0% found this document useful (0 votes)
422 views342 pages

Filter Synthesis Using Genesys S - Filter PDF

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
422 views342 pages

Filter Synthesis Using Genesys S - Filter PDF

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 342

Using Genesys S/Filter

Filter Synthesis
S/Filter includes tools beyond direct synthesis, including a wide variety of both
exact and approximate equivalent network transforms, methods for selecting
the most desirable out of potentially thousands of synthesized alternatives, and
a transform history record that simplifies design attempts requiring iteration.
Very few software programs are based on direct synthesis, and the additional
features of S/Filter make it a uniquely effective tool for filter design.

This resource presents a practical guide to using Genesys software for


microwave and RF filter design and synthesis. The book focuses on how to
use direct synthesis and the application of S/Filter features to solve common
filter problems. Both lumped element and distributed filters are discussed, with
extensions to dielectric and quartz crystal resonators.

Contents Overview:
Transmission Zeros; All-Pole Lowpass and Highpass; Lowpass with Finite
Zeros; Conventional Bandpass; Extraction Sequences; Customized Bandpass
Filters; Norton Transform; Bandpass with Resonators; TEM-Mode Resonators;
Piezoelectric Devices; Symmetry; S/Filter and Matching; Distributed Filters;
Distributed Lowpass Filters; Distributed Bandstop Filters; Distributed Bandpass
Filters; Distributed Highpass Filters; Electromagnetic Simulation; Classic
Bandpass Method

Randall W. Rhea is a leading RF and microwave engineering expert with


extensive industry experience working for Boeing Company, Goodyear
Aerospace, and Scientific-Atlanta. He is the author of popular books in the field
as well as numerous technical papers and tutorial CDs. He is a graduate of the
University of Illinois and Arizona State University.

Rhea
Include bar code

ISBN 13: 978-1-60807-802-8


ISBN 10: 1-60807-802-7

BOSTON LONDON
Randall W. Rhea
www.artechhouse.com
Filter Synthesis Using Genesys S/Filter
For a complete listing of titles in the
Artech House Microwave Library
turn to the back of this book.
Filter Synthesis Using Genesys S/Filter

Randall W. Rhea
Library of Congress Cataloging-in-Publication Data
A catalog record for this book is available from the U.S. Library of Congress.

British Library Cataloguing in Publication Data


A catalogue record for this book is available from the British Library.

Cover design by Vicki Kane

ISBN 13: 978-1-60807-802-8

© 2014 ARTECH HOUSE


685 Canton Street
Norwood, MA 02062

Cover image © Agilent Technologies, Inc. 2013–2014.

All rights reserved. Printed and bound in the United States of America. No part of this book
may be reproduced or utilized in any form or by any means, electronic or mechanical, including
photocopying, recording, or by any information storage and retrieval system, without permission
in writing from the publisher.
All terms mentioned in this book that are known to be trademarks or service marks have been
appropriately capitalized. Artech House cannot attest to the accuracy of this information. Use of
a term in this book should not be regarded as affecting the validity of any trademark or service
mark.

10 9 8 7 6 5 4 3 2 1
To Marilynn:

As just a boy, I made my choice,


I now look back, and proudly smile.
With wind in face, or at our backs,
hand in hand, we trod our paths,
some a thousand, some a mile.
Youthful dreams, some wise some poor,
but oh so sweet, the ones we store.
For Hera’, only change endures,
not so we say, tis love evermore,
hand in hand, tis love evermore.
Contents
Preface xiii
References xiv
1 Transmission Zeros 1
1.1 Determining TZ by Inspection 1
1.2 Filter Degree 4
1.3 Canonical Realization 4
1.4 Influence of TZs on the Response 4
References 6
2 All-Pole Lowpass and Highpass 7
2.1 Initial All-Pole Lowpass Parameters 7
2.2 Dual Topologies 9
2.3 Chebyshev Approximation with Even Order 10
2.4 All-Pole Highpass Example 11
References 12
3 Lowpass with Finite Zeros 13
3.1 Introduction 13
3.2 Alternative Topologies 15
4 Conventional Bandpass 17
4.1 Bandpass Transform 17
4.2 Classification Symmetry or Antimetry 17
4.3 A 75- to 125-MHz Bandpass 18
4.4 A 96- to 104-MHz Bandpass Filter 19
4.5 Comparative Analysis of the Wide and Narrow Filters 19
Reference 21
5 Extraction Sequences 23
5.1 The Extraction Tab 23
Reference 27
6 Customized Bandpass Filters 29
6.1 Custom Filter Specification 29
6.2 Partial Extractions of FTZs 33
6.3 Inexact Extractions 34
6.4 Inexact Example 34
7 Norton Transforms 39
7.1 Norton Series Transform 39
7.2 Removing a Transformer with the Series Norton 40
viii Filter Synthesis Using Genesys S/Filter

7.3 Norton Shunt Transform 43


7.4 Equal-Valued Inductor Bandpass 44
7.5 The History Tab 45
7.6 Equate All Ls 46
8 Bandpass with Resonators 47
8.1 Coupled Parallel-Resonator Filters 47
8.1.1 Exact Design of a Parallel Resonator All-Pole Filter 49
8.1.2 Termination Coupling Transforms 51
8.1.3 Find Dual Transform 53
8.1.4 Exact Design with Like Coupling Elements 55
8.1.5 The Equate All Shunt Ls or Shorted Stubs Transform 56
8.1.6 Termination-Coupled Bandpass 57
8.2 Coupled Series-Resonator Filters 58
8.2.1 The Basic Series-Resonator Bandpass 58
8.2.2 Tubular Bandpass 59
8.2.3 Manufacture of the Tubular Bandpass 61
8.2.4 Generalized Series-Resonator Bandpass 61
8.2.5 Tunable Constant-Bandwidth Bandpass 63
Reference 67
9 TEM-Mode Resonators 69
9.1 Filter Insertion Loss 69
9.2 Filter Using 50-Ohm Coaxial Resonators 70
9.2.1 Lumped to Distributed Equivalents 70
9.2.2 The Convert Using Advanced Tline Routine 72
9.3 Generalized Bandpass Using Ceramic Resonators 74
9.3.1 Creating Parallel Resonators 75
9.3.2 Shifting the Internal Impedance Level 76
9.3.3 The Pi to Tee Transform: Increasing Coupling Caps 77
9.3.4 Converting the Parallel L-C to Coaxial Resonators 77
9.3.5 Optimizing the Values 77
9.4 Ceramic Bandpass with Two FTZs 78
References 81
10 Piezoelectric Devices 83
10.1 Quartz-Crystal Device Model 83
10.1.1 Physical Form of the Quartz Crystal 83
10.1.2 Insertion Response of a Quartz Crystal 84
10.1.3 Modeling the Quartz Crystal 84
10.1.4 Calculating Model Parameters from the Response 85
10.1.5 The Quartz-Crystal Model and Filter Design 86
10.2 Quartz-Crystal Filter Approximate Design 86
10.3 Nulling the Static Capacitance 90
10.4 Design of a Lower-Sideband Crystal Filter 91
10.5 Upper-Sideband Quartz-Crystal Filter 97
10.6 Filters with TZs Above and Below the Passband 103
Contents ix

10.7 Wide-Bandwidth Quartz-Crystal Filters 107


10.8 Very Wide-Bandwidth Quartz-Crystal Filters 108
10.9 Ceramic-Piezoelectric Resonators 111
Reference 113
11 Symmetry 115
11.1 Physical Symmetry 115
11.1.1 A Lowpass Filter with FTZ Pairings 115
11.1.2 A Bandpass Filter with FTZ Pairings 117
11.2 Response Symmetry 119
11.2.1 All-Pole Symmetric Response Filters 120
11.2.2 Generalized Bandpass with Symmetric Response 120
11.2.3 Symmetry by FTZ Placement 123
11.3 Group-Delay Equalization 124
References 127
12 Matching with S/Filter 129
12.1 Matching Concepts 129
12.1.1 Complex Conjugate Match 130
12.1.2 Two-Element Matching Networks 130
12.2 Real Terminations 132
12.2.1 Exploiting Extraction Sequences 132
12.2.2 Exploiting Resonator Filters 138
12.3 Complex Terminations 139
12.3.1 Fano’s Limit 139
12.3.2 Example: Power Amplifier Match 140
12.3.3 Example: Broadband Antenna Match 142
References 144
13 Distributed Filters 145
13.1 Comparing Distributed and Lumped Filters 145
13.2 The Genesys Microwave Filter Module 146
13.3 Distributed Synthesis Concepts 149
13.3.1 TLEs 149
13.3.2 Richards Transform 150
13.3.3 Kuroda Identities 152
13.3.4 Ikeno Transforms 155
13.3.5 Kuroda-Minnis Transform 157
13.3.6 Half-Angle Transform 159
13.3.7 Interdigital Transform 161
13.3.8 Combline Transform 161
13.4 Lumped to Distributed Equivalent Transforms 162
13.5 Inverters 164
13.6 The Convert Using Advanced TLine Routine 165
13.7 Box Modes 166
13.8 Introduction to Distributed Filter Examples 166
References 167
x Filter Synthesis Using Genesys S/Filter

14 Distributed Lowpass Filters 169


14.1 Exact Methods 169
14.1.1 Lowpass with Redundant UEs 169
14.1.2 Stub TLEs and Contributing Unit Elements 175
14.1.3 Lowpass with Only Contributing UEs (Stepped-Z) 176
14.1.4 Generalized Lowpass Filter 179
14.2 Approximate Methods 180
14.2.1 All-Pole: Equivalent Series TLE and Shorted Stubs 182
14.2.2 Stepped Impedance Lowpass 183
14.2.3 Generalized Lowpass 187
14.3 Size Reduction by Penetration 190
14.4 Radial Stub Lowpass 192
14.5 Hybrid Lowpass 194
14.6 Distributed Lowpass Summary 196
Reference 198
15 Distributed Bandstop Filters 199
15.1 All-Pole with Stubs and Contributing UEs 199
15.1.1 Wide Bandwidth Bandstop 199
15.1.2 Moderate Bandwidth Bandstop 202
15.1.3 Narrow Bandstop with Ikeno Transforms 204
15.2 Generalized Narrowband Bandstop 205
16 Distributed Bandpass Filters 211
16.1 Tutorials of Bandpass by Synthesis 211
16.1.1 Edge-Coupled Using Richards Transform 211
16.1.2 Edge-Coupled Using Inverters 216
16.1.3 Interdigital Using Inverters 218
16.2 Unique Bandpass Designs 224
16.2.1 Combline with Capacitive External Coupling 224
16.2.2 Miniature Bandpass with Contributing UEs 228
16.2.3 Narrow Bandwidth with UEs and an FTZ 233
16.2.4 Penetrating Combline 238
16.2.5 Minnis Class-D Bandpass 245
16.3 Hybrid Bandpass 248
16.3.1 Penetrating Combline with Capacitors 248
16.3.2 Generalized Combline Hybrid 249
16.3.3 Direct-Coupled Bandpass with Capacitors 252
References 258
17 Distributed Highpass Filters 259
17.1 The Hybrid Highpass 259
17.1.1 The All-Pole Hybrid: Distributed Synthesis 259
17.1.2 The All-Pole Hybrid Highpass: Lumped Synthesis 261
17.1.3 The Hybrid Highpass with UEs 263
17.1.4 The Hybrid Highpass with an FTZ 266
17.2 Purely Distributed Highpass 268
Contents xi

17.2.1 Highpass with Three TZs at DC and a UE 268


17.2.2 Highpass with Three TZs at DC and Four UEs 270
17.3 The Highpass Synthesized as a Bandpass 272
17.3.1 Hybrid Highpass from an Eighth-Degree Bandpass 272
17.3.2 Hybrid Highpass from a 10th-Degree Bandpass 275
18 Multiplexers 277
18.1 Contiguous Multiplexers 277
18.1.1 Contiguous Lowpass-Highpass Diplexer 277
18.1.2 Contiguous LP/BP/HP Multiplexer 279
18.2 Noncontiguous Multiplexers 281
18.2.1 Noncontiguous LP/HP Diplexer with FTZ 281
18.2.2 Noncontiguous Distributed Combline Diplexer 284
Reference 287
19 Electromagnetic Simulation 289
19.1 Overview 289
19.1.1 The EMPower Program 290
19.1.2 The Momentum Program 291
19.1.3 The EMPro Program 292
19.2 Box Modes 292
19.3 EM Simulation of Distributed Circuits 295
19.3.1 EM Simulation of Penetrating Stepped-Z Lowpass 295
19.3.2 EM Simulation of a Combline Bandpass 298
19.3.3 EM Simulation of a Direct-Coupled Bandpass 300
19.4 Classic Method of Bandpass Design 302
19.4.1 Classic Method Fundamentals 302
19.4.2 Example: Determining K Values 304
19.4.3 Example: Determining Q Values 307
19.4.4 Filter Example Using the Classic Method 307
References 310
Appendix A: Example Summary 313
A.1 Lumped Examples 313
A.2 Distributed Examples 315
A.3 Hybrid Examples 316
A.4 Multiplexer Examples 317
Constants, Symbols, and Initialisms 319
About the Author 323
Index 325
Preface
Filter Synthesis Using Genesys S/Filter covers the design of practical
lumped element, distributed, and resonator based filters from the
perspective of transmission zeros. While this book is of some interest to all
filter designers, it is written for users of the synthesis program Genesys
S/Filter from the Agilent EEsof EDA division of Agilent Technologies.
In the 1920s, filters were designed by cascading k-type sections with
attenuation poles (transmission zeros) at infinite frequencies and m-type
sections with additional transmission zeros at finite frequencies. This
method was mathematically economic and suitable for the calculation tools
then available, but the filter impedances and responses were only
approximate. In the 1930s, theories of network synthesis evolved that
produced exact designs with polynomial approximations to the ideal zonal
(brick wall) response filter. By the 1950s, solutions to these more
mathematically involved methods were found and tabulated. These tables
only required simple impedance and frequency scaling to produce efficient
lowpass and highpass filters, and simple transformations for bandpass and
bandstop filters. This became known as the modern method and today it is
still the common filter design technique. A reference to this method for RF
and microwave applications is the venerable Microwave Filters,
Impedance-Matching Networks, and Coupling Structures [1]. The
popularization of the digital computer resulted in a plethora of software
programs, most of which employ the modern method.
The modern method which is based on only a subset of topologies
available from general network synthesis does not necessarily result in the
most efficient filter. Orchard and Temes solved certain synthesis problems
caused by the precision of digital computers [2], and Szentirmai is a
pioneer in the development of commercial software for the direct synthesis
of filters [3]. Filter synthesis is a beautiful example of applied mathematics,
arguably one of the better examples in engineering. However, this book is
not about synthesis theory. This subject is treated in depth by [4],[5]. Note
that this book is not a treatment of modeling, tuning, standard values,
tolerance, or Q of elements and their effects on filter performance. Those
topics are covered in depth in my other works [6],[7].
The focus of the book is common filter design problems and how to use
direct synthesis to solve those problems. The emphasis is not on the
S/Filter user environment. The software documentation provides that
information. Rather, this book covers the application of S/Filter features to
solving important and common filter problems. S/Filter’s strength is the
incorporation of features that help simplify the powerful, but confusing-to-
the-novice, process of direct filter synthesis. S/Filter includes tools beyond

xiii
xiv Filter Synthesis Using Genesys S/Filter

direct synthesis, including a wide variety of both exact and approximate


equivalent network transforms, methods for selecting the most desirable
structure out of potentially thousands of synthesized alternatives, and a
transform history record that simplifies design attempts requiring iteration.
Very few software programs are based on direct synthesis, and the
additional features of S/Filter make it a uniquely effective tool for filter
design. The book assumes the reader is familiar with basic RF and
microwave concepts.
The author thanks How-Siang Yap of Agilent Technologies for
suggesting the inclusion of a chapter on multiplexers and for other
comments. He also thanks Rulon VanDyke of Agilent Technologies for
leading the incorporation of improvements to S/Filter suggested during the
writing of the book. The author also acknowloges the professional staff at
Artech House for their assistance in finalizing the book.

References
[1] G. Matthaei, L. Young and E.M.T. Jones, Microwave Filters, Impedance-
Matching Networks, and Coupling Structures, Artech House, Dedham, MA, 1980.
[2] H. Orchard and G. Temes, “Filter Design Using Transformed Variables,” IEEE
Trans. on Circuit Theory, Vol. CT-15, December 1968, pp. 385-408.
[3] G. Szentirmai, “FILSYN – A General Purpose Filter Synthesis Program,” Proc. of
the IEEE, Vol. 65, October 1977, pp. 1443-1458.
[4] A. Zverev, Handbook of Filter Synthesis, John Wiley, Hoboken, NJ, 1967
[5] G. Temes and S. Mitra, editors, Modern Filter Theory and Design, John Wiley,
New York, 1973.
[6] R. Rhea, HF Filter Design and Computer Simulation, SciTech Publishing,
Raleigh, NC, 1994.
[7] R. Rhea, Practical Issues in RF Design (Three CD Series), SciTech Publishing,
Raleigh, NC, 2003.
1 Transmission Zeros
Central to filter design by direct synthesis is the specification of
transmission zeros (TZs) and understanding their influence on the filter
response. With the modern method refined in the 1950s, specified filter
parameters include the type of approximation, such as Butterworth or
Chebyshev, the passband frequencies, the termination impedance, and the
order of the filter. The TZs are fixed by the filter type. Fixed TZs simplify
the design process, but they restrict the characteristics of the filter. The
newer direct synthesis method described in this book supports flexible
specification of TZs, thus providing the designer with more control [1],[2].
The first step is to understand TZs and their influence on the filter.

1.1 Determining TZ by Inspection


Consider the fourth-order lowpass filter schematic in Fig. 1.1a. Also shown,
as open or closed switches, are the transmission characteristics of each of
the elements at DC and at infinite frequency (infinity). At DC, each of the
inductors pass signal and each of the capacitors is open, and therefore do
not shunt signal to ground. At infinity, each inductor is open and each
capacitor shunts signal to ground. This filter has no TZ at DC and it has
four TZs at infinity. Because this filter passes low frequencies and
attenuates high frequencies, it is a lowpass filter.
Consider the lowpass filter in Fig. 1.1b. The antiresonant series branch
to ground produces a finite-frequency transmission zero (FTZ) at its
resonant frequency. At infinity, the shunt inductor is an open and this
shunt branch is ignored. Therefore, inductor L1 and L2 are redundant
opens. They are redundant because two series inductors are replaced with a
single inductor whose value is the sum of the individual inductors.
Therefore, this filter has no TZs at DC, two TZs at infinity, and one FTZ.
The FTZ has removed one TZ at infinity. The FTZ is placed above the
passband. Filters with TZ at only DC or infinity are referred to as all-pole
filters. Filters with FTZ are referred to as generalized filters [1]. As
discussed in detail later, elliptic filters are a subclass of generalized filters.
Highpass filters have TZs at DC and if present, FTZs are below the
passband.

1
2 Filter Synthesis Using Genesys S/Filter

Figure 1.1 (a) All-pole lowpass filter showing DC and Infinite TZs and (b) a generalized
lowpass with also a TZ at finite frequency.

Consider the third-order (sixth degree) bandpass filter in Fig. 1.2a. At


DC, the shunt inductor in the first branch is a TZ, the series capacitor in the
second branch is a TZ, and the shunt inductor in the third branch is a TZ.
This filter has three TZs at DC. At infinity, the shunt capacitor in the first
Transmission Zeros 3

branch is a TZ, the series inductor in the second branch is a TZ, and the
capacitor in the third branch is a TZ. This filter has three TZs at infinity.
Because it has TZ at both DC and infinity, this is a bandpass filter.

Figure 1.2 (a) Conventional three-resonator all-pole bandpass filter and (b) generalized
bandpass filter with FTZs.

Consider the 45- to 85-MHz bandpass filter in Fig. 1.2b. At DC, all of
the shunt branches are open and they play no role. Capacitor C5 is shorted
by L3, so C5 plays no role. Capacitors C1, C4, and C7 are all in series and are
replaced with one series capacitor of smaller value. This filter has one TZ at
DC. At infinity, capacitors C4 and C5 form a short that places capacitors C3
and C6 in parallel. These latter two capacitors are replaced with one
capacitor of larger value. The branch containing L2 and C2 becomes an open
because of L2 and this branch plays no role. Therefore, L1, C3 in parallel
with C6, and L4 form three TZs at infinity. L2 and C2 series resonate at 30
MHz forming a short to ground resulting in a FTZ. C5 and L3 resonate at
102 MHz forming an open resulting in a FTZ. This filter has one TZ at DC,
three TZs at infinity, one FTZ below the passband, and one FTZ above the
passband.
4 Filter Synthesis Using Genesys S/Filter

1.2 Filter Degree


The degree of a filter [3] is the order of the transducer function polynomial,
H(s). Each TZ at DC adds one degree to the filter transfer function. In
addition, each TZ at infinity adds one degree to the filter. Each finite-
frequency TZ adds two degrees.

1.3 Canonical Realization


A canonical filter is a filter network realized using a minimum quantity of
elements. All filters as initially synthesized by S/Filter are canonic.
Although they possess the minimum quantity of elements, canonic filters
may have element values that are difficult to realize, or a topology that is
inconsistent with practical manufacturing methods. A powerful feature of
S/Filter is the inclusion of numerous transforms that are easily
implemented in the integrated environment of Genesys as the design
moves seamlessly from specification, to schematic, to layout, and to
electromagnetic (EM) simulation if desired.
Certain of these transforms, such as the well-known pi to tee, retain a
canonic form of the synthesized filter. Certain other transforms, such as the
powerful Norton transforms, add additional elements and nullify the
canonical status of the filter. Nevertheless, it is shown that these
noncanonical transforms may improve the realizability of a filter.

1.4 Influence of TZs on the Response


Each element in an all-pole lowpass filter results in one TZ at infinity. For
this type of filter, the number of TZs is equal to the filter degree. The solid
trace in Fig. 1.3 is the transmission amplitude response of a fifth order (five
element) Butterworth-approximation lowpass filter with a 0.05-dB cutoff of
100 MHz. In the transition region from 200 to 400 MHz, the attenuation
increases from 11.09 to 40.84 dB, or 29.75 dB. This is approximately five
times 6.02 dB per octave. Each TZ adds approximately 6.02 dB per octave
in the transition region.
Transmission Zeros 5

Figure 1.3 Lowpass filter responses of a five-element Butterworth (solid), a five-element 0.5-
dB ripple Chebyshev (dashed), and a five-branch Cauer-Chebyshev (dash-dot).

Next, consider a lowpass with five infinite TZ but a 0.05-dB passband-


ripple Chebyshev approximation. From 130 to 260 MHz, the attenuation
increases 36.3 dB, or 7.3 dB per octave. As the passband ripple is increased,
greater selectivity occurs near the cutoff. Deep into the stopband, the
attenuation rate approaches 6.02 dB per octave for each TZ, regardless of
the passband ripple.
Fig. 1.4 shows a generalized lowpass filter with one TZ at infinity and
two FTZs above the passband. This filter has a 0.18-dB passband ripple and
40-dB minimum attenuation in the stopband. The amplitude transmission
response is also plotted in Fig. 1.3. The FTZ are placed at 140 and 205 MHz
to provide equal stopband attenuation. This class of filter with equal ripple
in the passband and stopband is referred to as a Cauer-Chebyshev, or an
elliptic filter. The nomenclature elliptic is used not because the zeros of the
transmission function lie on an ellipse in the complex plane. This is true for
an all-pole Chebyshev. The nomenclature elliptic [4] is used because the
synthesis polynomials involve elliptic functions.
The elliptic response depicted in Fig. 1.3 has zeros specifically placed to
provide equal ripple in the stopband attenuation. This is a characteristic of
Cauer-Chebyshev elliptic filters. This class of filter can be designed using
the modern filter design method. However, it is overly restrictive because it
dictates equal-stopband ripple. The design of filters using direct synthesis
supports arbitrary placement of TZs and therefore provides custom
responses and potentially a more economic design.
6 Filter Synthesis Using Genesys S/Filter

There are rules governing the specification of TZ for synthesis in


S/Filter. Some of these rules are absolute and must be adhered to. Other
rules are required for exact synthesis but may be overruled if an
approximate response is acceptable. These rules are discussed later.

Figure 1.4 Lowpass filter with one TZ at infinity and two FTZ.

References
[1] R.W. Rhea, “Transmission Zeros in Filter Design,” Applied Microwave &
Wireless, January 2001, pp. 92, 94, and 96.
[2] R.W. Rhea, Filter Design by Transmission Zeros (CD), SciTech Publishing,
Raleigh, NC, 2003.
[3] G. Temes and S. Mitra, Modern Filter Theory and Design, John Wiley & Sons,
New York, 1973.
[4] A.I. Zverev, Handbook of Filter Synthesis, John Wiley & Sons, Hoboken, NJ,
1967.
2 All-Pole Lowpass and Highpass
The direct synthesis method offers no advantage over the modern
method for all-pole lowpass and highpass filters. Nevertheless, the basic all-
pole filters are important and are used in this chapter to introduce the
concept of direct-synthesis filter design.

2.1 Initial All-Pole Lowpass Parameters


Given in Fig. 2.1 is the Specification tab of the S/Filter program dialog
window in the Genesys platform of the Agilent EEsof EDA division of
Agilent Technologies [1]. It is launched by opening Genesys, right-clicking
Designs in the workspace tree, and selecting Add, Synthesis, Run S/Filter.
All programs in the Genesys platform comply with the concepts of the
Common User Access standard and Microsoft Windows, so operation of
S/Filter is intuitive for those who use other programs of this type.

Figure 2.1 S/Filter dialog window showing the Specification tab.

This example specifies a source and load resistance of 50 ohms and a


cutoff frequency of 100 MHz. The passband ripple is specified as 0.5 dB.
Table 2.1 relates filter reflection coefficient ρ , return losses, scattering
parameters (Snn), voltage standing-wave ratios (VSWRs), and ripples (LA-
dB) for a filter with lossless elements. Table 2.1 is used to select a ripple

7
8 Filter Synthesis Using Genesys S/Filter

based on the required passband return loss for an ideal filter. A passband
ripple of 0.5 dB corresponds to a return loss of 9.65 dB.
Table 2.1 Representative Values of Radial-Scaled Parameters
Snn(dB) VSWRn LA(dB)
ρn
0.010 -40.00 1.020 0.0004
0.032 -30.00 1.065 0.0043
0.056 -25.00 1.119 0.0138
0.100 -20.00 1.222 0.0436
0.158 -16.00 1.377 0.1105
0.200 -14.00 1.499 0.1764
0.251 -12.00 1.671 0.2830
0.316 -10.00 1.925 0.4576
0.333 -9.54 2.000 0.5118
0.398 -8.00 2.323 0.7494
0.500 -6.02 3.000 1.2496
0.600 -4.44 3.997 1.9365
0.707 -3.01 5.829 3.0106
0.794 -2.00 8.724 4.3292

These are referred to as radial-scaled parameters because they all


represent a vector length outward from the center of the Smith chart.

RL(dB) = − S nn (dB) = −20 log ρ n (2.1 )

1 + ρn
VSWR = (2.2)
1 − ρn

(
L A (dB) = −10 log 1 − ρ n
2
) (2.3)

where n is the port number.


In the S/Filter Specification tab, the ¼ Wave Freq is used only for
distributed filters and it specifies the frequency at which each distributed
element is a quarter-wavelength long. The process is lumped, meaning each
element is an inductor or capacitor with no electrical length. The Shape
refers to the transfer approximation in the passband. Max Flat is a
Butterworth approximation and Equiripple is a Chebyshev approximation.
During synthesis, S/Filter maintains these passband approximations
regardless of the quantity and frequency of the transmission zeros. Linear-
phase passband approximations are not synthesized by S/Filter.
The filter types synthesized by S/Filter include lowpass, highpass, and
bandpass. Lumped-element bandstop filters are designed in S/Filter by
parallel connecting lowpass and highpass filters. A more practical approach
to designing bandstop filters is using the Passive Filter, Microwave Filter,
and Active Filter modules of Genesys.
All-Pole Lowpass and Highpass 9

To specify a fifth-order lowpass, five TZs at infinity are specified in the


TZ table.

2.2 Dual Topologies


Given in Fig. 2.2 is the Extraction tab of the S/Filter window. Notice that
Series Element First is not selected. This forces synthesis with the first
element on the left as a shunt element, a capacitor in the case of a lowpass
filter. If this selection is checked, the first element will be in series, an
inductor in the case of a lowpass filter.

Figure 2.2 Extraction tab of the S/Filter window.

These two forms are given in Fig. 2.3. The two topologies are the dual
of each other. The transmission characteristics of these dual forms are
identical in all respects. The port-impedance characteristics as a function of
frequency are different, which is important for the diplexer designs
considered later. In addition, element values are different, and certain
forms are advantageous for practical reasons. The designer is free to choose
the desired topology. The remaining selections in the Extraction tab are
covered later.
10 Filter Synthesis Using Genesys S/Filter

L1 L2
L=97.85nH L=97.85nH

C1 C2 C3
C=54.3pF C=80.88pF C=54.3pF

L1 L2 L3
L=135.74nH L=202.19nH L=135.74nH

C1 C2
C=39.14pF C=39.14pF

Figure 2.3 Dual forms of the lowpass filter with shunt element first (top) and series element
first (bottom).

2.3 Chebyshev Approximation with Even Order


Consider the fourth-degree Chebyshev lowpass filter on the left in Fig. 2.4.
Four TZs at infinity are specified in the TZ table. The transmission
amplitude and reflection responses are given on the right in Fig. 2.4 as the
thicker lines. The transmission amplitude response for a true Chebyshev
lowpass approximation has a quantity of transmission-response slope zeros
equal to the degree of the filter. Likewise, the number of minimum and
maximum in the passband return loss equals the order of the lowpass. In
this case, notice that the quantity of transmission slope zeros is four. For
even order, this forces attenuation at DC equal to the ripple value.
All-Pole Lowpass and Highpass 11

ZO=50Ω ZO=25.13Ω

L1 L2
L=94.9nH L=66.99nH

C1 C2
C=53.17pF C=75.32pF

Figure 2.4 Even order (fourth) lowpass filter schematic (left), and the transmission and
reflection responses (right).

Again, consider the schematic. At DC, the capacitive reactance


approaches infinity and the capacitors effectively vanish. Similarly, the
reactance of the inductors approaches zero and the inductors vanish. Since
the source is therefore connected directly to the load, how can the
attenuation at DC equal the ripple value? For attenuation to exist at DC
there must be a mismatch [2]. Notice that this fourth-order Chebyshev
lowpass has a load termination resistance of 25.13 ohms rather than 50
ohms. Dissimilar terminations are required to achieve a true even-order
Chebyshev response.
The thinner lines in the response are with the load termination set at
50 ohms and all elements optimized to achieve equal ripple in the
passband. The attenuation at 140 MHz is reduced by about 1.5 dB. Notice
that the quantity of minimum and maximum values in the passband return
loss is three. The modified element values are substantially different: the
new value for C1 is 41.3 pF, for L1 is 122.5 nH, for C2 is 49.4 pF, and for L2 is
105.2 nH. However, optimization proceeds very quickly, so this is an
effective design procedure.

2.4 All-Pole Highpass Example


The all-pole highpass has all TZs at DC. Like the all-pole lowpass, the
highpass is easily designed using the modern method. The lowpass
prototype g-values are inverted, representing a 1/s transform, and
converting each shunt capacitor to a shunt inductor and each series
12 Filter Synthesis Using Genesys S/Filter

inductor to a series capacitor. Then the values are scaled for frequency and
termination resistance, just as with the lowpass.
Fig. 2.5 shows the schematic and responses of a 0.1-dB ripple
Chebyshev highpass with four TZs at DC. Notice that similar to the lowpass,
the even-order Chebyshev highpass requires dissimilar termination
resistance.

ZO=50Ω ZO=30.86Ω

C1 C2
C=25.08pF C=37.41pF
L1 L2
L=57.74nH L=38.71nH

Figure 2.5 Chebyshev all-pole highpass with four TZs.

References
[1] Agilent Technologies, GENESYS 2012.01 Documentation Set, www.agilent.com.
[2] R.W. Rhea, HF Filter Design and Computer Simulation, SciTech Publishing,
Raleigh, NC, 1994.
3 Lowpass with Finite Zeros
As previously stated, direct synthesis offers no advantage over the
modern method for all-pole lowpass filter design. However, this is not the
case for generalized lowpass filters with FTZs. In this chapter, instruction
on the use of S/Filter continues with the design of generalized lowpass
filters.

3.1 Introduction
The design of generalized lowpass filters will begin with a specific filter
requirement. Consider the following required filter specification:
1) Nominal termination resistances: 50 ohms
2) Passband cutoff frequency: 32 MHz
3) Maximum passband return loss: 16 dB
4) Minimum rejection at 35 MHz: 50 dB
5) Minimum rejection above 38 MHz: 20 dB
The goals are maximum economy with minimum inductor count. The
rejection requirements are indicated on the graph in Fig. 3.1 as a mark at 35
MHz and a shaded region above 38 MHz.
An all-pole 0.1-dB ripple Chebyshev filter that just satisfies the
requirements is 19th order, thus requiring nine inductors and ten
capacitors. The response is given in Fig. 3.1 as the dotted trace. A schematic
is not given here.
Next, an elliptic Cauer-Chebyshev filter is designed using the modern
filter design method. A schematic of the solution that just satisfies the
requirement is given in Fig. 3.2. It satisfies the 35-MHz rejection
requirement with frequency margin, but the next lower order filter is
insufficient. This filter requires four inductors and nine capacitors, a
significant improvement over the all-pole Chebyshev. It is incorrect to
assume that Cauer-Chebyshev filters always have superior economy to all-
pole filters. This comparison depends on specific filter specifications.

13
14 Filter Synthesis Using Genesys S/Filter

Figure 3.1 Rejection goals (mark and shaded region) and responses of a 0.1-dB ripple all-
pole Chebyshev filter (Chebyshev), an elliptic Cauer-Chebyshev (Cauer-Cheby), and a
custom filter designed by S/Filter that just satisfies the requirements.

Figure 3.2 A 0.1-dB passband ripple Cauer-Chebyshev (elliptic) filter that satisfies the above
filter specification.

Finally, S/Filter is used to design a filter, with a Specification tab given


in Fig 3.3. the Lumped, Equiripple, and Lowpass radio buttons are
selected, and the desired termination resistances, cutoff frequency, and
ripple are entered. A FTZ at 35 MHz is entered and then the Quantity of TZ
at infinity is increased to five until the rejection specifications are met. The
resulting filter is given in Fig. 3.4.
Lowpass with Finite Zeros 15

Figure 3.3 Specification tab for the S/Filter-designed custom filter.

This solution requires three inductors and five capacitors, a savings of


one inductor and four capacitors with respect to the Cauer-Chebyshev
solution. Other solutions are available that are more economic than the
Cauer-Chebyshev. For example, a filter that uses two TZs spread around 35
MHz can broaden the frequency range of the 35 MHz notch but the filter is
still more economic than the Cauer-Chebyshev by two capacitors.

C3
ZO=50Ω C=176.58pF ZO=50Ω

L1 L2 L3
L=360.59nH L=117.1nH L=360.59nH

C1 C2 C4 C5
C=119.53pF C=136.34pF C=136.34pF C=119.53pF

Figure 3.4 A 0.1-dB passband ripple custom filter designed using S/Filter to synthesize a
solution with user-specified TZs.

3.2 Alternative Topologies


If you duplicate the Specification tab in Fig. 3.3, the filter schematic
displayed by Genesys may be different than Fig. 3.4 because multiple
topologies realize a given transfer function. Another advantage of direct
synthesis is that all of these topologies are available for selection. Managing
this process is discussed in Chapter 5.
4 Conventional Bandpass
This chapter introduces the conventional bandpass and discusses some
of the difficulties that arise when designing bandpass filters. The design of
bandpass filters is significantly more challenging than the design of the
lowpass or highpass.

4.1 Bandpass Transform


Conventional bandpass filters are designed using the modern method by
converting each shunt capacitor in the lowpass prototype to a shunt parallel
resonator, and each series inductor in the lowpass prototype is transformed
to a series resonator in series with the transmission path. Therefore, the
conventional bandpass has twice the quantity of elements as the lowpass
prototype. The lower and upper cutoff frequencies are referred to as fu and
fl, respectively [1]. An important parameter is the absolute bandwidth
BW = f u − f l (4.1)

The bandpass transform operates geometrically. Therefore the


mathematical center frequency for this filter is not the average of the upper
and lower cutoff but

f0 = fu fl (4.2)

This frequency is used to scale the lowpass prototype values. Then the
fractional bandwidth is

bw = BW (4.3)
f0
The percentage bandwidth is the fractional bandwidth times 100%.
These parameters are used to calculate element values using the modern
method. The fractional bandwidth has significant repercussions in the
design of bandpass filters. Small fractional bandwidth results in extreme
element values and high dissipative insertion loss.

4.2 Classification Symmetry or Antimetry


Bandpass filters with an odd quantity of TZ at DC and an odd quantity of
TZs at infinity are classified as electrically symmetric filters. Bandpass
filters with an even quantity of TZs at DC and an even quantity of TZs at

17
18 Filter Synthesis Using Genesys S/Filter

infinity are classified as antimetric filters. Notice that both classes have an
even-quantity sum of TZs. Filters with an odd-quantity sum of TZs are
classified as parametric filters. This class of bandpass filter requires
approximate techniques and it is not designed by S/Filter.

4.3 A 75- to 125-MHz Bandpass


Given in Fig. 4.1 is the Specification tab for a three-section 75- to 125-MHz
passband conventional bandpass filter. The absolute bandwidth is 50 MHz,
the fractional bandwidth is 0.516, and the center frequency is 96.82 MHz. A
schematic of the bandpass is given on the left in Fig. 4.2. At DC, each shunt
inductor and each series capacitor forms a TZ, and at infinity, each shunt
capacitor and each series inductor forms a TZ. There are three TZs at DC
and three TZs at infinity for this filter.

Figure 4.1 Specification tab for the 75- to 125-MHz Chebyshev bandpass.

The transmission amplitude (S21), input reflection (S11), and group-


delay responses are given on the right in Fig. 4.2. With this example, the
element Qs, (unloaded Qs) are specified as 1000 for the capacitors and 100
for the inductors. This introduces dissipative insertion loss, primarily at
frequencies near the passband. The passband ripple is set at 0.0436 dB
resulting in a passband return loss of 20 dB. The group delay peaks at
approximately 19 nS near the lower cutoff and 10 nS near the upper cutoff.
Conventional Bandpass 19

L2 C2
ZO=50Ω L=174.68nH C=15.47pF ZO=50Ω

C1 L3
C=53.31pF L=50.69nH
L1 C3
L=50.69nH C=53.31pF

Figure 4.2 Schematic and responses of the 75- to 125-MHz bandpass filter.

4.4 A 96- to 104-MHz Bandpass Filter


Next, consider a bandpass with the same parameters except with cutoff
frequencies of 96 and 104 MHz. The geometric center frequency is 99.84
MHz and the fractional bandwidth is 0.0801, or 8.01%. As the bandwidth
narrows, the geometric center frequency approaches the arithmetic center
frequency. The schematic and responses for this filter are given in Fig. 4.3.
Notice the change of scale for the frequency-sweep axis and the group-delay
axis.

4.5 Comparative Analysis of the Wide and Narrow Filters


First, consider the dissipative insertion loss. The insertion loss of the 50
MHz bandwidth filter is barely discernable on the 6-dB per division scale. It
is 0.26 dB. However, the dissipative insertion loss of the 8 MHz bandwidth
filter is 1.65 dB. Interestingly, the shape of the dissipative insertion loss is
approximately equal to the shape of the group delay. For example, the
dissipative loss at the lower cutoff is 2.16 dB. This is a ratio of 1.3, equal to
the ratio of the group delay and 96 MHz and band center. The transmission
amplitude response and the group delay response are related by the Hilbert
transform.
20 Filter Synthesis Using Genesys S/Filter

L2 C2
ZO=50Ω L=1091.74nH C=2.32pF
ZO=50Ω

C1 L3
C=333.17pF L=7.62nH
L1 C3
L=7.62nH C=333.17pF

Figure 4.3 Schematic and responses of the 96- to 104-MHz bandpass filter.

The ratio of the insertion losses is equal to the ratio of the bandwidths.
The increase of the insertion loss with decreasing bandwidth is inherent
with bandpass filters and this loss does not depend on the filter topology or
whether the filters are designed using the modern method or direct
synthesis. Improving the insertion loss requires elements or resonators
with improved unloaded Q [1]. Achieving good resonator Q often requires a
manufacturing process other than lumped inductors and capacitors (L-C),
such as coaxial resonators and quartz resonators. This may require all
parallel or all series resonators forms. Direct synthesis of filters with all
equal-type resonators is covered in Chapter 8.
Next, consider the group-delay responses. The group delay at band
center is 8.06 nS for the wider filter and 54.54 nS, approximately inversely
proportional to the bandwidth.
Consider the symmetry of the group-delay responses and the rejection
in the transition regions. The wide filter is highly asymmetric in group
delay and the steepness of the transition region below the passband is
greater than above the passband. This asymmetry is topology-dependent,
and direct synthesis offers effective methods for dealing with these issues,
as is discussed in Chapter 11.
Finally, consider the element values. Notice that the ratio of the largest
to smallest inductor value is 3.45 with the wider bandwidth filter and the
ratio is 143.3 for the narrower filter. While the former ratio is manageable,
realizing the inductor values in the narrow filter is difficult. For lowpass
and highpass filters, the inductor and capacitor ratios are equal to the ratio
of lowpass prototype g-values. Element values are closely related to the
termination resistance for a given cutoff frequency, naturally resulting in
Conventional Bandpass 21

good realizability. For bandpass filters, narrow bandwidth significantly


decreases some, and increases other, element values.
In summary, wide bandwidth presents challenges to filter symmetry
while narrow bandwidth causes issues with element values and insertion
loss. Direct synthesis is used in later chapters to solve the symmetry and
element value problems, but the insertion loss problem is inherent.

Reference
[1] R.W. Rhea, HF Filter Design and Computer Simulation, SciTech
Publishing, Raleigh, NC, 1994.
5 Extraction Sequences
When a filter is synthesized, TZs are extracted into inductors or
capacitors. The extraction sequence of the previous filters with three TZs at
DC and three TZs at infinity is DC ∞ DC ∞ DC ∞. However, with direct
synthesis the TZs may be extracted in other sequences. This offers
alternative topologies and element values that may be more desirable from
a manufacturing standpoint. The Extraction tab of S/Filter is explored in
this chapter using the previous all-pole bandpass as an example.

5.1 The Extraction Tab


Fig. 5.1 shows the Extraction tab of S/Filter for the 75- to 125-MHz
bandpass filter introduced in Chapter 4. If N is the total quantity of TZs,
Nfinite is the number of FTZs, NDC is the quantity of TZs at DC, and N∞ is the
quantity of TZs at infinity, then [1]
N!
Permutations = (5.1)
N finite !× N DC !× N ∞ !

In this case there are six total TZs, no FTZs, three TZs at DC, and three TZs
at infinity, or 20 permutations. However, some permutations are not
unique. For example, the sequence DC ∞ DC ∞ DC ∞ and the sequence ∞
DC ∞ DC ∞ DC are essentially identical because they only change the order
of the elements in each shunt and each series resonator. In addition, for
symmetric filters, the left and right ports may be exchanged resulting in an
identical filter with the same element values and only exchanged positions.
For this filter, six unique permutations exist. The filter shown at the middle
left in Fig. 5.2 as DC ∞ ∞ DC DC ∞ has eight different extraction sequences
with an identical schematic because the elements may be reversed in each
branch, and because this filter is electrically symmetric. This is the all-pole
bandpass filter topology of the filter designed by the modern method. The
extraction sequence shown as DC DC DC ∞ ∞ ∞ has four different
extraction sequences with an identical schematic because the elements may
be reversed in the center branch, and because this filter has mirror
symmetry. The four remaining extraction sequences each have two
identical schematics because elements may be reversed in one branch.
These four sequences require transformers or dissimilar termination
resistances. Transformers are dealt with in a later chapter.

23
24 Filter Synthesis Using Genesys S/Filter

Figure 5.1 Extraction tab for the 75- to 125-MHz bandpass filter.

While the number of unique extraction sequences in this example is


only six, even moderately complex filters can have a large number of
extraction sequences, often reaching into the thousands. For example, a
seventh-order Cauer-Chebyshev has 40,320 total permutations of which
1,656 are unique. Therefore, the Extraction tab has the option Max
Solutions to Display to limit the time and data space required to compute
solutions. The maximum number of supported solutions to display is
25,000, which takes a few minutes to compute, or to sort.
Filters, which have FTZs, have certain constraints on the order of
extraction, or the resulting solution is inexact. Occasionally these inexact
solutions have desirable element values or topologies and the response is
sufficiently accurate. Therefore, there is an option to allow inexact
solutions. This topic is covered in Chapter 6. All extraction solutions for all-
pole filters are exact.
A dual topology exists for all L-C filters. With each of the filters in Fig.
5.2, the first element is shunt. By selecting the Series Element First option,
an additional set of 20 extractions and six unique solutions are available.
The duals of the previous bandpass filters without transformers are given in
Fig. 5.3. The topology on the bottom is the all-pole form designed by the
modern method.
Extraction Sequences 25

C1 L3 C2
L1
C=68.8pF L=39.3nH ZO=50Ω C=305.8pF ZO=50Ω
L=39.3nH

L1 L2 C2 C3 C1 L3 C3
L=50.7nH L=11.4nH C=237pF C=53.3pF C=53.3pF L2 L=2.6nH C=1053.9pF
L=11.4nH

T1
P=1
S=4.4

DC DC DC Inf Inf Inf Inf Inf DC DC DC Inf

C2 C1
L2 L2
C=15.5pF ZO=50Ω ZO=50Ω C=68.8pF ZO=50Ω
ZO=50Ω L=174.7nH L=8.8nH

L1 C1 L3 C3 L1 C2 L3
L=50.7nH C=53.3pF L=50.7nH C=53.3pF L=50.7nH C=237pF L=2.6nH
C3
C=1053.9pF

T1
P=1
S=4.4

DC Inf Inf DC DC Inf DC DC Inf Inf DC Inf

C2 C3
L3 L2
ZO=50Ω C=15.5pF ZO=50Ω ZO=50Ω C=3.5pF ZO=50Ω
L=776.7nH L=174.7nH

L1
L=50.7nH C3 C2 L3
C=2.7pF C=12pF L=1002nH
C1 L2 C1
C=53.3pF L=225.4nH C=53.3pF
L1
L=50.7nH

T1 T1
P=1 P=1
S=0.2 S=0.2

DC Inf DC DC Inf Inf DC Inf Inf Inf DC DC

Figure 5.2 The six unique solutions to the bandpass filter with three TZs at DC and three TZs
at infinity.

The checkbox Specify Permutation allows the user specify a particular


permutation sequence. Clicking the Edit Permutation button opens the
dialog given in Fig. 5.4. The sequence is modified by selecting a TZ and
moving that TZ earlier or later in the sequence.
With all-pole filters such as this example, selecting the extraction
sequence is of limited advantage. Unless the filter is used for matching,
obviously the most desirable filters are the ones without transformers.
Using S/Filter for matching is discussed in Chapter 12. Of the four filters
without transformers, in this case, the filters with the smallest spread of
inductor values are the dual forms of the conventional bandpass.
26 Filter Synthesis Using Genesys S/Filter

C2
L2
ZO=50Ω C=4.6pF ZO=50Ω
L=592.6nH

L3
C1 L=133.3nH
C=20.3pF
L1 C3
L=171.9nH C=15.7pF

DC DC DC Inf Inf Inf


C1 C3
ZO=50Ω C=20.3pF C=20.3pF

L1 L3
L=133.3nH L=133.3nH ZO=50Ω

L2 C2
L=38.7nH C=69.9pF

Inf DC DC Inf Inf DC

Figure 5.3 Duals with a series element first of the two filters in Fig. 5.2 that do not have
transformers.

Figure 5.4 Custom Permutation Order dialog box of S/Filter.


Extraction Sequences 27

As we explore complex filters with more TZ and FTZ, it becomes clear


that realizability is greatly improved by direct synthesis and the ability to
control the extractions. Finding the best solution to thousands of unique
permutations requires tools. This is the function of the features Use Goals,
Edit Goals, Customize Table, and Auto Fit Table.

Reference
[1] Genesys 2012: Synthesis, Agilent Technologies, Santa Clara, CA 2012, p. 292.
6 Customized Bandpass Filters
This chapter describes filter customization by the independent
specification of TZs at DC, infinity and finite frequencies. This chapter also
describes the tools within S/Filter used to find the optimum solution from
potentially thousands of unique topologies.

6.1 Custom Filter Specification


Consider the following required filter specification:
1) Nominal termination resistances: 50 ohms
2) Maximum passband return loss: 20 dB
3) Lower passband cutoff frequency: 3.5 MHz
4) Upper passband cutoff frequency: 29.6 dB
5) Minimum rejection below 2.0 MHz: 40 dB
6) Minimum rejection above 38 MHz: 70 dB
Manufacturability goals include no transformer, a minimum quantity
of inductors, and the smallest possible ratio of inductor values.
An S/Filter synthesis is added to Genesys. The filter passband is
specified as 3.4 to 29.8 MHz to offer some tolerance for tuning the
manufactured filter. An equiripple Chebyshev passband with a 0.043-dB
ripple is entered to achieve a passband return loss of 20 dB. Optimization
goals are added to easily visualize the passband and rejection specifications.
Next, TZs are added and the FTZ frequencies are adjusted until the
specification is satisfied. This is an interactive process, but because the
objectives are visualized, the process is quick. In this case, it is discovered
that four TZs at DC, two TZs at infinity, one TZ below the passband, and
three TZs above the passband satisfy the requirements. The frequencies of
the FTZs are listed in the Specification tab shown in Fig. 6.1.

29
30 Filter Synthesis Using Genesys S/Filter

Figure 6.1 Specification tab for the custom filter.

Next, the Extraction tab is utilized to select the optimum sequence. The
Customize Table button is clicked to open the Customize Permutation
Table dialog. The manufacturability goals for this filter involved inductors
and the transformer, so Maximum C Value and Minimum C Value are
selected in turn and shifted left to the Available column. Then Lmax/Lmin
Ratio and Transformer Turns Ratio are selected in turn and shifted right
to the Show in Table column and OK is clicked. The results are displayed in
Fig. 6.2.

Figure 6.2 Customize Permutation Table dialog after changes.


Customized Bandpass Filters 31

In the Extraction tab, the Auto Fit Table button is clicked to


automatically adjust the width of the columns in the displayed table. There
are 75,600 extraction sequences. To reduce solution time and data space,
8,000 is entered for the Max Solutions to Display. The objective is to locate
the permutation with the fewest number of inductors, a transformer turns
ratio of one, and a solution with a low Lmax/Lmin. An optimizer is
available to match adjustable goals, but the number of inductors is not an
available measurement, so the best technique is to sort the table. Clicking a
column title sorts that column. If the TRF Ratio column is clicked first,
followed by the #Inds column, all solutions with the fewest inductors are at
the beginning of the list and the TRF Ratio is in sorted order. The list is
scrolled until those entries with a transformer ratio of one are located. A
quick scroll through this list reveals the solution with the lowest
Lmax/Lmin. The Extraction tab scrolled to this entry is given in Fig. 6.3.
Two sets of solutions are available, one set with a series element first and
one set with a shunt element first. In this case, both sets offer a minimum
inductor count of seven with no transformer. The series element first offers
a slightly smaller Lmax/Lmin, 6.4 to 1. The extraction sequence is∞ 38 42
DC 54 DC 2.4 ∞ DC DC.

Figure 6.3 Extraction tab for the custom bandpass filter example.

The schematic is given in Fig. 6.4. Because of digital precision, the


S/Filter schematic may include the transformer even if the turns ratio is
unity. If so, the transformer is deleted. S/Filter may not always display
schematic objects in the exact extraction sequence if the displayed
schematics are exactly equivalent. For example, C3, C5, and L3 extract the
32 Filter Synthesis Using Genesys S/Filter

42-MHz TZ, while C4 extracts a TZ at DC. The listed extraction sequence


shows that the DC extraction is after the FTZ, while the schematic shows
them reversed. However, the schematic is valid because C4 and C5/L3 may
be reversed with no effective difference.

Figure 6.4 Schematic of the custom bandpass filter with transformer removed.

Figure 6.5 Amplitude transmission and reflection responses of the custom bandpass.

A Cauer-Chebyshev designed using the modern method requires ninth


order to just satisfy the 35-MHz rejection requirement. This over performs
in the lower stopband, and the filter requires 13 inductors and 13 capacitors
with Lmax/Lmin over 100.
Customized Bandpass Filters 33

6.2 Partial Extractions of FTZs


With all-pole filters, each TZ extracts into one inductor or capacitor.
However, each FTZ requires three elements. Therefore, the previous
custom filter with 10 TZs requires 18 elements. Generally, generalized
filters are economic only when excellent rejection is required near the
passband.
The extraction of a FTZ requires a pole at the FTZ frequency. This
requires forming a remainder impedance function at the FTZ frequency by
the partial extraction of a TZ at DC or infinity. Each partial extraction and
FTZ extraction results in a three-element network with one of the forms
given in Fig. 6.6.
Because a TZ at DC or infinity is only partially extracted, this process
requires that the remaining portion is extracted after the FTZ is extracted.
Therefore, the user must select an extraction sequence with the final TZ at
DC extracted after all the FTZs below the passband are extracted. Likewise,
the final TZ at infinity must be extracted after all FTZs above the passband
are extracted.

Figure 6.6 Four partial and FTZ extraction topologies.


34 Filter Synthesis Using Genesys S/Filter

6.3 Inexact Extractions


If the above partial extraction rules are violated, the resulting filter
response is inexact. If the violation occurs early in the extraction process,
impedance errors ripple through the filter causing significant response
degradation. If the violation occurs late in the extraction process, the
degradation may be minimal. S/Filter displays extraction sequences
including those with violations if the Allow Inexact box is checked. It is
advisable to explore this option, because the economy may be better, and
response errors can potentially be corrected by optimization.

6.4 Inexact Example


Given in Fig. 6.7 is the Specification tab for a 50- to 90-MHz bandpass
0.043-dB ripple filter with one TZ at DC, one TZ at infinity, TZs below the
passband at 22 and 31 MHz, and two TZs above the passband at 144 and
197 MHz. The TZs are specified to realize 70-dB rejection below 32 and
above 140 MHz. The capacitor Q and inductor Q are set at 1000 and 100,
respectively.

Figure 6.7 Specification tab for the 50- to 90-MHz bandpass filter.

The Extraction tab for the exact solutions is given in Fig. 6.8. The
Series Element First type is selected because the inductor values for its dual
are rather small. There are 720 permutations and 57 unique solutions. The
number of unique solutions is a small fraction of the total permutations
because there are only two nonfinite TZs and the elements in each of the
FTZs may be exchanged. To find the best solution, the Permutation Table is
sorted as before. One of the solutions with the minimum number of
inductors, no transformer, and the smallest Lmax/Lmin is selected. The
Customized Bandpass Filters 35

sequence selected is 22 31 DC 197 144∞. The schematic is given in Fig. 6.9.


Seven inductors are required and Lmax/Lmin is 51.6. This ratio is generally
difficult but manageable.

Figure 6.8 Exact extractions for the 50- to 90-MHz bandpass filter.

C1 C3 L3 C5 L5
C=33.7pF C=8.8pF L=3316.5nH C=1.7pF L=584.6nH ZO=50Ω

L7
ZO=50Ω L=140.9nH

L1 L2 L4 L6
L=261.3nH L=979nH L=86.5nH L=64.2nH

C2 C4 C6 C7
C=200.3pF C=26.9pF C=7.5pF C=19pF

Figure 6.9 Schematic of the exact solution to the 50- to 90-MHz bandpass filter.

The transmission amplitude and reflection responses are given in the


left graph of Fig. 6.10. The finite capacitor and inductor Q has resulted in a
0.5-dB insertion loss at band center.
36 Filter Synthesis Using Genesys S/Filter

Figure 6.10 Responses of the exact extraction (left) and the inexact extraction before
optimization (dashed traces on the right graph) and after optimization (solid traces on the right
graph).

Next, the Allow Inexact selection box is checked to explore if an inexact


solution has an advantage. This increases the quantity of unique solutions
to 601, and both exact and inexact solutions display in the S/Filter table. All
the permutations displayed in Fig. 6.11 are inexact and appear dark gray.
Again, the same sorting procedure is used to find the best solution with the
added criteria that while scrolling through the table the response is
observed to ascertain that the response degradation is minimal. The
solution chosen for this example is ∞ 22 144 31 197 DC and displays as dark
black in Fig. 6.11. This permutation violates extraction rules because the
last TZ at infinity is extracted before the FTZs above the passband are
extracted. The extraction of the FTZs below the passband are exact because
the TZ at DC is extracted after those FTZ.
The transmission amplitude and reflection responses for this inexact
permutation are given as dashed traces in the right graph of Fig. 6.10. The
inexact extraction has resulted in minimal degradation of the responses.
The transmission response is affected very little, and the return loss is
degraded from about 22 dB for the exact permutations to about 18 dB.
However, optimizing only the inductors in this filter, as shown by the solid
traces on the right in Fig. 6.10, has improved the return loss to near the
exact response.
Customized Bandpass Filters 37

Figure 6.11 Extraction tab with inexact permutations included in the display.

The schematic for this inexact permutation is given in Fig. 6.12. The
filter is significantly superior to the best exact solution. Although the
quantity of capacitors is increased from seven to nine, the quantity of
inductors is reduced from seven to five and Lmax/Lmin is only 3.0, 17
times lower than the exact solution, thus significantly enhancing inductor
design options.

Figure 6.12 Schematic of the inexact bandpass filter with values after optimization.
7 Norton Transforms
Filter economy and manufacturability are often enhanced by the
application of transforms. Scores of transforms are integrated into S/Filter,
thus simplifying their application to specific filters. They are accessed
through the Transform tab. Experience with transforms is essential to the
mastery of filter design. The transforms are listed in the S/Filter
documentation so a transform-by-transform description is not repeated
here. Some of the more important transforms used to solve common filter
design problems are illustrated throughout the remainder of the book. The
Norton transforms are essential and the basis of many other “macro”
transforms. For example, conversions between Pi, Tee, and L-network
conversions use Norton transforms. Therefore, this chapter is devoted to a
description of the Norton.

7.1 Norton Series Transform


The Norton transform has two basic forms: the series and the shunt. Given
in Fig. 7.1 is the Norton series transform. The elements Z may be a
capacitor, an inductor, a series L-C, or a parallel L-C.

Figure 7.1 Norton series transform.

The formulas relating these elements are


Z
Za = (7.1)
1− n
Z
Zb = (7.2)
n
39
40 Filter Synthesis Using Genesys S/Filter

Z
Zc = (7.3)
n (n − 1)
where n is the turns ratio of the transformer. In S/Filter dialog windows, n
is referred to variously as n or N. Notice that when n>1 then Za is negative,
and when n<1 then Zc is negative. The obvious question is why would it be
desirable to utilize the Norton when one reactor is replaced with a negative
reactor, two positive reactors, and a transformer. The answer is best
illustrated with an example.

7.2 Removing a Transformer with the Series Norton


Consider the 55- to 85-MHz 0.043-dB ripple bandpass filter shown in Fig.
7.2 with one TZ at DC, three TZ at infinity, one TZ at 35.7 MHz, and one TZ
at 107.2 MHz. All the permutations of the exact synthesis of this filter have
a transformer. The extraction sequence selected is ∞ 107.2 35.7 ∞ ∞ DC.

Figure 7.2 A 55- to 85-MHz bandpass filter that requires a transformer.

To remove the transformer, first L4 and C6 are swapped by selecting L4


in the schematic, and applying the Basic Operations: Swap Element
transform in the Transform tab (Fig. 7.3). This swap is necessary because
the Norton transform adds an additional element, and for practical reasons,
it is desirable to add a capacitor rather than an inductor. The Norton series
transform creates a negative valued capacitor on one side of the series
capacitor. To absorb the negative capacitor, the series capacitor must be
adjacent to a larger, positive shunt capacitor. C4 is the target absorbing
capacitor.
Norton Transforms 41

Figure 7.3 The S/Filter Transform tab with Swap element selected.

Next C6 is selected in the schematic and the Norton: Series Transform


is applied. This invokes the Norton Transform dialog given in Fig. 7.4. Four
options are available: choosing the left element Za, the right element Zc, the
transformer ratio n, or letting S/Filter find the transformer ratio that
cancels the existing transformer. In this case, the later option is selected.
The application of this transform displays the schematic given in Fig. 7.5.

Figure 7.4 Norton Transform dialog for specification of the transform parameter.
42 Filter Synthesis Using Genesys S/Filter

Figure 7.5 Schematic of the 55- to 85-MHz bandpass filter after application of the series
Norton.

Next, the Simplify Circuit button in the Transform tab is clicked. This
function combines all transformers and combines like-parallel and like-
series elements. S/Filter determines the resulting transformer has a unity
turns ratio and eliminates it. In addition, S/Filter absorbs the negative
capacitor C6 into C4, thus eliminating C6. The final schematic is given in Fig.
7.6. It is unnecessary to swap C6 and the L3/C5 branch before applying the
Simplify Circuit transform. The Simplify Circuit transform is schematically
global in that it searches the entire schematic during simplification.
Compare the original schematic in Fig. 7.2 with the final schematic in
Fig. 7.6. Inspection of the Norton series diagram in Fig. 7.1 suggests that
the Norton transform replaces a single capacitor with three capacitors and
a transformer. In reality, when applied at the proper position in the filter,
the affect is to replace a transformer with a more economical capacitor.
When applied to an inductor, the Norton replaces a transformer with an
inductor. In essence, the Norton transform is an impedance
transformation.
Norton Transforms 43

Figure 7.6 Schematic of the 55- to 85-MHz bandpass filter after simplification.

7.3 Norton Shunt Transform


The Norton shunt transform is the dual of the series transform applied to a
shunt reactor or L-C pair. The schematic of the Norton shunt transform is
given in Fig. 7.7.

Figure 7.7 The Norton shunt transform.

The formulas relating these elements are


 1
Z a = 1 −  Z (7.4)
 n 
Z
Zb = (7.5)
n
44 Filter Synthesis Using Genesys S/Filter

1− n 
Z c =  2 Z (7.6)
 n 
For the Norton shunt transform, if n<1 then Za is negative and if n>1
then Zc is negative.

7.4 Equal-Valued Inductor Bandpass


Because the Norton transforms shift the impedance level within the filter,
they are often used to change reactor values within a filter. For example,
repeated application of the Norton transform at various locations within a
filter is used to equate all inductors to an identical value. This is desirable if
only specific values of inductance are available.
Again, the design begins with the 55- to 85-MHz bandpass filter given
in Fig. 7.2. Starting on the left, C1 is selected for a Norton shunt transform.
The radio button Choose Transformer Ratio (n) is selected. The reactance
and therefore inductance value is proportional to n2. The square root of the
ratio of L2 to L1, or 1.5288, is entered in the dialog box. The resulting
schematic is given in Fig. 7.8. The application of the Simplify Circuit
transform absorbs T1 into T2 and the negative C3 into the positive C4. Then,
L1 and L2 are equal valued at 250.95 nH.

Figure 7.8 Schematic after the application of the Norton shunt transform to the original C1.

Norton transforms are applied to the left of each remaining inductor


until all four inductors are of equal value. The Norton transforms are
applied to capacitors to avoid increasing the quantity of inductors. Each
Norton transform increases the capacitor count by one. The resulting
schematic with all equal-valued inductors and three additional capacitors is
given in Fig. 7.9.
Norton Transforms 45

Figure 7.9 Final schematic of the 55- to 85-MHz bandpass filter with equal-valued inductors.

7.5 The History Tab


The History tab of the S/Filter dialog lists the selected permutation
sequence and each transform that was applied to the filter. The History tab
for the transforms of the previous bandpass filter is given in Fig. 7.10. The
History tab is convenient for not only reviewing the applied transforms, but
also to edit those transforms. Until experienced, you may apply
inappropriate transforms, or you may use incorrect parameters. The
History tab supports deleting, substituting, and editing of the transforms.

Figure 7.10 History tab of the transforms applied in S/Filter.

When a transform increases the element count, the element


designators after that transform are automatically reassigned. In the above
example, the first Norton shunt transform is applied to the original C1, the
46 Filter Synthesis Using Genesys S/Filter

second Norton transform is applied to the capacitor with the new


designator C4, and the final Norton transform is applied to the new C8.
The schematic at each step in the History tab of the transform process
is displayed by S/Filter whenever that line in the History tab is selected.
Prior to Genesys 2014, the S/Filter dialog is not saved in the Genesys
*.wsp workspace file. To save the S/Filter settings and the History tab it is
necessary to save a *.SF$ file using the Start tab of the S/Filter dialog. In
Genesys 2014 and later, the S/Filter settings are saved in the .wsx
workspace file.

7.6 Equate All Ls


The previous example illustrates the combined use of Norton transforms to
equalize all inductor values in the filter. This is a relatively common
requirement, and so similar functions are built into the S/Filter as
compound transforms. Four compound transforms are available: Equate
All Ls, Equate All Cs, Equate All Shunt Ls or Shorted Stubs, and Equate All
Shunt Cs or Open Stubs. The application of a single Equate All Ls to the
previous initial filter results in a schematic identical to that in Fig. 7.9.
8 Bandpass with Resonators
Consider the conventional bandpass filter in Fig. 4.2. This sixth-degree
filter has three resonators. Two of the resonators are shunt, and one of the
resonators is series. As the bandwidth narrows, the reactance of the shunt
elements becomes low and the reactance of the series elements becomes
high. This often leads to difficulty in the realization of the elements.
Furthermore, realization with specific resonator technologies such as TEM-
mode coaxial resonators, piezoelectric resonators, and other devices often
dictate that all resonators be of similar form; either all parallel or all series.
In this chapter, techniques for designing these topologies are illustrated.

8.1 Coupled Parallel-Resonator Filters


One of the most popular filter topologies for the 30- to 1000-MHz
frequency range is the parallel-resonator filter with coupling capacitors
between resonators. Although not canonic, this topology uses a minimum
quantity of inductors, inductors may all be equal valued, and either the
grounded inductors or capacitors may be tuned.
The classic text covering filters by Matthaei, Young, and Jones, still in
print after nearly 50 years [1], reviews in detail extensions of the modern
method for designing parallel-resonator and series-resonator filters, using
various resonator-coupling techniques. Similar techniques are used by
numerous software programs for designing these filters. Given in Fig 8.1
are the top-C coupled parallel-resonator bandpass (top of figure) and the
shunt-C coupled series-resonator bandpass (bottom) for a 0.0432-dB
ripple Chebyshev with a 90- to 110-MHz passband. These filters are quickly
and easily designed using the Passive Filter module of Genesys. This
program designs lowpass, highpass, bandpass, and bandstop filters with 21
different topologies and a wide variety of transfer approximation shapes.
When the performance and economy of filters designed by the modern
method are sufficient, Passive Filter is the tool of choice, since extraction
sequences and transforms are not utilized, thus simplifying the design
process.
The parallel resonator and series resonator topologies are workhorse
filters for the frequency range from HF through the lower microwave. They
are effective and possess many desirable attributes. However, the design
algorithms use impedance and admittance inverters that assume zero
bandwidth. Inverters are described in Section 13.5. Design errors in the
responses increase with increasing bandwidth. As with other bandpass
filters designed by the modern method, these filters utilize fixed TZ

47
48 Filter Synthesis Using Genesys S/Filter

placement and may not meet system requirements economically. The


consequences of both the design error and inflexible TZ placement become
worse with increasing bandwidth.
C1 C3 C5 C7
C=15.78pF C=5.47pF C=5.47pF C=15.78pF

ZO=50Ω ZO=50Ω

L1 L2 L3 C6
L=100nH L=100nH L=100nH C=7.89pF
C2 C4
C=7.89pF C=15.17pF

C2 C4 C6
L1 L2 L3
ZO=50Ω L=100nH C=57.86pF L=100nH C=42.56pF L=100nH C=57.86pF ZO=50Ω

C1 C3 C5 C7
C=48.95pF C=122.15pF C=122.15pF C=48.95pF

Figure 8.1 A 90- to 110-MHz top-C coupled parallel-resonator bandpass (top) and shunt-C
coupled series-resonator bandpass (bottom).

Given in Fig. 8.2 are the responses of the 90- to 110-MHz bandpass
filters designed in the Passive Filter module. The responses of the top-C
coupled bandpass are given on the left, and the responses of the shunt-C
coupled bandpass are given on the right. Inspection of the top-C coupled
bandpass schematic in Fig. 8.1 reveals that the quantity of TZs at DC is
seven and the quantity of TZs at infinity is one. Therefore, the selectivity of
the response on the left in Fig. 8.2 is much greater below the passband. In
the shunt-C coupled bandpass, the quantity of TZs at DC is one and the
quantity at infinity is seven. Therefore, it has greater selectivity above the
passband.
The return loss peaks in the passband are not uniformly 20 dB. This is
a consequence of errors in the inverter algorithms at 20% bandwidth.
Notice that the center frequency is shifted slightly downward for the top-C
coupled filter and slightly upward for the shunt-C coupled filter. The
Passive Filter module of Genesys includes correction algorithms not
included in Matthaei, Young, and Jones. Nevertheless, above 20%
bandwidth the inverter errors become significant, and the correction
algorithms begin to fail. In the next example, S/Filter is used for the exact
synthesis of a parallel-resonator bandpass filter.
Bandpass with Resonators 49

Figure 8.2 Responses of the 90- to 110-MHz top-C coupled bandpass (left) and the shunt-C
coupled bandpass (right).

8.1.1 Exact Design of a Parallel Resonator All-Pole Filter


The design begins in S/Filter with the same passband specifications but
with three TZs at DC and three at infinity. The extraction sequence DC ∞ ∞
DC DC ∞ is selected with a shunt element first. The resulting schematic is
given in Fig. 8.3.
L2 C2
ZO=50Ω L=438.92nH C=5.83pF ZO=50Ω

C1 L3
C=135.52pF L=18.88nH
L1
C3
L=18.88nH
C=135.52pF

Figure 8.3 Conventional 0.0432-dB 90- to 110-MHz passband filter with three TZs at DC and
three TZs at infinity.

Next, two series Norton transforms are used to convert the series
resonator to a parallel resonator. First, inductor L2 is selected and a series
50 Filter Synthesis Using Genesys S/Filter

Norton is applied. The option Choose Right Element is used to specify a


shunt inductor equal to L1, 18.88nH, resulting in the schematic given in Fig.
8.4.
L3 C2
ZO=50Ω L=82.08nH C=5.83pF ZO=50Ω

L4 L5
L1 L2 L=18.88nH L=18.88nH
L=18.88nH C1 L=-100.96nH C3
C=135.52pF C=135.52pF

T1
P=1
S=5.35

Figure 8.4 Conventional bandpass after application of a series Norton to the original L2.

After applying Simplify Circuit, the second series Norton is applied to


the series capacitor C2. The option Calculate n to Remove Existing
Transformer is chosen. After Simplify Circuit is applied, the parallel-
resonator bandpass in Fig. 8.5 is displayed.

L2 C3
ZO=50Ω L=82.08nH C=31.17pF ZO=50Ω

L3
L=18.88nH L4
L1 C1 L=18.88nH
L=23.22nH C=135.52pF C2 C4
C=135.52pF C=110.18pF

Figure 8.5 Exact 90- to 110-MHz parallel-resonator bandpass filter.

The responses of the exact 90- to 110-MHz parallel-resonator bandpass


are given in Fig. 8.6. Notice that the return loss is uniform and equal to 20
dB and the transmission response is more symmetric. The selectivity is
somewhat greater below the passband. Achieving true response symmetry
is addressed in Chapter 11.
Bandpass with Resonators 51

Figure 8.6 Responses of the exact parallel-resonator bandpass filter.

8.1.2 Termination Coupling Transforms


The inductor values in the previous example are low for realization with
good unloaded (element) Q at 100 MHz. This is resolved by adding series
capacitors to couple to the 50-ohm terminations. This effectively increases
the parallel termination resistance presented to the filter and allows a
higher internal impedance, resulting in increased inductor values. The
transform is given in Fig. 8.7. The shunt element is negative and must be
absorbed by an adjacent parallel element of the same type. The element
values are:

X series = nRs 1 − n 2 (8.1)

n 4 Rs2 + X series
2
X shunt = (8.2)
X series

where n is the transformer turns ratio, Rs is the source resistance, Xseries is


the reactance of the series element, and Xshunt is the reactance of the
negative shunt element. At the source, n values less than 1 are used.
52 Filter Synthesis Using Genesys S/Filter

Figure 8.7 Shunt inductor or capacitor to series-element termination-coupling transform.

Capacitor C1 in the previous filter is swapped with L1 and the Shunt C to


Series C Transform in the Termination Coupling group is applied to C1.
This transform is approximate and so it is displayed as red in the transform
tree in S/Filter. The filter center frequency of 100 MHz is entered in the
first option and a turns ratio of 0.40 is chosen for the final option. This
increases inductor values by 1 n 2 , or 6.25. A Shunt C to Series C transform
is then applied to the shunt capacitor at the load with a ratio of 1/0.40, or
2.5. After Simplify Schematic is applied, the schematic in Fig. 8.8 results.

C1 L2 C4 C6
C=13.9pF C=4.9875pF C=13.9pF
L=513.0034nH

ZO=50Ω ZO=50Ω

L1 L3 L4 C5
L=145.14nH L=118nH L=118nH C=6.2708pF
C2 C3
C=10.4121pF C=21.68pF

Figure 8.8 Parallel-resonator bandpass with series-termination capacitors to increase the


internal-filter impedance and the resulting inductor values.

The responses are given in Fig. 8.9. Although the transform is


approximate, for the moderate bandwidth of this filter, the return loss is
maintained near the intended 20 dB. Notice, however, that the center
frequency is shifted down approximately 2 MHz. This 2% shift is easily
correctable using optimization of the resonator inductance or capacitance.
The same optimization could be used to equalize the resonator inductor
values if desired.
Exact transforms do not change the TZs of a filter, while inexact
transforms may. The Shunt C to Series C transforms added two TZs at DC.
Bandpass with Resonators 53

These are false TZs, and their influence on the response is dependent on the
degree of impedance shift. Notice that the response in Fig. 8.9 has
somewhat greater selectivity below the passband.

Figure 8.9 Responses of the parallel-resonator bandpass with series termination capacitors.

8.1.3 Find Dual Transform


Another potential solution to the small inductor values of the filter in Fig.
8.5, which offers an exact solution, is the dual topology. This solution offers
an identical response with larger values of inductance. Duals are
automatically solved using the Find Dual of Schematic option in the
Compound Operations folder of the Transform tab. The dual found using
this transform is given in Fig. 8.10. The response is identical to Fig. 8.6.

C1 L1 C3 L2 C4 L4
C=9.29pF C=7.55pF C=7.55pF
L=338.81nH L=338.8nH L=275.45nH

ZO=50Ω ZO=50Ω

C2 L3
C=32.83pF L=77.93nH

Figure 8.10 Series dual of the parallel-resonator 90- to 110-MHz bandpass.


54 Filter Synthesis Using Genesys S/Filter

The History tab for transforms for the series-resonator bandpass is


given in Fig. 8.11. This topology can also be derived by starting with a Series
Element First extraction and applying two shunt Norton transforms to the
shunt resonator of original filter.

Figure 8.11 History tab for the exact series-resonator 90- to 110-MHz bandpass.

In Section 8.1.2, the shunt to series termination-coupling transform


illustrates using a series-coupling capacitor to increase the internal
impedance and inductor values in the parallel-resonator filters. A related
transform uses shunt-coupling capacitors to decrease the internal
impedance and decrease inductor values in a series-resonator bandpass, as
shown in Fig. 8.12. The element values are

n 2 Rs
X shunt = (8.3)
n2 − 1
−1
 1 X 
X series =  + 4shunt2 
 (8.4)
 X shunt n Rs 
where n is the transformer turns ratio, Rs is the source resistance, Xseries is
the reactance of the series element, and Xshunt is the reactance of the
negative shunt element. At the source, n values greater than 1 are used. The
resonator-inductor values in series-resonator filters are reasonable except
for wide bandwidth, and the termination-coupling transform is less
frequently required for this filter.
Bandpass with Resonators 55

Figure 8.12 Series capacitor or inductor-to-shunt-element termination-coupling transform.

8.1.4 Exact Design with Like Coupling Elements


This section describes the exact design of filters that have a topology
identical to the popular, but inexact, resonator filters. Consider the case
where parallel resonators and coupling capacitors are required. Four
resonators with eight TZs are chosen: seven TZs at DC and one TZ at
infinity. This results in greater selectivity below the passband. If greater
selectivity is required above the passband, parallel resonators with coupling
inductors or series resonators with shunt coupling capacitors are used.
Design begins with any extraction sequence with one parallel resonator
and alternating series capacitors and shunt inductors. An example is given
in Fig. 8.13. The extraction sequence is DC ∞ DC DC DC DC DC DC.
Because this filter has an odd number of TZs at both DC and infinity it is
classified as symmetric, and the final design can have equal input and
output terminations, so the transformer and difficult element values are
ultimately resolved. The shunt parallel resonator may be in any other
position, so other extraction sequences are useable.

C2 C3 C4
ZO=50Ω C=4.99pF C=98.6e-3pF C=3.27e-3pF
ZO=50Ω

L2
L=518.13nH
L1 C1
L=17.27nH C=147.41pF L3 L4
L=25883.85nH L=804909.95nH
T1
P=1
S=4.63e-3

Figure 8.13 Initial bandpass with seven TZs at DC and one at infinity.
56 Filter Synthesis Using Genesys S/Filter

Forming the next parallel resonator requires a capacitor in parallel with


L2. A series Norton is applied to C2 with the right element chosen similar to
C1. The negative element left of C2 is absorbed into C1. The value chosen is
not critical as the shunt inductors are equalized as a final step. The
schematic given in Fig. 8.14 results after the application of the first series
Norton transform.

Figure 8.14 Bandpass after application of the first series Norton.

After applying Simplify Circuit, a series Norton is applied to the next


series capacitor. This process is repeated until the final shunt inductor has
a parallel-resonating capacitor.

8.1.5 The Equate All Shunt Ls or Shorted Stubs Transform

Finally, the transform in the Compound Operation folder entitled Equate


All Shunt Ls or Shorted Stubs is applied. This adjusts the impedance level
throughout the filter to equate all shunt inductors. The final filter is given
in Fig. 8.15. If a unity ratio transformer is produced because of digital
precision, it is deleted or absorbed into the load by an S/Filter transform.
This design is exact.

Figure 8.15 Exact capacitor-coupled bandpass with equal inductors.


Bandpass with Resonators 57

8.1.6 Termination-Coupled Bandpass


Again, the small inductor values are difficult to realize with good unloaded
Q at 100 MHz. Therefore, the termination-coupling algorithm is applied as
described in Section 8.3. The resulting filter is given in Fig. 8.16 and the
responses are given in Fig. 8.17. As expected with most TZs at DC, the
selectivity below the passband is high. The termination coupling transforms
are approximate, and this design is not exact. The worst-case return loss is
18 dB rather than 20 dB, and the return loss is not equal ripple.
Optimization corrects the return loss if necessary.

Figure 8.16 Bandpass with like-coupling elements and termination coupling.

Figure 8.17 Responses of the filter with like-coupling elements and termination coupling.
58 Filter Synthesis Using Genesys S/Filter

8.2 Coupled Series-Resonator Filters


The parallel-resonator filter with coupling capacitors has greater selectivity
below the passband than above the passband. If greater selectivity is
desired above the passband, then coupling inductors are used. However,
this increases the quantity of inductors significantly. Another solution for
greater high-side selectivity is series-resonator filters with shunt coupling
capacitors. If design begins with a parallel-resonator top-C coupled
bandpass and the Find Dual transform as described in Section 8.1.3 is used,
the resulting filter has series resonators but shunt coupling inductors. The
design of series-resonator bandpass filters with shunt-coupling capacitors
is covered next.

8.2.1 The Basic Series-Resonator Bandpass


Consider a three-resonator 90 to 110 MHz bandpass (sixth degree) with
two shunt-coupling capacitors. The Series Element First option is selected,
and the canonic form synthesized by S/Filter with the extraction sequence
DC ∞ ∞ ∞ ∞ ∞ is given in Fig. 8.18.

C1
L1 L2 L3
C=7.64pF ZO=50Ω
L=338.83nH L=14.73nH L=0.6nH

ZO=50Ω

C2 C3
C=173.78pF C=4127.52pF

T1
P=1
S=23.73

Figure 8.18 Initial canonic bandpass with one TZ at DC and five at infinity.

Next, a shunt-Norton transform is applied to C2. A transformer turns


ratio is selected equal to the square root of L2/L1, or 0.2085, to equalize the
value of the first and second inductors. A shunt-Norton transform is then
applied to the last shunt capacitor with a turns ratio selected to eliminate
the transformer. Because the filter is naturally symmetric, this equates the
last inductor to the other two. Because of digital precision, the inductor and
capacitor values may be slightly different. They are manually equalized if
desired. The resulting schematic is given in Fig. 8.19.
Bandpass with Resonators 59

C1 C3 C5
L1 L2 L3
C=9.17pF C=12.08pF C=9.19pF
L=338.83nH L=338.83nH L=338.1nH

ZO=50Ω ZO=50Ω

C2 C4
C=36.23pF C=36.27pF

Figure 8.19 Series-resonator bandpass after application of two shunt Norton transforms to the
shunt capacitors.

These series inductor values are reasonable for a 100-MHz filter.


However, as the bandwidth is reduced, the inductor values become larger.
If the inductor values are too large, or if a specific inductor value is desired,
the Termination Coupling: Source: Series C to Shunt C is applied after
swapping L1 with C1. To set the series inductors to 180 nH, the transform is
applied with a turns ratio equal to the square root of 338.83/180, or 1.372.
The load termination coupling transform is then applied to the final series
capacitor with a turn ratio the inverse of the source ratio. Simplifying the
circuit to eliminate the transformers produces the schematic in Fig. 8.20.
L1 C2 L2 C4 L3 C6
ZO=50Ω L=180nH C=23.66pF L=180nH C=22.73pF C=23.74pF ZO=50Ω
L=179.61nH

C1 C3 C5 C7
C=29.9pF C=68.19pF C=68.27pF C=29.9pF

Figure 8.20 Series-resonator bandpass with termination coupling capacitors to reduce and set
the inductors at a specific value.

8.2.2 Tubular Bandpass


Another effective filter topology, popular in radio systems design, is the
tubular bandpass. These inline filters, with connectors on each end, offer
low loss, excellent temperature stability, and excellent stopband
performance at reasonable cost. They typically are used in the 50- to 5000-
MHz frequency range with bandwidths from 4% to 40%. They are
manufactured in a unique form that is described after the schematic is
generated.
60 Filter Synthesis Using Genesys S/Filter

Consider the basic series-resonator shunt-C coupled bandpass in Fig.


8.19. Capacitor C3 is split into two equal capacitors using the Basic
Operation: Split Series Element with the 2 Equal Pieces option. The left
capacitor is then swapped with L2. After swapping L3 and the final series
capacitor, two capacitor tees are formed with the shunt capacitors. The
transform Basic Operations: Pi-Tee-L: Tee: Tee to Pi is applied to each
capacitor Tee. The resulting topology is given in Fig. 8.21.

C2 C5
ZO=50Ω C=3.18pF C=3.19pF ZO=50Ω

L1 L2 L3
L=338.83nH L=338.83nH L=338.1nH

C1 C4
C=4.78pF C=12.58pF
C3 C6
C=12.58pF C=4.79pF

Figure 8.21 Filter topology known as the tubular bandpass created from the series-resonator
bandpass.

Only exact transforms are used to create this topology, so the


responses, as given in Fig. 8.22, are exact.

Figure 8.22 Responses of the sixth-degree tubular bandpass.


Bandpass with Resonators 61

8.2.3 Manufacture of the Tubular Bandpass


The internal structure used to manufacture this filter is given in Fig. 8.23.
The solid black material is metal that forms the outer shield and the
capacitive slugs. The capacitive slugs are isolated from the outer shield with
dielectric material (shown as gray) to form the shunt capacitors. Dielectric
disks isolate adjacent slugs to form the series capacitance. The inductors
are connected to the slugs and input/output connectors and are supported
by the dielectric rod. The inductors are modeled accurately as helical
transmission lines. The shielded inductors do not radiate and have high Q
when constructed with larger diameters. The slugs shield the inductors
from coupling, and the small housing size reduces package waveguide
modes, resulting in excellent stopband performance. Proper selection of
metallic and dielectric materials results in excellent temperature
performance. Tuning is accomplished by compressing or spreading the
inductors though small access holes in the shield. An outer shield covers
the tuning holes.

Dielectric support rod


Dielectric spacers
Metal slugs and housing
Wire helical inductor

Figure 8.23 Physical structure of the tubular bandpass filter.

8.2.4 Generalized Series-Resonator Bandpass


Fig. 8.24 shows the initial canonic filter for a 90- to 110-MHz generalized
shunt-C coupled bandpass of the sixth degree with 40-dB minimum
rejection in the lower stopband and 30-dB minimum rejection in the upper
stopband. The extraction sequence is 134 MHz ∞ 60 MHz DC.
62 Filter Synthesis Using Genesys S/Filter

L3 C2
ZO=50Ω C=1.57pF ZO=50Ω
L=1624.47nH

L1
L=135.68nH C4
C=23.93pF
L2 L4
L=233.42nH L=196.94nH

T1
C1 C3 P=1
C=6.04pF C=35.73pF S=1.39

Figure 8.24 Generalized bandpass of the sixth degree with FTZs below and above the
passband.

The element values are reasonable, although the value of L3 is


somewhat large. The transformer can be removed and the response almost
recovered by optimization. To create a filter with all series resonators, the
transformer is removed by applying a shunt-Norton transform to C3 and L4
with the Calculate n to Remove Transformer option. This transforms the
final branch from a series capacitor to a series L-C and it reduces L3
slightly. At this point, a shunt-Norton transform applied to C1 and L2 adds
only one more capacitor, significantly reduces L3, and results in all series
resonators. This choice is particularly attractive if a manufacturing
technology is available with high-Q resonators that are only series form.
The resulting schematic is given in Fig. 8.25 and the exact responses are
given in Fig. 8.26.

L1 C1 L3 C3 L5 C5
L=252.39nH C=12.09pF L=260.42nH C=9.52pF L=242.28nH C=8.71pF

ZO=50Ω ZO=50Ω

L2 L4
L=116.71nH L=136.58nH

C2 C4
C=12.09pF C=51.52pF

Figure 8.25 Generalized bandpass converted to series resonators.


Bandpass with Resonators 63

Figure 8.26 Responses of the generalized series-resonator bandpass.

8.2.5 Tunable Constant-Bandwidth Bandpass


Because the parallel-resonator bandpass filter with capacitive coupling
elements is popular, it is tempting to use this topology for the design of
tunable bandpass filters. Because each resonator has a shunt inductor,
replacing the shunt capacitors with back-to-back varactor diodes requires a
minimum of decoupling elements to apply the tuning voltage. An example
schematic for a 100-MHz bandpass with 10-MHz bandwidth is given in Fig.
8.27a, where capacitors C2, C4, and C6 would be replaced with back-to-back
varactors. The responses of this filter with these capacitors adjusted to tune
the filter to 75, 100, and 125 MHz are given on the left in Fig. 8.28. The
filter is synthesized for 100 MHz, and the response is as expected at that
frequency. However, tuning the capacitors to shift resonance does not
result in appropriate values for the other elements. As a result, the
responses are severely degraded with tuning.
The primary difficulty is that as the frequency is tuned lower, for
constant bandwidth, the coupling should be increased. At 100 MHz, the
bandwidth is 10%. At 75 MHz, 10% bandwidth is only 7.5 MHz. To achieve
constant bandwidth, the percentage bandwidth at 75 MHz must be 13.3%.
The reactance of the capacitive-coupling elements increases as the
frequency is reduced, which actually reduces rather than increases the
bandwidth. The inverse situation exists when the frequency is tuned higher.
64 Filter Synthesis Using Genesys S/Filter

Figure 8.27 Potential tunable bandpass filter topologies.

The situation is significantly improved by utilizing a filter with


inductive coupling elements as shown in Fig. 8.27b. In this case, as the
frequency is tuned lower, the reduced inductive reactance increases the
coupling and increases the percentage bandwidth. Responses are given on
the right in Fig. 8.28. This filter has the disadvantage of increased inductor
count, but it has much improved responses. The dual of this filter, a series-
resonator bandpass with shunt coupling elements also tunes well and
returns the inductor count to three, but applying a tuning voltage to the
series varactors requires additional coupling and decoupling elements.
Bandpass with Resonators 65

Figure 8.28 Responses of the parallel-resonator bandpass filter with tuning with capacitive
coupling elements (left) and with inductive coupling elements (right).

As can be seen on the right in Fig. 8.28, the bandwidth compensation is


not perfect, and the bandwidth is still too narrow as the frequency is tuned
lower. Placing capacitors in parallel with the coupling inductors, as shown
in Fig. 8.27c, increases the branch reactance as resonance is approached.
This further compensates the bandwidth. This filter could be synthesized
using a FTZ for each coupling branch. However, the correct frequency of
the FTZs is unknown. In this case, the filter is synthesized as shown in Fig.
8.27b and small capacitors are manually added in parallel with the coupling
inductors. Next, two additional instances of the original filter are copied
into the same schematic. These additional filters used ports three and four
and ports five and six. The tuning capacitors are optimized individually for
each filter, and all remaining elements are optimized to the same values
using variables. The optimization goals are given in Fig. 8.29.
The final values are given in Fig. 8.27c and the responses are given in
Fig. 8.30. Although not ideal, the bandwidth is relatively constant and a
better than 16-dB return loss is achieved with 60% tuning bandwidth. The
achievable performance degrades with increased tuning bandwidth.
66 Filter Synthesis Using Genesys S/Filter

Figure 8.29 Goals for the simultaneous optimization of the tunable filter at three frequencies.

Figure 8.30 Responses of the bandwidth-compensated and optimized tunable bandpass.


Bandpass with Resonators 67

Reference
[1] G. Matthaei, L. Young, and E.M.T. Jones, Microwave Filters, Impedance-
Matching Networks, and Coupling Structures, Artech House, Dedham, MA, 1980.
9 TEM-Mode Resonators
With narrow bandpass filters, insertion loss often dominates the design
process. The solution requires higher unloaded (element) Q. Element Q is
typically dominated by metal conductivity. Higher volume increases
element Q. For a given volume, solenoid inductors typically have a higher
element Q than distributed transmission line elements (TLEs). However,
parasitics limit the allowable physical size of wire-wound inductors.
Transmission lines have higher unloaded Q than inductors when they are
physically larger [1].
Hybrid filters combine L-C elements and TLEs. These filters are readily
designed using the Lumped to Distributed Equivalent group of transforms
in S/Filter. Lumped and distributed equivalents and their theory are
covered in HF Filter Design and Computer Simulation [2]. These
transforms are approximate, but over the narrow bandwidth typically
involved in this filter class, they are sufficiently accurate. In this chapter,
the parallel-resonators discussed in Chapter 8 are transformed into TLE
resonators in order to provide higher resonator unloaded Q.

9.1 Filter Insertion Loss


Well below the cutoff frequency, the dissipative insertion loss of all-pole
lowpass filters, due to finite unloaded Q, is given approximately by [2]
n= N
4.34
I .L.(dB) =
Qu − ave ∑g
n =1
n (9.1)

where Qu-ave is the average unloaded Q of the elements, N is the prototype


order, and gn are the lowpass prototype g-values of the transfer
approximation from prototype tables. The sum of the g-value is very
roughly equal to N+1.5 for Butterworth and low-ripple Chebyshev
approximations. The sum is 6.47 for a fifth-order Butterworth and 6.49 for
a fifth-order 0.0432-dB ripple Chebyshev. For a fifth-order lowpass with an
average element unloaded Q of 50, the dissipative insertion loss is roughly
0.56 dB.
The expression for the dissipative insertion loss of an all-pole bandpass
filter is similar, except the loss is multiplied by the filter loaded Q (Ql). It is
important to recognize the difference between loaded and unloaded Q.
Loaded Q is a design parameter dependent on the filter bandwidth. The
loaded Q is

69
70 Filter Synthesis Using Genesys S/Filter

f0
Ql = (9.2)
BW
At midband, the dissipative insertion loss of all-pole bandpass filters is
n=N
4.34Ql
IL( dB ) =
Qu − ave ∑g
n =1
n (9.3)

A fifth-order bandpass with a 4% bandwidth has a loaded Q of 1/0.04,


or 25. With the same average unloaded Q as before, the midband insertion
loss is 14 dB. Decreasing bandwidth requires ever-higher element unloaded
Q to avoid excessive insertion loss.

9.2 Filter Using 50-Ohm Coaxial Resonators


This example is a 900-MHz bandpass filter with capacitive coupling and
resonators constructed from shorted sections of 50-ohm coaxial line with
air dielectric. Consider the following filter specifications:
1) Nominal termination resistances: 50 ohms
2) Maximum passband return loss: 20 dB
3) Insertion loss midband maximum: 2.5 dB
3) Lower passband cutoff frequency: 895 MHz
4) Upper passband cutoff frequency: 905 dB
5) Minimum rejection <880 MHz: 60 dB
6) Minimum rejection >920 MHz: 60 dB
The quantity of TZs at infinity is set to one. Using the initial L-C filter
designed in S/Filter, the quantity of TZs at DC is increased until the
specification is met with an odd quantity of TZs. Nine TZs are required,
which is five resonators. The inductor Q is adjusted until the midband
insertion loss is just met. The required inductor Q is 505.

9.2.1 Lumped to Distributed Equivalents


The equivalence between a parallel L-C resonator and a shorted
transmission line stub is [2]
4Z 0
L= (9.4)
πω
where the transmission line is 900 long at the L-C resonant frequency.
For 50-ohm coaxial line, the required inductance is 11.26 nH. A five-
resonator filter is designed using the series Norton transform as described
in Sections 8.5 and 8.6. The resulting filter is given in Fig. 9.1.
REM-Mode Resonators 71

Figure 9.1 Five-resonator narrow bandpass filter to be converted to a coaxial resonator filter.

Next, the resonator inductors are scaled up to 11.26 nH by using the


transform Termination Coupling: Source: Shunt C to Series C at 900 MHz
and n equal to 0.09476, which is the inverse of the square root of the target
inductor value and the initial inductor value. The Load: Shunt C to Series C
transform is then applied with the inverse of the source value for n, or
10.55. The result is given in Fig. 9.2.

Figure 9.2 Narrow bandpass after the termination coupling transforms to increase the value of
the inductors.

The values of the internal coupling capacitors are very small. If the
coaxial resonators are to be coupled using capacitive probes, these values
are appropriate. If the coaxial resonators are to be mounted on a printed-
circuit boad (PCB), the coupling capacitors could be printed as interdigital
microstrip gaps. If chip capacitors are to be used, the values are
unreasonable. This is addressed in the next section covering microstrip
resonators.
The next step is converting the parallel L-C resonators to coaxial
resonators using the Lumped to Distributed Equivalents: Parallel LC to
Ground to Grounded Stub transform. The result is given in Fig. 9.3. The
transform in S/Filter uses a transform frequency equal to the resonant
frequency of the parallel L-C network. This is appropriate for many
applications. However, this results in a slightly different characteristic
impedance for each distributed resonator. Here, 50-ohm characteristic
impedance is desired for each resonator. This is resolved by manually
changing each resonator to 50 ohms and optimizing the response. The
72 Filter Synthesis Using Genesys S/Filter

parallel L-C to grounded stub algorithm is approximate, so optimization is


good practice anyway.

Figure 9.3 Narrow bandpass with parallel L-C resonators transformed to coaxial resonators.

Given in Fig. 9.4 is the filter after setting each resonator to 50 ohms
and optimizing the length of the resonators.

Figure 9.4 Narrow bandpass after setting each resonator to 50 ohms and optimizing the
resonator lengths.

9.2.2 The Convert Using Advanced Tline Routine


The final step is converting the grounded stubs into physical TLEs. This is
accomplished using the Convert Using Advanced TLine routine in the
Genesys environment. All five grounded stubs are selected and the Convert
Using Advanced TLine routine is selected in the Schematic tab of Genesys.
The dialog in Fig. 9.5 is displayed.
REM-Mode Resonators 73

Figure 9.5 Convert Using Advanced TLine dialog.

This routine coverts the grounded stub transmission line model with
electrical parameters into the selected physical transmission line model, in
this case coax, using the previously defined substrate material Air. The Air
material is defined by adding a Substrate in the workspace tree of Genesys.
The dielectric in this case is air. The coax outer diameter determines the
line loss and therefore the unloaded Q. Earlier, S/Filter was used to
determine that the required unloaded Q required to achieve a dissipative
insertion loss of 2.5 dB is 505. The unloaded Q of a quarter wave coaxial
resonator due to conductor loss is
b ln (b a )
Qc = 1.513 × 10 4 f ( MHZ ) (9.5)
1+ b a

where b is the outer diameter in meters, a is the center conductor diameter


in meters, and f(MHz) is the frequency in megahertz. For a 50-ohm coaxial
line with air dielectric, b/a is 2.3. To achieve an unloaded Q of 505 requires
coax with an outer conductor of 4.4 mm. For margin, an outer conductor
diameter of 5 mm is specified in the Convert Using Advanced TLine dialog.
The resulting responses are given in Fig. 9.7.
74 Filter Synthesis Using Genesys S/Filter

Figure 9.6 Coaxial bandpass with electrical line models converted to physical line models.

Figure 9.7 Responses of the 50–ohm coaxial resonator bandpass.

9.3 Generalized Bandpass Using Ceramic Resonators


The true power of direct synthesis lies in the ability to place transmission
zeros at arbitrary frequencies. In this example, ceramic-loaded coaxial
resonators are used with the placement of one FTZ above the passband to
increase selectivity close to the passband. A 950- to 1000-MHz bandpass is
required with attenuation of at least 50 dB below 840 MHz and above 1050
MHz. TransTech 1000 series high-profile (HP), ceramic-loaded, square-
coaxial resonators are used. The characteristic impedance of these
resonators is 25.3 ohms and the unloaded Q at 975 MHz is just over 900.
REM-Mode Resonators 75

A four-resonator filter is eighth degree. Each FTZ adds two degrees,


and each nonfinite TZ adds one degree, so one finite and six nonfinite TZs
are specified. The FTZ is manually adjusted to achieve the rejection
requirement at 1050 MHz. The initial S/Filter specification screen is given
in Fig. 9.8 and the initial L-C filter with the extraction DC 1050 ∞ DC DC
DC DC is given in Fig. 9.9.

Figure 9.8 Specification tab for the generalized bandpass using ceramic-loaded coaxial
resonators.

C2
C=6.74pF C4 C5
ZO=50Ω C=9.26e-3pF C=20.1e-6pF ZO=50Ω

L2
L=3.41nH
L4
C1 C3 L=1.33e+6nH
C=58.63pF C=0.96pF
L1
L=0.45nH L3
L=2888.85nH

T1
P=1
S=581e-6

Figure 9.9 Initial L-C filter for the proposed generalized ceramic-loaded coaxial resonator.

9.3.1 Creating Parallel Resonators


The first step is converting C3, L3, and L4 to shunt resonators. A series
Norton is applied to C2 and L2. It is valid to apply the Norton transform to
not only individual reactors but to a parallel or series pair as well. A turns
ratio greater than one is chosen so that the negative parallel L-C elements
76 Filter Synthesis Using Genesys S/Filter

are absorbed into L1 and C1 and positive L-C values occur on the right. The
turns ratio value is not critical as the shunt inductors are equalized later.
Here, n equal to two is entered. Series Norton transforms are then applied
to the two remaining series capacitors to place capacitance in parallel with
the remaining shunt inductors. Again, n is not critical, and a value of 10 is
specified. Next, Remove Existing Transformer is applied. Finally, the
transform Compound Operations: Equate All Shunt Ls or Shorted Stubs is
applied. The schematic with parallel shunt resonators is given in Fig. 9.10.

Figure 9.10 Generalized bandpass before the termination coupling and coaxial-resonator
transforms.

9.3.2 Shifting the Internal Impedance Level


Next, the source and load termination coupling transforms are applied to
shift the impedance level up so that the shunt inductors are 5.258 nH to
satisfy Eq. 9.4 and the 25.3-ohm characteristic impedance of the Trans
Tech resonators.

Figure 9.11 Generalized bandpass after impedance scaling by termination coupling to


increase the shunt inductor values.

The two internal series coupling capacitors are a small value. If this is
manageable, then the steps in the next section are skipped.
REM-Mode Resonators 77

9.3.3 The Pi to Tee Transform: Increasing Coupling Caps


If coupling-capacitance values are excessively small, the following
transforms are applied. Inductor L4 is swapped with capacitor C6. Both C4
and C6 are then split into additional capacitors with 0.5-pF capacitors on
each side of C5 by using the Basic Operation: Split-Shunt Element
transform. Then the Basic Operation: Pi to Tee transform is applied to this
pi of capacitors. The capacitor value is increased to 1.03 pF. The same
process is applied to capacitor C7, creating another Tee with a capacitor
value of 1.08 pF.

9.3.4 Converting the Parallel L-C to Coaxial Resonators


The parallel L-C resonators are next converted to transmission lines using
the Lumped to Distributed Equivalents: Parallel LC to Ground to Ground
Stub transform. Because S/Filter adjusts w in Eq. 9.4 to the frequency of
each parallel L-C, the characteristic impedance of each line is slightly
different from 25.3 ohms. Because the termination coupling and parallel L-
C to grounded stub routines are approximate, the final step is optimization.
Therefore, each ground stub is manually changed to a 25.3-ohm
characteristic impedance.

9.3.5 Optimizing the Values


The last step is optimization to correct the response after application of
approximate transforms and setting each coaxial resonator at 25.3 ohms.
Selected for optimization are the resonator line lengths, the FTZ branch
values of L1 and C2, the two shunt coupling capacitors, and the two
termination coupling capacitors. Because L1 is realized as a wound
inductor, its Q is set at 100. The unloaded Q of the resonators is
automatically determined by the conductivity, dielectric, and dimensional
parameters of the square coaxial resonator model. The final filter after
optimization is given in Fig. 9.12 and the responses are in Fig. 9.13.
78 Filter Synthesis Using Genesys S/Filter

Figure 9.12 Final schematic of the generalized ceramic-loaded coaxial resonator bandpass
with one FTZ.

Figure 9.13 Responses of the ceramic-loaded coaxial resonator bandpass.

9.4 Ceramic Bandpass with Two FTZs


In this example, the previous bandpass is extended to a filter with FTZs
both above and below the passband, resulting in steeper transitions on both
sides of the passband. Again, the target filter contains four ceramic
resonators. In this case, the specification begins with three TZs at DC and
one at infinity. The FTZs are manually tuned to 899 and 1064 MHz to
achieve the desired Amin of 50 dB. The initial schematic with an extraction
sequence of DC 1064 ∞ DC 899 DC is given in Fig. 9.14a.
REM-Mode Resonators 79

Figure 9.14 Transform steps for the four-section ceramic-resonator bandpass with FTZs
above and below the passband.
80 Filter Synthesis Using Genesys S/Filter

The following transforms complete the design.


1) A Norton: Series Transform is applied to C2 and L2 with a specified
transformer turns ratio of 2.0 followed by a Simplify Schematic. This
effectively places an inductor in parallel with the second shunt
capacitor.
2) A Norton: Series Transform with a transformer turns ratio of 20 is
applied to the series capacitor now designated C4 followed by a
Simplify Schematic. This places a capacitor in parallel with the third
shunt inductor.
3) A Norton: Series Transform is applied L5 and C6 with the option to
calculate n to remove the existing transformer. This places a capacitor
in parallel with the final shunt L.
4) The Compound Operations: Equate All Shunt Ls or Shorted Stubs is
applied. This filter is symmetric so if a transformer remains, the turns
ratio is near unity and the transformer is deleted. The resulting
schematic is given in Fig. 9.14b.
5) Basic Operations: Swap Element is applied to L1 to place C1 adjacent
to the input.
6) Termination Coupling: Source: Shunt C to Series C is applied to C1
and the complementary transform is applied at the load. Next,
Simplify Schematic eliminates the transformers. The resulting
schematic is given in Fig. 9.14c.
7) Lumped to Distributed Equivalents: Parallel LC to Ground to
Grounded Stub is applied to each parallel resonator. This creates
grounded stubs with electrical parameters.
The four-grounded stubs are converted to coaxial resonators with
square outer conductors using Advanced TLine. To realize Trans-Tech
high-profile resonators a substrate is created with a relative dielectric
constant of 10.5. Because S/Filter adjusts w in Eq. 9.4 to the frequency of
each parallel L-C, the characteristic impedance of each line is slightly
different from 25.3 ohms. Because the termination coupling and parallel L-
C to grounded stub routines are approximate, the final step is optimization.
Therefore, the outer-side dimension and inner diameter of each resonator
are manually changed to 12 mm and 3.3 mm, respectively, before
optimization of the resonator lengths. The final schematic is given in Fig.
9.14d and the responses are given in Fig. 9.15.
REM-Mode Resonators 81

Figure 9.15 Initial synthesized ideal lumped bandpass with two FTZs (right) and after
transformation to coaxial resonators (left).

References
[1] R. Rhea, “Technology Enables New Components,” Microwave Journal,
November 2006.
[2] R. Rhea, HF Filter Design and Computer Simulation, Noble
Publishing(SciTech), Raleigh, NC, 1994.
[3] G. Matthaei, L. Young, and E.M.T. Jones, Microwave Filters, Impedance-
Matching Networks, and Coupling Structures, Artech House, Dedham, MA, 1980.
10 Piezoelectric Devices
Quartz and ceramic bulk-mode vibrational resonators offer extremely
high-unloaded Q and excellent temperature and long-term stability. These
attributes are critical for very narrow bandwidth applications. However, the
equivalent electrical model parameters for these mechanical-vibration
devices pose unique issues. This chapter illustrates bandpass filters that
utilize these devices and includes a larger number of examples than
previous chapters because of the unique difficulties in synthesizing filters
with piezoelectric devices.

10.1 Quartz-Crystal Device Model


This section describes and models bulk-mode quartz piezoelectric
resonators, often referred to as quartz crystals.

10.1.1 Physical Form of the Quartz Crystal


Fig. 10.1 shows diagrams and photographs of the physical form of a typical
quartz crystal. The fundamental and overtone shear modes are illustrated
in the diagram.

Figure 10.1 Diagram of quartz-crystal resonator illustrating electrodes on a quartz blank, and
the fundamental and overtone shear vibrational modes (left), and a photograph [1] of the
resonator and a metallic housing (right).

83
84 Filter Synthesis Using Genesys S/Filter

The fact that mechanical vibration in quartz, and certain other


materials, produces an electrical signal is referred to as piezoelectricity. It is
reciprocal in that an electrical signal can induce a mechanical change. Even
a small mechanical displacement occurring at such a high frequency, and
the rigidity of the quartz, results in tremendous energy storage.

10.1.2 Insertion Response of a Quartz Crystal


Given in Fig. 10.2 is the transmission amplitude response swept in the
frequency range near the fundamental-mode of a 9.6-MHz quartz crystal,
inserted in series in a 50-ohm measurement system. The vertical scale is 10
dB/div and the horizontal scale is 10 kHz/div. The peak of transmission on
the left at 9.600 MHz is the series-resonant mode of the quartz crystal. The
null to the right of this peak at 9.620 MHz is the parallel-resonate mode of
the quartz crystal. The remaining inflections in the response to the right are
spurious modes. This entire pattern of series, parallel, and spurious modes
repeats at odd overtones of the fundamental frequency. They are referred to
as overtones rather than harmonics because they are approximately, but
not precisely, odd integer multiples of the fundamental frequency.

Figure 10.2 Transmission amplitude response of the fundamental-mode of a 9.6-MHz quartz


crystal.

10.1.3 Modeling the Quartz Crystal


An electrical model for the fundamental series and parallel-mode activity is
given in Fig. 10.3. The capacitor C0 results from the metallic electrodes on
the quartz blank. It is referred to as the static capacitance because it is not
derived from vibration. It is measured with a low-frequency capacitance
meter.
Piezoelectric Devices 85

Lm Cm Rm

Co

Figure 10.3 Electrical model for quartz-crystal resonators.

The parameters in the series R-L-C branch are referred to as motional


parameters because they represent the electrical equivalent of the
piezoelectric effect. Because the energy storage is tremendous, the value of
the motional inductance is extremely large, typically many milliHenries and
potentially as large as a Henry. Parasitics would preclude realizing these
inductance values as wire-wound inductors. The motional capacitance
resonates with the motional inductance at the series-resonant frequency of
the quartz crystal. The motional resistance ranges from approximately 500
ohms at 1 MHz downward to as low as 10 ohms at 20 MHz, and somewhat
higher for overtone modes. The unloaded Q is extremely high, not because
of low resistance, but because of extremely high reactance. More exhaustive
modeling of piezoelectric resonators is given in [1].
Increasing the resonant frequency requires decreasing the thickness of
the quartz blank. Above 20 to 30 MHz, the blank becomes so thin that it
becomes too fragile for practical application. Operation through the ninth
overtone extends the useful frequency range to about 200 MHz. A special
manufacturing technique can produce inverted-mesa resonators that
operate to about 600 MHz. Nevertheless, quartz-crystal filters are typically
confined to 140 MHz or less, and fundamental mode filters up to 20 MHz
are the most common.

10.1.4 Calculating Model Parameters from the Response


The four parameters of the quartz-crystal model in Fig. 10.3 are calculated
from the measured static capacitance and the quartz-crystal insertion
response. The static capacitance is measured with a low-frequency
capacitance meter. Then the motional capacitance is given by

 f  2 
p
C m = C0   − 1 (10.1)
 s 
f 
 
and
1
Lm = (10.2)
(2πf s )2 Cm
86 Filter Synthesis Using Genesys S/Filter

Then assuming the reactance of C0 is much greater than Rm

 IL 
Rm = 2Z 0 10 20 − 1 (10.3)
 
where Z0 is the characteristic impedance of the measurement system and IL
is the decibel insertion loss at the series-resonant transmission peak.

10.1.5 The Quartz-Crystal Model and Filter Design


The characteristics of wire-wound inductors are far different from the
quartz crystal. A design that mixes quartz and L-C resonators is dominated
by the poorer unloaded Q and stability of the inductors. Capacitor unloaded
Q and stability in the frequency range of quartz-crystal filters is often
acceptable. Therefore, quartz-crystal filter design typically includes the
piezoelectric resonators and capacitors. An exception is when discrete
inductors are used in quartz-crystal filters as described in Section 10.3.

10.2 Quartz-Crystal Filter Approximate Design


The extreme values of the motional branch elements, and the presence of
the static capacitance, require that the filter topology comply with the
model topology. This results in a more involved synthesis and transform
process. For certain applications, the design procedure can ignore the static
capacitance. The static capacitance is added after synthesis and before
simulation to assess the effect on the response and make corrections if
necessary. This design approach is simpler and is illustrated first.
The measured static capacitance of the 9.6 MHz quartz crystal in Fig.
10.2 is 5.45 pF. From Eq. 10.1, the motional capacitance is 22.73 fF and
from Eq. 10.2 the motional inductance is 12.0920062 mH. The insertion
loss at series resonance in the 50-ohm system is 0.95 dB. The reactance of
C0 at 9.6 MHz is 3042 ohms, significantly higher than Rm, so Eq. 10.3 is
valid and Rm is 11.56 ohms. The unloaded Q of the crystal resonator is the
motional reactance divided by the motional resistance, which is 63,100.
These parameters are the basis for the design of a 9.600- to 9.602-MHz
four-section quartz crystal bandpass filter. This is a loaded Q of
9.601/0.002, or 4,800.
The series resonator of the motional branch is the defining element, so
the static capacitance is temporarily ignored. Therefore, a series-resonator
filter topology with shunt coupling capacitors is selected. Design begins
with the S/Filter Specification tab in Fig. 10.4. The motional resistance is
modeled by specifying the unloaded Q of the motional branch as the
unloaded Q of the inductor. A maximally flat transmission-amplitude
approximation (Butterworth) is used with cutoffs defined as 3.01-dB
attenuation for the lossless case.
Piezoelectric Devices 87

Figure 10.4 Specification tab for the approximate design four-section quartz-crystal bandpass.

The initial synthesis with a series element first and the extraction
sequence ∞ DC ∞ ∞ ∞ ∞ ∞ ∞ is given in Fig. 10.5. The topology is identical
to the shunt-C coupled series-resonator bandpass filter introduced in
Section 8.2, and the design proceeds as before. Shunt-Norton transforms
are applied to the shunt capacitors to equate the series inductors and
remove the transformer. The element values are displayed with eight-digit
precision because of the extremely high Q of the design. The high stability
of quartz justifies this expectation of precision.

C1
ZO=50Ω C=0.0902371pF ZO=50Ω

L1 L2 L3 L4
L=3.0452454e+6nH L=0.0934436nH L=1.1876835e-9nH L=36.444144e-18nH

C2 C3 C4
C=2.9407491e+6pF C=231.36976e+12pF C=7.5401426e+21pF

T1
P=1
S=289.06617e+9

Figure 10.5 Initial synthesis of the four-section quartz-crystal bandpass.

The results after the three shunt-Norton transforms are given in Fig.
10.6. The value of the series inductors is approximately 3 mH, far too large
to realize as wire-wound inductors but consistent with quartz crystals.
88 Filter Synthesis Using Genesys S/Filter

Figure 10.6 Quartz-crystal bandpass after conversion to series resonators.

The next step is shifting the impedance level higher by the ratio of the
quartz-resonator motional inductance and these schematic values. In this
case, the termination resistance is increased to 198.5 ohms. The results are
given in Fig. 10.7.

Figure 10.7 Quartz-crystal bandpass after increasing termination resistance to 198.5 ohms.

At this low frequency, the higher impedance can often be


accommodated by higher-impedance amplifiers. Alternatively, a simple
matching network or a 4-to-1 impedance transformer could step 50 ohms
up to 200 ohms. Matching network design using S/Filter is covered in
Chapter 12.
The first and last resonators are loaded with the reactance of one
shunt-coupling capacitor while the internal resonators are loaded by two
coupling capacitors. This results in different resonant frequencies for the
two groups of resonators. The resonant frequency of each series inductor
and capacitor pair are given in Fig. 10.7. The filter could be built with two
different quartz crystals. Alternatively, provided the difference in resonant
frequency is small, a single quartz-crystal with a series resonant frequency
of 9599.618 kHz could be used with a capacitor in series with resonators
one and four to increase those branches’ frequency to 9600.159 kHz. This is
illustrated in Fig. 10.8. These frequency-shifting capacitors are then tuned
to achieve a maximally flat response by observing the return loss. The series
L-C networks are replaced with the quartz-crystal model, which includes
the motional resistance and static capacitance. The termination resistance
and capacitor values are set to standard values.
Piezoelectric Devices 89

X1 X2 X3 X4
R=11.6Ω R=11.6Ω R=11.6Ω R=11.6Ω
L=12.092mH L=12.092mH L=12.092mH L=12.092mH
Cm=.02273182pF C1 Cm=.02273182pF Cm=.02273182pF Cm=.02273182pF C7
Co=5.45pF C=200pF Co=5.45pF Co=5.45pF Co=5.45pF C=200pF

ZO=200Ω ZO=200Ω

C2 C4 C6
C=130pF C=200pF C=130pF

Figure 10.8 Quartz-crystal bandpass with standard values and identical quartz resonators.

The response of the 9.6- to 9.602-quartz-crystal filter is given in Fig.


10.9. The static capacitance adds four finite-frequency TZs at 9.62 kHz that
are not involved in the initial synthesis. Their effect is to increase the
selectivity above the passband and slightly reduce the bandwidth. The
narrow bandwidth of this filter could not be achieved with L-C resonators
because extremely high-unloaded Q is required in the resonators. The
loaded Q of the shunt coupling elements is not critical.

Figure 10.9 Responses of the four-section quartz crystal 9.6- to 9.602-MHz filter designed by
approximate methods.
90 Filter Synthesis Using Genesys S/Filter

10.3 Nulling the Static Capacitance


It is parallel resonance in the quartz crystal and the resulting TZ in the
response above the passband that causes the simplified design procedure in
Section 10.2 to be approximate. It is the extreme values of the motional
inductance and motional capacitance that causes very high unloaded Q in
the quartz crystal. The static capacitance is a moderate value and it can be
nullified with a moderate inductor value. For the 9.6 MHz quartz-crystal
example, the 5.45-pF static capacitance is nulled with a 50.326-uH inductor
in parallel with the quartz crystal. This value of inductance is practical at
9.6 MHz using inductors wound on cores with permeability greater than
one. The schematic of the previous quartz-crystal bandpass with nulling
inductors is given in Fig. 10.10.

Figure 10.10 A 9.6-MHz quartz-crystal filter with nulling inductors.

The responses of the quartz-crystal filter with nulling inductors is given


in Fig. 10.11. Compared to the quartz-crystal filter response without nulling
in Fig. 10.9, the response is more symmetric due to a lack of TZs above the
passband. Notice the bandwidth is now equal to the design value of 2 kHz
because the parallel resonances have not constricted the upper side of the
passband.
The responses in Fig. 10.11 are a composite of 100 Monte Carlo runs
with a Gaussian distribution of 5% standard deviation for the four nulling
inductors, and with an unloaded inductor Q of 100. The result illustrates
the relatively low sensitivity of the nulling inductors despite the loaded Q of
4800 for the filter.
Piezoelectric Devices 91

Figure 10.11 Responses of the quartz-crystal bandpass with inductors to null the static
capacitance. Monte Carlo analysis displays the results of multiple simulations with 5%
tolerance-nulling inductors.

10.4 Design of a Lower-Sideband Crystal Filter


As the bandwidth is increased, the frequency of the quartz-crystal parallel-
resonant mode approaches the upper cutoff of the filter. This results in
increasing error when ignoring the static capacitance during the design
process. In this section, the design of a 5-kHz bandwidth lower-sideband
crystal filter is illustrated using a crystal similar to the 9.6-MHz quartz
crystal modeled earlier in the chapter. The design considers the static
capacitance, which then improves the accuracy of the design.
The specified termination impedance is 50 ohms. The final termination
resistance is much higher and this is dealt with in a later step. The specified
unloaded Q is specified as one million and resonator loss is dealt with later.
The desired passband is 9.600 to 9.605 MHz. The desired stopband is 80
dB attenuation from 9.607 to 9.612 MHz. To create a topology consistent
with the quartz-crystal resonator model, no TZs are specified at DC or
infinity. All extractions are inexact because nonfinite TZs are unavailable
for extraction after the FTZs. This results in unequal passband ripple,
which is dealt with later in the design. The maximum allowable passband
ripple, 0.25 dB, is entered, and FTZs are added and adjusted until the
stopband attenuation requirement is achieved. Four FTZs result in four
quartz-crystal resonators. The S/Filter Specification tab is given in Fig.
10.12, the initial synthesis solution is given in Fig. 10.13, and the responses
are given in Fig. 10.14.
92 Filter Synthesis Using Genesys S/Filter

Figure 10.12 Specification tab for the four-section lower-sideband moderate-bandwidth


quartz-crystal filter.

Figure 10.13 Schematic of the four-section quartz-crystal filter as initially synthesized.

Crystal filters are often employed in communication receivers and


transmitters. The narrow bandwidth of amplitude-modulated voice,
continuous-wave (CW) channels, and low-data rate systems, combined with
the desirability of high IF frequencies for image rejection and IM
performance, require quartz-crystal filters. This filter is referred to as a
lower sideband because the greater rejection is above the passband in the
regions of the carrier and upper sideband.
Piezoelectric Devices 93

Figure 10.14 Responses of the four-section lower-sideband quartz-crystal filter.

The first series inductor in Fig. 10.13 is used to create a shunt


termination coupling inductor using the Termination Coupling: Source:
Series L to Shunt L transform. A transform frequency at mid-passband, or
9.6025 MHz is specified, followed by a turns ratio, n, of two. This value is
not critical but must be within the specified range. The Simplify Circuit
transform is applied followed by the Compound Operations: Find Dual of
Schematic transform. The resulting schematic is given in Fig. 10.15.

Figure 10.15 Schematic of the lower-sideband filter after the termination coupling and final
dual transforms are applied.

The next step creates capacitors in series with the first parallel-
resonant branch by applying a shunt-Norton transform to capacitor C2 in
Fig. 10.15. Any transform turns ratio is acceptable that results in a left
series capacitor that can be absorbed in capacitor C1. One option is selecting
94 Filter Synthesis Using Genesys S/Filter

n so that this negative capacitor is exactly absorbed in C1 resulting in


eliminating one capacitor. That value of n is just greater than
1
nnull = (10.4)
C2
1+
C1

which in this case is 0.88789169. If n is just slightly larger than this value,
the resulting capacitor value is extremely large and is deleted
The left side part of the schematic resulting from this transform is given
in Fig. 10.16. The form of the branch with L1, C3 and C4 is referred to as a
dipole. With appropriate values, it is exactly equivalent to the quartz-crystal
model dipole given in Fig. 10.3.
C4
C=207975.150197pF

C1
C=2.2843906e+9pF
C3
C=169.928106pF

L1
ZO=50Ω L=1.319633nH

C2 C5
C=21.455715pF C=810.530659pF

Figure 10.16 Left-side portion of the schematic resulting from the application of the shunt-
Norton transform.

Unfortunately, this dipole transform is not included in the many


S/Filter transforms. The transform is given in Fig. 10.17 and Eqs. 10.5
through 10.10.

Figure 10.17 Synthesized and quartz-crystal dipoles with transform designators.

C1 = C4 1 + C4  (10.5)
 C3 
Piezoelectric Devices 95

C 2 = C3 + C 4 (10.6)

 C  
2
L1 = L2  3   (10.7)
 C3 + C4  
 
C2
C3 = (10.8)
 C1 
1 + C 
 
C2
C4 = (10.9)
1 + C 2 
 C1 

2
 (C + C2 ) 
L2 = L1  1  (10.10)
 C2 
Substituting the values of the 9.6-MHz quartz-crystal resonator in
Section 10.2, the required values of the left dipole in Fig. 10.16 are
C1=1315.758 pF, C2=5.472670 pF, and L1=208.5877 nH. The value of L1 in
Fig. 10.14 is now 1.319633 nH. To realize this filter using quartz-resonators
similar to the 9.6-MHz quartz-crystal resonator requires that this value is
raised to 208.5877 nH. This is accomplished by increasing the termination
resistance in the Specification tab by the ratio of these values. Therefore,
the termination resistance is increased to 7903.247 ohms.
C3 C6 C9 C12
C=1315.757626pF C=1315.565905pF C=1315.114158pF C=1314.498515pF

L1 L2 L3 L4
ZO=7903.247Ω
L=208.587744nH L=208.587744nH L=208.587744nH L=208.587744nH

C2 C5 C8 C11 ZO=1481.042992Ω
C=1.389559pF C=3.245273pF C=3.353582pF C=3.018671pF

C1 C4 C7 C10
C=0.13574pF C=2.465795pF C=5.13692pF C=8.056066pF

Figure 10.18 Lower-sideband quartz-crystal filter after equating all inductors, deleting the
original C1 and absorbing the transformer into the load resistance.

The remaining series branches could be transformed to the correct


form by applying shunt-Norton transforms to C5, C7, and then C9. However,
applying the Compound Operations: Equate All Ls transform completes all
three remaining transforms automatically. The schematic after deleting C1
96 Filter Synthesis Using Genesys S/Filter

and absorbing the transformer into the loaded resistance is given in Fig.
10.18. The values of C2, C5, C8, and C11 result in static capacitance in the
quartz-crystal resonator dipoles that are smaller than the desired quartz-
crystal model. Therefore, these capacitors are split into two capacitors, one
that becomes the static capacitance and one that can serve as tuning
elements. Using Eqs. 10.8 through 10.10, the required value of capacitance
in series with L1 and C3 in Fig. 10.18 that results in a static capacitance of
5.45 pF in the quartz-crystal dipole is found to be 5.47267 pF. The
transform Basic Operations: Split Series Element: Specify Left Value is
applied with a left value of 5.47267 pF. The remainder is 1.862453 pF.
These steps are applied to the remaining three resonator branches. The
results after converting the dipoles to quartz-crystal resonator models are
given in Fig. 10.19.

Figure 10.19 Lower-sideband quartz-crystal 9.600- to 9.605-MHz bandpass.

The resulting termination resistances are high. If these resistances


cannot be accommodated by the source and load circuitry of the system
then matching networks are employed. Because the filter is narrowband,
broadband matching techniques are not required, and simple two-element
networks are used. Because no TZs at DC and infinity are used in this
design, the low- and high-frequency attenuation is poor. Therefore, both
highpass and lowpass matching networks are used. Genesys includes an
optional match synthesis program for designing these and much more
difficult matching networks. Chapter 12 discusses the use of S/Filter to
design matching networks. The final lower-sideband quartz-crystal
bandpass filter with 50-ohm terminations is given in Fig. 10.20. All
capacitors are placed on the nearest standard values. Because of the
extreme values of termination resistance, tuning of the matching network
inductors might be required in production if the return loss response is
critical.
Piezoelectric Devices 97

Figure 10.20 Lower-sideband bandpass with matching to 50-ohm terminations.

Given in Fig. 10.21 are the responses of the final lower-sideband filter
with unloaded inductor Q specified as 100. This Q has a small influence on
the response as it improves the return loss without adding significant
insertion loss. Unloaded capacitor Q in excess of 1000 has very little effect
on the responses.

Figure 10.21 Responses of the final lower-sideband quartz-crystal bandpass.

10.5 Upper-Sideband Quartz-Crystal Filter


The ladder filter with series quartz-crystal resonators in Fig. 10.20 is a
common form because it is relatively easy to design, at least for narrow
bandwidth, using the prevalent modern method of filter design. However, it
naturally results in a lower-sideband form. Upper-sideband quartz-crystal
filters are realized using the parallel-resonant mode of quartz crystals in
shunt. The series resonance then forms TZs below the passband. The
design of this type of filter is illustrated in this section.
98 Filter Synthesis Using Genesys S/Filter

Given in Fig. 10.22 is the initial Specification tab for the proposed
upper-sideband filter. The TZs are placed to achieve the desired responses
given in Fig. 10.23.

Figure 10.22 Specification tab for the upper-sideband quartz-crystal filter.

The initial schematic for the extraction sequence 9.599 9.5983 9.5948
DC ∞ is given in Fig. 10.24. Th e strategy is to apply a series Norton
transform to the first three series capacitors to place capacitance in parallel
with each series L-C to ground. These branches then have a topology that
conforms to the quartz-crystal model. However, applying a series Norton to
C1 with a positive capacitor to its right results in an unrealizable negative
capacitor in shunt at the input. Therefore, it is necessary to create a shunt
capacitor at the input that later will absorb the negative capacitance created
by the series Norton. To create the input shunt capacitor a Termination
Coupling: Source: Series C to Shunt C transform is applied to C1. This
transform is approximate, but at the extreme narrow bandwidth of a
quartz-crystal filter, the transform is nearly exact. The result with n=1.6 is
given in Fig. 10.25. The value of n is not critical and is selected between the
limits suggested by the S/Filter dialog.
Piezoelectric Devices 99

Figure 10.23 Responses for the upper-sideband quartz-crystal filter.

Figure 10.24 Initial schematic of the synthesized upper-sideband filter.

It is then feasible to begin the series-Norton transforms. In the


description of the remaining transforms, the designators refer to those that
are created after the application of the previous transform. A Simplify
Schematic is applied after any transform that adds a transformer. The
transforms beginning with the filter in Fig. 10.25 are:
1) A series Norton is applied to C2 with n>1 so that a positive capacitor is
placed in parallel with the first series L-C to ground. Larger n results
in a higher filter input impedance. n=1.2 is chosen.
100 Filter Synthesis Using Genesys S/Filter

2) The source and load resistance in the Specification tab are increased
by the ratio of the desired motional inductance, 12.092 mH, to the
current value of L1. The resistance is 345.743738 ohms. L1 now equals
the desired motional inductance. C3 is too large to equal the static
capacitance. This is dealt with later.
3) A series Norton is applied to C5 with n equal to the square root of the
ratio of L2 and the desired motional inductance. n is 2.378502692.
4) A series Norton is applied to C8 with n equal to the square root of the
ratio of L3 to the desired motional inductance. n is 4.099802122. At
this point, each of the shunt inductors equals the desired motional
inductance, but L4 is too large in value.
5) C10 is swapped with the L3/C9 branch.
6) C10 is split, specifying the left value as the quartz-crystal static
capacitance of 5.45 pF.
7) A shunt Norton is applied to capacitor C11 with n equal to the square
root of the ratio of L4 to the motional inductance. n is 8.035702256.
8) C3 is then split with the left value specified as 5.45 pF. It is then
swapped to be adjacent to the series L-C to ground.
9) C7 is treated in a similar fashion.
10) The transformer is absorbed into the load.
The result of these transforms is given in Fig. 10.26. The shunt
resonators are all of the form of quartz-crystal resonators with the desired
static capacitance and motional inductance. The final series resonator is
also the desired motional inductance. Each of these quartz crystals is
manufactured with the series resonant frequency equal to the resonant
frequency calculated from the motional inductance and capacitance of that
branch. The range of frequencies is 9594.727 to 9599.580 kHz. The series
resonator does not have a parallel capacitor corresponding to the static
capacitance. This is dealt with shortly.
Next, each corresponding group of elements is replaced by a quartz-
crystal model with motional resistance as shown in Fig. 10.27. The static
capacitance of the series resonator is included. In addition, each discrete
capacitor is replaced with the nearest standard value. These replacements
have only a minor effect on the return loss.
Piezoelectric Devices 101

Figure 10.25 The upper-sideband filter after application of the Termination Coupling transform
at the input.

Figure 10.26 Upper-sideband filter after application of transforms preparing for replacement of
resonators with quartz crystals.
102 Filter Synthesis Using Genesys S/Filter

Figure 10.27 The upper-sideband quartz-crystal filter with standard values for the discrete
capacitors.

The responses of this final filter are given in Fig. 10.28. The motional
resistance has introduced an approximately 1.7-dB insertion loss. The
traces labeled “with Co” are with the static capacitance added to the series
quartz crystal, and the traces labeled “original” are with a 50.3-uH nulling
inductor in parallel with the quartz crystal. The designer may choose the
more desirable response. The fact that the termination resistances are not a
standard value is dealt with as in Section 10.2.

Figure 10.28 Responses of the upper-sideband quartz-crystal filter. Transmission responses


are with the static capacitance (solid line) and with the static capacitance nulled by a 50.3-uH
inductor (dashed line).
Piezoelectric Devices 103

10.6 Filters with TZs Above and Below the Passband


When a more symmetric response is required, either the static capacitance
is nulled, or in the case of generalized bandpass filters, TZs are added both
above and below the passband. This section illustrates the design of a
quartz crystal bandpass filter with one TZ below and one TZ above the
passband. The filter is 1-kHz bandwidth centered at 9.605 MHz. TZ zeros
are specified as given in Fig. 10.29 to create the responses given in Fig.
10.30. The extraction sequence ∞ 9.5989 9.6021 ∞ ∞ ∞ ∞ DC results in the
initial schematic given in Fig. 10.31.
The elements C2, L2, and C3 are already the correct topology for a
quartz-crystal model. However, the elements L3 and C4 are nonconforming.
In the following list of transforms, the designators refer to those created
after the previous transforms is applied. Simplify Schematic is applied after
any transform that adds a transformer. The transforms are:
1) Both termination resistances in the Specification tab are increased by
the ratio of the desired motional inductance to the inductance of L1, or
53.595266 ohms.
2) A shunt Norton is applied to capacitor C2. The resulting negative
capacitor on the left must absorb into C1 with a positive result. n equal
to 0.1 is chosen.
3) A series Norton is applied to capacitor C3 to create a capacitor in
parallel with L2 and C4. The turns ratio is the square root of the ratio of
L2 to the desired motional inductance. n equals 2.405301029.
4) A shunt Norton is applied to capacitor C7 with n equal to 28.0222, the
square root of the ratio of L4, and the desired motional inductance.
5) A shunt Norton is applied to capacitor C10 with n equal to 13143.6, the
square root of inductor L5, and the desired motional inductance.
6) Capacitor C5 is split with the left value specified as the desired static
capacitance, or 5.45 pF.
7) Eqs. 10.8 through 10.10 are used to find the values of C7, C8, and L3
that convert to the desired quartz-crystal model parameters. C7 is
adjusted to so that the motional inductor in the equivalent quartz-
crystal model equals the desired motional inductance. In this case, C7
equals 13.25082 pF. Capacitor C7 is split with this left value.
8) Capacitor C7 is swapped with capacitor C8.
9) The transformer is absorbed into the load.
104 Filter Synthesis Using Genesys S/Filter

Figure 10.29 Specification tab for a quartz-crystal filter with one TZ below and one TZ above
the passband.

Figure 10.30 Desired responses of the quartz-crystal filter with two FTZs.
Piezoelectric Devices 105

Figure 10.31 Initial synthesized schematic of the quartz-crystal filter with two FTZs.

The resulting schematic is given in Fig. 10.32. L2, C4, and C5 are ready
for replacement with a quartz-crystal resonator. When the network of C8,
C9, and L3 is replaced with the quartz crystal using Eqs. 10.8 through 10.10,
the motional values are correct. The required parallel capacitance is
13.22806 pF, of which 5.45 pF is the quartz-crystal static capacitance and
7.77806 pF is an added capacitor in parallel with the quartz crystal. The
three remaining series resonators are replaced with quartz crystals, and the
static capacitance is nulled using parallel inductors as before.
The final schematic with quartz-crystal resonators and nulling
inductance is given in Fig. 10.33. In addition, discrete capacitors and the
termination resistance are set to standard values. An exception to standard
values is the capacitor C5 in parallel with X3. Unless the return loss is not
critical, this value is somewhat sensitive and possibly needs trimming in
production. The series-resonant frequency specified for manufacture of the
quartz-crystals range from 9593.848 to 9600.214 kHz.
The responses for the final quartz-crystal filter are given in Fig. 10.34.
Standard values, dissipation, and effects of the nulled static capacitance
shift the passband up about 50 Hz. This could be compensated by shifting
the resonant frequency specification of each quartz crystal down 50 Hz.
106 Filter Synthesis Using Genesys S/Filter

Figure 10.32 Schematic of the filter with two FTZs ready for quartz-crystal resonators.

Figure 10.33 Final quartz-crystal filter with FTZs below and above the passband.
Piezoelectric Devices 107

Figure 10.34 Responses of the final quartz-crystal filter with FTZ below and above the
passband.

10.7 Wide-Bandwidth Quartz-Crystal Filters


The bandwidth of quartz-crystal filters is generally restricted to less than
the series/parallel inter-resonance spacing. This is 20 kHz for the example
9.6-MHz quartz crystal. In practice, difficulties arise with bandwidths even
less than this theoretical upper limit. Given in Fig. 10.35 is a 10-kHz
bandwidth filter with one TZ at DC, five TZs at infinity, one TZ at 9.592
MHz below the passband, and one TZ at 9.618 MHz above the passband.
The design of this filter follows the same procedure as described in the
previous section. The difficulty is that the required quartz-crystal static
capacitance for X3 is 2.077596 pF. This is not feasible for the type of quartz
crystal exemplified by the 9.6 MHz sample. For this filter, the problem is
resolvable by nulling only a portion of the static capacitance. The
susceptance of the 5.45 static capacitance is 0.328907 mS. The susceptance
of the 2.0776-pF parallel capacitance required by the filter synthesis is
0.125383 mS. An inductor in parallel with the quartz-crystal with the
difference of 0.203524 mS causes the net effective capacitance to be 2.0776
pF. This is an 80.415-uH inductor as shown in Fig. 10.35.
108 Filter Synthesis Using Genesys S/Filter

Figure 10.35 Quartz-crystal resonator filter with a 10-kHz bandwidth.

The response of the resulting filter is given in Fig. 10.36. The finite
unloaded Q of the susceptance canceling inductor results in limited
attenuation at the TZ above the passband. Otherwise, the 10-kHz
bandwidth quartz-crystal design is successful.

Figure 10.36 Responses of the wide-bandwidth quartz-crystal filter with partial nulling of X3 to
realize a small static capacitance requirement.

10.8 Very Wide-Bandwidth Quartz-Crystal Filters


The next case considers very wide-bandwidth quartz-crystal filters where
the passband must exceed the series/parallel inter-resonant spacing of the
quartz crystal. The filter schematic given in Fig. 10.37 is a three-section
9.625- to 9.675-MHz quartz-crystal filter. The 50-kHz passband is a factor
of 2.5 times than the interresonant spacing of the example 9.6-MHz crystal.
Piezoelectric Devices 109

It is necessary to null the parallel-resonant modes of the quartz crystals


because the resulting TZs lie in the passband. Therefore, three 50.3-uH
inductors are placed in parallel with the quartz crystals.

L4 L5 L6
L=50.3μH L=50.3μH L=50.3μH

ZO=2914.5Ω X1 X3 X2 ZO=2914.5Ω
R=11.6Ω R=11.6Ω R=11.6Ω
L=12.092mH L=12.092mH L=12.092mH
Cm=0.022591pF Cm=0.022687pF Cm=0.022591pF
Co=5.45pF Co=5.45pF Co=5.45pF
C2 C4
C=5.306757pF C=5.306762pF

Figure 10.37 Schematic of the very wide-bandwidth quartz-crystal filter.

The responses are given in Fig. 10.38. The results are satisfactory from
about 9.25 to 10.05 MHz. However, well below and well above the passband
the rejection decreases. Examination of the schematic in Fig. 10.37 reveals
that when the static capacitance of the quartz crystals is considered, the
quantity of TZs at DC is zero and the quantity at infinity is only one.
Consequently, the attenuation at these frequency extremes is very poor.
Notice that the very wide-bandwidth results in a high termination
resistance. This is used to advantage in the next step by adding matching
networks with a loaded Q of 50 at the input and output to match this filter
to 50-ohm terminations. The schematic after optimization for best response
and placing elements on standard values is given in Fig. 10.39 and the
responses are given in Fig. 10.40 with an unloaded Q of 100 for all
inductors. The high Q of the matching networks results in most of the
passband insertion loss and probably requires tuning of one of the elements
in each matching network.
Although narrow bandwidth quartz-crystal filters are less susceptible to
stopband issues, whenever inductors are used to null the static capacitance,
it is wise to simulate the attenuation well into the stopbands.
110 Filter Synthesis Using Genesys S/Filter

Figure 10.38 Responses of the very wide-bandwidth quartz-crystal filter.

L3 L4 L5
C2 C5
L=49.2μH L=53.3μH L=48.5μH
ZO=50Ω C=330pF C=330pF ZO=50Ω

X1 X3 X2
R=11.6Ω R=11.6Ω R=11.6Ω
L1 L=12.092mH L=12.092mH L=12.092mH L2
L=1000nH Cm=0.022565pF Cm=0.022621pF Cm=0.022565pF L=1000nH
C1 Co=5.45pF Co=5.45pF Co=5.45pF C6
C=1500pF C3 C4 C=1500pF
C=6.2pF C=6.2pF

Figure 10.39 Very wide-bandwidth quartz-crystal filter with matching networks and standard
values for discrete elements.

The static capacitance of the quartz-crystal posses special issues for


filter design, particularly at wider bandwidths. However, filters with loaded
Q over 100 cannot use purely L-C elements. In this case, L-C networks are
used merely to clean the stopbands of a quartz-crystal filter.
Piezoelectric Devices 111

Figure 10.40 Responses of the completed very-wide quartz-crystal filter.

10.9 Ceramic-Piezoelectric Resonators


A more elegant solution to the problem of filter bandwidths that lie
between the optimum range for L-C and quartz-crystal resonators is to use
ceramic piezoelectric resonators. These piezoelectric resonators are not to
be confused with the ceramic-loaded coaxial resonators illustrated in
Chapter 9. A piezoelectric material such as lead zirconium titanate replaces
the quartz of quartz crystals. Ceramic resonators are typically used in the
0.4- to 60-MHz frequency range with common application for 455-kHz and
10.7-MHz IF filters. Multisection monolithic filters are common, but
individual resonators may also be also used to construct filters. The loaded
Q of ceramic resonators is typically in the 500 to 1000 range.
The electrical equivalent model is identical to that of a quartz crystal, so
the same terminology is used here. The parameters of an 11.607-MHz
ceramic resonator, determined from the measured transmission response
using Eqs. 10.1 through 10.3 are Rm=4.1 ohms, Lm=44.9805 uH, Cm=4.18
pF, and C0=28.6 pF [1]. The unloaded Q is the motional reactance divided
by the motional resistance, or 800 in this case.
Because the model is identical to the quartz-crystal model, the
constraints and available design techniques are similar. Consider the design
of a four-section 100-kHz bandwidth filter centered at 11.6 MHz using a
ceramic resonator similar to the unit just described. The design generally
follows procedures used for the filter in Section 10.6 with single TZs below
and above the passband but with three rather than five TZs at infinity.
Beginning with the extraction sequence ∞ 11.488 11.712 ∞ ∞ DC, the
transforms, minus step 5, result in the schematic given in Fig. 10.41.
112 Filter Synthesis Using Genesys S/Filter

Figure 10.41 Schematic of the ceramic-resonator filter after application of transforms to the
original extraction.

The first and last resonators are replaced with ceramic resonators. The
static capacitance is then nulled with 6.574 uH inductors. The dipole
consisting of C5, C6, and L2 is ready for replacement with a ceramic
resonator. When the dipole consisting of C7, C8, and L3 is replaced with the
piezoelectric dipole using Eqs. 10.8 through 10.10, the values are ready for
replacement with a ceramic resonator. However, the required static
capacitance is higher than provided by the resonator. This is corrected by
placing additional capacitance, C4, in parallel with the resonator. The
ceramic resonator filter, after placing discrete capacitors on standard
values is given in Fig. 10.42 and the responses are given in Fig. 10.43.

Figure 10.42 Ceramic-resonator filter after replacing L-C dipoles with ceramic resonators.
Piezoelectric Devices 113

Figure 10.43 Final responses of the ceramic-resonator filter.

The final filter has a TZ at DC and one at infinity, so the stopband


attenuation remains at least 50 dB through 50 MHz except for a false
passband about 15 dB down around 4.4 MHz. If necessary, this false
passband could be eliminated with matching networks that shift the 25.14-
ohm termination resistance.

Reference
[1] R. Rhea, Discrete Oscillator Design: Linear, Nonlinear, Transient, and Noise
Domains, Artech House, Norwood, MA, 2010.
11 Symmetry
Symmetry has two important manifestations in electrical filters:
physical symmetry with element types and values mirrored with respect to
the input and output, and response symmetry with equal selectivity above
and below the passband [1]. Physical symmetry reduces the number of
unique elements that are modeled, ordered, tested, stocked, picked, and
assembled. For distributed element filters that are electromagnetically
simulated, physical symmetry can reduce the required execution time and
memory by a factor of 16 in simulators that take advantage of symmetry.
The conventional all-pole bandpass filter with equal quantities 0f TZs
at DC and infinity has geometric response symmetry. If the transmission
amplitude response is plotted on a logarithmic-frequency scale, the slope of
the transition regions below and above the passbands are equal. However,
more often arithmetic symmetry is desired. Arithmetic symmetry results in
group-delay symmetry.

11.1 Physical Symmetry


Physical symmetry can be achieved using the modern method of filter
design for the first three classes of filters listed below, but direct synthesis
is required for the last two classes.
1) A Butterworth transfer approximation.
2) Odd-order Chebyshev transfer approximation.
3) Coupled-resonator bandpass filters with one TZ at DC and an odd
quantity of TZs at infinity, or an odd quantity of TZs at DC and one TZ
at infinity.
4) Lowpass filters with FTZ pairings and an odd quantity of TZs at
infinity.
5) Bandpass filters with odd and equal quantities of nonfinite TZs and
with paired FTZs below or above (not both) the passband.

11.1.1 A Lowpass Filter with FTZ Pairings


Consider case four in the above list. A lowpass filter is required that passes
DC to 32 MHz with a maximum passband ripple of 0.1 dB (16.4-dB return
loss) and a minimum rejection of 60 dB above 50 MHz. This requirement
requires a tenth-degree all-pole lowpass with five inductors and five
capacitors. This requires five unique values of inductors and five unique
values of capacitors and results in dissimilar input and output impedance.

115
116 Filter Synthesis Using Genesys S/Filter

The better solution is 11-th degree that adds only one capacitor (a total of
six), but requires only three unique values of inductors and capacitor, and
results in equal input and output impedance. An even better solution is a
symmetry preserving lowpass with three TZs at infinity and two TZs at 54
MHz. The Specification tab is given in Fig. 11.1 and the schematic for the
extraction sequence 54 ∞ ∞ 54 ∞ is given in Fig. 11.2. This solution only
requires three inductors. Given in Fig. 11.3 are the responses of this filter.
The rejection specification is achieved with considerable margin.

Figure 11.1 Specification tab for the value-symmetric lowpass created with two-paired FTZs.

C2 C5
C=32.82pF C=32.82pF
ZO=50Ω ZO=50Ω

L1 L2 L3
L=264.69nH L=400.52nH L=264.69nH

C1 C3 C4 C6
C=90.41pF C=190.03pF C=190.03pF C=90.41pF

Figure 11.2 Schematic of the value-symmetric lowpass.


Symmetry 117

Figure 11.3 Responses of the value-symmetric lowpass.

11.1.2 A Bandpass Filter with FTZ Pairings


With case five in the above list, paired FTZs may exist either above or below
the passband, but not both. Therefore, this case is useful when increased
selectivity is required either above or below the passband. Consider the
following 455-kHz IF bandpass filter with 30-kHz bandwidth. The S/Filter
Specification tab is given in Fig. 11.4 and the resulting schematic with the
extraction sequence 0.37 ∞ 0.37 DC is given in Fig. 11.5 . The responses of
the 455-kHz bandpass filter is given in Fig. 11.6.
Fortunately, for case five in the above list, the extraction sequences that
result in value symmetry have a unity transformer turns ratio.
Unfortunately, particularly for narrow bandwidth, this type of filter is
susceptible to extreme element values, and large inductor or capacitor
value ratios. For example, the inductor values are unreasonably small for
the shunt-element first version of this filter. The Series Element First form
has rather large inductor values, but this is consistent with realization using
toroid or pot-core inductors, which also offer better unloaded Q at this
frequency than air-cored inductors.
118 Filter Synthesis Using Genesys S/Filter

Figure 11.4 Specification tab for the 455-kHz bandpass with paired FTZs below the
passband.

L2 C3
ZO=50Ω L=8.44e+6nH C=14.475pF ZO=50Ω

C1 C5
C=1346.458pF C=1346.458pF
L1 L3
L=280533.943nH L=280533.943nH

C2 C4
C=659.556pF C=659.556pF

Figure 11.5 Schematic of the 455-kHz bandpass.


Symmetry 119

Figure 11.6 Responses of the 455-kHz bandpass.

11.2 Response Symmetry


The conventional all-pole bandpass has an equal number of TZs at DC and
infinity. As shown in Fig. 4.2, when plotted with an arithmetic frequency
scale, the response has greater selectivity below the passband. The response
is symmetric with a logarithmic frequency scale; however, arithmetic
symmetry is often required. The transmission amplitude response is
directly related to the transmission phase response for ladder networks
with no bridging elements. Arithmetic symmetry results in group-delay
symmetry. For topologies that are not symmetric, the selectivity inequity
worsens with increasing bandwidth. For narrow bandwidth filters,
symmetry may not be a significant issue. For wide bandwidth filters, a
solution is often required.
Blinchikoff [2] offers a proof that for bandpass filters transformed from
lowpass prototypes of order N, symmetry results when the quantity of TZs
at DC equals N/2. For example, with N equal to four, the degree of the
bandpass is eight. For symmetry, the quantity of TZs at DC is two and the
quantity at infinity is six. Another way of stating this is that the quantity of
TZs at infinity should exceed the quantity at DC by a factor of three.
Blinchikoff then proceeded to describe topologies and tables for
designing a few specific filters that meet this criteria. The ability to
arbitrarily specify transmission zeros using direct synthesis provides a
superior solution for designing symmetric filters.
120 Filter Synthesis Using Genesys S/Filter

11.2.1 All-Pole Symmetric Response Filters


Consider an eighth-degree, 140- to 260-MHz, 0.1-dB ripple, bandpass filter
with two TZs at DC and six at infinity. There are approximately ten unique
solutions to this specification with the Series Element First option, and ten
with a shunt element first. All require dissimilar terminations so the
automatic Remove Transformer transform is applied and the Extraction
table is canvassed for the topology with the most reasonable element
values. The selected topology is given in Fig. 11.7.

C2 C4
ZO=50Ω C=2.21pF C=34.99pF ZO=50Ω

L1 L2 L3
L=73.76nH L=374.97nH L=86.82nH

C1 C3 L4
C=7.56pF C=8.93pF L=24.19nH
C5
C=29.48pF

Figure 11.7 Schematic of the symmetric eighth-degree 140- to 260-MHz bandpass.

Fig. 11.8 shows the transmission, group-delay, and return-loss


responses for the symmetric-response bandpass in Fig. 11.7. Notice the
excellent symmetry in both the transmission-amplitude and group-delay
responses.

11.2.2 Generalized Bandpass with Symmetric Response


The Cauer-Chebyshev bandpass zigzag is one of the most inductor-
economic topologies [3]. This filter does not require direct synthesis and is
designed by Genesys Passive Filter module. This topology requires even-
order lowpass prototypes of order four or greater. Given in Fig. 11.9 is a 50-
to 90-MHz example designed from a fourth-order lowpass. It has two TZs
at DC and two at infinity, and one each below and above the passband.
While it has a low inductor count, it has two disadvantages for certain
applications: the Cauer-Chebyshev approximation requires specific FTZ
placements, and the response is not symmetric because there is an equal
quantity of TZs at DC and infinity.
Symmetry 121

Figure 11.8 Responses of the symmetric-response all-pole bandpass.

C5
C1 C4 C=10.3pF
ZO=50Ω C=28.07pF C=23.27pF
ZO=71.56Ω

L1 L3
L=223.59nH L=155.92nH

C2 L2 L4
C=67.61pF L=129.02nH L=90.14nH
C6
C=55.99pF

C3
C=153.18pF

Figure 11.9 Zigzag (minimum-inductor elliptic) designed by the Genesys Passive Filter
program.
122 Filter Synthesis Using Genesys S/Filter

Figure 11.10 Responses of the zigzag Cauer-Chebyshev bandpass.

The response is given in Fig. 11.10. The lack of arithmetic symmetry is


evident from the group-delay response. Direct synthesis may be used to
create a similar topology with corrected symmetry, or to place FTZs for a
custom stopband attenuation. These filters may also improve inductor
economy. S/Filter is used to synthesize a 50- to 90-MHz generalized
bandpass, also with four nonfinite TZs, but with one at DC and three at
infinity to achieve arithmetic symmetry. In this case, the FTZs are placed to
achieve a minimum stopband attenuation of 40 dB and a symmetric
response. The resulting schematic with the Remove Transformer transform
applied and the extraction sequence ∞ 22.5 114.2 ∞ ∞ DC is given in Fig.
11.11. The responses are given in Fig. 11.12.
Notice the dissimilar stopband attenuation that results when the FTZs
are placed to achieve group-delay and selectivity symmetry. The FTZs may
be placed to achieve specific stopband requirements at the expense of
symmetry, but the symmetry is still improved over that of the elliptic
Cauer-Chebyshev, because of better management of the nonfinite TZs.
Symmetry 123

Figure 11.11 Generalized response of the symmetric 50- to 90-MHz bandpass filter.

Figure 11.12 Responses of the generalized 50- to 90-MHz bandpass with symmetric
responses.

11.2.3 Symmetry by FTZ Placement


Response symmetry can also be controlled by the placement of FTZs. For
example, consider a symmetric requirement similar to the previous 50- to
90-MHz bandpass, but with greater attenuation in the stopband above the
passband. In this case, two TZs at both DC and infinity are specified. This
results in asymmetry with greater selectivity below the passband. This is
124 Filter Synthesis Using Genesys S/Filter

compensated by specifying an additional FTZ above the passband. The


placement of FTZs is adjusted to achieve symmetry and meet the required
stopband attenuation. The extraction sequence ∞ 38.3 122.5 DC 105.5 ∞ DC
is selected, and after applying the Remove Transformer transform, the
schematic given in Fig. 11.13 results.
The responses of this filter are given in Fig. 11.14. The symmetric
group-delay response is evidence of symmetry in the transmission
amplitude response. The additional FTZ above the passband increases the
stopband rejection in that stopband and provides symmetry even though
the ratio of the quantity of TZs at infinity and DC is less than three.

Figure 11.13 Schematic of the symmetric generalized 50- to 90-MHz bandpass with greater
stopband attenuation above the passband.

11.3 Group-Delay Equalization


Many communications systems require near-constant group delay in the
passband of filters. This can be achieved by filters that include elements
that bridge across nodes in the filter [4]. Filters of this class with arbitrary
TZ placement are economic, but they can be difficult to design and tune. An
alternative approach is to design the filter using the synthesis techniques
outlined in this book, followed by cascading the filter with a group-delay
equalization network.
Symmetry 125

Figure 11.14 Responses of the symmetric generalized 50- to 90-MHz bandpass with greater
attenuation above the passband.

The Genesys Equalization module automates the group-delay


equalization of any Genesys simulated circuit. The Equalization module
cascades a user-specified quantity of nine different types of L-C and op-
amp all-pass equalizer sections and optimizes the parameters of those
sections to flatten the group-delay response over a user-specified frequency
range. The original filter equalized in this case is the symmetric-response
all-pole bandpass designed in Section 11.2.1. Group-delay equalizers of type
three are added until a quantity of three achieved the desired error from a
flat group-delay response. The resulting three-section group-delay
equalizer is shown in Fig. 11.15. The original filter is automatically
represented by data in the two-port network N1.
The resulting cascade responses are given in Fig. 11.16. Both the
original and equalized group-delay responses are given as dashed
responses. The group-delay equalizer sections are all-pass networks and
with high element Q, they have no affect on the transmission amplitude
response. With finite-Q elements, these networks introduce midband
insertion loss, but because the loaded Q of these sections is modest, the
added insertion loss is typically rather small.
126 Filter Synthesis Using Genesys S/Filter

Figure 11.15 Three-section group-delay equalizer cascaded with the all-pole symmetric-
response bandpass.

Figure 11.16 All-pole symmetric response bandpass with group-delay equalization.

Tuning of the filter and equalizer sections is best accomplished in


phases. PCB jumpers allow tuning of the filter first, followed by tuning of
each equalizer section individually to a presimulated group-delay response
mask.
Symmetry 127

References
[1] R. Rhea, “Exploiting Filter Symmetry,” Microwave Journal, March 2001.
[2] H. Blinchikoff and A. Zverev, Filtering in the Time and Frequency Domains,
Noble Publishing(SciTech), Raleigh, NC, 2001.
[3] R. Saal and E. Ulbrich, “On the Design of Filters by Synthesis,” IRE Trans.
Circuit Theory, Vol. CT-5, December 1958.
[4] J. Rhodes, “The Design and Synthesis of a Class of Microwave Bandpass Linear
Phase Filters,” Trans. Microwave Theory and Techniques, Vol. 17, April 1969.
12 Matching with S/Filter
Matching herein refers to the use of a network to cause maximum
power transfer from an arbitrary source impedance to an arbitrary load
impedance. Even with complex impedances, matching at a single
frequency, or approximately over a narrow bandwidth, is a straightforward
process requiring simple networks with closed form equations [1],[2].
With wider bandwidth, the problem can be daunting. Genesys includes
Impedance Match, a module designed to match real or complex
impedances using eight built-in network algorithms, or a custom user-
specified network. One or more of these algorithms can find the optimum
solution to an impedance-matching problem. For increasing bandwidth,
the user specifies an increased quantity of elements. One of the Impedance
Match algorithms uses a direct synthesis similar to that in S/Filter. When
the problem is primarily a matching problem, Impedance Match is
recommended. This chapter addresses matching techniques used with
S/Filter when the problem is primarily a filtering problem, but the loads
are dissimilar or complex.

12.1 Matching Concepts


Impedance mismatch may be viewed from a network or a wave perspective.
For a network perspective, consider Fig. 12.1, with a 1-volt DC source with a
50-ohm source impedance and a 50-ohm load impedance.

SOURCE LOAD

Rsource
R=50Ω

VS1 Rload
VDC=1V R=50Ω

Figure 12.1 The matching problem from a network perspective.

129
130 Filter Synthesis Using Genesys S/Filter

The current is 1 volt over 100 ohms, or 10 mA. The voltage source
delivers 10 mW; 5 mW to the source resistance and 5 mW to the load
resistance. If the load is 25 ohms, the current increases to 13.33 mA, and
the 1-volt source delivers 13.33 mA. However, only 4.44 mW is delivered to
the load, while the majority is delivered to the source resistance. If the load
is 100 ohms, the current decreases to 6.67 mA, and the 1-volt source
delivers 6.67 mW. While the majority is delivered to the load, again it is
only 4.44 mW. Maximum transfer of power occurs when the load resistance
equals the source resistance.
From a wave perspective, when a signal propagating through a system
encounters an impedance shift, a portion of the signal is reflected, thus
reducing the transmitted signal. This is analogous to reflection in an optical
system. Mismatch occurs when the impedance shift is either resistive or
complex.

12.1.1 Complex Conjugate Match


When the load impedance is complex, the power delivered from a resistive
source to the resistive component of the load is reduced. When the load
impedance is R + jX, maximum power transfer is restored if the source
impedance is R - jX. This is referred to as a complex conjugate match. The
source may be purely resistive and a matching network may be used to
present an effective R - jX to the load. This network may also shift the
resistive component of the source to equal the load resistance. When the
matching network added between the source and load is purely reactive,
this network dissipates no energy and maximum power transfer occurs.

12.1.2 Two-Element Matching Networks


Two reactors may be used to match any complex source and load
impedance [1],[2]. Given in Fig. 12.2 are the matching-spaces supported by
the matching-network types one through eight. These types of diagrams
were first introduced by Smith [3]. Complex loads RL + jXL within the
enclosed spaces on the Smith chart are matched to a real source impedance
R0 using the indicated two-reactor network. Given below are the closed-
form equations for computing the reactor values of the matching-network
types one through eight in Fig. 12.2. At a design frequency f and with the
load admittance given by
RL XL
YL = GL + jBL = −j (12.1)
RL2 + X L2 RL2 + X L2

and then setting


Matching with S/Filter 131

GL
A= − GL2 (12.2)
R0

B = RL ( R0 − RL ) − X L (12.3)

C = − RL ( R0 − RL ) − X L (12.4)

the network values are


A
L1 =
(
2πf GL2 + A2 ) (12.5)

A − BL
C1 = (12.6)
2πf

1
L2 = (12.7)
2πf ( A + BL )

GL2 + A2
C2 = (12.8)
2πfA

B
L3 = (12.9)
2πf

B + XL
C3 =
2πf (RL2 + (B + X L )2 ) (12.10)

L4 = −
(R + (C + X L )2
2
L ) (12.11)
2πf (C + X L )

1
C4 = − (12.12)
2πfC

GL2 + A2
C5a = (12.13)
2πfA

A + BL
C5b = − (12.14)
2πf

A
L6 a =
(
2πf Gl2 + A2 ) (12.15)
132 Filter Synthesis Using Genesys S/Filter

1
L6b = − (12.16)
2πf ( A − BL )

1
C7 a = − (12.17)
2πfB

B + XL
C7b =
2πf ( RL2 + (B + X L )2 ) (12.18)

C
L8a = (12.19)
2πf

L8b = −
(R 2
L + ( X L + C )2 ) (12.20)
2πf ( X L + C )

Consider the match of a 50-ohm source to a load of 10 ohms in series


with 6.366 pF at 100 MHz. The load impedance is 10 – j250 ohms. Types
two, three, six, and eight are valid matching networks. For a lowpass type
passband, type three is selected. From Eqs. 12.3, 12.4, 12.9, and 12.10, L3 is
429.72 nH and C3 is 63.66 pF. The resulting S11 is given in Fig. 12.3.
The match is exact at only the design center frequency. If the reactance
is significant, or the ratio of Ro to RL differs significantly from unity, the
matched bandwidth is narrow. In our example, the 20-dB return loss
bandwidth is only 0.6 MHz, or 0.6% bandwidth. The problem of matching,
at least for lossless elements, is largely an issue of bandwith. Increased
bandwidth is achieved by using matching networks with an increased
quantity of elements. The remainder of this chapter covers direct synthesis
to create these matching networks.

12.2 Real Terminations


This section deals with S/Filter matching when the terminations are purely
resistive with no reactive element.

12.2.1 Exploiting Extraction Sequences


As demonstrated in examples in previous chapters, certain extraction
sequences result in a termination resistance different than the source
resistance. This is used to advantage when matching dissimilar resistance.
Matching with S/Filter 133

Figure 12.2 Match space for two-element matching networks.


134 Filter Synthesis Using Genesys S/Filter

Figure 12.3 Match response of the two-element matching network.

Fig. 12.4 shows plots of the maximum ratio of termination resistance


that are to be matched by a bandpass filter given a ratio of the upper
frequency to the lower frequency, for a match of 20-dB return loss in the
passband. The four traces represent the quantity of nonfinite transmission
zeros at DC and infinity. For example, the 2/4 trace represents a
synthesized network with two TZs at DC and four TZs at infinity. Recall that
for an all-pole network, six TZs result in a network with six elements when
the network is canonic.
Increasing the quantity of TZs increases the matching bandwidth for a
given termination resistance ratio. For example, when matching a 50-ohm
source to a 100-ohm load, the termination resistance ratio is two. The
termination resistance ratio is also two when matching a 50-ohm source to
a 25-ohm load. With two TZs at DC and at infinity, the maximum upper
frequency is almost twice the lower frequency, an octave bandwidth. With
six TZs at DC and at infinity, the bandwidth is a decade. Likewise,
increasing the quantity of TZs increases the achievable bandwidth for a
given termination resistance ratio. For example, over an octave bandwidth,
2/2 matches a resistance ratio of almost two while 6/6 matches a resistance
ratio of 100.
Matching with S/Filter 135

Figure 12.4 Matchable termination resistance ratios versus the ratio of the upper to lower
cutoff frequency for a passband return loss of 20 dB.

For matching, even quantities of TZs are more economic than odd
quantities. For example, 2/4 provides better bandwidth or higher
termination resistance ratios than 3/3. All even TZ choices of the same
quantity result in similar performance. For example, 2/6, 4/4, and 6/2
provide similar performance.
Fig. 12.4 is for the case with 20-dB passband return loss. A worse
return loss improves the bandwidth and increases the termination
resistance ratio.
With respect to matching, FTZs are not economic. For a given quantity
of elements, the specification of FTZs reduces the performance. FTZs are
employed in filters used for matching only if the stopband requirements
benefit.
The above discussion involves design to maximize the bandwidth or
termination resistance ratio. The general case of matching, however,
involves matching a source resistance to a specific load resistance. For
example, consider a required resistance ratio of 30 over an octave
bandwidth. This requirement lies between TZ placements of 4/4 and 6/6. A
number of techniques are employed:
1) If the design is purely a matching issue, the 6/6 synthesis is used and
the bandwidth is extended beyond the required bandwidth. This
reduces the sensitivity of the design to element tolerance.
136 Filter Synthesis Using Genesys S/Filter

2) If filtering is required and the stopband rejection must be retained,


the 6/6 synthesis is used and a Norton transform is applied to adjust
the load resistance.
3) If filtering is required and the stopband rejection must be retained,
the 6/6 synthesis is used and a different extraction sequence is used.
Fig. 12.4 is derived by selecting the sequence resulting in the greatest
performance.
4) The passband ripple is reduced until the extraction sequence results
in a TRF of unity.
Fig. 12.4 is provided to illustrate the relationships between network
complexity, the resistance ratio, and matching bandwidth. To illustrate how
the design process might actually proceed, consider the following example.
A 50-ohm source is matched to a 12.5-ohm load from 100 to 200 MHz. A
passband return loss of 12 dB is required. No filtering specifications are
involved.
Given in Fig. 12.5 is the Specification tab specifying the above
parameters. The load resistance is set at the required 12.5 ohms. The
passband ripple is adjusted until the response indicated a 12-dB return loss.
A review of Fig. 12.4 suggests that two TZs at DC and four TZs at infinity
should be sufficient.

Figure 12.5 Specification tab for matching example that exploits extraction sequences.

The Extraction tab is selected and the Lmax/Lmin column is canvassed


for a small inductor value ratio. The extraction sequence ∞ ∞ ∞ DC ∞ DC is
selected. Next, the Transform tab is selected and the Remove Transformer
button is clicked. The resulting matching network is given in Fig. 12.6. The
Norton transform added an inductor. Selecting the extraction sequence ∞
Matching with S/Filter 137

DC ∞ DC ∞ ∞ foll owed by the Remove Transformer option adds a


capacitor, thus reducing the inductor count to three.

ZO=12.5Ω
ZO=50Ω

L2 L4
L=25.78nH L=26.96nH C3
C=48.34pF
L3 C2
L=10.02nH C=147.34pF
L1 C1
L=194.21nH C=43.12pF

Figure 12.6 A 50- to 12.5-ohm matching network designed by exploiting extraction


sequences.

Alternatively, the bandwidth is increased by decreasing the Lower


Cutoff and increasing the Upper Cutoff until the TRF ratio for the selected
extraction sequence is unity. In this case, no Remove Transformer option is
required. The resulting schematic is given in Fig. 12.7 and the responses in
Fig. 12.8. Another option is to decrease the passband ripple until a Norton
transform is not required. In this case, a passband ripple of 0.16 dB with a
return loss of 14.4 dB results in a canonic solution very similar to Fig. 12.7.

ZO=50Ω ZO=12.5Ω

L1 L3
L=24.97nH L=22.44nH C3
C=59.11pF
L2 C2
L=12.03nH C=122.7pF
C1
C=35.85pF

Figure 12.7 Matching network with bandwidth expanded to avoid a Norton transform.
138 Filter Synthesis Using Genesys S/Filter

Figure 12.8 Responses of the 50- to 12.5-ohm matching network with expanded bandwidth.

12.2.2 Exploiting Resonator Filters


Coupled resonator filters typically employ impedance transformations that
may be exploited to match dissimilar terminations. This is particularly
effective for narrow bandwidth filters. Consider the following example of an
88- to 108-MHz three-section series resonator filters with matching of 100
to 50 ohms with a 20-dB return loss. Both source and load terminations are
specified as 100 ohms, the cutoff frequencies are specified, and a passband
ripple of 0.0432 dB is specified. As described in Section 8.2.1, we begin
with the Series Element First extraction sequence DC ∞ ∞ ∞ ∞ ∞ and apply
shunt Norton transforms to the first shunt capacitor. A turns ratio value of
n equal to 0.21082 results in equal first and second inductors. The resulting
schematic is given in Fig. 12.9.
C1 C3
ZO=100Ω C=4.67pF C=4.87pF ZO=100Ω

L1 L2 L3
L=666.37nH L=666.37nH L=27.75nH

C2 C4
C=18.22pF C=89.29pF

T1
P=1
S=4.9

Figure 12.9 Schematic of the 88- to 108-MHz bandpass used as a matching network.
Matching with S/Filter 139

Notice that the transformer turns ratio is 4.9, thus indicating that
without a transformer, this network matches 100 ohms down to about 4.2
ohms. The application of a shunt Norton to the final shunt capacitor
supports matching any resistance from 4.2 ohms up to 100 ohms.
Therefore, the termination resistance in the Specification tab is set to the
desired 50 ohms and a Norton Shunt transform is applied to C4, selecting n
to remove the transformer. The filter matching 100 to 50 ohms is given in
Fig. 12.10.

C1 C3 C5
ZO=100Ω C=4.671pF C=5.6216pF C=10.4779pF

L1 L2 L3
L=666.37nH L=666.37nH L=332.5nH ZO=50Ω

C2 C4
C=18.22pF C=25.79pF

Figure 12.10 Final schematic of the series-resonator bandpass used to match 100 to 50
ohms.

12.3 Complex Terminations


As illustrated by Fig. 12.4, very wide-bandwidth, even as high as a decade or
more, is readily achieved between resistive terminations. When the load is
complex, the matching bandwidth is more limited.

12.3.1 Fano’s Limit


Three definitions of Q are required to define the matching problem with
complex terminations. Loaded Q is the center frequency divided by the 3-
dB bandwidth of the response.
f0
Qloaded = (12.21)
f upper − f lower

It is a finite value even if the circuit is built using elements with infinite
Q. Element Q, or unloaded Q, is a measure of element quality. It is defined
as 2πf times the stored energy divided by the dissipated energy in the
element. It is as high as 200 for excellent inductors. Unloaded Q increases
with physical size, so modern miniature chip inductors have low unloaded
Q. Finally, Q of the load is a property of a complex termination. For series
impedance it is given simply by
140 Filter Synthesis Using Genesys S/Filter

XL
Qof load = (12.22)
RL

and for parallel admittance it is

BL
Qof load = (12.23)
GL

As the Q of the load increases, the achievable match bandwidth


decreases. In a classic paper, Fano [4] offers an elegant formula for
predicting the relationship between the bandwidth and the best achievable
reflection coefficient using a lossless, infinitely complex matching network.
−πQloaded
Qof load
Γmin = e (12.24)

where Γmin is the magnitude of the best reflection coefficient achievable


over a bandwidth f upper − f lower . The return loss of Γmin is

RLdB = −20 log Γmin (12.25)

Useful algebraic derivatives of Fano’s equation are


Qof load ln Γmin
Qloaded = − (12.26)
π
πQloaded
Qof load = − (12.27)
ln Γmin

An octave bandwidth is Qloaded = 1.5 . Eq. 12.25 indicates that the


reflection coefficient of 0.1 is a return loss of 20 dB. From Fano’s equation,
to achieve a 20-dB return loss over an octave bandwidth requires
Qof load less than 2.047. Fano’s formula provides an absolute limit that is
approached with an infinite quantity of elements. Fano’s limit is used to
discover if a solution is possible, thus avoiding effort on an unsolvable
problem.

12.3.2 Example: Power Amplifier Match


A transistor power amplifier output with 9.6 ohms resistance in parallel
with 40 pF is to be matched to 50 ohms from 200 to 400 MHz with a
worse-case return loss of 12 dB, or Γmin of 0.251. The Q of the load from Eq.
12.23 is 0.4825 at 200 MHz and 0.9651 at 400 MHz. Using the higher Q of
the load, from Eq. 12.24, for an octave bandwidth (loaded Q = 1.5), a
Matching with S/Filter 141

reflection coefficient of 0.00758 is theoretically achievable with an infinite


quantity of elements. Design begins with the Specification tab given in Fig.
12.11.

Figure 12.11 Initial Specification tab for the 200- to 400-MHz power-amplifier matching
network.

The matching network must include a shunt capacitor at the input that
is sufficiently large to absorb the 40-pF transistor output capacitance. The
extraction must therefore begin with a shunt element first. Furthermore,
the extraction must result in a topology with a transformer turns ratio less
than unity so that a Norton transform is available to remove the
transformer. The quantity of nonfinite TZs is increased until a satisfactory
extraction sequence is achieved. The lowest degree is six. Since harmonic
suppression is desirable in a power amplifier, the extraction with one TZ at
DC and five at infinity is selected with the sequence ∞ DC ∞ ∞ ∞ ∞.
Finally, to avoid adding an additional capacitor with a Norton
transform, to improve the design margin, and to retain the filtering
functions of the network, the requested passband ripple in the Specification
tab is reduced until the transformer turns ratio in the Extraction tab is
unity. The resulting passband ripple is 0.0282 dB, or a passband return
loss of 22 dB. The schematic for the power-amplifier matching network is
given in Fig. 12.12. Capacitor C1 is replaced with a 24.84-pF capacitor with
the remaining capacitance provided by the transistor output capacitance.
142 Filter Synthesis Using Genesys S/Filter

ZO=9.6Ω ZO=50Ω

L2 L3
L=7.68nH L=17.5nH

L1 C2 C3
L=6nH C=38.51pF C=12.47pF
C1
C=64.84pF

Figure 12.12 Schematic of the 200- to 400-MHz power-amplifier matching network. C1 is


reduced by the transistor output capacitance.

12.3.3 Example: Broadband Antenna Match


This example illustrates a matching network for a monopole antenna. The
design is required to center the resonant frequency at 3.75 MHz, shift the
resistive component up to 50 ohms, and broaden the bandwidth of the
return loss. An R-L-C model for the antenna is 34 ohms in series with a
series L-C of 19.5 uH and 87 pF. The VSWR of this antenna in a 50-ohm
system is given as the solid trace in Fig. 12.13. The VSWR is a radial-scaled
parameter often used with antennas. It is related to the magnitude of S11 by
the following equation:
1 + S11
VSWR = (12.28)
1 − S11

The minimum VSWR in this example is not unity because of the


mismatch of the 34-ohm antenna resistance to the 50-ohm system. The
minimum occurs when the L-C components of the model resonate.
Resonance does not occur at band center because the monopole antenna is
electrically shorter than a quarter wavelength at 3.75 MHz.
Matching with S/Filter 143

Figure 12.13 VSWR responses of the antenna before (solid trace) and after matching (dashed
trace).

The goal is to achieve a VSWR under 3:1 across the frequency span of
3.5 to 4.0 MHz. Design begins with the specification of the 50-ohm source
and 34-ohm load, frequency cutoffs of 3.5 and 4.0 MHz, and a passband
ripple of 1.25 dB that corresponds to a VSWR of 3:1. The series L-C
component of the model is incorporated into the final series branch of the
matching network. The quantity of TZs is increased and an extraction
sequence is selected so that the 3:1 VSWR is achieved over the desired
band, and the matching network contains a final series L-C branch with the
inductor greater than the antenna 19.5 uH, and a capacitor smaller than the
antenna 87 pF. The extraction sequence selected is DC∞ DC ∞ DC ∞. Then
a Remove Transformer transform is applied to set the output resistance at
34 ohms. The final matching network could be realized by absorbing the
antenna L-C into the matching network values. However, to eliminate the
matching network capacitor and to afford margin in the design, the ripple is
reduced to 1.0776 dB to improve the VSWR maximum and the cutoffs are
broadened to 3.48 and 4.02 MHz to widen the frequency response. This
sets the capacitor C4 to exactly match the antenna capacitance. The network
is shown in Fig. 12.14. The capacitor C4 is absent in the final network and
the inductor is 20.857 uH minus 19.5 uH, or 1.357 uH. If the antenna is
physically lengthened, this inductor is eliminated as well. The final
response is given as a dashed trace in Fig. 12.13.
144 Filter Synthesis Using Genesys S/Filter

C1 C2 C4
ZO=50Ω C=59.02pF C=32912.06pF C=87pF

L1 L3
L=30677.88nH L=20857.56nH ZO=34Ω

L2 C3
L=313.57nH C=7002.93pF

Figure 12.14 Matching and broadbanding network for the antenna. Refer to the text regarding
L3 and C4.

References
[1] R. Rhea, The Yin-Yang of Matching: Part 1 – Basic Matching Techniques, High
Frequency Electronics, March 2006.
[2] R. Rhea, The Yin-Yang of Matching: Part 2 – Practical Matching Techniques,
High Frequency Electronics, April 2006.
[3] P. Smith, Electronic Applications of the Smith Chart, SciTech Publishing,
Raleigh, NC, 1995.
[4] R. Fano, “Theoretical Limitations on the Broadband Matching of Arbitrary
Impedances,” Jour. Franklin Institute, January, 1950.
13 Distributed Filters
Distributed filters use TLEs to realize the transfer function rather than
using lumped inductors and capacitors. The reactance of ideal lumped
elements is a linear function of the frequency variable ω , and practical
elements exhibit this behavior over a wide bandwidth. When electrically
much shorter than a quarter-wavelength, TLEs behave much like lumped
reactors. The behavior of longer TLEs are a periodic function tan θ , where
θ is the electrical length of the TLE. Exact synthesis techniques are
available for distributed filters by replacing the linear frequency variable ω
with the periodic frequency variable tan θ . This is referred to as Richards’
transform [1]. Minnis is an excellent review of exact synthesis for
microwave circuits using the Richards transform [2].
Therefore, two different design methods are used for distributed filters.
One assumes electrically short TLEs, and it takes advantage of tan θ ≈ θ for
small θ . Although approximate, this method should not be dismissed.
Filters constructed with short TLEs provide better stopband performance.
Several practical filter topologies are easily designed using this approach.
Furthermore, while the initial synthesis process using the Richards
transform is exact, when realized with real transmission lines,
discontinuities result in inexact realization. For any method of design,
trade-offs and tuning or optimization are required.
The second method, exact synthesis using Richards’ transform,
requires all TLEs to be of equal length, referred to as commensurate. This
synthesis often results in series TLE stubs that are realizable only in
wireline, and not microstrip or stripline. This difficulty is resolved using
transforms and an additional type of TLE, the unit element. Mastery of
these techniques requires practice. Later chapters include many examples
to illustrate these techniques.

13.1 Comparing Distributed and Lumped Filters


The decision to use lumped or distributed filters is typically based on
frequency. Lumped filters are often employed below approximately 1 GHz,
and distributed filters are selected for microwave frequencies. However,
only a careful analysis of the requirements determines the optimum
approach. Lumped filters are effectively used at microwave frequencies
when small size is paramount, and distributed filters are sometimes used at
lower frequencies when low loss is critical.

145
146 Filter Synthesis Using Genesys S/Filter

Distributed filters offer a number of realization advantages that make


them attractive for certain applications:
1) Higher unloaded Q is achievable.
2) Element tolerance is often tighter, depending primarily on the
substrate dielectric constant and dimensional tolerances. This
eliminates tuning in some applications.
3) The absence of lumped elements and solder connections improves
reliability.
4) In certain applications, the absence of embedded elements is an
advantage.
However, TLEs also possess some significant disadvantages:
1) Higher unloaded Q is achieved only with larger physical size than
lumped elements [3].
2) Periodic behavior destroys stopband bandwidth. This is often the
limiting factor for distributed filters.
3) With larger size, box modes may further degrade stopband
performance.
4) Design is complicated by restricted TLE parameter range. For
example, the minimum and maximum practical values of TLE
characteristic impedances are roughly 20 and 120 ohms, respectively.
For coupled TLE, extremely close spacing may be required.
In summary, lumped filters are often limited by tuning requirements or
losses associated with finite element Q, while distributed filters are often
limited by size and poor stopband performance.

13.2 The Genesys Microwave Filter Module


Overcoming the previously listed TLE disadvantages typically requires
clever and unintuitive design steps and transformations. This can be
daunting to both the novice and experienced designer. A rich history of
modern-method solutions exist using specific topologies. An excellent
review of methods developed prior to 1964 is available in the classic book
Microwave Filters, Impedance-Matching Networks, and Coupling
Structures [4]. This book has served as an important reference for 50 years.
These methods use adaptations of the modern method and do not require
direct synthesis techniques. As such, they do not offer the flexibility of
direct synthesis, but for all-pole distributed topologies, they work well and
involve straightforward methods.
The Microwave Filter module of Genesys automates the design of
topologies derived from those reviewed in [4]. Alternative solutions can be
designed and compared in minutes with the click of a mouse button. Four
of the eight distributed bandpass topologies automated by Microwave
Distributed Filters 147

Filter are given in Fig. 13.1. The microstrip metal patterns are
approximately 3x actual size. These are five-section 6- to 6.2-GHz all-pole
bandpass filters with a 0.0432-dB passband ripple on the Rogers TMM6
25-mil thick substrate.

Figure 13.1 Four of the eight different distributed bandpass filter topologies automatically
designed by the Microwave Filter module of Genesys.

The edge-coupled topology is larger and therefore less desirable than


the folded version, referred to as hairpin. Each TLE resonator is electrically
a half-wavelength long. End-effect capacitance requires physical shortening
of the resonators. As the frequency is increased, the resonators become
shorter and folding into hairpins is impractical.
Some designers avoid the interdigital because of concern about the
affects of the via holes. However, the affects of via-hole discontinuities is
less than the open-end discontinuities. Open ends are particularly onerous
as they serve as radiators, more so than other discontinuities. Interdigital
resonators are only a one-quarter wavelength in length, resulting in a more
compact size with reduced box modes in a smaller housing. The most
148 Filter Synthesis Using Genesys S/Filter

compact topology is the combline. It requires capacitive loading of each


resonator. In this example, each of the resonators is 600 in electrical length.
Each of the filters shown in Fig. 13.1 utilizes transformer coupling at
the input and output. The Microwave Filter module also supports tapped
coupling to the first and last resonator, along with other options to
customize these filters. These topologies may be designed using S/Filter,
but many steps are involved. A more effective use for S/Filter is the design
of topologies not automated by the Microwave Filter module.
Given in Fig. 13.2 are four different five-section lowpass filters designed
by Microwave Filter with a cutoff frequency of 2.4 GHz on the same TMM6
substrate material. A minimum TLE characteristic impedance of 24 ohms
and a maximum TLE characteristic impedance of 120 ohms are specified.
The scale is approximately 3x actual size.

Figure 13.2 Four different distributed lowpass topologies automatically designed by the
Microwave Filter module of Genesys.

The stepped-Z filter cascades high-impedance TLEs acting like series


inductors with low-impedance TLEs acting like shunt capacitors. The stub
lowpass is similar except that parallel open stubs less than 900 long act like
the shunt capacitors. The elliptic lowpass has two FTZs with series
inductors and series L-C branches in shunt. The stepped penetrating
topology is described in detail later in this chapter.
Distributed Filters 149

13.3 Distributed Synthesis Concepts


This section covers fundamental concepts involved in the design of
distributed element filters using both short-line approximations, and exact
synthesis using Richards transform.

13.3.1 TLEs
Distributed synthesis utilizes six basic TLEs as shown in Fig. 13.3. A TLE in
the transmission path is referred to as a unit element (UE). It does not add
a TZ, but it may have an effect on the transfer function.

Figure 13.3 The six basic elements of distributed synthesis.

Each specified nonfinite TZ and each UE adds one degree to the filter
transfer function and one TLE to the topology. Each FTZ adds two degrees
to the filter and three TLEs to the topology. The open wireline and shorted
wireline are not realizable in a planar process such as microstrip. This is
resolved by transformation to one of the other basic elements.
TLE elements are defined by their characteristic impedance, Z0, and
electrical length θ . It is sometimes more convenient to use TLE
characteristic admittance, Y0. The characteristic impedance and electrical
length are related to the distributed inductance and capacitance per unit
length by

Z0 = L (13.1)
C
150 Filter Synthesis Using Genesys S/Filter

θ rad = ω × len LC (13.2)

LC = µε = (µ 0 µ r )(ε 0ε r ) =
µrε r (13.3)
c2
where len is the physical length of the line and c is the speed of light in a
vacuum. With µ r = 1 ,

θ 0 ≅ 1.2 × 10 −9 lenmm f ε r ≅ 3.05 × 10 −8 leninch f ε r (13.4)

εr
Z0 = (13.5)
cC
A parallel-coupled line (PCL) that is symmetric with equal self-
impedance lines is defined by the even-mode ( Z 0e ) and odd-mode ( Z 0o )
characteristic impedance and the even and odd-mode electrical length. For
pure TEM-mode propagation, θ e = θ o . This holds for stripline (triplate) and
coupled rods between ground planes (slabline), but is only approximately
valid for microstrip. Useful coupled-line relationships are

Z 0 = Z 0e Z 0o (13.6)

Z 0e − Z 0o
kv = (13.7)
Z 0e + Z 0o

Z 0e − Z 0o
kw = (13.8)
2 Z 0e Z 0o

13.3.2 Richards Transform


The Richards transform does not change the structure of a distributed
circuit, but rather it is a mathematical transformation of the linear
frequency variable to a form based on a trigonometric function. This
facilitates network synthesis of distributed filters. Given in Fig. 13.4 are the
input reactances of shorted and open stubs versus frequency. The
independent axis is scaled in multiples of the one-quarter wave frequency,
fq. The periodic behavior of the stubs is evident. Higher frequency
resonances are referred to as reentrance modes. This reentrance limits the
stopband performance of distributed filters.
At fq, the shorted stub has infinite reactance. This behavior repeats at
odd multiples of fq. Over limited bandwidth, this behavior mimics a parallel
L-C network to ground. The reactance of a parallel L-C network is plotted
as a dashed trace from 0.5 to 1.5 fq. At fq, the open stub has zero reactance.
This behavior also repeats at odd multiples of fq. Over limited bandwidth,
this behavior mimics a series L-C network to ground. The reactance of a
Distributed Filters 151

series L-C network is plotted as a dashed-dot-dot trace from 0.5 to 1.5 fq.
The shorted and open stubs reverse behavioral roles at even multiples of fq.
The behavior of ideal lumped elements is a linear function of frequency.
For example, the reactance of an inductor is ωL, where ω = 2πf . The
behavior of TLEs is a function of tan θ . Therefore, it is useful to define a
new variable for synthesis, Ω = tan θ . The Richards transform relates these
two frequency variables by
 πf 
Ω = tan  (13.9)
 2 fq 
 

Figure 13.4 Input reactance of a shorted stub (solid trace) and an open stub (dashed trace).
Also plotted near fq are the reactance of series and parallel L-C networks to ground.

The frequency mapping is illustrated in Fig. 13.5 using the transmission


amplitude responses of a fifth-degree generalized lowpass filter with 40-dB
Amin synthesized with ideal lumped elements (solid trace) and with ideal
TLEs (dashed trace). In this design, the lowpass cutoff frequency is 1/3 fq.
For lowpass, the center of the second passband occurs at 2 fq, so specifying
a higher fq increases the frequency of the second passband and improves
the stopband bandwidth. However, as fq is increased, the characteristic
impedance of the TLE eventually become too extreme to realize.
152 Filter Synthesis Using Genesys S/Filter

Figure 13.5 Responses of a synthesized lumped (solid trace) and distributed (dashed trace)
lowpass.

13.3.3 Kuroda Identities


A sampling of transforms based on Kuroda identities are illustrated in Fig.
13.6. The Kuroda Wireline Transfers: Full: Series Shorted Left (Fig. 13.6a)
converts the unrealizable series-shorted wireline, Z1a, into a structure with a
conventional open-stub TLE, Z3a. A similar series shorted right form is
available. All TLEs must be commensurate length. The characteristic
impedances of the transformed TLEs are higher than the original values.
The transformed characteristic impedances are:
Z 0a
na2 = 1 + (13.10)
Z1a

Z 2 a = na2 Z1a (13.11)

Z 3a = na2 Z 0 a (13.12)

For example, the application of this Kuroda transform to a series TLE


of 30 ohms cascaded with a shorted wireline of 80 ohms, results in n2 of
1.375, a series TLE characteristic impedance of 41.25 ohms, and an open
stub of 110 ohms.
Fig. 13.6b illustrates the shifting of a series open-right element to the
left of the UE. The transform is the Kuroda Wireline Transfer: Full: Series
Open Right. The defining formulas are:
Z1b
nb2 = 1 + (13.13)
Z 0b
Distributed Filters 153

Z 0b
Z 2b = (13.14)
nb2

Z1b
Z 3b = (13.15)
nb2

Fig. 13.6c illustrates that a transform may be applied to a portion of a


series wireline. This results in a second wireline on the mirrored side of the
series TLE. S/Filter offers the option of specifying the transformer turns
ratio, or the characteristic impedance of the resulting wireline or series
TLE. Depicted in Fig. 13.6c is the option Kuroda Wireline Transfer: Equal:
Series Open Right that results in two wirelines with equal characteristic
impedance. A series open left form is also available. Transforms that create
equal-valued wirelines on each side of a TLE are of the proper format to be
later transformed into coupled combline and interdigital filter structures.
The defining formulas for the Kuroda Wireline Transfer: Equal: Series
Open Right transform are:

Z1c
nc = 1 + (13.16)
Z 0c

Z 0c
Z 2c = (13.17)
nc

Z1c
Z 3c = Z 4c = (13.18)
nc (nc + 1)

Fig. 13.6d illustrates the Kuroda Wireline Transfer: Equal: Shunt


Grounded (shorted) Left, which creates a symmetric pair of shorted TLE
stubs from a single shorted TLE stub. A similar shunt shorted right form is
available. Formulas are:

Z 0d
nd = 1 + (13.19)
Z1d

Z 3d = Z 5d = nd (nd + 1)Z1d (13.20)

Z 4 d = nZ 0 d (13.21)
154 Filter Synthesis Using Genesys S/Filter

Figure 13.6 Sample of Kuroda-derived transforms.


Distributed Filters 155

Synthesis and later transforms may create a TLE pi network with


unequal shunt shorted TLE stubs as depicted on the left in Fig. 13.6e. The
transform Kuroda Wireline Transfer: Specify Transformer: Shunt Shorted
Right (or Left) is used to equalize the shunt shorted TLE stubs. Symmetry
about the series TLE is required to realize coupled-line pairs of equal width.
The transfer prompts for n. The following value equalizes the stubs.

Z 0e Z 2e + Z1e Z 2e
ne = (13.22)
Z 0e Z1e + Z1e Z 2e

Z 0e Z1e
Z 3e = (13.23)
Z 0e + Z1e − ne Z1e

Z 0e
Z 4e = (13.24)
ne

Z 0e
Z 5e = (13.25)
ne (ne − 1)

When the structure is simplified, Z5e and Z2e parallel combine to one
stub with characteristic impedance equal to Z3e.

13.3.4 Ikeno Transforms


When the synthesis of a distributed filter results in TLE open stubs with an
impractically high characteristic impedance, the Ikeno transform is used to
moderate the characteristic impedance of the TLEs. When the
characteristic impedance of Z2 in Fig. 13.7 is impractically high, the Ikeno
transform replaces Z1 with coupled lines and Z2 with an open stub with
lower characteristic impedance. The transform depicted in Fig. 13.7 is the
Coupled Lines: Ikeno Lines: Open, Open Stub. The dual transform Open
Stub, Open is the left/right mirror of the illustrated form. The descriptor
refers to the terminations present on the nonthrough ports of the resulting
coupled line.

Figure 13.7 Ikeno transform to resolve excessively high characteristic impedance of Z2.
156 Filter Synthesis Using Genesys S/Filter

Invoking this transform in S/Filter prompts the user for a parameter


Zc1. This is the even-mode characteristic impedance of the resulting coupled
lines, Z0e. The dialog indicates the minimum and maximum allowable
values. The maximum value is twice the characteristic impedance Z1. The
maximum and minimum values are:
Z 0e: max = 2Z 0 (13.26)

(
Z 0e: min = Y1 + Y0 − Y1 (Y1 + Y0 ) )
−1
(13.27)

where the admittances Yn are the inverse of the impedances Xn. Specified
values for Zc1 near the minimum value result in an open-stub TLE
characteristic impedance near 0 ohms. Specified values for Zc1 near the
maximum value result in an open-stub characteristic impedance near the
high-original value, and an unrealistically low value of Z00. If the
characteristic admittance of the final open-stub TLE is Ys, then with

YsY1 (Y0 + Ys )(Y0 + Y1 )


A= (13.28)
Ys − Y1

the even- and odd-mode characteristic impedances are


−1
 Y (Y + Y ) 
Z 0e =  s 0 1 − A  (13.29)
 Ys − Y1 
−1
 1 
Z 0o =  + 2 A  (13.30)
Z
 0e 
The Ikeno transform dialog then prompts for a value for n, which must
range from 1.0001 to 1000. Higher numbers result in lower values of the
characteristic impedance of the resulting open stub.
1
Z stub ∝ (13.31)
n2
Next, a message is returned with the characteristic impedance of the
resulting open stub. The user then indicates if they wish to change the stub
impedance. When Yes is selected, new values of Zc1 and n are requested.
The open stub characteristic impedance is reduced by the square of the
ratio of the new value of n to the original value of n. Iteration of the
specified values of Zc1 and n generally results in realizable values. Finally,
the relative dielectric contstant of the lines is prompted.
Distributed Filters 157

13.3.5 Kuroda-Minnis Transform


The Kuroda-Minnis transform is used to convert the shorted and open
wirelines in Fig. 13.8a into realizable elements. Several intermediate steps
are automatically invoked by S/Filter, with the result given as Fig. 13.8e.
Normally, the left three elements of this final result are later converted to a
combline PCL. S/Filter begins the Kuroda-Minnis transform by splitting
the shorted wireline Z2 into two elements. The value of the left element is
specified using Eq. 13.32. Next, a Kuroda Wireline Transfers: Full: Series
Shorted Right is applied to Z1 resulting in Fig. 13.8c. Next, Z2b and Z3 are
swapped and a Kuroda Wireline Transfers: Full: Series Open Right is
applied to Z9, resulting in Fig. 13.8d. Finally, the transformer is shifted to
the right and a Kuroda Wireline Transfers: Full: Series Shorted Right is
applied to Z8 resulting in Fig. 13.8e.
The resulting element values are

a + b 2 − 4ac
Z 2a = (13.32)
2a
where
a = 2 Z1 + Z 2 + Z 3 (13.33)

(
b = 2 Z12 + Z1Z 3 ) (13.34)

c = − Z1Z 2 Z 3 + Z12 Z 2 (13.35)

then

 Z 
Z 4 = 1 + 1  Z1 (13.36)
 Z 2a 

 Z 
Z 9 = 1 + 1  Z 2 a (13.37)
 Z 2a 
Z3
nT 1 = 1 + (13.38)
Z9

Z9
Z8 = (13.39)
nT 1

Z3
Z5 = (13.40)
nT 1
158 Filter Synthesis Using Genesys S/Filter

Figure 13.8 Kuroda-Minnis transform steps.


Distributed Filters 159

Z 2b
Zx = (13.41)
nT21

 Z 
Z 6 = 1 + 8  Z 8 = Z 4 (13.42)
 Zx 
Z 7 = Z x + Z8 (13.43)

The Kuroda-Minnis transform in S/Filter allows the user to split the


series open-wireline Z3 before applying the transform. The parameter
option k is the impedance fraction applied to the new Z3 used in the
transform. The remaining fraction, (1-k)Z3, becomes a series open-wireline
that remains after the transform.

13.3.6 Half-Angle Transform


Given in Fig. 13.9 are half-angle transforms. These transforms replace lines
with lines of one-half the original length. When applied to a subset of lines,
the filter is no longer commensurate and subsequent compound operations
such as Equate All Open Stubs are voided. Nevertheless, half-angle
transforms may result in better realization, and when combined with
original length lines, the stopband performance is improved.
In each case
θ1
θ 2 = θ3 = (13.44)
2
then the line characteristic impedance are
Z 2 a = Z 3a = Z1a (13.45)

Z1b
Z 2b = Z 3b = (13.46)
2
Z 2c = Z 3c = 2 Z1c (13.47)

Z1a
Z 2 d = Z 3d = (13.48)
2
Z 2e = Z 3e = 2 Z1e (13.49)

Z1 f
Z2 f = (13.50)
0.5 2 + 1

(
Z 3 f = Z1 f 0.5 2 + 1 ) (13.51)
160 Filter Synthesis Using Genesys S/Filter

Figure 13.9 Half-angle transforms.


Distributed Filters 161

13.3.7 Interdigital Transform


Given in Fig. 13.10a is the transform Coupled Lines: Interdigital Lines:
Open, Open, which converts two open wirelines and a UE to an interdigital
PCL. The formulas for the symmetric PCL with Z1 equal to Z3 are:
Z 0e = Z1 + 2Z 2 (13.52)

Z 0o = Z1 (13.53)

Figure 13.10 (a) Interdigital and (b) combline transforms.

13.3.8 Combline Transform


Fig. 13.10b shows the transform Coupled Lines: Comb Lines: Open, Open,
which converts a wireline and two open stubs to a combline PCL. The
formulas for the symmetric PCL with Z1 equal to Z3 are:
Z 0e = Z1 (13.54)

Z1Z 2
Z 0o = (13.55)
2 Z1 + Z 2
162 Filter Synthesis Using Genesys S/Filter

13.4 Lumped to Distributed Equivalent Transforms


Current flowing longitudinally in a TEM-mode transmission line develops
an encircling magnetic field that impedes the current flow. The
transmission line thus possesses series inductance. The electric field
developed in the insulating material between the center conductor and the
ground represents capacitance. This incremental inductance and
capacitance is uniform along the length of the transmission line, thus the
term “distributed.” Fig. 13.11 illustrates the relationships between short
TLE parameters and the equivalent lumped-element circuit. The
distributed and lumped parameters are related by:
θ
2πfL1 = Z 0 tan (13.56)
2
1 Z
= 0 (13.57)
2πfC1 sin θ

When the line is narrow, the inductance predominates and the


capacitance is ignored. In this case, the series inductance of the model is
the sum, or twice the value of L1. When the line is wide, capacitance
predominates and the inductance is ignored. In summary, a narrow line
appears inductive and a wide line appears capacitive.
TL1 L1 L1

Ζο
θ C1

Figure 13.11 Short TLE and lumped-element equivalent model.

Fig. 13.12 illustrates the relationships between short TLEs configured


as open or shorted stubs. When the stub is shorted, the TLE appears like an
inductor to ground. When the stub is open, the TLE appears like a capacitor
to ground. The TLE and lumped parameters are related by:
2πfL2 = Z 0 tan θ (13.58)

1 Z
= 0 (13.59)
2πfC3 tan θ
Distributed Filters 163

Figure 13.12 Short TLE shorted and open stub and lumped-element equivalent model.

Fig. 13.13 illustrates the relationships between resonant TLEs and


lumped-element equivalents. A shorted TLE that is electrically 900 long
appears like a parallel L-C to ground. An open TLE that is electrically 900
long appears like a series L-C to ground. Finally, a TLE in series with the
transmission path that is 1800 long appears like a series L-C. The parameter
relationships are:
4Z 0
2πfL5 = (13.60)
π
πZ 0
2πfL6 = (13.61)
4
πZ 0
2πfL7 = (13.62)
2
1
2πfLx = (13.63)
2πfC x

These circuits are only approximately equivalent. The accuracy of the


short transmission-line equivalence improves with shorter line lengths, and
the resonant TLE and lumped equivalents are exact only at the design
frequency f. Nevertheless, these relationships are useful for designing
distributed filters by beginning with a lumped-element circuit and
converting these elements into TLEs.
164 Filter Synthesis Using Genesys S/Filter

Figure 13.13 Resonant TLE and lumped-element equivalent model.

13.5 Inverters
An ideal impedance inverter behaves like a transmission line that is a
quarter-wavelength long at all frequencies [4]. If the line with characteristic
impedance K is terminated in an impedance Zb at one end, the impedance
seen looking into the other end is Za where

K2
Za = (13.64)
Zb

Likewise, for an admittance inverter of characteristic admittance J,


then

J2
Ya = (13.65)
Yb

Inverters are used to obtain more realistic element values by scaling the
impedance level of a filter. Inverters are also used to transform a parallel
resonator to a series resonator, and vice versa. This can be used to convert a
conventional bandpass filter into a topology with all parallel or all series
resonators. The process of inserting an inverter is exact and does not alter
the responses of the filter. However, ideal inverters do not exist. Inverters
are replaced with realizable structures that approximate an ideal inverter
over a limited bandwidth. For bandpass filters, the approximation is
adequate for a wide variety of applications. One realizable structure is a
transmission line of characteristic impedance K or admittance J that is a
quarter-wavelength long at the center frequency of a bandpass filter.
Certain pi, tee and “L” networks constructed using inductors or using
Distributed Filters 165

capacitors behave as inverters over wider bandwidth than do quarter-


wavelength long lines. Some of the element values of these L-C networks
are negative, but these networks are successfully used by absorbing the
negative values into adjacent elements. Formula for replacing inverters
with realizable networks are given in [4]. S/Filter includes over 30
transforms for inserting and replacing inverters. These transforms are
accessed through the Inverter folder in the Transform tab. This book
includes numerous examples of inverter use for both lumped and
distributed filters.

13.6 The Convert Using Advanced TLine Routine


Distributed synthesis generates a schematic with ideal transmission lines
defined by electrical parameters. The Genesys environment includes the
Convert Using Advanced TLine routine that automatically converts the
schematic to physical models for the TLEs. The routine handles all
transmission line processes supported within Genesys. The dialog that
displays when this routine is invoked is given in Fig. 13.14. This routine has
options to
1) Select from substrate definitions included in the schematic or libraries.
2) Select any or all of the TLEs.
3) Select the desired manufacturing process.
4) Using these first three options, to mix processes within the schematic.
5) Automatically add DisCos, which are discontinuities with parameters
that automatically track with adjacent lines that are tuned or
optimized.
6) Absorb DisCos into adjacent lines to attempt to preserve the response.
For example, the line terminated by an open end is shortened to
correct for the loading by the end-effect capacitance.
The combination of these features significantly simplifies the process of
transitioning a design from synthesis to a real design.
166 Filter Synthesis Using Genesys S/Filter

Figure 13.14 Dialog for the Advanced TLine routine.

13.7 Box Modes


A conductive housing that encloses a filter is a cavity. At the frequencies of
cavity resonance, TLE discontinuities such as open-ends, bends, and steps
easily couple energy into and out of the cavity. At these resonant
frequencies there is little attenuation from filter port to port, regardless of
the ideal response of the filter. The result is total destruction of the stop
bands. Between resonant frequencies, the ultimate stopband attenuation is
severely limited. Larger housings lower the resonant frequencies and
degrade the stopbands at lower frequencies. This is described analytically
in Chapter 19.

13.8 Introduction to Distributed Filter Examples


The next four chapters include numerous examples of distributed lowpass,
highpass, bandpass, and bandstop filters. The filters in these chapters were
chosen for their variety and to illustrate concepts applicable to similar
structures. Until the designer has considerable experience with distributed
filters, the appropriate initial extraction, and which transforms must be
applied, are far from obvious. An appropriate starting point for the novice
is by using an example from these chapters that is similar to the target
filter.
Distributed Filters 167

References
[1] P.I. Richards, “Resistor-Transmission-Line Circuits,” Proc. IRE, Vol. 36,
February 1948.
[2] B. J. Minnis, Designing Microwave Circuits by Exact Synthesis, Artech House,
Norwood, MA, 1996.
[3] R. Rhea, “Technology Enables New Components,” Microwave Journal,
November 2006.
[4] G. Matthaei, L. Young, and E.M.T. Jones, Microwave Filters, Impedance-
Matching Networks, and Coupling Structures, Artech House, Norwood, MA, 1980.
14 Distributed Lowpass Filters
This chapter covers the design of both all-pole and generalized
distributed lowpass filters by both exact and approximate methods. This
chapter also illustrates techniques for reducing the physical size of the
filters to potentially reduce box-mode degradation of the stopbands.
In this chapter, each example lowpass filter is fifth-degree designed
with a 0.0432-dB passband ripple and a cutoff frequency of 2400 MHz.
They are implemented on microstrip with a 50-mil-thick Rogers TMM6
substrate with quarter-ounce copper.

14.1 Exact Methods


This section describes distributed lowpass filters that are initially
synthesized using exact methods. This initial synthesis forms a foundation
for the later realization of practical forms. The initial synthesis is also an
important academic exercise. In this section, each example lowpass is
carried through the design process to filters realized with microstrip or
coaxial elements.

14.1.1 Lowpass with Redundant UEs


The design is based on Richards transform and utilizes commensurate-
length series and stub TLEs. Design begins with the Specification tab given
in Fig. 14.1. Notice that the Distributed radio button is selected and the ¼
Wave Freq (fq) for the TLEs is specified as 6000 MHz. To result in the
widest possible stopbands, this frequency should be as high as possible
without resulting in unrealizable TLE characteristic impedance. Unless
otherwise noted, extreme minimum and maximum characteristic
impedances of 20 and 120 ohms are used in this book. The ¼ Wave Freq
may require later iterative adjustment to optimally satisfy the minimum
and maximum impedances.

169
170 Filter Synthesis Using Genesys S/Filter

Figure 14.1 Specification tab for the exact lowpass filter with series and stub TLEs.

The resulting filter is shown in Fig. 14.2a. The characteristic


impedance, physical length in millimeters, and the effective relative
dielectric constant for each TLE is listed. The electrical length is 900 at 6
GHz, or 360 at the 2.4-GHz cutoff frequency. Unfortunately, the series wire-
line shorted stubs are unrealizable. To eliminate the series wire-line
shorted stubs, the following transforms are applied:
1) Two series 50-ohm transmission lines are added at the input and
output of the filter. Since the characteristic impedance of these lines
equals the termination impedance, they add only phase shift and do
not influence the amplitude responses. Their length must equal the
TLE commensurate length in Fig. 14.2a. This results in Fig. 14.2b.
2) Kuroda Wireline Transfer: Full: Shunt Open Right transform is
applied to TL2 in Fig. 14.2b. This converts TL2 and the open stub TL3
into a series wireline shorted stub and a transmission line. This creates
TL2 and TL3 in Fig. 14.2c.
3) Similarly, a Kuroda Wireline Transfer: Full: Shunt Open Left is
applied to TL8 in Fig. 14.2b.
4) The four unrealizable series wire-line stubs are eliminated by applying
four Kuroda Wireline Transfer: Full: Series Shorted Right or Series
Shorted Left transforms to each pair of series transmission line and
series-shorted stub beginning with TL1 and TL2. The resulting
structure is given in Fig. 14.2d.
Distributed Lowpass Filters 171

Figure 14.2 Transformation steps for the exact lowpass filter with series and stub TLEs.

The filter does not quite meet the desired minimum and maximum
characteristic-impedance limits in that TL1 and TL9 exceed the desired
maximum TLE characteristic impedance of 120 ohms. The ¼ Wave Freq
parameter could be increased to decrease the characteristic impedance of
TL1, but this has the affect of decreasing the characteristic impedance of
TL5.
The amplitude responses of the filter are given in Fig. 14.3. As can be
seen, the design is exact as evidenced by a cutoff at precisely 2.4 GHz and
an equal passband return loss of 20 dB. Also evident is a primary limitation
172 Filter Synthesis Using Genesys S/Filter

of distributed filters. TLE reentrance results in an additional passband


beginning at 9.6 GHz. Actually, a second passband removed by four times
the frequency of the desired passband is good stopband performance for a
distributed filter.

Figure 14.3 Responses of the five-section 2.4-GHz lowpass filter designed in S/Filter based
on Richards’ transform and transformed using Kuroda.

Given at the top in Fig. 14.4 is a schematic of the microstrip realization


of this filter. Fifty-ohm input and output lines are manually added to the
filter to provide interface with housing input and output connectors. This
schematic is created automatically using the Convert Using Advanced
TLine feature of Genesys. The conversion begins with the final electrical
parameters given in Fig. 14.2 and executes the following steps:
1) The electrical TLE parameters are converted to physical dimensions for
the user specified manufacturing process, such as microstrip, stripline,
and coax, on the user-specified substrate parameters.
2) If requested, discontinuities are added as necessary at points such as
line junctions, steps, bends, and open-ends. These automatic
discontinuities are called “DisCos.”
3) If requested, the line lengths are adjusted in an attempt to compensate
for the effects of the DisCos.
Notice that the lengths of the TLE are no longer commensurate. The TLE
stubs are shorter because of compensation for the stub microstrip open-end
discontinuities.
Distributed Lowpass Filters 173

Figure 14.4 The 2.4-GHz lowpass schematic (top) as realized in microstrip with dimensions
and discontinuities automatically added by the Genesys feature Covert Using Advanced
TLine. The layout is given at the bottom (2X scale).

Next, Add Layout is applied to the microstrip schematic in the Genesys


workspace tree. The resulting layout is given at the bottom of Fig. 14.4.
Notice the extreme dimensions of the series TLE to TLE stub Tee
discontinuity. In addition, the width of the microstrip open-end is as great
as the length of the TLE stubs. In fact, the Genesys program displays
warnings about layout dimensions.
The response computed by circuit theory simulation is given in Fig.
14.5. The discontinuities required for the realization of this filter
significantly disturb the response. The cutoff frequency is increased to 2.7
GHz, the return loss is degraded to 16 dB, and the reentrance modes are
modified.
Given at the top in Fig. 14.6 is the schematic of the 2.4-GHz filter after
optimization of TLE lengths to correct for the perturbations of the
discontinuities. Variables are assigned to the lengths of TL2 through TL6 to
retain physical symmetry and they are optimized. The optimized layout is
shown at the bottom of Fig. 14.6. Optimization slightly lengthens the filter.
174 Filter Synthesis Using Genesys S/Filter

Figure 14.5 Responses computed by circuit-theory simulation of the of the 2.4-GHz filter
realized in microstrip.

Figure 14.6 The 2.4-GHz lowpass schematic (top) with realized in microstrip after optimization
of dimensions. The layout is shown at the bottom (2X scale).

The response after optimization of line lengths is given in Fig. 14.7.


Optimization goals are a 20-dB return loss from DC to 2.4 GHz with the 40-
dB stopband rejection frequency limits manually adjusted to 4 to 7.2 GHz
to achieve the best available responses.
Distributed Lowpass Filters 175

Figure 14.7 Optimized responses of the 2.4-GHz filter initially designed using exact methods.

14.1.2 Stub TLEs and Contributing Unit Elements


The four 50-ohm TLEs added to the previous lowpass as illustrated in Fig.
14.2b are redundant. Driven or loaded by 50-ohm terminations, they do not
contribute to the selectivity of the filter. Although their impedance is
altered as they transform all elements to realizable form, they remain
redundant. These redundant elements add four TLEs of commensurate
length and therefore increase the physical size of the structure.
Rather than specifying five TZs at infinity, three TZs are specified at
infinity and two unit elements (UEs) are specified. The Specification tab is
shown in Fig. 14.8. If the Series Element First option is selected in the
Extraction tab, the synthesized topology includes three shorted wirelines
cascaded with the UEs. This structure is unrealizable. However, if the
Series Element First option is not selected, the UEs are cascaded with
open-stub TLEs as illustrated in Fig. 14.9a.
The characteristic impedances of the TLEs are less extreme than the
desired 20- to 120-ohm limits. In order to improve the stopband
performance, the ¼ Wave Freq is increased to 6557.5 MHz to increase the
characteristic impedance of TL2 and TL4 to 120 ohms. The result is shown
in Fig. 14.9b. This decreases the impedance of TL3 below 20 ohms. This is
dealt with by splitting TL3 into two shunt elements as shown in Fig. 14.9c.
TL4 is rotated 1800 in the schematic to facilitate realization of the structure.
The synthesis and transforms to this point are exact.
176 Filter Synthesis Using Genesys S/Filter

Figure 14.8 Specification tab for the lowpass with contributing UEs.

Next, the electrical schematic is converted into microstrip and


discontinuities using Convert using Advanced TLine and short lengths of
input and output transmission lines are added. The results are given in Fig.
14.9d. The conversion is approximate because discontinuities are required
to realize the physical structure.
The synthesized ideal responses of the resulting lowpass are given on
the left in Fig. 14.10 and the simulated responses in microstrip with
discontinuities are given on the right.

14.1.3 Lowpass with Only Contributing UEs (Stepped-Z)


A useful class of distributed filters consists entirely of cascaded
transmission lines of alternating low and high impedance. These are often
referred to as stepped-impedance or stepped-Z filters. They are particularly
important for realization in coaxial line because stubs are inconvenient to
manufacture.
The design for a 2.4-GHz fifth-degree filter begins with the
specification of no TZs at infinity and five UEs. The remaining parameters
in the Specification tab are unchanged. The desired coaxial minimum and
maximum characteristic-impedance targets selected are 10 and 110 ohms.
Shunt element first results in the more desirable initial impedances. The
initial design with a ¼ Wave Freq of 6 GHz is given in Fig. 14.11a. The ¼
Wave Freq is increased to 6.76 GHz so that the maximum characteristic
impedance is 110 ohms with the result given in Fig. 14.11b. The dimensions
Distributed Lowpass Filters 177

after converting with Convert Using Advanced TLine to coaxial elements


are given in Fig. 14.11c.

TL4
W=3.05mm
TL7 TL5 TL8
L=4.85mm
W=1.88mm W=0.18mm W=1.88mm
L=1.45mm L=5.91mm L=1.45mm

ZO=50Ω TL2 ZO=50Ω


W=0.18mm
TL1 TL3 TL6
L=5.91mm
W=3.57mm W=3.05mm W=3.57mm
L=4.8mm L=4.85mm L=4.8mm

Figure 14.9 Design sequence for the 2.4-GHz lowpass with contributing UEs (2X scale).
178 Filter Synthesis Using Genesys S/Filter

Figure 14.10 Initial synthesized responses (left) and microstrip responses (right) of the 2.4-
GHz lowpass with three TZs at infinity and two contributing UEs.

The relatively low ratio of maximum to minimum characteristic


impedance required to synthesize this filter results in very poor stopband
performance. The responses are given in Fig. 14.12.

Figure 14.11 Schematic with (a) a 6-GHz quarter-wave frequency, (b) a 6.76-GHz quarter-
wave frequency, and (c) realized in coax.
Distributed Lowpass Filters 179

Figure 14.12 Responses of the coaxial stepped-impedance 2.4-GHz lowpass.

A manufacturing process that allows a very thin center conductor with


high characteristic impedance, or higher order, improves the stopband
performance of this coaxial filter. Moding in coax is less of an issue than in
housed microstrip, so the additional length of a higher-order filter is less
problematic.
The stepped-impedance lowpass is a convenient structure for many
applications, and it is a common form of distributed-lowpass filters.
Approximate design techniques that result in better stopband performance
are illustrated in Section 14.2.

14.1.4 Generalized Lowpass Filter


Generalized filters offer improved skirt selectivity at the expense rejection
well into the stopband. Because reentrance modes lower the expectation of
stopband depth, the improved skirt selectivity of generalized filters is
particularly attractive.
Consider the following generalized fifth-degree Chebyshev passband
lowpass with one FTZ and 30-dB minimum stopband attenuation. The
initial specification includes a ¼ Wave Freq of 6 GHz, one TZ at infinity,
two UEs, and one TZ at 3490 MHz. The electrical schematic is shown in
Fig. 14.13a. Next, a Kuroda Wireline Transfer: Full: Series Shorted Right
transform is applied to TL1 and TL2, and a mirror Series Shorted Left is
applied to then TL5 and TL6. The transform Tlines: Stepped Resonators:
Finite Zero to Two Step Stub is applied to TL3 and TL4. The result is given
in Fig. 14.13b. The maximum TLE characteristic impedance is 98.25 ohms.
180 Filter Synthesis Using Genesys S/Filter

The ¼ Wave Freq is then increased to increase the maximum


characteristic impedance to approximately 120 ohms. This requires a small
adjustment in the FTZ frequency to retain the minimum stopband
attenuation of 30 dB. These two adjustments typically require a few
iterations to satisfy the maximum characteristic impedance and minimum
stopband attenuation. The final ¼ Wave Freq is 7540 MHz and the FTZ is
set to 3550 MHz. The resulting electrical schematic is given in Fig. 14.13c.
The high ¼ Wave Freq results in good stopband performance for a
distributed filter, as shown on the left in Fig. 14.14.
The electrical schematic is then converted to microstrip using Convert
Using Advanced TLine with the 50-mil TMM6 substrate. Again, 1.45-mm
long 50-ohm microstrip lines are added at the input and output. The
microstrip schematic is given in Fig. 14.13d and the layout is given in Fig.
14.15.
The circuit-theory simulated response for the microstrip form is given
on the right in Fig. 14.14. The ideal stopband performance from 3.6 to 11
GHz of the electrical design is significantly compromised in the microstrip
form. The large microstrip tees required to interface the wide TLE open
stubs at the input and output are largely responsible for this stopband
degradation. The Genesys simulator displays warnings about the physical
dimensions for this filter. Some of the limitations imposed on this design by
the commensurate requirements of Richards transform can be reduced by
the use of approximate methods. Approximate methods are the subject of
Section 14.2.

14.2 Approximate Methods


Previous examples show that the exact aspect of filter design is somewhat
academic because realization of structures in physical form introduces
discontinuities that have a significant effect on the responses. In addition,
the introduction of additional TLEs to resolve unrealizable wire-line
structures results in a filter that is significantly larger than necessary.
Finally, the commensurate requirement of Richards transform may
preclude the use of shorter lines to improve stopband performance. This
section describes methods that are approximate but that avoid the later two
issues.
Distributed Lowpass Filters 181

Figure 14.13 Transform steps for the generalized fifth-degree 2.4-GHz lowpass.
182 Filter Synthesis Using Genesys S/Filter

Figure 14.14 Responses of the electrical schematic (left) and microstrip form (right) of the
generalized lowpass.

Figure 14.15 Layout of the generalized 2.4-GHz lowpass (2X scale).

14.2.1 All-Pole: Equivalent Series TLE and Shorted Stubs

Consider the Series Element First form of the five-section all-pole lowpass
in Fig. 14.2a. The reason for shifting to the Series Element First form is
evident later in this section. The series-shorted TLE stubs are unrealizable,
and in the exact design, redundant UEs are added to transform these
series-shorted TLE stubs. This adds four TLEs of commensurate length,
thus substantially increasing the size of the filter. In this section, the initial
design of the 2.4-GHz lowpass filter uses approximate methods. Design
begins with the same Specification tab except the ¼ Wave Freq is adjusted
to 5858 MHz to set the maximum characteristic impedance of TL3 at 120
ohms. The result is given in Fig. 14.16a. The following transforms are then
applied:
1) Both shunt TLE open stubs are converted to shunt capacitors using
Distributed to Lumped Equivalents: Open Stub to Shunt C at the
cutoff frequency 2400 MHz. These transforms are approximate.
2) Both shunt Cs are split using Basic Operations: Split Shunt Element.
Distributed Lowpass Filters 183

3) Each shunt capacitor is converted to a shunt TLE open stub using


Lumped to Distributed Equivalents: Shunt C to Open Stub with the
Chose Theta option. The conversion frequency is 2400 and the chosen
TLE length is 180. This length causes the open stubs to be FTZs at
12,000 MHz.
4) Each shorted wireline is converted to a TLE using Tlines: Shorted
Wireline to TLine.
These steps result in Fig. 14.16b. The cutoff frequency is reduced to
approximately 2 GHz. This is corrected by optimization later. Using the
Genesys schematic editor, TL3 in Fig. 14.16b is rotated to vertical and
moved to the left to form a cross with TL2. The same change is applied to
the second pair of shunt TLE stubs. Short 50-ohm TLEs are added at the
input and output. The electrical schematic is then converted to microstrip
on TMM6 using Convert using Advanced TLine. The microstrip schematic
is given in Fig. 14.16c.
The responses are given on the left in Fig. 14.17. The discontinuities
have a strong influence on the resulting response. The cutoff frequency is
too low in frequency and the stopbands are degraded. The microstrip
crosses are the primary offenders in this filter. The lengths of the series
TLEs and the stubs, as well as the width of the stubs, are optimized in an
attempt to correct the cutoff and improve the stopbands. Variables for these
parameters are used to force symmetry. This also reduces the number of
variables that must be optimized. The resulting responses are given on the
right in Fig. 14.17.

14.2.2 Stepped Impedance Lowpass


The stepped-impedance lowpass in Section 14.1.3 is designed using an
exact method with cascaded UEs. The resulting commensurate filter has
characteristic impedances that are not as extreme as can be realized, thus
degrading the stopband. In this section, the stepped-impedance lowpass is
designed using approximate methods that result in TLEs with extreme,
user-selectable TLE characteristic impedance.
184 Filter Synthesis Using Genesys S/Filter

Figure 14.16 Series TLE and shorted-stub lowpass designed using approximate transforms.
Distributed Lowpass Filters 185

Figure 14.17 Responses before optimization (left) and after optimization (right) of the
microstrip series TLE and open-stub 2.4-GHz lowpass.

Figure 14.18 Layout of the optimized series TLE and open-stub lowpass (2X scale).

The design begins with the 2.4-GHz lumped element filter given in Fig.
14.19a. The series inductors are converted to series TLEs using Lumped to
Distributed Equivalents: Series L to TLine with a conversion frequency of
2400 MHz and a characteristic impedance of 120 ohms. The shunt
capacitors are converted using Lumped to Distributed Equivalents: Shunt
C to TLine with a conversion frequency of 2400 MHz and a characteristic
impedance of 20 ohms. Again, 50-ohm input and output lines are added.
The electrical schematic is given in Fig. 14.19b and the schematic after
conversion to microstrip on TMM6 is given in Fig. 14.19c. The layout is
shown in Fig. 14.19d at approximately 2x actual size.
Shown on the left in Fig. 14.20 are the responses of the stepped-
impedance 2.4-GHz lowpass with electrical TLE models and on the right
after conversion to microstrip with discontinuities on 50-mil TMM6.
186 Filter Synthesis Using Genesys S/Filter

L1 L2 L3
ZO=50Ω L=3.22nH L=5.97nH L=3.22nH ZO=50Ω

C1 C2
a C=1.82pF C=1.82pF

Figure 14.19 Transform history of the stepped-impedance 2.4-GHz lowpass design by


converting an L-C form to distributed equivalents. The layout is given at the bottom (2X scale).

Figure 14.20 Responses of the TLE stepped-impedance 2.4-GHz lowpass (left) and in
microstrip including discontinuity models (right).

The cutoff frequency is approximately 20% lower than desired. The


microstrip filter is optimized to correct the cutoff and attempt to improve
the stopband performance from 4800 to 9600 MHz. Symmetry is forced
Distributed Lowpass Filters 187

using variables. Narrower lines would improve the stopbands, but the
widths are at the desired minimum widths, so they are not optimized. It is
anticipated that optimization would shorten the TLE lengths to increase the
cutoff frequency and that it would widen TL2 and TL4 to improve the
stopbands, and that is what happens. The optimized microstrip schematic
is given in Fig. 14.21 and the optimized response in Fig. 14.22.

Figure 14.21 Schematic of the microstrip optimized stepped-impedance 2.4-GHz lowpass.

Figure 14.22 Responses of the optimized stepped-impedance microstrip filter.

14.2.3 Generalized Lowpass


In this section, the generalized lowpass filter designed with exact methods
in Section 14.1.4 is designed using approximate methods. Liberated from
the constraints of commensurate lengths, the approximate filter is 6%
shorter and 33% less high. The design begins with a lumped-element filter
with three TZs at infinity and one TZ at 3580 MHz. The resulting schematic
is given in Fig. 14.23a. Next, the series inductors and shunt capacitors are
transformed to TLEs using the Lumped to Distributed Equivalents: Series
L to TLine and Shunt C to Open-Stub.
188 Filter Synthesis Using Genesys S/Filter

L1 L3
ZO=50Ω L=3.72nH L=3.72nH ZO=50Ω
L2
C1 L=1.29nH C3
C=1.32pF C=1.32pF

C2
C=1.53pF

ZO=50Ω TL7 TL2 TL3 TL8 ZO=50Ω


W=1.88mm W=0.18mm W=0.18mm W=1.88mm
L=1.45mm L=5mm TL5 L=5mm L=1.45mm
W=0.18mm TL4
L=1.65mm W=7.33mm
TL1
L=2.81mm
W=7.33mm
L=2.81mm

TL6
W=7.33mm
L=3.29mm
c

ZO=50Ω TL7 TL2 TL3 TL8 ZO=50Ω


W=1.88mm W=0.18mm W=0.18mm W=1.88mm
L=1.45mm L=4.6mm [L2] L=4.6mm [L2] L=1.45mm

TL1 TL5 TL4


W=5.3mm [W6] W=0.18mm W=5.3mm [W6]
L=4.32mm [L1] L=1.55mm [L5] L=4.32mm [L1]

TL6
W=5.3mm [W6]
L=4mm [L6]
d

Figure 14.23 Design steps for the generalized 2.4-GHz lowpass by approximate design.
Distributed Lowpass Filters 189

Unfortunately, the transform routines in S/Filter do not include a


transform for a shunt inductor that is not grounded. Therefore, the
equations in Section 13.4 are used to manually convert L2 to a TLE.
Likewise, C2 is converted to an open stub. The electrical schematic is given
in Fig. 14.23b. This is converted to microstrip on 50-mil TMM6 using
Convert Using Advanced TLine and short 50-ohm input and output lines
are added. The schematic is given in Fig. 14.23c and the responses are given
on the left in Fig. 14.24.

Figure 14.24 Responses of the microstrip-generalized 2.4-GHz lowpass designed by


approximate methods (left) and after optimization (right).

The cutoff frequency is low and the passband return loss is poor.
Optimization is used to correct the passband and to improve the stopband
from 3600 to 9600 MHz. Symmetry is used to retain symmetry. The length
of each TLE and the width of the wide stubs are optimized. The responses
after optimization are given on the right in Fig. 14.24. The layout at
approximately 2x actual size is given in Fig. 14.25.

Figure 14.25 Layout of the microstrip-generalized 2.4-GHz lowpass designed by approximate


methods after optimization (2X scale).
190 Filter Synthesis Using Genesys S/Filter

14.3 Size Reduction by Penetration


Box-mode suppression and PCB space conservation motivate the
miniaturization of all filters. Reconsider the stepped-impedance lowpass in
Section 14.2.2. The length of the low-impedance TLEs contribute to the
structure length. These TLE provide capacitance to ground via their area.
The area of these lines may be retained while the high-impedance series
TLEs penetrate the area and meet near the middle of the low-impedance
TLE. The geometry is envisioned by examining the final layout in Fig.
14.28. Design begins with the lumped-element filter given in Fig. 14.19a.
The following transforms are then applied:
1) The series inductors are converted to 120-ohm TLE at 2400 MHz using
Lumped to Distributed Equivalents: Series L to TLine.
2) The shunt capacitors are converted to 20-ohm open-stubs at 2400
MHz with Lumped to Distributed Equivalents: Shunt C to Open-Stub.
3) Next, 0.54 long by 0.18 mm wide TLEs are manually added to the
schematic between the series TLE and the open-stub TLE.
4) Short 50-ohm input and output lines and discontinuity steps are
manually added to the schematic.
In essence, step two converted the stepped-impedance lowpass into a
filter with two low-impedance stubs. This results in the schematic in Fig.
14.26a. Convert Using Advanced TLine is then used to convert the
electrical design to microstrip on 50-mil TMM6 substrate. The schematic is
given in Fig. 14.26b. When the discontinuities are absorbed into the
microstrip TLE, the line lengths of TL6 and TL9 are modified. They are
manually returned to 0.54 mm in length.
The resulting response is given on the left in Fig. 14.27. The cutoff
frequency is too low. The length of the non-50-ohm TLE are optimized
while forcing symmetry using variables to define the lengths. The
stopbands are optimized to at least 30 dB from 4200 to 8700 MHz. The
resulting dimensions are given in Fig. 14.26c and the responses are given
on the right in Fig. 14.27.
The layout is given in Fig. 14.28. The high-impedance lines do not
simply cascade with the low-impedance stubs since the widths of the low-
impedance stubs do not contribute to the overall length of the filter.
Consequently, the total length of this filter is 16 mm, as opposed to the
29.7-mm length of the conventional stepped-impedance lowpass. An
alternative approach that significantly reduces the height of the filter is to
place both stubs on the same side of the high-impedance lines. In this case,
the stubs almost contact at the sides. This is easily corrected by manually
reducing the width of the stubs slightly before optimization of the filter.
This filter is investigated further using EM simulation in Chapter 19.
Distributed Lowpass Filters 191

Figure 14.26 Electrical TLE schematic of the penetrating double-stub 2.4-GHz lowpass filter.
192 Filter Synthesis Using Genesys S/Filter

Figure 14.27 Penetrating stepped-impedance 2.4-GHz microstrip response before (left) and
after (right) optimization of the line lengths.

Figure 14.28 Layout of the penetrating stepped-impedance 2.4-GHz lowpass (2X scale).

14.4 Radial Stub Lowpass


This unique lowpass requires approximate and manual design methods, but
the resulting filters are compact and effective. Radial stubs are fan-shaped,
open-stub TLEs with the narrow end connected to the through path. Dual
stubs on each side of the through path are referred to as butterfly radial
stubs, or butterfly stubs. Radial stubs offer a well-defined reference plane
and better stopband bandwidth than conventional stubs.
Unfortunately, computing the input impedance of radial stubs requires
a series in Bessel functions [1]. However, the Genesys simulator includes a
radial stub model, so designing filters by tuning or optimizing of manually
placed radial stubs is straightforward. Wide sweep angles, φ , for radial
stubs generally improve performance, but to avoid higher order modes, the
sweep angle is generally specified as 900 or less.
Distributed Lowpass Filters 193

Again, design begins with the lumped-element filter given in Fig.


14.19a. The following steps complete the design:
1) The series inductors are converted to 120-ohm TLE at 2400 MHz using
Lumped to Distributed Equivalents: Series L to TLine.
2) The shunt capacitors are deleted and replaced with microstrip radial
stubs (MRSs) above and below the intersection of the series microstrip
TLEs. Each radial stub is specified with φ = 900 and a length set by a
variable. The Width parameter at the narrow end of the radial stubs is
specified as the width of the series TLE.
3) The radial stubs are connected to the series TLE using microstrip cross
models.
4) In this case, due to symmetry, all radial stubs are equal length, so there
is one variable specified for each length. If the filter is higher order or
is asymmetric, suitable variables are used for each pair and equal sets
of radial stubs are substituted and tuned for best response before the
next shunt capacitor is replaced.
5) Input and output lines are added to interface to connectors.
Optimization of both the radial stub lengths and the lengths of the
series TLE is used to finalize the design. The schematic and layout of the
fifth-degree 2.4-GHz lowpass with radial stubs is given in Fig. 14.29. The
responses are given in Fig. 14.30.

Figure 14.29 Schematic and layout (2X scale) of the microstrip radial stub 2.4-GHz lowpass.
194 Filter Synthesis Using Genesys S/Filter

Figure 14.30 Responses of the microstrip radial stub 2.4-GHz lowpass.

14.5 Hybrid Lowpass


A motivation for the use of distributed filters is often the undesirability of
lumped inductors. The size of lumped inductors must be limited to avoid
destructive parasitic capacitance, but the small size limits the available
unloaded Q. The tolerance of TLEs is often better than lumped inductors.
However, this is generally more of an issue with bandpass filters because
lowpass filters are less critical for tolerance.
Some of the drawbacks associated with lumped inductors are less
severe with capacitors. Modern ceramic chip capacitors are small and have
excellent unloaded Q. Capacitors with low effective series resistance (ESR)
have high unloaded Q. An effective filter solution for some applications is to
use distributed inductors and lumped chip capacitors. These are referred to
as hybrid filters since they employ a hybrid of lumped and TLEs.
Consider the following 2.4-GHz hybrid lowpass filter. Again, design
begins with a lumped-element filter, but a shunt-first extraction is used to
minimize the quantity of distributed inductors, which reduces size and
improves the stopbands. The schematic is given in Fig. 14.31. Then the
series inductors are converted to 120-ohm TLE at 2400 MHz using Lumped
to Distributed Equivalents: Series L to TLine. Convert Using Advanced
TLine converts the electrical schematic elements to microstrip on 50-mil
TMM6 substrate. Footprints of size 0603 are used for the chip capacitors.
Microstrip steps are manually added between the series TLE elements and
the input and output lines. The end capacitors are shifted so their metal lies
on the input and output lines. The footprint metal of middle capacitor adds
Distributed Lowpass Filters 195

extra capacitance that is ignored for this simulation. The responses are
given on the left in Fig. 14.32.
Next, the length of the series TLE and the values of the lumped
capacitors is optimized, the capacitors are then fixed at the nearest
standard values, and a final optimization of the length of the series TLE is
completed. The final microstrip schematic and layout are given in Fig.
14.31. The optimized response is given on the right in Fig. 14.32.

L1 L2
ZO=50Ω L=4.55nH L=4.55nH ZO=50Ω

C1 C2 C3
C=1.29pF C=2.39pF C=1.29pF

TL1 TL2
Z=120Ω Z=120Ω
L=12.1mm L=12.1mm
ZO=50Ω K=1 K=1 ZO=50Ω

C1 C2 C3
C=1.29pF C=2.39pF C=1.29pF

TL3 TL1 TL2 TL4


W=1.88mm W=0.18mm W=0.18mm W=1.88mm
L=1.45mm STP1 L=6.06mm [L1] L=6.06mm [L1] L=1.45mm

ZO=50Ω STP2 ZO=50Ω


C1 C2 C3
C=1.1pF [C1] C=2pF [C2] C=1.1pF [C1]

Figure 14.31 Electrical and microstrip schematics and layout (2X scale) of the hybrid 2.4-GHz
lowpass.

This hybrid lowpass has small size and the good stopband performance.
The excellent stopband performance results partly from the absence of
reentrance modes in the lumped capacitors.
Further size reduction is possible by folding the series TLEs. Given in
Fig. 14.33 is the microstrip schematic resulting from manually folding the
series TLE and adding chamfered corners. An EMPower EM of the
196 Filter Synthesis Using Genesys S/Filter

compact hybrid lowpass is then used to consider the effects of the footprint
metal. The lumped elements are then optimized for the best response. The
results given in Fig. 14.34 illustrate the difficulty in maintaining stopband
performance beyond a few multiples of the cutoff frequency, even using a
hybrid approach. Electromagnetic simulation is described in detail in
Chapter 19.

Figure 14.32 Responses of the 2.4-GHz hybrid lowpass before (left) and after (right)
optimization.

14.6 Distributed Lowpass Summary


The design of multiple forms of a fifth-degree 2.4-GHz lowpass illustrates
design using both exact and approximate direct synthesis. Continuing the
design through realization in coax or microstrip with layout illustrates that
practical issues significantly modify results, and strict adherence to exact-
only methods is overly restrictive. The path to a final filter is not unique,
and the methods described in this chapter are chosen to help the reader
understand potential techniques. The choices of minimum and maximum
line impedance vary widely with manufacturing process. An ability to
maintain tight tolerance on very narrow lines is an advantage for the best
stopband performance. Small physical size is an important parameter that
drove the evolution of examples in this chapter. The designer should
consider unusual topologies, such as radial stub and hybrid lowpass.
Distributed Lowpass Filters 197

Figure 14.33 Microstrip schematic and layout (2X scale) of the compact hybrid lowpass with
lumped capacitors and folded series TLE.
198 Filter Synthesis Using Genesys S/Filter

Figure 14.34 Electromagnetic cosimulation responses of the compact hybrid lowpass.

Reference
[1] F. Giannini, M. Ruggieri, and J. Vrba, “Shunt-Connected Microstrip Radial
Stubs,” IEEE Trans. Microwave Theory and Techniques, Vol. MTT-34, No. 3,
December 1986.
15 Distributed Bandstop Filters
Examination of the responses on the left in Fig. 14.10 reveals that
reentrance modes cause lowpass filters synthesized using the Richards
transform to also possess bandstop responses centered on the ¼ Wave
Freq (fq). Unfortunately, many bandstop filters require narrow stopbands.
This causes unrealizable TLE characteristic impedance. This chapter
includes techniques for managing this problem.
In this chapter, bandstop filters centered at 5 GHz are designed. The
higher frequency is best implemented on thinner substrate, so 25-mil-thick
Rogers TMM6 with half-ounce copper is used for each example.

15.1 All-Pole with Stubs and Contributing UEs


This filter type is illustrated using three bandwidths to illustrate three
different sets of problems and solutions. These designs are concerned with
performance up to a maximum frequency of 10 GHz. A Lowpass type is
selected in the Specification tab.

15.1.1 Wide Bandwidth Bandstop


Consider a bandstop with passbands from DC to 2 GHz and 8 to 10 GHz.
This is about the maximum bandwidth for this type of filter. Wider
bandwidth results in extreme characteristic impedance. In the Specification
tab, a Distributed filter process with a Cutoff of 2000 MHz, a Ripple of
0.0432 dB, and fq of 5000 MHz are entered. Two TZs at infinity and three
UEs are specified. In the Extraction tab, a shunt element first and the
extraction sequence ∞ UE ∞ UE ∞ are chosen. In the Transform tab, 50-
ohm input and output lines of 1 mm in length are added using Basic
Operations: Insert Element: Transmission Line. Because the characteristic
impedance of these inserted TLEs equal the termination resistance, they
have no influence on the response. They offer metal to attach the first and
last stubs, and the input and output connectors. The resulting electrical
schematic is given in Fig. 15.1a and the responses are given on the left in
Fig. 15.2. The responses reveal symmetry about fq, thus providing a
bandstop response with approximately 48 dB of rejection from 4 to 6 GHz.
Greater attenuation or a wider stopband bandwidth is achieved using a
higher passband ripple or higher order.

199
200 Filter Synthesis Using Genesys S/Filter

TL1 TL3 TL5 TL7


W=0.93mm W=0.12mm W=0.12mm W=0.93mm
L=0.48mm L=7.79mm L=7.79mm L=0.48mm

ZO=50Ω ZO=50Ω
TL2 TL4 TL6
W=1.48mm W=3.39mm W=1.48mm
L=6.83mm L=6.49mm L=6.83mm

b
TL1 TL3 TL5 TL7
W=0.93mm W=0.12mm [W3] W=0.12mm [W3] W=0.93mm
L=0.48mm L=7.79mm [L3] L=7.79mm [L3] L=0.48mm

ZO=50Ω ZO=50Ω
TL8 TL9 TL10
W=.12mm W=.12mm W=.12mm
L=.12mm L=.12mm L=.12mm

TL2 TL4 TL6


W=1.48mm [W2] W=3.39mm [W4] W=1.48mm [W2]
L=6.83mm [L2] L=6.49mm [L4] L=6.83mm [L2]

TL1 TL3 TL5 TL7


W=0.93mm W=0.12mm [W3] W=0.12mm [W3] W=0.93mm
L=0.48mm L=7.76mm [L3] L=7.76mm [L3] L=0.48mm

ZO=50Ω ZO=50Ω
TL8 TL9 TL10
W=.12mm W=.12mm W=.12mm
L=.12mm L=.12mm L=.12mm

TL2 TL4 TL6


W=1.39mm [W2] W=3.21mm [W4] W=1.39mm [W2]
L=6.46mm [L2] L=6.27mm [L4] L=6.46mm [L2]

Figure 15.1 Design steps for the wide-bandwidth 2.4-GHz bandstop filter with three TLE open
stubs and two contributing UEs.
Distributed Bandstop Filters 201

Figure 15.2 Responses of the electrical (left) and microstrip (right) forms of the wide-
bandwidth bandstop filter.

Next, the filter is converted to microstrip using Convert Using


Advanced TLine with the resulting schematic in Fig. 15.1b. The responses
are given on the right in Fig. 15.2. The stopband is shifted higher in
frequency and the upper passband is significantly perturbed. This is
primarily caused by the microstrip tee discontinuities associated with
connecting the stubs to the UEs. The low impedance of the open-stubs
causes a significant length increase in the throughpath (UEs) of the filter.
Optimization of filter dimensions to correct these effects is unsuccessful.
The electrical schematic models this connection as occurring at a point.
The next step modifies the filter layout to approximate a point connection.
This is accomplished by adding narrow short TLE between the UEs and the
TLE open stubs as shown in Fig. 15.1c. These 0.12- by 0.12-mm TLEs are so
narrow and short as to approximate a point connection. The resulting
responses are shown on the left in Fig. 15.3. The perturbation is much less
severe, and optimization recovers the responses relatively well as shown on
the right in Fig. 15.3. The passband return loss is improved over the original
but there is a slight loss of attenuation just below 6 GHz. The layout at
approximately four times actual size is shown in Fig. 15.4. The input and
output lines are increased to 2 mm in length to provide clearance of the
TLE open stubs from the sidewall of the housing.
202 Filter Synthesis Using Genesys S/Filter

Figure 15.3 Responses of the wide-bandwidth bandstop filter modified to reduce the effects of
the microstrip tee discontinuities (left) and after optimization to improve the responses (right).

Figure 15.4 Layout of the wide-bandwidth bandstop filter (4X scale).

15.1.2 Moderate Bandwidth Bandstop


This moderate bandwidth bandstop has a lower passband from DC to 3.5
GHz and an upper passband of 6.5 to 13.5 GHz. The design begins similarly
to the wide bandwidth filter except the cutoff is specified as 3.5 GHz. The
electrical schematic is given in Fig. 15.5a. As the bandwidth is decreased,
the series UEs moderate downward in characteristic impedance and the
TLE open stubs increase in impedance. This narrower bandwidth results in
TLE open stubs near the desired maximum characteristic impedance of 120
ohms.
The microstrip schematic after Convert Using Advanced TLine is given
in Fig. 15.5b and the responses are given on the left in Fig. 15.6. The higher
TLE open stub impedance results in narrower line width and therefore
smaller microstrip tees. The resulting responses suffer only moderate
perturbation, so the connection jumpers used in the previous example are
Distributed Bandstop Filters 203

not necessary. The results after optimization of dimensions are given on the
right in Fig. 15.6.

TL1 TL3 TL5 TL7


W=0.93mm W=0.58mm [W3] W=0.58mm [W3] W=0.93mm
L=1mm L=7.41mm [L3] L=7.41mm [L3] L=1mm

ZO=50Ω ZO=50Ω
TL2 TL4 TL6
W=0.13mm W=0.59mm [W4] W=0.13mm
L=7.63mm [L2] L=7.19mm [L4] L=7.63mm [L2]

b
TL1 TL3 TL5 TL7
W=0.93mm W=0.56mm [W3] W=0.56mm [W3] W=0.93mm
L=1mm L=7.18mm [L3] L=7.18mm [L3] L=1mm

ZO=50Ω ZO=50Ω
TL2 TL4 TL6
W=0.13mm W=0.59mm [W4] W=0.13mm
L=7.69mm [L2] L=7.4mm [L4] L=7.69mm [L2]

Figure 15.5 Design steps for the moderate bandwidth bandstop filter.
204 Filter Synthesis Using Genesys S/Filter

Figure 15.6 Responses of moderate bandwidth microstrip bandstop filter before (left) and
after optimization (right).

15.1.3 Narrow Bandstop with Ikeno Transforms


In this example, the cutoff frequency is increased to 4.5 GHz, resulting in
an upper passband beginning at 5.5 GHz. In the Specification tab, two TZs
at infinity and three UEs are specified with the Series Element First option.
The 30-dB rejection bandwidth is approximately 50 MHz centered at 5
GHz. The electrical schematic is given in Fig. 15.7a. A 50-ohm, 1 mm long
UE is added at the input and output to interface with connectors. The
characteristic impedance of the series UEs moderate toward the
termination impedance as the bandwidth is narrowed. However, the TLE
open stub characteristic impedance is 490 ohms. This difficulty is resolved
by using the Ikeno transform described in Section 13.3.4. It replaces a
series UE and a TLE open stub with a commensurate length parallel-
coupled TLE and a TLE open stub of moderate characteristic impedance.
Applying the transform Coupled Lines: Ikeno Lines: Open, Open Stub
on TL2 displays the prompt for Zc1 with a minimum value of 75.0052 and a
maximum value of 113.449 ohms. The entered value is the even-mode
characteristic impedance of the coupled line section. In this case, 85 ohms
is entered. S/Filter then prompts for a value for n between 1.0001 and
1000. The transform is exact for the displayed default value of n of 1.98,
which is accepted in this case. The characteristic impedance of the resulting
open stub is inversely proportional to n. The even- and odd-mode
characteristic impedances of the coupled line section are not influenced by
n. Finally, S/Filter prompts for the relative dielectric constant of the even
and odd modes that are used to compute the physical line lengths. In this
case, 1 is entered for each and the circuit is converted to microstrip in a
Distributed Bandstop Filters 205

later step. The resulting electrical schematic, after applying the Ikeno
transforms to TL1 and TL5 and adding short 50-ohm TLEs at the input and
output, is given in Fig. 15.7b. The resulting responses are shown on the left
in Fig. 15.8.
The resulting microstrip schematic after Convert Using Advanced
TLine is given in Fig. 15.7c. As expected, the microstrip discontinuities
disturb the response. Because the open stubs are narrower than those in
lowpass filters, the tees connecting the open stubs to the through path are
less destructive. Nevertheless, optimization is used to correct the cutoff and
center frequencies of this bandstop filter. Because the width of the center
TLE and coupled lines are similar, they are set equal, and variables are used
to force symmetry during optimization. Because the automatic DisCos for
the corners may pick either adjacent line width, microstrip corner
discontinuities were added and set equal to the width of the coupled lines.
Unfortunately, this width does not equal the width of the narrow open
stubs so the corner discontinuities are geometrically incorrect. This is best
dealt with by electromagnetic simulation as described in Chapter 19. The
microstrip schematic after optimization is shown in Fig. 15.7d and the
responses are shown on the right in Fig. 15.8. The layout is shown in Fig.
15.9. The maximum stopband attenuation is limited by losses in the
microstrip realization of the filter. Folding the open stubs back toward the
filter input and output would reduce the physical height of the filter.

15.2 Generalized Narrowband Bandstop


This section describes a generalized, 20-dB return loss passband, fifth-
degree bandstop with 30-dB minimum attenuation in the stopband
centered on 5 GHz. The cutoff corners are 4 and 6 GHz and the minimum
attenuation stopband extends from 4.89 to 5.11 GHz. Design begins with
the specifications of a distributed lowpass with a cutoff of 4.5 GHz, fq of 5
GHz, ripple of 0.0432 dB, and one TZ at infinity, two UEs, and a FTZ at
4.827 GHz. The selected extraction is a shunt element first with the
sequence UE 4827 UE ∞. The resulting responses are given on the left in
Fig. 15.10 and the schematic is given in Fig. 15.11a.
206 Filter Synthesis Using Genesys S/Filter

Figure 15.7 Design steps for the narrow-bandwidth bandstop filter.


Distributed Bandstop Filters 207

Figure 15.8 Electrical schematic responses of the narrow bandstop (left) and the microstrip
responses after optimization (right).

Figure 15.9 Microstrip layout of the narrow bandstop with optimized dimensions (4X scale).

The resulting geometry and TLE characteristic impedances are


unrealizable. To correct these issues, the following transforms are then
applied:
1) A Kuroda Wire Line Transfer: Full: Series Shorted Right is applied to
TL1. This results in Fig. 15.11b. Later transforms will correct
unrealizable TLE characteristic impedances.
2) A Basic Operations: Half Angle transform is applied to each of the
original TLE. Each TLE is now one-half the original length. The half
angle transform must be applied to each TLE to retain commensurate
length for later transforms. These five transforms create the schematic
shown in Fig. 15.11c.
208 Filter Synthesis Using Genesys S/Filter

Figure 15.10 Responses of the synthesized generalized narrowband bandstop filter (left) and
the microstrip form after modifications and optimization (right).

3) A Coupled Lines: Ikeno Lines transform is applied to each of the four


UEs and its paired finite-zero stub. The first transform is the
Grounded Stub, Open form applied to TL3 with TL1 and TL2. The
transform prompts for a value of n. The value for n displayed in the
title bar, 2.21815, results in a symmetric parallel-coupled TLE, and this
value is advised.
4) The second transform is the Open, Grounded Stub form applied to TL4
with TL5 and TL6. The resulting schematic after all four Ikeno
transforms are applied is given in Fig. 15.11d.
The physical layout for the schematic in Fig. 15.10d is unrealizable
because the adjacent edges of TL4 and TL6 would contact. In addition, the
corner discontinuities are unsymmetric. The following manual steps resolve
these issues and aid physical realization of this filter.
1) The orientations of the second and fourth open-stub TLE are shifted
upward. This prevents edge contact of the second and third open
stubs.
2) The width of all four coupled lines is set to 0.8 mm.
3) Four 0.2-mm-long TLEs of width 0.8 mm are added between each
open-stub TLE and the coupled line to which it connects. These
additional lengths of line shift the stopband downward a small
percentage. A step Disco is added between each short TLE and the
open-stub TLE.
4) 1.6-mm-long TLEs are added to interface the filter input and output.
Distributed Bandstop Filters 209

Figure 15.11 Design steps for the generalized narrowband bandstop.


210 Filter Synthesis Using Genesys S/Filter

The resulting schematic, after optimization of the responses to correct


the effects of the discontinuities and added lines is given in Fig. 15.12. The
responses after optimization are given on the right in Fig. 15.10. The layout
at 4x scale is shown in Fig. 15.13, which illustrates the need to alternate the
open-stub TLEs.

Figure 15.12 Generalized narrowband bandstop filter layout after modifications and
optimization.

Figure 15.13 Layout of the generalized narrowband bandstop (4X scale).


16 Distributed Bandpass Filters
The bandpass filter is a mainstay of electronic systems. As with
lumped-element filters, the distributed bandpass filter often requires a
clever solution to avoid extreme parameter values, particularly with wide or
narrow passbands. This chapter includes a variety of bandpass topologies
to address these unique challenges.

16.1 Tutorials of Bandpass by Synthesis


As mentioned in Chapter 13, which introduced distributed filters, the
Microwave Filter program automates the design of many distributed
topologies. These topologies are historically very successful. Design of these
popular topologies by synthesis requires repeated application of many
transforms. The resulting filters are identical in form to filters
automatically designed in Microwave Filter. Given the additional effort
required by numerous transforms, direct synthesis of these filters is
primarily of academic interest. Therefore, for educational reasons, this
chapter begins with the direct synthesis of two popular distributed
bandpass topologies. The latter portion of this chapter covers structures
that are not designed by classic techniques and that solve important filter
problems, such as reduced size and custom placement of FTZs.

16.1.1 Edge-Coupled Using Richards Transform


This example is a three-resonator edge-coupled bandpass with a passband
from 4.5 to 5.5 GHz. Richards transform mirrors the highpass passband
around fq, which is specified as 5 GHz. A distributed, 0.0432-dB equiripple
highpass synthesis is specified a cutoff of 4.5 GHz. The procedure begins
with one TZ at DC and N+1 unit elements, where N is the quantity of
resonators. The Specification tab is given in Fig. 16.1.

211
212 Filter Synthesis Using Genesys S/Filter

Figure 16.1 Specification tab for the five-section edge-coupled bandpass.

A Series Element First extraction with the sequence UE UE DC UE UE


results in the schematic given in Fig. 16.2a. The responses are given on the
left in Fig. 16.3. This topology is symmetric and the transformer is a
consequence of limited computer precision and the extreme values. Later
transforms that moderate characteristic impedances eliminate the
transformer.
The following transforms are applied:
1) Basic Operations: Split Series Element is applied to TL3.
2) Kuroda Wire Line Transfers: Equal: Series Open Right is applied to
TL2 and then TL1.
3) Kuroda Wire Line Transfers: Equal: Series Open Left is applied to the
two TLEs on the right.
4) Remove Transformer is applied. The result is given in Fig. 16.2b.
5) The middle three series open TLE are split using Basic Operations:
Split Series Element: Two Equal Pieces. The result is given in Fig.
16.2c.
6) Coupled Lines: Interdigital Lines: Open, Open: Symmetric PCL is
applied to each UE. The result is given in Fig. 16.2d.
Distributed Bandpass Filters 213

Figure 16.2 Design steps for the three-section edge-coupled bandpass.


214 Filter Synthesis Using Genesys S/Filter

The even- and odd-mode characteristic impedances are excessively


high. To resolve this, the Source and Load termination resistance in the
Specification tab are iterated downward until reasonable even- and odd-
mode characteristic impedances are obtained. In this case, a termination
resistance of 12 ohms is selected. To transform the termination resistance
back up to 50 ohms, quarter-wavelength, 5-GHz TLEs are added at the
input and output to serve as impedance transformers. Their characteristic
impedance is 12 × 50 = 24.5 ohms. The resulting schematic is given in Fig.
16.2e and the responses are given on the right in Fig. 16.3. The quarter-
wavelength TLEs transform the termination impedance exactly at only 5
GHz, resulting in a degradation in the return loss near the passband edges.

Figure 16.3 Response of the three-section edge-coupled 5-GHz bandpass (left) and with
quarter- wavelength transformers to moderate even- and odd-mode impedance (right).

The top of Fig. 16.4 shows the microstrip schematic after applying
Convert Using Advanced TLine and optimization to recover the return loss.
The bottom of Fig. 16.4 shows the layout at approximately 2x scale. The
responses after optimization are given in Fig. 16.5.
Each resonator is electrically a quarter-wavelength long at 5 GHz,
resulting in the first reentrant passband at three times 5 GHz or, 15 GHz.
However, in microstrip the even- and odd-mode propagation velocities are
slightly different. This results in a spurious passband centered near two
times 5 GHz. This lower reentrance does not occur in pure TEM mode
media such as stripline.
Distributed Bandpass Filters 215

Figure 16.4 Optimized edge-coupled 5-GHz bandpass microstrip schematic (top) and layout
at bottom (2X scale).

Figure 16.5 Responses of the optimized microstrip edge-coupled 5-GHz bandpass.


216 Filter Synthesis Using Genesys S/Filter

16.1.2 Edge-Coupled Using Inverters


The previous example uses exact synthesis techniques up to the point
where input and output quarter-wavelength transformers are used to
moderate the characteristic impedances. In this example of that filter,
inverters are used that are approximate, but the resulting characteristic
impedances are more moderate and the resulting structure is shorter
because quarter-wavelength transformers are not required.
Design begins as a lumped element bandpass with N transmission
zeros at DC and N at infinity. The resulting filter is given in Fig. 16.6a. The
following transforms are then applied:
1) Inverters: Replace Element(s) with Inverter(s) is applied to each
series L-C network. This replaces the series L-C with a cascade of an
inverter, a shunt parallel L-C, an inverter, and a transformer. After
both transforms, a Simplify Circuit removes the transformers.
2) Each of the shunt parallel L-C networks is converted to a 50-ohm TLE
using Lumped to Distributed Equivalents: Parallel LC in Shunt to
Half Wavelength TLine. The resulting schematic is given in Fig. 16.6b.
3) Each of the series TLE is split into two equal length TLEs using Basic
Operations: Split Series Element. A Kuroda Wire Line Transfers:
Equal: Series Open Left is applied to the two TLEs on the right. Next a
series 50-ohm TLE of length 15.075 mm is added at the input and
output. The left half of the resulting symmetric circuit is given in Fig.
16.6c.
4) Each inverter with TLEs on its left and right is replaced with a coupled
TLE using Coupled Lines: Interdigital Lines: Open, Open. A
conversion frequency just below the upper displayed limited is
selected that results in the most moderatly valued coupled lines. In
this case, fo of 4974.9 is entered. The resulting schematic is given in
Fig. 16.6d.
The response of this edge-coupled bandpass is shown on the left in Fig.
16.7. Because inverter transforms are approximate, the return loss is better
than the expected 20 dB and the bandwidth is narrower than expected.
Convert Using Advanced TLine transforms the schematic to
microstrip. After optimization, the microstrip schematic is given in Fig.
16.6e and the resulting responses are given on the right in Fig. 16.7. As with
the previous example, the differential even- and odd-mode propagation
velocities results in a spurious passband near 2x the desired passband.
Comparing the microstrip responses in Fig. 16.5 and 16.7 reveals that the
input and output quarter-wavelength transforms required in the first
example provide some additional selectivity.
Distributed Bandpass Filters 217

L1 L3
ZO=50Ω L=6.78nH C1 L=6.78nH ZO=50Ω
C3
C=0.151pF C=0.151pF

L2 C2
L=0.2915nH C=3.5114pF

a
*

TL3
W=0.65mm [W2]
S=0.22mm [S2]
L=7.09mm [L2] TL4
TL1 W=0.31mm [W1]
W=0.31mm [W1] S=0.18mm [S1]
S=0.18mm [S1] L=7.28mm [L1]
L=7.28mm [L1] TL2
W=0.65mm [W2]
S=0.22mm [S2]
L=7.09mm [L2]
e

Figure 16.6 Design steps for the edge-coupled bandpass designed using inverters.
218 Filter Synthesis Using Genesys S/Filter

Figure 16.7 Responses of the electrical edge-coupled bandpass (left) and in microstrip after
optimization (right).

After adding short 50-ohm TLEs at the input and output, the layout at
approximately 2x scale is given in Fig. 16.8. The coupled-line widths are
somewhat wider in this version of the filter resulting in somewhat less
insertion loss in the passband.

Figure 16.8 Layout of the 5-GHz edge-coupled bandpass designed using inverters (2X scale).

16.1.3 Interdigital Using Inverters


The example illustrates the synthesis of a 4.5- to 5.5-GHz interdigital
bandpass filter with five quarter-wavelength resonators. As with the edge-
coupled bandpass, design uses Richards transform and a passband
mirrored about fq with highpass synthesis. One TZ at DC and N-1 unit
elements are specified, where N is the desired quantity of resonators. The
Specification tab is given in Fig. 16.9. External coupling is via an additional
coupled line section at the input and output, which is referred to as
transformer coupling. These sections are added as inverters and
transforms, thus creating seven parallel quarter-wavelength lines. This
facilitates changing the internal filter impedance level if required.
The extraction sequence UE UE UE UE DC results in the schematic of
Fig. 16.10a. The unrealizable TLE characteristic impedance and
Distributed Bandpass Filters 219

transformer are resolved as transforms are applied to the filter. The initial
responses are given on the left in Fig. 16.11.

Figure 16.9 Specification tab for the five resonator interdigital bandpass mirrored about fq of a
highpass synthesis.

The transforms are:


1) Kuroda Wireline Transfers: Equal: Shunt Grounded Right is applied
in sequence to TL4 and each UE to the left. After Simplify Schematic,
the schematic in Fig. 16.10b results.
2) Inverters: Scale Source and Load by Inserting Inverters. After
Simplify Schematic, the resulting left half is given in Fig. 16.10c. The
lack of precise symmetry is reduced in the next step.
3) The structure developed by the previous steps is not symmetric.
Asymmetry is reduced by a repeated applications of Kuroda Wireline
Transfers: Specify Transformer: Shunt Shorted Left (or Right). The
goal is all equal shunt-shorted TLEs. This is achieved by applying the
above transform to the series TLE that has the largest variation in the
shunt-shorted TLEs. In this case, six transform applications result in
shunt-shorted TLE characteristic impedance within 1% of each other.
More precision is unnecessary because a later optimization is required
for other reasons. The transformer turns ratio is near unity.
4) Next, the transform Coupled Lines: Interdigital Lines: Both Lines
Grounded is applied to each series TLE individually and in turn. The
resulting interdigital lines even- and odd-mode impedance are
recorded, the transform is deleted and the transform is applied to the
next series TLE in turn, and results are again recorded. During this
transform, the option Symmetric PCL (parallel-coupled line) option is
selected.
220 Filter Synthesis Using Genesys S/Filter

Figure 16.10 Initial extraction and transform sequence of the interdigital bandpass.
Distributed Bandpass Filters 221

Figure 16.11 Responses of the synthesized interdigital filter (left) and after optimization of the
multiple-coupled-line model (right).

The resulting interdigital filter with even- and odd-mode impedance,


and Z0 and kw for each PCL is given in Fig. 16.12a. The characteristic
impedance and wireline coupling coefficients are found using Eqs. 13.6 and
13.8.
This structure is then replaced with the Agilent Genesys multiple-
coupled-line model based on parallel-connected wireline models. This
model requires equal characteristic impedance of each section. The
significantly different coupling of the external transformer sections
compromises this assumption. Therefore, an intermediate characteristic
impedance is used, and it and the coupling coefficients are optimized to
obtain the best responses. This optimization also forces symmetry and the
removal of the transformer. The final optimized characteristic impedance
and coupling coefficients are given in Fig. 16.12b, and the resulting
responses are given in Fig. 16.11b.
Given in Fig. 16.13a is the microstrip interdigital bandpass with open-
end and via hole discontinuities after Convert Using Advanced TLine is
applied on TMM6 25-mil substrate. The responses are given on the left in
Fig. 16.14.
222 Filter Synthesis Using Genesys S/Filter

Figure 16.12 Multiple-coupled-line model of the interdigital filter.

The discontinuities result in poor return loss and wider than desired
bandwidth. This is corrected by the optimization of the line widths,
spacings, and the length. The results are given on the right in Fig. 16.14.
The optimized dimensions are given in Fig. 16.13b. The responses are
improved with dimensions near the initial values, but the best responses
require extensive optimization with significantly wider lines. This results in
closer line spacing, and the increased open-end capacitance reduces the
required line lengths.
Distributed Bandpass Filters 223

ZO=50Ω ZO=50Ω

W=0.4626mm
S1=0.1544mm
S2=0.7131mm

7
S3=0.8827mm
L=6.9288mm

ZO=50Ω ZO=50Ω

W=1.8676mm [W]
S1=0.1mm [S1]
S2=0.4111mm [S2]
1

7
S3=0.6113mm [S3]
L=6.3393mm [L]

Figure 16.13 Microstrip form of the interdigital filter created using (a) Advanced TLine and (b)
after optimization.

Figure 16.14 Microstrip form of the interdigital bandpass with initial dimension (left) and after
optimization (right).
224 Filter Synthesis Using Genesys S/Filter

Figure 16.15 Layout of the 5-GHz transformer-coupled interdigital bandpass (2X scale).

16.2 Unique Bandpass Designs


The remainder of this chapter deals with distributed bandpass structures
not available using classic techniques.

16.2.1 Combline with Capacitive External Coupling


As the bandwidth is decreased, the characteristic impedance of the
resonators in a combline becomes lower. In this example, a capacitive
transformer is used to couple the filter to the loads. This allows the
specification of the resonator characteristic impedance. Design of this three
resonator 950- to 1050-MHz bandpass begins with a specification of one TZ
at DC and 2 N − 1 TZs at infinity. The Specification tab is given in Fig. 16.16.
The resulting schematic for the extraction sequence ∞ ∞ ∞ DC ∞ ∞ is
given in Fig. 16.18a and the resulting responses are given on the left in Fig.
16.17.

Figure 16.16 Specification tab for the three-section, capacitive-coupled combline bandpass.
Distributed Bandpass Filters 225

Figure 16.17 Responses of the synthesized combline (left) and after transforms with
capacitors (right).

The following transforms are then applied.


1) Compound Operations: Equate All Shunt Ls or Shorted Stubs creates
two additional shunt-shorted TLE. The resulting schematic is given in
Fig. 16.18b.
2) Transformer Operations: Absorb in Rload removes the transformer,
which has a near-unity turns ratio. Distributed to Lumped
Equivalents: Open Stub to Shunt C replaces the open stubs with
loading capacitance. Finally, Inverters: Scale Source and Load by
Inserting Inverters prepares for the addition of external-capacitive
coupling. The result is given in Fig. 16.18c.
3) Inverters: Replace Inverter with: End Inverters to: Capacitive L adds
the coupling capacitors. Simplify Schematic, then Basic Operations:
Swap Element and Coupled Lines: Comblines: Both Lines Grounded
replaces the series-shorted wireline and adjacent shorted stubs with a
combline section as shown in Fig. 16.18d.
It is unnecessary to transform the second series-shorted wireline since
the result is identical to the first transform. The combline filter derived
from this schematic is manually added to Genesys. The schematic is given
at the top of Fig. 16.19 and the responses are given on the right in Fig. 16.17.
226 Filter Synthesis Using Genesys S/Filter

TL1
ZOE=60Ω
ZOO=42.3471Ω
LENGTH=37.500mm

C1 C5
C=1.5061pF C=1.507pF

TL2
Z=287.8653Ω
L=37.5mm
TL3
C2 C3 C4
Z=60Ω
C=1.9159pF C=3.6679pF C=1.9166pF
L=37.5mm

Figure 16.18 Initial extraction and transform sequences for the external-capacitance-coupled
combline bandpass.
Distributed Bandpass Filters 227

C3
C=3.67pF [C3]

C2 C4
C=1.92pF [C2] C=1.92pF [C2]

C1 C5
C=1.51pF [C1] C=1.51pF [C1]

3
ZO=50.4Ω [Zo]
K1=3.474e-3 [k1]
K2=3.474e-3 [k1]
LEN=37.5mm

C3
C=3.2309pF [C3]

C2 C4
C=2.1314pF [C2] C=2.1314pF [C2]

C1 C5
C=1.2742pF [C1] C=1.2742pF [C1]
1

W=0.882mm [W]
S1=0.9092mm [S]
L=17.42mm [L]

Figure 16.19 Combline form of the bandpass as transformed (top) and after conversion to
microstrip and optimization (bottom).

The schematic after conversion to microstrip on 25-mil-thick Rogers


TMM6 with half-ounce copper is given on the bottom in Fig. 16.19. The
values are after optimization of the responses. The responses are given in
Fig. 16.20. This filter is investigated further using EM simulation in
Chapter 19.
228 Filter Synthesis Using Genesys S/Filter

Figure 16.20 Responses of the optimized combline bandpass in microstrip.

16.2.2 Miniature Bandpass with Contributing UEs


This example uses an equal number of TZs at DC and infinity, and
contributing UEs, with user-selectable commensurate line lengths. This
2250- to 2750-MHz bandpass uses three TZs at DC and infinity, and four
unit elements. Miniature refers to the fact that the selected line lengths are
less than one-eighth wavelength, in this case 37.50 at band center. The
Specification tab is given in Fig. 16.21. The extraction sequence UE UE DC
∞ DC ∞ DC ∞ UE UE results in the schematic given in Fig. 16.22a. This
filter is symmetric and the transformer does not have a unity turns ratio
because of computer precision during synthesis. The transformer is
absorbed into the load resistance in a later step.
The designators referred to in the following description of transforms
are the designators before the application of the transform. The transform
may change designators, and the next transform description uses the new
designators. Because of symmetry, each transform is applied first at the left
and then the right side of the filter. The transforms are:
1) Basic Operations: Swap Element to TL3 and TL4
2) Kuroda Wireline Transfer: Full: Shunt Open Right to TL2
3) Kuroda Wireline Transfer: Full: Shunt Open Right to TL1. This shifts
the shunt-open stub to the far left. The same transforms are applied on
the right side of the filter.
Distributed Bandpass Filters 229

Figure 16.21 Specification tab for the 2250- to 2750-MHz miniature bandpass.

4) Kuroda Wireline Transfer: Full: Shunt Grounded Right to TL3. This


transform is mirrored on the right side.
5) Transformer Operations: Combine All Transformers.
6) Basic Operations: Split Series Element is applied with two-equal
pieces to the open wireline TL6.
7) Basic Operations: Swap Element is applied to TL6 and TL7. The result
of these transforms is given in Fig. 16.22b.
8) Kuroda Wireline Transfer: Full: Series Shorted Right to TL4 and the
mirror element TL8.
9) Basic Operations: Split Series Element is applied with two-equal
pieces to TL6.
10) Kuroda Wireline Transfers: Pi: Pi Shunt Right is applied to TL2 and
the mirror to TL12.
11) Transformer Operations: Combine All Transformers. The results are
given in Fig. 16.22c.
12) Coupled Lines: Comblines: Both Lines Grounded is applied to TL3
and the mirror to TL10 with the Symmetric PCL (parallel-coupled line)
option.
13) Kuroda Wireline Transfer: Full: Series Open Right is applied to TL4
and the mirror to TL7.
14) Transformer Operations: Combine All Transformers.
230 Filter Synthesis Using Genesys S/Filter

Figure 16.22 Transformation steps for the 2250- to 2750-MHz symmetric miniature bandpass.
Distributed Bandpass Filters 231

15) Kuroda Wireline Transfer: Equal: Series Open Right is applied to


TL3 and the mirror to TL9.
16) Transformer Operations: Combine All Transformers. The results are
given in Fig. 16.22d.
17) Coupled Lines: Interdigital Lines: Open, Open is applied to TL4 and
the mirror to TL7 with the Symmetric PCL (parallel-coupled line)
option.
18) Transformer Operations: Absorb in Rload removes the transformer
while changing the load resistance an insignificant amount.
19) Inverters: Scale Source and Load by Inserting Inverters. Enter an
inverter parameter of 40.
20) Inverters: Replace Inverter with: Quarter Wavelength TLine is
applied to each inverter with a quarter-wavelength frequency of 2500
MHz. The result is given in Fig. 16.22e.
The quarter-wavelength lines act as impedance transformers to
moderate the characteristic impedance of the TLE elements. The optimum
transform ratio is a function of the filter bandwidth and is determined
iteratively near the completion of the filter. These transformers are exact
only at band center. The schematic after minor optimization is given at the
top of Fig. 16.23 and the responses are given on the left in Fig. 16.24.

Figure 16.23 Electrical schematic of the optimized miniature bandpass (top) and microstrip
schematic after optimization (bottom).
232 Filter Synthesis Using Genesys S/Filter

The schematic is converted to microstrip on 25-mil-thick TMM6 with


Convert Using Advanced TLine, and it is again optimized to correct for the
effects of the discontinuities. The dimensions after optimization are given
in the bottom schematic in Fig. 16.23 and the responses are given on the
right in Fig. 16.24.

Figure 16.24 Miniature 2250- to 2750-MHz bandpass responses of the optimized electrical
schematic (left) and the optimized microstrip filter (right).

TL1 and TL10 are quarter-wavelength lines and TL5 and TL6 cascaded
are 750 in length. These longer length lines are folded to reduce the overall
length of the filter. A layout at 2x scale with folded lines is given in Fig.
16.25.

Figure 16.25 Microstrip layout of the miniature 2250- to 2750-MHz bandpass with folded lines
to conserve space (2X scale).

The responses of this folded filter, after minor optimization to correct


for bends inserted during the folding is given in Fig. 16.26. The frequency
sweep to 10 GHz illustrates the excellent stopband performance of this
structure. The short elements not only reduce the size of the filter but they
also significantly improve the stopband performance. The use of both
Distributed Bandpass Filters 233

interdigital and combline parallel-coupled lines also contribute to the


excellent stopbands. However, responses in Fig. 16.25 are computed by
circuit-theory simulation and do not include the effects of radiation,
surface, or box modes. This level of stopband performance is difficult to
achieve in practice. This is addressed in Chapter 19.

Figure 16.26 Responses of the folded miniature bandpass through 10 GHz.

16.2.3 Narrow Bandwidth with UEs and an FTZ


This type of filter begins with a commensurate-highpass synthesis utilizing
one TZ at DC, FTZs, and a quantity of UEs greater than the quantity of
FTZs. Synthesis using Richards transform results in a bandpass response
mirrored about fq. Physical symmetry is achieved if the quantity of either
FTZs or UEs is even and the other is odd. Subsequent half-angle transforms
reduce the length of internal lines to one-eighth wavelength. The input and
output impedance transforming lines remain one-quarter wavelength. The
reentrant passband occurs at 3x fq.
This 2400- to 2600-MHz passband example utilizes one TZ at DC, one
FTZ and two UEs for a fifth-degree bandpass with fq of 2500 MHz. The FTZ
is manually adjusted to achieve a minimum stopband attenuation of
approximately 40 dB. Line characteristic impedance is often unreasonable
so inverters are included for later adjustment. The Specification tab is given
in Fig. 16.27.
Fig. 16.28a shows the synthesized bandpass with a series element first
and the extraction sequence UE 2320 DC UE. The turns ratio is near unity,
so the transformer is eliminated by absorbing it into the load.
234 Filter Synthesis Using Genesys S/Filter

Figure 16.27 Specification tab for the narrow bandpass with one FTZ.

The transformations are:


1) Transformer Operations: Absorb in Rload.
2) Inverters: Scale Source and Load by Inserting Inverters.
3) Inverters: Replace Inverter With: Quarter Wavelength TLine is
applied to the input and output inverters. An inverter parameter of 50
ohms does not change the response. This is adjusted later to achieve
realizable characteristic impedance.
4) Kuroda Wireline Transfer: Equal: Series Open Right is applied to TL2
and then TL1. These transforms are also applied on the right side.
Apply Simplify Schematic. The result is given in Fig. 16.28b.
5) Basic Operations: Split Series Element with the Two-Equal Pieces
option is applied to TL3 and then TL11.
6) Coupled Lines: Interdigital Lines: Open, Open with the Symmetric
PCL option is applied to TL2 and each similar set of TLEs. The result
after Simplify Schematic is given in Fig. 16.28c.
7) Basic Operations: Half Angle is applied to TL2 and then TL9.
8) Coupled Lines: Comblines: Open, Open with the Symmetric PCL
option is applied to TL4 and then TL9.
9) Basic Operations: Half Angle is applied to TL5 and TL6.
10) Tlines: Stepped Resonators: Finite Zero to Two Step Stub is applied
to TL5 and TL6, and then TL7 and TL8.
Distributed Bandpass Filters 235

11) The inverters added in step are re-edited for the most reasonable set
of characteristic impedance for the filter. In this case K=85 is chosen.
The result is given in Fig. 16.28d.
In the final layout, one two-step stub is oriented up and the other is
oriented down. The responses after transformations are given on the right
in Fig. 16.29. The quarter-wave impedance transformers are approximate,
but the performance degradation is small because of the narrow bandwidth
of this filter.
The filter is converted to microstrip on Roger TMM6 50-mil substrate
using Convert Using Advanced TLine. The schematic after optimization is
shown in Fig. 16.30. The narrow 8% bandwidth and FTZs near the
passband result in high-sensitivity to line parameters. The shunt stubs are
low impedance and are wide. To avoid the disturbance of a tee
discontinuity with a wide through line, very short and narrow lines are
added between these stubs and the through lines. Optimization requires
close user monitoring to avoid divergence. To avoid divergence, portions of
the circuit are converted using Convert Using Advanced TLine and are then
optimized before additional portions are converted. The resulting layout
approximately to scale is given in Fig. 16.31. The gap created by adding the
0.2- by 0.2-mm lines result is barely visible in the layout. The final
optimized responses are given in Fig. 16.32.
236 Filter Synthesis Using Genesys S/Filter

Figure 16.28 Transform sequence for the narrow bandpass with one FTZ.
Distributed Bandpass Filters 237

Figure 16.29 Initial responses (left) and after transforms (right) of the narrow bandpass with
one FTZ.

Figure 16.30 Microstrip schematic of the 2400- to 2600-MHz bandpass with one-finite FTZ.
238 Filter Synthesis Using Genesys S/Filter

Figure 16.31 Layout of the narrow bandpass (to scale).

Figure 16.32 Responses of the microstrip narrow bandpass with FTZ after optimization.

16.2.4 Penetrating Combline


This bandpass is a progression from combline, to miniaturized stepped-
impedance resonators, and then a penetrating geometry. It offers the
compactness of the combline but does not require lumped-loading
capacitors. This progression is illustrated in Fig. 16.33.
Fig. 16.33a depicts a conventional combline bandpass filter with
capacitive loading. Resonance is achieved with an electrical line length, θ ,
less than 900 by using lumped-capacitance loading. Line lengths of 300 to
600 are typical. This forms a very compact structure. The capacitors replace
much of the electrostatic capacitance of the TLE open ends and the
corresponding capacitive coupling. Coupling is primarily magnetic.
Distributed Bandpass Filters 239

Figure 16.33 Progression of the (a) combline bandpass to (b) stepped-impedance combline
(c) to penetrating combline.

A conventional quarter wavelength transmission line consists of


continuously distributed inductance and capacitance. The inductance is
predominant at the grounded end where the current is high, and the
capacitance is predominant at the open end where the electrostatic-field is
high. If the grounded end is constructed with a narrower line and the open
end is constructed with a wider line, both the inductance and capacitance
are increased, and the resonant frequency is lowered. Such a stepped
impedance resonator is physically shorter than a uniform line and the need
for loading capacitance is eliminated [1],[2].
If the characteristic impedance of the high-impedance section is Z1 and
the low-impedance section is Z2, then
Z2
K= (16.1)
Z1

and resonance is achieved when


tan θ1 tan θ 2 − K = 0 (16.2)

where θ1 and θ 2 are the electrical lengths of the high and low-impedance
sections respectively. The total length, as a fraction of the uniform-
impedance length, is given by
240 Filter Synthesis Using Genesys S/Filter

l fraction =
4
π
tan −1 ( K) (16.3)

For example, with a characteristic-impedance ratio of 0.167, the length


of a stepped-impedance resonator is 49.4% of the length of a uniform-
impedance resonator.
Depicted in Fig. 16.33b is a coupled filter with stepped-impedance
resonators. The end-capacitance of the wide transmission-line open ends
further reduces the required physical length. The narrow high-impedance
sections and the close-proximity low-impedance sections reduce the
magnetic and increase the capacitive coupling. While the loading capacitors
are eliminated, the wider sections increase the width of the filter.
As depicted in Fig. 16.33c, the height of the filter is reduced by
penetrating the high-impedance lines into the low-impedance sections.
Each resonator is formed by loading each high-impedance section with two
low-impedance lines in parallel, thus effectively creating a very-low
impedance section. The pair of short lines that connect each high-
impedance section to the two low-impedance sections is dealt with
manually as a final optimization.
This filter structure is designed in S/Filter using a commensurate-
length bandpass synthesis. The Specification tab for a 950- to 1050-MHz
bandpass is shown in Fig. 16.34.

Figure 16.34 Specification tab for the 950- to 1050-MHz penetrating combline bandpass.

Each line is only half the length of the shortened resonators, so fq is


several times the passband center frequency. In this case, fq is specified at
3750 MHz. This results in excellent stopband performance for the filter.
Distributed Bandpass Filters 241

One TZ at infinity and 2 N − 1 TZs at DC are specified for this topology. The
initial synthesized filter with a shunt element first and the extraction
sequence DC DC DC ∞ DC DC DC is given in Fig. 16.35a and the responses
are given on the left in Fig. 16.36.
The follow transforms are applied:
1) Compound Operations: Equate All Shunt Cs or Open Stubs. This adds
a shunt-open stub in parallel with each shunt-shorted stub, and
equates all the open stubs.
2) Basic Operations: Split Shunt Element with the option Two Equal
Pieces is applied to TL5.
3) Basic Operations: Swap Element is applied to TL4. This prepares TL2,
TL3 and the new TL4 for conversion to a combline PCL. The result is
given in Fig. 16.35b.
4) The same operations are applied to the other internal shorted and
open-stub pair, in this case TL8 and TL9.
5) Coupled Lines: Comb Lines, Open, Open with the Symmetric PCL
option is applied to each series-open line.
6) Transformer Operations: Absorb in Rload is applied to the near-unity
ratio transformer. The result is given in Fig. 16.35c.
7) The characteristic-impedance values are unrealizable. Inverters: Scale
Source and Load by Inserting Inverters with a parameter of 130 ohms
is applied.
8) Inverters: Replace Inverter with Quarter Wavelength TLine at 1000
MHz is applied to each inverter.
The resulting schematic is given in Fig. 16.35d, and the responses are
given on the right in Fig. 16.36.
242 Filter Synthesis Using Genesys S/Filter

Figure 16.35 Transform sequence for the 950- to 1050-MHz penetrating combline filter.
Distributed Bandpass Filters 243

Figure 16.36 Initial synthesized responses of the penetrating combline filter (left) and after the
addition of the quarter-wavelength transformers (right).

The electrical schematic is converted to microstrip on Roger TMM6 50-


mil substrate using Convert Using Advanced TLine. The structure is
unsuitable for realization of the compact form depicted in Fig. 16.33c
because the shunt-shorted stubs and the combline PCL occupy the same
PCB area. This is resolved by adding short-narrow lines between each stub
and PCL. In addition, so that the corners are symmetric, short-narrow lines
are placed in series with the PCL. The left half of the symmetric microstrip
schematic after addition of these lines, and after optimization of the
responses, is given in Fig. 16.37. Symmetry is forced, and all line lengths
and narrow line widths are forced equal, by using variables to define
parameters. After a few trial optimization runs, the width of the narrow
lines is fixed at 0.3 mm and the line lengths at 11 mm.
The responses after optimization are given in Fig. 16.38 and the layout
is given in Fig. 16.39. The long and narrow quarter-wavelength
transmission-line transformers are 38.9-mm long and extend beyond the
figure. A final design could meander these lines. This is an extremely
compact structure for a bandpass filter with a 1-GHz center frequency
without lumped-loading capacitors. An alternative solution to meandering
the impedance transformers could be using lumped capacitors for
impedance transforming. This is covered in Section 16.3.1.
244 Filter Synthesis Using Genesys S/Filter

Figure 16.37 Left half of the symmetric penetrating combline 950- to 1050-MHz bandpass.

Figure 16.38 Responses of the optimized microstrip 950- to 1050-MHz penetrating-combline


bandpass.
Distributed Bandpass Filters 245

Figure 16.39 Layout of the penetrating-combline bandpass (2X scale). The quarter
wavelength transformer lines extend well beyond the figure (see text).

16.2.5 Minnis Class-D Bandpass


A moderately compact filter topology suitable for narrow bandwidth is the
Minnis type-D bandpass [3]. This example is a 975- to 1025-MHz bandpass
with four resonators. The resonators are coupled via combline-type PCL.
When the number of resonators is even, the center-coupling element is an
interdigital-type PCL.
Design begins with one TZ at DC, N TZs at infinity, and N-1 UEs, where
N is the quantity of resonators. The initial synthesized filter with an
extraction sequence ∞ UE DC ∞ UE ∞ UE ∞ and the Series Element First
option is given in Fig. 16.41a. The responses are given on the left in Fig.
16.40.

Figure 16.40 Minnis type-D narrow bandpass filter ideal response (left) and in microstrip after
optimization (right).
246 Filter Synthesis Using Genesys S/Filter

Figure 16.41 Transform steps for the narrowband Minnis Class-D bandpass.
Distributed Bandpass Filters 247

The following transforms are then applied:


1) Kuroda Wireline Transfers: Equal: Series Open Left is applied to TL5
followed by Transformer Operations: Shift Rightmost to the new
transformer.
2) Basic Operations: Swap Element is applied to TL3.
3) Kuroda Wireline Transfers: Full: Series Open Right applied to TL2
and Transformer Operations: Shift Rightmost again to the new
transformer. All transformers are then combined with Transformer
Operations: Combine All Transformers.
4) Basic Operations: Insert Element is used to insert a 50-ohm quarter-
wavelength line at the input. This line has no effect because the source
resistance is 50 ohms. The resulting schematic is given in Fig. 16.41b.
5) Kuroda Wireline Transfers: Minnis: Kuroda-Minnis CD Right is
applied to the new TL1 with a parameter of 0.45. By iterative trial and
error, this value is found to offer a good compromise of the line
impedances throughout the filter. The resulting transformer is shifted
to the far right.
6) Kuroda Wireline Transfers: Full: Series Open Left is applied to TL6
and the resulting transformer is shifted right.
7) Kuroda Wireline Transfers: Minnis: Kuroda-Minnis CD Right is
applied to the new TL5 with a parameter of 0.45. The resulting
transformer is shifted to the far right and Transformer Operations:
Combine All Transformers is applied. The resulting schematic is given
in Fig. 16.41c.
8) The resulting topology is symmetric about TL10. The mirror of the
above transforms are applied from the load end of the filter. If a
transformer remains, its turns ratio is near unity and it is deleted.
9) Coupled Lines: Combline: Open, Open is applied to TL1, TL2, and TL3
with the Symmetric PCL option. The same transform is applied to the
remaining three similar line sets.
10) Coupled Lines: Interdigital: Open, Open is applied to TL5, TL6, and
TL7 with the Symmetric PCL option. The final schematic is given in
Fig. 16.41d.
The narrow bandwidth of this filter increases the sensitivity of the
responses to discontinuities and to the differential velocity of the even and
odd modes of propagation in the coupled lines. The ideal schematic in Fig.
16.41d is converted to microstrip on Rogers TMM6 50-mil substrate using
Convert Using Advanced TLine. Conversion results in a significant
perturbation of the responses, and the large quantity of variables results in
unsuccessful recovery by optimization. Therefore, the circuit is converted
by one symmetric pair of elements at a time, followed by optimization
before the next pair is converted. Symmetry is forced by the use of
variables. The resulting left portion of the symmetric schematic is given in
248 Filter Synthesis Using Genesys S/Filter

Fig. 14.61e. The optimization is unconstrained except the spacing of the


first and last PCLs is set at a minimum of 0.1 mm. The resulting layout is
given in Fig. 16.42. The microstrip responses are given on the right in Fig.
16.40. The first reentrant passband is wide and begins at approximately 2
GHz. The stopbands can be extended by specification of higher fq. This also
reduces filter size but line impedances become difficult to realize.

Figure 16.42 Microstrip layout of the narrowband 1-GHz Minnis type-D bandpass (to scale).

16.3 Hybrid Bandpass


Lumped shunt inductors, series inductors, and shunt capacitors have
equivalent distributed transmission line equivalents. Conspicuously absent
is a distributed form of a series capacitor. Strip-type structures require an
extremely narrow gap to develop significant capacitance. Interdigital
fingers and broadside-coupled structures have capacitance to ground in
addition to their series capacitance. In many instances, combining lumped
capacitors with TLEs is a better solution than fighting the problem. Modern
chip capacitors offer high unloaded Q and good tolerance, generally much
better than is obtainable from chip or wire wound inductors. The
remainder of this chapter offers hybrid solutions containing both TLEs and
lumped capacitors.

16.3.1 Penetrating Combline with Capacitors


This example substitutes lumped-capacitor termination coupling for
quarter-wavelength transmission-line impedance transformers used in
Section 16.2.4. Because the internal structure of this filter uses 240 line
lengths, the quarter-wavelength lines add significant size to an otherwise
very compact filter. The Q and tolerance for coupling elements are not as
critical as resonating elements, so lumped capacitors function well for this
application. Design proceeds as in Section 16.2.4 through transform step 6.
Distributed Bandpass Filters 249

The following transforms are then applied:


1) Basic Operations: Swap Element is applied to TL1 so that the open
stub is the first element. The same transform is applied at the output.
2) Distributed to Lumped Equivalents: Open Stub to Shunt C is applied
to the open stubs adjacent to the input and output.
3) Termination Coupling: Source: Shunt C to Series C is applied at the
input with parameters Choose Theta equal to 240 and fo equal to 1000
MHz. The mirror transform is applied at the load.
4) Lumped to Distributed Equivalents: Shunt C to Open Stub is applied
to both shunt capacitors. The transformers are removed by Simplify
Circuit. The resulting schematic is shown in Fig. 16.40.
The microstrip layout is completed using the techniques described in
Section 16.2.4.

Figure 16.43 950- to 1050-MHz penetrating combline bandpass coupling capacitors.

16.3.2 Generalized Combline Hybrid


This bandpass utilizes the typical capacitive-loaded resonators of combline
filters, as well as interresonator capacitors to create FTZs and capacitor-
termination transformers. This example is a narrow 975 to 1025 MHz four-
resonator bandpass with 0.0432-dB ripple and fq = 2000 MHz. The initial
synthesis specifies one TZ at DC, one TZ at infinity, and a total of N-1 FTZs,
where N is the quantity of resonators. In this case, a FTZ is tuned to 881
MHz and two FTZs are tuned to 1071 MHz to produce 50-dB Amin. An
extraction sequence 1071 881 1071 ∞ DC, with the application of the
transform Compound Operations: Equate All Shunt Ls or Shorted Stubs,
results in the symmetric schematic shown in Fig. 16.44a. The responses are
given on the left in Fig. 16.45. If a transformer results during synthesis, it is
essentially unity ratio and is deleted. The initial termination resistance is
adjusted to 600 ohms to realize a moderate value of impedance for the
shunt-grounded stub.
250 Filter Synthesis Using Genesys S/Filter

The following transforms in S/Filter continue the design process:


1) Distributed to Lumped Equivalents: Open Stub to Shunt C with a
conversion frequency of 1000 MHz is applied to each shunt-open stub.
2) Distributed to Lumped Equivalents: Wireline/Stub to Single L or C
with the option SC Stub and conversion frequencies of 975 and 1025
MHz is applied to each of the three series open/shorted wireline pairs.
This replaces the open wireline with a capacitor while leaving the
shorted wireline.
3) TL1 and C1 are swapped using Basic Operations: Swap Element to
place C1 adjacent to the input. Termination Coupling: Source: Shunt C
to Series C is applied to C1 with n=0.288675 ( = 50 / 600 ). The mirror
transform is applied to the load with n=3.464103.
4) Transformer Operations: Absorb into Rsource is applied to the input
transformer and the mirror transform is applied to the output
transformer. This sets the termination resistances to 50 ohms.
This completes the transforms applied in S/Filter.
With a 1000-MHz passband band center and fq=2000 MHz, the
commensurate electrical lengths are 450. Replacement of TLEs over 300 in
length with series and shunt lumped capacitors results in significant
distortion of the responses. At this point, a partial correction is
implemented by retuning the FTZs to recover Amin to -50 dB. After tuning,
the new FTZ frequencies are 891 MHz and two at 1099 MHz. The responses
are given on the right in Fig. 16.45. This correction is only approximate and
the return loss is not equal ripple. Later conversion steps to combline
distort the response again, and optimization is used then to improve the
responses.
The next design step is conversion to combline. The form of TL1, TL2,
and TL3 is used to convert the filter to combline even- and odd-mode
impedances using Eqs. 13.54 and 13.55. This step is completed manually
because the presence of C3 voids S/Filter’s recognition of three-combline
forming elements. A four-line coupled-line element is placed in a new
schematic, and the characteristic impedance and coupling coefficients are
found using Eqs. 13.6 and 13.8 applied to each three-element shunt-stub
and series wireline set in Fig 16.44b. The lumped capacitors are copied and
placed in the new schematic. The resulting schematic is given in Fig.
16.46a. After optimization, the resulting schematic is given in Fig. 16.46b.
The quantity of parameters involved in the combline description result in a
nonunique solution. Therefore, the characteristic impedance is fixed while
other parameters are optimized. The resulting responses are given on the
left in Fig. 16.47.
Distributed Bandpass Filters 251

Figure 16.44 Transform steps for the generalized combline hybrid filter.
252 Filter Synthesis Using Genesys S/Filter

Figure 16.45 Response of the ideal generalized combline hybrid (left) and after transforms
and retuning of the FTZ (right).

Finally, the electrical-model schematic is converted to microstrip on


Rogers TMM6 50-mil substrate with Convert Using Advanced TLine. The
length is fixed at 18 mm, and after optimization, the dimensions are
rounded to two-significant digits and the series-lumped capacitors are
placed on standard values. Because of the narrow bandwidth of this filter, it
is necessary to tune the resonating capacitors in production. The resulting
schematic is given in Fig. 16.46c and the responses are given on the right in
Fig. 16.47.

16.3.3 Direct-Coupled Bandpass with Capacitors


Wideband bandpass filters often require impractically close-spaced coupled
lines. One solution that avoids coupled lines is the bandpass with quarter
wavelength shorted stubs and quarter-wavelength connecting lines [4]. A
similar but more compact commensurate structure with wider stopbands is
described in this section. In this octave-bandwidth example, open-stubs are
replaced with lumped capacitors, which further extends the stopbands.
Design of this 1000- to 2000-MHz four-resonator, eight-degree
bandpass begins with the Specification tab given in Fig. 16.48. There is 1 TZ
at DC, N TZs at infinity, and N-1 UEs, where N is the number of resonators.
With fq = 3500 MHz, the commensurate line lengths are 38.570. Shorter
lines improve the stopband bandwidth, but element values become
unreasonable. The responses are given on the left in Fig. 16.49.
Distributed Bandpass Filters 253

Figure 16.46 (a) Original combline form of the generalized bandpass, (b) after optimization of
parameters, and (c) after conversion to microstrip and another optimization.
254 Filter Synthesis Using Genesys S/Filter

Figure 16.47 Responses of the optimized electrical-model combline (left) and after conversion
to microstrip and optimization (right).

Figure 16.48 Specification tab for the 1000- to 2000-MHz hybrid direct-coupled bandpass.
Distributed Bandpass Filters 255

Figure 16.49 Ideal responses of the 1000- to 2000-MHz hybrid direct-coupled bandpass (left)
and after the transforms with quarter-wave transformers and lumped capacitors (right).

The initial schematic with a shunt element first and the extraction
sequence ∞ UE DC ∞ UE ∞ UE ∞ is given in Fig. 16.50a. The following
transform steps are then applied:
1) TL3 is split using Basic Operations: Split Shunt Element and the right
element is swapped with the open stub using Basic Operations: Swap
Element.
2) Kuroda Wireline Transfers: Specify Transformer: Shunt Sorted Left
is applied to TL6 with the parameter option n=1.846 which is the
inverse of the existing transformer turns ratio. After Simply Circuit is
applied, the transformer is eliminated and the structure is symmetric
as illustrated in Fig. 16.50b.
3) Compound Operations: Equate All Shunt Ls or Shorted Stubs
produces the schematic in Fig. 16.50c.
4) Inverters: Scale Source and Load by Inserting Inverters with an
impedance parameter of 40 is applied to improve element values.
5) Each inverter is replaced with a quarter-wavelength line using
Inverters: Replace Inverter With: Quarter Wavelength TLine. The
center frequency of 1500 MHz is specified.
6) Each shunt-open stub is replaced with a lumped capacitor using
Distributed to Lumped Elements: Open Stub to Shunt C. The resulting
schematic is given in Fig. 16.50d.
256 Filter Synthesis Using Genesys S/Filter

Figure 16.50 Transform steps for the 1000- to 2000-MHz hybrid direct-coupled bandpass.
Distributed Bandpass Filters 257

These transforms result in the responses given on the right in Fig.


16.49. Most of the transforms are exact, but the quarter-wavelength
transformers and the open-stub to shunt capacitor transforms are
approximate. The wide bandwidth reduces the quality of these transforms.
The return loss is degraded, and the bandwidth is somewhat wide.
However, the lumped-shunt capacitors improve the stopband bandwidth
and significantly reduce the size.
Next, the electrical models for the TLEs are converted to microstrip on
Rogers TMM6 50-mil substrate with Convert Using Advanced TLine. The
resulting schematic is given at the top in Fig. 16.51. Prior to the conversion,
very short lines are manually added between the lumped capacitors and the
junction of the shunt and series lines. This facilitates adding the
discontinuity models at this junction. The values given in the schematic are
after optimization of the line widths and capacitor values to recover the
response. Optimization reveals that the quarter wavelength transformers
tend to be higher in value, suggesting that less transformation is required.
Therefore, their length is manually reduced to the commensurate length to
reduce the size of the final filter, and optimization is restarted. The
responses after optimization are given in Fig. 16.52. The layout is given at
the bottom of Fig. 16.51 at approximately 2x actual size.

Figure 16.51 Hybrid 1000- to 2000-MHz bandpass microstrip schematic at top and layout at
bottom (2X scale).
258 Filter Synthesis Using Genesys S/Filter

Figure 16.52 Optimized responses of the microstrip hybrid 1000- to 2000-MHz bandpass.

References
[1] S. Yamashita and M. Makimoto, “Miniaturized Coaxial Resonator Partially
Loaded with High-Dielectric-Constant Microwave Ceramics,” Trans. MTT-31,
September 1983, p. 697.
[2] R. Rhea, HF Filter Design and Computer Simulation, SciTech Publishing,
Raleigh, NC, 1994.
[3] B. Minnis, Designing Microwave Circuits by Exact Synthesis, Artech House,
Norwood, MA, 1996.
[4] G. Matthaei, L. Young, and E.M.T. Jones, Microwave Filters, Impedance-
Matching Networks, and Coupling Structures, Artech House, Norwood, MA, 1980.
17 Distributed Highpass Filters
The L-C form of highpass filters is shunt inductors alternating with
series capacitors. The series capacitor is difficult to realize in distributed
form. Therefore, the highpass filter is often realized in distributed form by
utilizing the lower transition region of a bandpass filter and accepting a
finite limit to the upper passband frequency. An alternative approach is
using shunt stubs to realize the shunt inductors and using lumped
capacitors, or semilumped elements such as overlaid strips in multilayer
structures. Because of TLE reentrance modes, these hybrid highpass filters
also have a finite passband bandwidth. This chapter considers both
approaches with filters on Rogers TMM6 50-mil-thick substrate material.

17.1 The Hybrid Highpass


This section covers realization of the series capacitance using lumped
capacitors. Because chip capacitors are small and have high unloaded Q,
this is an advantage when the manufacturing process supports lumped
elements. Not only are chip capacitors smaller than distributed structures,
they do not possess reentrance modes.
Two basic approaches are used. The highpass is originally designed as
lumped, and the shunt inductors are replaced with shorted stubs, or the
original highpass is synthesized as distributed, and the resulting series-
open wirelines are replaced with lumped capacitors.

17.1.1 The All-Pole Hybrid: Distributed Synthesis


This example starts with a highpass synthesized as a distributed filter,
extracted with a shunt element first. The Specification tab is given in Fig.
17.1. Because all TZs are at DC, there is only one unique extraction
sequence. Two extraction sequences are listed in the S/Filter table because
the first listed is always the selected sequence. The fq is iteratively increased
until the highest shunt-shorted TLE characteristic impedance reaches the
maximum value that is desired for manufacture. In this case, fq equal to
3870 MHz results in a maximum characteristic impedance of 119.78 ohms.
The resulting schematic is given in Fig. 17.2a and the resulting response
is given on the left in Fig. 17.3. Next, the transform Distributed to Lumped
Equivalents: Wireline/Stub to Single L or C is applied to TL2 and TL4 with

259
260 Filter Synthesis Using Genesys S/Filter

a transform frequency equal to the cutoff frequency, 1000 MHz. The


resulting schematic is given in Fig. 17.2b.

Figure 17.1 Specification tab for the highpass synthesized initially as a distributed filter.

ZO=50Ω ZO=50Ω

TL2 TL4
Z=29.48Ω Z=29.48Ω
L=19.38mm L=19.38mm

TL1 TL3 TL5


Z=119.78Ω Z=64.59Ω Z=119.78Ω
L=19.38mm L=19.38mm L=19.38mm

a
C1 C2
C=2.32pF C=2.32pF

TL1 TL2 TL3


Z=119.78Ω Z=64.59Ω Z=119.78Ω
L=19.38mm L=19.38mm L=19.38mm

b
Figure 17.2 (a) Initial distributed highpass and (b) after transform of the series-open wirelines
to lumped capacitors.
Distributed Highpass Filters 261

The response of the hybrid highpass is given on the right in Fig. 17.3.
The passband degradation above approximately 2500 MHz is caused by
reentrance modes of the shunt TLEs. Synthesis produced commensurate
shunt TLE. Consequently, with the first and last TLE at the maximum
desired 120 ohms, the middle TLE characteristic impedance is more
moderate. Extreme characteristic impedance improves passband
bandwidth. Therefore, the alternative approach of starting with a lumped
design is considered next.

Figure 17.3 Responses of the initial synthesized distributed highpass (left) and after transform
of the series-open wirelines to lumped capacitors (right).

17.1.2 The All-Pole Hybrid Highpass: Lumped Synthesis


In this case, the synthesis setup is the same except that the Process is
specified as Lumped. The resulting schematic is given in Fig. 17.4a. The
transform Lumped to Distributed Equivalents: Grounded L to Grounded
Stub is applied to L1, L2, and L3. The Choose Theta option is selected, and a
θ of 240 at 1000 MHz is entered for L1 and L3. A θ of 140 at 1000 MHz is
entered for L2. An advantage of starting with a lumped synthesis is during
the transform of the shunt inductors to TLE, the shortest possible length
can be used for each transform. The lengths are not necessarily
commensurate. The resulting responses are shown on the right in Fig. 17.5.
The initial distributed synthesis results in a 20-dB return-loss passband to
approximately 2700 MHz, while the transform from an initial lumped-
element design results in a passband to approximately 3300 MHz.
262 Filter Synthesis Using Genesys S/Filter

ZO=50Ω ZO=50Ω

C1 C2
C=2.32pF C=2.32pF
L1 L2 L3
L=8.19nH L=4.42nH L=8.19nH

T1
P=1
a S=1

C1 C2
C=2.32pF C=2.32pF

TL1 TL2 TL3


Z=115.62Ω Z=111.33Ω Z=115.62Ω
L=19.99mm L=11.66mm L=19.99mm

b
C1 C2
C=2.32pF C=2.32pF

TL1 TL2 TL3


W=0.21mm W=0.24mm W=0.21mm
L=9.05mm L=4.73mm L=9.05mm

Figure 17.4 (a) The initial lumped highpass, (b) after shunt inductors transformed to TLE, (c)
after conversion of the TLE from electrical to physical, and (d) the final layout at 2X scale.
Distributed Highpass Filters 263

Figure 17.5 Responses of the initial lumped highpass (left) and after transform of the shunt
inductors to TLEs (right).

17.1.3 The Hybrid Highpass with UEs


In this example, UEs are added to the highpass illustrated in Section 17.1.1.
This adds length to the filter, but improves the selectivity and the stopband
performance. Synthesis begins with extracting a shunt element first, an odd
quantity of TZs at DC, and N-1 UEs, where N is the quantity of TZs at DC.
The extraction sequence that places a UE on each side of a series-open
wireline is selected. To obtain the best stopband performance, fq is
increased until the maximum TLE characteristic impedance approaches the
maximum value it is desired to manufacture. In this case, fq is specified as
3650 MHz. The resulting synthesized schematic for N equal to five is given
in Fig. 17.6a and the resulting responses are given on the left in Fig. 17.7.
Next, the series-open wirelines are converted to lumped capacitors at
1000 MHz using the transform Distributed to Lumped Equivalents:
Wireline Stub to Single L or C. The resulting schematic is given in Fig.
17.6b and the responses are given on the right in Fig. 17.7.
264 Filter Synthesis Using Genesys S/Filter

Figure 17.6 (a) Initial synthesized highpass with UEs, (b) after replacing the wirelines with
lumped capacitors, and (c) values after optimization. (d) Further optimization allowing shorter
TL2 through TL6 further improved the passband bandwidth.
Distributed Highpass Filters 265

Figure 17.7 Initial response of the distributed highpass with UEs (left) and after transforming
the series-open TLE to capacitors (right).

The transform has significantly degraded the passband bandwidth.


This is recovered by optimization of the capacitor values, and the
characteristic impedance of the TLEs TL2 through TL6. During the initial
optimization, it was noted that TL1 and TL7 tended to be higher than the
desired maximum TLE characteristic impedance, so these lines are fixed at
120 ohms and removed from the optimized variables. Symmetry is forced
by using variables during optimization. The result is given on the left in Fig.
17.8 and the final values in Fig. 17.6c.
The moderate characteristic impedance of TL2 through TL6 suggests
that more extreme impedances with shorter TLE lengths might extend the
passband bandwidth. Therefore, the lengths of these TLEs are added to the
optimized variables. The result is successful in extending the passband to
3500 MHz, as given on the right in Fig. 17.8, and is the widest passband of
the highpass filters designed so far. The final values are given in Fig. 17.6d.
266 Filter Synthesis Using Genesys S/Filter

Figure 17.8 Responses after optimization of the commensurate-hybrid highpass filter (left),
and extended passband bandwidth by optimization with shorter TL2 through TL6 (right).

17.1.4 The Hybrid Highpass with an FTZ


This example uses three TZ at DC and a FTZ at 665 MHz. Design begins
with a lumped synthesis, a shunt element first, and the extraction sequence
DC 665 DC DC. The schematic is given in Fig. 17.9a and the responses are
given on the left in Fig. 17.10.
The shunt inductors L1 and L3 are transformed to shorted TLE stubs
using Lumped to Distributed Equivalents: Grounded L to Grounded Stub
with the Choose Z0 option. A maximum TLE characteristic impedance of
120 ohms is chosen to obtain the shortest feasible TLE length. Because of
the presence of C2, S/Filter does not recognize L2 as a grounded inductor.
Therefore, L2 is manually replaced with a TLE using Eq. 13.56. Again, a
characteristic impedance of 120 ohms is used. The TLE lengths and
capacitor values are then optimized to recover the response. The resulting
schematic and values are given in Fig. 17.9b and the responses are given on
the right in Fig. 17.10. This hybrid highpass is compact and has the best
passband bandwidth of the hybrid highpass filters previously considered.
Distributed Highpass Filters 267

C1 C3
ZO=50Ω C=2.83pF C=2.83pF ZO=50Ω

L2
L1 L=6.89nH L3
L=8.13nH L=8.13nH

C2
a C=8.31pF

C1 C3
C=3.04pF [C1] C=3.04pF [C1]

TL3
TL1 Z=120Ω TL2
Z=120Ω L=15.94mm [L3] Z=120Ω
L=16.71mm [L1] L=16.71mm [L1]

C2
b C=8.05pF [C2]

Figure 17.9 Initial schematic and transformed hybrid highpass with a FTZ.

Figure 17.10 Initial responses of the 1000-MHz cutoff lumped highpass filter (left) and after
transformation of the inductors to TLEs (right).
268 Filter Synthesis Using Genesys S/Filter

17.2 Purely Distributed Highpass


The highpass filters in this section are designed as purely distributed
structures without the use of lumped capacitors.

17.2.1 Highpass with Three TZs at DC and a UE


Consider the Specification tab in Fig. 17.11. The initial schematic is given in
Fig. 17.12a and the initial responses are given on the left in Fig. 17.13. The
required transformations are then:
1) Kuroda Wireline Transfers: Equal: Series Open Right is applied to
TL2.
2) Simplify Schematic removes the transformer.
3) Coupled Lines: Interdigital Lines: Open, Open is applied to TL3 with
the Symmetric PCL option. The result is given in Fig. 17.2b.
4) The transformer, if present, is near unity ratio and is deleted.

Figure 17.11 Specification tab for the 1000-MHz highpass with three TZs at DC and one UE.
Distributed Highpass Filters 269

ZO=50Ω ZO=50Ω

TL3
TL2 Z=125.8251Ω
Z=74.2723Ω L=31.25mm
L=31.25mm
TL1 TL4
Z=72.3222Ω Z=194.8436Ω
L=31.25mm L=31.25mm

a T1
P=1
S=0.6091

Figure 17.12 Transformations of the 1000-MHz highpass with three TZs at DC and one UE.
270 Filter Synthesis Using Genesys S/Filter

Figure 17.13 Initial responses of the 1000-MHz highpass with three TZs at DC and one UE
(left) and after optimization of the microstrip form (right).

17.2.2 Highpass with Three TZs at DC and Four UEs


In this example, the selectivity of the previous highpass filter is increased
by increasing the quantity of UEs to four. Design begins with the extraction
sequence DC UE UE DC UE UE DC with a series element first. The
responses are given in Fig. 17.14. During the application of the following
transforms, if a transformer appears with a near-unity turns ratio, that
transformer is deleted. Pairs of nonunity ratio transformers are eliminated
using the Simplify Circuit transform. The initial schematic is given in Fig.
17.15a. The following transforms are applied.
1) Kuroda Wireline Transfers: Equal: Series Open Left is applied to TL2.
2) Kuroda Wireline Transfers: Equal: Series Open Right is applied to
TL7.
3) Simplify Schematic removes the transformers.
4) Basic Operations: Split Shunt Element is applied to TL5 with equal
elements.
5) Kuroda Wireline Transfers: Equal: Shunt Grounded Right is applied
to TL4.
6) Kuroda Wireline Transfers: Equal: Shunt Grounded Left is applied to
TL8.
7) Simplify Schematic removes the transformers.
8) Coupled Lines: Interdigital Lines: Open, Open is applied to TL2 and
then TL8 with the Symmetric PCL option.
The transformer, if present, is near unity ratio and is deleted.
Distributed Highpass Filters 271

Figure 17.14 Responses of highpass with three TZs at DC and four UEs.

Figure 17.15 (a) Initial schematic of the highpass with three TZs at DC and four UEs and (b)
after transforms that eliminate the series-open wirelines.
272 Filter Synthesis Using Genesys S/Filter

17.3 The Highpass Synthesized as a Bandpass


Consider the response of the octave bandwidth bandpass filter in Fig. 16.52.
Through 2000 MHz, the response is highpass and the filter may be used as
such. For bandpass filters, increasing the quarter wavelength frequency, fq,
of the elements extends the upper stopband bandwidth. This is not
necessary for highpass filters, and this provides a degree of freedom. A wide
bandpass may therefore serve as a highpass filter up to a certain frequency.
The filter design resulting in Fig 16.52 places only one TZ at DC, which
reduces the low side selectivity and the usefulness of this structure for a
highpass filter. The ideal bandpass used as a highpass places more TZs at
DC. These are the factors impacting the design of highpass filters as wide
bandpass filters.

17.3.1 Hybrid Highpass from an Eighth-Degree Bandpass


Consider the Specification tab in Fig. 17.16. Synthesis begins with a
distributed bandpass with four TZs at DC, two TZs at infinity, and two UEs.
A wide passband is specified to increase the upper highpass passband
frequency limit. The fq is selected as 4000 MHz to moderate the values of
the realizable TLEs. The extraction sequence, UE DC UE DC ∞ ∞ DC DC,
results in a unity turns-ratio transformer and is selected. After deleting the
transformer, the initial schematic is given in Fig. 17.17a and the responses
are given on the left in Fig. 17.18. Notice that the frequency span is
increased from the previous examples to 6000 MHz. This example has
excellent passband bandwidth.

Figure 17.16 Specification tab for the highpass filter synthesized as an eighth-degree
bandpass.
Distributed Highpass Filters 273

Unrealizable TLE such as series wirelines are generally dealt with in


distributed design by transformation to coupled-line elements. However,
with the wide bandwidth desired for highpass filter design, excessively tight
coupling values are encountered. For this example, a hybrid approach is
used by replacing the series-open wirelines with lumped capacitors. The
characteristic impedance of TL6 is excessively high, but this is managed
later.

Figure 17.17 Transformation steps for the hybrid highpass designed as an eighth-degree
bandpass.
274 Filter Synthesis Using Genesys S/Filter

The series-open wirelines are transformed to lumped capacitors using


Distributed to Lumped Equivalents: Wireline/Stub to Single L or C. The
series-shorted wireline is transformed to a series TLE using TLines:
Shorted Wireline to TLine with a chosen characteristic impedance of 60
ohms. This value is arbitrary and is chosen close to the characteristic
impedance of the adjacent TL3. The characteristic impedance of TL6 is too
high to realize. Therefore, this element is first transformed to a shunt
capacitor using Distributed to Lumped Equivalents: Open Stub to Shunt C,
and this capacitor is then converted to an open stub with a characteristic
impedance of 60 ohms. The resulting responses are given on the left in Fig.
17.19. Although the transforms have significantly degraded the return loss,
the 10-dB return loss to 5000 MHz suggests this filter could potentially
have excellent passband bandwidth.

Figure 17.18 Initial responses of the hybrid highpass designed as an eighth-degree


bandpass.

Therefore, the characteristic impedance of the commensurate length


TLEs, the length of the 60-ohm TLEs, and the capacitor values are
optimized to recover the responses. The results given on the right in Fig.
17.19 reveal excellent passband bandwidth.
Distributed Highpass Filters 275

Figure 17.19 Responses of the eighth-degree highpass after transformation of the wireline
TLEs (left) and optimization (right).

17.3.2 Hybrid Highpass from a 10th-Degree Bandpass


The previous example results in the widest passband of any of the highpass
filters considered thus far. However, the selectivity below the cutoff is
rather poor. To increase selectivity, a 10th-degree filter is designed. The
Specification tab is similar to the previous example except the quantity of
TZs at DC is increased to six, the upper cutoff is increased to 4000 MHz,
and fq is increased to 5000 MHz. The selected extraction is a shunt element
first with the sequence UE DC ∞ DC DC UE ∞ DC DC DC. The resulting
schematic is given in Fig. 17.20a and the initial responses are given on the
left in Fig. 17.21.
The transformer turns ratio is 0.99, so close to unity that the
transformer is deleted, and the slight degradation of the responses is
recovered by the final optimization. As with the previous example, the
series-open wirelines are transformed to series-lumped capacitors, and the
series-shorted wirelines are transformed to 60-ohm series TLEs. The
results are given in Fig. 17.20b. The resulting responses appear to have the
potential for an extremely wide passband, so the optimization goals are set
for a passband to 6000 MHz. Although unconventional, the resulting
highpass has good selectivity and has the best passband bandwidth of the
example filters in this chapter.
276 Filter Synthesis Using Genesys S/Filter

ZO=50Ω ZO=50Ω

TL3 TL4 TL7 TL8 TL10


TL1 Z=11.8Ω Z=21Ω TL6 Z=17.29Ω Z=35.51Ω Z=17.94Ω
Z=37.36Ω L=15mm L=15mm Z=35.56Ω L=15mm L=15mm L=15mm
T1
L=15mm L=15mm
P=1
TL2 TL5 TL9
S=0.99
Z=104.37Ω Z=60.4Ω Z=115.77Ω
L=15mm L=15mm L=15mm

Figure 17.20 Transformation steps for the 10th-degree hybrid highpass filter.

Figure 17.21 Initial responses (left) and after transformation and optimization (right) of the
10th-degree hybrid highpass filter.
18 Multiplexers
Multiplexers as considered in this chapter are three or more port
devices that route signals to alternate ports based on frequency. They are
used to either split signals and divert them to a given port, or to combine
signals from alternate ports at a single port. A three-port device is referred
to as a diplexer. Devices with four or more ports are referred to as
multiplexers. Multiplexers differ from splitters and combiners in that with
ideal elements, the signal loss is zero. Whereas with splitters, the responses
are broadband, but signal power is divided among output ports [1].
Passive linear filters are reciprocal so design concepts are the same
whether the multiplexer is used to split or to combine signals. The
individual filters are a combination of lowpass and/or highpass filters, with
any quantity of bandpass filters. Therefore, there are an infinite variety of
multiplexers. Multiplexers are created in Genesys by designing the
individual filters and pasting them into a common schematic. All of the
features of the simulator are available, such as statistical, tuning,
optimization, and electromagnetic analysis.

18.1 Contiguous Multiplexers


There are two classes of multiplexers. The cutoffs of contiguous
multiplexers share a common 3.01-dB frequency. The cutoff frequencies of
noncontiguous multiplexers are separated by a guard band, and the
attenuation at the crossover frequency is greater than 3.01 dB, typically
much so. The design procedure and the performance in the crossover
frequency range are different for these two classes of multiplexers. The
contiguous multiplexer is considered first.

18.1.1 Contiguous Lowpass-Highpass Diplexer


Consider the basic lowpass-highpass diplexer in Fig. 18.1. The individual
fourth-degree lowpass and highpass filters are designed using the Passive
Filter module of Genesys and pasted into a new schematic. Other than
connecting the filters at the common port and numbering the ports, no
other modifications are necessary. It is critical that each filter presents a
series element to the common port. The stopband of each filter is in the
frequency range of the passband of the other filter. A series element causes
a high impedance in the stopband of one filter to be presented to the
passband of the other filter, thus avoiding interaction and destruction of
the passbands. If these two filters are connected at the ports with shunt
277
278 Filter Synthesis Using Genesys S/Filter

elements, L3 shunts the passband of the lowpass, and C1 shunts the


passband of the highpass. In the case of bandpass sections, a series L-C
resonator must connect to the common port.

Figure 18.1 Basic lowpass-highpass diplexer.

The transmission amplitude responses of the basic diplexer are given in


Fig. 18.2. The selected transfer approximation for each filter is Butterworth,
with a 3.01-dB cutoff of 100 MHz for each filter. The isolation of ports two
and three degrade to 6.02 dB at the cutoff frequency.
The individual filters are designed as singly terminated, where the
transfer function is realized with a finite termination resistance at one port
and a zero impedance at the common port. Singly terminated filters do not
present zero ohms to this port. When used individually, singly terminated
filters are mismatched and deliver only current or voltage to the load, but
not power. But when two singly terminated Butterworth filters are parallel
connected and share a 3.01-dB cutoff frequency, the filters are said to be
complementary, and the input impedance at the common port is matched
at all frequencies. S11 is not displayed in Fig. 18.2 because the return loss is
infinite at all frequencies. S/Filter does not support singly terminated
filters, so the Passive Filter module of Genesys must be used to design
contiguous multiplexers. Only singly terminated Butterworth filters are
truly complementary. However, as illustrated in the next example, singly
terminated Chebyshev filters are approximately complementary.
Multiplexers 279

Figure 18.2 Transmission amplitude responses of the basic lowpass-highpass diplexer.

18.1.2 Contiguous LP/BP/HP Multiplexer


This example is a Chebyshev contiguous multiplexer with a lowpass
channel from DC to 3 MHz, a bandpass channel from 3 to 30 MHz, and a
highpass channel above 30 MHz. Given in Fig. 18.3 are the individual,
singly terminated filter schematics, and a schematic that assembles these
filters. Reusing the schematics to assemble the diplexer in Genesys allows
the user to modify the parameters of the individual filters without needing
to repaste a new schematic into the diplexer schematic.
Filters using Chebyshev transfer approximations are approximately
complementary. The best wideband, common-port return loss is realized
using a cutoff attenuation of approximately 3 dB. The reused schematic
build illustrated in Fig. 18.3 supports iteratively adjusting the cutoff of each
filter for the best wideband return loss. In this case, individual filter cutoff
attenuations of 3.5 dB resulted in the best return loss. In this example, the
individual filters are designed with a passband ripple of 0.25 dB, which
would result in a passband return loss of 12.5 dB. However, due to the
approximate complementary nature of this contiguous multiplexer, the
worst-case return loss at the common port is approximately 22.5 dB, as
illustrated in the multiplexer responses given in Fig. 18.4.
280 Filter Synthesis Using Genesys S/Filter

C1 C2
ZO=50Ω ZO=0Ω ZO=50Ω C=69.5pF C=70.64pF

L1 L2
L=4049.7nH L=3984.49nH ZO=0Ω

C1 C2 L1 L2
C=844.69pF C=1958.68pF L=333.2nH L=143.69nH

C2 C4
L2 L4
ZO=50Ω C=625.49pF C=635.72pF
L=449.97nH L=442.72nH

ZO=0Ω

L1 C1 L3 C3
L=2998.77nH C=93.85pF L=1293.23nH C=217.63pF

Port_2
ZO=50Ω

LP_Filter

Port_3
ZO=50Ω

Port_1
BP_Filter ZO=50Ω

Port_4
ZO=50Ω

HP_Filter

Figure 18.3 Lowpass (top left), highpass (top right), and bandpass (middle) components of
the multiplexer with the reused filters assembled into the multiplexer (bottom).
Multiplexers 281

Figure 18.4 Responses of the contiguous lowpass/bandpass/highpass multiplexer.

18.2 Noncontiguous Multiplexers


In this section, the design of noncontiguous diplexers and multiplexers is
illustrated. As with contiguous multiplexers, a series branch must connect
the individual filters to the common port. However, the individual channel
filters of noncontiguous multiplexers are synthesized as doubly terminated.
Therefore, the unique advantages of S/Filter for synthesizing the individual
filters is available for multiplexer design. Noncontiguous multiplexers
forego excellent return loss in the crossover region, but have the advantage
of better isolation between the noncommon ports.

18.2.1 Noncontiguous LP/HP Diplexer with FTZ


This example is a diplexer with a lowpass channel from DC to 30 MHz and
a highpass channel for 50 MHz and above, resulting in a 20-MHz guard
band. The individual filters are designed in S/Filter and are doubly
terminated with a 0.1-dB ripple Chebyshev passband and one FTZ. The
schematics are given in Fig. 18.5.
The responses of the individual filters plotted on one graph grid are
given on the left in Fig. 18.6. Each filter has approximately 16-dB return
loss and a stopband minimum attenuation of 30 dB. The FTZ are 42.2 MHz
for the lowpass and 35.5 MHz for the highpass. Given on the right in Fig.
18.6 are the responses after the individual filters are parallel connected to
form the diplexer. Interaction between the filters degrades the passband
282 Filter Synthesis Using Genesys S/Filter

return loss to approximately 15 dB and has shifted the cutoffs by several


megahertz.

Figure 18.5 Schematic of the LP/HP 30-/50-MHz noncontiguous diplexer with doubly
terminated channel filters.

To recover the responses, the left three elements of each filter are
optimized with goals of re-establishing the initial return losses and cutoffs.
The optimization converges quickly to the responses given in Fig. 18.7.
Unlike contiguous diplexers, the return loss is poor in the crossover region,
but a minimum isolation of 35 dB is maintained.
Multiplexers 283

Figure 18.6 Responses of the individual channel filters plotted on one graph (left), and
responses after parallel connection of the filters to form the noncontiguous diplexer (right).

Figure 18.7 Optimized responses of the LP/HP noncontiguous diplexer.


284 Filter Synthesis Using Genesys S/Filter

18.2.2 Noncontiguous Distributed Combline Diplexer


This example is a noncontiguous, distributed bandpass/bandpass diplexer.
Design begins with individual, capacitor-loaded, combline bandpass filters.
The lower bandpass covering 925 to 975 MHz is given in Fig. 18.8. This
filter is designed in the Genesys Microwave Filter module and then
optimized for the best responses, as shown on the left in the left graph of
Fig. 18.9.
Port_1

1
TL1
W=1.7mm [WLead]
L=8.45mm [Llead]
2
TL2
C1
W=1.7mm
C=1.84pF [Cap1]
S1=0.61mm [S1]
S2=2.6mm [S2] 3
S3=2.6mm [S2]
S4=0.61mm [S1] C2
L=25mm [L1] C=1.77pF [Cap2]

TL3 4
W=1.7mm [WLead]
C3
L=8.45mm [Llead]
C=1.84pF [Cap1]

5
Port_2

Figure 18.8 Combline sixth-degree 925- to 975-MHz bandpass filter.

A upper bandpass covering 1025 to 1075 MHz is similarly designed in


the Microwave Filter module and optimized. The responses are given on
the right in the left figure of Fig. 18.9. These two comblines are then
connected as shown in the schematic in Fig. 18.10. The responses are given
on the right side of Fig. 18.9. Filter interaction has narrowed the lower
channel passband, widened the upper channel passband, and significantly
degraded the return loss of both channels.
The coupled-line spacings and loading capacitors are then optimized to
recover the responses using variables to force symmetry in the resulting
values. The final values are given in the schematic of Fig. 18.10 and the
resulting responses are given on the left in Fig. 18.11. The cutoff frequencies
and return loss are recovered to approximately 20 dB. Nevertheless,
interaction remains and is visible in the transition regions of the channel
filters below approximately 25 dB of attenuation.
Multiplexers 285

Figure 18.9 Individual responses of two combline bandpass filters plotted on the same graph
(left) and responses after connection as a diplexer (right).

Prior to optimization, a susceptance compensation network, in the


form of a shorted stub approximately a quarter-wavelength long at the
crossover frequency, is manually added at the common port of the diplexer.
The length and width of this network is optimized simultaneously with the
other variables. Susceptance networks, in the form of shorted quarter-
wavelength stubs or series L-C networks to ground, are often added to
compensate for interaction effects in diplexers.
Given on the right in Fig. 18.11 is the channel isolation. This is the
amplitude transmission between ports two and three of the diplexer.
286 Filter Synthesis Using Genesys S/Filter

Figure 18.10 Optimized schematic of the combline bandpass-bandpass diplexer.


Multiplexers 287

Figure 18.11 Channel transmission amplitude and return-loss responses of the optimized
combline diplexer (left) and the isolation from port two to port three (right).

Reference
[1] G. Matthaei, L. Young, and E.M.T. Jones, Microwave Filters, Impedance-
Matching Networks, and Coupling Structures, Artech House, Norwood, MA, 1980.
19 Electromagnetic Simulation
In earlier chapters, simulation of the responses utilized closed-form,
analytical models for both lumped and TLEs. The emphasis in this book is
synthesis and transforms. Simulation is used to illustrate the success or
failure of these processes. Other works by the author illustrate the practical
aspects of filter realization [1-3] such as models, element (unloaded) Q,
losses, parasitics, tolerance, tuning, and standard values. For lumped
elements, the Genesys environment supports convenient substitution of
user developed and commercial models.
Distributed models in Genesys use refined and accurate industry-
published models. However, these models cannot consider a multitude of
factors including box modes, radiation, coupling between discontinuities,
and unusual geometries. These factors may significantly impact the
accuracy of the circuit theory simulation. Electromagnetic (EM) simulation
is a powerful tool for dealing with these issues. In this chapter, EM
simulation is used to further refine a few of the filters synthesized in this
book and to review a classic method of filter design newly enabled by EM
simulation.

19.1 Overview
Circuit simulators rely on network theory applied to impedance and
admittance models for individual elements. In the case of lumped elements,
simple reactance models, or models using a small subcircuit of elements,
are used. For TLEs, pure-TEM mode elements such as coax are modeled as
simple trigonometric functions. Models for quasi-TEM mode elements,
such as microstrip, are modeled by more complex functions that are based
on curve-fits to numeric data. In isolation, these models are accurate.
However, their accuracy is reduced by interactions, and models are often
not available for unusual geometries.
Rather than relying on network theory, EM simulators compute the
currents and fields produced in the metal of a circuit. Fundamentally, these
calculations rely on Maxwell’s equations, but to produce practical results,
involved mathematical concepts and tricks are employed. These concepts
vary among different commercial software programs.
Two-dimensional (2-D) simulators analyze only problems that are
infinitely continuous in one direction. Uniform transmission lines and
waveguides are analyzed with great numeric efficiency using this technique.

289
290 Filter Synthesis Using Genesys S/Filter

It is suitable for finding the characteristic impedance of unusual


geometries. Execution speed is excellent, but flexibility is limited.
2 ½-D simulators provide for arbitrary geometry in an x-y plane and
are suitable for analyzing discontinuities such as tees, bends, crosses, and
transmission line coupling. No currents are simulated in the z-direction.
3-D planar simulators support arbitrary x-y currents, and they support
some functionality for z-directed currents. This supports multiple planes
and via hole currents connecting those planes. Via holes are simulated as a
single current and must be small in diameter and short with respect to a
wavelength. Fields are 3-D, so antenna patterns can be computed.
Full 3-D simulators support arbitrary 3-D metal. They are truly flexible,
but require significant computational resources. They are inefficient for
solving 3-D planar problems.

19.1.1 The EMPower Program


Work began on EMPower at the Novosibirsk Electrical Engineering
Institute in 1987. This led to the development of the program TAMIC in
1991 with commercial use primarily in the Soviet Union. The principal
contributor, Yuri Shlepnev, joined Eagleware and led the integration of
EMPower into Genesys with the release of Version 6.5 in 1998.
EMPower is a 3-D planar EM simulator. It uses the method of lines
(MOL) technique. The initial planar 3-D geometry is reduced to a layered 2-
D problem by partial discretization of Maxwell’s equations in a grid
domain. The resulting matrix relating grid currents and voltages is reduced
to an immitance matrix of integral currents and voltages in ports. The
method of simultaneous diagonalization extracts the scattering matrix. The
MOL gives a self-regulating solution with only one variable: grid cell size.
This approach offers good convergence and accuracy. The primary
disadvantage is the requirement for a uniform grid. EMPower mitigates
this disadvantage by a number of automatic techniques including thinning
into groups of cells and detection of symmetry, which greatly reduces the
numeric cost. The comparatives of EMPower are:
1) Full integration into Genesys with automatic cosimulation. One
schematic and layout are used for the linear, EM, harmonic balance,
and time-domain simulators. This schematic is created either by a
synthesis process or by the user. EMPower automatically removes
discrete (lumped) elements, adds internal ports for the EM simulation,
and re-installs the discrete elements for an integrated simulation.
2) Fast and accurate simulation of metal patterns that can be placed on a
grid, or that are small with respect to a wavelength. This requires
discipline by the user to ensure that lines and metal patterns are a
multiple of a user-selected size. The benefit is fast execution. However,
the grid is a disadvantage for some structures, such as filters with wide
lines but narrow gaps.
Electromagnetic Simulation 291

3) Automatic detection of YZ, XZ, two-mirror and rotational symmetry.


This substantially reduces the computational cost of symmetric
structures common with distributed filters.
4) Supports the decomposition of a circuit into blocks. These blocks are
then connected with analytical transmission-line models. This
technique takes advantage of the real-time speed of circuit theory
simulation with the addition of EM simulation for those portions of
the circuit where it is required. Although decomposition requires
additional implementation steps, for large symmetric structures, like
high-degree filters, it can save substantial execution time.
5) Consideration of the effects of the housing. Box modes significantly
impact the response characteristics of both active and passive circuits.
6) Closed or open housing covers and antenna pattern plotting. However,
for circuits with a non-conductive housing, or for more advanced
planar antenna simulation, the program Momentum is a better choice.

19.1.2 The Momentum Program


Momentum is a 3-D planar EM simulator based on the method of moments
(MOM). It was originally developed by Alphabit, a spin-off of the Belgium
company IMEC, and was acquired by Agilent Technologies (then Hewlett-
Packard) in 1990.
MOM is a numerical discretization technique used to solve Maxwell’s
equations for planar metal structures that are embedded in a multilayer
dielectric substrate. First the substrate Green’s functions are calculated,
then the planar signal structures are meshed to form the MOM interaction
matrix that is solved for the unknown surface electric and magnetic
currents. S-parameters are then accurately derived from the solution after
port calibration and de-embedding similar to making a network analyzer
measurement.
The comparatives of Momentum are:
1) Full integration into Genesys with automatic cosimulation as described
with EMPOWER.
2) Advanced automatic and/or user-directed meshing techniques.
Conforming the circuit to a grid is unnecessary. Efficient for solving
circuits with combined wide-strips and narrow gaps common in filter
applications.
3) Adaptive frequency sampling for fast simulation. This is particularly
useful when resonant or critical frequencies are initially unknown.
4) Support for open (infinite-extent PCB), closed, and waveguide (two-
vertical conducting walls) environments. Accurate results for metal-
packaged, nonconducting packaged, and antenna circuits.
5) Full-wave and quasistatic modes. Full-wave is appropriate for
microwave integrated circuits (MICs) and monolithic-microwave
292 Filter Synthesis Using Genesys S/Filter

integrated circuits (MMICs) with lengthy or resonant transmission


lines such as filters. Quasistatic is efficient for large-scale circuits
comprised of electrically short metal.
6) 3-D display of currents in complex layered circuits with via holes and
thick metal.

19.1.3 The EMPro Program


EMPro is a full 3-D metal EM simulator that operates in two modes; a
frequency-domain finite element method (FEM), and a time-domain finite
difference time domain (FDTD) method. FEM uses volumetric meshing
where the geometry is divided into a large number of tetrahedral formed by
four equilateral triangles. The FEM simulator includes both direct and
iterative solvers with linear and quadratic basis functions. The FDTD solver
uses meshes built from rectangular “Y” cells. The FDTD method updates
field values while stepping time, following EM waves propagating through
the structure.
The EMPro FEM solver comparatives are as follows:
1) Simulation of true 3-D metal and dielectric problems such as bond
wires, connectors, RFIC packaging, dielectric objects, and 3-D RF
elements.
2) Full integration into the ADS environment and importation into
Genesys using S-parameter data files.
3) Handling of high-Q circuits.
4) Best solution for smaller 3-D objects.
5) Best solution for multiport problems.
6) Multi-threading for support on multicore CPUs.
The EMPro FDTD solver comparatives are:
1) Simulation of true 3-D metal and dielectric problems such as bond
wires, connectors, RFIC packaging, dielectric objects, and 3-D RF
elements.
2) Full integration into the ADS environment and importation into
Genesys using S-parameter data files.
3) Requires separate simulation for each port.
4) Less memory intensive for larger problems.
5) Highly multithreaded for support of multicore CPUs and large CPU
cards.

19.2 Box Modes


The first example to illustrate EM simulation is a study of box modes. A
conductive housing that encloses a filter is a cavity. At the frequencies of
Electromagnetic Simulation 293

cavity resonance, TLE discontinuities such as open ends, bends, and steps
easily couple energy into and out of the cavity, resulting in destruction of
the stop bands. The frequency of resonance in a rectangular cavity is [4]
2 2 2
v  l  m n
f lmn =   +  +  (19.1)
2 µrε r a  b  c

where l, m, and n are the TElmn-mode subscripts related to the half-sinusoid


number in the standing-wave pattern in the x, y, and z directions of the
cavity, with corresponding dimensions a, b, and c. µ r and ε r are the relative
permeability and permittivity of the cavity-filling material. For example,
the dominant-mode TE101 resonance of a 4- by 6-cm rectangular air-filled
cavity that is 1-cm high is 4.51 GHz. For the dominant mode, the smallest
dimension, in this case the heigth b, does not influence the resonant
frequency.
Fig. 19.1 depicts a 4- by 6-cm rectangular cavity with 2-mm wide input
and output microstrip lines on Rogers TMM6 ½ ounce electrodeposited
copper with a 50-mil thick substrate of nominal relative dielectric constant
6.0. This width line is approximately 50 ohms. The gap is far too wide to
support significant transmission from the input to the output via gap
capacitance.

Figure 19.1 A 4- by 6-cm rectangular housing with microstrip input and output lines
(approximately to scale).

Given in Fig. 19.2 is the amplitude transmission of this system from the
housing port on the left to the housing port on the right, computed using
the EMPower electromagnetic simulator module of the Genesys suite. To
resolve the narrow resonant modes, frequency steps every 20 MHz are
used, thus requiring 400 frequency steps. The simulation is completed in a
few minutes on a standard desktop PC with an Intel Core2 2.5-GHz CPU.
The TE101-mode resonance occurs at approximately 4.3 GHz. The simulated
resonant frequency is slightly lower than predicted by Eq. 13.1 because the
294 Filter Synthesis Using Genesys S/Filter

50-mil thick TMM6 substrate loads about 13% of the height of the cavity.
The presence of the microstrip metal pattern also loads the cavity.
Radiation from the microstrip open ends and/or the standing-wave
pattern on the microstrip TLE readily couples to the cavity resonances. The
sweep through 10 GHz reveals three resonances as well as antiresonances,
with degraded stopband performance between resonances. The cavity
resonances are almost lossless, and the displayed loss of all three
resonances is nearly zero when the number of frequency simulation points
is sufficiently large.
In this case, stopbands better than 40 dB are unavailable at frequencies
above one half the lowest resonant frequency. Resonant frequencies scale
directly with the housing size. It is clear that achieving good stopband
performance at higher frequency requires a small housing. Cavity
resonance is why small filter size is critical for good stopband performance.
Given on the right in Fig. 19.2 is the response of the same layout but on
an open PCB without a conductive housing, as computed by the Momentum
EM simulator module of the Genesys suite. Resonances are avoided, thus
improving the potential stopband performance of filters. Nevertheless,
surface modes and edge reflections severely limit the ultimate stopband
performance.

Figure 19.2 Transmission amplitude response of the microstrip TLEs enclosed in a 4- by 6-cm
housing computed by EMPower (left) and on an open 4- by 6-cm PCB computed by
Momentum (right).

EMPower is more efficient for solving the enclosed box-mode problem


because the metal is easily placed on a grid, the problem is symmetric, and
a large number of frequencies are required to simulate the narrow
resonances of the box modes. Momentum is used to solve the open PCB
Electromagnetic Simulation 295

case study because it more accurately supports open structures and the
lower Q requires fewer frequency points.
This case study also illustrates the author’s insistent preference
throughout this book for smaller filter topologies. Smaller filters naturally
require smaller housings, which increases the frequency of the first
resonant mode, and therefore the stopband performance of the filter.

19.3 EM Simulation of Distributed Circuits


This section illustrates the use of EM simulation to improve the accuracy of
predicted responses. Both the Agilent Technologies EM simulator
Momentum and the simulator EMPower are used for these examples.

19.3.1 EM Simulation of Penetrating Stepped-Z Lowpass

Consider the penetrating, stepped-impedance 2.4-GHz lowpass filter


described in Section 14.3. An Agilent Technologies Momentum simulation
is added to the Genesys workspace tree and the general simulation
properties illustrated in Fig. 19.3 are entered. The values of other
parameters in the Momentum options tabs are left at default values, except
the Use Box option is selected in the Simulation Options tab.
The box size is set at 1.58-cm long by 1.20-cm wide. As with the original
filter, the PCB material is Rogers TMM6 half-ounce copper on a 50-mil
substrate. The height of the enclosure is 0.64 cm, with a conducting cover.
The dominant mode resonance for this enclosure, if unloaded with a
substrate, is 15.6 GHz. The mesh automatically generated by Momentum
for EM simulation is visible as gray lines in the layout in Fig. 19.4.
Responses for this lowpass are given in Fig. 19.5. The solid traces are
computed by the linear circuit-theory simulator in Genesys. The dashed
traces are the responses computed by the Momentum simulation The
required tee models are small and of little consequence. However, the
asymmetric step at the input/output lines, and the large steps associated
with the high-impedance lines to the low-impedance stubs, are severe.
Nevertheless, the excellent agreement of the cutoff frequency, and the
return-loss responses through the cutoff frequency, suggests that the static
model quality of the circuit theory simulator is good. A small error easily
produces return-loss disagreements of 20 and 30 dB.
296 Filter Synthesis Using Genesys S/Filter

Figure 19.3 General simulation properties for the penetrating, stepped-impedance 2.4-GHz
lowpass filter.
Electromagnetic Simulation 297

Figure 19.4 Layout of the original penetrating, stepped-impedance 2.4-GHz lowpass filter
overlaid with the mesh automatically generated by the Momentum simulator (left), and after
narrowing and lengthening the open stubs (right).

Figure 19.5 Responses of the penetrating, stepped-impedance 2.4-GHz lowpass filter (right
graph) by circuit theory simulation (solid) and Momentum EM simulation (dashed), and the
responses after narrowing and lengthening the open stubs (right).

Agreement of the EM and circuit-theory responses is fair up to


approximately 8 GHz. EM simulation predicts a somewhat lower frequency
of the transmission zeros, and interaction of the two resonances to produce
different frequency zeros. This also produces a somewhat steeper
transition. Cascaded stubs typically produce interactions of this form. The
layout after narrowing the open stubs, and then lengthening them to return
the capacitance, is given on the left in Fig. 19.4. The agreement is improved
through 12 GHz. The resonance at 8400 MHz in the EM simulation of the
298 Filter Synthesis Using Genesys S/Filter

original design is possibly caused by interaction of the closer-spaced open


stubs of the original layout.
The Momentum responses are similar when the thick metal option is
selected, suggesting that the 0.02 mm thickness of the half-ounce copper is
of little consequence in this circuit.
In this case, circuit-theory simulation is adequate for design, but EM
simulation improves the accuracy of the predicted responses. Of course, the
final test is measured results, but experience shows that properly set up EM
simulations are extremely accurate.

19.3.2 EM Simulation of a Combline Bandpass


Consider the 950- to 1050-MHz, three-resonator combline bandpass in
Section 16.2.1. The layout for this filter is given in Fig. 19.6. The three larger
footprint pads are for miniature trimmer capacitors to load and tune the
filter. With 10% and less bandwidth, the tolerance of fixed capacitors is
unlikely to be sufficient, and trimming is required. The smaller footprints
near the input and output are for smaller, fixed-chip capacitors. Coupling-
capacitor values are less critical.
Overlaid on the layout is the mesh created by the Momentum
simulator. When all the filter resonators are loaded with capacitors to
achieve resonance, the metal portion of the circuit is nonresonant.
Consequently, the characteristics of the unloaded structure change slowly
with frequency. In this case, few simulation frequencies are required.
Resonance and the responses are realized when the capacitors are
cosimulated with the metal. This saves EM simulation time.
The responses of the combline computed by circuit-theory simulation
are given as the light traces on the left graph of Fig. 19.7. The responses of
the Momentum EM simulation are given as the heavy traces. Inspection of
the layout reveals that the required trimmer and chip-capacitor pads add
metal at the open ends of the combline, thus adding capacitance to ground.
This additional capacitance lowers the passband of the filter. Optimization
of capacitance values is used to recover the passband response as illustrated
on the right in Fig. 19.7. Optimization reduces the coupling capacitors from
1.27 to 1.21 pF, the end-resonator loading capacitors from 2.13 to 1.7 pF,
and the center loading capacitor from 3.23 to 2.67 PF. The cosimulation
environment of Genesys supports optimization of the lumped elements
without requiring the EM simulation of the metal.
Electromagnetic Simulation 299

Figure 19.6 Layout for the 950- to 1050-MHz combline bandpass overlaid with the mesh
automatically created by the Momentum EM simulator.

Figure 19.7 Responses of the 950- to 1050-MHz combline simulated with circuit-theory
models (light traces) and by Momentum (heavy traces). The Momentum simulation with initial
capacitor values is on the left and after optimization of the capacitors is on the right.
300 Filter Synthesis Using Genesys S/Filter

The primary coupling of an ideal combline structure is inductive. The


capacitor pads at the open end of this combline provide capacitive coupling
between the resonators. This capacitive coupling in parallel with the
coupled-line inductive coupling results in finite-transmission zeros above
the passband, as observed in the responses in Fig. 19.7.

19.3.3 EM Simulation of a Direct-Coupled Bandpass


Consider the wideband direct-coupled bandpass in Section 16.3.3. In this
section, the accuracy of the circuit-theory simulation for this 1000- to
2000-MHz bandpass is tested by EM simulation and corrections to the
design are applied, as required. In the original design as shown in Fig.
16.51, the footprint pads of the chip capacitors are placed over the metal of
the TLEs. This is done to minimize metal added to the layout that might
impact the responses. However, EM simulation induced corrections to the
responses modify the width of the series TLEs. This could cause the series
TLE to short to the grounded portion of the chip-capacitor pads. Therefore,
the chip-capacitor pads are placed at the end of narrow TLE connecting
lines that are 0.6 mm long. When the automatic-connect feature in the
Genesys layout module is used, these connecting lines are sufficiently long
to avoid the series TLE contacting the capacitor pads. The modified layout
is given in Fig. 19.8.

Figure 19.8 Layout of the 1- to 2-GHz bandpass after modification of the placement of the
chip-capacitor pads.

The EM simulated responses of the 1- to 2-GHz bandpass after


modifying the locations of the chip-capacitor pads but before tuning is
given on the left in Fig. 19.9. The EMPower and Momentum EM simulators
support optimization of the layout. By using the Connect Parts before
Simulating option, as TLE dimensions are optimized, the elements remain
connected.
Electromagnetic Simulation 301

Figure 19.9 EM-simulated responses of the 1- to 2-GHz bandpass before (left) and after
tuning to correct the responses (right).

In order to correct the return loss, the center frequency, and the
bandwidth of the filter, optimization must adjust several parameters of the
filter. Because each pass is computationally expensive for EM simulation,
and because of the requirement to adjust multiple parameters, optimization
requires significant time. A difficulty with bandpass filters is the
nonindependence of parameters. For example, increased line length might
adjust the bandwidth, but it also reduces the resonant frequency.
Therefore, to correct the responses, tuning rather than optimization is
applied. Iterative actions by an intelligent user, while tedious and time
consuming, is more effective than optimization in this case. If tuning is
simply trial and error, it is ineffective. However, certain principles guide the
tuning process. Principles common to many filter structures are:
1) Lengthening lines shifts the passband down.
2) Widening the spacing between coupled TLEs decreases the bandwidth.
3) Increasing the characteristic impedance of connecting lines in direct-
coupled filters decreases the bandwidth.
4) Adjusting the characteristic impedance of lines connecting to the
terminations changes the return loss properties.
5) Steps of one parameter at a time reveals behavior.
Notice in the responses on the left in Fig. 19.9 that the filter bandwidth is
too narrow and the center of the passband is too low. The author used
principles 1), 3), 4) and 5) in this case to tune the responses. The resulting
responses are given on the right in Fig. 19.9. The final tuned values are
given in Fig. 19.10.
302 Filter Synthesis Using Genesys S/Filter

Figure 19.10 Schematic of the 1- to 2-GHz bandpass with TLE dimensions after tuning to
correct the responses.

19.4 Classic Method of Bandpass Design


The focus of this book is filter design using direct synthesis. As such, this
section is off topic. However, a motivation for writing the book is the fact
that direct synthesis is underutilized. Another underutilized filter design
technique is the classic method of bandpass filter design, so a description
and example are included here [5]. The classic technique is referred to by
Puglia [6] as the general procedure because it enables the design of
bandpass filters of almost any type. The classic method was used
extensively before analytical models for microstrip and other quasi-TEM
mode transmission lines were available and when the primary engineering
computational tool was the slide rule.
The classic technique requires building prototypes of a few select
resonators and plotting certain measured data. From these plots, the filter
dimensions are easily synthesized. Modern EM simulation breathes new
life into this classic technique by eliminating the need to construct
prototypes. Unusual resonators such as compact planar, MEMS, and
multilayer structures are increasingly used to solve challenging design
requirements, but they are poorly modeled by analytical formula. In
addition, the prototype turnaround time for some of these processes is
lengthy. The classic technique is particularly well suited for such structures.

19.4.1 Classic Method Fundamentals


This article emphasizes the application of EM to the procedure. Puglia [6]
gives a comprehensive description of the general procedure. For
completeness, the procedure is briefly reviewed here.
Electromagnetic Simulation 303

Bandpass filters entail only three first principles: 1) resonators must


exist, 2) the resonators must couple to each other, and 3) the structure
must couple to the terminations. The classic method is based directly on
these first principles. The designer selects a form of resonator. Essentially
the only restrictions are resonator realizability, and some method of
coupling must exist. Next, data is acquired that relates the degree of
resonator coupling to a variable parameter. Finally, data is acquired that
relates termination coupling to a variable parameter. From this data, the
filter is synthesized using simple analytical expressions.
Filter tables are usually published in the form of prototype g-values for
a lowpass filter with 1-radian cutoff frequency and a source impedance of 1
ohm. Many authors and Puglia provide a simple formula for finding
Chebyshev prototype values for any passband ripple and order.
Alternatively, S/Filter may be used to design a target lowpass with these
terminations, cutoff, and passband ripple, and the resulting element values
are the prototype g-values. For example, the values for a third-order
0.0432-dB ripple Chebyshev are go=1, g1=0.8515, g2=1.1031, g3=0.8515,
and g4=1.
The normalized source termination is g0, and gN+1 is the normalized
load termination. For odd order Chebyshev, g0 = gN+1. For even order, gN+1
increases for increasing passband ripple. There are N reactive values for an
Nth order prototype.
The classic method utilizes k and q values rather than lowpass
prototype g-values. k values relate to resonator couplings and q values
relate to end resonator loaded Qs. k and q values are easily derived from
lowpass prototype values.
1
k n , n +1 = for n = 1to ( N − 1) (19.2)
g n × g n +1

q1 = g0 × g1 (19.3)

qN = g N × g N +1 for N odd (19.4)

gN
qN = for N even (19.5)
g N +1

For the example 0.0432-dB ripple Chebyshev, q1=q5=0.8515, and


k12=k23=0.9393. These k and q values are normalized by the filter fractional
bandwidth, bw , by
f upper − f lower
bw = (19.6)
f0

where fupper is the passband upper cutoff frequency, flower is the passband
lower cutoff frequency and
304 Filter Synthesis Using Genesys S/Filter

f upper + f lower
f0 = (19.7)
2
Then the actual filter couplings and loaded Qs are
K n , n +1 = bw × kn , n +1 (19.8)

q1
Q1 = (19.9)
bw
qN
QN = (19.10)
bw
Capitalized symbols refer to actual couplings and loadings while
noncapitalized symbols refer to normalized values.

19.4.2 Example: Determining K Values


On the left in Fig. 19.11 are two PCB resonators edge-coupled by their
proximity. This design takes advantage of two signal layers in a multiplayer
PCB. The top metal layer is depicted in the figure as the darker-gray
objects. The wider, lighter-gray objects are metal on layer 2, separated from
the top-metal layer by 15-mil-thick Rogers TMM6 substrate material. The
objects on the two signal layers are connected by via holes at the top of the
objects in the figure. The darker-gray objects that connect to ports on the
wall of the enclosure are 50-ohm lines that loosely couple to the resonators
by their proximity.

Figure 19.11 Resonator pair for finding the resonator coupling (left) and tapped resonator for
finding the external coupling (right).

The resonators are stepped-impedance (stepped-Z) to reduce resonator


length. Inductance predominates at the grounded end of a quarter-
wavelength, open-circuit transmission-line resonator, where the current is
highest. At the open end, capacitance predominates where the electric field
is greatest. By narrowing the line at the grounded end, and increasing the
Electromagnetic Simulation 305

line width at the open end, both the inductance and capacitance are
increased, thus lowering the resonant frequency. This supports a physically
shortened resonator [1]. Using two metal layers allows the open-end
portion of the resonator to be folded under the grounded end, thus halving
the physical length and further reducing the size. The resulting resonator is
very compact.
This structure is used by EM simulation to find the coupling between
resonators. In this case, the EMPower simulator is used because the
problem is easily placed on a large grid, and the natural x-y symmetry
results is fast simulation. The grid is 0.25 mm square.
Given in Figure 19.12 is the resulting transmission amplitude response
with a gap between resonators of four grid cells, or 1.o mm. The resonator-
coupling coefficient, K, is given by
f upper − f lower
K= (19.11)
f0

where fupper and flower are the frequencies of the system transmission peaks,
in this case 1802.8 and 1780 MHz, and f0 is the aritmetic mean of those
frequencies. In this case, K=0.01273. The external coupling during this test
should be light so as not to disturb the resonant frequencies of the pair. The
coupling is sufficiently loose if the attenuation between the peaks exceeds
20 dB.
Table 19.1 shows the results for EMPower simulations of gap spacings
0.25 to 1.0 mm. Ideally, the center frequency would be constant, but in this
case, it increases approximately 2% with increasing gap spacing. Given as
data points on the dashed trace in Fig. 19.13 are the K values. A line-fit
through these points is used to find the necessary gap spacing for any
required K in this range. A feature of the classic method is that K is typically
a monotonic function of the variable parameter and a smooth curve is
easily drawn through a small number of data points.
306 Filter Synthesis Using Genesys S/Filter

Figure 19.12 Transmission amplitude response of the coupled resonators (left) and
transmission amplitude response of the tapped resonator (right).

Table 19.1 EM Simulation Data for Coupling of the Resonators Shown on the Left in Fig.
19.11
Gap (mm) f upper (MHz) f lower (MHz) f 0 (MHz) K
0.25 1844.00 1666.00 1755.00 0.1014
0.50 1824.50 1738.00 1781.25 0.0486
0.75 1811.25 1766.25 1788.75 0.0252
1.00 1802.80 1780.00 1791.40 0.0127

Figure 19.13 EM-simulated resonator coupling, K, versus gap spacing (dashed line) and
singly loaded Qext versus tap location (solid line).
Electromagnetic Simulation 307

19.4.3 Example: Determining Q Values


The next step is determining the termination loading of the filter as
measured by the external Q. Consider the resonator on the right in Fig.
19.11. The resonator is identical to the coupled resonators except the input
and output are coupled to the resonator by tapping the high-impedance
line. Any form of coupling is used that can be defined by a variable. In this
case, the variable is the length of the grounded, high-impedance line from
the center of the via hole to the bottom of the input and output tapping
lines. The 1-mm-wide lines have a characteristic impedance of
approximately 50 ohms. To better define the tap point, higher-impedance
lines are used to tap the resonator.
The amplitude transmission response of this system with a tap value of
2.5 mm is given on the right in Fig. 19.12. This test resonator is doubly
loaded: once by the input and once by the output. In the final filter, the
resonators at the input and output are singly loaded. Therefore, the
external-loaded Q of a singly terminated resonator is given by
2 f0
Qext = (19.12)
f upper − f lower

In this case fupper=1944 MHz and flower=1657.5 MHz, so Qext=12.78.


The classic method involves repeating this data for a few different tap
values. The results with tap values from 1.0 to 3.0 mm are given in Table
19.2. The data points on the solid trace in Fig. 19.13 are the EMPower
simulated Qext for tap values from 1.0 to 3.0 mm. As is the case with K, Qext
is typically a monotonic function of the variable parameter and a smooth
curve is easily drawn through a small number of data points. We now have
a plot of Qext for any tap value in this range.
Table 19.2 EM Simulation Data for the Tapped Resonator Shown on the Right in Fig. 19.11
Tap (mm) f upper (MHz) f lower (MHz) f 0 (MHz) Qext
1.0 1837.50 1783.5 1813.00 67.1
1.5 1854.25 1744.5 1807.25 32.9
2.0 1891.00 1705.0 1814.50 19.5
2.5 1944.00 1657.5 1829.50 12.8
3.0 2009.00 1599.0 1847.00 9.0

The curves in Fig. 19.13 are sufficient for designing 1800-MHz


bandpass filters within the data range of any bandwidth, order, and transfer
approximation for which g-values are available.

19.4.4 Filter Example Using the Classic Method


In this section, a 1750- to 1850-MHz, three-resonator bandpass filter is
designed using the EMPower data from the previous sections. A bandwidth
308 Filter Synthesis Using Genesys S/Filter

of 100 MHz at 1800 MHz is a fractional bandwidth, bw, of 0.556. The


resulting values are K12=K23=0.0573 and Q1=Q4=15.33. From Fig. 19.13 we
find a gap spacing value of 0.44 mm and a tap location of 2.26 mm.
Because the structure is symmetric and easily gridded, the EMPower
EM simulator is used. To satisfy the required dimensions, the y-direction
grid remains at 0.25 mm. To satisfy the gap-spacing requirement, the x-
direction grid is changed to 0.15 mm, and the spacing is set to 0.45. This
requires the vertical narrow line width to be decreased from 0.5 to 0.45
mm. The frequency shift caused by the narrower lines is compensated by
adjusting the length of the wide lines. The resulting initial filter layout is
given in Fig. 19.14 at a 4x scale.

Figure 19.14 Initial layout of the 1800-MHz stepped-impedance bandpass filter designed
using the classic method (4X scale).

The initial responses computed by EMPower are given on the left in


Fig. 19.15. The center frequency is approximately 5.5% low and the return
loss is only 5 dB. Given that the center frequencies are different in the
simulated K and Q values in the previous tables, and that the narrow line
width is reduced by about 10%, errors in the center frequency and return
loss are expected.
Using the techniques described in Section 19.3.3, the dimensions of the
filter are iterated and EMPower simulations are repeated. To increase the
frequency, the length of the wide lines is reduced. This intelligence, and a
lucky guess that the tap point should be reduced, led to rapid progress in
the iterative process, and after a few steps the responses given on the right
in Fig. 19.15 are obtained. To correct the return loss and fine-tune the
frequency, small metal tips are added at the open ends of the wide lines,
and the length of the center resonator is differentiated from the end
resonators. The resulting schematic and layout are given in Figures 19.16
and 19.17, respectively.
Electromagnetic Simulation 309

Figure 19.15 Initial responses of the 1800-MHz bandpass computed by EMPower (left) and
after tuning to correct the responses (right).

Figure 19.16 Schematic of the 1800-MHz bandpass after tuning of dimensions.


310 Filter Synthesis Using Genesys S/Filter

Figure 19.17 Final layout of the 1800-MHz bandpass after tuning (4X scale).

Given in Fig. 19.18 is a wider sweep EMPower simulation of the final


1800-MHz bandpass filter to illustrate the stopband performance. The
compact size of the filter contributes to excellent stopband performance.
The notch just above the passband is characteristic of stepped-impedance
resonators and contributes to excellent selectivity above the passband.
However, this characteristic complicates the design of wideband filters that
use stepped-Z resonators.

Figure 19.18 EM-simulated stopband performance of the 1800-MHz bandpass.

References
[1] R. Rhea, HF Filter Design and Computer Simulation, SciTech Publishing,
Raleigh, NC, 1994.
Electromagnetic Simulation 311

[2] R. Rhea, Filter Techniques: 3 DVD Set, SciTech Publishing, Raleigh, NC, 2003.
[3] R. Rhea, Practical Issues in RF Design: 3 DVD Set, SciTech Publishing, Raleigh,
NC, 2003.
[4] T. Ishii, ed., Handbook of Microwave Technology, Vol. 1, Components and
Devices, Academic Press, San Diego, 1995.
[5] R. Rhea, “EM Enables Classic Filter Technique,” Microwave Journal, February,
2007.
[6] K.V. Puglia, “A General Design Procedure for Bandpass Filters Derived from
Low Pass Prototype Elements,” Microwave Journal, Part I, December 2000, Part
II, January 2001.
Appendix A: Example Summary

This appendix is a listing of the example filters used throughout the


book to illustrate solutions to filter design problems. These tables are used
to quickly find an example based on which filter design problem is being
illustrated, or to find examples using a specific transform. To segment
searching in these tables, they are categorized as lumped (including
resonator based designs), distributed, hybrid using both lumped and
distributed elements, and multiplexers. Not all occurrences of basic
transforms such as split element, swap element, remove transformer, and
simplify circuit are listed. The names of the transforms may be abbreviated.

A.1 Lumped Examples


Table A.1 List of Lumped-Element Examples
Type Description Illustrates Transforms Page
Lowpass All-pole and Finding TZs by inspection None
generalized 1
Lowpass All-pole and Influence of TZs on None
generalized response 4
Lowpass All-pole Dual extractions None 9
Lowpass All-pole Even-order terminations None 10
Lowpass Elliptic and Custom FTZ placement None
generalized 14
Lowpass Generalized Symmetric element None
values 115
Highpass All-pole Even-order terminations None 11
Bandpass All-pole and Finding TZs by inspection None
generalized 2
Bandpass 100-MHz all-pole Effects of bandwidth None 18
Bandpass 100-MHz Alternate extractions None 24
Bandpass 3– to 30-MHz Selecting extractions None
generalized 29
Bandpass 70-MHz generalized Inexact extraction None 34
Bandpass 70-MHz generalized Removing a transformer Norton series,
Simplify circuit 40
Bandpass 70-MHz generalized Equalizing inductor values Norton shunt 44
Bandpass 100-MHz all-pole Resonator-based filters None 47
Bandpass 100-MHz all-pole Exact design with parallel Norton series
resonators 49
Bandpass 100-MHz all-pole Termination coupling Termination coupling 51
Bandpass 100-MHz all-pole Dual topology Find dual 53
Bandpass 100-MHz all-pole Same-type coupling Norton series,
elements Simplify circuit 55

313
314 Filter Synthesis Using Genesys S/Filter

Bandpass 100-MHz all-pole All equal inductors Equate all shunt Ls 56


Bandpass 100-MHz all-pole Improving inductor values Termination coupling 57
Bandpass 100-MHz all-pole Exact series resonators Norton shunt 58
Bandpass 100-MHz all-pole Reduced inductor values Termination coupling 59
Bandpass 100-MHz all-pole Tubular topology Split element,
Tee to pi 59
Bandpass 100-MHz generalized Series resonator with Norton shunt
FTZs 61
Bandpass 100-MHz tunable Multiband optimization None 63
Bandpass 900-MHz all-pole Coaxial resonators Norton series,
Termination coupling,
Parallel LC to grounded
stub,
Advanced TLine 70
Bandpass 975-MHz generalized Complete generalized Norton series,
ceramic filter design with Equate all shunt Ls,
one FTZ Termination coupling,
Pi to tee,
Par LC to grounded
stub 74
Bandpass 975-MHz generalized Previous with two FTZs See previous 78
Bandpass 9.6-MHz crystal Quick approximate design Norton shunt 86
Bandpass 9.6-MHz crystal All equal crystals None 88
Bandpass 9.6-MHz crystal Nulling parallel resonance None 90
Bandpass 9.6-MHz crystal Lower sideband filter Termination coupling,
Find dual,
Norton shunt,
Equate all Ls 91
Bandpass 9.6-MHz crystal Upper sideband filter Termination coupling,
Norton series 97
Bandpass 9.6-MHz crystal Generalized Norton shunt,
Norton series,
Split element,
Swap element, 103
Bandpass 9.6-MHz crystal Wideband generalized See previous 107
Bandpass 9.6-MHz crystal Very wideband See previous
generalized 108
Bandpass 11.6-MHz ceramic Generalized See previous
piezoelectric 111
Bandpass 455-kHz generalized Symmetric element None
values 117
Bandpass 200-MHz all-pole Symmetric responses None 119
Bandpass 70-MHz generalized Symmetric responses Remove transformer 120
Bandpass 70-MHz generalized Symmetric by FTZ Remove transformer
placement 123
Bandpass Matching network Matching by extraction Remove transformer
sequence 132
Bandpass Matching network Bandwidth expansion to None
minimize element count 137
Bandpass Matching network Power amplifier match None 140
Bandpass Matching network Broadband antenna None
match 142
All-pass 200-MHz equalizer Group-delay equalization None 124
Example Summary 315

A.2 Distributed Examples


Table A.2 List of Distributed-Element Examples
Type Description Illustrates Transforms Page
Lowpass 2.4-GHz all-pole Redundant UEs, adding Kuroda wireline
a layout 169
Lowpass 2.4-GHz all-pole Stubs, contributing UEs Split element 175
Lowpass 2.4-GHz all-pole Stepped-Z all UEs None 176
Lowpass 2.4-GHz Lowpass with FTZ Kuroda wireline,
generalized Finite zero to two-step stub 179
Lowpass 2.4-GHz all-pole Advantages of Open stub to shunt C,
approximate methods Split shunt element,
Shunt C to open stub,
Shorted wireline to TLine 182
Lowpass 2.4-GHz all-pole Advantages of Series L to TLine,
approximate methods Shunt C to TLine 183
Lowpass 2.4-GHz Advantages of Series L to TLine,
generalized approximate methods Shunt C to TLine 187
Lowpass 2.4-GHz all-pole Size reduction by Previous
penetration 190
Lowpass 2.4-GHz all-pole Radial stubs, Series L to TLine
optimization 192
Lowpass 2.4-GHz all-pole EM simulation None 297
Bandstop 5-GHz all-pole Wide bandstop using Insert element
lowpass synthesis 199
Bandstop 5-GHz all-pole Effect of microstrip tees None 201
Bandstop 5-GHz all-pole Reducing high stub Ikeno lines
impedance 204
Bandstop 5-GHz generalized Narrow bandstop with Kuroda wireline,
FTZ Half angle,
Ikeno lines 205
Bandpass 5-GHz all-pole Edge-coupled tutorial Split series element,
Kuroda wireline,
Interdigital lines 211
Bandpass 5-GHz all-pole Edge-coupled with Replace with inverter,
inverters LC in shunt to half-wave line,
Kuroda wireline,
Split element,
Interdigital lines 216
Bandpass 5-GHz all-pole Interdigital using Kuroda wireline,
inverters Scale source and load by
inserting inverters,
Interdigital lines 218
Bandpass 1-GHz all-pole Combline with capacitor Equate all shunt Ls,
coupling Absorb transformer in load,
Scale source and load,
End inverter to capacitive L,
Comblines 224
316 Filter Synthesis Using Genesys S/Filter

Bandpass 2.5-GHz all-pole Miniaturization by short Kuroda wireline,


line lengths Combine all transformers,
Kuroda wireline,
Comblines,
Interdigital lines,
Absorb transformer in load,
Scale source and load,
Replace inverter with quarter
wave line 228
Bandpass 2.5-GHz Narrowband with FTZ Absorb transformer in load,
generalized Scale source and load,
Kuroda wireline,
Interdigital lines,
Half angle,
Finite zero to two-step, stub 233
Bandpass 1-GHz all-pole Combline without Equate all shunt Cs,
capacitors Comblines,
Absorb transformer in load,
Replace inverter with quarter
wave line 238
Bandpass 1-GHz all-pole Minnis class-D Kuroda wireline,
miniature Kuroda Minnis CD right,
Comblines,
Interdigital lines 245
Bandpass 1.8-GHz all-pole Folded, multilayer None
resonators, EM
simulation 307
Highpass 1-GHz all-pole Purely distributed Kuroda wireline,
highpass Interdigital lines 268
Highpass 1-GHz all-pole Increased selectivity Kuroda wireline,
with additional UEs Interdigital lines 270

A.3 Hybrid Examples


Table A.3 List of Hybrid-Element Examples
Type Description Illustrates Transforms Page
Lowpass 2.4-GHz all-pole Advantages of hybrid Series L to TLine
elements 194
Bandpass 1-GHz all-pole Penetrating combline with Open stub to shunt C,
capacitive coupling Shunt C to series C,
Shunt C to open stub 248
Bandpass 1-GHz Hybrid combline with FTZ Open stub to shunt C,
generalized Wireline to single C,
Termination coupling,
Absorb transformer in load 249
Bandpass 1.5-GHz all-pole Direct-coupled with Kuroda wireline,
capacitive loading Equate all shorted stubs,
Scale source and load,
Replace inverter with
quarter-wave line
Open stub to shunt C 252
Bandpass 1-GHz all-pole EM cosimulation None 298
Bandpass 1.5-GHz all-pole EM cosimulation None 300
Example Summary 317

Highpass 1-GHz all-pole Distributed synthesis of Wireline to single C


hybrid highpass 259
Highpass 1-GHz all-pole Lumped synthesis of Grounded L to grounded
hybrid highpass stub 261
Highpass 1-GHz all-pole Increasing selectivity with Wireline to single C
UEs 263
Highpass 1-GHz Hybrid highpass with FTZ Grounded L to grounded
generalized stub 266
Highpass 1-GHz all-pole Bandpass used as Wireline to single C,
highpass Shorted wireline to TLine,
Open stub to shunt C 272
Highpass 1-GHz all-pole Increased selectivity See previous 275

A.4 Multiplexer Examples


Table A.4 List of Multiplexer Examples
Type Description Illustrates Page
Contiguous Lowpass/highpass Butterworth complementary filters 277
Contiguous Lowpass/bandpass/highpass Chebyshev approximately
complementary 279
Noncontiguous Lowpass/highpass Guard band, optimization 281
Noncontiguous Bandpass/bandpass Distributed diplexer 284
Constants, Symbols, and Initialisms
Note: Not all variables are listed. Variables defined locally for use only in
nearby equations are not listed here.
bw fractional bandwidth
BW absolute bandwidth
c vacuum speed of light 2.99792458 × 108 meters/s
C value of capacitance
C0 value of static capacitance
Cm value of motional capacitance of a piezoelectric resonator
CPU central processing unit of a digital computer
DC zero hertz
ε0 vacuum permittivity 8.8542 × 10 −12 farads/meter
EM electromagnetic
εr relative permittivity of a material
ESR effective series resistance of a capacitor
FDTD finite difference time domain EM simulation technique
FEM finite element method EM simulation technique
fl lower cutoff frequency
f0 center frequency
fp frequency of the parallel-resonant mode of a resonator
fq frequency of the quarter-wavelength resonance of a TLE
fs frequency of the series-resonant mode of a resonator
FTZ finite-frequency transmission zero
fu upper cutoff frequency
g value of a lowpass prototype element
HF high-frequency spectrum band
IF intermediate frequency in a radio
IL insertion loss
IM intermodulation distortion product

319
320 Filter Synthesis Using Genesys S/Filter

k mn normalized coupling coefficient between resonators m and n


K mn absolute coupling coefficient between resonators m and n
L value of inductance
LA insertion loss due to signal reflection
L-C comprised of inductors and/or capacitors
Lm value of motional inductance of a piezoelectric resonator
len physical length
MOL method of lines EM simulation technique
MOM method of moments EM simulation technique
µ0 vacuum permeability 4π × 10 −7V ⋅ s /( A ⋅ m ) = 1.256637 × 10 −6 H/m
µr relative permeability of a material
n transformer turns ratio (may be referred to as n or N in S/Filter
dialogs)
N Order of a filter, or the quantity of transmission zeros
ω radian frequency, equal to 2πf
Ω Richards transform variable
PC personal computer
PCL parallel-coupled transmission line
π pi; 3.14159
PCB printed-wiring board
Qc unloaded Q defined by conductor loss
Qext external Q
Ql loaded Q
qN normalized loaded Q caused by the load resistance
QN absolute loaded Q caused by the load resistance
q1 normalized loaded Q caused by the source resistance
Q1 absolute loaded Q caused by the source resistance
Qu unloaded (component) Q
R value of resistance
RFIC radio-frequency integrated circuit
ρ n reflection coefficient at port n
RL return loss
Rm value of the motional resistance of a piezoelectric resonator
Constants, Symbols, and Initialisms 321

S mn scattering parameter with index m and n


θ angle, or electrical length of a TLE
TLE transmission line element
TEM transverse-electric mode of propagation
TRF, xfmr abbreviation for transformer
UE unit element (series transmission line)
VSWR voltage standing-wave ratio
X value of reactance
Y value of admittance
Z value of impedance
Z 0 characteristic impedance of a transmission line
Z 0e even-mode impedance of coupled transmission lines
Z 0o odd-mode impedance of coupled transmission lines
About the Author
Randall W. Rhea was born to Noble (Bill) Rhea and Emma Jane Wright
in 1947 in Findlay, Illinois. He was licensed as radio amateur WN9FFO in
1962. He received a BSEE with honors from the University of Illinois, and
married Marilynn Sue Thomas in 1969. He attended Seattle University and
they both graduated from Arizona State University (ASU) in 1973. His
MSEE thesis at ASU was the construction and operation of an earth station
that received Unified S-Band communications from the Apollo 16 and 17
Command Modules in lunar orbit.
He worked at the Boeing Company, Goodyear Aerospace, and Scientific
Atlanta, where he was conferred principal engineer. His engineering
experience includes amplifiers, antennas, CATV equipment, earth stations,
filters, modems, oscillators, radar, receivers, synthesizers, and
management. In 1985, Marilynn and Randall founded Circuit Busters (later
Eagleware-Elanix Corporation). He wrote the initial releases of the
Eagleware circuit simulator and the filter, oscillator, and transmission line
synthesis programs. In 1994, they founded Noble Publishing. He is the
author of numerous technical papers on antennas, amplifiers, components,
filters, matching, modeling, oscillators, simulation, synthesis, and the
history of the microwave industry. He is the lecturer in numerous CD-ROM
tutorials published by SciTech Publishing, and the author of the books
Oscillator Design and Computer Simulation, first and second editions; HF
Filter Design and Computer Simulation; Discrete Oscillator Design:
Linear, Nonlinear, Transient and Noise Domains; Cable Television Signal
Distribution; and Chapter 6 of the Handbook of Microwave Filter
Technology. Eagleware-Elanix Corporation was sold to Agilent
Technologies in 2005 and Noble Publishing was sold to SciTech Publishing
in 2006. He has taught full-day seminars on filter and oscillator design to
over a thousand engineers at trade shows, the Georgia Institute of
Technology, and companies worldwide.
Randall and Marilynn live at Susina, an antebellum plantation home
near Thomasville, Georgia, where they entertain visits from two adult
children, six grandchildren, family, and friends. In the summer of 2003,
Randall toured 48 states by motorcycle. His current amateur radio license
is N4HI and he has DXCC and first-place CW operating awards. Randall
also enjoys technical writing, astronomy, and wine making.
Randall may be reached at [email protected] for questions,
comments, and the reporting of errata. From time to time, an errata sheet
may be published and supplemental material may be available.

323
Index
Add layout, 173 Distributed, 147
Allow inexact, 34 combline transform, 164
All-pass, 125 compare lumped and distributed, 147
All-pole, 1 half-angle transform, 162
Antimetry, 17 Ikeno transforms, 155
Bandpass, 2, 17 interdigital transform, 161
Bandwidth introduction, 147
absolute, 17 Kuroda identities, 152
fractional, 17 Kuroda-Minnis transform, 157
wide or narrow, 19 reentrance, 152
Blinchikoff, 119 Richards transform, 151
Box modes, 166 TLEs, 151
Canonic, 4 Distributed bandpass, 211
Cauer-Chebyshev, 13 combline capacitive coupled, 226
Cavity resonance, 299 combline hybrid, 249
Ceramic piezoelectric, 111 direct-coupled hybrid, 258
Chebyshev, 10 edge-coupled tutorial, 214
Classic design method, 307 edge-coupled with inverters, 218
example design, 317 generalized combline, 252
finding K by EM simulation, 313 generalized narrow, 238
finding Q by EM simulation, 316 hybrid, 253
fundamentals, 312 interdigital tutorial, 221
k & q values, 312 miniature with UEs, 228
Coaxial ceramic loaded Minnis class D, 250
generalized, 74, 79 unique, 229
Coaxial resonator Distributed bandstop, 199
from parallel L-C, 71 generalized narrow, 206
unloaded Q, 73 moderate bandwidth, 203
Combline narrow bandwidth, 205
capacitive coupled, 224 wide bandwidth, 199
diplexer, 284 Distributed highpass, 259
EM example, 307 bandpass synthesis, 272
generalized hybrid, 249 distributed synthesis, 259
hybrid penetrating, 253 generalized hybrid, 269
penetrating, 243 high degree, 277
transform, 161 hybrid, 259
Complementary multiplexers, 286 hybrid 10th degree, 283
Complex conjugate match, 130 hybrid 8th degree, 279
Compound transforms, 46 hybrid with UEs, 263
Conventional bandpass, 17 lumped synthesis, 260
Convert Using Advanced TLine, 73, 168 purely distributed, 268
Coupled lines, 159 Distributed lowpass, 169
Coupling coefficient, 152 approximate, 180
Customize exact, 169
permutation table, 30 generalized, 179
Degree, 4 hybrid, 194
Diplexer. See Multiplexer penetrating, 190
Dipole transform, 94 radial stub, 192

325
326 Filter Synthesis Using Genesys S/Filter

redundent UEs, 169 Match


stepped-Z, 176, 295 antenna example, 142
stepped-Z approximate, 187 complex conjugate, 130
stubs and contributing UEs, 175 complex terminations, 139
Dual topology, 9, 24, 53 exploiting resonators, 138
Edge-coupled distributed, 147, 211–218 exploiting sequences, 132
Electromagnetic simulation, 289 Fano's limit, 139
box modes, 292–295 matchable resistance ratios, 135
combline example, 300 power amp example, 140
direct coupled example, 300 two element networks, 96
EMPower, 290 Matching, 129
EMPro, 292 Minnis. See Kuroda-Minnis
Momentum, 294 Minnis class-D distributed, 245
overview, 289 Mutliplexers
stepped Z example, 295 combline noncontiguous, 281, 284
surface modes, 294 complementary, 279
Elliptic, 1, 5 contiguous, 277
Equal inductors generalized noncontiguous, 281
by Norton, 42 LP/BP/HP contiguous, 279
compound transform, 46 LP/HP contiguous, 281
Equalizers, 125 Norton transform, 39
Extraction remove transformer, 40
inexact, 34 series, 39
partial, 33 shunt, 43
rules, 33, 34 Order, 4
sequences, 23 Parallel coupled lines, 233
tab, 23 Parametric filters, 18
Fano, 139 Partial extraction, 33
Fractional bandwidth, 17 Penetration
FTZ. See TZ combline, 238
lowpass example, 13 lowpass, 190
Generalized filters, 1 Permutations
Genesys, xiii, 7 goals, 27
Group delay, 18 maximum displayed, 24
Group delay equalization, 124 quantity, 23
Half-angle transform, 159 specify, 25
Highpass, 1, 11 Piezoelectric, 83. See Quartz crystal
History tab, 45 Q
Ikeno transforms, 155 loaded, 70
Impedance matching. See Match unloaded (component), 18
Inexact extraction, 34 Q of load, 140
Insertion loss Quarter-wave frequency, 8
bandpass, 69 Quartz crystal, 83
lowpass, 69 approximate filter design, 87
vs. bandwidth, 19 calculating parameters, 86
vs. ripple, 8 dipole, 94
Interdigital generalized filters, 104
transform, 161 lower sideband filter, 92
Interdigital distributed, 218 model, 83
Inverters, 164 nulling static capacitance, 90
scaling impedance, 218 response, 84
k & q values, 304, 307 upper sideband filter, 98
Kuroda indentities, 152 wide bandwidth filter, 108, 109
Kuroda-Minnis transform, 157 Radial parameters, 8
Layout, 173 Radial stub, 192
Lowpass, 1 Reflection coefficient, 7
Lumped/Distributed equivalents, 182 Resonator-based bandpass, 47
Index 327

classic forms, 47 Transformer


coaxial, 70 remove, 40
coaxial ceramic, 74, 79 turns ratio, 30
exact, 49 Transforms
generalized, 61, 79 combline, 161
like-coupling elements, 55 compound, 46
quartz crystal, 83 dipole, 94
series-type, 58 equate all shunt Ls, 56
TEM-mode, 69 find dual, 53
tubular, 59 half-angle, 159
tunable, 63 Ikeno, 155
Return loss, 7 interdigital, 161
Richards transform, 150 Kuroda, 152
Ripple, 7 Kuroda-Minnis, 157
permutations lumped to distributed, 162
customize table, 30 Parallel L-C to grounded stub, 72
Scattering parameters, 8 pi to tee, 77
Sequences, extraction, 23 Richards, 152
Series element first, 9 series Norton, 39
Simplify Circuit, 42 shunt Norton, 43
Singly terminated, 278 tab, 39
Smith chart, 130 termination coupling, 51, 54
Specification tab, 7 Transmission zero. See TZ
Surface modes, 294 Tubular bandpass, 59
Symmetry, 115 Tunable bandpass, 63
all-pole, 120 TZ
by FTZ placement, 123 by inspection, 1
classification, 18 definition, 1
generalized, 120 finite frequency, 1
group delay, 20 response, 4
group delay equalization, 124 rules, 18, 33
physical (element), 115 Unit Element (UE), 149
response, 119 VSWR, 8
zigzag, 122 Wireline, 149
TEM-mode resonators, 69 Zigzag, 120
Transform tab, 39

You might also like