romeu, and X. Garc a, Fractal and
Multifractal Antennas, Patent ES 9501019.
28. C. Borja, C. Puente, and A. Medina, Modelo sencillo para el
estudio de los para metros de entrada de una antena fractal de
sierpinski (Simple model to study input impedance of the
Sierpinski fractal antenna), Proc. XII Symp. Natl. URSI,
Bilbao, 1997, pp. 363371.
29. C. Puente, C. Borja, M. Navarro, and J. Romeu, An iterative
model for fractal antennas: Application to the sierpinski gas-
ket antenna, IEEE Trans. Anten. Propag., 48(5):713719
(May 2000).
30. C. Borja, C. Puente, and A. Medina, Iterative network model
to predict the behaviour of a sierpinski fractal networks, IEE
Electron. Lett., 34(15):14431445 (July 1998).
31. C. Borja and C. Puente, Iterative network model to predict the
performance of sierpinski fractal antennas and networks,
Proc. IEEE Antennas and Propagation Society Int. Symp.,
Orlando, FL, July 1999, pp. 652655.
32. J. Soler, C. Puente, and J. Anguera, Extended multiperiodic
traveling wave model for the accurate analysis of the radia-
tion performance of the sierpinski fractal-like multiband an-
tenna, Microwave Opt. Technol. Lett., 36(1):6770 (Jan. 2003).
33. J. Soler, C. Puente, and J. Anguera, Results on a new extend-
ed analytic model to understand the radiation performance of
mod-p sierpinski fractal multiband antennas, Proc. IEEE An-
tennas and Propagation Society Int. Symp., Columbus, OH,
June 2003.
34. J. Soler, J. Romeu, and C. Puente, Mod-p sierpinski fractal
multiband antenna, Proc. AP2000 Millenium Conf. Antennas
and Propagation, Davos, Switzerland, 2002.
35. J. Soler, D. Garc a, C. Puente, and J. Anguera, Novel com-
bined mod-p structures: A complete set of multiband anten-
nas inspired on fractal geometries, Proc. IEEE Antennas and
Propagation Society Int. Symp., Columbus, OH, June 2003.
36. J. Parron, J. M. Rius, and J. Romeu, Numerical analysis of
highly iterated fractal antennas, Proc. IEEE Antennas and
Propagation Society Int. Symp., San Antonio, TX, 2002.
37. D. H. Werner, P. L. Werner, and K. H. Church, Genetically
engineered multiband fractal antennas, IEE Electron. Lett.,
37(19):11501151 (Sept. 2001).
38. D. H. Werner, P. L. Werner, J. W. Culver, S. D. Eason, and R.
Libonati, Load sensivity analysis for genetically engineered
miniature multiband fractal dipole antennas, Proc. IEEE An-
tennas and Propagation Society Int. Symp., June 2002, Vol. 4,
pp. 8689.
39. M. F. Pantoja, F. G. Ruiz, A. R. Bretones, R. G. Mart n, S. G.
Garc a, J. M. G. Arbesu , J. Romeu, J. M. Rius, D. H. Werner,
and P. L. Werner, On the design of small antennas using GA,
Proc. 27th ESA Antenna Technology Workshop on Innovative
Periodic Antennas, March 911, 2004.
40. J. M. Gonza lez, M. Navarro, C. Puente, J. Romeu, and A.
Aguasca, Active zone self-similarity of fractal-sierpinski an-
tenna veried using infra-red thermograms, IEE Electron.
Lett., 35(17):13931394 (Aug. 1999).
41. C. Puente, M. Navarro, J. Romeu, and R. Pous, Variations on
the fractal sierpinski antenna are angle, Proc. IEEE Anten-
nas and Propagation Society Int. Symp., Atlanta, June 1998.
42. M. Navarro, C. Puente, R. Bartolome, A. Medina, J. Romeu,
and R. Pous, Modicacion de la antena de sierpinski para el
ajuste de las bandas operativas (Modication of the Sierpin-
ski antenna to allocate operating bands), Proc. XII Symp.
Natl. URSI, Bilbao, 1997, pp. 363371.
43. C. T. P. Song, P. S. Hall, H. Ghafouri-Shiraz, and D. Wake,
Sierpinski monopole antenna with controlled band spacing
and input impedance, IEE Electron. Lett., 35(13):10361037
(June 1999).
44. C. T. P. Song, P. S. Hall, and H. Ghafouri-Shiraz, Perturbed
sierpinski multi-band fractal antenna with improved feeding
technique. IEEE Trans. Anten. Propag., 51(5):10111017
(2003).
45. C. Puente, J. Romeu, and A. Cardama, Fractal-Shaped An-
tennas, Frontiers in Electromagnetics Series, D. H. Werner
and R. Mittra, eds., IEEE Press, 2000.
46. Fractus, Fractal antennas: New generation of antennas for
compact and versatile telecommunication services. A practi-
cal application to mobile GSM 900 and 1800 networks,
Microwave J. (Jan. 2000).
47. C. Puente, J. Anguera, C. Borja, and J. Soler, Fractal-shaped
antennas and their application to GSM 900/1800, J. Inst. Br.
Telecommun. Eng. 2 (JulySept. 2001).
48. J. Soler, C. Puente, and A. Puerto, A dual-band bidirectional
multilevel monopole antenna, Microwave Opt. Technol. Lett.,
34(6):445448 (Sept. 2002).
49. J. Romeu and Y. Rahmat-Samii, Fractal fss: A novel dual-
band frequency selective surface, IEEE Trans. Anten. Prop-
ag., 48(7):10971105 (July 2000).
50. J. Romeu and Y. Rahmat-Samii, A fractal based fss with dual
band characteristics, Proc. IEEE Antennas and Propagation
Society Int. Symp., Orlando, FL, July 1999, Vol. 3, pp. 1734
1737.
51. J. Romeu and Y. Rahmat-Samii, Dual band fss with
fractal elements, IEE Electron. Lett., 35(9):702703 (April
1999).
52. D. H. Werner and D. Lee, Design of dual-polarized multiband
frequency selective surfaces using fractal elements, IEE Elec-
tron. Lett., 36(6):487488 (March 2000).
53. E. S. Siah, B. L. Ooi, P. S. Kooi, and X. D. Xhou, Experimental
investigation of several novel fractal antennasvariants of
the sierpinski gasket and introducing fractal fss screens,
Proc. Asia Pacific Microwave Conf., 1999, Vol. I, pp. 1710
1713.
54. J. Soler, Novel Techniques for Small and Multifrequency An-
tennas, Ph.D. dissertation, Dept. Signal Theory and Commu-
nications, Univ. Polite`cnica de Catalunya, in progress.
55. J. Romeu and J. Soler, Generalized sierpinski fractal multi-
band antenna, IEEE Trans. Anten. Propag., 49(8):12371239
(Aug. 2001).
56. J. Soler and J. Romeu, Dual-band sierpinski fractal monopole
antenna, Proc. IEEE Antennas and Propagation Society Int.
Symp., Salt Lake City, July 2000.
57. J. Soler and C. Puente, Loaded Antenna, Patent PCT/EP01/
11914.
58. R. Quintero and C. Puente, Multilevel and Space-Filling
Groundplanes for Miniature and Multiband Antennas,
Patent Appl. PCT/EP01/10589.
59. J. Soler, C. Puente, and J. Anguera, Multifrequency proper-
ties of monopole antennas using multilevel ground planes in-
spired on the sierpinski fractal shape, Proc. IEEE Antennas
and Propagation Society Int. Symp., Columbus, OH, June
2003.
60. S. R. Best, Operating band comparison of the perturbed sier-
pinski and modied parany gasket antennas, IEEE Anten.
Wireless Propag. Lett., I:3538 (2002).
61. S. R. Best, On the radiation pattern characteristics of the si-
erpinski and modied parany gasket antennas, IEEE Anten.
Wireless Propag. Lett., I:3942 (2002).
62. C. T. P. Song, P. S. Hall, and H. Ghafouri-Shiraz, Multiband
quasi-fractal multiple ring monopole antenna, IEEE Trans.
Anten. Propag., 51(4):722729 (2003).
63. J. Soler, C. Puente, and J. Anguera, Solutions to tailor the
radiation patterns of 2d and 3d multiband antennas based on
the sierpinski fractal, Proc. IEEE Antennas and Propagation
Society Int. Symp., Columbus, OH, June 2003.
64. C. Borja, C. Puente, J. Romeu, and J. Anguera, Fractal
multiband patch antenna, Proc. Millennium Conf. Antennas
and Propagation AP2000, Davos, Switzerland, April 2000.
65. C. Borja, C. Puente, J. Anguera, J. Romeu, and R. Pous, Estu-
dio experimental del parche de sierpinski (Experimental
study of the Sierpinski patch antenna), Proc. XII Symp.
Natl. URSI, Santiago de Compostela, 1999.
66. C. Borja, Fractal Microstrip Patch Antennas with Fractal Pe-
rimeter and Self-Afne Properties, Ph.D. dissertation, Dept.
Signal Theory and Communications, Univ. Polite`cnica de Cat-
alunya, 2001.
67. C. Borja and J. Romeu, Multiband sierpinski fractal patch
antenna, Proc. IEEE Antennas and Propagation Society Int.
Symp. Salt Lake City, July 2000, pp. 17081711.
68. J. Anguera, Fractal and Broadband Techniques on Miniature,
Multifrequency, and High-Directivity Microstrip Patch Anten-
nas, Ph.D. dissertation, Dept. Signal Theory and Communi-
cations, Univ. Polite`cnica de Catalunya, 2003.
69. C. Borja, C. Puente, and J. Anguera, Dual-Band Dual-Polar-
ized Antenna Array, Patent PCT/EP01/04288.
70. J. Anguera, C. Puente, C. Borja, G. Font, and J. Soler, A sys-
tematic method to design single-patch broadband microstrip
patch antennas, Microwave Opt. Technol. Lett., 31(3):185188
(Nov. 2001).
71. J. Anguera, C. Puente, and C. Borja, A procedure to design
stacked microstrip patch antenna based on a simple network
model, Microwave Opt. Technol. Lett., 30(3):149151 (Aug.
2001).
72. S. E. El-Khami, M. A. Aboul-Dahab, and M. I. Elkashlan, A
simplied koch multiband fractal array using windowing and
quantization techniques, Proc. IEEE Antennas and Propaga-
tion Society Int. Symp. Salt Lake City, July 2000, Vol. 3, pp.
17161719.
73. X. Liang, W. Zhensen, and W. Wenbing, On the synthesis of
fractal patterns from the concentric-ring array, IEE Electron.
Lett., 32:19401941 (Oct. 1996).
74. D. Baldacci and D. H. Werner, An efcient recursive proce-
dure for calculating the driving point impedance of linear and
planer fractal arrays, Proc. IEEE Antennas and Propagation
Society Int. Symp. Boston, July 2001, Vol. 3, pp. 620623.
75. D. H. Werner and P. L. Werner, A general class of self-scalable
and self-similar arrays, Proc. IEEE Antennas and Propaga-
tion Society Int. Symp., Orlando, FL, July 1999, Vol. 4, pp.
28822885.
76. R. L. Haupt and D. H. Werner, Fast array factor calculations
for fractal arrays, Proc. 13th Annual Review of Progress in
Applied Computational Electromagnetics (ACES), Naval
Postgraduate School, Montereu, CA, March 1997, Vol. I, pp.
291296.
77. H. A. Wheeler, Fundamental limitations on small antennas,
Proc. IRE 14791488 (Dec. 1947).
78. L. J. Chu, Physical limitations of omnidirectional antennas, J.
Appl. Phys., 19:11631175 (Dec. 1948).
79. H. A. Wheeler, The radiansphere around a small antenna,
Proc. IRE 13251331 (Aug. 1959).
80. J. Anguera, C. Puente, and J. Soler, Miniature monopole an-
tenna based on the fractal hilbert curve, Proc. IEEE Antennas
and Propagation Society Int. Symp., San Antonio, TX, June
2002.
81. J. Anguera, E. Mart nez, C. Puente, and J. Soler, El monopolo
de hilbert: una curva fractal de dos dimensiones (The Hilbert
monopole: A two dimensional fractal curve), Proc. 17th Natl.
Symp. Scientific International Union of Radio, Alcala de Hen-
ares, Spain, Sept. 2002.
82. C. Puente, E. Rozan, and J. Anguera, Space-lling miniature
antennas, Patent WO0154225.
83. K. J. Vinoy, K. A. Jose, V. K. Varadan, and V. V. Varadan,
Resonant frequency of hilbert curve fractal antennas, Proc.
IEEE Antennas and Propagation Society Int. Symp., Boston,
July 2001, Vol. 3, pp. 648651.
84. J. Zhu, A. Hoorfar, and N. Engheta, Feed-point effects in hil-
bert-curve antennas, IEEE Antennas and Propagation So-
ciety Int. Symp. URSI Digest, San Antonio, TX, June 2002,
p. 373.
85. P. Thang and P. Wahid, Scaling property of the koch fractal
dipole, Proc. IEEE Antennas and Propagation Society Int.
Symp., Boston, July 2001, Vol. 3, pp. 150153.
86. D. Gala, J. Soler, C. Puente, C. Borja, and J. Anguera, Min-
iature microstrip patch antenna loaded with a space-lling
transmission line based on the fractal hilbert curve, Micro-
wave Opt. Technol. Lett., 38(4):311312 (Aug. 2003).
87. J. McVay, N. Engheta, and A. Hoorfar, High-impedance
metamaterial surfaces using hilbert-curve inclusions,
IEEE Microwave Wireless Compon. Lett., 14(3):130132
(March 2004).
88. C. T. P. Song, Novel Antenna Design for Future Mobile Sys-
tems, Ph.D. dissertation, School Electronic and Electrical
Engineering, Univ. Birmingham, UK, 2001.
89. C. T. P. Song, P. S. Hall, H. Ghafouri, and I. Henning. Short-
ed fractal sierpinski monopole antenna, Proc. IEEE Anten-
nas and Propagation Society Int. Symp., Boston, July 2001,
Vol. 3, pp. 138141.
90. J. P. Gianvittorio, Fractals, MEMS, and FSS Electromagnet-
ic Devices: Miniaturization and Multiple Resonances, Ph.D.
dissertation, Univ. California, Los Angeles, 2003.
91. J. P. Gianvittorio and Y. Rahmat-Samii, Fractal antennas: A
novel antenna miniaturization technique, and applications,
IEEE Anten. Propag. Mag., 44(1):2036 (Feb. 2002).
92. J. Anguera, C. Puente, C. Borja, and J. Romeu, Miniature
wideband stacked microstrip patch antenna based on the
sierpinski fractal geometry, Proc. IEEE Antennas and Prop-
agation Society Int. Symp., Salt Lake City, July 2000.
93. J. Anguera, C. Puente, and C. Borja, Miniature Broadband
Ring-like Microstrip Patch Antenna, Patent Appl. PCT/
EP01/01287.
94. G. J. Walker and J. R. James, Fractal volume antennas, IEE
Electron. Lett., 34(16):15361537 (Aug. 1998).
95. B. Sapoval, Th. Gobron, and A. Margolina, Vibrations of
fractal drums, J. Am. Phys. Soc., 67(21):29742977 (Nov.
1991).
96. B. Sapoval and Th. Gobron, Vibrations of strongly irregular
or fractal resonators, J. Am. Phys. Soc., 47(5):30133024
(May 1993).
97. C. Borja and J. Romeu, Fracton vibration modes in the sier-
pinski microstrip patch antenna, Proc. IEEE Antennas and
Propagation Society Int. Symp., Boston, July 2001.
98. J. Anguera, C. Puente, and C. Borja, Broadside High-Direc-
tivity Microstrip Patch Antennas, Patent NPCT/EP03/00757.
99. C. Borja and J. Romeu, On the behaviour of the koch island
fractal boundary microstrip patch antenna, IEEE Trans.
Anten. Propag., 51(6):12811291 (June 2003).
100. J. Anguera, C. Puente, C. Borja, and R. Montero, Bowtie
microstrip patch antenna based on the sierpinski fractal,
Boston, July 2001, Vol. 3, pp. 162165.
101. J. Anguera, C. Puente, C. Borja, and R. Montero, Antena
microstrip miniatura y de alta directividad basada en el
fractal de sierpinski (Miniature and high-directivity micro-
strip antenna based on the Sierpinski fractal), Proc. 16th
Natl. Symp. Scientific Int. Union of Radio, Madrid, Spain,
Sept. 2001.
102. G. Montesinos, J. Anguera, C. Puente, and C. Borja, The si-
erpinski fractal bowtie patch: A multifracton-mode antenna,
San Antonio, TX, June 2002.
Modos multifracton en la antena microstrip bowtie fractal
de sierpinski (Multifracton nodes in a microstrip antenna
based on the bowtie Sierpinski fractal), Proc. 17th Natl.
Symp. Scientific Int. Union of Radio, Alcala de Henares,
Spain, Sept. 2002.
A microstrip array operating in a fracton mode, Proc. IEEE
Antennas and Propagation Society Int. Symp., Columbus,
OH, June 2003.
Agrupacion submuestreada operando en modo fracton ut-
ilizando antenas microstrip inspiradas en el fractal de sier-
pinski (Undersampled array operating in a fracton mode
using microstrip antennas inspired in the Sierpinski frac-
tal), Proc. 18th Natl. Symp. Scientific Int. Union of Radio, La
Corun a, Spain, Sept. 2003.
106. J. Anguera, C. Puente, C. Borja, and J. Soler, Scanning prop-
erties in an under-sampled microstrip array using sierpin-
ski fractal-inspired elements, Proc. 27th European Space
Agency (ESA) Antenna Workshop on Innovative Periodic
Antennas, Santiago de Comnpostela, Spain, March 2003.
107. D. L. Jaggard, On fractal electrodynamics, in D. L. Jaggards
and H. N. Kritikos, eds., Recent Advances in Electromagnet-
ics Theory, Springer-Verlag, New York, 1990, pp. 183224.
108. X. Sun and D. L. Jaggard, Wave interactions with general-
ized cantor bar fractal multilayers, J. Appl. Phys., 70(5):
25002507 (Sept. 1991).
109. N. Cohen and R. G. Hohed, Array sidelobe reduction by
small position offsets of fractal elements, 16th Annual Re-
view of Progress in Applied Computational Electromagnet-
ics, Naval Postgraduate School, Monterey, CA, March 2000,
Vol. 2, pp. 822828.
110. D. H. Werner and S. Ganguly, An overview of fractal antenna
engineering research, IEEE Anten. Propag. Mag., 45(1):38
57 (Feb. 2003).
111. D. L. Jaggard, Fractal electrodynamics: From super anten-
nas to superlattices, in J. L. Vehel, E. Lutton, and C. Tricot,
eds., Fractals in Engineering Series, Springer-Verlag, New
York, 1997, pp. 204221.
FREQUENCY CONVERTERS AND MIXERS
ILTCHO ANGELOV
ERIK L. KOLLBERG
HERBERT ZIRATH
Chalmers University of
Technology
Goteburg, Sweden
An essential part of most microwave receivers is the
frequency converter, a device that converts the frequency
of an incoming signal to another frequency. The output
frequency may be downconverted, upconverted, multi-
plied, or divided. Important and common is the mixer,
which downconverts a high-frequency input signal f
s
to a
much lower and more manageable signal f
IF
, preserving
information concerning the amplitude, frequency, and
phase of the input signal. The devices used are nonlinear
(i.e., the relationship between current and voltage is not
FREQUENCY CONVERTERS AND MIXERS 1635
linear). The devices may have either two terminals
(diodes) or three terminals (transistors).
Mixers are important for very high frequencies, where
ampliers are not available and direct amplitude or phase
detection is difcult. A mixer can be used to downconvert,
say, a terahertz frequency signal to a microwave frequency
one, where electronic methods are readily available for
amplication and any kind of demodulation. In fact, in
almost any radio set or mobile telephone receiver or base
station, there are several mixers and other types of
frequency converters.
Both two-terminal devices (diodes) and three-terminal
devices (transistors) are used in mixers. The frequency
conversion is accomplished by using the nonlinear proper-
ties of the device. Virtually all semiconductor devices, such
as diodes and transistors, show nonlinear properties in
certain bias ranges. Common devices for microwave mixer
applications are the Schottky diode, the eld-effect tran-
sistor (FET), and the bipolar transistor. There are many
other devices available as well, for example, the super-
conducting tunneling device (SIS) and the superconduct-
ing hot-electron device for low-noise (high-sensitivity)
millimeter- and sub-millimeter-wave receivers. For infra-
red wavelengths, metalinsulatormetal (MIM) devices
and, for optical frequencies, photoconducting devices
have been used. Note that ordinary resistors, capaci-
tances, and inductances are linear components.
Figure 1 gives an example of how mixers are used in a
receiving system. The antenna is connected directly to the
mixer. For example, in a TV satellite receiver, an amplier
is placed just after the antenna to increase the signal
amplitude. In reality there are many other systems (e.g.,
radar, radio, measurement systems) where mixers are
used to frequency downconvert the input signal.
To give a simple illustration of how a diode mixer may
work, consider the detector circuit shown in Fig. 2 using
an ideal diode (zero resistance in the forward direction
and innite resistance in the backward direction). If a
sinusoidal voltage V(t) V
LO
cos(2pf
LO
t) is applied, it will
be rectied (detected) by the diode. The voltage over the
resistance v
R
(t) will be a constant DC voltage proportional
to V
LO
. Next, add a small signal voltage dv
s
cos(2pf
s
t), that
is, V(t) V
LO
cos(2pf
LO
t) dv
s
cos(2pf
s
t). Assuming that
dv
s
/V
LO
51, the resulting voltage V(t) will become ampli-
tude-modulated, as shown in Fig. 2. The detected voltage
v
R
(t) over the load resistance R will be proportional to
the envelope V
R
(t) (assuming t RC51/o
IF
) (see Fig. 3),
(a)
(b)
Antenna
Signal input
ov
s
cos(o
s
t)
L
o
w

p
a
s
s

f
i
l
t
e
r
1
3
Nonlinear
three-
terminal
device
IF
amplifier
IF
out
Local
oscillator
V
LO
cos(o
LO
t)
2
High
pass
filter
High
pass
filter
Antenna
Signal input
ov
s
cos(o
s
t)
D
i
p
l
e
x
e
r
H
i
g
h

p
a
s
s

f
i
l
t
e
r
L
o
w

p
a
s
s

f
i
l
t
e
r
1 2
Nonlinear
two-
terminal
device
IF
amplifier
IF
out
Local
oscillator
V
LO
cos(o
LO
t )
Figure 1. Typical mixer block diagram show-
ing layouts for a two-terminal device (diode) (a)
and a three-terminal device (transistor) (b). At
the input there is a highpass lter that will
prevent any low-frequency IF power from es-
caping to the mixer input (left). The lowpass
lter will stop any input signal or LO power
from going to the IF circuit (right).
R C V(t)
i
d
(t) i
R
(t)
v
R
(t )
Figure 2. A rectifying diode circuit. If the capacitance C and the
resistance R are large enough, v
R
(t) will follow the envelope of v(t)
(see Fig. 3).
Envelope = V
R
(t )
+1
1
0
v(t )
V
LO
ov
s
(t)
V
LO
t
2
Figure 3. An example with a drive voltage v(t)/V
LO
cos(o
LO
t)
(dv
s
/V
LO
) cos(o
s
t), where dv
s
/V
LO
0.2 and o
LO
/o
s
1.18. The IF
is 0.18o
LO
.
1636 FREQUENCY CONVERTERS AND MIXERS
that is
v
R
t V
LO
1
dv
s
V
LO
cos2pf
IF
t

1
where f
IF
|f
LO
f
s
| is the intermediate frequency (IF).
The DC part of Eq. (1) is the detected local oscillator
(LO), and the alternating part is identical to the IF
voltage and, in this example, is equal in amplitude to
the input signal. This IF signal is fed into an amplier as
described in Fig. 1.
Note that in the simplied example above we have not
correctly accounted for a number of parameters, such as
the impedances of the LO and signal sources. This means
that the IF voltage will not become equal to dv
s
. A more
detailed description of a more correct calculation is given
in the Section 3. For details concerning microwave mixers,
see Refs. 1 and 2.
1. GENERAL PROPERTIES OF TWO-TERMINAL
NONLINEAR DEVICES
Below we describe some general results obtained when a
two-terminal device such as the Schottky diode is excited
with a sinusoidal signal. The impedance of the Schottky
diode is voltage-dependent. The speed of this device is
indeed high. It shows a nonlinear behavior up to several
terahertz.
1.1. Frequency Multiplication
Consider a nonlinear diode device exposed to LO (or
pump) power yielding a large voltage swing V(t) V
LO
cos(o
LO
t) over the diode. (Note: Below we use o2pf.)
Since the relation between current and voltage is not
linear (i.e., Iaconst. V), the resulting current will not
have a sinusoidal shape like that of the input voltage.
However, the current in this case is still a periodic
function versus time with the same periodicity, t
2p/o
LO
, as the LO frequency and can consequently be
expressed as a Fourier series with harmonics of the
LO, no
LO
:
It
X
1
n0
i
n
cosno
LO
t j
n
2
The component i
0
is the DC component. To obtain power at
a particular harmonic, for example, the third harmonic, it
is required that the current component at 3o
LO
pass
through a resistance R
3
delivering a power of
1
2
i
2
3
R
3
. To
avoid any power being delivered at other harmonics, it is
necessary to ensure that the device is reactively termi-
nated at these harmonics.
In reality, the impedance of most nonlinear devices is
complex with both the real and the imaginary parts
voltage (or current)-dependent. For a more detailed the-
ory, see Section 3.2.
1.2. Frequency Conversion
If two signal voltages at o
LO
and o
s
are simultaneously
interacting with the nonlinear diode impedance, the
resulting current can be expressed in a more complex
Fourier series:
It
X
1
m1
X
1
n1
i
m;n
cosmo
LO
no
s
t j
m;n
3
In the case when both the signal voltage and the local
oscillator voltage are large, a large number of frequency
conversion products are obtained. Obviously it is possible
to generate power at any frequency mo
LO
7no
s
. If for
mn1 the required output frequency is higher than the
signal frequency, one has frequency upconversion, and if
the output frequency is lower than the signal frequency,
one has frequency downconversion.
1.3. Linear Mixing
As already mentioned, a mixer receiver usually handles
small signals. We have a small signal case if the ampli-
tude of the signal at o
s
is much smaller than the LO
amplitude at o
LO
. In this case only harmonics of o
LO
are
important, and we are left with
It %
X
1
m1
i
m
cosmo
LO
o
s
t j
m
4
In most mixer applications, we are interested in the
intermediate frequency, o
IF
|o
LO
o
s
|. An IF load
resistance in the circuit will allow power at the IF
frequency to be extracted (compare with Fig. 2). Equation
(4) suggests that power may go not only to the IF but also
to harmonic sidebands, as illustrated in Fig. 4. The only
way to prevent this power loss is to make sure that the
Power to upper sideband
(image frequency)
Power to harmonic sidebands
Signal in
at lower sideband
0
o
IF
o
i
o
s
o
LO
2o
LO
3o
LO
Figure 4. Power ow in a lower sideband
mixer. The input signal power is distributed
not only to the IF but also to the harmonic
sidebands.
harmonic sideband current components are facing impe-
dances that are purely reactive.
Note that an IF signal (Fig. 4) will be created if either
o
su
o
LO
o
IF
or o
sl
o
LO
o
IF
. The former frequency
o
su
is the upper sideband, and the latter o
sl
the lower
sideband. For a lower sideband mixer the upper sideband
is denoted the image frequency, and vice versa for an upper
sideband mixer. For a lower or upper sideband case, some
signal power may go to the image frequency. In an image-
reject mixer a lter prevents the image frequency from
entering the mixer. In the image-enhanced mixer, the
image terminal is terminated reactively so that the con-
version loss is reduced.
At high signal powers, there may be confusion because
the mixer may produce output signals in the IF band (see
Section 1.5 below) for
jmo
LO
no
s
j o
IF
5
1.4. Harmonic Mixer
In a harmonic and small-signal linear mixer one has o
IF
|no
LO
o
s
|. A spectrum analyzer always uses a harmo-
nic mixer to analyze the signal, and the harmonic number
n can be very high (410). The power ow in a harmonic
mixer is shown in Fig. 5.
1.5. Intermodulation
All mixer products created by two or more signals are
called intermodulation (IM) products. Most IM products
are unwanted.
For example, if two signals at slightly different fre-
quencies o
s1
and o
s2
and with a power of the same order of
magnitude as the LO are interacting with the nonlinear
device, the current will contain frequency products as
shown in the following equation:
It
X
1
m1
X
1
n1
X
1
k1
i
m;n
cosmo
LO
no
s1
ko
s2
t
j
m;n;k
6
All signals at frequencies fullling the requirement
o
s1
% jmo
LO
no
s2
o
IF
j 1om; no1 7
may create IM products in the IF band. Indeed, inter-
modulation must be considered a potentially serious pro-
blem in all applications where strong signals may occur.
2. SYSTEMS ASPECTS
In a system, the following properties are important: (1)
conversion loss, (2) noise properties, and (3) intermodula-
tion properties.
2.1. Conversion Loss
An important property of a mixer is the conversion loss L,
dened as
L
P
s
P
IF
signal power available at the input

IF power delivered to the IF load
8
In most practical mixers the conversion loss is larger than
one. However, it is possible to obtain gain owing to
parametric amplication caused by a nonlinear capaci-
tance. In certain congurations using transistors (see Ref.
1 or 4) conversion gain can be obtained. Superconducting
mixers, however, due to quantum phenomena, can show
stable conversion gain.
There are several loss mechanisms causing the conver-
sion loss:
1. Losses due to absorption in the nonlinear device
2. Losses due to power lost to harmonic sidebands
3. Losses due to reection at the input port
4. Losses due to reection at the IF output port
See also Section 3.
2.2. Mixer Receiver Noise
The important noise measure in practical applications is
always the total receiver noise temperature. The contribu-
tions approximately in order of importance are (1) mixer
device noise, (2) the IF amplier noise, (3) thermal noise
from the mixer circuit, and (4) LO noise. There are two
different noise measures usually cited in the literature:
the single sideband (SSB) noise temperature and the
double sideband (DSB) noise temperature.
When calculating the noise temperature of a mixer, it is
advisable to always start adding up all noise contributions
at the IF amplier input (see Fig. 6). The temperature
Signal in
at lower sideband
0
o
IF
o
i
o
s
o
LO
2o
LO
3o
LO
Figure 5. Power ow in a second-harmonic
lower sideband mixer.
then becomes
T
in
T
l
L
s
T
l
L
i
X
1
n2
T
l
L
n
T
l
L
n

T
M out
T
IF
9
where L
s
, L
i
, L
n
, and L
n
are the conversion losses at
the signal, image, and upper harmonic sidebands and
lower harmonic sidebands, respectively. T
l
is the noise
temperature of the input resistance, T
M out
is the noise
from the mixer diode entering the IFamplier. As the next
step, identify the signal to noise ratio:
P
s
P
in
P
s
=L
s
kT
in
Df

P
s
kT
syst
Df
10
where T
syst
is by denition the system noise temperature
for a single-sideband receiver. It is assumed that the
useful and interesting signal enters only one sideband.
Hence
T
syst;SSB
T
in
L
s
T
l
1
L
s
L
i
X
1
n2
L
s
L
n
L
s
L
n

" #
T
MXR;SSB
L
s
T
IF
11
where we have dened the equivalent noise temperature
of the mixer itself, T
MXR,SSB
T
M out
L
s
.
For the double-sideband case, one assumes a useful
signal to enter both the upper and the lower sidebands.
Hence the signal-to-noise ratio for this case should be
dened as
P
s
P
n
P
s
1=L
s
1=L
i
kT
in
Df

P
s
kT
syst
Df
12
that is,
T
syst;DSB
T
in
1=L
s
1=L
i
T
l
1
L
s
L
i
L
s
L
i
X
1
n2
1
L
n
1
L
n

" #
T
MXR;DSB
L
s
L
i
L
s
L
i
T
IF
13
where the double sideband noise temperature of the mixer
itself is dened as
T
MXR;DSB
T
M out
L
s
L
i
L
s
L
i
14
Note that if L
s
L
i
, both the single-sideband mixer and
system noise temperatures are twice as large as for the
double-sideband case. The LO noise (if important) can be
taken into account by adding a certain amount at the
input port. Also note that the noise entering the mixer at
the harmonic sidebands may considerably inuence the
total receiver performance.
3. SCHOTTKY DIODE MIXERS
The Schottky diode mixer is the most common type of
mixer for frequencies from megahertz to terahertz.
3.1. Schottky Diode for Mixer Applications
A schematic diagram of a common design of millimeter-
wave Schottky barrier diode is shown in Fig. 7. Note that
the radiofrequency (RF) current is owing from the diode
contact at the surface of the diode chip to the back contact.
Hence the RF series resistance is slightly larger for RF
than for DC. The currentvoltage (IV) characteristic of
the junction itself can be calculated from
i
d
I
0
exp
qV
j
ZkT

1
!
15
where I
0
is the saturation current, V
j
the junction voltage,
q the charge of the electron, k Boltzmanns constant, T the
physical temperature, and Z the ideality factor, which
for good diodes at room temperature is between, say,
1.03 and 1.10.
The small-signal RF junction properties of a Schottky
diode can be modeled as a nonlinear resistance r
j
in
parallel with a nonlinear capacitance C
j
ri
d

@V
j
@i
d
kZT
qi
d
CV
j

C
0
1
V
j
V
b
r 16
where C
0
is the zero bias capacitance and V
b
is the built-in
voltage of the diode. For more details, see Ref. 3.
Mixer
Input
load
T
I
IF amplifier
T
IF
T
in
Figure 6. Receiver conguration for calculating the receiver
noise. For the calculation it is wise to determine the noise
temperature at the input of the IF amplier T
in
.
Schottky contact
Passivation Depleted region
W
Back contact
Epitaxial layer
Highly doped
region n
++
r (i
d
)
i
d
V
j
R
s
C(V
j
)
Figure 7. Simple design of a GaAs Schottky barrier diode and
the corresponding equivalent circuit. The higher the frequency,
the smaller the diodes required.
A common measure of the high-frequency properties is
the cutoff frequency, which is dened for the diode at zero
bias,
f
c
1
2pR
s
C
0
17
where R
s
is measured using the DC IV characteristic. For
good mixer performance this cutoff frequency must be
much larger than the signal frequency. The series resis-
tance uses up signal power and should for obvious reasons
be made as small as possible. For the conversion efciency
the nonlinear capacitance is of much less importance than
the nonlinear resistance. The diode is typically designed
as a thin n-doped active layer of 1000A
(Mott diode),
which leaves one with a diode with reduced capacitance
variation and a minimum series resistance.
GaAs is preferred for millimeter-wave diodes. The main
reason is that the high mobility of GaAs yields a low series
resistance and consequently a high cutoff frequency. An-
other advantage (to silicon) is that carriers do not freeze
out when the diode is cooled to cryogenic temperatures in
order to improve the mixer noise properties [see Eq. (7)].
The noise performance and the conversion efciency
are the prime properties of the diode for use in mixer
applications. Noise properties are discussed next.
3.1.1. Shot Noise. The shot noise is due to uctuations
in the electron particle current between cathode and
anode of the diode [3]. The root-mean-square (RMS)
uctuations in the current are di
2
2eIDf, where Df is a
small frequency interval. The noise power of the Schottky
diode can then be calculated as
P
n
di
2
4
@I
d
@V
j
1
2
kZTDf 18
Identifying this equation with the ordinary Johnson noise
expression, P
n
kT Df, it is seen that the equivalent noise
temperature of the Schottky diode is
T
sh
1
2
ZT 19
Note that the noise temperature decreases linearly with
decreasing physical temperature. However, the charge
transport at room temperature over the Schottky barrier
is due to thermionic emission and decreases when the
temperature is lowered. Hence for temperatures on the
order of 50100K and below, temperature-independent
tunneling becomes the dominant process for electrons
passing the barrier, and the equivalent temperature ZT
of Eq. (6) becomes
T
tunnel
qh
k
N
d
4em
r
20
where N
d
is the doping concentration in the epitaxial
layer, e the dielectric constant of the semiconductor, and
m
the effective mass of the electron. In practice, T

tunnel
is 50 K for N
d
310
16
. This doping concentration is
recommended for mixers operated at 15 or 20 K, a
typical temperature for commercial cryogenic cooling ma-
chines.
3.1.2. 1/f Noise. There is excess noise at low frequen-
cies (of the order 100kHz and lower for a good diode),
which is related to surface phenomena at the metal
semiconductor interface. This noise is normally not im-
portant in mixers for millimeterwave receivers. However,
it is an important limiting factor for certain radar and
communications systems.
3.1.3. Thermal Noise. The series resistance is essen-
tially an ordinary resistor and consequently causes ordin-
ary thermal (Johnson) noise. The main noise contribution
of this type comes from the substrate (corresponding
resistance R
sub
) and is denoted as T
sub
.
3.1.4. Hot-Electron Noise and Intervalley Scattering
Noise. Since the diode area is very small in order to
render the capacitance reasonably small, a high current
density is required to make r
j
R
s
. This means that
electrons may obtain energies larger than the energy
related to their thermal movement and hot-electron noise
is obtained. The increase in energy also means that
electrons can be transferred from the main G valley in
the Ek diagram (for details, see Ref. 3) to the upper L
valley, causing uctuations in the electron velocity and
intervalley scattering noise is obtained. The hot electron
and intervalley scattering noises occur essentially in the
undepleted part of the epilayer (resistance R
epi
) and are
denoted as T
epi
.
3.2. Large-Signal Analysis by Harmonic Balance
When the nonlinear device is pumped by the LO, harmo-
nic currents are created [46]. The equivalent circuit of a
Schottky diode mixer is shown in Fig. 8. Note that the
series resistance is assumed to be linear and will be
included in the embedding circuit.
The current I
e
(t) contains harmonics of the LO pump
frequency, ko
LO
. For a Schottky diode, this current con-
sists of two parts: one that is associated with the nonlinear
resistance i
d
(t), and one with the parallel nonlinear capa-
citance i
c
(t). We have [using the complex notation; of
Eq. (2)]
I
e
t i
C
t i
d
t
X
k1
k1
I
ek
e
jho
LO
t
I
ek
I
ek
21
This current I
e
(t) ows through the embedding
circuit, creating voltages at harmonic frequencies
at ko
P
p:
V
j
t
X
k1
k 1
V
k
e
jko
LO
t
V
k
V
k
22
The boundary conditions set by the embedding circuit
require that
V
k
I
ek
Z
ek
ko
LO
R
s
ko
LO
k 2; 3; . . . ; 1
23
V
1
V
LO
I
e1
Z
e1
o
LO
R
s
o
LO
24
V
0
V
DC
I
e0
Z
e
0 R
s
0 25
where V
LO
and V
DC
are the LO and DC bias voltages,
respectively. The frequency dependence of R
s
is due to the
skin effect. If V
j
(t) is known, i
d
(t) and i
C
(t) (see Fig. 8) can
be calculated from Eqs. (15) and (16). We now have a
nonlinear problem to solve in order to determine I
ek
and
V
ek
. Several iteration type methods have been suggested.
See Refs. 4 and 5 for more details.
Having determined the components I
k
and V
k
, we have
V
j
(t) and I
e
(t), and we can determine i
d
(t) i
C
(t), g
d
(t), and
C
j
(t):
g
d
t
X
k1
k1
G
k
e
jko
LO
t
G
k
G
k
26
C
j
t
X
k 1
k1
C
k
e
jko
LO
t
C
k
C
k
27
These equations together with the embedding impedance
Z
e
(o) allow us to determine the small signal properties of
the mixer.
3.3. Small-Signal Analyses
The relation between the small-signal current and voltage
vectors dI and dV can be expressed in a more general
form as
dI Y dV 28
where
Y
mn
G
mn
jo
0
mo
LO
C
mn
29
where for convenience we use o
0
for o
IF
. It is convenient to
form an augmented Y matrix, Y
0
, as indicated in Fig. 9.
This augmented network contains the whole mixer, in-
cluding diode and embedding network, but does not con-
tain signal sources associated with these terminations.
Since we dene the signal sources as current sources, the
augmented network in this case is open-circuited. For the
augmented network
dI
0
Y
0
dV 30
and
Y
0
Y diag
1
Z
em
R
sm
!
31
Inverting Eq. (23) yields
dV Z
0
dI
0
32
where
Z
0
1
Y
0

.
.
.
.
.
.
.
.
.
. . . Z
0
11
Z
0
10
Z
0
11
. . .
. . . Z
0
01
Z
0
00
Z
0
10
. . .
. . . Z
0
11
Z
0
10
Z
0
11
. . .
.
.
.
.
.
.
.
.
.
2
6
6
6
6
6
6
6
6
6
6
4
3
7
7
7
7
7
7
7
7
7
7
5
33
3.4. Mixer Port Impedances
The port impedance Z
mm
, dened in Fig. 9, can be
determined if the corresponding embedding impedance is
open-circuited, that is
Z
m
Z
0
mm;1
34
where the subscript Nmeans that Z
0
mm
is evaluated for
Z
em
N. The IF output impedance becomes
Z
out
Z
0
R
s0
Z
0
00;1
R
s0
35
R
s
(o)
LO
dc
o
LO
I
e
(t)
Z
e
(o)
C
j
(v
j
)
V
j
g
d
(i
d
)
+
Diode
Intrinsic diode
i
c
i
d
Embedding
circiut
Figure 8. Equivalent circuit of a mixer. For
the intrinsic diode, C
j
and g
d
are nonlinear and
are characterized in the time domain, while
the diode series resistance R
s
and the embed-
ding impedance Z
e
are linear and can be
described in the frequency domain.
3.5. Conversion Loss
The conversion loss of a mixer is dened as [cf. Eq. (8)]
L
power available from source Z
e1
power delivered to load Z
e0
36
yielding the conversion loss
L
1
4jZ
0
01
j
2
jZ
e0
R
s0
j
2
ReZ
e0
jZ
e1
R
s1
j
2
ReZ
e1
37
where Z
ek
is dened in Fig. 9 (see gure caption) and Z
0
01
is
the 01 diagonal element of Eq. (33). A more general
expression for the conversion loss from any sideband j to
any other sideband i is
L
ij
1
4jZ
0
ij
j
2
jZ
ei
R
si
j
2
ReZ
ei
jZ
ej
R
sj
j
2
ReZ
ej
38
3.6. Equivalent-Noise Temperature of the Mixer
The equivalent input noise temperature of the mixer T
M
is
dened as the temperature that the real part of the Z
el
(lower sideband) must have in order to generate the same
noise voltage as the diode itself generates during mixer
operation. It is
T
M
hdV
2
N0
i
4kDf
jZ
e1
R
s1
j
2
jZ
0
01
j
2
ReZ
e1
39
where /dV
2
N0
S is
hdV
2
N0
i Z
0
0
hdI
0
s
dI
y
s
iZ
y
0
Z
0
0
hdI
0
T
dI
y
T
iZ
y
0
40
where
hdI
0
Tm
dI
0
Tm
i
4kT
eq
R
sm
Df
jZ
em
R
sm
j
2
mO0
4kT
eq
R
sm
Df
jZ
0
j
2
m0
41
and
hdI
0
sm
dI
0
sn
i 2eI
mn
Df 42
I
mn
is the current component at the harmonic (mn)
and Z
0
is the zero row of the matrix Z
0
, Eq. (33). Z
el
is
dened in Fig. 9.
Z
m
Z'
em
oI
m
oV
m
oI '
m
Z
1
Z'
e1
oI
1
oV
1
oI'
1
Z
out
Z
o
Z'
eo
oI
o
oV
o
oI'
o
Z
1
Z'
e1
oI
1
oV
1
oI'
1
Z
2
Z'
e2
oI
2
oV
2
oI'
2
Augmented network [Y']
o
0
2o
p
o
0
o
p
o
0
+ o
p
o
0
+ mo
p
Diode
(excluding R
s
)
Figure 9. The small-signal representation of
the mixer as a multifrequency linear multi-
port network. Note that R
s
is included in
Z
0
Z
ek
R
s
. In a more exact model, R
s
is
frequency- and bias-dependent. In this text
we use R
s1
for the series resistance at the
input frequency (fundamental mixer) and R
s0
for the series resistance at the IF. Index 0
indicates parameters at the IF; o
0
is the IF.
The noise temperature as dened in Eq. (42) is the
single-sideband noise temperature for the lower sideband.
Knowing the diode parameters, we can analyze the
Schottky mixer with a high degree of accuracy using
commercial software.
4. DIODE MIXER TOPOLOGIES AND MIXER DESIGN
One may consider a number of different approaches in
designing mixers. Symmetry properties are one way of
dening basic types of mixers:
1. The single-ended mixer with one diode and a com-
mon input port for the signal and the LO
2. The single-balanced mixer with two diodes and
separate ports for signal and LO
3. The double-balanced mixer with four diodes and
separate ports for signal and LO
4. The double-double-balanced mixer with eight diodes
and separate ports for signal and LO
For millimeter- and submillimeter-wave applications,
types 1 and 2 have been implemented experimentally
and are described in the literature, while all four types
are common in microwave frequency applications.
The performance of either mixer type depends on the
impedances seen at the signal, at the IF, and at the
harmonic sidebands. The LO should experience a reason-
ably good match in order to reduce the LO power require-
ment. In practice, the impedances at the harmonics of the
LO and at the harmonic sidebands are very difcult to
control. The exception is the impedance at the image
frequency, which can often be controlled. The importance
is illustrated by referring to properties of a typical broad-
band mixer (when L
s
L
i
) used in a single-sideband
application. In such a mixer a considerable amount of
the signal power (E25%) may end up at the image
frequency.
It is also obvious that the noise entering the mixer at
the image port will be converted with the same efciency
as the signal at the signal port, adding to the system noise.
4.1. Single-Ended Mixer
A single-ended mixer has one input port, used for both the
signal and the LO. Hence it is necessary to incorporate a
circuit in front of the mixer itself for injecting the LO. This
circuit should not signicantly attenuate the signal. For
example, using a 10-dB directional coupler will attenuate
the signal B10% and add noise (see Fig. 10). A common
way of introducing the LO in microwave- and millimeter-
wave mixers is to use a narrow band diplexer, for example,
a lter structure in the input waveguide or a quasioptical
interferometer in the signal path in front of the mixer
(e.g., see Ref. 10).
4.2. Single-Balanced Mixers
In single-balanced mixers, the signal and the LO enter the
mixer through different ports, isolated from each other.
Either 901 or 1801 hybrids or baluns are used (Fig. 11a)
(see Ref. 1 for details). Figure 11b shows a low-frequency
equivalent circuit of a single-balanced mixer. The paths of
the signal current i
s
and the LO current i
LO
indicate that
they add in one diode and subtract in the other. This
causes an imbalance in A, which will slowly cycle at a
frequency equal to the IF. Hence the IF power can be
subtracted between A and ground. Note that if the LO is
noisy, this will not cause any output noise at the IF port.
The use of two diodes rather than one means that the
mixer can handle twice as much power for the same
intermodulation as for the single-ended mixer. In sum-
mary, for the single-balanced mixer:
*
The signal, LO, and IF ports are isolated from each
other.
*
The LO noise cancels at the IF port.
*
The power handling is superior to the single ended
mixer.
Mixer diode
IF out
IF filter
Matched
load
Local oscillator
Directional coupler
10 dB coupling
Signal in
Figure 10. Example for a single-ended mixer conguration. Note
that the directional coupler used for injecting the LO couples only
10% of the LO power to the mixer diode, and that the coupler
consequently attenuates the signal by a factor of 0.9.
(a)
IF
A
D1
Signal
D2
180 or 90
3 dB
hybrid
LO
D1
(b)
IF
Signal
D2
Low pass filter LO
i
s1
i
s2
i
LO
z/4
Figure 11. Basic design of the single-balanced mixer (a) and
equivalent circuit of the 1801 hybrid mixer, (b).
4.3. Double-Balanced Mixers
Essentially, a double-balanced mixer is constructed from
two single-balanced mixers, coupled in parallel and 1801
out of phase. The diodes can be arranged in either a star or
ring conguration (see Fig. 12). The ring can be arranged
very compactly as a monolithic circuit.
If the diodes are perfectly identical, the symmetry
ensures perfect isolation between the signal and the LO
ports. The topology also yields cancellation at the IF port
of the even harmonics of both the signal and the LO
frequencies. This also means that intermodulation is
reduced as compared to the mixers mentioned above.
Hence the advantages of the double-balanced mixer are:
*
Excellent isolation between the signal, LO, and IF
ports
*
LO noise cancellation at the IF port
*
Superior power handling compared to the double
balanced mixer
*
Superior intermodulation properties compared to the
double balanced mixer
4.4. Double-Double-Balanced Mixers
Double-double-balanced mixers are constructed using two
double-balanced mixers. Eight diodes are used, leading to
further power-handling capacity and still better intermo-
dulation properties.
4.5. Image Rejection and Image Enhancement
The system properties of a single-sideband mixer receiver
can be improved by introducing a proper circuit at the
image frequency. It is of particular importance to reac-
tively terminate the image frequency so that no signal
power is lost at the image frequency and no noise (or any
other unwanted signal) at the image frequency can be
converted to the IF frequency. Furthermore, if the reac-
tance at the image frequency is chosen properly, the
signal will be reected back into the mixer, such that
the conversion is enhanced and/or the noise properties are
improved. A stopband lter can be added in the input
transmission line to prevent one sideband from reaching
the diode. The distance to the diode is chosen to optimize
the mixer conversion toss. The conversion loss becomes
several decibels lower than the typical 5 dB for a common
broadband microwave mixer.
A most elegant method to realize a single sideband
mixer is shown in Fig. 13. By using two balanced mixers
and two 901 hybrids, it is possible to arrange that the
upper sideband and the lower sideband exit the mixer at
different ports (see Ref. 1 or 4 for details).
5. HARMONIC MIXERS
In a harmonic mixer a harmonic of the LO frequency,
no
LO
, is used for mixing; that is, the IF is obtained as
o
IF
jno
LO
o
s
j 43
Harmonic mixers are practical when it is difcult to
realize LO power at a frequency near the signal frequency.
They are particularly useful at millimeter and submilli-
meter waves.
Large harmonic numbers n are often used when max-
imum sensitivity is not required. For example, in spec-
trum analyzers large harmonic numbers may be used.
5.1. Two-Diode Subharmonically Pumped Mixers
If two diodes are used in an antiparallel conguration (see
Fig. 14), the small-signal conductance will vary with twice
the LO frequency. Hence the mixer will convert signals
located near 2f
LO
, and no conversion will occur near f
LO
.
The advantage of the two-diode over the single-diode
subharmonically pumped mixers is that no conversion
can occur at the fundamental frequency. Moreover, the
LO noise will contribute less, since the frequency differ-
ence between the signal and the LO is of the order f
LO
.
Another advantage is inherent self-protection against
large peak reverse voltage burnout.
6. PARAMETRIC FREQUENCY CONVERSION
In a parametric frequency converter, a nonlinear reac-
tance, such as a backward-biased Schottky diode, is
used. Common parametric components are frequency
downconverters, frequency upconverters, and frequency
Signal
IF
LO
Figure 12. A double-balanced mixer conguration: the ring
mixer.
T
Multi-path components
T
g
Sampling start
t
max
T
x
T
Figure 13. A single-sideband mixer using two balanced mixers
and two 901 hybrids. Note that at one output port the upper
sideband appears, while at the other output port the lower side-
band appears.
multipliers. In a frequency downconverter, a strong pump
signal f
LO
and signal f
s
(strong or weak depending on
application) are applied to the device. The output fre-
quency is
f
out
jf
LO
f
s
j frequency downconversion 44
For a frequency upconversion, we have
f
out
jf
LO
f
s
j frequency upconversion 45
The signal frequency in this case may be much lower than
the LO (or pump) frequency; that is, the output frequency
is not far from the LO frequency.
Note that if the device has no resistive parasitics, no
power is lost in the device itself, and 100% efciency is
theoretically possible. However, power may go to harmo-
nics or harmonic sidebands and there are always some
parasitic resistances present, for example, the series
resistance in a Schottky diode.
If we select a large ratio f
LO
/f
s
, the frequency upcon-
verter may have high gain. This is possible since, by
appropriate choice of the circuit parameters, parametric
amplication is achieved [7]. In the case of gain, one has
negative resistance in the circuit and one may face
stability problems. However, it is very difcult to make a
broadband parametric upconverter, since proper impe-
dances have to be realized at f
s
, f
LO
, and f
out
. Parametric
converters are used much less today than a few decades
ago. The reason is that the quality of mixer diodes and
FETs has improved signicantly and it is much easier to
make diode (or FET) mixers very broadband. This is the
reason why resistive mixers are preferred in most applica-
tions. Note that FET mixers can be designed for a conver-
sion gain greater than one.
A classical reference concerning varactor circuits is the
book by Peneld and Rafuse [7]. A parametric down-
converter, like the Schottky mixer, can be analyzed using
commercial software.
7. NEGATIVE-RESISTANCE DIODE MIXERS
In the currentvoltage (IV) characteristic of, say, the
Esaki tunnel diode or the resonant tunneling diode, there
is a region that has a differential negative resistance. This
means that the mixer can have conversion gain. Tunnel
diode mixers have been built and tested. However, a large
junction capacitance made the frequency range quite
limited, which, together with poor power handling, stabi-
lity problems, and less favorable noise properties, means
that these mixers have very little practical use today.
8. SELF-OSCILLATING MIXERS
The negative resistance devices can as well promote an
oscillation. Hence it is possible to design circuits where the
LO is delivered by the same device that is performing the
mixing. Besides the devices mentioned already, Gunn
diodes have also been used in self-oscillating mixers. The
sensitivity of such mixers is limited. The advantage may
be in applications where the best performance is less
important and the lowest price is required, as, for in-
stance, in low-priced Doppler radar applications.
9. BOLOMETER MIXERS
Bolometer mixers have been constructed since the 1950s.
Since the electromagnetic absorption in bolometer devices
can be essentially frequency-independent, it should be
possible to do mixing to several terahertz. In this type of
mixer, one is using the fact that when two signals at
slightly different frequencies are superimposed the result-
ing signal can be described as a signal that is amplitude-
modulated with the difference frequency. The rst useful
bolometer mixer was based on InSb devices cooled to
temperatures of a few kelvins. When the device absorbs
the modulated signal, the electron temperature becomes
modulated, leading to a modulation in the device resis-
tance. The theory is described in more detail in Section
10.2 and by Arams et al. [8].
However, the thermal time constant for the InSb device
is long, allowing a maximum IF of only about 2MHz. A
more recent bolometer mixer is based on a two-dimen-
sional electron gas in HEMT materials allowing an IF to
about 1 GHz [9]. However, the most successful hot-elec-
tron bolometer mixer so far is the superconducting hot
electron bolometer mixer.
10. MIXERS BASED ON SUPERCONDUCTING DEVICES
Room-temperature mixers for frequencies from about
100GHz to a few terahertz THz frequencies use only
Schottky diodes. However, if sensitivity is an issue (e.g.,
as in radioastronomy), there is a better alternative in
(b)
I
V
(c)
t
V
LO
(t)
g
m
(t)
(a)
LO in
IF out
Signal in
LO
filter
IF filter
Signal
filter
Figure 14. Subharmonically pumped mixer using antiparallel
diodes, showing (a) the mixer circuit, (b) the DC IV character-
istic, and (c) the resulting waveforms for the LO voltage and the
time-dependent small signal conductance.
mixers based on superconducting devices cooled to a few
kelvins. Low-noise superconductorinsulatorsupercon-
ductor (SIS) mixers have excellent performance up to about
1THz [10]. Superconductor hot-electron bolometer (HEB)
mixers are the best alternative for frequencies above 1THz.
Figure 15 shows a schematic of a receiver (except for
the input quasioptics) based on superconducting devices.
10.1. The SIS Mixer
The SIS mixer is also called the quasiparticle mixer.
Because of the extremely strong nonlinearity in the IV
characteristic of the SIS device, quantum effects are
important. Indeed, for certain choices of embedding im-
pedance network, this can result in a conversion gain (see
Ref. 11 for details).
In a superconductor and at a temperature below the
superconducting transition temperature, electrons form
pairs, called Cooper pairs. When they do so, the energy of
the electrons near the Fermi energy is lowered by a
certain amount D. Hence, to break up a Cooper pair, an
energy of 2D is required. This can be described in terms of
a bandgap with the energy 2D, as shown in Fig. 16.
In the SIS mixer, a tunneling phenomenon is used. In
the device, two superconducting lms are separated by a
thin (B20-A
) layer of insulator. Under bias, Cooper pairs

on one side of the isolator break up into two electrons
(quasiparticles) that individually tunnel through the iso-
lator and recombine on the other side of the insulator. This
is illustrated in Fig. 16. It is interesting to note that the
density of states near the band edges becomes innite.
This is one important reason why there is such a sharp
increase in the current when the device is biased to a
voltage V2D/e (e is the charge of the electron). The IV
characteristic is shown in Fig. 17. Note that the voltage
scale is in millivolts, and that 1 meV corresponds to
240 GHz. The steps in the IV curve correspond to DC
bias voltages, where exactly eV2Dhf (VE1.35 mV).
Since the IV characteristic is strongly nonlinear within
a fraction of a millivolt, the mixer is operating in the
quantum regime. See also Fig. 18, where the IV character
of an SIS device is compared with a Schottky diode. The
Schottky diode obviously is not very nonlinear within a
voltage interval of 1mVand is therefore operating fully as
a classical mixer for frequencies up to several terahertz. It
should also be mentioned mat the required LO power is
very low, on the order nanowatts. This is of great impor-
tance for sub-millimeter-wave mixers, where substantial
LO power is difcult to obtain.
The best SIS devices are realized in so-called Nb
trilayer technology [11]. The device structure is Nb/
Al
2
O
3
/Nb, where the E20-A
-thick Al
2
O
3
serves as the
insulator in the SIS device. For frequencies above
B700 GHz, one is trying to develop devices based on
NbN, which has a higher bandgap (E1.2THz) than does
Nb. So far these attempts have not been very successful.
The basic noise in the SIS mixer is shot noise. Compar-
ing the IV characteristics of the SIS device and Schottky
diode and using the classical theory described in Section 3,
one can see that the mixer noise and the conversion loss
are essentially lower for the SIS mixer. However, the
theory must include quantum effects (see Ref. 11 for
more details). This leads to a conversion gain that is
possibly larger than one, a fact that has been demon-
strated in practice.
IF out
Device
IF amplifier
Window
Signal
and
LO in
Spiral
antenna
Substrate
with
antenna
Liquid helium, = 4 K
Hyperhemispherical
lens
Figure 15. Schematic description of typical Dewar setup for a
sub-millimeter-wave SIS or HEB mixer. The LO and the signal
are entering together through the Dewar window. The radiation
is focused on the antenna using a hyperhemispherical lens,
downconverted to the IF in the nonlinear device, and nally
amplied by the IF amplier. In this gure the antenna is
illustrated as a spiral antenna. There are many other possible
planar antenna structures available (see Ref. 9 for details).
One photon assists
tunneling
2
eV = 2
eV = 2hf
LO
Figure 16. The SIS device under bias: (a) no bias; (b) for a voltage bias V42D/e electrons will
tunnel from right to the left; (c) tunneling is assisted by a photon with energy hf
LO
for the bias
voltage V(Dhf
LO
)/e.
10.2. Hot-Electron Bolometer Mixer
A bolometer consists of an absorber that is heated by
radiation and a temperature-dependent resistance as a
thermometer. The bolometer has a thermal time con-
stant t
0
limiting the maximum detectable modulation
frequency of the absorbed power. This means that the
maximum feasible IF is f
IF
1/(2pt
0
). When the LO and
signal are added together, the instantaneous power varia-
tion is described by [V
LO
cos(o
LO
t) v
s
cos(o
s
t)]
2
(1/R
RF
)
(see Fig. 3). If the bolometer can respond to o
IF
but not to
(o
LO
o
s
), 2o
LO
, and 2o
s
, the bolometer temperature and
the resistance will be approximately proportional to
P
LO
P
s
2

P
LO
P
s
p
coso
IF
t. Note that the slow re-
sponse of the bolometer device means that there are no
harmonics of the LO created and no signal power is
transformed to the image frequency, as is the case for
both the SIS and Schottky mixers.
A superconducting hot-electron bolometer (HEB)
consists of one or several superconducting thin-lm strips
in parallel, deposited on a substrate, for example, silicon,
single-crystalline quartz, or sapphire. The strips
are cooled to the superconducting state and then heated
by DC and microwave power to temperatures near
the superconducting to normal transition temperature,
where the superconductor will gradually become normal
(Fig. 19).
The maximum IF is determined by the electron tem-
perature relaxation time t
0
, that is, f
IF
o1/(2pt
0
), and a
major issue is to nd ways of making the time constant t
0
short enough. Figure 20 indicates how cooling occurs in
the so-called phonon-cooled and the diffusion-cooled bol-
ometer, respectively [10,12].
When operating the mixer, the device is absorbing LO
power (P
LO
) and signal power (P
s
) as well as power from
the DC bias supply (P
DC
V
0
I
0
). When the power in-
creases, obviously the electron temperature increases
and the resistance of the device increases as DR(dR
0
/
dP)DPC
0
DP. Figure 21 shows a simple equivalent cir-
cuit of the mixer, where the device is biased by a constant
DC current. Consequently, the modulation at the IF of the
resistance will cause an IF voltage to appear across the
device, causing an IF current through the IF load resis-
tance R
L
. The IF current DI is superimposed on the DC
bias current through the mixer device R
0
(P) and will cause
a modulation of the DC power DP
DC
cos(o
IF
t). The total
power dissipated in the device is then
Pt P
0
DPt
P
DC
DP
DC
coso
IF
t P
LO
P
s
2
P
LO
P
s
p
coso
IF
t
46
Assuming that DC and RF power affect the resistance by
the same amount, the IF modulation of the device resis-
tance becomes C
0
DP(t). There is a resulting bias point of
the device VV
0
and I I
0
. Dening the device DC
resistance R
0
as the time average of R
0
(P), that is, R
0
SIS
SIS
Schottky
at 20 K
Schottky
at 20 K
V
1 V
mV
I
2 4
50 jA
Figure 18. Comparing the IV characteristic of a SIS element
with that of a Schottky diode. Note the enormous difference in
nonlinearity.
Voltage
Operating bias point
With LO
No LO
C
u
r
r
e
n
t
R
N
R
0
I
c
I
0
V
0
Figure 19. Currentvoltage (IV) characteristic of a HEB with
and without a LO. At the operating point, V
0
/I
0
R
0
. For large
bias voltage the whole strip is normally conducting (RR
N
). I
c
is
the critical current, the maximum current in the completely
superconducting state.
mV
Current
Voltage
1 2 4 3
V = 2 /e
50 jA
Figure 17. Typical IV characteristic of a SIS element. The thin
line indicates the shape of the pumped IV characteristic, where
f
LO
hE1.4 meV (f
LO
325GHz), while the thick line is for the
unpumped device.
V
0
/I
0
, one obtains (8) the conversion gain
G
P
IF
P
s
2C
2
0
P
LO
P
DC
R
L
R
0
2
.
R
L
R
0
1 C
0
P
DC
R
0
.
R
L
R
0
R
L
R
0

2
47
where C
0
dR
0
/dP; R
L
is the IF load resistance and P
IF
,
P
s
, P
LO
, and P
DC
are the IF, signal, LO, and DC power,
respectively, dissipated in the device.
The commonly assumed fundamental limit of 6 dB
gain for hot-electron mixers is not valid if a negative
differential resistance of the unpumped IV curve is
available (see Ref. 13 for details). The load resistance for
maximum gain is equal to the differential resistance of the
IV curve at the bias point of the pumped mixer.
10.3. Experimental Results
In experiments on phonon-cooled NbN HEB mixers, noise
temperatures of about 400K (DSB) have been obtained at
600 GHz, and 1000 K (DSB) at 900GHz. At 2.5THz a noise
temperature of about 1400K has been obtained. The
conversion loss is typically 10 dB, including losses from
the optics in front of the mixer.
The noise of these mixers is caused by thermal uctua-
tions in the bolometer device (causing resistance uctua-
tions) and by Nyquist noise. For diffusion-cooled mixers a
noise temperature of 650 KDSB at 533GHz was measured
by Skalare et al. [14]. These experiments indicate that an
IF bandwidth of at least 3 GHz is achievable. Later
experiments show that it should be possible to obtain at
least 50% higher IF bandwidths in practical mixers. A
crucial number is the maximum IF bandwidth. For a
phonon-cooled HEB mixer, a 3.7 GHz IF bandwidth
(a)
Radiation
NbN or NbCN bolometer
ph
ph
Au
contact
pad
t = 30100
w = 11.5 jm
Si, SiO2 or sapphire substrate
L=0.11jm
(b)
Radiation
Nb or NbC
L < Dt
eph
L=0.070.2 jm
Au
contact
pad
t 100
Figure 20. Two types of bolometer devices:
(a) phonon-cooled; (b) diffusion-cooled.
dc Block
LP filter
P(t)
R
L R
0
(P(t)) I
I
0
L =
Figure 21. Equivalent circuit of bolometer with load.
( 3 dB reduction in conversion gain) has been measured
and diffusion-cooled between 2 and 6GHz. While this
bandwidth is dened for a conversion sloss increase of
3 dB, the bandwidth dened for when the noise tempera-
ture has increased by 3dB is about 1.5 times larger
(B5.5GHz for the phonon-cooled one).
The LO power needed is less than 100nW, which is
much lower than that needed for Schottky diode mixers.
BIBLIOGRAPHY
1. S. A. Maas, The RF and Microwave Circuit Design Cookbook,
Artech House, Norwood, MA, 1998.
2. E. L. Kollberg, Mixers and detectors, in K. Chang, ed., Hand-
book of Microwave and Optical Components, Wiley, New York,
1990.
3. K. S. Yngvesson, Microwave Semiconductor Devices, Kluwer,
Norwell, MA, 1991.
4. S. A. Maas, Microwave Mixers, 2nd ed., Artech House,
Norwood, MA, 1993.
5. S. A. Maas, Nonlinear Microwave Circuits, IEEE Press,
Piscataway, NJ, 1997.
6. D. N. Held, Conversion loss and noise of microwave and
millimeter-wave mixers, Part 1Theory, and Part 2Experi-
ment, IEEE Trans. Microwave Theory Tech. 26:4961 (1978).
7. P. Peneld and R. P. Rafuse, Varactor Applications, MIT
Press, Cambridge, MA, 1962.
8. F. Arams et al., Millimeter mixing and detection in bulk InSb,
Proc. IEEE 54:308318 (1966).
9. J.-X. Yang et al., Wide-bandwidth electron bolometric mixers:
A 2DEG prototype and potential for low-noise THz receivers,
IEEE Trans. Microwave Theory Tech. 41:581589 (1993).
10. R. Blundell and C. E. Tong, Submillimeter receivers for radio
astronomy, Proc. IEEE 80:17021720 (1992).
11. M. J. Wengler, Submillimeter-wave detection with supercon-
ducting tunnel diodes, Proc. IEEE 80:18101826 (1992).
12. J. E. Carlstrom and J. Zmuidzinas, Millimeter and submilli-
meter techniques, in W. R. Stone, ed., Reviews of Radio
Science 19931995, Oxford Univ. Press, London, 1996.
13. H. Ekstrom et al., Conversion gain and noise of niobium
superconducting hot-electron-mixers, IEEE Trans. Micro-
wave Theory Tech. 43:938947 (1995).
14. A. Skalare et al., A heterodyne receiver at 533GHz using a
diffusion cooled superconducting hot electron mixer, IEEE
Trans. Appl. Superconduct. 5:22362240 (1995).
FREQUENCY-DIVISION MULTIPLEXERS
RICHARD V. SNYDER
RS Microwave
Butler, New Jersery
1. INTRODUCTION
Frequency multiplexers are networks used to distribute
an input spectrum, either through separation into spectral
subcomponents or by combination of subcomponents. In
the rst case, a given input spectrum is separated by fre-
quency-sensitive elements into discrete (but possibly con-
tiguous) segments. In the second case, a multiplicity of
discrete (but possibly contiguous) input segments are sim-
ilarly combined into a composite output spectrum. In most
situations, it is required to maintain desirable passband
characteristics, such as match to source and load imped-
ances, amplitude atness, low ripple, and minimum group
delay distortion, for all ports of the multiplexer. Certainly
the separate spectral portions have to be isolated by main-
taining specied stopband slopes for each portion. Because
of interactions between the separate portions, these re-
quirements are very difcult to realize. However, there are
many techniques that have been developed for frequency
multiplexing. Coordination of direct synthesis and math-
ematical optimization provides the tools necessary to solve
the problem.
2. STATEMENT OF THE PROBLEM
It is required to separate a contiguous spectrum into spec-
ied portions. In most cases, it is necessary to avoid re-
ections from the ports associated with the specied
portions in order to preserve the total energy contained
in the initial contiguous spectrum. The associated problem
is combination of specied spectral segments into a con-
tiguous output spectrum, again with the low port reec-
tion criterion pertaining to the combining scheme. These
closely related problems are shown in a two-port congu-
ration in Fig. 1. Note that simply minimizing port reec-
tions is normally inadequate, because the intention of the
multiplexing schemes is availability of total spectral en-
ergy at the respective ports, rather than internal dissipa-
tion as an alternative to external reection. However,
examples of important exceptions to this condition will
be presented in later sections.
Consideration of Fig. 1 reveals a deceptively simple
conguration for the three-port multiplexer, known as a
diplexer. It consists of two selective lters and an inter-
connection network. The idealized response is shown in
Fig. 2. It is required to determine the criteria for the
selective lters and the interconnection network. The l-
ters can employ almost any of the available design ap-
proaches, including lumped element, crystal, dielectric
resonators, evanescent mode, waveguide, cavity, and
SAW. The interconnection network must be physically
compatible with the chosen lter approach, but can also
consist of a very wide variety of transmission lines, lumped
Common
port
(F1-F4)
Isolated
port
(F1-F2)
Isolated
port
(F5-F6)
Selective filter
(passes F1-F2)
Selective filter
(passes F5-F6)
Interconnection
network
Figure 1. Threeport multiplexer (a diplexer). The common
port can be an input or an output, while the isolated ports are,
respectively, outputs or inputs.
FREQUENCY-DIVISION MULTIPLEXERS 1649
Next Page

06d

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

06d

Transféré par

Droits d'auteur :

Formats disponibles

15. S. D.

Gedney, An anisotropic perfectly matched layer absorb-

Figure 11. Effect of metallization thickness on the dispersion

bertragung 31:4044 (1977).

bertragung 35:287292 (1981).

bertragung 37:375380 (1983).

Visible range at f/27

Figure 28. Array pattern of a binomial distribution and a Cantor

signal power available at the input

the effective mass of the electron. In practice, T

) layer of insulator. Under bias, Cooper pairs

Vous aimerez peut-être aussi