IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 51, NO.

1, JANUARY 2003 1

Advanced Coupling Matrix Synthesis Techniques

for Microwave Filters
Richard J. Cameron, Fellow, IEEE

AbstractA general method is presented for the synthesis of

the folded-configuration coupling matrix for Chebyshev or other
filtering functions of the most general kind, including the fully
canonical case, i.e., prescribed finite-position transmission
zeros in an th-degree network. The method is based on the
+2 transversal network coupling matrix, which is able
to accommodate multiple input/output couplings, as well as the
direct sourceload coupling needed for the fully canonical cases.
Firstly, the direct method for building up the coupling matrix for
the transversal network is described. A simple nonoptimization
process is then outlined for the conversion of the transversal
matrix to the equivalent +2 folded-configuration coupling
matrix. The folded matrix may be used directly to realize mi-
crowave bandpass filters in a variety of technologies, but some
of these could require awkward-to-realize cross-couplings. This
paper concludes with a description of two simple procedures
for transforming the transversal and folded matrices into two
novel network configurations, which enable the realization of ad-
vanced microwave bandpass filters without the need for complex
inter-resonator coupling elements.
(a)
Index TermsAsymmetric filtering functions, Chebyshev
characteristics, circuit synthesis methods, coupling matrix,
microwave filters, transversal network.

I. INTRODUCTION (b)
N
I N [1], a recursive method for deriving the transfer and
M k
Fig. 1. Canonical transversal array. (a) resonator transversal array
reflection polynomials for Chebyshev filtering functions including direct sourceload coupling . (b) Equivalent circuit of the th
low-pass resonator in the transversal array.
with prescribed finite-position transmission zeros (TZs) was
presented. This was followed by the synthesis methods for
The matrix has the following advantages, as compared
the corresponding coupling matrix, ready for the
realization of a microwave filter with resonators arranged as with the conventional coupling matrix.
a folded cross-coupled array. It was mentioned in [1] that, Multiple input/output couplings may be accommodated,
although the polynomial synthesis procedure was capable i.e., couplings may be made directly from the source
of generating TZs for an th-degree network (i.e., fully and/or to the load to internal resonators, in addition to the
canonical), that a maximum of only finite-position main input/output couplings to the first and last resonator
zeros could be realized by the coupling matrix. This in the filter circuit.
excluded some useful filtering characteristics, including those Fully canonical filtering functions (i.e., th-degree char-
that require multiple input/output couplings, which have been acteristics with finite-position TZs) may be synthe-
finding applications recently [3]. sized.
In this paper, a method is presented for the synthesis of the During certain synthesis procedures that employ a
fully-canonical or folded coupling matrix, which over- sequence of similarity transforms (rotations), it is some-
comes some of the shortcomings of the conventional times convenient to temporarily park couplings in the
coupling matrix. The or extended coupling matrix has outer rows or columns, whilst other rotations are carried
an extra pair of rows top and bottom and an extra pair of columns out elsewhere in the matrix.
left and right surrounding the core coupling matrix, The paper begins by detailing the procedure for synthesizing
which carry the input and output couplings from the source and the coupling matrix from the transversal array circuit rep-
load terminations to resonator nodes in the core matrix. resentation of the filtering function (see Figs. 1(a) and 2), which
follows on from the methods originally established in [4][7]
and later extended in [1]. The new method is actually simpler to
Manuscript received November 2, 2001.
The author is with COM Dev International Ltd., Aylesbury HP22-5SX, U.K. derive than those used to synthesize the coupling matrix,
Digital Object Identifier 10.1109/TMTT.2002.806937 not requiring the GramSchmidt orthonormalization stage. The
0018-9480/03$17.00 2003 IEEE
2 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 51, NO. 1, JANUARY 2003

where , is the
prescribed return loss in decibels, and it is assumed that the poly-
nomials , , and have been normalized to their
respective highest degree coefficients. Both and are
th-degree polynomials, is the degree of the filtering func-
tion, whilst , which contains the finite-position prescribed
TZs, is of degree , where is the number of finite-posi-
tion TZs that have been prescribed. For a realizable network,
must be .
is unity for all cases except for fully canonical filtering
functions, where all the TZs are prescribed at finite frequencies,
i.e., . In this case, the value of (in decibels) is
finite at infinite frequency, and if the highest degree coefficient
of the polynomials , , and are each normalized to
unity, will have a value slightly greater than unity as follows:

(2)

N +2 [M ]
N 2N
Fig. 2. fully canonical coupling matrix for the transversal array. It is also important to ensure that the transfer and reflection vec-
M =M
The core matrix is indicated within the double lines. The matrix is
symmetric about the principal diagonal, i.e., . tors are orthogonal in order to satisfy the unitary conditions for
the scattering matrix [8]
reduction of the transversal coupling matrix to the folded
cross-coupled array coupling matrix is then outlined, following
much the same procedure as in [1]. A demonstration of the use
of the techniques to synthesize the coupling matrix for a fully (3)
canonical filtering function is included.
From (3), it may be shown (see [2, p. 177]) that the phases
Finally, the direct synthesis of two novel filter configurations
, , and of the vectors , , and , respec-
are presented; one starting with the transversal coupling matrix
tively, are related by the following:
and the second based on the folded coupling matrix. Both are ap-
plicable to the design of microwave bandpass filters in a variety
(4)
of technologies, but the second, in particular, has some impor-
tant implementation advantages that should considerably ease
where is an integer.
the design and production of high-performance filters for space
Equation (4) shows that the difference between the phase
or terrestrial communications systems.
of the vector, and the average of the phases of the and
vectors must be an odd multiple of rad. For this con-
II. SYNTHESIS OF THE TRANSVERSAL dition to be satisfied at any value of the frequency variable ,
COUPLING MATRIX the TZs of must be positioned symmetrically about
The approach that will be employed to synthesize the the imaginary ( ) axis or upon the imaginary axis itself. Sim-
transversal coupling matrix will be to construct the two-port ilarly, the pattern of the zeros of must either be co-
short-circuit admittance parameter matrix for the overall incident with those of on the imaginary axis, or form
network in two ways; the first from the coefficients of the ra- mirror-image pairs about the imaginary axis with corresponding
tional polynomials of the transfer and reflection scattering pa- off-axis zeros of . In this way, the sum of the phases of the
rameters and , which represent the characteristics individual vectors that make up the overall phases of the vectors
of the filter to be realized, and the second from the circuit el- , , etc., will be multiples of rad.
ements of the transversal array network. By equating the Since , , and share a common denom-
matrices as derived by these two methods, the elements of the inator polynomial , it is only necessary to consider their
coupling matrix associated with the transversal array network numerator polynomials as far as (4) is concerned. The multiples
may be related to the coefficients of the and poly- of rad referred to above therefore depend upon the number
nomials. of finite-position transmit (Tx) zeros for the numer-
ator polynomial , and the degree of the filtering func-
A. Synthesis of Admittance Function From the Transfer tion for the and numerator polynomials (
and Reflection Polynomials and , respectively). With this in mind, it follows that, for
The transfer and reflection polynomials that are generated in the left-hand side of (4) to produce an odd multiple of rad,
[1] for the general Chebyshev filtering function are in the form the integer quantity must itself be odd. Thus, to ensure
orthogonality between the and vectors, i.e., is
an odd multiple of rad, it is necessary to multiply the
(1)
polynomial by whenever is an even integer.
CAMERON: ADVANCED COUPLING MATRIX SYNTHESIS TECHNIQUES FOR MICROWAVE FILTERS 3

The numerator and denominator polynomials for the The process for building up [see (5)] results in its highest
and elements of may be built up directly from the degree coefficient having a value of and, since the
transfer and reflection polynomials for and [1]. highest degree coefficient of , the value of may be
For a double-terminated network with source and load termina- found as follows:
tions of 1
(8)

The new numerator polynomial may now be determined

and as follows:
for even (9)
which will be of degree , and the residues of
may now be found as normal.
and

for odd B. Synthesis of Admittance Function Circuit Approach

The two-port short-circuit admittance parameter matrix
where for the overall network may also be synthesized directly
from the fully canonical transversal network, the general form
of which is shown in Fig. 1(a). It comprises a series of
individual first-degree low-pass sections, connected in parallel
(5) between the source and load terminations, but not to each other.
The direct sourceload coupling inverter is included to
and and , are the complex coef- allow fully canonical transfer functions to be realized, ac-
ficients of and , respectively. The and cording to the minimum path rule, i.e., , the maximum
polynomials for single-terminated networks may be number of finite position TZs that may be realized by the
found by a similar procedure [1]. network , where is the number of resonators
Knowing the denominator and numerator polynomials in the shortest route through the network between the source
for and , their residues and , and load terminations. In fully canonical networks
may be found with partial fraction ex- and, thus, , the degree of the network.
pansions, and the purely real eigenvalues of the network Each of the low-pass sections comprises one parallel-con-
found by rooting the denominator polynomial common nected capacitor and one frequency invariant susceptance ,
to both and , which has purely imaginary roots connected through admittance inverters of characteristic admit-
(see [1, Appendix]). Expressing the residues in matrix tances and to the source and load terminations, respec-
form yields the following equation for the admittance matrix tively. The circuit of the th low-pass section is shown in Fig. 1(b).
for the overall network:
Fully Canonical Filtering Functions
The direct sourceload inverter in Fig. 1(a) is zero ex-
cept for fully canonical filtering functions, where the number of
finite-position zeros equals the degree of the filter. At infinite
frequency ( ), all the capacitors become parallel
short circuits, which appear as open circuits at the sourceload
ports through the inverters and . Thus, the only path
(6)
between source and load is via the frequency-invariant admit-
tance inverter .
where the real constant , except for the fully canonical If the load impedance is 1 , the driving point admittance
case where the number of finite-position TZs in the filtering looking in at the input port will be (Fig. 3)
function is equal to the filter degree . In this case, the degree
of the numerator of is equal to
its denominator , and has to be extracted from Therefore, the input reflection coefficient at is
first to reduce the degree of its numerator polynomial
by one before its residues may be found. Note that, (10)
in the fully canonical case, where the integer quantity
is even, it is necessary to multiply by to ensure that the Substituting for in the conservation of energy equation
unitary conditions for the scattering matrix are satisfied. using (10)
Being independent of , may be evaluated at as
follows:

(7)
4 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 51, NO. 1, JANUARY 2003

trix for the direct sourceload coupling inverter

Fig. 3. Equivalent circuit of transversal array at s = 6j1.

Solving for

(14)

At infinite frequency C. Synthesis of the Transversal Matrix

because, for a fully canonical filtering function, and
will both be th-degree polynomials with their Now the two expressions for , the first in terms of the
highest degree coefficients normalized to unity. Similarly, residues of the transfer function (6), and the second in terms
. Therefore, of the circuit elements of the transversal array (14), may be
equated. It may be seen immediately that , and for
the 21 and 22 elements in the matrices in the right-hand
sides of (6) and (14)

Since is slightly greater than unity for a fully canonical net- (15a)
work, choosing the negative sign will give a relatively small
value for
(15b)

(11)
The residues and and the eigenvalues have already
been derived from the and polynomials of the desired
and correctly gives for noncanonical filters, where filtering function [see (5)] and, thus, by equating the real and
. It can be shown that the positive sign will give a second imaginary parts in (15a) and (15b), it becomes possible to relate
solution , but since this will be a large number, them directly to the circuit parameters
it is never used in practice [8].
and
Synthesis of Two-Port Admittance Matrix
and
Cascading the elements in Fig. 1(b) gives an transfer
matrix for the th low-pass resonator as follows:

and
(12)
(16)

which may then be directly converted into the equivalent short- It may be recognized at this stage that and constitute
circuit -parameter matrix the unscaled row vectors and of the orthogonal matrix
, as defined in [1, Appendix].
Since the capacitors of the parallel networks are all
unity, and the frequency-invariant susceptances ( ,
representing the self couplings ), the input
couplings , the output couplings , and the direct
sourceload coupling are all now known, the reciprocal
transversal coupling matrix representing the
network in Fig. 1(a) may now be constructed. ( ) are
the input couplings and occupy the first row and column of
(13)
the matrix from positions 1 to (see Fig. 2). Similarly,
( ) are the output couplings and they occupy the last
The two-port short-circuit admittance matrix for the par- row and column of from positions 1 to . All other entries
allel-connected transverse array is the sum of the -parameter are zero. and are equivalent to the terminating
matrices for the individual sections, plus the -parameter ma- impedances and , respectively, in [1].
CAMERON: ADVANCED COUPLING MATRIX SYNTHESIS TECHNIQUES FOR MICROWAVE FILTERS 5

and these are shown in Table I. Being fully canonical,

and may be found from (2). Note that, because
and is, therefore, an even number, the coefficients of have
been multiplied by in Table I.
Now the numerator and denominator polynomials of
and may be
constructed using (5). The coefficients of , , and
(a) , normalized to the highest degree coefficient of ,
are summarized in Table II.
The next step is to find the residues of and
with partial fraction expansions. Since the numerator of
is one less in degree than its denominator ,
finding the associated residues is straightforward. How-
(b)
ever, the degree of the numerator of is the
Fig. 4. Folded canonical network coupling matrix formfifth-degree same as its denominator , and the factor has
example. (a) Folded coupling matrix form. s and xa couplings are zero for
symmetric characteristics. (b) Coupling and routing schematic. to be extracted first to reduce in degree by one.
This is easily accomplished by first finding by evalu-
ating at , i.e., equals the ratio of the highest
Reduction of the Transversal Matrix to the Folded degree coefficients in the numerator and denominator polyno-
Canonical Form mials of [see (7) and (8)] as follows:
With input and output couplings, the transversal topology
is clearly impractical to realize for most cases and must be trans-
formed to a more suitable topology if it is to be of practical use.
A more convenient form is the folded or reflex configuration
which may be seen is the highest degree coefficient of
[9], which may be realized directly or used as the starting point
in Table II. Alternatively, may be derived from (11).
for further transformations to other topologies more suitable for
may now be extracted from the numerator of [see
the technology it is intended to use for the construction of the
(9)] as follows:
filter.
To reduce the transversal matrix to the folded form, the formal
procedure, as described in [1], may be applied, working on the
matrix instead of the coupling matrix. This proce- At this stage, will be one degree less than and the
dure involves applying a series of similarity transforms (rota- residues may be found as normal. The residues, the eigen-
tions), which eliminate unwanted coupling matrix entries alter- values [where are the roots of ], and the associated
nately right to left along rows and top to bottom down columns, eigenvectors and are listed in Table III.
starting with the outermost rows and columns and working in- Note that, for double-terminated lossless networks with equal
wards toward the center of the matrix, until the only remaining source and load terminations, will be positive real for a
couplings are those that can be realized by filter resonators in a realizable network, and .
folded structure (Fig. 4) Now knowing the values of the eigenvalues , the eigenvec-
As with the matrix, no special action needs to be taken tors and , and , the transversal coupling
to eliminate unneeded and couplings in the cross-di- matrix (Fig. 2) may be completed as shown in Fig. 5.
agonalsthey will automatically become zero if they are not Using the same reduction process as described in [1], but op-
required to realize the particular filter characteristic under con- erating upon the matrix, the transversal matrix may be
sideration. reduced to the folded form with a series of six rotations, anni-
hilating the elements , , , , , and finally
Illustrative Example in order (see Table IV). The resulting folded configuration
To illustrate the matrix synthesis procedure, an example coupling matrix is shown in Fig. 6(a), and its corresponding cou-
is taken of a fully canonical fourth-degree asymmetric filtering pling and routing schematic is shown in Fig. 6(b).
function with 22-dB return loss and four TZs, two at The analysis of this coupling matrix is shown in Fig. 7. It
and , which produce two attenuation lobes of 30 dB may be seen that the return loss and rejection characteristics are
each on the lower side of the passband, and two at unchanged from those obtained from the analysis of the original
and , producing a lobe of 20 dB on the upper side. and polynomials.
Applying the recursive technique of [1, Sec. II] yields the
coefficients for the numerator and denominator polynomials of III. TRANSFORMATIONS OF THE COUPLING MATRIX
and A microwave filter may be realized directly from the folded
coupling matrix, the topology and strengths of its inter-resonator
couplings directly corresponding to the nonzero elements of the
coupling matrix. However, it is sometimes necessary to apply
6 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 51, NO. 1, JANUARY 2003

TABLE I
44 FILTERING FUNCTIONCOEFFICIENTS OF E (s), F (s) AND P (s) POLYNOMIALS

TABLE II
44 FILTERING FUNCTIONCOEFFICIENTS OF NUMERATOR AND DENOMINATOR POLYNOMIALS OF y (s) AND y (s)

TABLE III
44 FILTERING FUNCTIONRESIDUES, EIGENVALUES, AND EIGENVECTORS

(a)
Fig. 5. Transversal coupling matrix for 44 fully canonical filtering function.
The matrix is symmetric about the principal diagonal.

TABLE IV
FOURTH-DEGREE EXAMPLEPIVOTS AND ANGLES OF THE SIMILARITY
TRANSFORM SEQUENCE FOR THE REDUCTION OF THE TRANSVERSAL
MATRIX TO THE FOLDED CONFIGURATION. TOTAL NUMBER OF
TRANSFORMS R = n=6 (b)
Fig. 6. Fully canonical synthesis example. Folded coupling matrix for 44
filtering function. (a) Coupling matrix. Matrix is symmetric about the principal
diagonal. (b) Coupling and routing schematic.

Here, two novel realizations are introduced; parallel-con-

nected two-port networks and the cul-de-sac configuration.
The first may be derived by grouping residues and forming
separate two-port subnetworks, which are then connected in
a further series of rotations to the matrix, to transform it into parallel between the source and load terminations. The second
a form more convenient or more practical to the application in is formed by a series of similarity transforms operating upon
hand, e.g., [10][12]. the folded coupling matrix.
CAMERON: ADVANCED COUPLING MATRIX SYNTHESIS TECHNIQUES FOR MICROWAVE FILTERS 7

Grouping residues and yields the folded matrix for

the second-degree subnetwork shown in Fig. 8. Now grouping
residues and yields the folded coupling matrix for
the fourth-degree subnetwork shown in Fig. 9.
Superimposing the two matrices yields the overall matrix
shown in Fig. 10.
The results of analyzing the overall coupling matrix are
shown in Fig. 11(a) (rejection/return loss) and Fig. 11(b) (group
delay), which show that the 25-dB lobe level and equalized
in-band group delay have been preserved.
Other solutions for this topology are available depending on
the combinations of residues that are chosen for the subnet-
works. However, whatever combination is chosen, at least one
Fig. 7. 44 fully canonical synthesis example: analysis of folded coupling
!6 1=
matrix. Rejection as s j 20 log (") = 30:407 dB.
of the input/output couplings will be negative. Of course, the
number of topology options increases as the degree of the fil-
tering function increases, for example, a tenth-degree filter may
A. Parallel-Connected Two-Port Networks be realized as two parallel-connected two-port networks, one
Being closely related to short-circuit admittance parameters, fourth degree and one sixth degree, or as three networks, one
the eigenvalues and corresponding residues of the filtering func- second degree and two fourth degree, all connected in parallel
tion may be separated into groups and subnetworks constructed between the source and load terminations. Also, each subnet-
from them using the same procedures as described above. The work itself may reconfigured to other two-port topologies with
subnetworks may then be connected in parallel between the further transformations, if feasible.
source and load terminations to recover the original filtering If the network is to be synthesized as parallel-coupled
characteristics. The transverse array itself may be regarded as pairs (see Fig. 12 for a sixth-degree example), a rather more di-
a parallel connection of single-resonator groups. rect synthesis route exists. Starting with the transversal matrix, it
Although the choice of residue groupings is arbitrary, it is only necessary to apply a series of rotations to annihilate half
will be found that difficult-to-realize couplings will be created the couplings in the top row from positions back to the mid-
within the subnetworks, and between the internal nodes of the point of this row , i.e., rotations (see Fig. 2). Due
subnetworks and the sourceload terminations if the choice to the symmetry of the values in the outer rows and columns of the
of filtering function and of residue groupings is not restricted. transversal matrix, the corresponding entries to in
The restrictions are: 1) filtering functions may be fully canon- the last column will be annihilated simultaneously.
ical, but must be symmetric and even degree and 2) residue The pivots of the rotations to annihilate these couplings start
groups must consist of complementary pairs of residues and at position [ ] and progress toward the center of the matrix
eigenvalues, i.e., if the residues with indexes and ( , until position [ , ]. For the sixth-degree example,
and , ) constitute a group or are part of a group, this is a sequence of rotations according to Table VI
then and . This implies that only and applied to the transversal matrix:
networks double-terminated between equal-value source and After the series of rotations, the matrix, as shown in
load terminations can be synthesized. Fig. 12(a), is obtained, which corresponds to the coupling and
If these restrictions are observed, the overall network will routing diagram in Fig. 12(b). In every case, at least one of the
consist of a number of two-port networks, the number corre- input/output couplings will be negative. An interesting example
sponding to the number of groups that the residues have been of a fourth-degree implementation of this topology realized in
divided into, each connected in parallel between the source and dielectric resonator technology is given in [13].
load terminals. If the filtering function is fully canonical, the di-
rect sourceload coupling will also be present. B. Cul-de-Sac Configurations
Once the residues have been divided into groups, the syn- The "cul-de-sac" configuration [14] is restricted to double-
thesis of the sub-matrices and their reduction to the folded form terminated networks and will realize a maximum of TZs.
follows exactly the same process as for a single network, as de- Otherwise it will accommodate even- or odd-degree symmetric
scribed in Section II, working on each subnetwork individually. or asymmetric prototypes. It has an important advantage over
To illustrate the process, an example is taken of a 23-dB return other configurations in that, whatever the prototype filtering
loss sixth-degree characteristic, with two symmetrically placed function, there will be only one negative coupling in the en-
TZs at 1.3958 producing lobes of 25 dB on either side of tire network and there will be no diagonal cross-couplings,
the passband, and a pair of real-axis zeros at 1.0749 to give which are sometimes awkward to realize in practice. Moreover,
group-delay equalization over approximately 50% of the pass- its form lends itself to a certain amount of flexibility in the phys-
band. This filter will be synthesized as two subnetworks, one of ical layout of its resonators.
degree 2 and one of degree 4. A typical "cul-de-sac" configuration is shown in Fig. 13(a) for
Following the procedure of Section II results in a set of a tenth-degree prototype with the maximum-allowable seven Tx
residues and eigenvalues for the characteristic as shown in zeros (in this case, three imaginary-axis and two complex pairs).
Table V. There is a central core of a quartet of resonators in a square
8 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 51, NO. 1, JANUARY 2003

TABLE V
6-2-2 SYMMETRIC FILTERING FUNCTIONRESIDUES, EIGENVALUES, AND EIGENVECTORS

(a)

(b)
Fig. 8. Coupling sub-matrix and coupling/routing diagram for residues k =1
and 6. (a) Coupling matrix. (b) Coupling and routing diagram.

(a)

(a)

(b)
Fig. 9. Coupling sub-matrix and coupling/routing diagram for residue group
k = 2; 3; 4; and 5. (a) Coupling matrix. (b) Coupling and routing diagram.
(b)
Fig. 11. Analysis of parallel-connected two-port coupling matrix.
(a) Rejection and return loss. (b) Group delay.

(a)

(a)

(b)
Fig. 10. Superimposed second- and fourth-degree sub-matrices. (a) Coupling
matrix. (b) Coupling and routing diagram.

(b)
formation [1, 2, 9, and 10 in Fig. 13(a)], straight-coupled to each Fig. 12. Symmetric 6-4 filter examplerealized as parallel-coupled pairs.
other (i.e., no diagonal cross-couplings). One of these couplings (a) Coupling matrix. (b) Coupling and routing diagram.
CAMERON: ADVANCED COUPLING MATRIX SYNTHESIS TECHNIQUES FOR MICROWAVE FILTERS 9

TABLE VI TABLE VII

SIXTH-DEGREE EXAMPLESIMILARITY TRANSFORM SEQUENCE FOR THE PIVOT COORDINATES FOR THE REDUCTION OF THE N +2
FOLDED MATRIX
REDUCTION OF THE TRANSVERSAL MATRIX TO THE PARALLEL-COUPLED TO THE CUL-DE-SAC CONFIGURATION
PAIRS FORMAT

(a)

(b)

(a)

(c)
Fig. 13. Cul-de-sac network configurations. (a) 10-3-4 network. (b) 8-3
network. (c) 7-1-2 network.

is always negative; the choice of which one is arbitrary. The (b)

entry to and exit from the core quartet are from opposite corners
Fig. 14. Cul-de-sac configurationseventh-degree example. (a) Original
of the square [1 and 10, respectively, in Fig. 13(a)]. folded coupling matrix. (b) After transformation to cul-de-sac configuration.
Some or all of the rest of the resonators are strung out in cas-
cade from the other two corners of the core quartet in equal num- where are the coordinates of the pivot and also of the ele-
bers (even-degree prototypes) or one more than the other (odd- ment to be annihilated, is the angle of the similarity trans-
degree prototypes). The last resonator in each of the two chains form, and is an arbitrary integer. Note that, for cross-pivot an-
has no output coupling, hence, the nomenclature cul-de-sac nihilations of ( 0), where the self-couplings ,
for this configuration. Other possible configurations are shown . It is also allowable to have for when
in Fig. 13(b) (eighth degree) and Fig. 13(c) (seventh degree). , which will give a slightly different configuration alter-
native. For odd-degree filters, the angle formula takes the more
C. Synthesis of the Cul-de-Sac Network conventional form
Fortunately, the synthesis of the "cul-de-sac" network is very (18)
simple and is entirely automatic. Starting with the folded cou-
Table VII gives the pivot coordinates and angle formula to
pling matrix, elements are annihilated using a series of regular
be used for the sequence of similarity transforms to be applied
similarity transforms (for odd-degree filters), and cross-pivot
transforms (for even-degree filters), beginning with a main line to the folded coupling matrix for degrees 49, and a general
coupling near the center of the matrix, and working outwards formula for the pivot coordinates for any degree 4.
An example is made of the double-terminated version of the
along or parallel to the antidiagonal. This gives a maximum of
transforms for even-degree prototypes and seventh-degree prototype that was used in [1]. This character-
for odd-degree prototypes. istic had 23-dB return loss, a TZ at 1.2576 to give a rejection
The cross-pivot similarity transform for even-degree filters lobe level of 30 dB on the upper side of the passband, and a com-
is one where the coordinates of the element to be eliminated are plex pair of Tx zeros at to give group-delay
the same as the pivot of the transform, i.e., the element to be equalization over approximately 60% of the passband.
annihilated lies on the cross-points of the pivot. The angle for After following the procedure of Section II, the folded
the annihilation of an element at the cross-point is different to matrix shown in Fig. 14(a) is obtained. Applying a series of
two similarity transforms at pivots [3, 5] and [2, 6] (Table VII)
that of a regular annihilation and is given by
with angles according to (18) results in the coupling matrix of
Fig. 14(b). The corresponding coupling and routing diagram is
(17)
given in Fig. 13(c).
10 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 51, NO. 1, JANUARY 2003

to employ. Two examples of such reconfigurations are included

in the paper: the parallel-coupled two-port network configura-
tion and the cul-de-sac filter configuration. The latter fea-
tures some important constructional simplifications that should
ease the volume production process for high-performance mi-
crowave filters for the wireless industry.

ACKNOWLEDGMENT
The author is grateful to Dr. J. D. Rhodes, Filtronics plc,
Saltaire, U.K., and Dr. C. Ernst, Lorch Microwave Inc., Salis-
bury, MD, for useful discussions that aided considerably in the
development of the theory presented in this paper.
(a)
REFERENCES
[1] R. J. Cameron, General coupling matrix synthesis methods for Cheby-
shev filtering functions, IEEE Trans. Microwave Theory Tech., vol. 47,
pp. 433442, Apr. 1999.
[2] R. E. Collin, Foundations for Microwave Engineering. New York: Mc-
Graw-Hill, 1966.
[3] I. C. Hunter, J. D. Rhodes, and V. Dassonville, Dual-mode filters with
conductor-loaded dielectric resonators, IEEE Trans. Microwave Theory
Tech., vol. 47, pp. 23042311, Dec. 1999.
[4] A. E. Atia and A. E. Williams, New types of bandpass filters for satellite
transponders, COMSAT Tech. Rev., vol. 1, pp. 2143, Fall 1971.
[5] , Narrow-bandpass waveguide filters, IEEE Trans. Microwave
Theory Tech., vol. MTT-20, pp. 258265, Apr. 1972.
[6] A. E. Atia, A. E. Williams, and R. W. Newcomb, Narrow-band mul-
tiple-coupled cavity synthesis, IEEE Trans. Circuits Syst., vol. CAS-21,
pp. 649655, Sept. 1974.
[7] M. H. Chen, Singly terminated pseudo-elliptic function filter,
(b) COMSAT Tech. Rev., vol. 7, pp. 527541, Fall 1977.
Fig. 15. Seventh-degree "cul-de-sac" synthesis exampleanalysis of folded [8] S. Amari, Direct synthesis of folded symmetric resonator filters with
coupling matrix. (a) Rejection and return loss. (b) Group delay. sourceload coupling, IEEE Microwave Wireless Comp. Lett., vol. 11,
pp. 264266, June 2001.
[9] J. D. Rhodes, A low-pass prototype network for microwave linear
The results of analyzing this coupling matrix are presented in phase filters, IEEE Trans. Microwave Theory Tech., vol. MTT-18, pp.
Fig. 15, confirming that the rejection lobe level and group-delay 290300, June 1970.
equalization performances have been preserved intact. [10] H. C. Bell, Canonical asymmetric coupled-resonator filters, IEEE
Trans. Microwave Theory Tech., vol. MTT-30, pp. 13351340, Sept.
As was noted above, all the couplings are positive, except for 1982.
one in the core quartet. This may be moved to any one of the [11] R. J. Cameron and J. D. Rhodes, Asymmetric realizations for
dual-mode bandpass filters, IEEE Trans. Microwave Theory Tech.,
four couplings for the greatest convenience and implemented as vol. MTT-29, pp. 5158, Jan. 1981.
a probe, for example, if the filter is to be realized in coaxial-res- [12] R. J. Cameron, A novel realization for microwave bandpass filters,
onator technology where the other couplings are inductive irises ESA J., vol. 3, pp. 281287, 1979.
[13] V. Pommier, D. Cros, P. Guillon, A. Carlier, and E. Rogeaux,
or inductive loops. Also, there are no diagonal couplings even Transversal filter using whispering gallery quarter-cut resonators,
though the original prototype was asymmetric. If it is feasible in IEEE MTT-S Int. Microwave Symp. Dig., Boston, MA, 2000, pp.
to implement a diagonal coupling between the input and output 17791782.
[14] A. E. Williams, J. I. Upshur, and M. M. Rahman, Asymmetric response
of the core quartet, then an extra TZ may be realized, bringing bandpass filter having resonators with minimum couplings, U.S. Patent
the maximum number realizable by this topology to . This 6 337 610, Jan. 8, 2002.
coupling in the "cul-de-sac" core will have the same value as in
the folded coupling matrix.
Richard J. Cameron (M83SM94F02) was
born in Glasgow, U.K., in 1947. He received the
IV. CONCLUSIONS B.Sc. degree in telecommunications and electronic
engineering from Loughborough University, Lough-
In this paper, a simple and general method for the synthesis of borough, U.K., in 1969.
the coupling matrix in the folded cross-coupled array In 1969, he joined Marconi Space and Defence
configuration has been presented. The coupling matrix is Systems, Stanmore, U.K. His activities there
included small earth-station design, telecommuni-
applicable to symmetric or asymmetric, single- or double-termi- cation satellite system analysis, and computer-aided
nated, and even- or odd-degree filtering functions, and will ac- RF circuit and component design. In 1975, he joined
commodate the fully canonical and multiple-input/output cou- the European Space Agencys technical establish-
ment (ESTEC, The Netherlands), where he was involved in the research
pling configurations. and development of advanced microwave active and passive components
The folded coupling matrix may be used directly for and circuits, with applications in telecommunications, scientific, and earth
the design of a microwave filter if it is convenient to do so, or observation spacecraft. Since joining COM DEV International Ltd., Aylesbury,
used as the starting point for the application of a further series of U.K., in 1984, he has been involved in the software and methods for the design
of a wide range of high-performance components and subsystems for both
similarity transforms to reconfigure it into a topology more con- space and terrestrial application.
venient for the technology or production process it is intended Mr. Cameron is a Fellow of the Institution of Electrical Engineers (IEE), U.K.