Efficient Design of Chebychev Band-Pass Filters with

Ansoft HFSS and Serenade

Tutorial

Dr. B. Mayer and Dr. M.H. Vogel

Ansoft Corporation

Contents

Abstract

1. Introduction

2. Circuit Representation of the Filter

3. Relationships between Circuit Components and Physical

Dimensions in the Microwave Filter
10
4. Initial Filter Design in HFSS

14

5. Curve Fitting in Serenade

16

6. Corrected Filter Design in HFSS

20

7. Additional Information from the 3D Field Solver

21

7a. Effects of Internal Losses

21

7b. Maximum power-handling capability

23

7c. Mechanical Tolerances

24

References

25

Appendix ADerivation of the Circuit

Appendix B

The Physical Meanings of K and Q

26
43

Abstract
An efficient method is presented to design coaxial Chebychev band-pass filters.
The method involves a 3D full-wave field solver, Ansoft HFSS, teaming up with a circuit
simulator, Serenade. The authors show how for a practical case, a 7-pole band pass filter
with a ripple of only 0.1 dB, an accurate design is obtained in a matter of days, as
opposed to weeks for traditional methods.
The method described is also applicable to even more challenging designs of
elliptic filters and phase equalizers realized in dielectric, waveguide or coaxial
technology.

1 Introduction
In this paper, we will describe an efficient method to design a filter. The method
involves a 3D full-wave field solver teaming up with a circuit simulator. The basic idea
has been explored by others [1] but a different circuit was used in the circuit simulator.
We will explain our procedure by presenting in detail how we design a Chebychev band
pass filter with the following specifications:
Center frequency
400 MHz
Ripple bandwidth
15 MHz
Ripple
0.1 dB
Out-of-band rejection
24 dB at 390 MHz and at 410 MHz
In order to achieve the out-of-band rejection, we will need seven poles.
The desired filter characteristic is shown in Fig. 1.

Fig. 1

Desired filter characteristic

As the basic geometry for this filter we have chosen a cavity with seven coaxial
resonators, as shown in Fig. 2. In the figure, the buckets have been drawn as wire
frames for clarity, to show that the cylinders dont extend all the way to the bottom.

Fig. 2

Basic filter geometry

This geometry is symmetrical with respect to the central cylinder. In this kind of
filter, the walls of the cavity, the long cylinders, the buckets under the cylinders and the
disk-shaped objects near the first and last cylinder are all made of metal. The long
cylinders are connected to the top of the cavity; the buckets are connected to the bottom
of the cavity. Cylinders and buckets dont touch. The disk-shaped objects near the first
and last resonator are connected to the input and output transmission lines and provide
the necessary coupling to the source and the load. We will call these objects antennas in
this document. They are near the first and last cylinders, but never touch them. Each
cylinder-bucket combination is a resonating structure. At this stage, without restricting
ourselves, we can choose many dimensions in the filter relatively freely. We make the
following choices:
Cavity dimensions
280 x 30 x 120 mm
Resonator diameter
10 mm
Buckets inner diameter
12 mm
Buckets outer diameter
16 mm
Buckets height
15 mm
Antennas diameter
26 mm
Antennas thickness
6 mm

Six dimensions remain, and these six will be crucial in obtaining the desired filter
characteristic:
The length of the first and last resonating cylinder (both have equal length)
The length of the five interior cylinders (all five have equal length)
The distance between an antenna and its nearest cylinder
Three distances between neighboring cylinders (remember the filter is symmetric)
With traditional filter design methods, obtaining the correct dimensions is a timeconsuming task that commonly takes several weeks. Filter design with a circuit simulator,
on the other hand, is relatively straightforward. Filter theory provides the values for the
lumped inductors and capacitors that are needed to obtain the desired filter characteristic.
First, we will show how to design a circuit that not only has the desired filter
characteristic, but also lends itself to implementation with microwave components. In
such a circuit, we use series L and C for each resonator, i.e. the cylinder-and-bucket
combinations, and impedance invertors to represent the distances between adjacent
resonators. Second, we will show how one can determine relationships between
components in the circuit and dimensions in the physical filter. Third, we will present an
iterative procedure between the electromagnetic field solver and the circuit simulator to
optimize the design. The procedure converges very quickly.

2 Circuit Representation of the Filter

In order to design an order-seven band-pass filter around 400 MHz with a 0.1 dB
ripple, filter theory tells us to start with an order-seven low-pass filter, normalized to 1
radian/s. The normalized filter is to have a 0.1 dB ripple, like the desired band pass filter.
The source and load impedances of the normalized low pass filter are normalized to 1
Ohm. This circuit is shown in Fig. 3 and its characteristic in Fig. 4. Filter theory provides
us with the values for the inductors and the capacitors, denoted by g1 through g7 in the
figure. These values are in our case
g1=g7=1.1812 H
g3=g5=2.0967 H
g2=g6=1.4228 F
g4=1.5734 F.

Fig. 3

Normalized low-pass filter circuit, starting point for design procedure

Fig. 4

Filter characteristic for the normalized low-pass filter in Fig. 3

The step-by-step procedure from this normalized low-pass filter circuit to the final
band-pass filter circuit is presented in detail in Appendix A. Here, we show an outline of
the major steps.
An important step is the replacement of shunt capacitors by series inductors and
impedance inverters. Basically, an impedance inverter transforms impedances in the same
way as a quarter-wave-length transmission line, but independent of frequency. The
resulting circuit is shown in Fig. 5. This is still a normalized low-pass filter with the same
characteristic as the circuit in Fig. 3. The reason for this change is that at microwave
frequencies it is often impossible to realize the ladder circuit consisting of series
inductors and shunt capacitors. Depending on the basic structure either series elements or
shunt elements are easily realizable but often not both in the same structure. Taking
advantage of impedance inverters, it is possible to transform shunt capacitors into series
inductors. In the physical filter these impedance inverters will be realized by couplings
between the coaxial resonators.

Fig. 5

Normalized low-pass filter without shunt capacitors

Following a standard procedure, we take the following steps to derive the desired
band-pass filter model:
(1) De-normalize the low-pass cut-off angular frequency from 1 rad/s to bw rad/s.
(2) Transform the low-pass filter to a band-pass filter with a relative bandwidth of
bw and a center angular frequency of 1 rad/s by inserting a 1 F capacitor in
series with every 1 H inductor.
(3) De-normalize the center frequency to 400 MHz by choosing

L=1/(24E8) H and C = 1/(24E8) F.

(4) De-normalize the port impedances from 1 Ohm to the usual 50 Ohm by
introducing impedance inverters at the input and output with coupling coefficients
of 50.
(5) Introduce finite quality factors to the individual resonators by adding a series
resistor to each resonator.
(6) Introduce individual resonant frequencies to the first and last resonators to be
able to be able to take the frequency shift due to the coupling antennas into
account.
(7) Add a homogeneous transmission line of length ZUL between filter
input/output and port 1 / port 2 to be able to adjust the phase due to the
connectors.
This gives us the filter shown in Fig. 6. The procedure outlined above is presented in
more detail in Appendix A.

Fig. 6

Final filter circuit, representing the desired band pass filter

In this circuit, every LC pair resonates at 400 MHz. Further K12, K23, K34 and QL
have been defined as
bw
Kij =
gigj
(1)
and
g1
Q =
bw
(2)
where bw is the relative bandwidth and gi is the ith g value from filter theory.
Notice that, since the g values are known from filter theory, we still know the
values of the all the components in the circuit, even through the components have
changed considerably in the process.
Filter theory [2] tells us that Ki,i+1 and QL have important physical meanings. Ki,i+1
is known as the coupling constant between adjacent resonators. If we have just two
resonators in the cavity, with a very light coupling to the source and the load, then the
relation between coupling constant K12 and resonant frequencies f1 and f2 is given by
K12 = 2(f2-f1) / (f2+f1) .

(3)

QL is known as the loaded Q of the circuit. If we have just one resonator in the cavity,
coupled to source and load, the relation between QL , resonant frequency fR and 3-dB
band width BW3dB is given by
QL = fR / BW3dB

(4)

In the next section, we will link the components of this circuit to dimensions in
the physical geometry of the filter.

3 Relationships Between Circuit Components and

Physical Dimensions in the Microwave Filter
As explained in the previous section, every LC pair resonates at 400 MHz. In the
microwave filter, we must choose the length of each resonator such that it resonates at
400 MHz. That will determine the length of each of them.
Further, Ki,i+1 (i=1,2,3) are the coupling coefficients between adjacent resonators.
Therefore, these three coefficients are related to the distances between adjacent
resonators.
Finally, QL is the loaded Q of the circuit. Therefore, in an otherwise lossless
circuit, it is directly related to the distance between the first or final resonator and the
antenna that couples it to the source or the load.

3a Relation Between Resonator Spacing and Coupling

Coefficient
The model in Ansoft HFSS that was used to determine the coupling coefficient K
as a function of resonator spacing is shown in Fig. 7. Two resonators have been placed in
a closed metal cavity. This cavity has the same height and the same front-to-back depth as
the cavity to be used in the real filter. The left-to-right width has been chosen large
enough to make its influence on the results negligible. There are no transmission lines nor
ports for signal input and output, since the resonances of this structure are to be
determined through an eigenmode simulation. In the Setup Materials menu, resonators
are modeled as perfect conductors; the cavity is filled with air. Further, symmetry has
been exploited through the use of a Perfect-H boundary condition. As can be seen in the
figure, this cuts both the resonators and the cavity in half.

Fig. 7

Model used to determine the coupling coefficient K

By embedding this HFSS project in Optimetrics, dimensions can be varied easily.

First, the length of the cylinders was adjusted such that the resonances are centered at 400
MHz. Then, the distance between the resonators was varied, and for each distance the
eigenmode solver in HFSS computed the two eigen frequencies and obtained K. In order
to get very accurate results, more accurate than necessary, twelve adaptive passes were
run in each simulation, resulting in models with 63,000 tetrahedra. Total run time for each

point was 38 minutes on a 1.2 GHz PC. The simulation of the lossless structure required
557 MB of RAM. The relation between the resonator spacing and K is shown graphically
in Fig. 8.

Fig. 8

Relation between resonator spacing and coupling coefficient K

With this graph, for any coupling coefficient required by filter theory, the spacing
to be applied between resonators in the physical model can be readily determined.

3b Relation Between Antenna Distance and Loaded Q

The model in Ansoft HFSS that was used to determine the loaded Q as a function
of antenna spacing is shown in Fig. 9. An antenna-resonator combination has been placed
in a closed metal cavity. The 50- transmission line is present, but in order to perform an
eigenmode analysis, it has been terminated by a Perfectly Matched Layer (PML) of
absorbing material. This was done by replacing the final 20 mm of dielectric in the
coaxial cable by PML material. A macro, named pmlmatsetup, in the Materials-Setup
menu supplies the material parameters. This construction will give use the same resonant
frequency and loaded Q as the corresponding structure with a real 50- load would. The
cavity has the same height and the same front-to-back depth as the cavity to be used in
the real filter. Again, the left-to-right width has been chosen large enough to make its
influence on the results negligible, and symmetry has been exploited through the use of a
Perfect-H boundary condition.

Fig. 9

Model used to determine the loaded Q

This HFSS project has been embedded in Optimetrics. The antenna distance and
the cylinder length were varied simultaneously, since both influence the resonant
frequency and the loaded Q. As an example of the results, the relation between antenna
spacing and loaded Q is shown graphically for a constant cylinder length of 113.4 mm. In
order to get very accurate results, maybe a little more accurate than strictly necessary,
fifteen adaptive passes were run for each point. This results in simulations with 50,000
tetrahedra, requiring 830 MB of RAM. Total run time per point was 50 minutes on a 1.2
GHz PC.

Fig. 10

Relation between antenna spacing and loaded Q at a resonator length of

113.4 mm

With results like these, for any loaded Q and resonant frequency required by filter
theory, the antenna spacing and cylinder length to be applied in the physical model can be
readily determined.

4 Initial Filter Design in HFSS

Now that we have the circuit and we know the relations between circuit
components and physical dimensions, we can construct the filter in the field solver,
Ansoft HFSS. Filter theory tells us we need to achieve the following parameters:
Resonant frequency of the outermost resonators
fR1= 400 MHz
Resonant frequency of the inner resonators
fR2=400 MHz
Loaded Q
Coupling coefficients K12=0.02893 , K23=0.02171 , K34=0.02065
The calibration projects above tell us that the dimensions of the filter, as shown in Fig. 2,
are to be
Length of the two outermost resonators
= 113.399 mm
Length of the five inner resonators
= 114.69 mm
Antenna distance
= 1.879 mm
Distances between resonators are 25.513 mm , 28.291 mm , 28.767 mm
This filter has been modeled and simulated in Ansoft HFSS. The model is shown
in Fig. 11.

Fig. 11

Initial design in HFSS

Notice that only half the geometry is actually simulated. Symmetry has been
exploited through the use of a Perfect-H boundary condition. Further, all materials and
boundaries in the model are lossless for now. This requires less RAM and less CPU time.
The resulting filter characteristic is shown in Fig. 12. Notice that the center frequency and
the ripple bandwidth are almost perfect. We see the correct number of ripples, but the
ripple is 0.3 dB rather than 0.1 dB.

Fig. 12

S21 results for the initial design in HFSS

5 Curve Fitting in Serenade

The HFSS results, shown in the previous section, have been exported to Serenade,
the circuit simulator. This was done through Post Process / Matrix Data / File / Export /
Touchstone. In Serenade, we can determine through curve fitting what the actual
parameters of this initial design are. This curve fitting is done through the Serenade setup
shown in Fig. 13. All the variables are defined in the top level schematic. Circuit model
and HFSS results are defined via the sub circuits denoted as MODEL and MEASU,
respectively. To utilize all the available information for the optimization process, the
optimization goal is to end up in a complex S-Matrix identical for the model and the Smatrix resulting from the HFSS simulation. This is defined in Serenade via an OPT block
and the goal definition S=MEASU in the sub circuit defining the circuit model as shown
in Fig. 14. To match the phases of the S parameters of the Serenade and HFSS
simulations, homogeneous transmission lines of length ZUL are attached to ports 1 and 2
in the Serenade model. Optimization is done by starting with the optimum filter
parameters given in Fig. 6 and performing 1500 iterations with the random optimizer. The
solution was found without any manual interaction.
It took 35 minutes on a PC with a clock speed of 400 MHz. Figs. 15 through 17
show curve fitting results. Notice, in Fig. 15, that there is still a few hundredths of a dB
difference between the HFSS results and the best fit in Serenade. This indicates that this
design method is accurate to a few hundredths of a dB.

Fig. 13

Fig. 14

Serenade setup used for curve fitting: top level schematic

Serenade setup used for curve fitting: Model definition as sub circuit

Fig. 15

Result of curve fitting, magnitude

Fig. 16

Result of curve fitting, phase

Fig. 17

Results of curve fitting, complex S11 Blue and green lines are S_11 and
S_22 from HFSS, which have the same magnitude but slightly different
phases; red line is the best fit.

The result of the curve fitting procedure is as follows: we have built a filter with
Resonant frequency of the outermost resonators
fR1= 400.058 MHz
Resonant frequency of the inner resonators
fR2=399.926 MHz
Loaded Q
QL=30.368
Coupling coefficients K12=0.02825 , K23=0.02173 , K34=0.02068
Notice that the largest discrepancies occur in K12 and QL. Apparently, the
calibration project that determines the coupling coefficient by simulating two identical
resonators is not quite representative of the two outermost pairs of resonators, where one
resonator is coupled through an antenna to the source or the load. Also, the calibration
project that determines the loaded Q by simulating one resonator-antenna combination is
not perfectly representative of the real situation where this resonator is coupled to a
neighboring one.
Nevertheless, the calibration projects tell us how much correction is needed to
achieve the desired characteristic. For example, noticing that QL is too low by a certain
amount, we will aim for a QL that is higher by this amount the second time. Caution is
needed when adjusting the antenna distance, since that also changes fR1. We have to
change antenna distance and resonator length simultaneously, and aim for the correct QL
and fR1.

Keeping this in mind, with the aid of the calibration projects we find that the
dimensions of the filter are to be
Length of the two outermost resonators
= 113.44 mm
Length of the five inner resonators
= 114.684 mm
Antenna distance
= 1.928 mm
Distances between resonators are 25.286 mm , 28.3 mm , 28.78 mm
Hence, the dimensions that undergo the largest changes are the antenna distance and the
distance between the first and second resonator.

6 Corrected Filter Design in HFSS

The corrected filter was modeled and simulated in Ansoft HFSS. The resulting
characteristic and the corresponding Smith chart are shown in Figs. 18a and 18b. Note
that the ripple, which was 0.3 dB in the initial design, is better than 0.13 dB now. The
target is 0.1 dB.

Fig. 18a

S21 results in HFSS for the corrected design

Fig. 18b

Smith chart in HFSS for the corrected design

In order to obtain this result, the mesh was refined adaptively until it had 180,000
tetrahedra. With a mesh that size, the calculation of each frequency point required 1.28
GB of RAM and 9.5 minutes real time on a 1.2 GHz PC with one processor. Seventeen
frequency points were needed for an interpolating frequency sweep, bringing the total
time needed for the sweep to two hours and forty minutes. An identical model with only
119,000 tetrahedra (see below) provided results within a few hundredths of a dB in the
pass band and saved almost half the time.

7 Additional Information from the 3D Field Solver

7a

Effects of Internal Losses

All simulations thus far have been performed with lossless filters. A simulation
without loss results in computations with real numbers only, as opposed to computations
with complex numbers. This reduces the RAM requirement and the CPU time
significantly. Once the design has been finalized, however, one can easily change the
material parameters and boundary conditions to go from perfectly-conducting metals to
lossy metals like copper or silver. The software enables you to select materials from a
database or specify the conductivity. A plot comparing a lossless and a silver filter is
shown in Fig. 19. Notice that, due to the seven consecutive resonances, even with a very
good conductor like silver the insertion loss will be between 0.5 and 1 dB.

Fig. 19

Comparison lossless filter and silver-plated filter

Also note that the center frequency of the silver filter is slightly lower than the
center frequency of the perfect filter. A careful inspection of the data shows that this shift
is between 0.07 and 0.08 MHz. A model with just one resonator shows the same shift.
Further investigation reveals that this shift is due to the imaginary part of the surface
impedance of the silver. According to electromagnetic theory, the conductivity of the
silver translates into an equivalent surface impedance, provided that the metal thickness
is much larger than the skin depth. This surface impedance has a real and an imaginary
part, which are both equal to (f0R/), where f is the frequency, 0R is the
permeability of the material, and is the conductivity of the material. In the case of silver
at 400 MHz the surface impedance is Zsurface = 5(1+j) m/square. A simulation in HFSS
with Zsurface = 5 m/square shows no frequency shift at all relative to the perfect
conductor case, while a simulation with Zsurface = 5j m/square shows the same shift as in
Fig. 19.

Curve fitting with Serenade shows that replacing perfect conductors by silver in
HFSS is equivalent to introducing an unloaded Q of 2,800 in each resonator in Serenade.
According to filter theory, the introduction of an unloaded Q shifts the resonant frequency
downward by fr/(2Q), which in this case equals 0.07 MHz. Hence, HFSS has predicted
this frequency shift very accurately.
In order to account for this shift, designers should first determine the magnitude
of the shift with an HFSS simulation involving just one resonator. Then, they should
design a perfectly lossless filter around a frequency that is higher by this amount. The
center frequency of the filter with internal losses will thus come out just right.
The computer requirements were as follows. These computations have been
performed with a model with 119,000 tetrahedra. In the lossless case, this took 810 MB
RAM, 207 MB disk, and 86 minutes real time on a PC with a clock speed of 1.2 GHz and
one processor. In the lossy case, it took 1,332 MB RAM, 1,600 MB disk and 396 minutes
real time. The large time difference is due to the change from a real to a complex solver
and to the time needed for disk access. The disk access in this case is spill logic, which
is a deliberate process, performed under the softwares control. It is not to be confused
with the very inefficient swapping which is done by the operating system when a
process is too large for the available RAM.

7b

Maximum Power-Handling Capability

It is important to know how much power the filter can handle. The maximum
power handling capability can be obtained easily with the help of a field plot. Fig. 20
shows a close up of the fields around a resonator in the region where they are strongest.

Fig. 20

Fields around a resonator

The HFSS 3D Fields Post Processor tells us that, with 1 W input power, the
electric field strength between the cylinder and the bucket is 105 kV/m. You can change
the input power in the post processor (Data/Edit Sources). The filter would cease to
operate when the fields are strong enough to cause arcing in the air. This phenomenon
occurs at 3 MV/m, although, with a wide safety margin, 1 MV/m is commonly used as
the maximum acceptable field strength. Therefore, the fields can be allowed to be 9.5
times as strong as they are now, which implies that the maximum power handling
capability is 9.59.5 = 90 W.

7c

Mechanical Tolerances

Once the dimensions are known that provide a filter with the desired
specifications, it is important to establish mechanical tolerances. With the HFSS model
fully parameterized in Optimetrics, it is an easy task to explore the effects of small
dimensional changes on the filter characteristic. An example is shown in Fig. 21. There,

the distance between two resonators was made 0.08 mm larger and 0.08 mm smaller. The
original characteristic and the two modified ones are shown. In this case, a manufacturing
inaccuracy of 0.08 mm in the distance between two resonators results in a change of up to
0.05 dB in the filter characteristic. This way, mechanical tolerances, depending on the
accuracy requirements of the filter characteristic, can be specified.

Fig. 21

Example of the effects of manufacturing tolerances

References
[1] Daniel G. Swanson and Robert J. Wenzel, Fast Analysis and Optimization of
Combline Filters Using FEM, presented at the IEEE MTT Society 2001 International
Microwave Symposium, May 2001.
[2] Randall W. Rhea, HF Filter Design and Computer Simulation
McGraw-Hill, Inc., 1995
ISBN 0-07-052055-0

Appendix A

Derivation of the Circuit

In this appendix, the well known low-pass prototype method for filter designs is
repeated. Only the necessary facts for the present example are given. For more details a
standard book on filter designs should be used [A1].
The starting point of this method is the low-pass prototype as shown in Figs. A1
and A2. These filters are normalized to a cut-off angular frequency of 1 rad/s and a
generator impedance of 1 Ohm. In the case of a Chebychev filter some care has to be
taken regarding the order of the filter. For an odd number of elements the load impedance
is also 1 Ohm. For an even number of elements, however, the load impedance depends on
the order of the filter and the ripple. Therefore, the two cases are treated separately.

g0 = 1

Fig. A1

g1

Fig. A2

g N-1

g3

gN

g N+1 = 1

Low-pass prototype for the case N=odd

g0 = 1

g2

g2

g1

gN

g3

Low-pass prototype for the case N=even

g N-1

g N+1 = 1

In the past, the prototype values g i (i = 1 N+1) were read from tables, but
nowadays it is more convenient to use a filter design program. Chebychev filters are
defined by the filter order N and the in-band ripple. Closed-form expressions exist for the
g-values. An example is given in Fig. A3. This MathCAD program is valid for even as
well as odd order filters for any ripple value. Essentially the g-values are defined by a
recursive relation. Only for the last value a special treatment for the even and odd order
case is necessary. This is considered by an ifstatement with the modfunction as
condition. Exact definitions of these functions are given in the MathCAD handbook [A3].
The response of this filter is shown in Fig. A4.
N 6

ripple 0.1
ripple
10

eps 10
sin

2 ( i 1) 1

K( i) 4

1
sin

2 N

asinh 1
eps

sinh
N

G( i) if i

sin
1 2

1 G( i)

g0nplus1( i) if ( i
g ( i) if [ ( i

sinh

G( i)

gi( i 1)

( i 1)

N

2 N
K( i)
1
1

asinh

N
eps

gnplus1 if mod ( N 2)

0 g0 gnplus1)

N 1) ( i

sin

gi( i) if i

2 N

2 ( i 1) 1

g0 1

eps

1 eps

0) g0nplus1( i) gi( i) ]

j 0 N 1
g ( j)

j
1

1.1681111

1.4039709

2.0562117

1.5170948

1.9028879

0.8618448

0.7378106

Fig. A3

MathCAD program to derive the g-values for Chebychev filter responses.

Example: N = 6, ripple = 0.1 dB.

Fig. A4

Chebychev filter response, N = 6, ripple = 0.1 dB

At higher frequencies it is often impractical to realize both series elements and

shunt elements in one circuit. To avoid this, a new element, the impedance inverter, is
introduced. In the following a procedure is given to transform each shunt capacitor in Fig.
A2 to a series inductor with impedance inverters on each side. In the practical band-pass
circuit, these elements are realized by coupling structures between the individual
resonators.

ZL

K
Z IN =
ZL
Fig. A5

Impedance transformation with an impedance inverter

An impedance inverter is an element with a behaviour like a quarter-wave

transmission line but independent of frequency. Fig. A5 shows the symbol of an
impedance inverter. An impedance at the output of the circuit appears as the inverted
impedance at the input, scaled with the square of the impedance inverter constant K. In
addition the phase of a signal travelling through the circuit is shifted by 90. Both the
impedance change and the phase shift are frequency independent.
In Serenade such a circuit is realized by a transmission line with adaptive
electrical length. Fig. A6 shows the schematic. As the frequency f is a global variable it is
an easy task to create the impedance inverter with the electrical length defined by the
appropriate formula.

Fig. A6

Realization of impedance inverter in Serenade

Prior to the transformation of the shunt capacitors to series inductors some basic
relationships of impedance inverter circuits are derived. Starting with the chain circuit of
a first impedance inverter, an inductance and a second identical impedance inverter, an
equivalent impedance is derived. This is demonstrated in Fig. A7. Effectively, the
equivalent input impedance seen by the generator is a parallel circuit of a capacitor and
the original load impedance as shown in Fig. A8. To derive an identical circuit, a phase
shift of 180 is introduced. This phase shift may be omitted in the case of Chebychev
filters as these types of filters are pure chain circuits. However, in more advanced circuit
like elliptic filters employing cross couplings this could be of importance. The procedure
shown here proves only the equivalence of the input impedances rather than the
equivalence of the entire circuits. However, the latter could be proven easily as well by
comparing the chain matrices.

ZG

L
ZL

ZG

L
2

K
ZL

ZG

K
jL +
ZL
K

1
2

K
jL +
ZL

j
C

Fig. A7

First identity, part 1

L
K

1
ZL

ZG
K

1
2

K
jL +
ZL

L
K

1
ZL

C
ZG

C=

ZL

Fig. A8

First identity, part 2

ZG
K

1
2

K
jL +
ZL

1
L x
j
2+
ZL
(K x )

C
ZG

Fig. A9

Lx
ZL
K x

K x

Second identity

By multiplying the series inductance by a factor of x and simultaneously

multiplying the impedance inverter constant K by x, the input impedance remains
constant. This leads to the second identity shown in Fig. A9. Again, the proof is only
based on an identical input impedance seen by the source. The complete proof could be
derived easily by comparing the chain matrixes. With these identities it is a
straightforward procedure to replace the shunt capacitors by series inductors in the
prototype low-pass filters. This is illustrated in Fig. A10 for the case N = odd. Both
circuits are identical.
Next, the cut-off frequency is de-normalized by dividing the inductance values by
the relative bandwidth bw. According to the identity jL = j( bw)L/bw, the impedance
values shift from frequency to frequency bw when we replace the inductance L by
inductance L/bw as shown in Fig. A11. Fig. A12 shows the results for the cases bw = 1
and bw = 0.5.

At this point, the inductors still have different values. In order to arrive at a circuit
with identical elements, the inductance values are set to one and the impedance inverter
constants are adjusted according to the second identity illustrated in Fig. A13.

g0 = 1

g2

g1

g0 = 1

Fig. A10

g N-1

g3

g1

gN

g2

g N+1 = 1

gN

Low-pass prototype with series elements only; N = odd

g0 = 1

g 1 / bw

g2 / bw

gN / bw

Fig. A11

Low-pass prototype with de-normalized cut-off frequency; cut-off

frequency bw.

Fig. A12

Chebychev low-pass filter response: N = 7, ripple = 0.1 dB,

Cut-off frequency bw = 0.5 (blue traces) and bw = 1 (red traces)

g 1 / bw

Fig. A13

bw

bw / g1

1
g1 g2

1
gi gj

g0 = 1

gN / bw

bw / gN

g2 / bw

bw

g0 = 1

Low-pass prototype with identical reactive elements; N = odd.

In the following step each inductance is replaced by a series resonator consisting

of an inductance of 1 H and a capacitance of 1 F to design a band pass filter around the
angular frequency of 1 rad/s and a relative bandwidth bw. Fig. A14 shows the circuit and
Fig. A15 shows the filter responses before and after this change.

Fig. A14

bw / gN

1
g i gj

g 1 g2

bw

bw / g1

bw

g0 = 1

Normalized band-pass filter with relative bandwidth bw and center angular

frequency of 1 rad/s

Fig. A15

Chebychev low-pass filter response, red and blue traces:

N=7, ripple=0.1 dB, cut-off angular frequency bw=0.25 rad/s and
Chebychev band-pass filter response, pink and green traces:
N=7, ripple=0.1 dB, center frequency 1 rad/s, relative bandwidth bw=0.25.

To arrive at the final filter model the center frequency is de-normalized to get a
center frequency of fr. Starting by the well know formula for the resonant frequency
f r

1
2 L C

the inductor values and capacitor values must be chosen properly. When scaling the
inductor L1 = 1 H and the capacitor C1 = 1 F of the normalized band-pass filter by the
same factor s, the relative bandwidth stays constant. This is because the normalized
impedance of the individual resonators,
Z1() = j (L1 1/C1),
and the impedance of the scaled resonators
Zs () = j( (L1/s) 1 / (C1/s) )
have the relation

Z1 () = Zs (s),
i.e. the complex resonator impedance values of Z1 at the angular frequency are
identical to the complex impedance values of the scaled complex resonator impedance at
the frequency s. The resonant frequency is s times larger while the relative bandwidth is
not altered. Therefore, we choose the following values for the scaled inductance Ls and
the scaled capacitance Cs
Ls = 1 / (2 fr) Henry,
and
Cs = 1 / (2 fr) Farad.
Additionally, the port impedance is de-normalized to the usual 50 Ohms by
impedance inverters at the input and the output. To be able to compare HFSS Sparameters and model parameters, feed lines are also introduced. For the impedance
converter constants the usual notations Kij = bw / (gi * gj)0.5 for the inner ones and Q1 =
g1 / bw for the outer ones are introduced.
One last step is necessary: the definition of a finite unloaded quality factor Q0 of
the individual resonators. This is taken into consideration by an appropriate series
resistor. The value of this resistor is Rs = 2 fr Ls / Q0 [A2]. In our circuit, since Ls = 1 /
(2 fr),
Rs = 1 Ohm / Q0 .
For filters, where the losses are caused by a metallic casing, there exists also a frequency
shift due to the imaginary part of the surface impedance. If the normal skin effect is
applicable, real part and imaginary part of the surface impedance are identical. Therefore,
in this case, it is possible to express this frequency shift by the unloaded quality factor Q0.
According to electromagnetic field theory the relative frequency shift f is
f = -1 / 2Q0 .
Fig. A16 shows the final Serenade filter model. Fig. A17 shows the corresponding filter
response for the lossless case and Fig. A18 shows the filter response in the case when
metallic losses are present and the normal skin effect is applicable. This model is not only
useful to asses insertion loss and frequency shift, but could also be used to gain an equal
ripple filter in the case of internal losses by optimizing the K and Q values properly.

Fig. A16

Final Serenade filter model, N = 7, ripple = 0.1 dB,

center frequency fr = 400 MHz, relative bandwidth bw = 0.0375

Fig. A17

Chebychev band-pass filter response, N = 7, ripple = 0.1 dB,

center frequency fr = 400 MHz, bandwidth 15 MHz

Fig. A18

Chebychev band-pass filter response, N = 7, ripple = 0.1 dB,

center frequency fr = 400 MHz, bandwidth 15 MHz . No internal losses
(blue trace) and Q0=2700 (red trace).

Notice that, in the circuit in Fig. A16, we have introduced quantities K and Q in
the impedance inverters. K and Q are defined by
bw
Kij =
gigj
and
Q =

g1
bw

where bw is the relative bandwidth and gi is the ith g value from filter theory. In Appendix
B we will show that K has the physical meaning of coupling between two resonators and
Q has the physical meaning of loaded Q of a circuit with one resonator. These physical
meanings are important parts of the overall design strategy, as they are used in the HFSS
calibration projects.
The derivation of the final Chebychev band pass filter of even order out of the
appropriate low-pass prototype is done by the same procedure. Special treatment is only
necessary at the output. Adding an impedance inverter at the output with the impedance
inverter constant of [g (N+1) ] 1/ 2 realizes a constant load impedance of one. Fig. A19
explains this special step. After this step the identical procedure as in the case of oddorder filters is used to arrive at the final Serenade filter model. After carrying out all steps
it turns out that the filters of even order are also symmetric although the original g-values
show no symmetry at all. The circuit is shown is Fig. A20 and the simulation results for
N = 6 and a ripple of 0.1 dB are given in Fig. A21.

g N-2

g1

g0 = 1

Fig. A19

gN-1

g1

g N-1

g N-2

g N+1

Z L= 1

gN

Low-pass prototype with series elements only; N = even

g N+1

g0 = 1

Fig. A20

Final Serenade filter model, N = 6, ripple = 0.1 dB, center

frequency fr = 400 MHz, relative bandwidth bw = 0.0375 (15 MHz)

Fig. A21

Chebychev band-pass filter response, N = 6, ripple = 0.1 dB,

center frequency fr = 400 MHz, bandwidth 15 MHz

References Appendix A
[A1]
G. Matthaei, L. Young, E.M.T. Jones
Microwave Filters, Impedance Matching Networks, and Coupling Structures
ARTECH HOUSE, INC. 1980, ISBN 0-89006-099-1
[A2]
D.M. Pozar
Microwave Engineering, second edition 1998
John Wiley & Sons, Inc., ISBN 0-471-17096-8
[A3]
MathCAD, Users Guide, MathCAD 2000 Professional
MathSoft, Inc.
101 Main Street
Cambridge Massachusetts 02142
USA
http://www. mathsoft.com/

Appendix B

The Physical Meanings of K and Q

In the derivation of the filter circuit in Appendix A, we introduced two quantities,

K and Q, defined as
bw
Kij =
gigj
and
Q =

g1
bw

where bw is the relative bandwidth and gi is the ith g value from filter theory. As the
derivation in Appendix A shows, K is the impedance inverter constant of an impedance
inverter between two neighbouring resonators, and ( 1 / Q )0.5 is the impedance inverter
constant of an impedance inverter between the first (or last) resonator and the 1 source
(or load).
These K and Q, seemingly introduced merely for notational convenience, have
important physical meanings. K is equal to the coupling constant between the two
resonators, which is defined commonly as
K = 2(f2-f1)/(f2+f1),
where f1 and f2 are the resonant frequencies of a circuit that consists of just two coupled
resonators. Q is equal to the loaded quality factor QL of a circuit that consists of only one
resonator coupled to a normalized load impedance.
In this appendix, we will prove that K and Q, as used in Appendix A in the
impedance inverters, indeed have these physical meanings of coupling constant and
loaded quality factor. These physical meanings of K and Q are crucial parts in the filter
design strategy, because the calibration projects in HFSS, where relations are established
between K and Q and certain physical dimensions in the filter, rely entirely on these
meanings.
We start with the physical meaning of K. Consider the circuit depicted in Fig. B1,
consisting of two coupled resonators without a source or a load. The resonators are
coupled through the impedance inverter with constant K. We need to prove that this
circuit has resonant frequencies f1 and f2 that are related to K via the equation
K = 2(f2-f1)/(f2+f1).


j C

j C

j L

j L

+ j L
j C
Fig. B1
Circuit used for the determination of the relation between impedance inverter constant K
and resonant frequencies f1 and f2
Applying Kirchhoffs Voltage Law to the loop left of the impedance inverter leads to

1
I0
jL
jC

1
jL
jC

=0

An obvious trivial solution to this equation is I0=0. We are looking for resonances, so we
require that non-trivial solutions exist. This is the case if

1
jL
jC

1
jL
jC

=0

Rewriting this so every term obtains the same denominator leads to

1 LC 2 K 2 2 C2
=0
2
jC 1 LC

To derive a solution which makes physical sense, one has to solve

1 LC 2 K 2 2 C2 = 0

Upon multiplying out the bracket term and dividing by LC a standard quadratic equation
in 2 results:
2

K
1

LC L2 ( LC) 2 =

The two solutions are

1 2

1 2
K

2 LC L2

1 2
K
1

2
4 LC L2

( LC)
2

+/-

The application of Vietas theorem gives the two relations

 1 2 2 2
1 2
2

=
=

2
K

LC L2
1
( LC)

For the resonator models the condition

L
= 1
C

has been chosen. Together

2 with the second Vieta relation this leads to

=
1 2

Now the first Vieta condition is solved for the impedance inverter constant K as function
of the two eigenvalues

2 1 2
2 1 2 K
2

This can be rewritten as

1 2
1 2

2 1 2

Replacing the geometric mean by the arithmetic mean leads to the well-known formula:

K
or

1 2

2 f 1 f 2
f1 f2

In practical cases, where the eigenfrequencies are close together the arithmetic mean and
the geometric mean give almost identical results. This completes the proof for K.
We now proceed with the physical meaning of Q. Consider the circuit depicted in Fig.
B2, consisting of one resonator, coupled to a 1- load impedance through an impedance
inverter. The impedance inverter has coupling constant 1/Q, denoted here for brevity as
K. We need to prove that the quality factor of this circuit is equal to Q.

j C

j L

Fig. B2
Circuit used for the determination of the relation between impedance inverter constant
and quality factor
Effectively, the setup is a series resonant circuit with a series resistor of K2 Ohm, an
inductance L and a Capacitor C. According to basic electrical-engineering theory, the
quality factor of such a circuit is defined as

Q=

2 f rL
K

and the resonant frequency is equal to

1
2 L C
Substituting the latter expression for fr in the equation for Q gives

Q=

1
K

L
C

For our resonators we have chosen the condition

L
= 1
C

Therefore the impedance inverter constant K, depicted in Fig. B2, is equal to

K=

1
Q

where, as said earlier, Q is the quality factor of this circuit. This completes the proof that
the Q, as introduced in Appendix A, has the physical meaning of the loaded quality factor
of the circuit of Fig. B2.