EMI Filter Design

EMI Filter
Design
Second Edition
Revised and Expanded
Richard Lee Ozenbaugh

Consultant
Hesperia, California
Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.
ISBN: 0-8247-8924-5
This book is printed on acid-free paper.
Headquarters
Marcel Dekker, Inc.
270 Madison Avenue, New York, NY 10016
tel: 212-696-9000; fax: 212-685-4540
Eastern Hemisphere Distribution
Marcel Dekker AG
Hutgasse 4, Postfach 812, CH-4001 Basel, Switzerland
tel: 41-61-261-8482; fax: 41-61-261-8896
World Wide Web
http://www.dekker.com
The publisher offers discounts on this book when ordered in bulk quantities. For more
information, write to Special Sales/Professional Marketing at the headquarters address above.
Neither this book nor any part may be reproduced or transmitted in any form or by any
means, electronic or mechanical, including photocopying, microfilming, and recording, or
by any information storage and retrieval system, without permission in writing from the
publisher.
Current printing (last digit):
10 9 8 7 6 5 4 3 2 1
PRINTED IN THE UNITED STATES OF AMERICA
This is dedicated to my wife, Pansy, for her moral support. Also, I wish to thank
her for the computer time and for the seminar time she lost (and for not
complaining when she had the right to), especially during this rewrite.
Preface
Preface
Over the last twelve years, I have made friends with many EMI gurus across
the United States, but primarily in the Southwest. Many have attended my EMI
Design seminars and I have met others as an Applications engineer in the field.
My desire was to increase my knowledge of EMI filter design by learning new
and better techniques. I concluded, after some years, that very few of them had a
concrete method for the design of these low pass filters (it is black magic, isnt
it?). I got to know most of these engineers quite well. They would give me a very
strange look when I would ask them this leading question: What is your design
technique? I finally concluded that in most cases it depended on what was handy
or readily available. They would have a certain capacitor or inductor on their
bench or in stock (or readily available from a supplier). At best, they would
calculate the needed inductor(s), or capacitor(s), to arrive at the desired loss.
Some would give me profound statements such as, We design for maximum
loss! or We use [a certain program] to get the values needed. To accomplish
this latter method, the guru would reach the desired loss by backward-engineering
the networkmodifying the inductor or capacitor values or changing the number
of components to get the apparent desired loss. Some designs had so much
capacitance across the line that the EMI filter inrush current, added to the power
supply filter capacitors charging currents, would trip the circuit breaker. Others
would spend days measuring the line and load impedance using network analyzers so the EMI filter could be matched to these two end sections. Unfortunately, the end sections would change from application to application so that the
filter center piece would not function properly, because of the mismatch existing
in the band pass.
vi
Preface
Another technique I have found is for the engineer to grab any core and start
winding wire around it for its inductor. Usually they could not even tell me the type
of coreMPP, HF, tape wound ferrite? Just an available core, something off the
shelf. This does not give a reference point from which to start. Whether the filter
works or fails, the filter inductor core must be known. The fallacious statement often
made is that a given core was used by one of the filter manufacturers and therefore
must be the correct type and about the correct number of turns.
A doctor from Sierra Vista, Arizona, wrote a very fine article in the
late 1980s on commercial filters that included a good method for calculating
the component values. The title of the article was something like Gus, the
Electrician. His filter component values came from M-derived filter design
techniques
L =
Rd
F
C =
1
FRd
where L is the inductance, C is the capacitance, Rd is the design impedance, and

F is the cutoff frequency. He felt that most of the original filter designers allowed
their cutoff frequencies to fall much too close to power line frequency. This
happens in the 461 specifications requiring high dB losses in the 10 to 14 kHz
region. Otherwise, few EMI filter cutoff frequencies drop as low as indicated in
the article. This is especially true for commercial filters. Gus would often switch
the ground and return wire because he thought that ground was ground. This was
a very good article and a very good design method for the component values.
He did not offer a method to calculate the cutoff frequency.
Other engineers have correctly used the energy spectrum of the switcher
frequency to determine the amount of loss required. There is trouble transforming
this information when the engineer selects the filter component values. Many
designs use very small capacitors between line and ground with very high values
of resistors in series with them. Some have used capacitors and resistors in series
such as 2000 pF in series with 470 ohms. The balanced circuit requires two of
these networks wired from both inputs to ground. One network wired between
the high, or hot, to ground and the other network wired between the neutral and
ground. So at the higher frequencies (in the case stated, above 2 MHz), the line
to line would be 940 ohms. Why? The line impedance would be somewhere
between 50 and 130 ohms, depending on the line. The energy content of the
switcher is low at these frequencies but, again, if this energy was high, why the
high series resistance?* Is there a time when this is a good technique? Yes, but
*This was written when switchers worked at 80 kHz maximum but now some are in the MHz region.
Preface
vii
the resistors are quite small in value. Usually, these values fall in the order of 10
ohms and the capacitors are in the area of 1 F. This is done to reduce the full
circuit Q to lower the tendency of the filter to oscillate at the filters normal
resonant rise frequency.
It has been very informative to be in the field, over the years, seeing these
filters in operation, especially the higher wattage power line filters. Many of these
filters were running very hot to the touch at much lower power ratings than the
design specification. Others would have a heavy peak-to-peak voltage at some
higher frequency, usually at some odd multiple of the line frequency, superimposed on the line voltage. Many of these troubles are caused by the filter cutoff
frequencies too close to the line frequency, as mentioned earlier, and excessively
high filter circuit Q allowing the filter to oscillate at the resonant rise frequency.
Both of these conditions allow the filter to heat, and this in turn heats the
capacitors. This will then blow the capacitors, destroying the filter. Some filter
companies have rebuilt these filters using the same designed-in capacitor, only to
have them fail again in a short period of time.
The impetus to develop simple design parameters and techniques came
from the situations described above. Hopefully, the information supplied here will
make life easier for those responsible for designing EMI filters. The general goal
of this book is to provide proper design techniques that will help to nip all of the
above practices in the bud. This must be done quickly because the EMI world is
bad enough now and the worst is yet to come, because the specifications will
become more involved and harder to meet as time moves on. The FCC will move
to greater losses at lower frequencies, to follow VDE, and meanwhile VDE will
move to tighter requirements. The same will hold for the MIL F 15733 moving
to 28861, and 461 will move to rev D and then E. The 220A specification will
also be updated. Other test methods will follow suit and be made harder to
accomplish. The medical will move out of the industrial specifications to tighter
requirements. The medical leakage current specification for equipment directly
touching the patient will decrease below the present 100 Asome are now
asking for 50 A.
The primary goal of this book is to provide a quick, but effective, design
method for the filter design engineer, equipping the filter engineer to design
the filter in minimal time and have the filter function with minimal adjusting
of the prototype. Even if the filter cant be adjusted enough to meet the specification and requires major changes, the time lost will be minimalnot weeks
or months.
The first chapters are basic but the EMI person should still scan them. This
book is based on two main principles. The first is called the engineer syndrome.
At what point is a project or product engineered? In the EMI arena, its a different
core, different capacitor type or shape, different style filter (from Ls to Ts), or a
viii
Preface
different mechanical shape of the filter. This book advocates getting the filter
designed, built, and tested as soon as possible.
The second is the KISS principle. The matrices of the various filter types
are discussed in Chapters 13 and 14. Much of the material covered here is based
on close approximations along with heuristic and empirical knowledge. The
abbreviated method is as follows:
1. Decide the best filter type from Chapter 12.3.
2. Find the filter component valuesboth differential and common
modeeither from a) the matrix equations of Chapter 13 or 14 or b)
the section in 18.8, F0The Easy Way. Method b will get the filter
designer onto the proper ballpark property while a) will get the designer
onto the ballpark playing field. Small adjustments to either of these will
get the designer to home plate.
3. Design the components for the filter based on Chapter 6.
4. Design and build the case or container and have it silver plated using
Chapter 17 to helpthis assumes a military specification.
5. Install the components and test the filter in the open container.
6. Adjust the filter for the desired loss if needed by:
a. Adding lossy components (4.4).
b. Adding small line-to-line capacitors in parallel to the existing
capacitors (keep the lead length as short as possible).
c. Add ferrite beads if the current is low enough (typically 5 amp
limit).
d. Add several turns on the inductors (watch for saturation). (If this
filter is a T, whatever turns are added to the central inductors of the
multiple T, increase the turns by half on the two outer inductors
because they are half the value of the inner inductors.)
e. Add an RC shunt (a resistor and capacitor in series across the line)
or a Cauer filter to tune one of the series inductor. See section 5.7.
7. Make sure that the product can be manufactured and is repeatable for
production. Adjusting the filter as described above will move the
designer from the ballpark property to home plate with 2a or 2b above.
Since the first edition of this book was published, I have determined
that basic information and definitions of EMI are mixed in the field. Some
have developed steadfast rules over time that other EMI people violate all
the time.
One example is the common mode inductor. Some will insist that these two
windings must be separated into sections so the two windings are apart on the
core with gaps at each end. Others will wind these bifilar with good results.
Preface
ix
RFI uses both techniquessometimes one method is better than the other. The
first method creates high leakage inductance while the other method does not.
This leakage inductance may ring with a nearby circuit capacitor, causing
problems. The second method adds capacitance between the two windings that
may also cause other problems. So, both have their place and neither is absolutely
correctuse the one best for the situation.
Richard Lee (Oz) Ozenbaugh
Acknowledgments
Acknowledgments
I wish to thank: Colonel W. T. McLyman, NASA Jet Propulsion Laboratory, now

retired, for his KGMAG computer programs, his technical help, and the transparencies; Armando Valdez, Aerojet General, Azusa, California, for all his recommendations and consulting; Bob Rudich, Airesearch, Torrance, California, for
another design approach; Robert Hassett, now retired, RFI Corp., Long Island,
New York, along with the rest of the RFI engineering group, for all their support,
patience, graphs, suggestions, papers, and critiques; Mitchell Popick, for all his
help, suggestions, and argumentation, which often forced me to think in different
terms; the people at Powerlab, Pomona, California, for their backing and help,
primarily with the power supply equations; the people of the James Gerry Co.,
especially Jim Gerry, who helped set up the seminars that started this whole thing;
Harvey Gramm, URAD, for all the special computer time, ideas, transparencies,
and the printing; my son, Richard Lee Ozenbaugh II, for the help with the EMI
Design computer program and the printing along with the computer help after the
hard drive crash. Finally, George Apergis, the present VP of engineering at RFI,
for capacitor equations for split windings and better C core design equations
and other worthwhile suggestions.
Contents
Contents
Preface
1. Why Call EMI Filters Black Magic?

1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
What Is EMI?
A Standard Filter Company Contrasted with
an EMI Company
Power Density Spectrum or Envelope
Power Transfer
Specifications: Real or Imagined
The Inductive Input for the 220A Test Method
The 400-Hz Filter Compared with the 50- or 60-Hz Filter
Three-Phase Filters
2. Common and Differential ModeDefinition, Cause,

and Elimination
2.1
2.2
2.3
2.4
Common and Differential Mode Definitions

What Creates Common Mode Noise on the Line Side?
What Creates Common Mode Noise from the Equipment Side?
What Eliminates Common Mode Noise from the Line and
Equipment?
xiv
2.5
2.6
Contents
What Creates Differential Mode Noise?

Three-Phase Virtual Ground
3. EMI Filter Source Impedance of Various Power Lines

3.1
3.2
3.3
3.4
3.5
3.6
Skin Effect
Applying Transmission Line Concepts and Impedances
Applying Transmission Line Impedances to Differential and
Common Mode
Differences Among Power Line Measurements
Simple Methods of Measuring AC and DC Power Lines
Other Source Impedances
4. The Various AC Load Impedances

4.1
4.2
4.3
4.4
4.5
4.6
Resistive Load
Off-Line Regulator with Capacitive Load
Off-Line Regulator with Inductor Before the Capacitor
The Power Factor Correction Circuit
Transformer Load
The UPS Load
5. DC CircuitLoad and Source

5.1
5.2
5.3
5.4
5.5
Various Source Impedances

Switcher Load
DC Circuit EMI Solutions or Recommendations
Lossy Components
Radiation Emissions
6. Typical EMI FiltersPros and Cons

6.1
6.2
6.3
6.4
6.5
6.6
6.7
6.8
6.9
The Filter
The T Filter
The L Filter
The Typical Commercial Filter
The Dissipative Filter
The Cauer Filter
The RC Shunt
The Conventional Filters
MatrixTest Specification and the Filter to Use
Contents
xv
7. Filter ComponentsThe Capacitor

7.1
7.2
7.3
7.4
7.5
Capacitor Specifications
Capacitor Construction and Self-Resonance Frequency
Veeing the Capacitor
Margin, Creepage, and CoronaSplit Foil for High Voltage
Capacitor Design
8. Filter ComponentsThe Inductor

8.1
8.2
8.3
8.4
Inductor Styles and Specifications

The Powder Cores
Inductor Design
Converting from Balanced to Unbalanced or the Reverse
9. Common Mode Components

9.1
9.2
9.3
9.4
9.5
9.6
9.7
Capacitor to Ground
Virtual Ground
Z for Zorro
Converting Common Mode to a Differential Mode Filter
Equations for the Common Mode Via the Differential Mode
Common Mode Inductor Used for Differential Mode
Other Wave Shapes Besides Sine Waves Work
10. The Transformers Addition to the Filter

10.1
10.2
10.3
10.4
10.5
10.6
10.7
10.8
Transformer Advantages
Isolation
Leakage Current
Common Mode
Voltage TranslationStep Up or Down
The Transformer as Part of an EMI Package
Skin Effect
Review
11. Electromagnetic Pulse and Voltage Transients

11.1
11.2
11.3
The Three Theories

The Location of the Arrester
How to Calculate the Arrester
xvi
11.4
Contents
The Gas Tube
12. What Will Compromise the Filter?

12.1
12.2
12.3
12.4
12.5
12.6
12.7
12.8
12.9
12.10
SpecificationsTesting
Power Supplies Either as Source or Load
Transformers: 9- and 15-Phase Autotransformers
Neutral Wire Not Part of the Common Mode Filter
Two or More Filters in CascadeThe Unknown Capacitor
Poor Filter Grounding
The Floating Filter
Unknown Capacitor in the Following Equipment
Input and Output Too Close Together
Gaskets
13. Waves as Noise Sources

13.1
13.2
13.3
13.4
13.5
13.6
13.7
The Spike
The Pulse
The Trapezoid
The Quasi-Square
Why Differentiate?
The Power SpectrumdB A/MHz
MIL STD 461 Curve
14. Study of the Off-line Regulator

14.1
14.2
14.3
With or Without Critical Value of Inductance

Size and Weight Difference of the Filter
The Added Power Line Harmonic Content Caused
by the Off-Line Regulator
Keith Williams Method
15. Initial Filter Design Requirements

15.1
15.2
15.3
15.4
15.5
15.6
Differential Mode Design Goals

Input Impedance of the Differential Mode Filter
Output Impedance of the Differential Mode Filter
Input and Output Impedance for a DC Filter
Common Mode Design Goals
Estimate of Common Mode Load Impedance
Contents
15.7
xvii
Methods of Reducing the Size of the Inductor Due to

Inductor Current
16. MatricesReview of A Matrices

16.1
16.2
Chain Matrix A: Transfer Functions

Review of A Matrices
17. The Filter Design Technique

17.1
17.2
17.3
17.4
17.5
17.6
17.7
The Unit Matrix

The Rs Matrix
The LINESIM Matrix
The LISN Matrix
The DIN and DOUT Matrices
The RCSHU Matrix
The Series Inductor, LSER, and the Shunt Capacitor,
CSHU
17.8 The L Matrix
17.9 The Matrix
17.10 The T Matrix
17.11 The Cauer Matrix or Elliptic Filter
18. Matrix Applications
18.1
18.2
18.3
18.4
18.5
18.6
18.7
18.8
18.9
Single-Phase AC Filter
Three-Phase Filter
Telephone and Data Filters
Impedance-Matched FiltersWhat Is the Impedance
Limit?
Pulse RequirementsHow to Pass the Pulse
The DC-to-DC Filter
Low-Current Filters
F0the Easy Way
Remote High-Voltage Supply Fed from a Local
DC Power Supply
19. Applications Using Round or Square Conducting Rods

19.1
19.2
Very High Current Filters

High-Current Second Method
xviii
19.3
19.4
19.5
19.6
Contents
High-Current Method Three

Review of High-Current Filters
Three in Parallel
Conclusion
20. Packaging Information

20.1
20.2
20.3
20.4
The Layout
Estimated Volume
Volume-to-Weight Ratio
Potting Compounds
21. Design Examples

21.1
21.2
21.3
Southeast Asia Filter for the Navy

The Faulty 400 Hz Source
The Round Rod Filter in Chapter 19
22. Questionable Designs

22.1
22.2
22.3
22.4
22.5
28 Volts at 35 Amps
120 Volts, 60 Hz, with Transzorbs
The 28 V DC Filter
120 V AC 400 Hz
Review
23. Review of Filter Design

23.1
23.2
23.3
23.4
Filter Design Review

Filters in Tandem
Q
Testing the Filters
1
Why Call EMI Filters Black Magic?
Most engineers, both designers of electromagnetic interference (EMI) devices and

others, call EMI black magic. There are four main reasons for this. First, there
has not been a well-defined design method. Second, the input and output
impedances are not constant over the band of interest. The impedances are usually
just good guesses, at best, because the measurements are costly to make and are
rarely repeatable. Third, the filter insertion loss test method specifications often
confuse or influence the design method. In the 220A specification, requiring
50-ohm source and load impedances, does the filter engineer design the filter for
the 50 ohms or for real-world impedances? If the design passes the Military
Standard (MIL STD) 220A, 50-ohm, test, what will happen in the real world? Is it
the designers responsibility to meet the specification or to stop the real-world
emissions? Fourth, the design methods outlined in various publications are
generally very complex. Most require measurements that are difficult to make and
time consuming. Often expensive additional equipment is required to obtain the
needed parameters.
1.1. WHAT IS EMI?

EMI is electromagnetic interference and is called conducted emissions or radiated
emissions. This book covers mainly conducted emissions (CEs), which means any
unwanted signal or noise on the wiring or copper. The reason for the reference to
power cabling is that EMI filters are part of the power wiring and are designed
to remove these unwanted properties from the copper wiring. Why is this from
the wiring? What does this have to do with magnetic fields? The reason is that
Chapter 1
any current flow creates an associated magnetic field. You cannot have one
without the other. Therefore, this high-frequency unwanted signal creates a
magnetic field that can interfere with surrounding equipment. It is the filters
function to remove this current so that its associated magnetic field will not
interfere. This noise can originate either from the line or from the associated
equipment that the filter is built into. From the equipment side, the noise could
be coming from computer clock frequencies, parasitic oscillations in the switcher
power supply inductors or transformers, power supply diode noise, harmonics of
the line frequencies due to the high peak current charging the power supply
storage capacitor, and many other sources. From the line, the noise could be due
to flattening of the sine wave voltage caused by the high peak currents slightly
ahead of 90 and 270 degrees due to the total of the power supplies fed from the
line without power factor correction circuitry. This generates odd harmonics that
feed the EMI filter. Other sources of noise from the line are other equipment
without any filtering and heavy surges of equipment being turned on and off.
Lightning and EMPs (electromagnetic pulses, possibly from nuclear explosions)
create other line problems for the filter.
To review, EMI is any unwanted signal from either the power line or the
equipment and must be removed to prevent a magnetic field from interfering with
closely associated equipment or to stop a malfunction of the equipment containing the filter. For example, if you or a loved one was in the hospital, you would
not want the heart monitor to dip every time the x-ray machine was used just
because the same copper connected them. Here, the heart monitor filter would
remove the pulse from the x-ray machine. Or, better yet, the x-ray machine filter
would stop the noise from leaving.
1.2. A STANDARD FILTER COMPANY CONTRASTED WITH
AN EMI COMPANY
Most of the energy in the stop band (the frequency area to be attenuated) of the
filter is reflected to its source. This fact is often overlooked in both standard filter
and EMI technology. The remaining energy is expended in the inductors through
the DC resistance of the coil, the core losses (eddy currents and hysterisis), and
the equivalent series resistance of the capacitors. All engineers have learned this
in the past but often forget it somewhere along the way. Whereas this is damaging
to the wave filter people, it is an aid to the EMI group. The dissipative filter
(discussed in Sec. 6.5) dissipates the energy in the stop band. Standard filter
designers have several excellent filter design programs in their computers, such
as Butterworth, elliptic, Chebyshev, and M derived. They know the input and
output impedances of the source and load (usually the same), the allowed
band-pass ripple, the 3-dB point, and the stop frequency (the first frequency with
the required amount of loss). The wave filter designer is provided with passband
ripple dB and other information that dictates a particular filter topology. When
the computer design is complete, the designer will be very close to the results
required. I am not saying that this is easy or that the standard filters never have
to be reworked. The filter may have to be altered by adding stages or through the
use of a different technique, or topology, to achieve the desired results.
Most EMI filter manufacturers design only the low-pass filters (all pole
networks) needed for the required EMI attenuation. Rarely do they build bandpass or other conventional filters. The technology used in conventional, or
standard, filters is truly different from that used in the EMI filter. The EMI filter
design is very loose compared with that used by the conventional filter manufacturer, especially if the EMI filter designer uses the techniques mentioned in the
preface of this book. These techniques would upset most of the normal filter, or
wave filter, designers. The EMI filter component values are very flexible, so the
engineer can use standard values. These filters are adjusted only to meet the
required insertion loss specification assuming the rest of the specification is met.
The languages spoken by the two groups are also different. True filter
houses often speak of poles, zeros, group delay, predistortion, attenuation, and
terms such as the order of the filter. The EMI filter designer thinks in terms of
attenuation, insertion loss, filter voltage drop, filter voltage rise, and the number
of filter sections required to meet the insertion loss. Although the power source
may have harmonics, the actual power supplied to the device through the filter is
restricted to the fundamental frequency. Such a harmonic content is especially
true for power supplied locally by shipboard generators and remote stations where
the generator is near or well past peak power. So flat frequency response, low
phase distortion, or low peak-to-peak ripple across the filter band pass is not an
issue here. These power line harmonics furnish no power to the load, so the EMI
filter designer is not concerned with them. As a result, terms such as group delay,
ripple, and phase distortion are never heard.
To summarize, the requirements of the conventional, or wave, filter house
are entirely different from EMI requirements. The technologies are completely
different. The conventional, or wave, filter house component value is critical. This
requires a better grade of components and often requires tuning. EMI houses often
tune, such as with a Cauer filter, but the reasons are different.
1.3. POWER DENSITY SPECTRUM OR ENVELOPE
The EMI filter designer is not blessed with a density spectrum as described above.
The designer knows the power line frequency and that it will have some
harmonics. The lower frequency harmonics should be passed on to the load. Most
are aware that the harmonic power provides no power to the load, but neither
should the EMI filter attenuate them. The devastating effects of the magnetic
fields are not as drastic as the higher frequencies, so this filtering is not as
Chapter 1
important. Such attenuation of these low frequencies would call for larger filter
capacitors, and capacitor currents, and larger cores. These cores would be subject
to greater eddy currents and hysterisis losses. Then each core would develop more
heat within the filter. The filters must pass high power at the line frequency and
also the power of some of the odd harmonics of this frequency. This demands
larger components to handle the energy passed to the load, together with the larger
wire diameter required for the current. This dictates that the filter will increase in
cost, size, and weight. However, if it is a requirement to attenuate these frequencies, a silicone tapewound toroid transformer could be added to the filter
components and the low-frequency loss would be great.
Summarizing, for the reasons just given and for other reasons covered later
in this book, the cutoff frequency usually starts around 4000 Hz for 60-Hz
systems and around 8000 Hz, if possible, for 400-Hz systems to keep the
components small for better performance (see Sec. 1.7).
1.4. POWER TRANSFER
The power to be attenuated is typically minimal compared with the power
transmitted or throughput. Exceptions would occur when there are larger electromagnetic pulses from lightning (EMPs are discussed in Chapter 11). These
conditions must also be treated or attenuated. This typical noise, called flea
power, could be from any or all of the following: (1) the switcher frequency plus
the odd harmonics, (2) the power supply and/or the switcher diode noise, and
(3) the parasitic oscillations of the switcher inductor or transformer. This noise
energy is usually much lower in level than the power supplied to the load. To
move this power from the line to the load, larger filter components are needed.
The transfer of power creates (1) larger filters with more weight and cost
and (2) much lower self-resonant frequency (SRF) values. The weight, size, and
cost of the inductors and capacitors are directly proportional to the power. The
SRF value of these filter components is indirectly proportional to the power
demanded by the load. This limits the filters usefulness at the higher frequencies.
These conditions will dictate that other components must be added to the filter to
compensate for this lack at higher frequencies. Unfortunately, this increases the
weight, size, and cost of the filter. The assumption here is that the SRF values of
the inductors and capacitors are well below the specified upper frequency limit
that must be filtered. These additions to the filter could be handled by placing
small ceramic or extended foil Mylar or polypropylene (depending on the power
frequency) capacitors in parallel with the existing capacitors. The second capacitor will raise the SRF. Capcon and ferrite beads could also be used.
To review, smaller components have higher SRFs, making them perform
better at higher frequencies. Several stages in tandem reduce the size of the
components, resulting in higher SRFs and lower component costs.
1.5. SPECIFICATIONS: REAL OR IMAGINED

Specifications are another thing that makes EMI black magic. Some test
specifications cloud the design and make it overspecified. For example, one
company had been using a particular filter for years without any problems
meeting the EMI qualifications. Then the test specification was changed from the
220A to the current injection probe (CIP) method and the filter never passed.
These specifications are such that they conflict with reality. The 220A specification calls for losses within the filter with a source and load impedance of 50 ohms.
The filter will, in reality, feed a power supply that is rarely close to 50 ohms and
work into a source of rarely 50 ohms. In the preceding example, the system was
for aircraft where the power feed would be very short. For this line impedance
to approach 50 ohms, the line frequency would be in the upper kHz range.
What does the EMI filter designer do: match the 50 ohms or meet the real-world
specification? What is meant here by the real-world specification is similar to
the quality (qual) test that may follow. This test measures the noise out of the
system through the filter. The line impedance simulation network (LISN) is often
used as the source impedance for these tests and is closer to the real-world
requirements.
Most specifications that call for an LISN require 50 ohms output impedance. Unfortunately, the 50 ohms is not reached until well above 100 kHz. If the
only concern of the designer is to match this 50 ohms, the filter will be matched
to this source impedance of 50 ohms. There are two concerns here. First, what is
the lowest frequency and loss required? Is it below 100 kHz? The LISN output
impedance drops rapidly from 50 ohms and the filter is then mismatched. Second,
what happens in the real world when neither the source nor the load impedance
is close to 50 ohms? Figure 1.1 shows a typical LISN where the line impedance,
whatever it is, is shunted by the input network at frequencies above the point
where the 22.5-F capacitors impedance is equal to 1 ohm. This is at 7000 Hz,
so at frequencies above 10,000 Hz the line impedance is reduced to the 1-ohm
value, while the load side of the LISN looks like 50 ohms, following the same
logic, also above 10,000 Hz. The impedance of the inductor is low at 10,000 Hz
(3.5 ohms). At 10 times these frequencies, the inductive reactance is only 35.2
ohms and reaches 50 ohms at 142,000 Hz. The entire network at 142 kHz
generates a source impedance for the filter of 35 ohms at 44 degrees. It is obvious
that the LISN will not look like 50 ohms until well above 150 kHz. The proposed
LISN is shown in Fig. 1.2.
The real intent of the filter is to attenuate conducted emissions of differential and common mode origins from both the device and the line. The test
specifications rarely prove that the filter will pass with any degree of satisfaction
within the system specification or real-world specification. The filter can often
pass the insertion loss in the test laboratory and fail when tested along with the
Chapter 1
L1
56 H
C1
22.5 F
LINE
R1
1 Ohm
LOAD
C2
22.5 F
50 Ohm term
R2
1 KOhm
FIG. 1.1 Typical LISN.
system. Often, using a filter that passed the full test as a target, or bogie, gives
disappointing end results. If the filter in question appears to pass the bogie, this
filter is often been tagged as bad later by the system tester. The 461 specification
(Fig. 1.7) is more realistic than the 220A, and Robert Hassett, the vice president
of engineering at RFI Corp (now retired), has given several presentations for the
Institute of Electrical and Electronics Engineers (IEEE) and other groups that
show the advantage of moving away from the 220A test method to the CIP
method. Mr. Hassett has tested many filters by both methods and shown the
difference in the insertion loss curves (Fig. 1.3). The curves in Fig. 1.3 show the
difference using an L filter with the capacitor facing the line as compared with
the inductor facing the line (Fig. 1.4). The same is also true for the pi filter due
to its input capacitor facing the line. Either filter will look good under the 220A
tests. This is due to the 50-ohm source impedance. The CIP utilizes a 10-F
capacitor, and this will reduce the filter loss by 6 dB, especially at the lower
frequency end.
In the case of Fig. 1.5, the signal generatornow normally from the
spectrum analyzer tracking generatorhas an output impedance of 50 ohms and
feeds a coaxial switch (not shown) and the load impedance is the receivers input
impedance (also 50 ohms) fed through a second coaxial switch. A calibration path
is provided between the two coaxial switches. This test method makes the pi filter
function as a true three-pole filter, giving 18 dB per octave, 60 dB per decade.
The real-world initial losses at the low-frequency end would shunt out the input
filter capacitor. This would give initially 12 dB per octave of loss until well over
L1
LINE
C1
2 F
R1
5 Ohms
50 H
LOAD
C2
.25 F
50 Ohm term
R2
1K Ohm
FIG. 1.2 The proposed 461 LISN.
100 kHz (depends on the line length). This test method masks this flaw of the pi
filter or the L filter with the capacitor facing the low impedance.
Figure 1.6 shows the method used by Robert Hassett at RFI.
In the 461 specification, the current probes feed the measuring equipment
to compare the two currents. This method shunts the input pi filter capacitor
copying the real world up to the frequency where the 10-F capacitors SRF takes
effect. This method makes the pi filter loss 12 dB per octave, 40 dB per decade,
instead of 18 dB per octave, 60 dB per decade.
Other test methods have been suggested. Mitchell Popick, previously with
Axel Corp. (now with The James Gerry Co. in Alhambra, California) and a
member of SAE EMI group and the dB Bunch, recommends that the load side of
the filter face a diode bridge that is properly loaded. This is much better for those
that will feed a power supply, which is about 95% of the time. This also shows
how the filter handles diode noise. The diode noise is the leading noise competing
with the switcher noise. These are primarily the odd-order harmonics of the power
line frequency and spikes during turn on and turn off. It works well in three phase
systems as well, but make sure that the total inductive reactance of the filter
inductors is much lower than the primary inductance of the transformer. This is
especially true when multiphase transformers are to be used. These are often
autotransformers and the primary inductance is much lower than for the isolation
transformer type. The inductance of the filter and the primary inductance of the
transformer can form a voltage divider reducing the voltage feeding the load.
Chapter 1
100
CIP INDUCTOR
INPUT
90
220A
80
70
CIP CAPACITOR
INPUT
60
50
40
220A
30
20
10
0
0
.01
.1
10
100
1000
FIG. 1.3 Curves for L filter with capacitor facing low-impedance source.
This is another reason to avoid the commercial filters, where one filter fits all.
These may work fine in some applications and fail in others.
In Fig. 1.7 the spectrum analyzer sees the diode, switcher, and parasitic
noise that is allowed to pass through the filter under test. If the load is the system
rather than the load resistor, as in the drawing, the analyzer will get the full noise
signature of the equipment. This is much more of a real-world test because the
filter must handle all the noise sources at the same time, which is what happens
in the real world. In other words, the filter could saturate under this condition
while the CIP and 220A are looking at a single frequency from the tracking
generator. Of course, both 220A and CIP pass the full power from the line, but so
does the test method above.
Reviewing, try not to use a capacitor input filter for the CIP test method
because the loss is 0 dB per octave for this component whereas it should be 6 dB
per octave. This component costs money, demands filter room, and adds weight
without performing until the frequency is very high.
L filter with capacitor facing the low-impedance source.

FIG. 1.4
Input Capacitor
Input Inductor
10
Chapter 1
DC
Buffers
Signal
R1
R1
L1
50 Ohms
Filter under test
50 Ohms
C1
C1
Reduces to:
Signal
R1
L1
50 Ohms
Filter under test
50 Ohms
C1
C1
FIG. 1.5 220A test setup.
R1
11
1.6. THE INDUCTIVE INPUT FOR THE 220A TEST METHOD

This is similar to the capacitor of the pi filter in the CIP method but not as severe.
With the 50 ohms impedance in the 220A test system, what is the inductor
impedance going to add? At least the 50 ohms is there and the inductive reactance
adds to it at 90 degrees. We are speaking here of either an L or a T filter. They
are not responsive until the impedance of the inductor reaches 50 ohms. Regardless, this takes effect orders of magnitude ahead as compared with the capacitance
to ground in the CIP method. Both L and T perform very well in the CIP method
but are somewhat limited in the 220A method. Calculate the frequency at which
the inductor is 50 ohms. This will be the starting point where the inductance will
start to function. This explains why most filters are the pi type required to pass
the 220A specification.
Summarizing, watch the inductors of the L or T filters in the 220A test
method. See the frequency at which they reach 50 ohms.
Current Probe
POWER
10 F
Current Probe
L1
Filter under test
C1
C2
Current Injection Probe

Reduces to:
Current Probe
Current Probe
L1
Filter under test
C1
C2

FIG. 1.6 CIP test setup.
12
L1
POWER
LISN
FILTER
UNDER
TEST
C1
TRANS1
L1
DIODES
SPECTRUM ANALYZER
FIG. 1.7
The diode test method.
Chapter 1
13
1.7. THE 400-HZ FILTER COMPARED WITH THE 50- OR

60-HZ FILTER
The problem with the 400-Hz power frequency is the voltage rise. Again, we are
speaking of a system requiring substantial serious loss and load current. It is the
cutoff frequency of the EMI filter that creates the problem. There is always a
substantial voltage rise ahead of this cutoff frequency that pushes up the frequency spectrum at 400 Hz. This creates a severe voltage rise at 400 Hz. At 60
or 50 Hz, the rise is so much smaller that other factors will compensate for it,
making the output voltage the same as the input or a slight voltage drop. This will
be handled later in the book, but the main answer is getting the cutoff frequency
as high as possible. This requires multistage filtering and impedance matching.
For the same amount of loss (80 dB at 100 kHz), as the number of stages grows,
the cutoff frequency increases, dropping the gain at 400 Hz. However, as the
current level decreases through the filter, this often enhances the voltage rise
problem. There is also a technique referred to as RC shunt that decreases resonant
rises, say here at 6000 Hz. The resonant rise could be 10 to 15 dB. The resistor
could be 10 ohms (covered later in the book) and the capacitive reactance at
6000 Hz would be 10 ohms. Would this lower the resonant rise? Of course it
would, but now you have 5 dB at 4000 Hz. So what is the gain at 400 Hz? About
the same as before, but I have seen cases where this fix made the condition worse.
The resonant frequency dropped in dB but was also moved to a lower frequency,
negating the fix.
Another method is the Cauer, or elliptic, filter. In a multistage filter one (or
two) of the inductors is tuned to, in this example, 100 kHz. This adds many dB
of loss, allowing all the filter component values to decrease until the filter loss is
brought back to the needed 80 dB at 100 kHz. This pulls the resonant rise
frequency farther away from the 400 Hz and reduces the voltage gain. Of course,
each change in the inductor values requires a change in the tuning capacitor so
that it still resonates at 100 kHz.
Reviewing, 400-Hz filters with high loss requirements at low frequencies
demand special handling to get the loss low enough and still have little voltage
gain at 400 Hz. The limit appears to be a 3% tolerance. The near-perfect design
via computer will miss the mark but should fall easily within the 3% range. I have
seen systems with 118 V in and 126 V out.
1.8. THREE-PHASE FILTERS

There are several styles of three-phase filters. For the larger current requirements,
there is the individual removal filter, which others call the insert filter. Each
filter of this type has large capacitance to ground. They are mounted directly to
ground and are usually installed within an enclosure. These EMI filters are called
14
Chapter 1
three wire, four wire, and five wirethree for the delta, four for a wye with a
neutral, and five with an added ground wire. All three-phase filters and, in case
of a wye, a fourth filter of the same size for the neutral are included within this
enclosure. Any leakage current, reactive current to ground, specification must be
removed because these large capacitors are directly tied to ground. This type
requires a very good grounding system, and if mounted off ground, the filter will
not function properly. RFI Corp. has built these filters for as much as 1000 A per
phase. Here, the individual filters must pass the specificationusually 220A
and are mostly of the multiple pi arrangement.
The next style is for the lower current types, typically below 100 A per
phase. In some of these, leakage current specifications are given either by a
current-to-ground limit or the maximum capacitance to ground. The three phases
are in the same box. Here, the design is based mostly on how the filter is going
to be tested. If the test is based on testing each leg to chassis (ground), the other
filter legs will be grounded on both input and output terminals. Now the
capacitors between the lines will work to ground, doubling the capacitance and
making it easier to pass the tests. The feed-through capacitors to ground add to
this capacity. In common mode (covered later), all the input terminals are tied
together, as are the output terminals, so the three are working in parallel. Here,
the line-to-line capacitors do not help the common mode loss.
A worse condition exists if the power line frequency also happens to be
400 Hz (see the preceding section). The best possible solution is a multistage in
each leg with Cauer filters to bring the resonance rise as high in frequency as
possible to reduce the voltage rise.
The common mode inductor should have the neutral (if required) wire
wound on the core as well as legs A, B, and C. The unbalanced current will flow
through the neutral wire and through the common mode inductor, canceling the
unbalance within the common mode core.
To review, if this is a three-phase 400-Hz filter requiring substantial
differential mode loss, the easiest to design is the individual removal filter or the
insert filter. The types with all phases and the neutral in the same enclosure are
the hardest to design. Virtual ground techniques can be used to reduce the leakage
currents if the three voltages feeding the three wires are nearly equal. This
technique greatly enhances the common mode loss. The virtual ground adds to
the feed-through capacitors to ground, adding common mode loss. The question
is how much capacitance you want between line to line and line to ground. Often
these values are part of the specification. Again, use multistage sections to
increase the filter resonant frequency. This will lower the 400-Hz voltage rise.
2
Common Mode and Differential
ModeDefinition, Cause, and
Elimination
There is a wide range of opinion about the definition, cause, and elimination of
common mode noise and differential noise. This should cover most of these ideas.
2.1. COMMON AND DIFFERENTIAL MODE DEFINITIONS
A basic definition of common mode and differential mode is required. Differential
mode means the normal transfer of energy down the line. In fact, this is also called
normal mode. A voltage across the line with a current flowing in one direction in
one wire and the opposite direction in the other wire is normal mode. In this case,
the subject is differential mode noise. In other words, it flows just like normal
power in the line energy.
Common mode means a voltage impressed across both, or all, lines. This
voltage is between all these lines and ground. If there is only one line, then the
pulse is still between this line and ground. In this unbalanced case, differential
mode and common mode act the samebetween line and ground.
A current flows in the same direction in all the lines and the return is
ground. Again, the subject is common mode noise. I have found that opinions
vary from EMI guru to guru with little agreement. I hope this section will, at least,
achieve some agreement among the various groups. Many claim that if the
common mode noise voltage impressed on these lines is not exactly equal, then
it is not common mode.
16
Chapter 2
Two signals cannot be equal on both lines because of differences between

the lines, the different spacing between the lines and ground, different capacitance
to ground between the lines, and so forth. Even the EMI filter input feed-through
capacitors, MOVs, transzorbs, and the like upset the common mode according to
some groups. Therefore, using that definition, common mode does not exist.
But common mode does exist, so this definition must be lacking or faulty.
To sum up, differential mode noise voltage is impressed between the lines
whereas the common mode noise is across the linestypically twoand ground
(Fig. 2.1).
2.2. WHAT CREATES COMMON MODE NOISE ON THE
LINE SIDE?
The simple definition of common mode noise is a pulse of voltage on both power
lines of equal value (Fig. 2.1). This pulse is between the power line wires and
ground. The EMI filter should be designed to handle this energy. A lightning strike
on the power line side will create a magnetic field that will cut the two, or more,
power line wires. This voltage is impressed between the lines and ground.
This strike will be several quick high-voltage pulses typically around 50 kHz.
The spacing between the lines may be 3 or 4 feet, depending on the voltage and
location, creating a slightly different voltage in the two power lines. This voltage
will be added algebraically to the AC power line voltage on all the lines. All of
this section assumes that the lines do not fuse and that transformers will take this
pulse without failing. If any failure occurs on the line, this reduces the high-pulse
problem at the filter and equipment following, but the power will be down. It will
be the difference in the two line voltages feeding the transformer that is transformed to the secondary. This difference between the lines is now transferred to
differential mode noise. This difference voltage will be transformed (stepped
down) to the user side. There will be extra transformer losses due to the
high-frequency core losses. These noise pulses are at higher frequencies, accentuating core losses. The skin effect within the transformer and on the lines where
this high-frequency pulse is being conducted adds to the pulse losses. The primary
FIG 2.1 V1, differential mode, and V2, common mode.
Common Mode and Differential Mode
17
to secondary capacitance of the transformer (Fig. 2.2) carries the common mode
part of this pulse.
However, the secondary has many paths to ground through the winding
capacity to the center taps at the transformer and each utility power service panel.
The capacity between primary and secondary is further reduced if the transformer
has a Faraday shield or screen. The high frequency of the pulse is further reduced
by the interwinding secondary capacitance of the transformer.
Most of the voltage transferred from the primary to the secondary through
the capacitance will be shunted to various service grounds (Fig. 2.3). It is
primarily the difference voltage across the transformer that will carry the pulse to
the utility service users. However, this voltage is differential mode and is carried
by the two outer legs.
Often, this voltage from the power company is called two phase because it
is 180 degrees out of phase. If the equipment works off both outside lines (220 V),
the filter must handle this reduced unwanted common mode pulse. Both lines
from the power panel are black (sometimes two different colors are used), and a
safety green wire goes back to the service ground. Usually, the filter will have
three transzorbs (see Chapter 9): one from line to line and the other two from each
line to equipment ground and carried by the green wire. These transzorbs will
have two different ratings. The line-to-line transzorbs will be rated at 250 or
275 V RMS and the two line-to-ground transzorbs will be rated at 150 V RMS.
The transzorbs are rated, or listed, by their RMS value. A Harris transzorb
V150LA20B has a rating of 150 V RMS and will fire at voltages around 212 to
240 V. The purpose of the MOVs, or transzorbs, is to eliminate these different
noise pulses. The line-to-line MOV, or transzorb, eliminates differential mode and
FIG. 2.2 The line transformer with the CM capacitors and center-tap grounds.
18
Chapter 2
FIG. 2.3 Service, showing other services, power meter, and busses.
the line-to-ground MOVs or transzorbs eliminate common mode and help with
the differential mode noise.
If the power for the system works off either side to the central ground, the
pulse is differential mode. Most equipment has three lines to it from the service:
the hot wire, typically a black wire, tied through the circuit breaker at the service;
the neutral, typically white, tied directly to the service common ground; and the
safety ground, typically green, also tied directly to the service common ground.
The hot black wire carries the unwanted pulse to the equipment and the white
common ground carries it back to the common ground. It is the filters job to
handle this pulse of noise that is now differential mode noise. Typically, the filter
will have a transzorb from the hot to the filter case, or equipment, groundthe
green wireand another tied between the hot wire and return.
If the pulses are between the transformer and the user, the magnetic pulses
cut the wires in the same way but the central wire is grounded repeatedly at all
the services and transformer. Then the two outside lines are carrying the common
mode pulses. If the equipment is tied to the outside lines, plus the grounded green
wire, the filter must handle the common mode problem. The three transzorbs will
be sized as before with two rated above 120 V RMS and the one from line to line
rated above 220 V RMS, typically 250 V RMS. If the equipment is tied from one
line to ground, the noise energy is differential mode carried by the hot and
return wires, typically black and white leads. The transzorb would be rated above
120 V RMS.
2.3. WHAT CREATES COMMON MODE NOISE FROM THE
EQUIPMENT SIDE?
Storage capacitors in most power supplies are hooked between the diode outputs
and ground. This wire is the chassis green wire from the service ground. As the
voltage on the storage capacitor raises and falls with respect to ground, the in-
19
coming power lines (black and white wires) follow this with respect to the ground.
This creates common mode from the equipment side back toward the line.
Switchers do the same thing. An input transformer would eliminate this, and so
would power factor correction circuits. Isolating the input supply from ground by
placing the storage capacitor across the diode bridge and then following with an
isolated switcher also works to remove the common mode noise.
Figure 2.4 shows the isolated supply with the storage capacitor (SC), the
switch (Q1), and the load resistor (R1). Besides removing the common mode
problem, the ground (green) wire has little current on it, referred to as leakage,
or reactive, current. The EMI filter in front of this supply can be balanced with
the differential capacitors from line to line, not to the green ground line, and
leakage current will be minimal. Only small capacitors to ground, feed-through
style capacitors, with a common mode choke will handle the common mode
problem if the system is working on the 120-V side. More common mode loss is
required if the system is off the 220-V side. However, larger capacitors to ground
may be used because the system is balanced. See Secs 2.4 and 2.5.
2.4. WHAT ELIMINATES COMMON MODE NOISE FROM
THE LINE AND EQUIPMENT?
Because common mode noise is between the lines and ground, capacitors to
ground are required. Also, common mode inductors are used and will be discussed
later. Here, the reactive capacitor current to ground, also called leakage current,
will be the main subject. This ground current is the difference in current in the
ground capacitors from both sides of the lines. A good isolation transformer
eliminates both the leakage current and the common mode noise.
FIG. 2.4 Simplified isolated input section of the power supply.
20
Chapter 2
Figure 2.5 shows two capacitors to ground for 220 V AC balanced lines.
These capacitors could be feed-through type or leaded capacitors. If the voltage
is equal and opposite (180 degrees out of phase) and the capacitors are equal, the
ground current is zero at the line frequency. This works for capacitors with leads
as well as for feed-through capacitors. If you have read the beginning of this
chapter, you are aware that there is no such thing as 100% balance, so there will
be some current to ground. An example would be 115 V from line to ground, 230
line to line, 5% capacitor tolerance, 5 mA allowed to ground and 60 Hz. Note that
if the capacitance is a limit specified by the customer and the customer simply
takes a capacitor measurement from either line to the common ground, this
method will not pass. If the current limit is specified, isolate the filter and load
from ground and measure the green wire current, and this method will pass. The
current through either capacitor to ground is
I1 = 2VFC1
(I1 I2) = 0.005 = 2VF(C1 C2)
The difference in the capacitors is double the percentage, here 0.1C (one 5% high
and one 5% low),
5000
0.005 106
=
= 0.1C
2VF
2VF
5000
= 1.15 F
C =
0.1 2 115 60
changing the capacitor value to F and substituting the 0.1C.
Therefore, two feed-through or leaded capacitors of 1 F will work well if
the capacitor tolerance is 5% or less. Remember, the values used in this configuration works only in the line-to-line system and a much smaller single value of
capacitance to ground would be required in the 120 V to ground arrangement to
meet this low current value. Also, this should never be used for medical equip-
FIG. 2.5 Balanced line-to-line capacitors for 220-V balanced lines.
21
ment touching patients. If one of the wires opens, the full line-to-ground voltage
(120 V) of the one remaining line is impressed across the one capacitor and the
current to ground through it is 45 mA, well above the patient limit.
Another arrangement for the two-phase balanced system, which also eliminates using the feed-through capacitors, is to replace the ground at the common
point with a capacitor to ground. This is used for common mode attenuation. Here
V is the line-to-ground voltage, C1 and C2 are the line-to-common capacitors
(assuming C1 is the larger of the two), C3 is the junction-to-ground capacitor, e
is the junction voltage, and IL is the maximum leakage current to ground
requirement through C3. The equations follow. If t is the tolerance, here from
above 0.05, then the maximum difference in the capacitance is 2tC, but the
addition of these two is assumed to be 2C. Assuming a common mode pulse of
equal amplitude on both lines to ground, the two lines to junction add (in parallel),
giving 2C. The total capacitance to ground would be the two in parallel and then
C3 in series. Simplifying,
e =
V(C1 C2)
2VtC
=
(C1 + C2 + C3)
2C + C3
C3 =
IL(C1 C2)
2FV(C1 + C2) IL
C3 =
2tILC
4FVtC IL
CT =
IL
2FVt
The most practical solution along with the best overall performance is to have
the three
C =
3IL
4FVt
capacitors equal. Make C3 equal to C in the preceding equation. Round down the
three capacitors to a convenient standard value. Just make sure that the tolerance
limit of the two line-to-junction capacitors is correct. The junction to ground is
not as critical. Another method, but a costly one, is to sort, or grade, the capacitors
into smaller difference percentages. Use the matched ones for the two line-tojunction capacitors and the oddballs for the junction to ground.
A difficulty arises if someone uses this 220 line-to-line filter for a 120-toneutral filter. Now the junction-to-ground capacitor and the junction-to-neutral,
or the return, capacitor are in parallel, and the voltage from junction to ground is
V divided by 3. In this situation, the current through C3 to ground is
22
Chapter 2
IG =
3VIL2F
IL
=
3 4FVt
2t
where IL is the original design leakage current and IG is the new ground current
and it is assumed that the actual calculated values of the capacitors are used.
Otherwise, IL is the resultant leakage current, which is lower than the specified
value. In the preceding case, with IL equal to 0.005 and t equal to 0.05, IG is equal
to 50 mA. This is well out of specification. Hopefully, the leakage current is not
specified for this requirement or is a larger value for this application. A 220 V AC
balanced filter, designed as above, should be marked on all documentation and
the system should not be used for a 120 V AC line-to-ground application.
If the system is being built in house where the filter, power supply, and the
rest of the system are under engineering control, build the filter in as part of the
supply and design a current transformer for the ground lead of the capacitor. Use
this to shut the system down if a voltage imbalance occurs or there is excessive
current to ground for any reason. Put this network reasonably close to the line
input with a current transformer on either side of the capacitor to ground. Design
this to operate a relay that opens the system after the filter. I would suggest that
the relay energize only in the excessive ground current.
Summarizing, the balanced line for 220 V AC for two phases with leakage
current specifications can be met with the three-capacitor arrangement. This
technique was developed for common mode noise. The normal voltages are 180
degrees out of phase whereas common mode is in phase. So the capacitors buck
for normal mode and add for common mode. This adds to the loss of the common
mode inductor. It is not advisable to use this in any medical equipment that would
touch a patient. A way around this is a current probe that would shut down the
equipment, and it would be better if this also opened the leads to the patient. If
this cannot be done, mark all documentation never to use this system on a 120
system working between line and ground or for any medical application.
This application is similar to the virtual ground for three-phase systems.
2.5. WHAT CREATES DIFFERENTIAL MODE NOISE?

In Sec. 2.2, the common mode pulse was discussed. A transformer was placed
between the lightning strike and the filter. For the most part, this created
differential mode energy at the filter. Thus, it is recognized that common mode
can create differential mode noise. I am not familiar with the reverse condition.
On the line, differential mode noise is also created with inductive equipment on
turn on and mainly turn off. This is a voltage pulse between the lines. Likewise,
transformers between the output of the filter and equipment, or the isolated
switcher transformer (see Fig. 2.4), produce differential mode for the filter to
23
FIG. 2.6 Balanced line-to-line capacitors with extra capacitor.
handle. For the line side, MOVs, or transzorbs, help to clip the higher pulses from
the line and equipment, and the differential filter section must handle the rest.
2.6. THREE-PHASE VIRTUAL GROUND
This technique can be used only for the three-phase type all within the same
enclosure. Types requiring individual insert filters with all the capacitance to
ground cannot use this technique.
This is very similar to the preceding two-phase application. A capacitor is
tied from each phase to a common point, making a virtual ground. If the RMS of
each phase voltage is the same (this is not a function of load current) and the three
capacitors are the same, then the junction voltage is zero, or a virtual ground.
If a fourth capacitor is tied between the junction and ground, the current through
this capacitor is zero. If the unit is tested for ground currentnot capacitance to
groundby isolating the equipment (or dummy load) and the filter from ground
and measuring the current on the ground wire, typically green, the current should
be well below the specification. Again, the capacitor values are equal for best
overall results. Solve equations similar to the preceding equations. The voltage at
the junction is equal to the line voltage if one phase fails and the phase angle will
be between the two remaining phases60 degrees from each. Therefore, this is
risky for medical use involving application to a patient. However, usually
three-phase high-power equipment never touches the patient. Use a current probe
to monitor the current and open some relay to remove the power from the
equipment or the patient or both. The latter assumes the engineer has control over
this and not just design of the EMI filter. This situation is similar to that in
Sec. 2.4, and it would be advisable also to read this section.
The best part of this is that the three-phase voltages give nearly zero current
to ground, but for the common mode voltages, the three capacitors are in parallel
with the fourth in series. This gives very low impedance to ground to eliminate
the common mode noise.
This technique plus other small feed-through capacitors along with a
properly designed common mode inductor will surely eliminate common mode
24
Chapter 2
problems. Remember, if there is a neutral wire, there should be equal windings

for the three phases plus an additional winding for the neutral wound on the ferrite
common mode inductor. Any unbalance in the phase currents is carried by the
neutral, so the total magnetic flux will be zero in the common mode core.
The theory is that the neutral current is low in a balanced system. Any multiple
of the third harmonic that exists (3, 6, 9, etc.) in the system adds back in phase
and adds on the neutral. So, the total third-order currents and the unbalanced
current add on the neutral. Make sure the wire size can handle at least the peak
phase currents. I have had many calls about neutral filters overheating well above
the temperature of the other phase filters, and this is due to unbalance current and
the third-order harmonics. Measure the current on the four wires to see if the
neutral current is not well above the other three.
The equation for the leakage current follows. E1 and C1 are the voltage and
capacitor associated with the first phase, and so on with E2, C2 and E3, C3, while
C4 is the capacitor between the junction and ground.
2
4
2FC4 E1C1 sin(2Ft) + E2C2 sin 2FT +
+ E3C3 sin 2Ft +
3
3
C1 + C2 + C3 + C4
To review, what removes common mode noise? The answer is common mode
inductors, capacitors to ground, transformers, and transzorbs.
3
EMI Filter Source Impedance of
Various Power Lines
One of the leading questions asked by people who have an EMI problem or are
joining the EMI field is related to the source impedance. Several senior scientists
and doctors of engineering have asked about the transfer function of a certain
filter that RFI Corp. proposed for their application. They were going to measure
the line impedance along with the load impedance and wanted to know the
transfer function of the EMI filter that was designed in. The transfer function can
be calculated, but most EMI filter manufacturers rarely measure them. If the filter
manufacturer, here RFI Corp., could provide this information, this company was
going to add another section between the filter and the load. The purpose would
be to notch out resonant rises and other instabilities of the entire system. This
procedure would work very well if they were assured that the line length,
conductor spacing, or diameter of the conductor would not change much from
installation to installation. These parameters require precious time to determine
and could require expensive rental equipment for most companies. This would
work well if their system, using the EMI filter device, were to be installed on one
type of ship or aircraft, where these cable dimensions would be nearly the same.
The technique would fail to work if the unit was installed on different vessels or
used in various other applications. Also, they would need to have a look at all the
possible load conditions, not just the peak load. If the device would go into some
standby mode, the instabilities could shift and the filter along with the rest of the
equipment could oscillate.
The harmonic content of the power line frequency varies from line to line.
In the past, most commercial lines had little harmonic content because of the very
26
Chapter 3
low line impedance at these frequencies. This was discussed with a large power
company in Southern California in 1985. They stated that 85% of the power was
then used for power and lights, that is, motors and lamps and, at that time,
incandescent lighting. With the proliferation of computers, computer printers,
scanners, copiers, fax machines, televisions, sound systems, and similar equipment now tied to these linesand very few with power factor correction
circuitsthe voltage waveform is less sinusoidal. This is due to the voltage drop
caused by the high current spikes that these machines demand a little ahead of 90
and 270 degrees. The power factor correction capacitors placed across the line by
the power company do help to reduce the current gulp, but the voltage is still less
sinusoidal. The harmonic content of generators at remote sites and shipboard
installations is much greater because of the higher resistance of the generator and
lines. The voltage supplied to the end users in these applications is less sinusoidal. A
power consultant in Southern California found 100-A spikes above the nominal
sinusoidal current in small office buildings that had power problems. Again, these
spikes were a little ahead of 90 and 270 degrees, typically around 85 and 265 degrees.
3.1. SKIN EFFECT
As the frequency increases on the line, the depth of conduction is reduced. The
wire cross-sectional area decreases because the radius of conduction decreases.
The higher the AC resistance, the greater the dissipation of this unwanted energy.
Skin effect can take its toll on the higher frequency energy on the power lines.
This helps dissipate the Electromagnetic Pulse (EMP) and other higher frequency
noise traveling on the power line in either direction. These power lines were
constructed to handle power at very low line frequencies and not for the higher
frequencies creating the loss. Although the characteristic impedance of the line
may be 150 ohms, the loss of the line per unit length increases with frequency.
For copper the equation of the skin effect depth in centimeters is
D = depth (cm) = 6.61
---------F
(3.1)
The cross section of the conducting area (CA) of the wire for frequencies above
the skin depth is
CA = |R2 (R D)2|
= D(2R D)
(3.2)
where R is the radius of the wire in centimeters and D is the skin depth from
above, also in centimeters. As the frequency increases, D decreases such that the
D term is much smaller than 2R. Equation (3.2) at these upper frequencies is
reduced to
EMI Filter Source Impedance of Power Lines
CA = 2RD
27
(3.3)
The original cross-sectional area, R2, compared to the value of CA, times the
original DC resistance, will give an approximate value of the AC resistance at
these upper frequencies due to the skin effect.
Rac =
=
R2Rdc
2RD
(3.4)
RRdc
2D
the lowest frequency would be well above the value of

F =
6.612
R2
(3.5)
Below this frequency, the skin effect is the full radius of the wire. The frequencies
we are discussing here should be several times this lower frequency. Replacing
D with its equation above,
Rac =
RRdc
F
13.22
(3.6)
If Rdc is the resistance in ohms for a small distance along the line, then Rac will
be the approximate AC resistance for this short section. Rac, along with L, C and
G, the conductance across the line, will form a short segment of this line, say a
meter, Then the characteristic impedance of the line is
Z0 =
R ac + L
--------------------G + C
(3.7)
Substituting (3.6) into (3.7):
Z0 =
0.07564RR dc F + L
-----------------------------------------------------G + C
(3.8)
Simplifying yields:
Z0 =
0.07564RR dc F + 2FL
------------------------------------------------------------G + 2FC
deleting G because it is much smaller than 2FC at the frequencies discussed in

this section. The square root of F is removed in the first term in the numerator:
28
Chapter 3
Z0 =
0.07564RR dc + 2L F
--------------------------------------------------------2C F
(3.9)
To find the limit with respect to frequency (F), differentiate the numerator and
the denominator separately:
2L
0 + ----------2 F =
-------------------2C
----------2 F
2L- =
---------2C
L--C
The first term of the square root numerator vanishes and the Fs in the last term
of the numerator and denominator cancel.
Z0 =
L--C
(3.10)
Therefore, the skin effect term has little or no effect on the characteristic
impedance. The values of L and C are the dominant terms. This is the fundamental
equation of the characteristic impedance of coaxial cables. This all shows that the
normal characteristic impedance equation still dominates at the higher frequencies and this characteristic impedance does not change with skin effect, although
the loss per unit length does.
This behavior is similar to that of coaxial cable except that coax is designed
to handle higher frequencies. There are many different coaxial cables with the
same characteristic impedance. These lines have different inside and outside
diameters. Some coax lines have very small diameters and others have larger
diameters. The dB loss per 100 feet varies from coax to coax and also varies with
frequency. Think of it as a pad. The impedance of the pad may be 50 ohms but
the loss of the pad varies from pad value to pad value. The main difference here
is that this power line impedance also varies with frequency. It is this line
impedance that dissipates the unwanted energies, not the characteristic impedance
of the transmission line.
Skin effect also applies to the wire used for the inductors, the transformers (if used), and the rest of the EMI filter wiring. The purpose of the filter is
to rid the system of unwanted signals or noise. What better way is there than
to dissipate it? Use the skin effect to do this. For the filter inductor, never
use Litz wire or strands for the turns. Allow the wire to help dissipate these
upper frequencies. The same is true for the rest of the hookup wire. Is there an
exception? Yes. Some larger current filters have capacitors in parallel to ground.
Here, Litz or braided wire should be used so that the capacitors can shunt these
signals to ground. Otherwise, the capacitors will be limited in performance and
have a lower self-resonant frequency (SRF) due to the lead inductance.
29
It makes no difference how long the inductor leads are, and they are always
the self-leads of the inductor. This lead is often a single strand as covered earlier.
The leads properties add to the loss of the inductor. These leads add to the
inductance, and their skin effect also adds to the wire loss. The square blocks in
Figs. 3.1 and 3.2 represent quality feed-through capacitors in the range 3.5 to
10 F. The oval shapes stand for larger capacitors in the range 10 to 30 F that
provide low-frequency loss but for best results should be a braided wire. Figures
3.1 and 3.2 are top views.
The wires from the inductors are self-leads, but the wires to the oval
capacitors should be better quality braids. If the inductor self-lead is attached to
the capacitor, this is fine. But the capacitors to ground need the lowest impedance
to work. Leads with inductance and skin effect between capacitors lower the SRF
of the capacitor, degrading the performance.
Figure 3.3 shows a terminal hooked directly to an inductor. This lead could
have been longer. The inductors longer lead is tied to a feed-through that is part
of the output terminal. However, there is a second capacitor tied in parallel to the
feed-through. This last lead should be a braided lead that offers a very low skin
effect and low inductance. Even with a quality lead, this lead should be as short
as possible. There is no perfect lead here, and the cure is to make these leads as
short as possible. This larger high-voltage style of capacitor is soldered to the
container. Try to place this capacitor as close to the other component as possible
to keep the tie leads short. See veeing the capacitor in Chapter 7.
3.2. APPLYING TRANSMISSION LINE CONCEPTS
AND IMPEDANCES
Transmission line concepts (Fig. 3.4) may sound like a strange subject to
introduce when discussing EMI filters, but the filter designer needs to have a basic
understanding of this subject for several reasons. The first is to understand why
high impedance is needed at the filter input end for EMP applications. The second
is to understand that the power line energy loss is due not to the characteristic
FIG. 3.1 Typical high-current power line filters.
30
Chapter 3
FIG. 3.2 Internal wiring of typical power line high-current filter.
impedance of the line but to the resistance elements along the line. These losses
are due to elements such as skin effect, the DC resistance of the lines, and the
conductance across the line. The third reason is that, especially in multipleelement filters, the filter has characteristics similar to those of transmission lines.
The shorter power lines start taking on their varying characteristic impedance at much higher frequencies. The characteristic impedance of these power
lines varies all over the place; it varies with frequency and is not as constant as
that of coax, twin lead, and twisted pair. The characteristic impedance of the open
wire type normally varies typically between 90 and 180 ohms due to the spacing
between the conductors and the diameter of the wire. The paired type, or twisted
wires, enclosed in conduit is 50 to 90 ohms for the same reason as before with
the added capacitance between the wires and the conduit. The conduit adds little
shielding because of the thinness of the material. In about 95% of the cases,
several different power line sections will be in tandem. These different power line
impedances vie back and forth as the different power sections approach resonant
lengths. The velocity of propagation is very low due to being constructed to carry
only the power line frequency, not radio frequency. The electrical lengths of these
lines appear to be about eight times their actual length. If the power line is struck
with a pulse, the power line will dissipate some of this energy in the resistive
FIG. 3.3 Lead layout example.
L
C
31
L
EQUIVALENT
4R
4L
C
FIG. 3.4 The transmission line and differential mode equivalent.
elements mentioned earlier. The pulse will travel toward the filter end at the lower
velocity of propagation of these cables. The fundamental frequency is around 50
kHz for lightning and soon assumes the characteristic impedance of the power
line cable rather than the free-space impedance of 377 ohms. If the filter input
impedance is high or looks like an open impedance to this pulse, the voltage soars
(could double), aiding the arrester to function quickly. If the filter impedance is
low, or a short to this pulse, the voltage drops to zero and the line current doubles.
This slows or keeps the arrester (varistor or MOV or transzorb) from functioning.
This is covered in Chapter 9.
I have seen graphs from the IEEE literature showing some line impedance
spikes of almost 500 ohms at one frequency in the MHz range. Figure 3.5 shows
the typical impedance range that the various types of power lines fall between for
various twisted-pair cables. The graph is like an average of impedances, and the
variations fall primarily between the two curves shown. A very minor percentage
was above or below the curve.
The major part of these losses is due in part to the skin effect and the DC
resistance, especially at the lower frequencies, for the longer lines. Short lines
32
Chapter 3
FIG. 3.5 Impedance range of a short linetwisted pair.
appear to form link coupling. The point here is that at frequencies below 10 kHz,
both the longer and shorter lines look resistive and close to the DC resistance
value. This value is near zero ohms for the better commercial power companies,
but for most remote power systems, the resistance can be much higher. As the
frequency increases, these lines have an output impedance that reaches 4 ohms at
10 kHz. The impedance then ripples its way to 50 ohms near 100 kHz for the
longer lines and 250 kHz for the shorter lines. It should be obvious that the losses
required for the Military Standard (MIL STD) 461 specification are such that the
filter must be designed to have the proper losses with the line impedance in
the very low or the zero ohms region. This should make it clear why the pi filter
has trouble meeting the loss for the real-world and 461 requirements. The same
is true using a 10-F capacitor across the power line in the common current
injection probe testing. The losses that the filter must meet are at frequencies in
the region 10 to 14 kHz in the 461 specification and may be as much as 100 dB
for power line filters. This is the type of filter used in secure room applications.
The skin effect applies within the filter. Regardless of filter type, often
many feet of wires are used, so the skin effect will dissipate the higher frequencies
within the filter body. The wires used for inductors are selected to be sufficient
to carry the power line current at the power line frequency. The voltage drop and
33
inductor temperature rise will dictate the wire gauge used. Stranded wire is used
only to ease winding techniques rather than for high-frequency loss considerations. In fact, these high-frequency losses are beneficial to high-frequency filter
loss requirements. In addition to the skin effect losses, core losses at the higher
frequencies are beneficial in increasing the filter losses. These are dissipated and
not reflected to the source.
3.3. APPLYING TRANSMISSION LINE IMPEDANCES TO
DIFFERENTIAL AND COMMON MODE
The transmission line losses for the differential mode (Fig. 3.6) are all the resistive
elements of the line including the reciprocal of G, the conductance across the line.
These losses include the skin effect and some very minor losses due to the
equivalent series resistance (ESR) of the capacitor. The DC resistance (DCR) of
the inductor adds no loss because the DCR is already included in the wireit is
the wire. The impedances all along the line in each section will dissipate any
differential mode pulse, motor, or inductive load switch, propagating down the
line toward the line filter.
The difference in the requirement for common mode (Fig. 3.7) and differential mode attenuation is that G and the ESR of the capacitor have no effect.
A pulse traveling down the line, lightning or EMP, toward the filter is partially
consumed by the resistance of the line and the lines skin effect. Again, the DCR
of the inductor(s) is included with the wire losses (this is the wire). The
characteristic impedance also changes because the two lines are in parallel, so the
capacitor across the line is out and so is G. The two inductors are in parallel, about
1.5 H per meter on each line, or 0.75 H, and the capacitance to ground
determines the characteristic impedance. This makes the characteristic impedance
quite high. This information explains why the EMI common mode filter section
can be made very high.
4R
4L
C
FIG. 3.6 Equivalent differential mode line.
34
Chapter 3
EQUIVALENT
2R
2L
2R
2L
EQUIVALENT
4R
4L
C
C to Ground
FIG. 3.7 Equivalent common mode transmission line.
3.4. DIFFERENCES AMONG POWER LINE

MEASUREMENTS
Many groups have measured the line impedance and there has been little
agreement when comparing their results. The reason is that the lines being
compared vary so much. The velocity of propagation is quite slow in all these
lines, making the lines appear much longer than they really are. In some lines,
35
the ratio is eight times their actual length. Devices such as generators or
transformers terminate the line at the higher frequencies. This is due to the
capacity across these devices, which shunts the line at these higher frequencies.
These lines were designed to carry DC, 50-, 60- or 400-Hz power and not these
higher frequencies.
3.5. SIMPLE METHODS OF MEASURING AC AND
DC POWER LINES
Power line impedance can be easily measured, with caution, if the designer
wishes to do so. There should not be any need to do this, especially when using
the techniques used in this book. This measurement can be accomplished with
inexpensive test equipment on both AC and DC lines, but it would be much faster
and more accurate using a network analyzer. Either line, AC or DC, should be
resistive loaded and several runs will be needed. The first readings are with
maximum loading, followed by medium loading and then low loading.
To measure the AC line, start well above the line frequency by at least a
factor of 10 to keep as much of the AC harmonic power from the readings as
possible. This is done to assure accurate readings and to protect the measuring
equipment. The frequencies required will be over a wider frequency range well
into the megahertz area. A frequency-selective level meter (FSLM) should be
used along with a signal generator and a blocking capacitor. The capacitor should
be high impedance to the line frequency and low to the frequencies to be
measured. If the line frequency was 400 Hz, the lowest intended frequency to be
measured should be well above 4000 Hz. If the impedance of the capacitor is
1000 ohms at 400 Hz, the impedance would be 100 ohms at the lowest frequency
to be measured. The FSLM should be used in its narrowest input filter selection,
if available. The generator frequency should still not be tuned to any multiple,
especially the odd harmonics, of the line frequency until 20 times the line
frequency is reached. The loss across the capacitor cancels except at the lowest
frequencies, anyway, so that the 400-Hz loss could be still higher. The only
component remaining is a resistor of some value, say 400 ohms, and this should
be noninductive to well above the highest frequency reading. The resistor,
capacitor, and signal generatorin that orderare tied across the line with the
resistor at the hot end and the signal generator to the neutral. The signal generator
is tuned to the desired frequency, the FSLM has its low side tied to the neutral,
and two points of measurement are taken. First, the voltage reading is taken
between the capacitor and resistor where the FSLM is tuned to peak at the
generators frequency. The second reading is taken at the high line without
readjusting the FSLM frequency.
Make sure that the FSLM front end can withstand this AC line voltage
without blowing the front end. Many of the frequency-selective voltmeters or
36
Chapter 3
level meters have precision input pads that would quickly generate some ugly
blue, and expensive, smoke if the hot line is touched. A high-pass filter with a
series capacitor input, not an inductive shunt input, could be built in a probe to
be installed at the test lead to protect the FSLMs input (Fig. 3.8). The cutoff
frequency of this filter must be at least half the lowest frequency to be measured.
This filter does not have to be special or does not require a flat Butterworth
wave filter and can have ripple in the passband area of the high-pass filter.
The filter errors are canceled as shown in the following equation and seen in the
drawing later in this section. The ratio of the two readings will be the same
because the errors cancel in the equation at the same frequency. As an example,
at 400 Hz, the lowest reading should be 4000 Hz. The reason for the multiple of
10 is to avoid the potential high-level harmonic content on some power lines.
The high-pass filter should have a cutoff frequency of 2000 Hz. Then the
equations for the values of the inductors and capacitors in the high-pass filter
above are
L =
C=
Zfslm
50
=
= 0.004
22000
2Fc0
(3.11)
1
1
=
= 1.592 106
22000 50
2Fc0Zfslm
This assumes that the input impedance of the frequency-selective level meter is
50 ohms. Record both of the readings and move on to the next frequency to be
measured. The formula is
Z0@F =
VbR1
Va
C1
C1
L1
FIG. 3.8 High-pass filter.
L1
FSLM
37
The voltage level, Va, is recorded at F0 between the capacitor and resistor, and
Vb is the high line voltage reading, also at F0. R1 is the value of the series resistor
between the high side and the capacitor. F is the high-pass filter installed ahead
of the FSLM to protect the input pads. The impedances at these various frequencies can be plotted for the various loads and compared (Fig. 3.9). It is often better
to use a battery-operated FSLM and signal generator to avoid ground loops.
The DC measurements are almost the same, except that lower frequencies
can be read and the highest frequency needed will be in the lower kHz range
(compare Figs. 3.10 and 3.11). Obviously, the high-pass filter is not required.
These resistive readings can again be plotted on the graph. The readings will level
out (resistive) at a low frequency in the kilohertz range. Here, you could replace
the FSLM with an AC voltmeter. This setup does not require as much caution as
in the AC readings. Make sure, again, that your FSLM or AC voltmeter can
handle the DC output or line voltage. If they cannot, simply use a capacitor (this
replaces the high-pass filter) in series with the meter probe to take both readings.
Figure 3.12 shows the changes for the DC line showing either a frequencyselective level meter or an AC voltmeter and the blocking capacitor, if required.
The same equation holds. Again, the various impedances would be plotted on the
100K
10K
L
100
Ohms
10
1
C
.1
1
10
100
1K
Hz
10K
FIG. 3.9 Enlargement of impedance-frequency chart.
100K
1M
38
Chapter 3
Vb
R
LINE
Va
AC
C
H - P Filter
Signal Generator
FSLM
FIG. 3.10 AC line impedance measuring setup.
frequency graphs. These DC plots will normally start with readings in the low
milliohm range at the lower frequencies up to around 10 Hz, followed by a slope
of approximately 45 degrees upward (inductive). In some power supplies, the
lower frequencies may have negative resistance readings. These plotted points
will be on both sides of the best straight line, but the best straight (45 degree) line
can be drawn and the inductance read directly off the frequency-impedance graph
(Fig. 3.12).
Z0 =
Vb R
Va
The DC power supply readings will be a smoother line, making it much

easier to read directly off the graph, and flatten out at a much lower frequency.
These points will then level out flat at the upper frequencies. The upper flat area,
in ohms, is then the line output impedance of these upper frequencies. The
operating frequencies of the switchers, clocks, or oscillators, which this power
supply feeds, will fall well into this upper frequency range. This could starve
some circuits being powered by this supply. A line simulation network (Fig. 3.13)
can be developed using the value of the low ohms in series with the inductor read
39
Vb
R
DC
LINE
Va
C
C
Signal Generator
FSLM
FIG. 3.11 DC line impedance measuring setup with equation.
from the graph. Then these two series elements will be in parallel with the higher
upper frequency ohms.
Some power supply designers cure this problem with a capacitor tied
across the output. This capacitor can oscillate with the inductance, read from the
graph, causing ringing at the output of the supply. The best way is to calculate
the maximum resistance that the following circuits can tolerate, say 2 ohms. Draw
the 2-ohm line across the same graph used above, and read the frequency where
the two lines cross. The first line is the 2-ohm line and the second is the inductive
line from the graph, and say this cross point is 1800 Hz. Divide this by 2.5, giving
720 Hz here. Then calculate the capacitor needed to equal 2 ohms at this
frequency. The answer for this problem is 110 F in series with the 2-ohm resistor
tied across the power supply output line. If the curve was plotted again using the
same setup as above, the curve would start in the same way and head up the 45
degrees along the inductive direction to above 720 Hz, then head 45 degree
downward in the capacitive direction and flatten out on the 2-ohm line. A good
source for this is the catalog Power Conversion Design Guide and Catalog, Calex
Manufacturing Co., Pleasant Hill, CA 94532.
40
Chapter 3
100K
L
10K
100
Hfr
10
L
1
C
.1
1
10
100
1K
10K
100K
FIG. 3.12 Impedance-frequency chart with plots.
LINESIM Network
Rs
R hf
FIG. 3.13 Simulation network.
1M
41
The output impedance of these supplies is important in EMI design because

the wrong simulation impedance network could be used. Without the decoupling
capacitor and resistor, the line simulation network looks like a small milliohm
resistor in series with a small inductor. These two components are paralleled with
a resistor value of the upper flattened out resistance. With the decoupling network
used, the output could be replaced with a fixed resistor, as in our case above, using
2 ohms. The best way to measure the line impedance is still with a network
analyzer if you have access to one.
3.6. OTHER SOURCE IMPEDANCES
A lightning strike will generate common mode energy between the line, or lines,
and ground. Assuming this occurrs some distance down the line, the voltage
divided by the current will be the characteristic impedance developed between
the line and ground. This is the square root of the inductance divided by the
capacitance between this line and ground. The inductance is around 1.5 H per
meter, and the capacitance per meter would be less than 1 pF. It should be obvious
that this impedance will be quite high.
4
The Various AC Load Impedances
The load impedance varies with different types of loads. Very few loads are truly
resistive. Most are capacitive or inductive. Some loads generate high-current
pulses at twice the line frequency, whereas others require high-frequency currents.
All these statements, and the rest of the information in this section, assume that
none of these loads have any input filtering or any other circuitry to offset the
conditions mentioned here.
4.1. THE RESISTIVE LOAD
The resistive load is the easiest for the EMI filter to handle. This assumes that the
storage capacitor is left out in the circuit in Fig. 4.1. The only error would be
the crossover error. The diodes are missing in Fig. 4.2, although either circuit can
be with or without the transformer. The power factor of the load is near unity and
the inductors of the EMI filter can be designed to handle the normal RMS current.
This means that the normal design method most magnetic engineers use to design
the inductors will give the desired results. These inductors will not saturate at the
peak current (see Appendix). If the load does use the diodes, this will add severe
noise problems to the already severe switcher noise. The EMI filter must attenuate
the entire load noise spectrum.
4.2. OFF-LINE REGULATOR WITH CAPACITIVE LOAD
The most common circuit used today is the off-line regulator (Fig. 4.3 with or
without the transformer). The output of the diodes feeds a large storage capacitor
44
Chapter 4
FIG. 4.1 Resistive load with diodes.
that in turn feeds a switcher(s). The load impedance also varies from nearly short
to open, depending on whether the rectifier diodes are turned on or off. This
depends on what instantaneous part of the sine wave voltage cycle is being fed
to the diodes. The diodes are turned on if the sine wave voltage, plus the diode
voltage drop plus IR losses, is greater than the capacitor stored voltage at that
instant. High-current pulses charge the storage capacitor during turn-on (Fig. 4.4).
The high-current pulses on the capacitor side of the diodes are often called sine
wave pulses. But the curve shows that the main current pulse is well ahead of the
90-degree point (ahead of 1.57 radians on the plot) and the shape is not sinusoidal.
On the diode side, this is rich in even-order harmonics, whereas the line side is
rich in odd-order harmonics. Figure 4.4 starts at the angle of conduction of the
diodes (in this case 0.8 radian), and a few degrees later shows the angle where
the charge and discharge currents of the storage capacitor are equal, which is the
lowest storage voltage point of the cycle feeding the load. Next follows the peak
current angle that is well ahead of the 1.6-radian grid line. Past the 90-degree
point where the charge and discharge are again equal is the maximum stored
voltage point feeding the load. The cutoff point (in this example, 2 radians)
follows this. The curve repeats in the next half-cycle. Just add 0.8 and 2 to for
the diode side and the second half is negative for the line side. If the load
requirement drops, the start angle increases while the peak load current also
drops. The stop angle stays about the same and may even increase slightly.
The curve was developed using the following values: Rs = 1 ohm, Rl = 22
ohms, C = 0.001 farad, F = 60 Hz, Em is the peak line voltage fed to the diodes,
and N = 1 for full-wave rectification.
FIG. 4.2 Transformer with resistive load.
45
FIG. 4.3 Power supply with capacitive load.
The question at this time may be why this is being discussed here. It is to
show what the true peak current is so that the inductor will not saturate and what
the true RMS current is for sizing the wire gauge. This really is the actual
requirement for the EMI filter. See the Appendix. The RMS current may be 8 A,
while the peak current could be well above 25 A using this circuitry. The EMI
filter must be designed to handle the peak current.
Figure 4.4 was developed by the following method (see Fig. 4.5).
30 30
20
F(X)
10
0 0
0.8
.8
1.2
1.4
X
1.6
FIG. 4.4 Capacitor charging current through the diodes.
1.8
2
2
46
Chapter 4
DIODE
LOAD
C
TRANS
DIODE
FIG. 4.5 The capacitive input filter.
Equations for capacitor rectifier circuits.* In the equations, A is the start

angle of conduction in radians where the diode is just turned on; B is the diode
turn-off angle, also in radians; and N is 1 for full-wave and 2 for half-wave
rectification.
sin(A) = sin(B)e
(B A N)
CRL
(4.1)
where
CRSRL
Y = tan1(CRL) X = tan1
RS + RL
Equation (4.1) assumes the engineer knows the line frequency F, so (2F) can
be calculated, C is the storage capacitor in farads, Rs is the line resistance
(typically 1 ohm), and RL is the load resistance (usually the lowest value of
resistancehighest current). The values of X and Y are then calculated along with
tan(X). Substitute these into the first two equations. In one equation guess a value
for A and solve for B. Insert B into the second equation and solve for A. Average
the two values for A and start the process again by substituting this new value in
the first equation. A spreadsheet is handy for this, and Keith Williams has a basic
computer program for it. Once A, the start angle, and B, the stop angle, are known,
U1, the first minimum voltage angle guess, V1, the first maximum voltage angle
guess, and P1, the first peak current angle guess, can be estimated.
*From Keith L. Williams, Grand Transformers, Grand Haven, MI, from O. H. Schade graphs
developed in 1943.
47
Rs + R1
U1 = sin1
sin(A)
R
1
Rs + R1
sin(B)
V1 = sin1
R
1
P1 =
(4.2)
(A + B)
2
Then U, V, and P can be solved through iteration. Substitute the known and the
estimated U1, V1, and P1 into one term and solve for the other U, V, and P.
Average the two and resolve.
sin(Y + X B) = sin(Y + X A)e
(A B)
tan X
(RS + RL) sin(Y + X A)e

cos(X U) =
sin(Y)
cos(V X) =
(RS + RL) sin(Y + X

sin(Y)
(A U)
tan X
(A V)
A)e tan(X)
(4.3)
(A P)
cos(Y + X P)
= e tan(X)
cos(Y + X A)
Then Eu, the minimum voltage at angle U (in radians), and Ev, the maximum
voltage at angle V (again in radians), can be found.
EU =
EMR1 sin(U)
Rs + R1
EV =
EMR1 sin(V)
Rs + R1
(4.4)
where Em is the peak line input voltage. Also, the peak current at angle P can be
calculated.
IM
sin(P Y X) + sin(Y + X A)e

= EM cos(X)
(RS + RL) cos(Y)
(A P)
tan(X)
(4.5)
Then the equation of the current during conduction is

sin(t Y X) + sin(Y + X A)e
I = EM cos(X)
(RS + RL) cos(Y)
(A t)
tan(X)
(4.6)
Equation (4.6) is what is plotted in Fig. 4.4. Note here that the value of t lies
between A and B. The voltage during conduction is
48
Chapter 4
(A t)
RS sin(Y + X A)e tan(X)

EMRL cos(X)
sin(t X)
(4.7)
E =
RL
cos(Y)
(RS + RL)
However, the requirement for the proper design of the EMI filter is the value of
IM. This is the current the inductor must handle without saturating. Then:
BKG =
1.55LHIM102
NAC
(4.8)
Where IM is the peak current just calculated, LH is the inductance in microhenrys, N is the turns, and AC is the iron area, here in square inches, of the core.
Power factor:
PF =
E1I1 cos()
true power
=
apparent power
VA
E1 V
A =
(IN)2
(4.9)
E 1 I1 cos ( )
I 1 cos ( )
- = --------------------P F = ---------------------------2
2
( IN )
E1 ( IN )
Once the peak current angle [P in Eq. (4.3)] is known from the equations, subtract
P from /2 (90 degrees) to get . The current fundamental frequencys peak must
coincide with the peak draw of IM in Eq. (4.5). This () is the angle of lead. To
finish the power factor, use the RMS value of the fundamental current (I1) and
the summation of the total current squared. This is continued in Sec. 7.5 of
Chapter 7.
These approximate equations still hold true even if a transformer is inserted
ahead of the off-line regulator. This current pulse is not truly a sine wave pulse,
as most people describe it. These equations were developed for several reasons.
The main reason was for the switcher noise to show the relationship between the
total size and weight of the filter with and without an inductor of critical value
(Figs. 4.6 and 4.7). This is discussed further in Chapter 12.
The total size and weight are smaller with the critical inductor for all serious
specifications but might not hold for U.S. Federal Communications Commission
specifications. The second reason is explained in Chapter 16 and the Appendix.
This is also very important when discussing the three-phase filter in Chapter 16
for voltage rise. Finally, these equations were developed so that the full 3rd, 9th,
15th, and so on harmonic currents could be demonstrated for the neutral leg of
the three-phase Y for capacitor design. See Chapter 7.
Stored energy feeds the switcher(s). The RMS value of the current is usually
what the filter designer is given to design the filter inductors. Designers are not
given this high peak current needed to design the inductors. The question is, what
will this high-current pulse do to the filter inductors? Most inductor designers
49
FIG. 4.6 Off-line regulator with the added critical inductance and transformer.
design the inductor somewhere around half-flux density at the RMS value of the
current. This is the wrong approach for the design of this AC EMI filter inductor
because the inductor will saturate. The equation is
H =
0.4NIp
Mpl
(4.10)
where N is the number of turns, Ip is the peak current feeding the charging
capacitor, Mpl is the magnetic path length in centimeters, and H is the magnetizing
force. The maximum flux density, Bm, is a constant for the core material. The
relationship between B and H is the permeability, . Thus, as H increases due to
the large current pulse, Ip, the permeability drops. The permeability () is a key
player in determining the inductance, L. The core material is spoken of as soft
because of the BH curve. These are S shaped, or sigmoid, as in Fig. 4.8 and are
not made using square loop material. These cores require a strong magnetizing
force, H, to drive the core into saturation. A hard core, or square loop core,
is driven quickly into saturation. The soft core is the type of core material
L1
Sw
LOAD
STORAGE
CAP
T1
STORAGE
CAP
DIODES
FIG. 4.7 Off-line regulator without the critical inductance capacitive load.
50
Chapter 4
FIG. 4.8 BH curve of a soft (sigmoid) core.
needed for EMI filter inductors. Cores that have square loop characteristics are
gapped, if they are to be used at all, reducing the hard magnetic core to a soft
core (Fig. 4.9). Soft cores are also often gapped to make them even softer (less
sigmoid) and for DC applications. The equation for the inductance is
L =
0.4N2Ac108
Mpl
(4.11)
where the only new term is the cross-sectional core area, Ac, in square centimeters. So as the value of drops, the value of L drops and so does the inductive
reactance.
What happens if the inductor saturates during this current pulse peak? The
answer is that some switcher, diode, and other noise can ride through the saturated
filter inductor during this time. These noise spikes can thus show up on the peak
of the AC sine wave voltage. This is because these unknown high-current pulses
saturate the inductor. The inductor design specification was underspecified. The
only possible redeeming quality would be achieved if high-quality capacitors
were used, producing a very low impedance to this noise as compared with the
load and line impedance in parallel, but this is not likely at higher frequencies at
the point where they are above their SRFs. This subject is continued in Chapter
11. Another way to avoid this is to design an inductor using a gapped core
51
FIG. 4.9 Toroid showing Wa and Ac.
(Fig. 4.10). This tilts the BH curve, requiring a much higher magnetizing force to
drive the inductor into saturation. It also makes the permeability much lower but
more constant. This technique keeps the inductance more constant for low,
medium, and high current demands throughout the conduction cycle.
The point is that the EMI filter will be bigger, it will weigh more,
and it will cost more when using the off-line regulator because of the highcurrent pulses.
FIG. 4.10 Gapped core. The BH curve is more linear and less sigmoid.
52
Chapter 4
4.3. OFF-LINE REGULATOR WITH INDUCTOR BEFORE

THE CAPACITOR
Adding an inductor ahead of the power supply filter capacitor widens the current
pulse width and lowers the peak current pulse (Fig. 4.11). If the value of the
inductor is equal to or greater than the critical inductance, the current flows all
the time and the current through the inductor is the average current. The normal
inductance design method will work, again, in this application. The critical
inductance is
Lc =
R0
6F
(4.12)
where R0 is the load resistance calculated using the lowest, worst-case, current
and the highest line voltage and F is the line frequency. The disadvantage is that
the stored voltage across the storage capacitor drops from the peak value of the
AC voltage to the average voltage. This also adds to the weight of the power
supply but cuts down on the EMI filter weight because it is not necessary to
handle this high peak pulse current.
Some power supply company designers have been very proud of their light,
compact power supplies. The power supply designers would ask the company to
design a filter for this small power supply, but they would not understand why
the filter is larger than their supply and weighs more. To keep their power supplies
this small and light, they have eliminated all the normal internal EMI protection.
Therefore, the output noise is high and the EMI filter is big! This is covered more
fully in Chapter 11.
4.4. THE POWER FACTOR CORRECTION CIRCUIT
Today in Europe (Canada and the United States are following their lead) power
factor correction circuits are a must (Fig. 4.12). I have seen power factors as low
CRITICAL
L1
L2
Sw
LOAD
STORAGE
CAP
T1
STORAGE
CAP
DIODES
FIG. 4.11 Off-line regulator with critical inductance capacitive load, inductive input
power supply.
53
L1
Sw
LOAD
STORAGE
CAP
T1
DIODES
FIG. 4.12 The power factor correction circuit.
as 0.43 resulting from the power factor of the power supply and the EMI filter
because of larger capacitors. Power supplies that use off-line regulators have
power factors as low as 0.7 by themselves, and improper filtering only adds to
the problem. The power factor correction circuits work by switching the diode
output voltage without initially storing the energy in a large filter capacitor. Thus,
the switcher current is in phase with, or follows, the line voltage. The switcher
creates high-frequency pulses that follow the AC voltage sine wave returning the
power factor back to near unity. This means that the output impedance of the EMI
filter must be very low compared with the conducting load of the power factor
correction circuit switcher frequency. The inductor conducts current in the same
direction when the switch is open and closed, so the current through the diodes
is a sine wave with a small triangular wave superimposed on top of it at the
switcher frequency. It is the job of the EMI filter to attenuate this switcher
frequency without starving the switcher. This is another reason why a filter
designed for very similar specifications may work well for one group and fail for
another. The first group could be using an off-line regulator and the second group
a power factor correction circuit and the output impedance of the filter may be
inductive (maybe a T). The inductive reactance would be low to the harmonics
to the off-line regulator but be high to the power factor correction circuit. This
would starve the power factor correction circuit.
The disadvantage of power factor correction circuits is that they are not
100% efficient. One of the main reasons for demanding power factor correction
circuits was to allow more devices to be plugged into the wall sockets. These
people did not account for the lower efficiencies of these circuits. So, little more
equipment can be added to the wall outlets. The power required to drive the
equipment using the power factor correction circuits is now greater, giving little
gain for the wiring and circuit breaker. The initial efficiencies were around 70%
but are close to 90% today.
54
Chapter 4
There is also a power factor correction coil (Fig. 4.13). In most designs, the
EMI filter looks very capacitive at the power line frequency, and some specifications demand a near-unity power factor for the filter for two reasons: obviously,
for power factor correction, and also for leakage current problems. This technology is archaic, at best, and is mainly seen on 400-Hz power lines.
At the line frequency, the impedance of the inductors in the EMI filter is
very low, so the capacitors add in parallel to a value, in Fig. 4.13, of 2C. The
inductive reactance of the power factor correction coil must be equal to the total
capacitive reactance at the line frequency. This returns the power factor to near
unity and the leakage current is reduced. If C1 is equal to 3 F, then Eq. (3.9) will
yield the following.
L =
1
1
=
= 0.026
4240026 106
42F2C
(4.13)
where C is the total capacitance in farads of the filter to ground. This is a rather
large inductor and the current through it is the same as the leakage current of the
total capacitance (but 180 degrees out of phase). This technique converts leakage
current of the capacitors to circulating current in this newly formed tank circuit.
There are two conflicting factors to consider in dealing with the Q of this tank
circuit. To reduce the leakage current as much as possible, the highest value of Q
is required. On the other hand, due to aging of the filter components and working
stresses on the filter over time, a lower Q is required. This is also true in
installations where the power line frequency drifts over a wide range, such as
remote power generators. This frequency drift would cause the network to be
off-tuned from the center frequency and be operating on the side skirts of the
L1
Ll
Power factor coil

C1
FIG. 4.13 Power factor correction coil.
C1
55
impedance curve of the parallel tank circuit. Whatever the Q, high or low, the
equation is
Icirc = Q Iline
(4.14)
where Icirc is the circulating current in this tank, the same as the leakage current
prior to addition of the power factor correction coil, and Iline is the new leakage
current. The reason for the concern about leakage current is the danger to anyone
touching the filter, or the unit that the filter is mounted in, if the safety ground
has been removed or broken off. The electric shock could be lethal when this
ground is cut. Equation (4.13) shows that the leakage current has been reduced
by a factor of Q.
RFI Corp., on Long Island, tunes the power factor correction coil for each
filter before shipping. This technique allows them to ship these units as matched
pairs with higher Q values. Otherwise, the Q should be limited to about 10. RFI
attaches these power factor correction coils on the load side; others mount these
power factor correction coils at the front end on the line side. I know of no
functional electrical difference.
The specifications are leaning toward measuring the actual ground current
rather than specifying the capacitance to ground. Using this trick would allow
larger capacitors to ground and enhance the common mode action. Using a Q
value above of 10, the 0.02 F, for 400 Hz, could be changed to 0.2 F. This
would reduce the value and size of the common mode inductor, here called the Z
(Zorro). This would enhance the common mode attenuation so that the filter easily
passes the insertion loss requirement while reducing the value of the Zorro
inductor. The only concern here is that the components and the line frequency
must be stable to ensure that the units are not detuned or the leakage current will
rise. This also returns the power factor back to near unity. The power factor
correction coil becomes monstrous in size for 50 and 60 Hz and is rarely used.
4.5. TRANSFORMER LOAD
If the filter designer knows that the load is transformer fed, knowledge of the
transformer is necessary. Try to get the customer or the transformer manufacturer
to tell you the transformer primary inductance. The reason is that the total filter
inductance in series across the filter must be much lower than the primary
inductance. Otherwise, the filter and transformer inductance forms an inductive
voltage divider.
Typically, the primary inductance is well into the millihenrys, possibly 50
mH or more, so this is not a problem, but it should be checked. Where the problem
really shows up is with autotransformers and multiphase transformers that often
employ autotransformers. Autotransformers typically have much lower primary
56
Chapter 4
inductance that often causes this problem. The autotransformer is smaller than the
isolation transformer, and this allows the smaller inductance. The total EMI filter
inductance should be less than 2% of the primary inductance to reduce the voltage
divider effect.
4.6. THE UPS LOAD
Another load to consider is the UPS load. At least in the past, the typical UPS
had zero crossing spikes that challenged the EMI filter. These spikes are very high
frequencies and require quality capacitors of the feed-through type giving very
low impedances to the 10th harmonic of the fundamental spike frequency. These
spikes are typically at 25 kHz and higher, requiring a large capacitor with SRF to
well above 250 kHz. I have known power specialists and consultants who
analyzed power systems and concluded that a UPS was needed, only to find out
that the UPS was creating the problem. Leaded capacitors cannot be used here,
especially for the final capacitor facing the UPS because of the low SRF caused
by the higher equivalent series inductance (ESL) and equivalent series resistance
(ESR) of the leads. Often, a feed-through capacitor of 3.5 F is enhanced with a
leaded capacitor, often referred to as a hang on, to bring up the total capacitance
value. The leads must be very short.
To sum up this section, attempt to find out as much as possible about the
load. Is the load a power supply and, if so, what type? If the power supply has a
power factor correction circuit or an inductive input filter, the EMI filter must
have low output impedance. The same is true if the power supply has a capacitive
filter but the inductors of the EMI filter must be designed to handle high-current
pulses. If the filter feeds a transformer, check the primary inductance and make
sure the total EMI filter inductance is less than 2% of this primary inductance.
5
DC CircuitLoad and Source
There is a difference between AC and DC circuit designs for filters. It is

understood that many off-the-shelf electromagnetic interference (EMI) filter
manufacturers sell filters specified to work for both alternating and direct current.
This adds to the difficulty in design because the sources are very different. Filter
capacitors must be designed to handle the AC voltage and the total harmonic
current. This calls for larger margins at 400 Hz. AC capacitors are typically
designed to handle 4.2 times the RMS working voltage, whereas DC capacitors
must handle 2.5 times the working voltage. Take a line with 250 V specified to
ground. If the system is AC, this equates to 1050 V margin, which means the
initial hipot will be 1800 V. This would increase the margins of the capacitor for
the higher AC voltage. The inductor should be gapped to handle the direct current;
otherwise the inductor will easily be driven into saturation. This dual requirement
adds to the overall cost of the filter. In the preceding example, the DC 250-V line
requires 625 V, calling for an initial hipot of 1000 V.
The filter manufacturer helps to cut the cost of these filters by building
sizable in-house orders for shelf stock. These filters are usually grouped in
families of different current values with the same specifications such as leakage
current and dB attenuation. The same parts are used for all the different current
values when possible, and this drives down the component costs. This makes the
parts less expensive and reduces the cost to the customer. The filter customer can
cut costs by having more than one application requiring the same filter. This
increase in the number of filters required over a period of time cuts the filter price.
This is true for all filter types.
58
Chapter 5
The loads are also different. Some are resistive, as in heaters that maintain
mountaintop repeaters above a certain temperature. Some equipment may use this
supply voltage directly; others may be switchers, for which the inductors should
be gapped for pot cores or C cores, but MPPs and powder cores are already
gapped as a result of the manufacturing process. Gaps for the MPP, Hiflux (HF),
powdered iron, ferrite, and now the new CMI cores have distributed gaps
throughout the core.
This chapter discusses recommendations for elements within the circuit that
reduce the EMI so that the insertion loss requirement of the filter is reduced, reducing
the size, weight, and cost of the filter. Other techniques are also discussed.
5.1. VARIOUS SOURCE IMPEDANCES
The DC power line feed is very short, providing link coupling well up into the
megahertz range. Often, it uses the normal chassis ground, making a balanced
circuit impossible. This is where tubular filters play a role, using feed-through
capacitors without line-to-line capacitors. Some have capacitance only; others
make up the Ls and s and the rest are Ts. This DC power is the type found in
aircraft, shipboard, telephone company and mountain repeater sites. This power
is furnished by various arranged systems such as battery racks, standby generators, solar panels, diesel generators, and wind generators. The battery feeds stored
energy to the various systems and helps to regulate the voltage. These large
batteries are called deep cycle batteries. Deep cycle batteries have more clearance
between the inside battery bottom and the bottom of the battery plates. This
allows a deeper discharge because of the increased clearance room for plate
material and debris. These can feed the entire system from several hours to many
days, depending on the system. They are designed in this way where the
conditions warrant the higher costs or are referred to as life threatening, the
highest cost system requiring more standby batteries, generator fuel storage, and
the like. These types of conditions can be caused by any outside power failures,
such as from a solar panel on a series of cloudy days, downed power lines, fuel
supply outages, and bad weather conditions.
The plates of the battery rack act as a capacitor and shunt the middle
frequencies of the unwanted conducted emission noises to ground, but the high
frequencies are not attenuated because of the inductance of the cable feed and the
battery plates. The radio frequency (RF) current on the power lead is radiated as
an H field, and this is what should be avoided by filtering. The DC output
normally feeds some switchers and these switchers create most of the RF noise,
along with the diodes following the switcher and the parasitic oscillations of the
switcher transformer(s).
The other type of DC system is from an AC power supply, and again the
feed is very short. This DC normally feeds a switcher(s). The difference is that
59
this output impedance is high at the switcher frequency even though the output
impedance is only a few milliohms from DC to 10 Hz or so. The output
impedance is inductive above this point and rolls off flat or resistive at, say,
5 kHz. This power supply looks like a milliohm resistor in series with an
inductance, and this is shunted with the higher resistor value (see Sec. 3.5).
Without being fixed, this situation starves the switcher because the inductive
reactance of the inductor at the switcher frequency, and especially the harmonics,
will be a high impedance equal to Rhf (Fig. 5.1). Is this true for all DC power
supplies? No, in some rare cases the customer informs the designer long before
the design is complete. Such specifications as the power supply must have low
output impedance at a specific high frequency and other conditions that the supply
must provide. The power supply people can make this output impedance low at
switcher frequencies around 100 kHz. Does the filter designer need to know this
output impedance of the DC supply feeding the filter before designing the new
filter for the remote switcher?
What happens if the output impedance of the DC supply is low without the
filter? The switcher will not be starved but the high RF current pulses will radiate,
producing a very high H field. What happens if the output impedance of the DC
supply is high without the filter? It has been stated repeatedly in this book that
the switcher will then be starved but the weak RF current pulses will produce a
very weak H field. The point here is that the filter fixes either condition, so the
filter designer could care less about the output impedance of the power supply.
This is easy to design and is treated in the next section.
5.2. SWITCHER LOAD
Most loads are of the switcher type. The output impedance of the filter at the load
or switcher end must be such as not to starve the load. This assumes that the
switcher does not have a capacitor at the switch to lower this impedance (or at
least not the proper type of capacitor). The capacitors close to the switch are
classed as part of the filter. If this capacitor is in the circuit without the filter
FIG. 5.1 Line simulation circuit.
60
Chapter 5
designers knowledge, it could detune the output network, lowering the cutoff
frequency of this filter. The capacitor values add because they are in parallel. This
assumes that the connecting wire is not a long lead. This would appear as an
inductance tending to split the capacitors.
This DC application would require a single L filter so that the capacitor
facing the load is as high as possible (Fig. 5.2). This is to reduce the drop in the
DC voltage while the switcher is turned on. If double L filters are used, the value
of the final capacitor is smaller than half the original. This could allow the voltage
feeding the load to drop more than for the single L filter. This is still true with the
double L even though the ripple voltage on the feed wire is the same (see
Sec. 19.2). If the return is through the chassis, the capacitor facing the load should
be a feed-through type. Otherwise, it should be a line-to-line capacitor in case the
system has a return power lead. The input inductor of the filter has high
impedance and reduces the H field no matter what the source impedance of the
power supply happens to be. These are easy to design and the method is as
follows. The switcher frequency is 60 kHz, the DC voltage is 60 V, and the peak
on current is 1.2 A. Divide the voltage by the switcher on current to find the on
impedance of the switcher, and divide this on impedance by 10 to minimize the
voltage drop. Figure out the capacitor value by making the capacitive reactance
equal to this impedance at the fundamental switcher frequency, 60 kHz. For the
inductor, multiply the on impedance by 10 and then solve for the inductor
reactance equal to this impedance at, as before, 60 kHz.
60
= 5
1.2 10
C =
1
= 0.53 F
260,000 5
60 10
= 500
1.2
L =
500
= 1.326 MH
260,000
L1
LINE
LOAD
C1
FIG. 5.2 Single unbalanced L filter.
61
This is rather a large inductor but should be easy to design because the main
current is DC. If the duty cycle is 50%, the circuit carries only 0.6 A DC with a
small I AC current. However, if the circuit is balanced, then two inductors of
670 H each are needed and must meet the same current requirement. The small
capacitor of 0.6 F is wired between the two inductors and faces the load.
To calculate the actual inductor current, the pulse width is needed. Every pulse
in this example is 4 s times two for the two pulses per cycle and the 60-kHz
cycle is 16.66 s. So, 1.2 A times 2 times 2 divided by 16.666 is the average DC
current through the inductor(s):
2 IPTon
2 1.2 4
=
= 0.57 A
16.66
Ttotal
The 1.4 MH would have to carry only about .57 A DC, making this possible even
if the filter is unbalanced. This can be made into a simple formula:
C =
10Ip
10 1.2
=
= 5.305 107 = 0.5305 F
260,000 60
2FswVdc
(5.1)
10Vdc
10 60
L =
=
= 1.326 103 = 1.326 MH
260,000 1.2
2FswIp
The equations give inductance in henries and capacitance in farads. The 0.6 F
capacitor may be paralleled with a smaller capacitor (0.01 F) to increase the
high-frequency response.
5.3. DC CIRCUIT EMI SOLUTIONS
OR RECOMMENDATIONS
This is included here because most problem solutions deal with DC power feeds
or are part of the power supply. Some of these may conflict, but use what works.
Most systems grow by these means: The inception of the idea is based on
a customer request, product application, or new technology. The various concepts
are listed, followed by choosing a method. The various design methods and
design problem solutions are selected, and the various circuits or circuit boards
are combined to test the system. This is followed by system corrections and then
the power supply, oops, the input transformer, and double oops, the EMI filter (in
the 2.25 1.6 0.75 inch or any other small leftover volume they could not fit
anything else into). Usually, as described here, the EMI filter is thought of last.
Then people wonder why the filter is so big and bulky and why it cannot fit in
that little leftover volume next to the input transformer. The plan for EMI should
begin almost at the inception of the idea and should be required on each board or
circuit that is applicable. If this is done correctly, the EMI filter might fit the
62
Chapter 5
leftover volume and the cost, along with the weight of the filter, might be
greatly reduced.
5.3.1. Some Ideas for the Initial Power Supply
Include an inductor of critical value in front of the storage capacitor to remove
the high current peaks. Lack of this inductor increases the size of the EMI filter,
it weighs more, and it costs more, especially for the military, whose specifications
are harder to meet. Isolate the off-line regulator to reduce the common mode, and
remember that a capacitor is not a capacitor at all frequencies. Power supply
storage capacitors of larger size are inductive by 20 kHz. Parallel them with a
good-quality extended foil or ceramic capacitor. Use the proper snubbers in the
switchers to remove spikes and stop reverse currents through diodes, for example.
This reduces diode noise. Keep the various supplies isolated, and use twisted pairs
to power remote units rather than a ground return. Filter the isolated leads, or
closely spaced traces on the printed circuit board, at the device end. In this way,
the primary current in the traces or twisted pair is direct current rather than pulses
at the switcher frequency. Lay out the power supply transformer, input, or
switcher(s) for minimum magnetic coupling from the transformer to a susceptible
receptor by increasing the distance between these devices or finding the best
orientation of the transformer.
Sometimes the magnetic field radiating from or around the transformer is
stronger in one direction than anotherthe lines of magnetic force may extend
farther. Go to a toroid transformer, which is known to be quieter magnetically.
Use single-ended converters rather than a flyback transformer, and do not use
switchers known to be noisy, such as SCRs (this should be obvious). Place
sections known to be noisy within a container or shield, which helps to stop the
H and E fields. Cold-rolled steel must be quite thick to reduce the magnetic field
generated by a transformer or any other device. These containers are often silver
plated inside and out to enhance the surface conduction. The better the conduction, the better the H field is attenuated and the lower the current on the outer
skin. This may require a mu metal can or foil which would be thinner and lighter
for the same attenuation of the H field but would cost much more.
Moves to increase the self-resonant frequency of the inductor or transformer
of the switcher also reduce the parasitic or at least raise the frequency of the parasitic.
This parasitic frequency often feeds all the way back to the filter where it is
attenuated. It can be attenuated with a smaller value of capacitance at the interface
of the power supply and switcher. Filter these as close to their source as possible.
5.3.2. Other Parts of the System
Clocks and other sources generate more than their share of noise. Sometimes the
noise does not start or originate in the clock but is created either in the power
63
supply or in other circuits that the power supply feeds. Make sure each feed of
the supply is filtered. Filter each feed such as the clock, DAC, or other device as
close to the device as possible.
Separate ground systems for digital and analog. Separate grounds for the
noisier systems so that the noise current on the ground lead does not induce that
voltage in the quieter circuits and be amplified. Shield the noise generators so that
magnetic radiation of the H field is a minimum.
Keep the noise within the EMI filter inductors and other inductors in nearby
areas, even though the cores are toroids (which are quieter and radiate less), by
mounting these inductors in quadrature to reduce their coupling. This is especially true if the cores are not toroids because they are known to have more
flux outside the core material. This flux can induce currents in other susceptible
nearby devices.
Figure 5.3 shows a balanced double L where the first two inductors feed
the capacitor using the vee technique (see Chapter 7). The last two inductors in
quadrature also use the vee technique. The lower left inductor could be magnetically coupled to the upper right inductor, but the increased distance and the
capacitor between them reduce this.
5.4. LOSSY COMPONENTS
Another technique is to use lossy components. One such technique is from
transmission lines where high frequencies are absorbed within the dielectric.
Coaxial cable developers continue research to find dielectrics with lower and
lower losses, whereas others have used this phenomenon to enhance the losses to
make lossy systems. One such company, in Long Island, New York is Capcon,
Inc. (Ed Reeves). Their material is tubular and can be threaded with the filter
FIG. 5.3 Toroids in quadrature.
64
Chapter 5
hookup wire; it provides substantial losses above 10 MHz and well into the GHz
region. This depends on the total length of the material within the filter. Various
inside diameters are available, allowing easy threading for capacitor, inductor,
and other hookup leads. Capcon material functions better than most ferrite beads
because most ferrite beads saturate around 5 A. These beads have a self-resonant
frequency of about 50 MHz. The lossy component mentioned above does not fail
either of these requirements. One foot of this lossy line material gives as much
as 100 dB at 10 GHz, whereas the ferrite bead gives 20 dB loss at the peak of
50 MHz, and the loss falls to zero by 100 MHz. This material is also available in
sheets with various thicknesses for enclosures and is very dissipative. Radiated
energy can be reduced by covering the filter enclosure, or a noisy device, with
this material.
Figure 5.4 shows the approximate loss of lossy suppressant tubing shielded
(LSTS) and LST, which is the same tubing without shielding, measured per MIL
STD 220A.
5.5. RADIATION EMISSIONS
The subject of this book is the design of the filter for conducted emissions. It is
generally true for radiated emissions that if the conducted emissions are reduced,
FIG. 5.4 Capcon lossy tubing and sheets.
65
the radiated emissions are also often reduced. If the conducted emissions are
nearly eliminated, there is often little left to radiate. The opposite condition is
even truer: if the conducted emissions are not eliminated, the device radiates.
A condition that makes this statement erroneous is that the outer skin of the
device is not a good conductorplastic in the case discussed. Although the conducted emissions were greatly reduced, the radiated emission was still almost as
offensive. If the case is to be plastic, there are conductive sprays that the enclosure
can be coated with that help to reduce the emissions.
6
Typical EMI FiltersPros and Cons
Let us discuss the typical EMI filters that are used today. These are primarily the
, T, and L, with Cauer, RC shunt, and dissipative filters sometimes included.
They include double and triple filters and sometimes even quadruple uses of the
mentioned types of filters. Each EMI filter has some positive and negative
attributes. Each has its best spot where it functions very well and other spots
where it fails to the first degree. This chapter should clarify all of this. In addition,
the transformer, if used, will add to the filter loss.
6.1. THE FILTER

We start with a discussion of the pi filter (Fig. 6.1). This filter looks very good
under the 220A 50-ohm test specification. The line-side capacitor will work into
the 50-ohm line impedance at the low frequencies. If this is the only test
requirement specified by the customer, then the will test well with this test
technique. The filter will be passed with flying colors by the source inspector
(if any). This is especially true for the three-phase type, for which the specification states Measure one phase with the other two phases tied to ground.
The input and output capacitors of each pi section are then doubled in value.
This makes it easier for the filter designer to meet the specified loss within a given
weight and volume. It means that the filter can easily pass the attenuation
requirement with smaller values, package size, and weight.
Note that the center capacitor is twice the size of the end capacitors.
All inboard capacitors are twice either of the two end capacitors for all multiple
filters (Fig. 6.2). Figure 6.3 shows the 220A test setup.
68
Chapter 6
L1
LINE
LOAD
.5C1
.5C1
FIG. 6.1 The filter.
The peak current is supplied to the filter through buffers or extended

buffers. These are large values of inductance to keep the EMI test signal out of
the DC power supply. The DC current supplied through the buffers is the peak
AC current the filter must handle, not RMS. This is done to ensure that the filter
inductors do not saturate at the full peak current. Two sets of 50-ohm matching
pads from 6 to 10 dB each are omitted in Fig. 6.3.
On the other hand, this type of filter looks bad under the 461 or naval tests
(Fig. 6.4). This is where a 10-F capacitor shunted from line to line or to ground
is required. The reason is that a filter capacitor facing the 10 F is out of the
L1
L1
LINE
LOAD
.5C1
FIG. 6.2 Multiple .
C1
.5C1
DC
69
Buffers
Signal
R1
R1
L1
50 Ohms
Filter under test
50 Ohms
C1
C1
Reduces to:
Signal
R1
L1
50 Ohms
Filter under test
50 Ohms
C1
C1
R1
FIG. 6.3 The 220A test setup.
circuit unless the cap is an order of magnitude greater than the 10-F capacitor.
This value would be much too high for good filter design. Large capacitors have
very low self-resonant frequency (SRF) and would have to be shunted with
smaller capacitors of good quality to make up for this low SRF. The receiver or
spectrum analyzer is switched back and forth between the two current probes to
determine the filter loss.
70
Chapter 6
Current Probe
Current Probe
POWER
10 F
L1
Filter under test
C1
C2

Reduces to:
Current Probe
Current Probe
L1
Filter under test
C1
C2

FIG. 6.4 The CIP test setup.
The three-element looks like a two-element L filter at the lower frequencies in the 461 and naval tests. The loss is about 6 dB per octave for each element,
so the 18-dB filter is now a 12-dB loss per octave filter; 18 dB is for the single
and the double is 30 dB, for example.
The might also function well in some DC systems if the switcher
frequency is high enough so that the capacitor impedance facing the load is
small enough not to starve the switchers. Also, the capacitor impedance should
not cause excessive voltage drop. This assumes that the switcher circuit has
not handled this problem. The filter is easily balanced by placing only half
of the inductor needed in the high line and the other half in the neutral line
(Fig. 6.5). This changes the shape from the to a square or box shape. It
does happen that the system must pass a specification after several prototypes
are finished.
A qual test may have to be passed (Fig. 6.6). Now the earlier , which
passed all those costly tests, may not perform properly. This is because the filter
loses the front, or line-side, capacitor at the lower frequencies. Some will wonder
71
HOT
.5L1
.5C1
LINE
.5L1
C1
.5L1
.5C1
LOAD
.5L1
NEUT
GRND
FIG. 6.5 Multiple balanced .
why the filter is not doing the job when it properly passed the tests at the EMI
test laboratory! The real world is not reflected by the 220A or 461 specifications,
but the 461 specification is closer to reality.
6.2. THE T FILTER
The T filter gives 18 dB per octave, and the double T gives 30 dB per octave
(6 dB per element) (Fig. 6.7). T filters work best in low-impedance lines (high
current requiring small inductors). The line impedance is very low up to at least
100 kHz, but the 461 loss specifications start at either 10 or 14 kHz. The inductive
input impedance of the T adds to the low line impedance. This gives the capacitor
an input impedance to work into. These are also best in the higher current loads
POWER
LISN
L1
FILTER
UNDER
TEST
C1
TRANS1
L1
DIODES
SPECTRUM ANALYZER
FIG. 6.6 System or diode test method.
72
Chapter 6
.5L1
.5L1
LINE
LOAD
C1
.5L1
L1
.5L1
LINE
LOAD
C1
C1
FIG. 6.7 The T and multiple T drawings.
if the design method does not call for too high values of these T inductors. This
could result in the voltages soaring or dropping, feeding the load. I have seen
115-V AC 60-Hz lines as high as 132 V feeding a light load requirement. This
happened because the resonant rise occurred at a very low frequency. This is
usually caused by higher inductive values.
The T should never be used in the DC system if the load utilizes any
switchers because the high impedance of the output inductor facing the load will
starve the switchers. The switcher designer may have taken this into account by
lowering the impedance with a capacitor at the switch input. This really makes
the filter into a maladjusted double L because the capacitor shunts the inductor
facing the load, but this is less troublesome in DC than AC.
Note here that the central inductor is twice the size of either of the two end
inductors. All inboard inductors are twice the size of either of the end inductors.
The T filter can be balanced by removing half of the inductors values and placing
this half in the neutral leg, forming an H pad (Fig. 6.8).
.25 L1
73
.5L1
C1
LINE
.25L1
.25L1
C1
.5L1
LOAD
.25L
FIG. 6.8 Multiple balanced T filter.
6.3. THE L FILTER

The L type is the filter most often used. The and the T are three element, or
more if multiples are used, and should give 18 dB of loss per octave. The L, being
only two elements, provides 12 dB of loss per octave. All of these loss figures
refer to the loss starting above the cutoff frequency. A single L filter works best
in the DC mode if the load has switchers. The inductor faces the DC source and
the capacitor (of high quality with a high SRF) would provide a low impedance
for the switcher frequency (Fig. 6.9). Why not a double L? The two inductors
required for the same amount of loss would total less than the single inductor, and
the same is true for the capacitor values. See Sec. 23.2. This smaller output
capacitor, smaller than 0.5 of the original, may not furnish the needed energy
storage, creating a larger peak-to-peak voltage drop feeding the switcher. This
peak-to-peak voltage would be at the switcher frequency or twice the switcher
frequency if this feeds the center tap of a class B amplifier or Royer. The double
L can be used as long as the drop is not excessive or the switcher frequency is
SINGLE "L" FILTER
L1
LINE
LOAD
C1
FIG. 6.9 Single L filter.
74
Chapter 6
high enough. This statement is true even though the attenuation is improved for
the DC source or power supply voltage. The L and multiple L work well in higher
power applications (Fig. 6.10). Again, to balance, split the inductors and put the
other half in the neutral (Fig. 6.11). The double L has 24 dB loss per octave.
6.4. THE TYPICAL COMMERCIAL FILTER
This is the type used in test equipment, computers, and other commercial
electronic equipment. Here the manufacturer has to pass tests for UL, TUV, VDE,
CSA, and the FCC. The tests are conducted by EMI test houses that help the
manufacturer with all the documentation needed for the various agencies. These
filters are the balanced type and are often purchased from outside suppliers and
often built offshore. The filters are mainly common mode in appearance, with a
capacitor across the input and output from hot to neutral and two other capacitors
to ground that must meet the leakage current specifications for whatever agency
has the toughest requirement. The leakage inductance is often made high by
adding washers to the center of the pot cores separating the two windings. This
is also accomplished by winding the two windings as far apart as possible on a
ferrite toroid core. The feed-through capacitors are grounded directly to the case
(Fig. 6.12). See Chapter 7. Some of these techniques add differential mode to this
common mode filter by increasing the leakage inductance or adding inductors to
both lines. This then makes a balanced type with both differential and common
Mode. This all works because the losses specified for the FCC start at 450 kHz.
The inductors and capacitors can be quite small to accomplish these tasks.
Another point to remember is that the current through the leakage inductance
DOUBLE "L" FILTER
L1
L1
LINE
LOAD
C1
FIG. 6.10 Multiple L filter.
C1
75
.5L1
.5L1
R2
LINE
C1
RC SHUNT
C1
LOAD
C2
.5L1
.5L1
FIG. 6.11 Balanced multiple L with RC shunt filter.
cannot saturate this inductor because most of this is not through the core but in
the surrounding air. This is because the common mode inductor is wound on
ferrite cores with high Al values. Some of these filters do not use the feed-through
type of capacitor so the circuit changes to that in Fig. 6.13. These capacitors
are less expensive but the self-resonant frequency is lowered by the added
lead length.
6.5. THE DISSIPATIVE FILTER
This filter is rarely seen in the EMI arena today. It consists of one inductor and
one capacitor along with two resistors (Fig. 6.14). The two resistors are tied in
.02 F
CF
LINE
C1
C1
LOAD
CF
FIG. 6.12 Commercial filter with Z and feed-through capacitors.
76
Chapter 6
C2
LINE
C1
C1
LOAD
C2
FIG. 6.13 Commercial filter with Z and leaded capacitors.
series across the inductor terminals, and the capacitor is tied to the center of the
resistors and then either to the other line or to ground. These filters are similar to
the line impedance stability networks (LISNs). This filter can be balanced by
using half the inductor and resistor values to be split on both legs and the
capacitor tied between the center points of the four resistors (Fig. 6.15).
The main disadvantages are that this filter appears to give 12 dB per octave
but really gives only 6 dB. Another is that this adds cost and volume to the entire
filter for only 6 dB loss for this section. The resistors should be noninductive and
the resistor(s) facing the line should be at least 2 watts for electromagnetic pulse
(EMP) purposes.
The advantages are as follows:
1. Using two of these at each end can match this filter to line and load
L1
R1
C1
FIG. 6.14 Dissipative filter.
R2
77
.5L1
.5R1
.5R1
C1
.5R2
.5R2
FIG. 6.15 Balanced dissipative filter.
impedance (Fig. 6.16). Any filter can be used between these two end
units and designed to match the impedance of the inbound resistor
values. This can work well to meet the 220A specification using four
50-ohm resistors and designing the middle unit for 12 dB less loss.
2. This filter dissipates the stop band energy. If any higher frequency
comes down the line, most of the energy will be dissipated by the
line-side resistor. The load-side resistor will dissipate most of the
energy of the switcher or any other noise from the load. If this is the
type described, the energy not dissipated would be reflected by the
center filter and the opposite resistor of the same dissipative filter and
most of this energy would be dissipated.
3. A filter would have a filter very similar to an inexpensive LISN would
be included within the filter, and two if the unit is the type described
and pictured in Fig. 6.16.
6.6. THE CAUER FILTER
The Cauer, or elliptic, filter is best used in very low impedance circuits
(Fig. 6.17). These filters are usually used with multiple Ls and Ts. In any case, a
capacitor is normally shunted across the center inductor. This is used to fix a
problem frequency such as 14 kHz. The network is tuned to slightly above the
problem frequency. Granted, the trouble may be fixed, but this center section will
not be in the circuit above this problem frequency. The network will pass all the
upper frequencies. Often, a resistor is placed in series with this capacitor and
78
DIN & DOUT WITH ANY FILTER

L1
L2
ANY FILTER
R1
FIG. 6.16
R2
C1
R3
C2
R4
The DIN and DOUT dissipative filter for both ends.
Chapter 6
R = Rd ?
BALANCED CAUER
R = Rd ?
LINE
CAUER
2C
.5L
LOAD
LINE
LOAD
R = .5Rd?
2C
.5L
FIG. 6.17
Cauer and balanced Cauer.
79
80
Chapter 6
the resistor limits the amount of bypass. The value of the resistors is typically
around 10 ohms and is often the value of the design impedance of the filter.
The design impedance is the lowest RMS line voltage divided by the highest
RMS line current.
If this filter is to be balanced, use half the calculated value of inductance
on the two lines, and parallel each inductor with two capacitor-resistor networks.
The value of the two capacitor networks uses twice the capacitance and half the
resistor values, and these would, again, be tied across the two inductors.
6.7. THE RC SHUNT
Another technique is preferred to the Cauer but is better when used in highimpedance, low-current circuits. This filter, called the RC shunt, uses fewer
components and is automatically balanced across the line to start with, if required
(Fig. 6.18). This is formed with a capacitor and a series resistor. Normally, the
filter has a resonant rise lower in frequency than the trouble frequency. This is
especially true if the filter is a multiple filter such as a double or triple L, , or T.
Usually, the number of resonant rises is one less than the multiple number,
meaning that the single L, pi, or T would not have any resonant rise but the quad
would have one less, equaling three. This holds true only if the circuit Q is low
enough. The higher Q has a resonant rise for each network. Find the frequency
of the lowest resonant rise and pick the capacitor value at this frequency that
equals the filter design impedance. This will attenuate each resonant rise above
the first and also the trouble frequency. If the resonant rise frequencies are of no
concernwell above the fifth harmonic of the power line frequency and well
C
R = Rd ?
FIG. 6.18 Balanced and unbalanced RC shunt.
81
below 10 kHzchoose the capacitor to equal the design impedance at the trouble
frequency.
As an example, the design impedance is calculated by dividing the highest
current required by the load into the lowest anticipated line voltage. In this case,
say, 100 V is the lowest line voltage at 10 A. This is the highest current at this
lowest line voltage. The resonant rise frequency is 14 kHz. Then,
Rd =
100
= 10 ohms
10
C =
1
= 3.979 F
24000 10
(6.1)
The capacitor is 3.979 F or 4 F in series with the 10-ohm resistor tied across
the line. This should remove the bump at 4 kHz and attenuate a problem
frequency at 14 kHz. If the resonant rise is in an area of little concern, change the
frequency to the problem frequency of 14 kHz. Then recalculate the value of
capacitor needed at that frequency.
C =
1
= 1.137 F
2 14,000 10
(6.2)
The 10-ohm resistor is in series with a 1.2-F capacitor tied across the filter. This
lends itself to multiple L filters where the preceding network can be tied across
any one of the capacitors. The closer it is to the load, the more it tends to minimize
or reduce the impedance swings of the load.
6.8. THE CONVENTIONAL FILTERS
The filter houses rarely use conventional wave filters. Some gurus do use these
filter types, but they require constant input and output impedance to work
properly, especially for the low-frequency losses required by the military. The line
and load impedances in EMI just do not provide this condition at these low
frequencies. However, a filter would test well in the 220A test method if the filter
was designed for 50 ohms. These filters would work very well between the two
dissipative filters discussed in Sec. 6.5 (Fig. 6.19) because the noninductive
resistors would give the proper match needed by these filters. This is especially
true if the cutoff frequency of the dissipative filter is about half that of the
regular filter.
6.9. MATRIXTEST SPECIFICATION AND THE
FILTER TO USE
This section covers the filter matrix to use for both the line and load conditions.
This is mainly empirical, and these should be used as general guidelines. The first
82
L1
R1
FIG. 6.19
L2
R2
C1
R3
C2
R4
Conventional filter used with DINs and DOUTs.
Chapter 6
83
TABLE 6.1 The Line-Side Matrix

Single phase
Three phase,
1 Can
Three phase,
Inserts
Single L,
L, , or T
L, , or T
L, , or T?
Single L, L, T?
Single L or L
L, , or T
L or T?
L or T?
L, , or T
L or T
L or T
NA
NA
DC to DC
FCC VDE
TUV
220A
461
System qual
is the line side covering DC to DC, single phase, and three phase (Table 6.1).
The other dimension is FCC, VDE, 220A, 461 and the full system specification.
The next part of the matrix is for the load side (Table 6.2). This is for 60 and
400 Hz. The other row is the off-line regulator, with inductor, the power factor correction circuit, and the power line filter. These are general recommendations only.
TABLE 6.2 The Load-Side Matrix

Source
DC
Off-line regulator,
capacitive load
NA
Off-line regulator,
inductive load
NA
PFCC
NA
Power line filter
NA
DC to DC
Filter output
impedance
must be low at
switcher F
NA
Transformer
50/60 Hz
400 Hz
Inductors must
handle the
peak current
Inductor must
handle only
RMS current
Filter output
impedance
must be low at
switcher F
Inductor must
handle the
peak current
NA
Inductor must
handle the
peak current
Inductor must
handle only
RMS current
Filter output
impedance
must be low at
switcher F
Inductor must
handle the
peak current
NA
Watch primary
inductance
Watch primary
inductance
7
Filter ComponentsThe Capacitor
The differential mode components must be of the high-Q type. The individual
component Q must be high while the circuit Q must be lowered to the point where
the filter does not oscillate. Additional circuits are often added to lower the Q
below 2.
7.1. CAPACITOR SPECIFICATIONS
The main specification for the capacitors is MIL STD 15573. The capacitor must
meet various voltage ratings. For AC capacitors, the level must be 4.2 times the
RMS voltage of the system. For example, in a 220-V RMS system, the capacitor
must be designed to handle 924 volts, usually rounded up to 1000 V. For the DC
capacitor, the multiplier is 2.5 times the system voltage. For example, a 50-V DC
capacitor must be designed to handle 125 V DC, probably made up to 150 V.
There may be specifications on certain creepage distances and corona specifications if the voltage is high enough. In the build to print type, the capacitor values
may be specified as a minimum or something like 10%. Regardless of how well
the filter satisfies the insertion loss, the filter will be returned as being out of
specification if these values are not adhered to. In most filters, if the capacitor
value is reasonably higher, the filter will work better giving more insertion loss,
but that will not suffice if the value is limited.
86
Chapter 7
7.2. CAPACITOR CONSTRUCTION AND

SELF-RESONANCE FREQUENCY
Figures 7.1 and 7.2 show the self-resonant frequency (SRF) of polyester (Mylar)
and ceramic capacitors. Robert Hassett (now retired) of RFI Corp., Long Island,
New York, provided both drawings. Designs are covered in a later section. Both
are feed-through type capacitors giving higher SRF values. Figure 7.3 shows a
method used to raise the SRF of either the wrap and fill type or any other style
of capacitor. This is called veeing the cap.
The advantages of the feed-through type are the low equivalent series
resistance (ESR) and equivalent series inductance (ESL) due to short lead lengths.
This, in turn, means a much higher SRF for the capacitor. The graphs show SRFs
up to 1 GHz on the smaller values of ceramic caps. The problem is that the higher
the power line RMS voltage becomes, the bigger the margin must be to eliminate
creepage and corona. This requires a bigger capacitor and increases the cost and
the container cost. As the line frequency increases, the line harmonic current
content also increases. This also increases the loss due to dv/dt. If the margins
were near the lower limits initially, now that the line frequency has increased, the
initial margin may have to be increased. In other words, the margin size is not a
function of frequency unless the capacitor is close to the margin limit.
Some filter companies use the large nonpolarized can-type filter capacitors.
Some of these are oil impregnated. These may be very good capacitors for power
supplies and other applications requiring nonpolarized capacitors, but watch out
for their self-resonant frequencies. Most of these types have a very low SRF, of
FIG. 7.1 Self-resonant frequency chart for Mylar.
87
the order of 50 kHz or less. This is where the old power supply trick may come
in handythe Ye olde paralleling capacitor trick.
This calls for paralleling the capacitor with another capacitor of 0.05 the
value of the original. This smaller capacitor would have a much higher SRF and
lower ESR and ESL. An article in the IEEE magnetic manuals showed that only
6 dB of gain is realized by these two capacitors in parallel. This theory, though,
assumed that the lead length would be close to the same. This would almost make
the second capacitor have an ESR and ESL the same in proportion to the original.
A feed-through type for the second capacitor would guarantee a workable system
because of the very low ESR and ESL and much higher SRF of this high-quality
type of feed-through capacitor. Experience shows that paralleling the capacitors
often gives serious peaks because of the feed-through oscillating with the ESL of
the original capacitor. The purpose of this nonpolarized style of capacitor is to
handle the low-frequency requirement of the EMI filter and it should not be
required over their SRF point. They often cost less for the capacity and working
value; 10, 15, 20 and 30 F at 480 V AC are available. Another method for
lowering the ESL and raising the SRF of the capacitor is as follows.
7.3. VEEING THE CAPACITOR
In Fig. 7.3, there are four inductors that are often deemed insignificant. This is
where the three Ls join at the top and represents some tie point or splice, with a
FIG. 7.2 Ceramic self-resonant frequency chart.
88
Chapter 7
L1
L1
L
C1
L
FIG. 7.3 The T filter used to demonstrate veeing.
fourth at the bottom. The two that make up the capacitor leadsthe two top and
bottomadd to the ESL and ESR, lowering the SRF of this capacitor. The two
leads next to the inductors add to the inductor value. This increases both
inductors, but they are orders of magnitude lower in value. This is similar to the
concept of keeping the lead lengths as short as possible.
Alhough this technique has been around a long time, the new name for it
is veeing the capacitor. This is the old keep the leads as short as possible trick,
especially on the capacitor side. The leads on the inductor just add a small amount
of inductance to the two inductors. However, the leads facing the capacitor, the
vertical leads in Fig. 7.3, increase the ESR and ESL of the capacitor, lowering
the SRF.
Now, in Fig. 7.4, all the inductance in either leg of the vee adds to the
inductors on both sides. The common tie point, or splice, is eliminated and the
self-leads of the inductor are wired directly to the capacitor as close to the
capacitor body as possible. The large nonpolarized capacitors discussed above,
such as 20 F at 480 V AC, have terminals on the top for wiring directly to. The
capacitor ESL and ESR decrease, increasing the SRF. The same can be done to
the bottom half if the circuit is balanced (Fig. 7.5).
L1
89
L1
C1
L
FIG. 7.4 Veeing the capacitor.
7.4. MARGIN, CREEPAGE, AND CORONASPLIT FOIL

FOR HIGH VOLTAGE
First, margin, creepage, and corona must be understood. The margin requires a
space calculated using 16 V per mil. If the working voltage of an AC capacitor
is 220 V AC, the quality multiplier for most military specifications is 4.2 for AC;
220 4.2 = 924. Divide this by 16 = 57.75 mils or 0.05775 minimum margin.
Use 3/32 of an inch. See Fig. 7.6. The two plates, or foils, extending to the left
will be swedged together. The margin places a gap between the plates. If the
margins are adequate, everything is okay.
If the capacitor is used at a voltage over its design rating, the voltage will
creep over the dielectric slowly, carbonizing a path as it goes until it arcs to the
swedged plates, ruining the capacitor. This will take some time to happen, and
the carbonized path will grow slowly over time.
Corona, on the other hand, is high voltage related. To avoid corona, smooth
surfaces are required, not sharp points. Capacitor winders require splitters, which
spilt dielectric material to the proper width required for the capacitor. In this way,
they buy only the wider widths and split the dielectric material for the desired
width. The point here is that in the process of splitting, if a point is formed in a
high-voltage section, corona will develop. One way to end this is to fold over the
edge on the margin side forming a rounded smooth end facing the margin. Using
90
Chapter 7
L1
L1
C1
L1
L1
FIG. 7.5 Balanced vee.
this technique, the capacitor becomes thicker with an increased diameter. But this
does eliminate all sharp points on the foil. See Fig. 7.7. In high-voltage capacitors,
the plates, or foils, are wound in series to divide the voltage.
The voltage in Fig. 7.8 divides across the sections, but the active area is
reduced. With two sections shown, the voltage is divided by 2, but the capacitor
value must be multiplied by 2. The gap in the middle to separate the two foils
must be greater than the value used for the margin. Use 32 mils per volt to
calculate this distance. The right and left ends are swedged together. There is no
connection to the center foil or plate. These types of capacitors are large and there
may be several in series with more than one split or gap.
FIG. 7.6 Capacitor plates and margin. T is the dielectric thickness in mils, and t
is the thickness of two dieletrics and two foil plate thicknesses in inches.
91
FIG. 7.7 Capacitor with rounded foil.
7.5. CAPACITOR DESIGN

7.5.1. Wrap and Fill Type
Three methods are used to build these capacitors and two subgroups. The initial
method was to attach leads to the foil at the start before wrapping, or winding,
the capacitor. These capacitors are now called inductive capacitors or chicklets
because of the current flow through the foil to charge or discharge the entire
length of the capacitor. The ESR and ESL are very high in these chicklets, so the
SRF is quite low. As an example, for a foil 0.00023 inch thick and 0.5 inch wide,
the cross-sectional area in square inches is 0.000115. Dividing this by 7.854
107 gives 146.42 circular mils. Dividing 17 by 146.42 gives 0.1161 ohm per foot.
If the 0.23 mil aluminum foil is 100 feet longan average size capacitorthe
electron flow average length is half this, but there are two plates. The resistance
is 11.61 ohms. The inductance would be of the order of 45 H based on
approximately 30 meters.
These should never be used for the main EMI filter capacitors. They can be
used for RC shunts if the frequency is low enough. They can also be used for the
Cauer types for low-frequency tuning capacitors, again depending on the frequency. The tuning frequency must be at least half the chicklets SRF. The
chicklets should not be used for the dissipative filter. If the filter has highfrequency problems that the RC shunt must also aid, the quality of the capacitor
dictates the higher quality of the extended foil type.
The two subgroups are metallized and foil. For the metallized style, the
dielectric is coated with a thin spray of metallized aluminum that becomes the
plate of the capacitor, and the other is aluminum foil. The typical aluminum foil
thickness is 0.00023 inch, or 0.23 mil. This foil is much thicker than the film,
FIG. 7.8 Plates in series, high-voltage capacitor.
92
Chapter 7
which is measured in microns. Foil will carry much more current and is therefore
better for pulse applications and EMI filters where there are high harmonic
currents from off-line regulators and similar harmonic generators. Most EMI
filters are built with this type of construction and have a thicker aluminum foil if
higher currents are expected. The metallized film has several advantages, however: this capacitor can be much smaller for the same capacitor value, and this
type is self-healing. All dielectrics have small pinholes throughout their length.
When the applied voltage stresses the film, the film often shorts out through one
of the pinholes, causing the film to melt. The aluminum will reform, making the
capacitor self-healing. Another advantage of this subgroup is that the metallized
aluminum can be sprayed on both sides of the dielectric. This adds to the smaller
size, promotes self-healing, and is better for extended life. To avoid the pinhole
problem, several dielectrics are used. Typically, the thickness is 0.24 mil each,
giving a total dielectric of 0.48 mil. The odds of two holes occurring together are
remote. Film capacitors can be used for DC filters, but the capacitor facing the
load must be foil.
The type used mostly in the EMI filter is the extended foil type. The foil
extends beyond the winding arbor so that one plate, or foil, extends to the left and
the opposite plate extends to the right. The area on the right of the first foil, or
the left of the second foil, makes up the margins for both plate ends. It should be
obvious why this type is called the extended foil type. The extension is typically
3/32 and each end is soldered (called swedging) for the contacts. The ESR and
ESL of extended foil capacitors are both very low because the current flow travels
only the average width of the capacitor foils, called the gauge, or height, of the
capacitor. Also, the turns are in parallel, so the average diameter times the number
of turns times the thickness gives the square inches of the aluminum. Convert this
to circular mils (divide by 7.854 107), and divide this into 17 for aluminum
(10.374 for copper) to obtain the ohms per foot. Dividing this by 12 and
multiplying by the gauge (the height of the capacitor) give the approximate
resistance of the capacitor. This would be a close approximation of the ESR.
Inductance is approximately 1.5 H per meter. Divide 1.5 by 39.36 to get 38.1
nH per inch. This times the gauge gives a first-order approximation of the ESL.
It should be apparent why a large capacitor diameter to length, or gauge, is
desirable. A 2:1 ratio of the diameter to the height, or gauge, is the optimum ratio.
If the ratio increases too much over this optimum, the capacitor starts to wobble
on the winding arbor and the active plate area is diminished. The capacitance
value drops but the gauge is also more than desired. The SRF is very high for this
optimum type and should be used by the EMI filter designer.
The last method is similar to the extended foil type but is used for high
capacitor current applications. This is called the tab type. The size and number of
tabs depend on the current. These tabs are thin strips of conductor that are placed
in the winding as the capacitor is being wound. These tabs are insertedone for
each plateevery so many turns and so many degrees are added to each tab so
93
that the tabs end up uniformly spaced around the sides of the capacitor. The tabs
extending out of each side of the capacitor are folded over and soldered together
to form the contact.
Most new people in the EMI design arena are knowledgeable about power
supply design, where the main concerns about the capacitors are their working
voltage and capacitor value. Others are familiar with derating of the capacitors,
and this quite helpful. The point is that AC capacitors must be designed to handle
the total AC currents at the line frequency and the odd harmonics. If this is a
single L or T, the capacitor should be of the foil type, not metallized film. In
and multiple filters, at least the last capacitor, the capacitor closest to the load
side, should be of the foil type. See the next section.
Some capacitors are not designed properly to handle the full AC current
flow or were designed for DC applications. This is especially true if the capacitors
are the metallized film type. The capacitor designer may not have designed the
capacitor to handle the total harmonic current. Capacitors must handle the
harmonic current from either the line frequency side or the load. This is especially
true for the harmonic current created by the off-line regulator or any power supply
using a capacitor input filter. The foil making up the capacitor plates can be too
thin, measured in microns, to handle this current. This raises the ESR losses, so
the capacitor will heat and will fail in the months ahead. Many factories take the
defunct unit apart and replace the blown capacitors with the same capacitor,
stating, Weve always used this capacitor in these units. They are not aware that
the capacitor has the wrong rating for this application. Commonly, the capacitor
was designed for DC operation and not for AC by the capacitor manufacturer and
was selected in error by the original filter designer. The filter capacitor must be
selected to handle the harmonic currents of the off-line regulator and any other
pulse type with high harmonic currents (Figs. 7.9 and 7.10).
If the line harmonic current is neglected, the line voltage is approximately
sinusoidal and the load current equation, per Sec. 4.2, is
A t
sin(t Y X) + sin(Y + X A)e tan(X)

I = EM cos(X)
(RS + RL) cos(Y)
(7.1)
See Sec. 4.2 for the method of calculating the various constants. Even with all
the constants inserted in the equation, further analysis is difficult. There is an easy
method from days gone by that is not even taught now. This method is included
on the disk provided with this book.
167 cos(0.346087) sin ( t 2.03688 ) + sin ( 1.235 )e
t
0.801779
---------------------------------
0.361
-------------------------------------------------------------------------------------------------------------------------------------------------------------------23 cos ( 1.69079 )
(7.2)
94
Chapter 7
Rs
I(t)
L1
C1
FIG. 7.9 The line voltage, Es, the source impedance, Rs, filter, and I(t).
Further reductions give
57.059 sin ( t 2.03688 ) + 0.944e
0.801779 t-
--------------------------------
0.361
(7.3)
However, this is still no fun even with Fourier analysis. Here, I have left the work
to a computer spreadsheet. With the speed of computers today, this is an easy task.
I put all the constants at the top of the spreadsheet. In column A, I listed the angles
from 0 to 2 in 360 steps. In column B, I listed the equation using IF statements
{IF (angle<A, 0, If (angle >B, O, Equation}. Here, the start angle A and stop angle
I(t)
L1
C1
FIG. 7.10 Total harmonic current without the source resistance, Rs.
95
B must be known. The rest of the columns are An values followed by the Bn values
and the calculation of Cn. See the Appendix and disk. The value of each harmonic
and phase angle is given. The spacing can be changed so that only odds or evens
or both odd and even can be looked at. Starting with a step of one, both the odd
and even appear. If the evens are extremely small, eliminate them by changing
the step from one to two (or vice versa). The fundamental current phase angle
and the true to apparent current ratios and hence the power factor are also given.
Two different spreadsheets are givenone for the line side, with all odds, and
one for the storage capacitor side, with all evens. These Lotus programs can be
used for any other wave if an equation can be developed.
The source resistance is low over the range of harmonics of interest and can
be eliminated. The inductors and capacitors making up the filter can be lumped
or added because the frequency of interest is well below the cutoff frequency of
the filter for the most part. But, as the harmonic number increases, the capacitive
reactance decreases and the capacitor current increases.
If this is a double L, the two inductors can be added (2L) and so can the
capacitors (2C). (if they are the samethey should be).
The harmonic current for the line sideodd harmonicsfirst through 25
are as follows: 11.97, 8.56, 4.39, 1.06, 1.17, 0.86, 0.30, 0.5, 0.28, 0.2, 0.26, 0.11,
1.6. The square root of the sum of the squares is 15.65 and the RMS line current
is 11.06. The difference is 4.59 or 4.6 A through the capacitors. An extended foil
should be able to handle this in most cases, but can a metallized one? The
metallized capacitor would depend on the size but probably not hold up in the
long run. As for the chicklet capacitors, they would not stand the current for any
length of time.
The solution here may be to replace this extended foil capacitor with the
tab type if the total capacitor current is excessive. (See Table 7.1 through 7.4.)
TABLE 7.1 Chart of Voltage Ratings

Maximum AC rating
Maximum DC rating
220
440
660
399
600
1,000
1,500
2,000
2,500
4,000
6,000
10,000
15,000
Minimum margin (inch)

3/32
1/8
3/16
3/16
1/4
1/4
7/16
1/2
3/4
1
96
Chapter 7
TABLE 7.2 RFI Corp. Dielectric Rating

DC volts
AC volts
100
400
600
120
250
Mylar
Paper
1X 0.00025
2X 0.00025
2X 0.00035
2X 0.00025
3X 0.00025
3X 0.0003
Source: Robert Hassett, VP Engineering, retired.
For 3/32 inch and above use 1/8 if possible. For paper more than 4 inches wide,
add 1/16 inch to all readings.*
Note that there are two foils and two dielectrics per thickness in the three
columns. Also, some extra allowance for bends and so on is included in the
preceding chart.
As the line frequency rises, the margins must be increased if the margins
were on the ragged edge or if the loss is due to dv/dt increase. The margins also
increase with the voltage as the information in the tables suggests. This must be
done to handle creepage. These charts and drawings are included to help the
designer. In the design, the first step is to determine the dielectric.
Polycarbonate is good without derating from 55 to +125C. and has a low
value of the dissipation factor, giving a higher Q capacitor. This type of
dielectric is easily contaminated by impurities either in storage or during
winding. However, the use of this material is greatly reduced because of
the high cost.
Polyester (Mylar) is good for the same range if the voltage is derated by
50%. Polyester is also rated from 55 to +85C without derating. This
has a higher dissipation factor or the lowest Q. It is used more for 60 Hz
and DC, especially the metallized type. For constant DC operation, there
is no current flow though the capacitor, eliminating heat buildup in the
metallized capacitor.
Polypropylene is also rated from 55 to +85 and to 105C with a 25%
voltage derating. The Q is higher, or there is a lower dissipation factor,
than for Mylar (polyester). The cost of this material is now less than that
of Mylar. This material is suggested for 400-Hz capacitors.
Paper in oil-filled caps is also good to +125C, and the quality of the
capacitor is similar to that of Mylar. However, with the costs being lower
for polypropylene and Mylar, this is rarely used. This style must be
*All information in this section was provided by RFI Corp., which has been very helpful for many
years. This is especially true of all the people in engineering. Without their help, this book, and much
of the seminar material, would not have been available.
97
impregnated, whereas Mylar, polypropylene, and polycarbonate do not

require impregnation.
The required thickness of the dielectric depends on the voltage. This is true
with or without derating, and the dielectric material is most often double, triple,
and quadruple lapped as the preceding tables suggest. This is to find the proper
thickness for the voltage. The best way is to make sure that there is no more than
400 volts RMS per mil of dielectric material regardless of the dielectric material
used. Each has a different value of K, the dielectric constant, needed to calculate
the active length of the capacitor. Next, the full thickness of the dielectric plus
the film must be determined based on the voltage derating listed in the tables.
If 200 V is needed at 105C, the thickness must be good for 266 V because
of the derating based on polypropylene and foil.
TABLE 7.3 RFI Corp. Oil-Impregnated CapacitorsDielectric Paper to +125C

Test
Ply thickness X
DC
volts
Flash
2X
2X
2X
3X
2X
1X
2X
3X
2X
3X
2X
2X
4X
3X
3X
4X
3X
2X
4X
4X
4X
5X
5X
65
100
160
200
200
215
240
400
500
600
750
800
850
850
1000
1200
1300
1500
1600
1800
2000
2200
2500
160
250
400
500
500
540
600
1000
1250
1500
1875
2000
2125
2125
2500
3000
3250
3750
4000
4500
5000
5500
6250
0.0002
0.00025
0.0003
0.0002
0.00035
(0.00035
0.0004
0.00025
0.0003 +
0.0003
0.0003 +
0.0004 +
0.00025
0.00035
0.0004
0.0003
0.0005
(0.00035
0.0004
0.00045
0.0005
0.0004
0.0005
+ 0.0004)
0.00025
0.0004
0.0003
+ 0.0004)
Source: Robert Hassett.
Test
1 min
AC
volts
Flash
1 min
130
200
320
400
400
430
480
800
1000
1200
1500
1600
1700
1700
2000
2400
2600
3000
3200
3600
4000
4400
5000
35
55
85
95
120
130
175
200
240
300
330
340
360
360
400
440
500
550
600
660
720
880
1000
90
140
215
240
300
315
440
500
600
750
825
850
900
875
1000
1100
1250
1375
1500
1650
1800
2000
3500
70
110
170
190
240
260
350
400
480
600
650
680
720
700
800
880
1000
1100
1300
1320
1440
1600
2000
98
Chapter 7
TABLE 7.4 RFI Corp. Mylar with Aluminum Foil Thickness of

0.00025 t in Equations
Mylar thickness
Plate pressed
Flat
Round
0.00040
0.00045
0.00050
0.00055
0.00060
0.00065
0.00070
0.00075
0.00080
0.00085
0.00090
0.00095
0.00100
0.00105
0.00110
0.00120
0.00130
0.00140
0.00150
0.00160
0.00180
0.00200
0,00210
0.00220
0.00230
0.00240
0.00250
0.00280
0.00300
0.00350
0.00400
0.00450
0.00500
0.00550
0.00600
0.00134
0.00144
0.00154
0.00164
0.00174
0.00184
0.00194
0.00204
0.00214
0.00224
0.00234
0.00244
0.00259
0.00264
0.00274
0.00294
0.00314
0.00334
0.00354
0.00374
0.00414
0.00454
0.00474
0.00494
0.00514
0.00534
0.00554
0.00614
0.00654
0.00754
0.00854
0.00954
0.01054
0.01154
0.01254
0.0014
0.00150
0.00161
0.00172
0.00182
0.00193
0.00204
0.00215
0.00225
0.00236
0.00247
0.00258
0.00268
0.00279
0.00289
0.00311
0.00332
0.00354
0.00375
0.00396
0.00439
0.00482
0.00503
0.00523
0.00546
0.00568
0.00589
0.00653
0.00696
0.00803
0.00910
0.01017
0.01124
0.01231
0.01338
0.00142
0.00153
0.00164
0.00175
0.00186
0.00197
0.00208
0.00219
0.00230
0.00241
0.00252
0.00263
0.00274
0.00285
0.00296
0.00318
0.00340
0.00360
0.00364
0.00406
0.00450
0.00494
0.00516
0.00538
0.00560
0.00582
0.00604
0.00670
0.00714
0.00824
0.00934
0.01044
0.01154
0.01264
0.01374
Source: Robert Hassett.
99
7.5.2. Round Capacitor Calculation

A 4-F feed-through capacitor is required for 220 AC, 60 Hz, requiring an arbor
0.125 inch in diameter (see Figs. 7.11 and 7.12). First determine the material, here
Mylar. The 4 F almost eliminates metallized Mylar, but these are made by many
companies and would be cheaper than building in house, unless the quantities are
high enough. So extended foil is the choice. Divide the 220 RMS volts by 400 V
per mil and get 0.55 mil. The available thicknesses are typically 0.24 and 0.32
mil. Two at 0.24 mil give 0.48 mil, which is too small. One 0.24 and one 0.32
give 0.56 mil and are usable if the capacitor designer is pressed for room. Here,
two at 0.32 should be used, giving 0.64. This is the value of t in the formula, the
total dielectric thickness.
Next, the foil thickness is determined and typically is 0.23 mil. The
winding machine will then have six rolls of materialfour Mylar and two foil.
The value of the setup, in inches, is 2.2 t (in inches) + 2 0.00023 = 2.2
0.00064 + 0.00046 = 0.00187. The 2.2 is to allow for the growth due to
winding. The foil width is the dielectric width of the Mylar less one margin
plus an extension. Here, 1 0.09375 + 0.09375 = 1 inch. Solve for the active
foil width:
A(FW) = FW 2 EXT = 1 2 0.09375 = 0.8125
where Fw is the foil width and Ext is the extension, both in inches.
FIG. 7.11 Extended foil capacitors.
(7.4)
100
Chapter 7
FIG. 7.12 Close-up view of extended foil plates with dielectric. T is the dielectric
thickness in mils, and t is the thickness of two dielectrics and two foil plate
thicknesses in inches.
Next, the active foil length is calculated, but first the capacitor constants,
K, are needed for round feed-through capacitors. The typical K values are as
follows:
Polyester (Mylar)
Polypropylene
Polycarbonate
Paper (resin or PBT)
900
1200
840
580 (wet), 550 (dry)
Solve for the active length of the foil:

900 4 0.64
KC t 1000
=
= 2835.7
0.8125
AW
(7.5)
where K is the constant, C is the capacitance, and t is the total dielectric thickness.
Next is the foil area. This is 1.27 setup the active length.
1.27 0.00187 2835.7 = 6.73
(7.6)
From this value and the arbor diameter, the outer diameter (OD) of the capacitor
is calculated. This is the square root of the foil area plus the arbor squared.
(7.7)
At this time, it is wise to check the gauge. This is the Mylar width plus two
extensions, assuming there is no problem with the diameter for fit at 2.6 inches.
Extra room is required for installation.
Gauge = DW + 2 Ext = 1 + 2 0.09375 = 1.1875
(7.8)
Compare the gauge with the diameter, here 2.6 /1.1875 = 2.2. This is close to the
optimum ratio, giving close to the minimum ESR and ESL and resulting in the
highest SRF. The winding is practical.
101
Next is the number of turns.

N =
(OD ID)
(2.6 0.125)
2.475
=
=
= 662
2 Su
2 0.00187
0.00374
(7.9)
7.5.3. Pressed Capacitor

The arbor is removed after winding and the capacitor is then pressed flat with
weights overnight. To calculate the flat sides from the round above, a stacking
factor of 0.95 is needed. The small flat is
S(flat) = 2 (N + 2) SUSF =
2 664 0.00187 0.95 = 2.36
(7.10)
and the large flat is

S(flat)
A(rbor)
+
LF = 1.1
= 1.1
2
SF
= 1.1
0.125
2.36
0.95 +
2
(7.11)
(2.48 + 0.196) = 2.94
This calculation was based on the round capacitor. If a flattened capacitor is being
designed, the K constants do change:
Polyester (Mylar)
Polycarbonate
Paper (resin or PBT)
800
750
515 (wet)
Other equations needed for the winding for either the flat or the round
capacitors are:
Dielectric width = Aw + 2M
Foil width = Aw + M + P
Gauge = Aw + 2M + 2P
where P is the foil overhang that is sputtered over to make the lead connections
on each end or form the end area for the feed-through type.
For the feed-through type (Fig. 7.13), the arbor tube is left in place and the
maximum filter current, not the capacitor current, must be known. A conductive
bolt, or a threaded rod, probably not copper because copper lacks the strength
necessary to handle the torque required by the nuts, is silver plated by a military
specification plating company (assuming a military contract). This bolt is designed to carry the full filter current and must fit through the center of the arbor.
The diameter needed for this material to handle the current with low heat rise is
often the factor determining the inside arbor diameter. This bolt is often plated to
102
Chapter 7
FIG. 7.13 Feed-through capacitor construction showing the case.
help handle the current and normally forms the outside stud for electrical
connections for the output.
The construction is as follows. A plated conductive bolt is threaded through
the contact washer or terminal, the plated conductive washer, and the body of the
capacitor through the arbor hole. The washer presses against the inside sputtered
end for good contact. The diameter of the conductive washer is the full diameter
of the capacitor, so the silver-plated washer covers the entire extended foil area
that has been swedged. Another conductive washer follows this full arrangement.
It is then placed through the larger container hole. An insulator washer, usually
made of ceramic, is slipped on the end bolt sticking through the can, followed by
a nut tightened to a specific torque. This presses the last washer and the end of
the capacitor to ground, forming one terminal. The hot side is formed by both the
bolt on the outside of the can and the connection on the inside of the can. Some
use a plate, called a ground cup and also ferrules, that has been stippled to press
into the extended foil area to replace the two silver-plated washers. These
numerous indentations or stipples help to maintain prolonged electrical contact.
8
Filter ComponentsThe Inductor
The EMI filter requires what is termed a soft core. This really means that the
core is driven into saturation slowly rather than abruptly as required for pulse
transformers and magnetic amplifiers. A hard core can be made soft by gapping
the core. This technique tilts over the BH curve (Fig. 8.1), making the core a soft
coreharder to drive into saturation.
8.1. INDUCTOR STYLES AND SPECIFICATIONS
Inductors for EMI filters come in three styles: tape-wound, toroids, C cores, and
slugs. E cores, pot cores, and RM cores are rarely seen. There are several
subgroups. For toroids, the styles are ferrite, powdered iron, MPP, and high-flux
(HF). For C cores, or cut cores, the styles are various steel mil thicknesses as well
as steel types. The same is true for the tape-wound toroid. For the slug, other than
size, there are various mixes.
As far as specifications are concerned, some filter customers specify the
core, the wire size, and the inductor value, usually as a range such as 10%.
Others specify creepage distance or list a specification that calls out the distance.
In this case, the clearance listed is between the top and bottom wires to the core.
For the tape-wound toroid this does not apply. This is more like a margin along
the coil form. Sometimes a wire size is specified.
104
Chapter 8
FIG. 8.1 BH curve for soft core material.
8.2. THE POWDER CORES

The core types used for EMI filters are often Molypermalloy toroids and
sometimes the HF type. Powdered iron toroids are also sometimes used but often
cause trouble at 400 Hz because they overheat. The main advantage is that all
three types of cores can have very high Q values. This decreases the losses and
also allows the coils to be tuned if necessary. The permeability of MPP or HF
types should be limited to 125 and below at 400 Hz for the same reason. This is
developed as follows using MPP cores from Magnetics, Inc. through Col. W. T.
McLyman, Jet Propulsion Laboratory, Pasadena, California (now retired).
Core manufacturers provide important information about their cores such
as outer diameter, inner diameter, height, window area (Wa), cross-sectional area
(Ac), weight, and magnetic path length (Mpl). If the window area is given in
circular mils, as Magnetics, Inc. does, the designer can choose as many circular
mils per ampere as needed. The maximum number of turns that will fit in the core
window can be calculated. As the current required increases, the turns that will fit in
the window drop, and as the needed current falls, the number of turns increases.
For any circular mils per ampere that are chosen, NI is a constant, making
H a constant for the core. From H, B can be found utilizing the information
provided by the manufacturer and using the winding or fill factor of 0.4 for the
window of the toroid. Please note here that this is not the proper way to design
the inductor. This is done only to ensure that all the cores in the series are equally
wound with the same fill factor.
N =
105
0.4Wa
0.4Wa
0.4NIrms
0.16Wa
NI
=
H =
=
CMIrms rms
Cm
Mpl
MplCm
where Wa is the window area in circular mils, Cm is the circular mils per ampere
needed for proper wire sizing and low DC resistance (DCR), Mpl is the magnetic
path length in centimeters, and 0.4 is the typical window fill factor for toroids.
Magnetics, Inc. lists core number 55438 for permeability 125. The window
area in circular mils is 842,700, the magnetic path length (Mpl) in centimeters is
10.74, and the weight in pounds is 0.4. Using 600 circular mils per ampere, H
follows.
0.16 842,700
= 0.65734
10.74 600
for each permeability using the BH curves furnished by the manufacturer yields
approximate values of B (Table 8.1). The given, or initial, permeability is listed
as a reference and B was estimated from the BH curve. With this information, the
approximate value of the actual permeability of the core was calculated. The watts
per pound follow from these formulas, also provided by the manufacturer
for MPP cores (Table 8.2). From the preceding equations the next column,
watts/pound, was determined and the last column followed on multiplying the
W/lb by the pounds furnished by the manufacturer, 0.4 here.
The watts per pound and total loss columns show a hefty jump from
permeability 60 to 125 and continue hefty increases upward. Using permeability
125 is questionable. The same thing happens with powdered iron at even lower
permeability. Reducing the frequency back to 60 Hz makes a large difference and
TABLE 8.1 The Permeability and Flux of a Core

Given u
Actual u
400 Hz
(W/lb)
Loss
(total)
14
26
60
125
147
160
173
200
300
550
13.69
25.86
45.64
83.67
86.71
95.84
99.64
103.45
109.53
117.14
900
1700
3000
5500
5700
6300
6550
6800
7200
7700
0.0377
0.1827
0.3091
1.0576
1.3471
1.6806
1.8315
1.8448
2.3278
3.9327
0.0151
0.0731
0.1236
0.4230
0.5388
0.6722
0.7326
0.7379
0.9379
1.5730
106
Chapter 8
TABLE 8.2 Watts per Pound for MPP Cores

14
26
60
125
140, 160, 173
200
300
0.312
0.360
0.828
0.489
0.425
0.788
0.890
1011
1010
1011
1011
1011
1011
1011
F 1.32
F 1.16
F 1.23
F 1.28
F 1.23
F 1.35
F 1.26
B 2.25
B 2.07
B 2.12
B 2.14
B 2.21
B 2.06
B 2.11
all the different permeabilities can be used. If this idea is violated, the core can
overheat. This concept can be overlooked if the inductor design is very conservative and if the winding factor is low, with the turns well below the maximum
or the window area of the core less filled. An example occurs where there is a
single wire layer on the core. This is given as fair warning of things that have
happened at several EMI filter companies. Refer again to Table 8.1.
Note that the flux density is too high at 125 and above anyway.
Another way to show the same thing is to require the same inductance at
the same current. The following equation of Al values for Table 8.3 was developed
using the same core as above at 5 A RMS at 380 H. See the next section for a
discussion of Al values.
Ip =
2 Irms
B = H
H =
0.4NIp
Mpl
The wire size is constant using the 600 circular mils per ampere, so No. 16 AWG
is used. Ip is 7.07 A to determine H.
The B term was determined using the initial permeability, and the error
increases with increased permeability. These could have been read off the BH
curve, and the values would be less than those listed in Table 8.3. Regardless of
how the value of B was determined, these should not be used much above 3000
gauss, so permeability 125 might be okay, but 147 and above should be avoided.
Another toroid is the ferrite one. This is used mainly for common mode.
The ferrite core lacks the distributed gap so the BH curve is too square looped for
a differential mode inductor. The ferrite toroids saturate at the coercive force
level. However, they work well for common mode, where the sum total of
currents through the center is zero and creates a balance of flux that cancels.
Another disadvantage is that they are very noisy when driven into saturation. This
shows that they should not be used as differential mode inductors.
107
TABLE 8.3 Constant Current at 5 A for MPP Core

Showing Al Values
Permeability
Al
14
26
60
125
147
160
173
200
300
32
59
135
281
330
360
390
450
674
109
81
53
37
34
32
31
29
24
90
67
44
31
28
26
26
24
20
1260
1742
2640
3875
4116
4160
4498
4800
6000
C cores are used especially at the higher current lines, mainly the power
line filters used for shelters, EMI test houses, screen rooms and secure communication applications. The disadvantage is their low Q values, and they are usually
not effective much above 100 kHz. The thickness of the magnet core tape used
to wind these C cores should be able to handle the fifth harmonic of the power
line frequency. This dictates a thinner core material but increases the Q and gives
further life at the higher frequencies. The potential harmonic current from off-line
regulators is also improved using the thinner core material. This should be done
to avoid heating the core. This is because of the harmonics on the line and from
these off-line regulators. If the filter is to be used in areas where the power feed
is more resistive with less harmonic content, this principle can be violated. Also,
if the inductor was designed with a low temperature rise, then the core can take
the extra heat. In this way, the wire and the core loss add to the filter via
dissipation. Where the high current spikes can flatten the voltage, creating high
levels of harmonic current and also some harmonic currents from the load, this
principle, or the following option, cannot be ignored. The other option is to design
the inductor using very low flux density and allow the line harmonics to heat
the core.
The costs of the core and filter are increased by either method. The first
allows a smaller core that is more expensive, and the second allows a less
expensive type of core but is often offset by the larger size and weight. One other
factor not to overlook is the smaller thickness of tape used in the first method,
which has a lower stacking factor, decreasing Ac, the core cross-sectional area.
The value of Ac, then, must be increased. The higher switcher frequencies and a
parasitic oscillation from the load side should be allowed to dissipate their energy
within this core material via eddy currents and hysteresis whatever the thickness
108
Chapter 8
of the tape. Additional losses for these higher frequencies can be gained in the
skin effect of the wire. Allow both methods to dissipate this high-frequency
energy.
The core material thickness that is normally applied is 11 or 12 mil for
60 Hz, 4 mil for 400 Hz, and 2 mil for DC for typical applications. Core
thicknesses available are 1, 2, 4, 7, 11, and 12. These often drop down one size
as described earlier. This reduces the core loss, lowers the temperature rise, raises
the Q, and increases the self-resonant frequency (SRF). The tape-wound toroid of
the same material has about the same constraints except that Q is higher and the
turns are typically lower.
High currents demand large wire diameters. Welding cable is often employed because it has many strands such as 833 strands of No. 30. This makes
the wire flexible, allowing it to be wrapped around the coil form. I have known
many engineers who wanted to use Litz wire and high-frequency cores, which
would increase the cost several orders of magnitude. But, for EMI filters, either
Litz or welding cable should be used only for ease of winding the coil form.
If the coil is to be tuned, the Q must be high at the tuned frequency. This calls
for a core with thinner material and stranded wire, so the AC resistance will be
low at and somewhat above the tuned frequency. For example, a 60-Hz inductor
is to be tuned to 25 kHz. Change the core from a National Arnold CA to a CH.
This moves the material from a 12- to a 4-mil core. The radius of conduction in
centimeters is 6.62 divided by the square root of the frequency. Up the frequency
to 30 kHz and take the square root of it, which gives 173.2. Multiply by 6.62 and
divide by 2.54 times the 173.2, and the wire diameter in inches is now known.
This is 0.030 inch. A No. 20 wire is 0.032, which is close enough, and this wire
has 104.2 circular mils. If 10 A is to be carried, approximately 5000 circular mils
is needed, which would require 48 strands of No. 20 wire. It should be apparent
that Litz may be the best choice.
The C core is easy to gap using shims of half the gap value in each leg,
showing that this would be good for DC operations and would make the core
softer. They are easy and quick to wind on a bobbin or coil form, whereas toroids
are much slower to wind. If many of these inductors are needed, several C core
bobbins or coil forms may be wound at the same time, depending on the wire
size, whereas a single toroid must be wound at a time. Usually, there is not enough
demand for this inductor to warrant the time to set this production up. Therefore,
C cores are quicker to manufacture.
Slugs, or plain cores, are also used for EMI filters but are harder to calculate
because the flux path is not complete as in toroids or C cores. The trouble is that
the flux can be routed through surrounding pieces or through the filter container.
Give them extra clearance from the container and other surrounding components
to reduce the flux through them. If another magnetic material is too close, the flux
paths shift and also increase the inductance. This stray flux can heat the container
and other objects, if violated. Usually, this type of inductor is in the low
109
microhenry range. I have designed them to 158 H and placed them literally
hanging in the middle of the container and they have functioned well. The slug
size was 1 inch in diameter and 2 inches long using Micrometals P6464-140.
The equations are:
(0.8UEFF(RN)2
L =
(6R + 9T + 10B)
1 L(6R + 9T + 10B) 2
N =
(8.1)
R
0.8UEFF
This is for multilayer windings, where UEFF is the effective permeability, R is the
radius, T is the length, D is the core diameter, and B is the coil build (Fig. 8.2).
The value of UEFF is a low percentage of the value of because the T-to-D ratio
used in most EMI filters is so low, typically 1:1 or 2:1. The estimation in most
cases is a UEFF of 5.
The SRF of the inductor is a function of the size and number of turns. There
are several ways to raise the SRF for the toroid. The first is to use the progressive
winding technique, such as six winds forward around the core followed by five
turns back over the previously wound turns, then another six turns forward over
the previously counterwound turns followed by another five turns back (Fig. 8.3).
Continue until all the turns are on the core and the number of turns can be placed
around the core without making much more than 320 degrees of the circumference of the core window. The 6/5 ratio can change to place the last turn at
approximately 20 degrees separation from the start lead. Place a barrier or space
between the start point and the finish point. This barrier is used so that the two
ends cannot slide together and could be anything such as tape. This technique
decreases the turn-to-turn capacitance that is then in series across the inductor.
FIG. 8.2 Slug.
110
Chapter 8
1 2 3 4
5 4 3 2 1 5 2
1 2 3 4 5 6 6 1
CORE
FIG. 8.3 The progressive wind, a higher SRF.
This works because the capacitance varies with the voltage difference between
the turns and the voltage difference is small for the nearby turns using this
technique. The disadvantage is that this looks as if a 5-year-old child wound it.
People normally cover it with tape to help hide this because if uninformed
customers saw it they might question the ability of the company.
Windings that are side by side where there is less than a full 360 degrees
around the core are also satisfactory, giving lower capacitance from turn to turn
for the same reason. The voltage from turn to turn is very low. Where multiple
layers are needed, it is obvious that layer 2 to layers 1 and 3 has much higher
turns, making a higher voltage gradient and giving much higher capacitance from
turn to turn. This guarantees low SRF, which compromises the filter action. The
second way is to keep the winding off the core. Tape the core with several
thicknesses to keep the turn-to-core capacity as low as possible. Toroids that are
already coated can be purchased. Often a layer of tape is needed the raise the wire
farther off this core. This capacitance is from turn to core and is somewhat in
parallel, so the capacitances add and are more directly in parallel with the
inductor. Figure 8.4 shows the core with tape and a single layer of turns.
If multiple layers are needed, use the progressive technique just described, along
with this technique. If a single layer is adequate, try to separate the turns slightly
to lower the turn-to-turn capacitance. Do not allow the two ends to come too close
together; try to keep them at least 20 degrees apart.
Another way to raise SRF is to use heavier insulation. This forces higher
spacing from turn to turn and helps raise the copper farther off the core and
decrease the turn-to-turn capacity (Fig. 8.5).
111
TAPE
CORE
CORE
TAPE
CORE
TAPE
FIG. 8.4 Taped core for better SRF.
8.3. INDUCTOR DESIGN

The design of the inductor has been covered for the most part in Chapter 3 (see
Sec. 3.2). Another common way is to design with Al values. These are listed as
millihenrys per 1000 turns and sometimes as microhenrys per 100-turns. Nanohenrys per turn is also used but rarely for the type of filters designed in this book.
This technique can be used for all inductors including common mode as long as
the Al value is listed by manufacturer or known by some other means. This was
developed from the following:
L =
0.4N2Ac 108
Mpl
(8.2)
In this equation, the needed inductance, L, is known along with 0.4, , Ac, Mpl,
and the permeability for the core that the designer would like to use. What is not
known is N or N squared. These known core quantities make up Al.
Al =
0.4Ac 108
Mpl
(8.3)
The manufacturer provides the Al values.

L = AlN2
(8.4)
The needed inductance is known and the turns need to be found. The Al value is
given, so
112
Chapter 8
(8.5)
where N2 is the needed turns, N1 is the known turns (1000), L1 is the needed
inductance in the same units as Al (millihenrys), and, again, Al is the known
millihenrys associated with the turns given (1000).
For example, the Magnetics, Inc. 55351 MPP core is listed as having an Al
value of 51 mH per 1000 turns. The designer needs 0.8 mH (800 H). L1 is equal
to 0.8 and Al is equal to 51. Then 0.8 divided by 51 is equal to 0.01569 and the
square root of this is 0.1252. This value times 1000 equals 125 turns.
(8.6)
The current is multiplied by the circular mils to determine whether the window
area is adequate, that is, to see if the turns will fit in the core window area.
This is a small core but we need 1 A, and using 600 circular mils per ampere
would lead to 23 AWG at 650 circular mils per ampere. The 650 times 125, the
number of turns just calculated, equals 81,250 circular mils, but the winding
CORE
start
TAPE
to keep spacing
FIG. 8.5 Core showing the start-to-finish spacing.
finish
113
factor for this toroid-type core is only 0.4. The 81,250 divided by 0.4 yields
203,125 circular mils.
650 125
= 203,125
0.4
Magnetics, Inc. lists the window area as 293,800 circular mils, so the windings
will fit. The wire size could be changed to 22 AWG, giving 253,125, and still fit.
This would reduce the copper losses and lower the temperature rise.
I recommend Col. W. T. McLymans programs for selecting the proper
inductor core and turns. The heat rise is a common filter problem and the filter
design should be rather conservative, and McLymans programs meet these
requirements.
8.4. CONVERTING FROM BALANCED TO UNBALANCED
OR THE REVERSE
Conversion from balanced to unbalanced, or the reverse, is required because the
equations are all based on unbalanced circuits. It is better to balance the filter if
the supply or equipment has not already been grounded. This happened to a
customer who specified the complete filter design. We tried to tell the customer
that the bottom half of his filter was not working, but (alas poor customer, I knew
you well!) it was never changed because it went through qual that way and would
have cost almost $100,000.00 to change it.
FIG. 8.6 Multiple L network converting from unbalanced to balanced.
114
Chapter 8
To balance the filter, as in the 800 H discussed in the last section, use a
double L with some value of capacitorsay 1 Fwith a limit of 0.1 F to
ground. The picture would be 800 H series, 1 F Y capacitor shunt, followed
by two just like the first two, then finish with the two feed-throughs on both lines.
Simply divide both of the 800-H inductors in half so that the circuit ends up
with 400 H on both the hot and neutral with the X capacitors of the same values
across the lines. The inductors are smaller and the SRF should be higher. Mount
the four inductors in quadrature (the two inductors that would be farthest apart
would have to be mounted in the same way, hoping that the extra distance would
reduce the coupling), and keep the capacitor leads as short as possible. Figure 8.6
shows the before and after drawings.
9
Common Mode Components
The causes of spikes or electromagnetic pulses (EMPs) on the line are lightning,
large inductive equipment shutting off, and magnetic pulses created by nuclear
activity. Lightning and nuclear activity create common mode pulses between the
lines and ground, whereas the equipment type creates a differential mode type of
pulse between the lines. On the load side, the switcher parasitic and the diodes
are the leading culprits generating noise. Any current pulse, on the load side of
the off-line regulator, to ground appears as common mode to the line. Adding a
transformer or keeping the power supply isolated reduces these common mode
conducted emission pulses. The one exception is the primary to secondary
capacitance of the transformer, but this is so small that it can be neglected.
Power supply filter capacitors pumping current to ground should be eliminated.
To eliminate the common mode, the EMI filter employs common mode chokes
and feed-through capacitors to ground. The problem is another specification
limiting the leakage current. This is the capacitive reactance current flow through
the capacitor between line and ground. This is another misnomer; it is really
reactive current.
9.1. CAPACITOR TO GROUND

For 400 Hz, the limit of the capacitor to ground is 0.02 F for Military Standard
461. Worse yet, the medical leakage current specification is harder to meet. If the
device touches a patient, the total system leakage is limited to 100 A. This means
that most of the power supply people want the filter restricted to 20 to 40 A.
It is difficult to have the common mode losses meet the common mode loss
116
Chapter 9
specifications with capacitors to ground this small. A transformer would help. See
Chapter 10. There are two schools of thought on this because the specifications
governing this, as usual, are not clearly written. The first group thinks that this is
the total capacitor to ground. In a three-phase, four-wire circuits, the capacitor
limit value for 400 Hz is 0.02 F. This capacitor value would then be shared, or
0.02
= 0.005 F
4
This means 0.005 F for each of the four legs to ground. The second group thinks
that this is the maximum per line whatever the number of lines. If the system is
well balanced, the current on each leg would nearly cancel through these
capacitors at the ground point. (See the next section on virtual ground.) I hold
with the latter group because it makes my job as a filter designer easier. However,
ground fault equipment would not allow any capacitance to ground that would
produce a current above the current threshold of the ground fault device, or it
would have to be in the circuit after the filter. Again, a transformer would
circumvent this.
9.2. VIRTUAL GROUND
In a two-phase system, where the two lines are 180 degrees out of phase, a virtual
ground can be employed. This is the common method for power. Either leg to the
common ground gives 120 V RMS and the two outer legs 230 V line to line and
is really two phase. If the two line voltages are nearly equal and opposite, then
two capacitors of nearly equal value from each line to ground have nearly equal,
but opposite, currents. Whatever current flows in one capacitor to ground, the
other capacitor has that much current flowing from ground. The two currents
almost cancel. The small ground current is due to the slight differences in the two
voltages and the differences in the capacitor values.
The same is true for three-phase systems. If the three RMS voltages are
nearly equal and the capacitors to ground are nearly equal, the current to ground
will be small. If the line voltages are the same and the capacitors are the same,
the ground current is zero.
One way to help remove some of the difference currents is to employ the
virtual ground technique. Tie the junctions of the capacitors together to form a
virtual ground. Under ideal situations, the junction voltage to ground would be
zero. Tie a capacitor of equal value from the junction to ground. Ground current
will flow through the added capacitor based on the junction voltage. This
technique further reduces the current on the ground lead.
Why are capacitors to ground necessary? Common mode reduction most
likely requires them even with a transformer. The three capacitors in two-phase
systems and the four in three-phase systems have a reasonably high impedance
117
to ground at 50, 60, and 400 Hz, but what about at 14 kHz and above? The two,
or three, capacitors from the lines to the junction are in parallel and are in series
with the capacitor from the junction to ground.
9.3. Z FOR ZORRO
Ferrite toroid cores are often used for common mode inductors because they have
the very high Al values required for these common mode filters. For common
mode testing, all lines are tied together in parallel and all the differential inductors
are in parallel in the balanced design. Capacitors between the lines are of no value
for common mode, but the capacitors from line to ground add in parallel. In the
preceding three-phase, four-wire case, the total capacitor to ground (my method)
would be
0.02 4 = 0.08 F
To get the needed common mode loss, a common mode choke usually must be
added. The total inductance and the total capacitance to ground normally do not
give the required loss. This problem is often solved by placing a well-grounded
barrier or shield across the filter center (Fig. 9.1). Split the value of the feedthrough capacitor limit to ground by two. Two of these smaller feed-through
capacitors are then installed in this shield. For best results with this method, put
the Zorro inductor at the low-impedance end, or the line side, and the feedthrough capacitors on the high-impedance end, or load side. In this case, where
they are being split, the first Zorro is placed at the line and the first feed-throughs
Zorro
Shield
CF
Zorro
outside case
CF
LINE
LOAD
CF
FIG. 9.1 The double-Zorro common mode filter.
CF
118
Chapter 9
are the last in the sectionmounted in the shield. The next section starts with the
Zorro and ends with the feed-throughs at the end. Try to use an even number of
differential mode filters so that they can be split evenly in the two cavities. Say
two L filters are required so that the first cavity would start with the Zorro,
followed by the first L, then followed by the feed-through capacitors. The second
section would be the same.
Of course, the line-to-line capacitors are not shown because they contribute
nothing to common mode. However, they would follow the differential mode
inductors. Put a second common mode inductor in the filter. This puts one
common mode inductor, Zorro, in the front half and another one in the second
half. This technique forms a double L in the common mode filter section, reducing
the common mode inductor greatly in size, and now the specification can be
reached most of the time. Remember that the inductors magnetic fields buck or
cancel for the differential mode and they have a high magnetic gain for the
common mode. Figure 9.2 shows a three-phase delta with spacing between
the windings.
FIG. 9.2 The RFI Corp. three-phase, three-wire common mode inductor. Note:
(1) Three windings on one core wound as shown. (2) Wire to be tightly wound and
evenly distributed over segment.
119
For the three-phase wye filter, another set of windings is required, demanding a larger core for the same form. These are excellent for creepage, but the
leakage inductance is greater, giving a differential inductance. These separate
windings create 0.5 to 1% leakage or differential inductance. Figure 9.3 shows
them wound together, reducing the leakage inductance, but this could be subject
to creepage and corona problems.
Some designers have gone from the separate core windings to the quadfilar
type to eliminate ringing. Leakage inductance and the stray capacitance in the
inductor and other wiring caused this.
9.4. CONVERTING COMMON MODE TO A
DIFFERENTIAL MODE FILTER
The common mode inductor has one value assigned to it. The inductance value
is written above the Z for Zorro, say 10 MH, which is a typical value. Either
winding should read the indicated inductance if the measurement is made with a
good inductance bridge. The reason comes from the inductance formula:
L =
0.4N2Ac 108
Mpl
(9.1)
If an inductance bridge is used to read each winding is this example, the reading
of both windings is 10 MH. If the inductance bridge is used to measure the two
FIG. 9.3 The RFI Corp. tightly coupledlow leakagecommon mode inductor.
Note: (1) Quadfilar winding on one core as shown. (2) Wire to be tightly wound
and evenly distributed. (3) Lead length to be 4 inches minimum.
120
Chapter 9
windings, aiding, what would the aiding inductance be? From Eq. (9.)1, the turns
are squared, so twice the turns gives four times the inductance, or 40 MH.
However, the windings are split, half on each line, so each half is 20 MH. These
two windings are now measured in parallel, so we are back to 10 MH! In other
words, either winding would measure X and both measured in parallelaiding
would measure X. If this was measured with the windings opposing, it would read
the leakage inductance.
A common mode inductor using a ferrite toroid core can be designed using
the Al value of the core. This would be used, though, only for bifilar types
requiring very low leakage inductance. The only difference between designing
the differential mode and the common mode is that the winding or window fill
factor is no longer 0.4 but now is reduced to 0.2 to fit the two windings. Divide
the number of leads into 0.4 to get the winding factor.
The single-layer ferrite toroid winding can also be found when the core Id
and the wire size (American wire gauge, AWG) needed to handle the current are
known. Get the wire diameter, Wd, in the same units as the Id of the core. The
total turns, Nt, is equal to
Nt =
(Id Wd)
Wd
(9.2)
If the wire diameter is much smaller than the core diameter (and it always is
except for the type discussed in Chapter 19), this approaches
Nt =
Id
Wd
(9.3)
Divide the turns by two, or whatever the number of wires is, and use the integer
value to solve for the inductance. Knowing the Al value of the core, the Al number
of turns (1000 typically), and the turns that the core can support, the inductance
value can be calculated. This would be in a single layer wound less than halfway
around. If this is not greater than the needed inductance, pick another core,
usually the next bigger, with a larger Id. Once a core is found for which the
inductance is larger than needed, resolve the number of turns required by
using the normal Al equation. Find the number of turns necessary, and use the
higher integer.
For example, 8 MH is needed with a current of 2 A peak. The AWG is No.
18, picked for the current specified. The diameter of the wire in inches is 0.0429.
Now, pick the core. Here, 42915-TC is selected from Magnetics, Inc. The Id of
this core in inches (same units) is 1.142
Nt =
1.142
= 83.629
0.0429
121
Divide this figure (83.629) by 2, obtaining 41.81, and use the integer. Here the
integer is 41 turns. The Al value of this core is 3868 MH per 1000 turns using F
material. Here, the backward formula for Al is used for 41 turns.
L =
Al412
10002
= 6.502 MH
This is a little low, so a larger core or a material with a higher Al must be selected.
The diameter of the wire is still 0.0429 inch. Pick a new larger core, here
43615-TC from Magnetics, Inc. The Id of this core, in inches (same units),
is 1.417.
Nt =
1.417
= 103.768
0.0429
Again, divide this by 2, obtaining 51.88 turns, and use the integer.
L =
4040 512
= 10.508 MH
10002
The integer is 51 turns. The Al value is 4040 using F material as before. This is
2.5 MH more than our goal of 8 MH. Solve for the turns needed with the normal
Al formula.
Use the upper integer or 45 turns for each half of the windings. Keep the
end gaps between the two halves as far apart as possible. This creates a visible
winding gap on the core and makes this gap as large as can be. The difference
between 51.88 and 45 gives this gap spacing between both ends of the two
windings; 6.88 times the wire diameter of 0.0429 inch gives approximately this
spacing. This is on an arc and the whole turn will not touch, eating some of this
circumference, so the value will be less than the 0.295 inch calculated. Again, I
would tape, or fill, these spaces so that both gaps between the two windings are
secure (Fig. 9.4). Now the next question: will it fit the required box? Add 2.2
times the wire diameter to the OD to get the outside diameter, which may not fit
the box. Try two smaller cores stacked together; the Al value doubles.
The problem is, now that the method of obtaining the proper core size is
known, how was the value of 10 MH determined for the Zorro? Two things must
be resolved. The first is how to convert from the common mode to the differential
modereally, from balanced (as the common mode is) to unbalanced, often
called normal mode. If several balanced normal mode networks follow the
122
Chapter 9
winding
CORE
GAP
TAPE
to keep spacing
LINE
GAP
start
LOAD
finish
winding
FIG. 9.4 Single-layer common mode inductor showing the spacing between
windings.
common mode inductor, this must be converted back to unbalanced differential

mode. All this is done in order to ease the calculations.
In Fig. 9.5, the Z is first followed by two L filters in turn, followed by
feed-through capacitors. The common mode has been discussed and converts to
a single inductor of 10 MH value. The following two inductors are in parallel
with a value of 0.5L each, giving an effective value of 0.25L. The following
capacitor, C1, is not in the circuit of the common mode, so the capacitor equals
zero. The following two inductors give the same result as the last, 0.25L. C1 again
equals zero, leaving the final component of C2, which is doubled in value. The
results are as shown in Fig. 9.6.
The total for the two differential mode coils is 0.5L but often ends up as
part of the overall headroom for the common mode part of the filter. The reason
CF
LINE
LOAD
C1
.5L1
C1
.5L1
CF
FIG. 9.5 The common mode and differential mode filter.
123
LINE
LOAD
2C2
FIG. 9.6 The common mode converted to the unbalanced filter of Fig. 9.5.
is that the total value of these two coils may be around 400 H, which is an order
of magnitude lower than for the Zorro inductor. C2 may be limited in size by the
specification, such as MIL STD 461, where the maximum for 400 Hz is 0.02 F
but now totals 0.04 F. This is now in a form where the actual value of the
common mode inductor can be solved using techniques in Chapters 16 and 17
(Fig. 9.7).
The second point to discuss is the test setup normally used to test the
common mode inductor (Fig. 9.8). The two inputs are shorted together. The input
is fed from a generator with 50 ohms output impedance. The two filter outputs,
also shorted together, feed the load. The load impedance is also 50 ohms. This is
the input impedance of the spectrum analyzer. This was another reason to convert
from the balanced to the unbalanced filter as before.
This all works out to the circuit in Fig. 9.9, and the equations are in the
next section.
10 MH
LINE
LOAD
2C2
FIG. 9.7 Final conversion of the common mode to the unbalanced filter of Fig. 9.6.
124
CF
LOAD
C1
.5L1
C1
.5L1
CF
Rs = 50 Ohms?
Filter Case Ground
Generator
Rl = 50 Ohms?
FIG. 9.8
The 220A test method for common mode section of the filter.
Chapter 9
125
FIG. 9.9 The converted test method to calculate the common mode values.
9.5. EQUATIONS FOR THE COMMON MODE VIA THE

DIFFERENTIAL MODE
The method used to arrive at Fig. 9.9 can be converted to the circuit here, where
C is equal to two times C2, L is the common mode inductor needed for the proper
insertion loss, Rs is the source impedance, and Rl is the load impedance. Because
the equation includes R, L, and C, equations based on Q (charge) rather than I
(current) of the two networks are the easiest to work with, and this generates the
matrix in Eq. (9.4). This is from impedance matrix equations.
2
1
1
LS + RSS +
Vin
C
C Q1
0 =
1
1 Q
RlS + 2
C
C
(9.4)
The delta, determinant, of the matrix is next:

(9.5)
Substitute in the initial requirements of Vin and 0 and solve for Q2:
2
1
LS + RsS + Vins
C
and solve for
(9.6)
126
Chapter 9
Q2s =
Vins
=
C
Vins
(R
(Rs + Rl)
R
C
s l + L)
LCRlSS2 +
+
LCRl
LCRl
(9.7)
but Q2 is not the goal, V0 is. In reality, the goal is the ratio
V0s
Vins
Q2sS = l2s
Vos
=
Vins
l2sRl = V0s
(9.8)
1
(RsRlC + L)
(Rs + Rl)
2
+
LCS +
LCR
LCRl
l
This equation has been published in many articles, but most often they do not
include Rs, the source impedance. Most do not use the two feed-through capacitors in parallel, which are now doubled in value.
Most solve this by completing the square. This means, in the case of
Fig. 9.5,
(RsRLC + L)
2LCRl
(9.9)
must be added to complete the square of the first two terms in the main
denominator and subtracted from the last term. This term is always much greater
in the common mode application than the last term in the main denominator,
making the new last term, for omega squared, negative. This makes the solution
a hyperbolic function and very lossy, as suggested by the test setup. Most leave
out Rs, making the value of a reduce to 1/(2RC), the dampening factor. This also
reduces the last term of the denominator to 1/(LC) 2.
(RsRLC + L)
a =
2LCRl
(9.10)
giving the following answer where a is the dampening factor.

1 at
|e sinh(t)|
LC
(9.11)
127
This may now look easy, but there is another way, knowing that the common
mode will always be lossy: forget the sin or sinh solution and set the main
denominator of Eq. 9.8 to
(S + a)(S + b)
(9.12)
So that a+b is equal to the middle term and ab is equal to the end term of the
main denominator of Eq. (9.8) repeated here.
Vos
=
Vins
1
(R
(Rs + Rl)
R
C
+
L)S
s l
+
LCS2 +
LCRl
LCRl
(9.13)
This is a simple solution, and both a and b are included within the same quadratic.
Either a or b can be assigned the positive square root, but the solution is better
with b being the more positive
(9.14)
and takes the form

eat ebt
(b a)
(9.15)
The term b a reduces to

(9.16)
This cancels LC, so the final answer is
(9.17)
In the normal test arrangement, Rs and Rl are both 50 ohms and C, because of
leakage current, is whatever the specification requires. Remember, the value of
the capacitor is doubled here because the two feed-throughs are in parallel and,
therefore, add. Now that the common mode is reduced to a simple single L filter,
128
Chapter 9
the required value of the common mode inductor can be easily solved. See
Chapters 17 and 18.
9.6. COMMON MODE INDUCTOR USED FOR
DIFFERENTIAL MODE
Some filter manufacturers have designed common mode inductors that also
function partially as differential mode inductors. This is done as before with very
wide winding spacing that generates the leakage inductance needed to provide
this differential inductance. Some flux of one coil fails to cut the other creating
this leakage inductance. Another way this is accomplished is with pot cores
(Fig. 9.10). They use a split bobbin so that some flux in one half of the bobbin
fails to cut the other half. This has been expanded to finding two separate bobbins
that fit in the core with additional room to place a washer between the two
bobbins. This washer is cut, or split, in order to avoid the washer acting as a
shorted turn. The material of the washer has little to do with the differential
inductance created. It is the separation of the two windings that causes the leakage
inductance.
The leakage inductance is easy to measure with an inductance bridge. Shunt
one winding of the common mode inductor, and read the inductance of the other
winding. If all the flux lines cross or cut the other turns, the reading is zero. This
FIG. 9.10 Gapped pot core and assembly.
129
is truly impossible to accomplish because the turns cannot all be so tightly

coupled. The difference is the leakage inductance measured by the inductance
bridge. Another way is to measure both legs together, opposing, and the inductance bridge reads the leakage inductance. Some state that the leakage inductance
is in the air, not the core, so it cannot saturate. This is true to some degree, but it
saturates a minor amount because not all is in the air.
9.7. OTHER WAVE SHAPES BESIDES SINE WAVES WORK
Some engineers to whom I have spoken think that the common mode inductor
requires a balanced sine wave for the common mode to work properly. In the
single-phase common mode inductor, the sum of the magnetic fields caused by
the currents will still cancel no matter what the wave shape. Because the current
is equal and opposite, these two flux fields still cancel. If the currents are not
equalfor example, a current difference created by the capacitors to groundthe
two fields almost cancel and give some differential mode inductance. This is the
problem in speaking to purists: most of them claim that the feed-through caps
unbalance the common mode because they are not matched. All of this is true,
but to what degree? This is also true of any arresters from line to ground because
they have different values of capacitance across them and leakage currents
through them.
This statement is very true in three-phase circuitry if the phases are exactly
120 degrees apart. Assuming this is not happening, the wave shape still need not
be sinusoidal. Any current of any shape gives a magnetic field that cancels in the
single-phase, three-phase, or DC system. If the reverse was true, common mode
would not work for three phases. Because of the imbalance of the phases and the
harmonic content, the voltage is not very sinusoidal; ergo, the common mode will
not work. The three-phase Zorro requires three windings for the delta and four
wires for the added neutral in the wye. If the three legs are not balanced, the
neutral carries the difference current and magnetic flux will still cancel because
of the fourth neutral winding.
10
The Transformers Addition to the Filter
The transformer is overlooked as a filter element. This is because of the obvious

disadvantages of weight, size, and cost. Using off-line regulators and switchers
can often eliminate transformers. On the other hand, switchers often complicate
EMI issues because of the high-frequency switching noise they generate. This is
worse as the switcher frequency increases. Also, switchers, because of their
present-day glut of components, defeat the MTBF requirements. The question
for the system designer is, do the disadvantages of transformer overshadow
the advantages?
10.1. TRANSFORMER ADVANTAGES
The main advantages of the transformer, without considering their EMI advantages, are isolation, voltage translation, common mode rejection, and the potential
of low-leakage current. Another advantage that is overlooked is the ruggedness
of this device. Transformers can handle voltage spikes without difficulty.
If the transformer is an autotransformer, isolation, common mode rejection,
and leakage current advantages are done away with because of electrical crosscoupling of the secondary to the primary. This chapter assumes that a standard
isolation transformer is used.
10.2. ISOLATION
Isolation is accomplished because the primary and secondary are coupled magnetically rather than physically or tied together electrically. Therefore, whatever
132
Chapter 10
the wiring arrangement of the secondary is, such as one output leg tied to chassis
ground, it is not coupled electrically to the primary. The green safety chassis
ground lead does not carry any current back to the service ground from the
secondary even thought the secondary is tied to this chassis ground.
10.3. LEAKAGE CURRENT
There are various specifications regarding leakage current. In commercial applications, the leakage current specification is typically 5 mA for the system. The
medical requirement is 300 A (0.3 mA) for the system, but the transformer is
usually specified as 100 A. Powertronix, a transformer house in Foster City,
California, can deliver a transformer with less than 50 A. Leakage current
through the transformer is, in theory, zero but in reality is a function of spacing
and shielding between the primary and secondary. The leakage current reduction
is also a function of a Faraday screen, if used. For transformers with bifilar
winding, where the primary and secondary are wound together at the same time,
the leakage current is drastically higher. The reason for all this is that the leakage
current is a direct function of the primary-to-secondary capacitance. If the
windings are bifilar, there is little spacing other than the wire insulation. Obviously, no Faraday screen or shield can be used, so the leakage current is due to
the capacity from wire to wire and is high.
10.4. COMMON MODE
Common mode tracks directly with the leakage current of the transformer.
Common modeequal voltage on both lines of either the primary or the
secondarydoes not create a magnetic field across the primary or secondary of
the transformer. The only coupling is through the capacitance from primary to
secondary. Any reduction in the leakage current by reducing the capacitance also
reduces the common mode.
10.5. VOLTAGE TRANSLATIONSTEP UP OR DOWN
Step-up and step-down requirements were the main reason for their development
years ago and so are not discussed here.
10.6. THE TRANSFORMER AS PART OF AN EMI PACKAGE
The problem with EMI filter design is opposite stated requirements. One of these
is a heavy common mode requirement specified along with an impossible leakage
current requirement. These two items conflict because, without a transformer, the
filter requires sizable capacitors to ground on both the hot and the return wires
Transformers Addition to the Filter
133
and a common mode inductor to remove the common mode noise, but the leakage
current requires little or no capacity to ground. A quality transformer can come
to the rescue. The transformer gives the filter designer common mode loss and
also large differential mode loss.
If the EMI filter is first in the power stream, care must still be taken in the
capacitors-to-ground values for leakage requirements. However, the transformer
should have removed most of the common mode noise anyway, easing the filter
part of the noise that it must reduce. So small capacitors and common mode
inductors should be sufficient to do the job. If the EMI filter follows the
transformer, any reasonable capacitor values could be used because the transformer will eliminate this ground current from the primary side because this is
common mode.
Is this all the transformer would do for the filter designer? No. Power
transformers using laminations, C cores, and tape-wound toroids with steel
thicknesses of 12, 11, 7, and 4 mils exhibit very high wattage losses in watts per
pound at higher frequencies. This enhances the differential mode section of the
transformer. These graphs are hard to read, but National Arnold 12 Mil Selectron
C Core lists the following approximate loss for 900 gauss. Then the loss per
octave and decade can be calculated.
Hertz
Watts per pound
1,000
2,000
10,000
20,000
0.27
1.1
24
92
10 log
0.27
= 6.1
1.1
10 log
24
= 5.8
93
10 log
0.27
= 19.5
24
10 log
1.1
= 19.22
92
Here, the weight in pounds cancels, leaving watts divided by watts. The first two
are for the octave (frequency doubling) loss using the data from the table. Both
are near 6 dB per octave, and the next two are for the decade (10 times the
frequency) loss, which gives 20 dB per decade. This loss, or cutoff frequency,
starts near the third harmonic for the steel type. Twelve-mil steel is proper for 60
134
Chapter 10
Hz, and this octave, or decade, loss should start by 180 Hz. This means that the
tester should expect to see something close to 6 dB by 360 Hz and close to 20 dB
by 1800 Hz. However, the cutoff frequency would vary from transformer to
transformer. This differential mode loss is dissipated, not attenuated. Another
way to evaluate this is through the core manufacture estimated loss equations.
These have the form
Watt/lb = CFABE
where C is a constant (possibly to a power), F is the frequency at some power A,
and B is the flux density at some power E. Because the weight (lb), flux density
(B), and constant C of the core remain the same, they cancel. Therefore the
standard dB equation can be used.
A
C F1 BE
Watt(1)/lb
=
C FA2 BE
Watt(2)/lb
FA1
F1
Watt(1)
= A =
Watt(2)
F2
F2
dB = 10 A log
F1
F2
Because the engineer is interested in the loss per octave or decade, the ratio of F
is either 0.5 or 0.1 and this is to the A power. Any frequency ratio can be
calculated, however. Armco 14-mil steel is listed as 1.68 for A. This yields
5.05 dB per octave and 16.8 dB for the decade. The problem is that with these
equations it is difficult at best to get a near fit to the listed data, but this loss for
Armco is in the ballpark. Nevertheless, this core loss adds greatly to the
differential mode loss of the filter.
The last equation can be worked backward.
dB = 10 A log
A =
F1
= 10A log(FR)
F2
dB
10 log(FR)
where FR is the frequency ratio, which is 0.5 for 6 dB. Back-solving for A,
knowing dB is 6 and FR is 0.5, gives 1.993. As an example, say a transformer has
6 dB of loss at 600 Hz. The designer needs 60 dB at 20 kHz. The frequency ratio
20,000/600 equals 33.33. The logarithm of 33.33 to the 1.993 power is 3.035, and
this times 10 is 30.35 dB. Adding back the 6 gives 36 dB. The designer needs
60 dB at 20 kHz, so 24 dB more is needed. A double-L filter added to the system
would do nicely. Four elements at 6 dB each give 24 dB. The cutoff frequency
Transformers Addition to the Filter
135
for the double L would, in theory, be 10 kHz. These would prove to be reasonably
small components giving very high self-resonant frequencies (SRFs). Is there any
other action within the transformer to aid in the loss?
10.7. SKIN EFFECT
The higher frequencies of either common or differential mode are also dissipated
within the high-frequency resistance of the wires. However, this does not come
into play until 30 kHz and above. The radius of conduction, in centimeters, is
RC =
6.62
F
10.8. REVIEW
In other words, the isolation transformer adds the same as any other EMI filter
element6 dB per octave or 20 dB per decade. A further advantage is that the
frequency cutoff point is so much lower than for the normal EMI filter components. Heavy common and differential mode loss is realized by using the isolation
transformer. Again, disadvantages are the added weight, size, and possibly cost.
If the transformer eliminates one or more filter sections, the increase in cost may
be compensated by eliminating the costs of these components. For example, if a
12-mil (for 60 Hz) steel core gives only 6 dB at 600 Hz, an additional 20 dB by
6000 Hz, and another 6 dB by 12,000 Hz, the total loss is 32 dB at 12,000 Hz,
and we can add a few for 14,000 Hz. This may be more than enough loss so that
no other filter elements may be needed.
On the other hand, the transformers effectiveness diminishes as the
primary-to-secondary equivalent capacitance comes into play. The approximate
primary-to-secondary equivalent capacitance is
C =
1
100 106
=
= 2210 pF
2EF
2 120 60
where E is the input volts at 120, F is 60 Hz, and I is 100 A. The maximum
value would be 2210 pF, and 50 A would be half that value. In a 220A test setup,
the transformer would be effective to above 700 kHz. This is where the capacitor
would be equal to the source plus load impedance, here 100 ohms. However, the
transformer effectiveness would be much lower than this frequency because of
the SRF caused by leakage inductance. But then all filter components suffer from
this dilemma, not just the transformer.
11
Electromagnetic Pulse and
Voltage Transients
This chapter develops a ratio according to the ideas stated in the following. The
circuit discussed here consists of a battery, a transmission line, and a switch.
The end piece is different, either a short or an open. The battery has an output
impedance of Z that equals the impedance of the transmission line. When the
switch is closed, a step function travels down the line at the velocity of propagation. The voltage divides between the characteristic impedance of the line and
the source impedance. The main interest is in what is at the far end of the line.
If the line is open at the far end, the voltage doubles. The wave travels back, at
the same velocity, to elevate each segment of the transmission line to full voltage.
At the time the wave reaches the source, or battery, end, the current drops to zero
and the full battery voltage is impressed across the transmission line (Fig. 11.1).
If a short is at the far end instead of an open, the current doubles when the
wave reaches the shorted end. One half of this current depletes the initial stored
voltage, segment by segment, while the other half continues to flow through the
line from the battery. This discharges each segment. When the pulse reaches
the battery end, the line is fully discharged and the double current flows from the
battery, or source. The full voltage is dropped across the internal source impedance, Z. If the line and battery Zs are not equal, the pulses iterate back and forth
until equality is reached. The main point here is to see the difference between the
two far end conditionsone near open and one near short. In the link, or short
line, the source impedance is thought to be very low and, again, the question is
what is at the filter end of the line: a shorted condition or nearly open impedance
condition. If the initial condition is high impedance for the load, the full strike
138
Chapter 11
OPEN
Generator step function
U?
INDUCTOR1
FIG.11.1
Open-ended transmission line applied to the transzorb.
voltage is impressed across the suppressor, aiding turn-on or firing. The device
then drops to low impedance and a high current is carried through the suppressor.
The suppressor and the line impedance will dissipate the energywe hope. If the
energy is too great for the suppressor, the suppressor normally first shorts and
then blows open.
The power in the pulse is dissipated in the MOV, line, and source impedance. If the condition is near short, the voltage divides, according to the ratio, and
delays the firing of the protector (Fig. 11.2).
Some engineers place capacitors across the transzorb, or MOV, making the
condition similar to the preceding shorted condition. Transzorbs and MOVs are
much faster on turn on, these days, but the capacitor must charge to well past the
turn-on voltage before the arrester can act. Another point is that the current
through the capacitor is of the order of twice the initial line current or several
times 100 A. All of this action slows, or delays, the turn-on of the transzorb. This
can also blow the capacitor, especially if the capacitor is the metallized film type,
if it was not rated for this amount of current regardless of its construction, and
again if the capacitor voltage rating was exceeded. All of this can stress not only
Electromagnetic Pulse and Voltage Transients
139
SHORTED TRANSMISSION LINE
SHORT
L
C
FIG.11.2
Shorted transmission line applied to the transzorb.
the capacitor but also all the remaining filter components. These engineers argue
that all they are trying to do is protect the arrester, but when a high-energy pulse
occurs, both the capacitor and arrester can be blown. If the filter is designed to
withstand this stress, it may well hold up, but this also adds to the cost of the
filter. If the far end (filter input end) is opentranszorb directly across the line
followed by an input series inductorthe voltage will rise quickly. The voltage
will not necessarily double; the impedance ratio is not known. The voltage will
help the arrester to fire, speeding up turn-on. The transzorb may blow, but the
filter and the following equipment will be protected. Most of these arresters
conduct in less than a microsecond under these conditions, so the peak current
through the inductor and the voltage stored across the following filter capacitor
are reduced.
11.1. THE THREE THEORIES
The first theory is from the purists. They say that the arrester is not necessary and
it is the job of the filter to handle these pulses of energy. This might be true if the
components were designed to handle them. In this case, the inductor must be
designed to handle the full pulse voltage and pulse duration without arcing.
The following capacitor must withstand twice the pulse voltage without blowing.
140
Chapter 11
The inductor must be wound with spacing between the turns or with very heavy
insulation. To calculate this, the total number of turns and the maximum spike
voltage must be known.
Maximum spike = 10,000V = Vmax and Turns = 50 = N
(10.1)
The volts per turn is

Vmax
10,000
=
= 200
50
N
Most wire insulation withstands 500 V and should not require special insulation,
and the turns, here, could touch each other. This is easy to measure. Take the
length required for the inductor and strip one end. Bury this length of wire, with
both ends exposed, in an aluminum basket of shot pellets or any other small
conductive container. Apply the test potential to the stripped wire end and the
other high-voltage lead to the shot container. What would happen if there were
only 10 turns? The volts per turn is
10,000
= 1000
10
This is marginal, so special insulation should be used and, as an extra margin of
safety, the turns should not touch. Granted, the wire should withstand 1000 V
from wire to wire, or turn to turn, but this is too risky. This spike inductor should
also have only one layer of wire and be raised off the core to protect the windings
from arcing to the core. Many are wound on forms without cores and wound in
or on a solenoid form. In the 10-turn case, these techniques must be employed.
In the 50-turn case, the same extra measures with respect to the core should be
employed: use better wire insulation, keep the turns apart so that they cannot
touch, and keep the turns as far as possible away from the core, if any (Fig. 11.3).
Also, watch the tape. Problems have been caused by tape that had too low an
isolation voltage. There is an additional theory about this. The inductor acts as a
transmission line, and the full spike voltage is impressed across the first turn, one
at a time. This charges the capacity to ground for each turn and then moves on to
the next turn. Many inductors impressed with high voltage seem to burn out on
the first few turns.
The second group wants to protect the equipment and not the filter, so the
arresters are placed at the equipment end of the filter. They also argue that they
do not want to exercise the arresters on every pulse that comes down the line. The
questions asked are, What if the pulse blows the filter? Can the equipment still
operate? These engineers answer the questions with a question, If the transzorb
blows, can the equipment still operate? The answer to their question could be
yes for two reasons.
141
TAPE
CORE
IF ANY
TAPE
FIG.11.3
TAPE
The solenoid, or core with tape, and insulated wire with spacing.
1. The arresters often blow shorted; then, with the high current through
them, they blow open. If the arrester is then open, the equipment can
still function.
2. I like placing the arresters on the outside of the filter at the input. With
proper access to the filter, these blown units, shorted or open, can be
cut out and the equipment can operate until new arresters can be
installed.
The third group locates the arresters at the front end of the filter to protect
the entire equipment. The diatribe of the opening comments also pertains to this
solution. I like the MOV at the front end with an inductive input to follow as part
of the filter (Fig. 11.1).
When the pulse reaches the arrester and inductor, the pulse sees high
impedance from the inductor; the voltage rises rapidly, firing the MOV. The initial
line current continues through the arrester, and the pulse is dissipated in the MOV
and the line impedance. The voltage that the filter sees is the MOV clamping
voltage. Also, the following capacitor should withstand at least twice the MOV
or transzorb clamping voltage.
There is a fourth group that is a combination of groups one and three. They
want the input inductor split with the arrester tied to the junction of the inductors
(Fig. 11.4). The first half limits the arrester current but must be able to withstand
the pulse. This first inductor is often wound on a bobbin without a core to
eliminate the potential for any arcing to the core.
142
Chapter 11
FIG.11.4
Split input inductor to limit the transzorb current.
There is a fifth group that uses two arresters, one on each end. This is for
very high pulses on the order of 100 kV. Often the arrester is sandwiched between
two input inductors and the last is across the output capacitor. The filter will take
the remaining energy not handled by the initial MOV and spread it over time,
lengthening the pulse width but reducing the peak energy. The last MOV, or
whatever, will handle the remaining energy.
11.2. THE LOCATION OF THE ARRESTER
From the preceding section, it should be clear that I prefer the MOV at the head
end of the filter with the filter having a series inductor at the filter input
(Fig. 11.5). I also prefer them to be mounted outside the filter, where there should
be access to change them if the need ever arises.
In the single-phase balanced circuit, three transzorbs are required: one from
hot to ground, one from neutral to ground, and one from line to line. The first two
protect the equipment from common mode pulsespulses from both lines to
ground. The last is any pulse between the lines. In the three-phase filter, where
FIG.11.5
Single phase with arresters.
143
all three lines are treated within the one filter enclosure, six transzorbs are
required. Three are wired from the three lines to ground, and three are wired from
line to line. If the units have internal arresters, the entire filter must be replaced
after an EMP or other pulse that blows the transzorbs. This can create an inventory
problem requiring expensive filters in stock. In the other method, where the
arresters are located outside, the blown arresters can be cut out. The equipment
can be operated without protection in an emergency until the transzorbs can
be replaced.
11.3. HOW TO CALCULATE THE ARRESTER

The specification of the shape and size of the strike is usually furnished by the
end user of the device. The job of the filter designer is to calculate the joules
needed. The style of transzorb is selected, and the clamp voltage for the peak
current is found from the transzorb specification sheet. From this and the various
shapes, the joules are figured out. Once the joules are known, check that the
transzorb initially selected will handle this amount of joules. The method follows.
The peak current is 50 A, and the circuit is three phase with 208 V line to line
and 120 V line to ground. Six transzorbs are needed, with three from line to line
and three from lines A, B, and C to ground. The latter three can be changed to fit
the 120 V line, say 150 V, but I recommend keeping all the same because
production or installation people do not know the difference. The radial lead is
selected because of ease of installation. The voltage is specified for the RMS
voltage, so 230 is the lowest we can use and 250 is the next. With the 230 V
transzorb, the clamp voltage according to the catalog is 700 V at 50 A. Next is
the shape of the wave, and each shape has a K factor. This shape factor, K, is
needed next (Fig. 11.6).
The K factors for the pulses in Fig. 11.7 are as follows: ramp = 0.5, a
constant height = 1.00, and a sine pulse = 0.637. The K factors based on the pulses
in Fig. 11.7 are as follows: dampened sine wave = 0.86; dampened exponential
= 1.4. If the drawing or specification has a ramp followed by the dampened
exponential with the time of 5 s for the full ramp time and a total of 50 s to
the 50% current point of the exponential (Fig. 11.8), the calculation is as follows.
Energy of the ramped wave = KVcIp = 0.5 700 50 106 = 0.087
Energy of the dampened wave = KVcIp = 1.4 700 50 (50 5)
106 = 1.4 700 50 (50 5) 106 = 2.205
where the time is the difference. This gives
Total = 2.292
144
Chapter 11
Ip
I
I
Ip
Tau
Tau
Ip
I
Tau
FIG.11.6
The K factor calculated using the full time of the pulse.
FIG.11.7
Dampened sine wave based on the time to 50% of peak current.
FIG.11.8
145
Exponential based on the time to 50% of peak current.
Multiply this by 1.2 for a safety factor:

J = 2.2925 1.2 = 2.751
and round this to 2.75 joules. The joules are listed as 20 W-s or J, giving good
headroom.
This was calculated using a Harris V230LA4. The typical pulse calculation
is shown in Fig. 11.9.
Ip
I
Tau
FIG.11.9
Tau
Typical pulse calculation.
146
Chapter 11
11.4. THE GAS TUBE

Besides transzorbs and MOVs, there are gas tubes. These handle 25,000 A over
short periods of time. One of the leading companies is Joslyn Electronics, located
in Goleta, California, near Santa Barbara. These are best used in DC systems,
where the DC voltage will stabilize back to a constant voltage soon after the
pulse. The tube will then deionize quickly. In AC systems, because of the
continuous change sinusoidal voltage, the gas tube does not fully deionize.
In some cases, the system must be shut down for a short time to rid the tube of
the ionization.
12
What Will Compromise the Filter?
The filter may test well using the specified test method and still fail to work as
designed for many reasons, the most prevalent of which are discussed here.
As mentioned in an earlier section, some filters are compromised by ground
faults. The case discussed was a balanced filter with the bottom half shunted
around. This reduced the loss by 50%. It was caused by lack of communication
between two different engineering groups, resulting in double the weight, size,
and cost for half the performance. More candidates for failure are discussed here.
12.1. SPECIFICATIONSTESTING
Filters are usually designed to pass the more prominent 220A test with 50 ohms
load and source. An EMI filter was designed for the 220A specification but the
new customer wanted to test using the current injection method described by
Military Standard (MIL STD) 461. The filter failed. This test setup requires two
10-F capacitors to ground on both the supply and return power leads. The filter
was unbalanced with all the components on the supply, or hot, side. The test setup
was detecting common mode noise, and even though common mode was added,
the filter failed the tests, so the filter had to be redesigned.
12.2. POWER SUPPLIES EITHER AS SOURCE OR LOAD
Power supplies are tricky either as the source of supply or as the load. As the
source, most designs are inductive output, not at the DC output, but become
inductive as the frequency increases. From 0 to 10 or 50 Hz, depending on the
148
Chapter 12
supply, the output impedance is in the milliohm range, and it starts climbing after
this and looks inductive. If the EMI filter following the DC supply is capacitive
input, the inductance of the supply and the input filter capacitor can ring.
However, if the EMI filter following the DC supply is inductive input, the stored
energy of the inductor has been known to blow the supply on turn-off. The best
arrangement for this application is to make the first stage of the EMI filter a T.
The inductor facing the supply is half the value of the inductor in the L. A resistor
at the input to ground furnishes a discharge path for the storage capacitor on the
supply side and reduces the voltage rise of the filter inductor. The resistor value
would be of the order of 25 to 100 K ohms.
If the EMI filter is to feed a power supply, low output impedance is required.
This means the preceding solution with the T, cannot be the only filter component.
This really requires an L with a quality capacitor, preferably the feed-through type,
for the output element. The central inductor can be the total of the T inductor plus
the L inductor. This is assuming the input in both cases is DC. In Fig. 12.1, the central
inductor is three times the line-side inductor because the central inductor is really two
inductors. The T has two inductors equal to half the value of the L inductor each.
So the central design value would call for 1.5 the calculated value.
What about AC power supplies? Many companies use AC switcher supplies
to generate AC at other frequencies such as 50 Hz for Europe and 400 Hz. Some
of these generators provide three-phase power outputs. Watch these closely if an
EMI filter design is required; they are usually rich in harmonics. One such 400-Hz
supply had a strong 2400-Hz component and the filter had a resonant rise close
to this. The filter overheated with little or no load current. The filter was not
designed to handle the 2400 Hz because no one thought to tell the filter designer
this fact. However, to be fair, the customer was probably not aware of this either.
12.3. TRANSFORMERS: 9- AND 15-PHASE
AUTOTRANSFORMERS
Autotransformers can cause problems by themselves if the EMI filter requires
substantial low-frequency loss. This occurs when the ratio of the total filter
FIG. 12.1 The input T with the output L.
149
inductance even approaches as little as 2% of the primary inductance. This forms

a voltage divider, so the output voltage is below requirement. Again, the customer
was either not aware the following device was an autotransformer or not aware
of the problem. The 9 and 15 phase often start with an autotransformer yielding
low primary inductance.
The cure for this is to add a filter stage or two to reduce the required total
inductance of the filter. For example, in going from a single stage to two stages,
the values of both the total inductance and capacitance drop drastically. If this
was a single L section with 12 dB per octave, it would change to 24 dB per octave,
reducing all the component values greatly. In addition, raising the capacitance,
allowing lower total inductance in series across the line, may be required.
However, this can play havoc with leakage current specifications.
12.4. NEUTRAL WIRE NOT PART OF THE COMMON
MODE FILTER
In three-phase systems, the currents are always unbalanced. The neutral wire
carries this unbalanced current. This neutral must be part of the common mode
choke. This means that there are four equal windings on the common mode core.
Without this, the ferrite core is driven into saturation because the difference
current is not present to balance the core. In this mode, the ferrite core is heated
and generates noise that masks the actual noise of the load. Therefore, the noise
is sometimes worse with the filter.
The same is true in single-phase systems. Both the hot and return legs are
wound on the common mode core. But in many cases there are grounds ahead of
the core, compromising equal current flow through this core. Now the core is
unbalanced and in saturation, again generating more noise than the load. If this
condition exists and cannot be avoided, remove the ferrite core. It will create more
problems than it can cure.
12.5. TWO OR MORE FILTERS IN CASCADE
THE UNKNOWN CAPACITOR
This happens when more than one filter type follows another (Fig. 12.2). For
example, in a secure room or screen room, a large three-phase filter powers the
entire room. Each power leg and neutral has an insert rated to carry 100 A or
more, and these inserts in turn are enclosed in a larger cabinet. This could be fed
to cabinets in the room to power a full 19-inch rack of equipment through a filter
in each rack, and the individual equipment mounted in the rack is also filtered.
This situation places three filters in tandem or in cascade. If the power line filter
insert feeding the rack required 100 dB at 10 kHz, this may have forced the
designer to cut well below the normal recommended cutoff frequency. This would
150
Chapter 12
FIG. 12.2 Three filters in cascade or tandem.
be done to get the proper attenuation or insertion loss for reasonable cost, size,
and weight. The filter in the bottom of the rack may be a single filter and be
chased by another in the equipment. These filters detune each other, especially
if any, or all, have higher circuit Q values of 2 or better. The higher circuit Q
increases the potential of these filters to oscillate, and this would move the cutoff
frequency farther into the normal passband. The latter problem has been known
to reduce the line voltage to the point that the rack equipment failed to work.
If any of the filters were designed incorrectly initially, this would accentuate
the problem. It would be worse if there were multiple double feeds with other
filters in cascade off the same power line filter. The cascaded capacitors would
total and cause higher line and harmonic currents that would add heat to the
filters. This would increase the number of resonant rises within the filter chain
along with added suck outs or resonant drops. The problems, as mentioned before,
that this situation could create should now be obvious.
12.6. POOR FILTER GROUNDING
A properly designed filter may check out well in the EMI test laboratory or at the
EMI filter design house. The reader may have been in these laboratories and have
seen all the grounding techniques that are necessary to test the filter and system
equipment (Fig. 12.3). The test bench is covered with a sheet of copper that is
well grounded. The equipment, or filter, under test is often C clamped tightly to
the copper sheet. Most filters are designed to be mounted directly to a ground
151
FIG. 12.3 Filter with feed-throughs showing the grounding through the mounting.
plate through input feed-through studs or through the connectors. The filter is
mounted through chassis holes with EMI gaskets used on both sides of the
chassis. The filter is tightened down to ground with the proper nuts and washers.
The gaskets give thousands of ground points with this technique. This also carries
the chassis ground plane through the cutout holes in the chassis. If this ground is
not provided, the filter fails to live up to its dB rating. Some people scream that
the filter is no good even though they were at the laboratory when the filter was
checked. Without a good ground, the filters feed-through capacitors and other
components to ground cannot work (Fig. 12.4). The transzorbs and MOVs tied to
ground cannot function.
It is easy to see that the filter in Fig. 12.4 does not function properly if the
filter case is not grounded or is poorly grounded. The two load-side feed-throughs
would be out of the circuit, along with the two line-side transzorbs to ground. The
two similar capacitors on the input are noncapacitive input terminals or connectors.
12.7. THE FLOATING FILTER
This follows from the topic in Sec. 12.6. The filter was designed according to the
method just described. This filter was to be mounted to ground as described, and
the users complained that the filter was not functioning. Sure enough, the filter
was mounted or hung in air through a plastic hanger. A nice 6-inch green wire
(normal hookup wire and not Litz) ran from one lug on the filter through a cable
152
Chapter 12
C2
LINE
REST OF FILTER AND LOAD
C2
FIG. 12.4 Filter showing lack of a good ground.
harness to a ground. This was not even Litz wire, which could carry most of the
upper frequencies. It was common hookup wire, which acts as a good point to
radiate an H field to the surrounding wires in the same cable form. I was never
able to have these users make a change, although I was able to show some of
them the folly of their ways. They were way behind schedule, with their customer
pressing them to ship, and could not find time to make changes. I think they must
have gotten a release from the EMI specification from their customer.
The green wire just mentioned has the following properties (Fig. 12.5):
1. A skin effect adds to the AC, or RF, resistance, making the ground more
resistive.
HOT
C2
LINE
REST OF FILTER AND LOAD
NEUT
C2
GRND
FIG. 12.5 Drawing of the little green wire.
153
2. A slow velocity of propagation, making the apparent length about eight

times longer, adds inductance to the lead, further impeding the RF
current.
3. From 1 and 2, a series RL circuit is formed.
4. Whatever the RF current in the lead, this would radiate to the surrounding areas and wires.
The green wires inductance would be close to 1.5 H because of the slow
velocity of propagation. This would make the 6 inches almost eight times the
length, or close to 48 inches. The full length would equate to at least a meter, and
the typical inductance is 1.5 H/m.
12.8. UNKNOWN CAPACITOR IN THE
FOLLOWING EQUIPMENT
This is similar to Sec. of 12.5 discussing filters in tandem and applies only to DC
filters. This is also another reason why a filter may be praised by one group and
cursed by another. Granted, this double capacitor may not compromise every
situation in DC filtering, but the additional component adds cost, volume, and
weight (Fig. 12.6). The added capacitor detunes the filter.
12.9. INPUT AND OUTPUT TOO CLOSE TOGETHER
I have seen filters designed by in-house people who put the input and output studs
of the filter on the same front face. Try not to do this because the inputs and
outputs are too close together. One in question was a face that measured 2 3
inches. This means that the filter components that make up the input and output
sections of the filter are too closely spaced. This makes a filter designed for 40 dB
a 26 dB or so filter because it is easier for the RF to radiate from input to output
DOUBLE "L" FILTER

DC
L1
L1
LINE
LOAD
C1
C1
ADDED CAP
FIG. 12.6 The added unknown or hidden capacitor.
154
Chapter 12
or vice versa. I have seen designs where the input and output are in the same
connector and the input and output wires are in the same cable.
The best design is a long filter body with the terminals on the opposite
smaller faces. In this way, the unwanted energy is cleaned up as the signals travel
through the filter sections toward the opposite end. Figure 12.7 shows an input
connector with threads so that EMI gaskets can be used on both sides of the
enclosure material. The internal gasket should be the full width and height, and
the outside gasket should be the same diameter as the mounting washer. The
mounting nut follows this. The output terminals and feed-through capacitors are
on the far end.
If you cannot escape this condition, say in the 2 3 inch section just
mentioned, place a shield between the two halves. Do not allow the harnesses
feeding the filter to carry both feeds. There are two ways to divide the 2 3 area,
and the choice is determined by the component size. The first is to divide the
3-inch width by two and install a shield that runs well to the other end of the filter.
This makes the shield 2 inches high (Fig. 12.8). The input or output section is
now 2 1.5 inches. The input studs and components should be installed on one
side of the filter, and the additional components should be installed toward the
rear and then turn the corner and head toward the front again. If some components
are larger than the compartments formed by the shield, the second method of
using the shield to split the 2-inch height can be used. Continue with the same
technique.
If the components still do not fit, it is best to change the layout so that the
filter can run from end to end without doubling back as in Fig. 12.7.
12.10. GASKETS
EMI filters are normally mounted through small holes in the case, and gaskets are
required to give a very good ground to the filter case (Fig. 12.9). Often, the gasket
is omitted, and the filter has a poor ground return. Any feed-through capacitors
FIG. 12.7 Filter using a connector: length compared to width.
155
FIG. 12.8 Input and output on same filter face.
and MOVs to ground or case within the filter are compromised. Sometimes, the
filter is removed for various reasons or replaced and the same gaskets are reused.
This again reduces the effectiveness of the ground, and the components to case
ground of the filter are less effective. On the other hand, it is better to reuse the
gasket than to be without any gasket.
Another reason to use gaskets is to complete the missing ground path
through the holes through which the filter is to be inserted. The proper way to
mount the filter requires two gaskets: a gasket between the case and filter and the
second set between the outside case and washer(s).
FIG. 12.9 EMI gasket used to install a feed-through input filter.
156
Chapter 12
There is a proper torque for an EMI gasket, supplied by the gasket

manufacturer, that ensures excellent grounding without collapsing the gasket.
This torque gives about the minimum DC and AC resistance between the filter
and ground.
When two objects are tightened down without gaskets or with old, reused
gaskets, few points touch, giving a resistive path. When new gaskets are used,
thousands of points touch, giving a low-impedance path.
13
Waves as Noise Sources
The waves more commonly encountered within systems are discussed here. These
waves are not as pure as drawn here; some parasitic oscillations from the
transformer and other components will be superimposed on the waveform. These
waveforms also have added rise and fall times, some of which are shown as pure
step functions that are impossible to achieve. Each voltage, or current, waveform
is from Fourier equations and is listed along with its differential.
13.1. THE SPIKE

Spikes are one of the common waves, or noise sources, encountered (Fig. 13.1).
They are seen when double-ended, or Royer, circuits are used, with one switcher
turning off while the other is turning on. The two currents add and, as far as the
switchers are concerned, the current nearly doubles. Other elements, such as
diodes, also add to this current, and often the total spike current is many times
the average switcher current. This spike occurs twice per cycle of the switcher
frequency, so the frequency of the spike is double the switcher frequency. Without
some correction in the switcher circuit in a DC system, this high spike current
can greatly reduce the DC voltage. This impairs switcher operation, and most of
this spike energy is passed back to the EMI filter.
With an off-line regulator, the storage capacitor is often very large, in the
500 to 2000 F range or more. If the spike frequency is 60 kHz, twice the 30 kHz
switcher frequency, this large storage capacitor may be well above its selfresonant frequency (SRF) and be either open or inductive. If this section is not
isolated at this point, a good-quality feed-through capacitor should be tied in
158
Chapter 13
a
FIG. 13.1 The spike.
parallel here. If this section is isolated, a good line-to-line capacitor (X capacitor)

should be tied across the storage capacitor. Another well-placed capacitor should
be used at the junction of the B+ and the switcher transformer. This should be a
good quality feed-through if the switcher is not isolated or a good-quality line to
line if it is isolated.
An easy way to calculate this feed-through or line-to-line capacitor is as
follows. Estimate the lower DC storage voltage, and divide this by twice the
normal switcher current (about the peak of the spike current) for the worst-case,
or lowest, working impedance. The line-to-line added capacitor impedance
should be 0.1 this impedance at the spike frequency, 60 kHz here. The leads of
the line-to-line capacitor should be as short as possible. Keep this capacitors SRF
well above 10 times the spike frequencyagain 60 kHz hereor 600 kHz.
Round the calculated value up to the next higher standard value of capacitance.
This is a pulse, and it requires the ability to handle the higher pulse currents.
Therefore foil, not metallized film, should work with the proper derating. Ceramic
capacitors work very well here. The leads must be short because articles from the
Institute of Electrical and Electronics Engineers (IEEE) state that a capacitor with
the same lead length will not give the higher SRF required. The equation of the
spike is as follows:
Ea
Ea
+
T
T
sin (X)
2Nt
X cos T
N = 2,4,6,8
where a is the pulse width and
(13.1)
X =
159
2Na
T
and the differential is:

E
2a
n = odd
1 sin (X)
2nt
sin
n X
T
(13.2)
13.2. THE PULSE

The pulse is similar to the quasi-square but pulses in the same direction, like the
spike (Fig. 13.2). This is similar to the Royer, where the dwell time for both halves
is not on for the full half-period. Therefore, this generates pulses of current twice
per period, once for each half, and is again at twice the switcher frequency. The
design and considerations are the same as for the spike.
The equation of the pulse is:
2E
Ea
+
N = 2,4,6,8
1
Na
cos (Nt) sin
N
T
(13.3)
where a is the pulse width. The differential is

4E
T
Na
sin (Nt) sin
T
N = 2,4,6,8
a
FIG. 13.2 The pulse.
(13.4)
160
Chapter 13
13.3. THE TRAPEZOID

The trapezoid (Fig. 13.3) is more realistic than the pulse or the quasi-square
because both have a step function or zero rise time and fall time. The trapezoid
has a rise and fall time like that in the real world. Also, the energy in the EMI
spectrum is less with the trapezoid. The disadvantage of the trapezoid is that the
efficiency of the power supply is reduced. More power is dissipated in the switch
during the rise and fall time.
A capacitor, found by the method shown for the spike, can be paralleled
with the main storage capacitor. Follow the same method; T0 is the pulse width
and Tr is the rise and fall time. The equation of the trapezoid is
2E(T0 + TR)
T
nTR/T
sin [n(T0 + TR/T]
2nt
sin
cos
(13.5)
+
T
)/T
/T
n(T
nT
0
R
R
T
n = odd
The differential is
4E
n(T0 + TR
nTR/T
2nt
sin
sin
sin
T
(nTR/T)
n = odd
To
Tr
FIG. 13.3 The trapezoid.
Tr
(13.6)
161
13.4. THE QUASI-SQUARE

The quasi-square is the wave applied to the gate, or control, of the switch, and the
reciprocal is applied to the opposite control (Fig. 13.4). The pulse width is a and
the dwell angle is one half the angle between the pulses. The dwell angle occurs
before and after each pulse, or there are four dwells per period. This turns on the
opposite device every half-cycle, and each is turned on for less than a half-cycle.
The output is the current pulse, and the pulse is twice the quasi-square frequency.
The pulse is filtered by the technique described in Sec. 13.1 for the spike.
This quasi-square wave and its output wave, the pulse, are not real-world
waves. Both lack the rise and fall time that any wave has. This is due to small
capacitors and inductors in the path. All devices such as circuit boards, the natural
capacitance of wiring, and the input capacitance of the switch have properties that
require time to discharge or change. This results in ramps in the wave shape,
changing the so-called square, quasi-square, and pulse into trapezoids that show
the rise and fall time. The quasi-square wave should look more like Fig. 13.5.
The equation of the quasi-square wave is
4E
N = 1,3,5,7
1
sin (nt) cos (N)
N
(13.7)
cos (Nt) cos (N)
(13.8)
The differential is
8E
T
N = 1,3,5,7
THETA
FIG. 13.4 The quasi-square.
162
Chapter 13
THETA
FIG. 13.5 The quasi-square with rise and fall times.
13.5. WHY DIFFERENTIATE?

Each pulse listed up to now has been differentiated for good reason. If each of
these waves represents current pulses, the H field needs to be calculated. If each
wave is only a small change in the DC current state, the radiated H field is small.
A steady-state DC current generates only a steady-state H field. This H field cuts
the surrounding surfaces only during turn-on and turn-off. These waves are often
pulses with peak amplitudes of many amperes. They generate strong varying H
fields, leading to radiated energy (Fig. 13.6). The formula is
0.4NIP
Mpl
(13.9)
where N is the number of turns, Ip is the peak current, and Mpl is the magnetic
path length. If this is a wire carrying this Ip current, the turns are then equal to
one and Mpl is the circumference. The radius is from the wire to the surface in
question (usually to the closest outer metal cover or another wire). If the length
between the wire and cover, or the part to which this is radiating is Rr, then this
equation is reduced to
0.2IP
RR
(13.10)
What is needed is the differential of H with respect to t, and the only

variable that depends on t in Eq. (13.10) is Ip. Therefore, dH/dt is equal to
163
conducting wire
Rr
Magnetic field
FIG. 13.6 The radiated field.
0.2 dIP
RR dt
(13.11)
So, 0.2 divided by RR, n = 1, times the differential from the proper wave in
question should give the approximate magnetic field intensity in amperes per
meter if RR is in meters.
13.6. THE POWER SPECTRUMdB A/MHz
How much power is in the various waves discussed in the earlier sections? These
form envelopes that provide the peak power, which varies with the amplitude of
the current, Ip. If the current pulse width is a and T is the period, the equation for
the pulse power dB/MHz is as follows (Fig. 13.7):
2Ea 106
20 log
(13.12)
where E is the amplitude. This gives a flat line across the frequency spectrum to
164
Chapter 13
FIG. 13.7 The dB A per megacycle.
the 20 dB per decade, or 6 dB per octave, breakpoint. This point starts at the
frequency
1
Hz
a
(13.13)
where a is the pulse width in seconds and the decibel level after the breakpoint is
1
20 log
dB
Fa
(13.14)
There is also a 40 dB per decade breakpoint, 12 dB per octave, which depends

on the rise time. The larger the rise time, the sooner the breakpoint occurs. The
point here is that EMI energy can be greatly reduced by using this principle. The
efficiency of the power supply decreases but the EMI energy is less. This is
usually a better trade-off for smaller and lighter units and should reduce the total
cost. If the rise time is 10% or better, this becomes, in essence, an L filter added
to the existing EMI filter at frequencies above the 40 dB breakpoint.
1
Hz
The added loss after the 40 dB per decade breakpoint is
(13.15)
1
40 log 2 2 dB
F
165
(13.16)
where F is in hertz and is in seconds.

13.7. MIL STD 461 CURVE
This specification is in dB A/MHz rather than insertion loss (Fig. 13.8). It is
rather difficult to convert from this to insertion loss. If the impedances were equal,
the conversion would be the obvious 120 dB. Insertion loss is stated in terms of
how much loss is required, whereas dB A/MHz, or dB V/MHz, is in terms of
how much noise is allowed. There is a 13 dB conversion factor due to changing
from 50 ohms to a probe estimated to be 1 ohm. This subtracts from the 120,
giving 107 dB, and then the probe correction factor is needed. This is a function
of the probe, the frequency bandwidth. This last correction is added back to the
107 dB. Usually, this method comes with the probe.
FIG. 13.8 MIL STD 461 C specification. Upper curve, Navy, Air Force; bottom
curve, Army.
14
Study of the Off-Line Regulator
Col. W. T. McLyman (Jet Propulsion Laboratory, National Aeronautics and Space

Administration; now retired) has worked out the approximate weight and volume
relationships using the area product, Ap. These depend on the type of core and the
winding technique. The purpose here is to show that the entire weight and volume
are reduced if an inductor of critical value is included in the linear power supply.
This inductor is placed between the diodes and storage capacitor. This mainly
applies to filters that have serious loss requirements at very low frequencies such
as specified in Military Standard (MIL STD) 461. Years ago, these were called
inductive input power supplies. This also applies to power factor correction
circuits that remove the high current spikes required to charge the storage
capacitors twice per cycle. Regardless of which technique is used, the size and
weight of the EMI filter are reduced, especially for 461 and similar specifications.
Also discussed in this chapter is what the high harmonic content does to the
line voltage, which affects other equipment operating on the line. In the last
section, the Keith Williams method is discussed, which is used primarily throughout this chapter and the book. The following section uses the same equations to
show that the added harmonic content on the power lines is increasing. This
happens today because most of the electronic equipment making up the load does
not employ power factor correction circuits or any other technology to remedy
this problem. Another reason for the higher harmonic content is the lack of the
inductive input technology discussed in Sec. 14.1. This condition will improve as
the loads meet the newer standards requiring power factor correction circuits, but
the equipment making up the load will take years to replace.
168
Chapter 14
FIG. 14.1 Power supply with capacitive load without critical inductor.
14.1. WITH OR WITHOUT CRITICAL VALUE

OF INDUCTANCE
SIZE AND WEIGHT
DIFFERENCE OF THE FILTER
In referring to this section, many have said that they know that the critical
inductor will make the EMI filter smaller, yet rarely is this application seen in
practice. In other words, the weight and volume of the EMI filter without the
inductor are much greater than those of the filter and inductor together using the
critical inductor (Figs. 14.1 and 14.2). This is especially true if the line frequency
is 400 Hz. The reason is seen in Eq. (14.1), where F is in the denominator. The
inductor Lc is much smaller. From another standpoint, this is true if the EMI filter
specification demands high attenuation at very low frequencies, as required by
the 461 specifications (Fig. 14.3). This may not be a good trade-off if the only
goal is to meet Federal Communications Commission (FCC) and similar specifications, especially at 50 or 60 Hz. If the dB loss of the filter must be 20 dB, or
greater at 10 kHz, this should be looked into regardless of the line frequency.
The formula for the critical inductance, Lc, is
FIG. 14.2 Power supply with critical inductor, L1.
169
FIG. 14.3 Simplified power factor correction circuit.
Lc =
R0
6F
(14.1)
The value of R0 is the highest DC supply voltage divided by the lowest load
current, and F is the line frequency, not the ripple frequency. The addition of the
inductor lowers the storage capacitor average voltage, but the power supplied to
the switcher(s) must be the same. This raises the current minimum, decreasing
the value of R0 and, therefore, Lc. This subject is covered in W. T. McLyman,
Transformer and Inductor Design Handbook (Marcel Dekker). This is handy if
the cores for the Lc inductor and the filter are not known. Otherwise, the weight
and size are known.
If the area product is known in centimeters to the fourth power, the
approximate weight, in grams, and the volume, in cubic centimeters, can be
estimated from Figs. 14.4 and 14.5. These have been assembled from McLymans
book. The Kv factor multiplied by the Apvalue gives cubic centimeters, and Kw
multiplied by Ap gives grams. This is useful if the inductor has not been fully
designed. If the inductance is known, the energy can be calculated from Eq.
(14.2). From this, Ap can be found from Eq. (14.3) knowing the core type. From
this and Figs. 14.4 and 14.5, the volume and weight can be estimated.
Laminations and tape-wound cores are rarely used for EMI filters but are
included in Figs. 14.4 and 14.5. The lamination information is needed for the
critical inductance in some applications. The area product is the cross-sectional
area of the core in square centimeters times the wire area or window area, also in
square centimeters. The Ap must be multiplied by the stacking factor, if applicable
(laminations and C cores). Continuing with McLymans book, the value of the
critical inductance must be found from Eq. (14.1). The next step is to find the
energy stored in the inductor:
Lc(IP)2
2
(14.2)
170
FIG. 14.4 Ratios of area product (Ap) to weight and volume.
FIG. 14.5 Value of Kj for the various core types.
Chapter 14
171
where Ip is the peak current the inductor must carry. The smallest load current
determines the value of Lc, but the peak of the largest charging current of the
storage capacitor, Ip, through the inductor determines the energy that the inductor
must handle. The minimum area product, AP, required to handle the energy is
then, according to McLyman,
[2(energy) 104]X
BMKuKj
(14.3)
Here Bm is the maximum flux density in teslas (gauss/10,000), Ku is the winding

factor for the type of core configuration chosen (typically 0.4 for most configurations), and Kj (C) comes from Fig. 14.5 (from McLymans core configuration
table on page 106 of the book cited earlier).
As an example from McLyman, figure the powder cores at 20 A with 100
H. For the powder core, Kv is 13.1, Kw is 58.8, and X is 1.14. The energy is 0.02.
From the energy and the Kj value of 403, AP can be calculated as 19.13.
Multiplying by Kv and Kw yields 2.48 pounds and 15.29 cubic inches. The design
should be for 25C to reduce the temperature rise of the entire filter, and this may
be good advice for the critical inductor too. The saving of volume and weight
would be greater if the 50C temperature rise can be tolerated for the critical
inductor because Kj is in the denominator, that is, if the surrounding circuits can
withstand it. X is repeated here for convenience.
If the current pulse equations, from Sec. 4.2, for charging the storage
capacitor, without the critical inductor, are solved for the approximate peak
current, the energy can be found. These are repeated here.
(A B)
sin(Y + X B) = sin(Y + X A)e tan X

sin(A) = sin(B)e
(B A N)
CRL
(14.4)
where
Y = tan1(CRL)
CRSRL
X = tan1
RS + RL
(14.5)
Equations (14.4) and (14.5) assume the engineer knows the following: the
line frequency, F, so that (2F) can be calculated; C, the storage capacitor, in
farads; RS, the line resistance (typically 1 ohm); and RL, the load resistance (the
lowest value of resistance, which is the highest current). The angles are in radians,
but degrees are given here for the most part. The values of X and Y are then
calculated, along with tan(X). Substitute these into the first two equations. In one
equation, guess a value for A and solve for B. Insert B into the second and solve
for A. Average the two values for A, and restart the process by substituting this
172
Chapter 14
new value into the first equation again. A spreadsheet is handy for this, with each
row being a new estimate. If the engineer is going to write a spreadsheet of this
type, it is advisable to multiply the initial guess by 49, add the calculated value
to it, and then divide by 50. Do this until the values are similar, and then multiply
by 9 and divide by 10. The reason is that the changes may be too abrupt and throw
the value of A into the wrong court. See the Appendix and the disk. Once A, the
start angle, and B, the stop angle, are known, U1, the first minimum voltage angle
guess; V1, the first maximum voltage angle guess; and P1, the first peak current
angle guess, can be estimated.
(Rs + Rl)
U1 = sin1
sin(A)
R
l
(R
+
R
)
s
l
V1 = sin1
sin(B)
Rl
P1 =
(14.6)
(A + B)
2
Then U, V, and P can be solved through iteration. Use similar techniques

for the averaging, as before, otherwise the changes may be too great. Again, a
spreadsheet is handy and all can be estimated within the same spreadsheet.
Substitute the known values of A and B and then the estimated U1, V1, and P1
into one term and solve for the other U, V, and P. Average the two and solve
again. A spreadsheet is included on the disk.
(Rs + Rl) sin(Y + X A)e
cos(X U) =
sin(Y)
(A U)
tan X
(A V)
(Rs + Rl) sin(Y + X A)e tan (X)

cos(V X) =
sin(Y)
(14.7)
(A P)
cos(Y + X P)
= e tan (X)
cos(Y + X A)
Then EU, the minimum voltage at angle U (in radians), and EV, the maximum
voltage at angle V (again in radians), can be found.
EU =
EMRl sin(U)
Rs + Rl
EV =
EMRl sin(V)
Rs + Rl
(14.8)
where EM is the peak line input voltage (or the stepped-up or stepped-down peak
voltage from a transformer). Also, the peak current, at angle P, can be calculated.
This peak current is the value required for this section. The beauty of this is that
173
the appendix program will solve any of these for you with other useful information such as the power factor and peak-to-peak ripple.
As an example, EM is 167 V, F is 400 Hz, Rs is 1 ohm, and Rl is 10 ohms.
The storage capacitor is 0.0002 F (200 F), and N is 1 for a full-wave rectifier.
It follows that is 2513.3. So both X and Y can be solved; X is 0.428626 and Y
is 1.767176, and their sum is 2.195802. The initial guesses for A and B are /3
and /1.6, respectfully. Through iteration, A and B can be determined using Eqs.
(14.4). The start angle, A, is 36.98775 degrees and the stop angle, B, is 123.7759
degrees. This accuracy is not required but is given here to show the values
calculated. The initial guess for the peak current angle, P, is the average of A and
B, which is 80.35581 degrees. Of course, these are all calculated with radians, but
I give the terms in degrees for clarity. The final value for the peak current angle
is 71.55862 degrees. With A, B, Y, P, X, EM, Rs, and Rl known, substitute into Eq.
(14.9) for the value of IM. From the appendix equations and the disk provided,
the peak-to-peak ripple is 42.2 V, the valley voltage is 99.78, the fundamental
current is 16 A and the leading angle is 13.33 degrees. However, the total
harmonic RMS current is 19.26 A. The power factor is 0.81854.
(A P)
IM
sin(P Y X) + sin(Y + X A)e tan (X)

= EM cos(X)
(Rs + Rl) cos(Y)
(14.9)
The peak current (P) is 38.54 A. Again, this is the value needed to design
the filter inductor. The inductor must not saturate at this peak current of 38 A.
This is not the worst situation, where the RMS current of 16 A is compared with
the peak of 38 A. The reason is that this system is approximately 1500 W. The
conduction angle is 86.73 degrees, so the peak current is low. If Rs was 20 ohms,
the conduction angle would be less, the peak would be 26.93 A, the peak-to-peak
ripple would have been less, and so on. But the ratio of the peak current to the
RMS would be much greater. In the initial problem, the EMI filter designer would
be given the 16 A value and the inductors would saturate.
If, in our example, the filter required several 100-F inductors for the 16
A, two MPP 55716s taped together with 27 turns with three strands of AWG 15
un2 would give the desired results. The flux density would be 3348 gauss with
an approximate temperature rise of 16.2C. It would be well below saturation if
this was a true sinusoidal wave at 400 Hz. This wave is neither sinusoidal nor
pulse but has properties of each. However, for the total harmonic current of 19
A, the flux density is still only 3976 gauss with 21.6C. The flux density is a little
high but is acceptable for a 50, 60, or 400 Hz sinusoidal wave. However, 38 A is
well above saturation at 7951 gauss with a rise of 68.7C and would be calculated
if the wave was a pulse. This temperature rise is not realistic because this is a
peak current, not RMS. The core would saturate but not overheat. The proper way
174
Chapter 14
to avoid this is to determine the true total RMS, using the equations provided here
or on the CD. Add a little headroom; here, go to 20 A. Keep the flux density well
below 50% saturation. For an MPP, saturation is 7000 gauss. Strive for a flux
density of 3000 gauss or below. Going to the next size bigger core, MPP 55111,
and using two cores taped together, as before, with three strands of AWG 14 with
39 turns will meet the inductance requirement. The flux density is 2509 and the
temperature rise is 24C. Now the 38 A peak should not drive the core into
saturation. However, the wound diameter is 2.5 and the height is 2 inches. The
volume consumed is 2.5 2.5 2 = 12.5, which is not too far from the McLyman
estimate of 15.29. These cores would be expensive, especially if several of these
inductors were needed. The powder core program on the CD calculated the MPP
core used here.
It should be apparent that a different core must be used. This calls for a
switch to a C core or tape wound using the same steel. Again, this program is on
the CD. The C core style is a 4 mil, as normally specified for 400 Hz. This core
gives adequate Q and lower core losses. Better yet, use a 2-mil core for better Q
and better tuning, if required. But this change adds to the cost. The core chosen
is CH-44. Two coil forms are wound with two AWG 15 un2 wires with 32 turns.
The coils are in parallel on each leg, and there are three layers on each coil form.
The required gap is 40 mils or 20 mils per leg. The flux density is less than 10.4
kG and the temperature rise is 30C. This is based on the necessary 20 A. The
approximate size is 2.3 1 1.75, and with one core per inductor it should be
less expensive than the two stacked MPP cores.
If the designer placed an inductor (Lc) between the diodes and the storage
capacitor of 1.33 MH, the filter inductors would drop in value and could go to
almost 4000 gauss. However, the 1.33 MH inductor requires some headroom,
especially if the load current will drop below the value needed or if the value of
R0 increases well above 10 ohms. If the load is reasonably constant, 2 MH should
do the job. See Eq. (14.1) and Fig. 14.2. This would create a constant current of
12.13 A, further reducing the size and cost of the filter cores. Lc would require a
4-mil CH-61 core with two coils in parallel with 75 turns each of AWG 14.
The gap is 36 mils with 18 mils in each leg and the temperature rise is 25C.
This change allowed one core for each filter inductor instead of two as before.
Also, the core was changed to MPP 55110, which has a higher Al value and
is less expensive. There are 37 turns with three strands of AWG 14. The temperature rise is 10C.
With Lc being primarily a DC inductor and a constant current of 12.13 A,
the storage capacitor can also be reduced to at least half value. The current leaving
the capacitor is the same as that charging the capacitor. This gives a further
175
reduction of cost, size, and weight of the entire power supply and EMI filter. The
current in the filter is a square wave with a peak of 12.13 A.
14.2. THE ADDED POWER LINE HARMONIC CONTENT
CAUSED BY THE OFF-LINE REGULATOR
Continuing with the system in Sec. 14.1, 1500 W can be powered by normal wall
plugs (i.e., a regular wall outlet) and does not require special wiring from the
service. However, the total harmonic current is over 19 A. Now the system leans
toward a dedicated line. Adding the fact that the peaks go to over 38 A, the line
is definitely a dedicated line with its own circuit breaker. If the equipment is listed
as 120 V, 12.5 A and is now wall fed, the odds are that the breaker will trip. The
question is, what is the total resistance from the input of the power supply, diodes,
and transformer (if any) through the filter, the line, and the breaker? In the power
supply equation, Rs was 1 ohm. At the peak current instant, the voltage drop is
38.54 V (1 38.54), and this is at 71 degrees. The sin of 71 degrees is 0.9455.
This times the peak of 167 is 157.90. Now subtracting the drop at this angle gives
119.36 V.
The point here is that this type of equipment is on the increase. Not all gulp
a peak current of 38 A and not all are at 71 degrees. In 1985 and before, 85% of
the total power was for lights and power for motors for refrigerators, lathes, and
so on. Only 15% was for radios, televisions, copiers, and some computers. Now
homes have fax machines, computers, several TV sets, and other devices requiring some peak current well above RMS. The medical field with magnetic
resonance imaging (MRI) equipment, x-ray equipment, heart monitors, and so on
all adds to the voltage distortion ahead of 90 and 270 degrees. So the voltage is
no longer just the fundamental frequency. The same is happening to the threephase systems with so much odd-order harmonics; the third harmonics, and their
multiples, are heating the neutral and the neutral filters because of the unbalance
currents plus the third order. These add back in phase (120 degrees times 3, 6, 9,
12, 15, etc.; all are multiples of 360 degrees or 0 degrees) on the neutral. The
people at RFI have received many calls asking why the neutral filter is so hot.
They are asked to measure the currents on the legs. The current is then reported
to be greater than the phase currents. The next question asked is whether RFI can
design a filter to correct this to remove the third-order currents. When the
customers are told the size, weight, and approximate cost, they look for other
solutions. Some of these solutions are to rebalance the loads, change to power
factor correction type power supplies where possible, and increase the size of the
neutral to handle the current required.
The best solution is to use power factor type front ends, but this is hard to
176
Chapter 14
do at higher powers. The critical inductor (Lc) also helps, especially at nearly
constant current requirements. The inductor is the smallest possible with this
arrangement.
14.3. KEITH WILLIAMS METHOD
Keith Williams works for Grand Transformers in Grand Haven, Michigan, and
well known throughout the magnetic community. He developed these equations
from O. H. Schade graphs developed in the mid-1940s. These equations are
complex trigonometric transcendental functions and can be solved only through
iteration. In other words, these equations cannot be solved by algebraic means.
Keith Williams has a program that solves these complex equations.
15
Initial Filter Design Requirements
The goals of this chapter are as follows:

1.
2.
3.
4.
Differential mode design goals

Common mode design goals
Methods for reducing the size of the inductor related to inductor current
Estimate of common mode load impedance
This chapter discusses these goals and what can happen if they are not met.
15.1. DIFFERENTIAL MODE DESIGN GOALS
The idea is to make the differential mode filter transparent to the line. This is ideal
for both DC and AC systems and easy to accomplish at DC, 50 and 60 Hz.
However, it is a demanding task at 400 Hz in any system requiring substantial
loss at low kilohertz frequencies. The requirement stated really means that the
load impedance is transferred to the input of the filter at the line frequency and
most of its lower harmonics. The harmonic content depends on the quality of the
line and the load. The higher the line impedance, the more effect the odd-order
harmonic content has in distorting the sine wave voltage shape. These harmonics
are odd harmonics. The filter cutoff frequency to accomplish this goal should be
above the 15th harmonic because the level of any harmonic above the 15th, even
for the poorest quality line and load, is insignificant. This is why the rule is set
at this harmonic of the line frequency. This is easy to do for 50 and 60 Hz, and
the cutoff will be well above this goal. The problem is 400 Hz. Actually, the 15th
harmonic is not high enough for 400 Hz because of the resonant voltage rise at
178
Chapter 15
FIG. 15.1 Resonant rise at 2252 Hz: effect at 400 Hz.
400 Hz. Try to get the cutoff as close as possible to 8 kHz. However, this requires
too many stages. In the past, the cutoff was formulated at the 10th harmonic, but
this did not take 400 Hz into account.
In Fig. 15.1, the cutoff frequency is well above 4000 Hz but has a serious
resonant rise at 2252 Hz. This gives a 0.281 dB gain at 400 Hz, which translates
to a 3.3% voltage rise. But this is calculated for ideal conditions, and the rise is
much more than 3.3% in operation. Note that this condition is not a problem at
50 or 60 Hz. The dB gain is zero. Actually, this filter was designed for 60 Hz, so
there would be no problems. There is insufficient power at these 60-Hz harmonics
to cause problems in the 2250 area.
15.2. INPUT IMPEDANCE OF THE DIFFERENTIAL
MODE FILTER
The equation is simply
Zin at NH = Rload
(15.1)
where Zin is the input impedance of the filter, NH is the harmonic number and is
set equal to 15, and Rload is the load impedance at the same frequency. Practically
speaking, the load impedance is constant over the frequency range of interest in
this discussion.
This goal is hard to reach, especially for the high-current filters with which
the required loss is heavy at a low-frequency end. The filter designer should strive
179
to reach it. If the cutoff frequency allows the filter to attenuate the lower
harmonics, higher capacitor currents result. These heat the capacitors because of
the current through the equivalent series resistance (ESR). The low-frequency
cutoff also increases the harmonic currents through the inductors and increases
their heat through the inductor DC resistance (DCR) and higher core losses. This
raises the operating temperature of the filter. These filters are normally power line
type and are often too hot now without this aspect. This low cutoff frequency
would lower the resonant rise frequency and raise the circuit Q. Either one of
these facts would increase the odds that the filter will oscillate as well as overheat.
15.3. OUTPUT IMPEDANCE OF THE DIFFERENTIAL
MODE FILTER
This section follows the same logic as for the input impedance of the filter. The
filter should be transparent to the load. If the input impedance goal is met, the
output impedance goal is normally met. Meeting these two goals makes for better
filter operation.
Z0 at NH = Rs
(15.2)
where Z0 is the filter output impedance and Rs is the line impedance at the same
frequency. The other terms are the same. The line impedance here is the basic DC
resistance of the line. This holds true for most lines to 5 kHz before any rapid
increase to higher line impedances is reached. At 10 kHz the impedance is about
4 ohms on most lines.
15.4. INPUT AND OUTPUT IMPEDANCE FOR A DC FILTER
Both requirements of the two preceding sections are easily met for a DC system
unless the load is a switcher. Here, the output impedance of the filter must be very
low at and above the switcher frequency. This statement rests on the premise that
the switcher designer has not anticipated this and corrected for it up front, but it
is still something that should be done.
Z0 at Fsw << Rl at Fsw
(15.3)
where Fsw is the switch frequency and the rest of the terms are the same. The
same holds true at the 10th harmonic of the switcher frequency. The switch may
not be all that starved at the fundamental and yet be starved at the 9th or 11th
harmonic if the output impedance is slightly inductive or the output capacitor is
above its self-resonant frequency (SRF). Little power is carried by the higher
frequencies above the ninth harmonic, so these may be attenuated.
180
Chapter 15
If this goal is not met, the switch is starved. The output impedance of the
filter should be of the order of 0.1 of the load impedance during conduction or
on time. This is also a function of the pulse width. The main goal is to make sure
that the drop is not excessive so that the switcher can function properly. As an
example, Fsw = 80 kHz, Vs = 28 V, and Ion = 8 A during the on time.
The on impedance followed by the required capacitor impedance and then
followed by the capacitor value is
28
= 3.5
8
3.5
= 0.35
10
C =
1
= 5.68 F
280,000 0.35
(15.4)
The output impedance of the filter at 80 kHz to meet the requirement is, from
Eq. (15.4), 0.35, and the capacitor value is 5.68 F. Use a high-quality capacitor
close to this value. If the pulse duration is 5 s and the period, for 80 kHz, is 12.5
with one pulse per cycle, the value of the capacitor can be lowered. In other
words, the capacitor must support the pulse for 5 s. Dividing 5 by 12.5 gives
0.4. Multiplying 0.4 by the capacitor value of 5.68 F gives 2.272 F, which is
rounded up to 2.5 F. Again, this must be a good quality wrap and fill, ceramic,
and would be better if a feed-through type. Also, this would require a very good
quality capacitor with a self-resonant frequency above at least the 10th harmonic
of the switcher frequency, 800 kHz in the example.
15.5. COMMON MODE DESIGN GOALS
The common mode does not have to meet any of the requirements discussed for
the differential mode. The cutoff frequency can be as low as desired, cutting well
into power harmonic frequencies. Watch for leakage inductance in the Zorro
inductor(s). This generates a high amount of differential mode inductance and
can saturate the inductors at the current peaks. Some state that because the
leakage inductance is in air, it cannot saturate. The answer is that the leakage is
not all just in air.
The disadvantage is that the common mode grows to very large sizes as the
cutoff frequency is lowered, but there is no lower frequency limit for band-pass
or other reasons. The real limit is set by the current rating and size of the common
mode inductor. These inductors should be designed to have little effect on power
factor correction circuits, switchers, or any other load. Again, this assumes little
differential mode inductance within the common mode inductor. The typical
amount is 1 to 2% and can be of the order of the differential inductance. Common
mode inductors are often in the range of 10 MH. At 2%, this equates to 200 H,
which is often more than the differential inductor value. Therefore, the common
mode often helps with the differential losses, but the reverse is also true. Some
companies do this on purpose by separating the windings and thereby reducing
181
the coupling. The ferrite toroid, in which the common mode windings are each
distributed over half the core, have this characteristic.
15.6. ESTIMATE OF COMMON MODE LOAD IMPEDANCE
If the DC system is balanced, using the hot wire and a return wire, common mode
inductors can be used without hindering the DC load. This assumes that the
differential mode properties are low.
The common mode impedance of the typical nonisolated circuit in Fig.
15.2, using the switcher as the common mode noise source, is as follows: G (the
switcher etc.) is now the noise source, and the circuit is drawn backward from
that in Fig. 15.2. This keeps the format used throughout this book of having the
source, or line, to the left and the load to the right in each drawing.
The storage capacitor is out of the circuit by now (Fig. 15.3), depending
on the size and the frequency of interest. The source could be the parasitic
oscillation from an inductor at some frequency above 300 kHz. The diodes and
the transformer look capacitive, and the switcher inductor remains the same.
The current probe measures the noise. The circuit looks something like that
shown in Fig. 15.4.
Figure 15.4 shows the four capacitors across the diodes and the output
transformer capacitor. If the transformer has a Faraday screen, the final capacitor
is even smaller. All of this makes the common mode circuit impedance much
higher than the differential mode impedance (Fig. 15.5).
Also, the circuit impedance is greater than the current probe until the upper
frequencies are reached. If the circuit lacks the transformer, the diode capacitance
to ground is still much higher. Figure 15.6 shows that the common mode design
impedance is higher than the differential mode impedance in most applications.
This information allows the following technique. Calculate the cutoff frequency
as before, and then determine the inductor and capacitor values with the same
equations.
FIG. 15.2 The switcher as the noise source.
182
CIP
SW
Source
FIG. 15.3 Capacitors around the power supply components.

Chapter 15
183
Sw noise source
FIG. 15.4 Power supply components reduced to capacitors.
L =
Rd
2F0
C =
1
2F0Rd
where Rd is the design impedance, which is the same as for the differential mode,
and F0 is the cutoff frequency. If there is a leakage current specification, calculate
the capacitor value of the capacitor to ground, or the specification may state the
maximum value of capacitance to ground. Divide the needed value of capacitance
by the maximum value, and multiply the inductor by this value. The cutoff
frequency remains the same, but the impedance grows by the multiplier.
CIP
Sw noise source
FIG. 15.5 Final switcher reduced to capacitors.
184
Chapter 15
CIP
Sw noise source
FIG. 15.6 Simplified common mode load impedance.
As an example, the 400 Hz maximum to ground per line is 0.02 F. In a

single-phase circuit, two of these would be in parallel for the common mode
application. With the design impedance of 10 ohms and a cutoff frequency of
3350 Hz, the common mode inductor would be 475 H and the capacitor 4.75
F using the preceding equations. This is much too big to meet the leakage current
specification! An easier way to calculate the capacitor would be to divide the
inductance value calculated by the square of the impedance, 10, or 100 here.
Divide this value of capacitor by the maximum value of 0.04 F, the maximum
allowed value of the capacitor to groundthe total of 0.02 times the two lines.
This gives 118.8, so the inductor is multiplied by this value, which equals 56.4
MH. The value of inductance is excessive at almost any current, and the common
mode should be split into two sections. The value of F0 jumps and makes the two
common mode inductors fall to reasonable values of inductance. This reduces the
size, weight, and cost and results in a better self-resonant frequency. The
maximum capacitance to ground for each line section drops to 0.01, making the
two in parallel 0.02 F. This technique of changing the values should not be used
in differential mode.
15.7. METHODS OF REDUCING THE SIZE OF THE
INDUCTOR DUE TO INDUCTOR CURRENT
Some of these filters may require high current. One way to help the design is to
balance the circuit. Here, half of the inductor is in the hot and the other half is in
185
the return. Cutting the value in half drastically improves the ability to design this
inductor with a low temperature rise, reasonable flux levels, and, possibly, a
reasonable cost. Other approaches with C cores use parallel windings. Each arm,
or side, is wound to carry half the current. This reduces the wire size, so the wire
is easier to wind. Where the system allows, using larger capacitors lowers the
inductance, which also helps to ease the inductor current problem. Make sure
the proper ratios are not seriously violated, otherwise a resonant rise will occur
at the line frequency.
16
MatricesReview of A Matrices
Matrices of various forms exist. Besides the A matrix, Z, Y, H, G, B, and now

the scatter parameter (S) are used. The advantage of the A, or chain, matrix is
that each element of a circuit can be chained, or in tandem, as the elements of
the circuit appear, as in the circuit in Fig. 16.1. R1 and R3 are series elements,
and R2 and R4 are shunt elements. Each of these resistors can make up one
complete matrix and each matrix can be multiplied, or chained, by the following
element. Continue through each following element as they appear in the circuit,
such as here.
R1 R2 R3 R4

Each matrix is formed by four elements, such as
A B
C D
FIG. 16.1 Matrix series and shunt elements.
188
Chapter 16
The series elements, R1 and R3, make up the matrix element B:

1 R1 1 R3
0 1 0 1
and the shunt elements make up the matrix element C but are the reciprocals of
their values:
1 0 1 0
1 1
1
1
R2 R4
These four matrices make up the following matrix equation:
Vin
1 R1 1 0 1 R3 1 0 Vo

=
1
1
1 0 1
1 Io
Iin
0 1

R2
R4
The matrix multiplication of the first two matrices and the last two together yields
(for matrix multiplication, see Sec. 16.1)
R3
R1

1 + R R1 1 + R R3 Vo
Vin
2
4

=
1
1
I
I
1
1
R2
R4
o
in

If all the Rs are equal, the two 2 2 matrices reduce to
Vin
2 R 2 R Vo

= 1 1 1
Iin
1 Io

R R
and the final matrix multiplication reduces to
5 3R Vo
Vin

=

3

2 Io
Iin

R

189
If the load resistor is also R, then Io is equal to Vo/R and the voltage and current follow.
Vin = 5Vo + 3
Iin =
RVo
= 8Vo
R
3Vo
2Vo
5Vo
+
=
R
R
R
The input impedance of the resistive pad is then 8R/5 or 1.6R. The only difference
between the preceding text and the text to follow is that the EMI filters use
reactive components, adding J factors, or imaginary components to the system.
16.1. CHAIN MATRIX A: TRANSFER FUNCTIONS
The advantage of the A matrix is that it transfers the output to the input. If there
are five A matrices in tandem, each transfers the quantity from the right to the
left. This was the style of matrices used years ago for analog servo systems. A
water load was transferred to a pump, the pump transferred the load to the
electrical motor, which transferred the load to a control panel and feedback
system. The feedback was from the water system. What this system does is to
compare the output level to the input level, which is equated to so much dB loss.
16.2. REVIEW OF A MATRICES
This section reviews the chain matrix to take advantage of this tool to calculate
the insertion loss and other filter properties. These can be chained to form filter
modules such as T, , L, Cauer, dissipative, and other filter elements along with
their multiples. The matrix includes the load and source impedances, allowing
direct calculation of the insertion loss.
Knowledge of this section should allow easy additions by the reader as
needed. All the unit filters, except the dissipative and Cauer filters, can be handled
with four elements. These two would also have four elements, but each term
would have two terms, one real and one imaginary, for a total of eight. Therefore,
all the filter equations must be equally complete so that each matrix solution can
use the same form, or template, for each solution. This is like an overlay, and
because each matrix fits the overlay, continuous solutions can be calculated.
Because any of these units are in the matrix equation form, these elements can be
cascaded or chained to form the entire new matrix transfer function. The user can
place all of the elements as needed in tandem, and as long as the chain matrix
does not run out of room (in terms of the computer program), the matrices can be
chained ad infinitum.
The filter designer must know the voltage, current, required insertion loss,
190
Chapter 16
operating frequency of the power line, and other information about the load. In
these equations our only concern is the filter loss at a known frequency, such as
30 dB at 14 kHz. Often, the required filter loss is specified by an EMI test
laboratory after running the proper tests on the unit or system. In other cases, it
is just a good educated guess by the equipment or power supply designer using
equations given in various chapters earlier. The filter designers goal is to meet
the specified needed loss with some added headroom. From the final matrix of
all the combined elements, the loss at a frequency can be calculated with an
estimated load and source impedance. The government normally specifies the
insertion loss of a filter at 50 ohms load and source impedance in the 220A
specification. The designer can insert these values for any load and source needed.
In our case, the designer would pick a filter arrangement, such as a double T, and
use some program to find the cutoff frequency needed to get the loss. Programs
such as GOAL-SEEKER, an old shareware program, or the newer Lotus What If
and Backsolver programs can be used. Again, these programs are used to detect
the cutoff frequency giving the needed loss with about 6 dB headroom. If the
cutoff frequency, F0, is too low, cutting too deep into the frequency passband
(15 times the line frequency is suggested for 400 Hz), another stage must be
added (here another T). GOAL-SEEKER, or whatever method you use to iterate
between the needed loss and the cutoff frequency, is then used again to find the
new higher cutoff frequency to make sure that the filter is still not too close to
the passband as before. Usually, the passband should be at least 10 times the line
frequency for 400 Hz for higher loss requirements.
All the matrices listed here are 2 2 square matrices giving four elements.
As the number of elements or stages grows, the complexity of the terms grows,
not the number of elements. Thus, a simple single element has four terms but the
terms are easily calculated. A four-element L section also has four terms, but the
terms may be algebraically complex. Also, the terms of the Cauer and dissipative
filters are made up of both real and imaginary terms. Thus, the matrix may have
four elements but can have eight terms. All the layouts should be the same to use
the same matrix equation solutions, such as
A JB C JD
E JF G JH
(16.1)
where A, C, E, and G are real and B, D, F, and H are imaginary and where A
through H may be quite complex algebraically. The determinant of the chain
matrix, called delta (), must always equal one, and the imaginary term is reduced
to zero. Close counts here, though, as in the imaginary reading of 3.8E-7, are near
enough to zero to qualify. Changing from one style of matrix to another leads to
other cases creating answers close to one or zero. Some networks have parallel
branches, such as the dissipative filter. The two branches were each initially set
up using the A matrix and then each was converted to a Y matrix (admittance
191
matrix). These were then added and converted back to the A matrix form. This
often creates round-off errors. The of the determinant is slightly off from one
and leads to an imaginary term different from zero. That is,
(A + JB)(G + JH) (C + JD)(E + JF) = 1
(16.2)
This is true for all the matrices in the chain, as well as the resultant matrix, except
for the dissipative, Cauer, and other filters because of the matrix conversions,
adding some ambiguity to some elements. The order of progression of the
matrices is the same as that in which the whole filter types are laid out. The
order cannot be changed without arriving at an incorrect answer, but combining
them can be worked in any order as long as you are combining or reducing
adjacent matrices.
The solution of a two-element column matrix composed of the two matrices
as before is
A + JB C + JD K + JL M + JN Vo
Vin
I = E + JF G + JH . P + JQ R + JS I

o
in
(16.3)
Vo
Vin
(A+JB)(K+JL) + (C+JD)(P+JQ) (A+JB)(M+JN) + (C+JD)(R+JS)
I = (E+JF)(K+JL) + (G+JH)(P+JQ) (E+JF)(M+JN) + (G+JH)(R+JS) I
in
(16.4)
The first term (the A, real, and B, imaginary, position) is equal to

AK + JAL + JBK BL + CP + JCQ + JDP DQ
(16.5)
These can be separated into real and imaginary components.

(AK BL + CP DQ) + J(AL + BK + CQ + DP)
(16.6)
The same follows for the rest of the terms. All the values of A through S reduce
either to numbers or to basic algebraic equations. These filters are based on two
terms, K and Rd, where K is the normalized frequency, so that these can be plotted,
and Rd is the design impedance. The first element just rendered reduces to a new
A + JB and so do the rest of the elements. The new combined matrix is
Vi
A + JB C + JD Vo
I = E + JF G + JH I
i
(16.7)
Again, the determinant of the matrix equals 1, which means that the J terms
vanish for those derived purely via A matrix components.
AG BH CE + DF = 1
J (AH + BG CF DE) = 0
(16.8)
Note, as stated before, that close counts for the J term, equaling zero on
some of the derived matrices because of round-off errors, are created especially
when moving between different matrices such as converting to Y matricesdoing
192
Chapter 16
some additionand then converting back to the A matrix form. Also note that
Vo/RL = Io. If this is the final form after multiplying all the matrices, it can be
transformed to complex algebra by means of the following:
Vi =
Vo(Rl(A + JB) + C + JD)

Rl
(Rl(A + JB) + C + JD)

Vi
=
Vo
Rl
(16.9)
RlA + C + J(BRl + D)
Vi
=
Rl
Vo
So the voltage ratio
Vo
Rl
=
Vi
(16.10)
Dropping the primes for clarity,

Rl
Vo
=
Vi
Rl
Vo
=
2
Vi
(ARl + C) + (BRl + D)2

This voltage ratio is compared with the voltage ratio of the source and load
impedances without the filter (Fig. 16.2). The Insertion loss is
Rl
Rs + Rl
.
2
Rl
(ARl + C) + (BRl + D)2

Rs + Rl
(ARl + C)2 + (BRl + D)2
= insertion loss
FIG. 16.2 Insertion loss with and without the filter.
(16.11)
(16.12)
193
where Rs and Rl are the source and load impedances. The dB loss is given by
Rs + Rl
20 log
2
2
(ARl + C) + (BRl + D)
(16.13)
The angle of lag is

tan1
BRl + D
ARl + C
(16.14)
Engineering manuals and the IEEE literature confirm that there are many
reasonably priced programs (others not so reasonable) available that list these
components. These plot, list dB losses, give phase angles, etc., but I do not know
any that give the values of the components needed for a specific loss. Of course,
there are programs for conventional filter design techniques in which known input
and output impedances are mandatory. Some use these programs to back-engineer
an EMI filter by changing the component values until the goal of so many dB
loss at a particular frequency is reached. This and most other similar methods give
random values of components that yield the needed loss for the 50 ohm test setup
but can do strange things to the passband in the real world. This method can also
give a resonant rise at a low harmonic of the line frequency, oscillate, heat the
filter, or present such a low input impedance that the circuit breaker may trip on
turn on. Also, for 400 Hz, this technique can give a serious voltage rise at 400
Hz. There have been cases in which 120 V in produced 126 V out at 400 Hz.
If the inductors and capacitors are pure, the series element of inductive
reactance is placed in C + JD and any shunt or parallel elements reciprocal
impedance is placed in E + JF. In this case, being pure, the C and E terms are
zero. If the inductance has significant resistance, or DC resistance (DCR), this
resistance value is placed in term C. If the shunt capacitor has added resistance, as with an RC shunt, It is 1 over R + JXc. This gives the following for the
E and F terms:
R2
JXc
R
+
2
2
R + (Xc)2
+ (Xc)
(16.15)
A simple L formed with a series inductor followed by a parallel, or shunt,

capacitor would form a small chain matrix such as this. Note that C and E are
both 0 because their small values of equivalent series resistance (ESR) and DCR
are neglectedor treated as pureand the inductor faces the line while the
capacitor faces the load. But the system is missing the zeros or the unused terms
are gone This deletes the following terms: B, C, F, and H in the inductor section
and B, D, E, and H in the capacitor section.
194
Chapter 16
1 0
1 JXl
Vo
0 1 . J I

1 o
Xc
(16.16)
The new matrix becomes

Xl
JXl
1
X
Vo
c
I
o
J
1
Xc
(16.17)
This might look bad, but the first term reduces to 1 K2 as follows. The inductance and capacitance equations are
L =
Rd
2F0
C =
1
2F0Rd
(16.18)
The inductive reactance using Eq. (16.18) is

Xl = 2Fl =
2FRd
FRd
=
2F0
F0
(16.19)
The capacitive reactance, also using Eq. (16.18), is

Xc =
2F0Rd
F0Rd
=
2F
F
(16.20)
But
F = KF0
(16.21)
The inductive and capacitive reactance becomes

Xl = KRd
Xc =
Rd
K
(16.22)
where K is the normalized frequency, Rd is the design impedance, and F0 is the

cutoff frequency. The matrix of (16.17) is
1 K2 JKRd
Vo
I
JK
1 o
Rd
(16.23)
Note that terms 1 and 4 are real, resistive, and 2 and 3 are imaginary. Also note
that the preceding matrix yields a delta value of 1. What happens if the order is
195
changed? This places the shunt capacitor on the line side and the series inductor
facing the load.
1 0
1 JXl Vo
J 0 1 I
o
1
Xc
(16.24)
Substituting these values for Xl and Xc:

1 0
1 JKRd Vo
JK 0 1 I
o
1
Rd
(16.25)
Multiply the two matrices:

1 JKRd
Vo
I
JK
1 K2 o
Rd
(16.26)
Note the difference in the two answers, showing what happens when elements are
processed in the wrong order. Compare (16.22) and (16.26) (Fig. 16.3). Either
answer is correct, depending on what element the designer wants facing the line
or the load. Most EMI filter designers want the inductor on the line side,
FIG. 16.3 Comparison of the LC and CL filter matrices.
196
Chapter 16
especially for 461 specifications or any specification requiring high insertion

losses at frequencies near 14 kHz.
A multiple L of the type in Eq. (16.23) with the inductor on the line side
can be formed by
1 K2 JKRd 1 K2 JKRd
Vo
JK
I
JK
1
1 o
Rd
Rd
(16.27)
This reduces to
1 2K2 + K4 + J2K2 2JKRd JK3Rd
V0
2
o
JK (2 + K )
2K2 + 1
J
Rd
(16.28)
Substituting for J squared (1) gives

1 3K2 + K4 2JKRd JK3Rd
Vo
I
2
JK (2 + K )
o
2+1
K
Rd
(16.29)
The delta of the matrix is 1. If the cutoff frequency, F0, is 4000 and the frequency
of interest is 400, then K is equal to 400/4000, which is 0.1, and if Rd is 20, the
values in Eq. (16.29) would be
3.98 J Vo
0.9701

3
9.95 10 J 0.99 Io
(16.30)
If the load resistance is also 20, then Io = Vo/RL and Vi is equal to

0.9701Vo + 0.199 JVo
and Ii is equal to
9.95 103 J Vo + 4.95 102Vo
Then Zin is equal to
19.614 + 7.7672 102 J
Zin equals 19.614 at 13.613 minutes.
The value of the impedance is not too far from the expected 20 and there
is, in essence, no phase angle.
197
Various values of K can be plotted, or any one frequency can be quickly

checked. The main advantage is that the filter element values are given for the
required insertion loss. The various filter sections needed are listed for each term
of the matrix, and the method for designing other units is included in the next
chapter. All the equations are listed to follow the format of (16.7) without the
primes.
Vin
A JB C JD Vo
I = E JF G JH I
in
(16.31)
17
The Filter Design Technique
Here we discuss how to build the matrices for each filter element in the following
list. The names can be changed if the designer wishes to build these filters into a
computer program. These names are only those I used initially in my programming years ago with Lotus.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
Unit matrix
Rs matrix
LINESIM matrix
LISN matrix
DIN and DOUT matrices
RCSHU matrix
LSER matrix and CSHU matrix
The L
Pi filter
T filter
Cauer matrix
17.1. THE UNIT MATRIX

The simplest matrix is the unit matrix. This is the equivalent of scalar multiplication by one. The unit matrix requires no input. Make sure that all the unused
elements of the matrix field have unit elements in them and not other filter
elements. These other elements in these sections give erroneous higher losses,
making the filter appear to have the required specified loss but be well out of the
required specification when it is built and tested.
200
Chapter 17
The values are as follows. Unit matrix:

1 + J0 0 + J0
0 + J0 1 + J0
(17.1)
In reality, the unit matrix is a square diagonal matrix as follows.

1
0
0
1
The off-diagonals are all zero, and again, this is like multiplying the matrix field
by one. In the beginning, the Lotus program had seven matrix multipliers. The
first element was the Rs matrix, followed by the rest of the elements, and the rest
of the multipliers were filled with the unit matrix. For example, here seven
multipliers started with three L matrices, Rs, three Ls, and four units in this order.
The cutoff frequency and component values were calculated from the bottom. If
the cutoff frequency was too low, one of the units was replaced with another L.
17.2. THE Rs MATRIX
The value needed comes from the Common Terms section and is used for the
input-side line impedance Rs (Fig. 17.1). This term is the lower frequency line
resistance or source impedance. In the 220A tests, this would be 50 ohms. This
is fine for the frequencies under 100 kHz, where skin effect has not yet taken over
and the filter is being designed for a real-world insertion loss around 14 kHz.
Most of the various groups who have tested power lines have agreed that
Rs
filter
GEN
FIG. 17.1 The Rs element.
Rl
Filter Design Technique
201
the line impedance is around 4 ohms at 10 kHz. Here, the value of 4 ohms is
assigned.
1 + J0 4 + J0
(17.2)
0 + J0 1 + J0
which is reduced to the following (but all the matrices must have the eight terms
so that they can overlay the equation format and the solutions may be repeated
whatever the matrix placed at the multiplication spot):
1 4
(17.3)
0 1

No further information is needed for this unit. This is part of the line, not the filter.
17.3. THE LINESIM MATRIX
This matrix is used to match a known line or the output impedance of a DC power
supply (Fig. 17.2). The small series resistor is the value of Rs from the Common
Terms, but the values of the shunt resistor Rhf and the series inductor L must be
given. Rs, K, and F0 should be available from the Common Terms area and are
needed here. Rhf, the high shunt resistance, and L must be known. Rs acts as the
low-value series impedance of the source. A different symbol can be used. This
value must be provided if Rs is not used.
The frequency of interest is
F = KF0
(17.4)
This makes Xl equal to
Rs
R hf
FIG. 17.2 The LINESIM network.
202
Chapter 17
Xl = 2FL = 2KF0L
Rs, Rhf, and L1 are given so that values can be inserted directly into the matrix.
This is so that the line simulation network impedance follows the normalized
frequency. Thus, as K varies, the inductive reactance of this network varies along
with the filter losses.
The LINESIM is
Rhf + Rs JSl
Rs + JXl
+
Rhf
Vi
Vo
Rhf
(17.5)
I =
I
1
i
o
1 + J0
Rhf + J0
As an example, F0 is 4000, Rs is 4, Rhf is 50, L is 90 H, and K floats, so it can

be plotted. Xl = 2.262K, so the matrix becomes
1.08 + 0.0452JK 4 + 2.262JK
(17.6)
0.02 + 0J
1 + 0J
When this matrix is multiplied by the rest in the series, the various values of K
can be plotted. The LINESIM network is not part of the filter and accounts for
the line impedance.
17.4. THE LISN MATRIX
The input section, formed by Rin and Cin, is in series shunt across the line. The
midsection is made up with one inductor in series, L1. The output section, with
Ro and Co in series shunt, is also across the line. All these values must be known
and should be provided by the manufacturer. The initial values are from a standard
LISN but can be changed. The default values in any of these sections can always
be used. The engineer can input the main values LISN used repeatedly and save
these values so that this information need not be reentered each time this LISN
is used. If more than one LISN is used, several LISNs can be added by copying
the equations into each new section and changing the known values. These would
be called LISN1, LISN2, and so forth.
These units are used by the EMI test laboratories and are often required for
the various test specifications (Fig. 17.3). If the test specification calls for a LISN,
this matrix can be handy for the design of the EMI filter. Here, the LISN is used
as the input feed to replace the Rs matrix. These are also used to evaluate the
full system to determine whether the system meets the specification. These units
(Fig. 17.4) can be purchased from companies such as Solar Electronics, Hollywood, CA 90038 (Attn. Al Parker).
The LISN matrix is the hardest to format and is formed by the following
multiplication of the three matrices.
L1
POWER
LISN
FILTER
UNDER
TEST
LOAD
RES1
C1
TRANS1
L1
DIODES
SPECTRUM ANALYZER
FIG. 17.3
The LISN used for testing.
203
204
Chapter 17
L1
LINE
C1
22.5 F
56 H
R1
1 Ohm
LOAD
C2
22.5 F
50 Ohm term
R2
1 KOhm
FIG. 17.4 The common LISN.
1
0
0
1
1 JXl1
0 1 R + JX
R + JX
c1
c2
2
1
1
1
R2 + X 2
R2 + X 2
c1
c2
1
2
(17.7)
Solving the first two,

1
JXl1
1
0
R + JX
R + JX
(R
+
JX
)
JX
c1
l1 1
c1 2
c2
1
1
R2 + X 2 1 +
R21 + Xc21 R22 + Xc22
c1
1
(17.8)
The full solution is
R2 + JXc2
JXl1
1 + JXl1 2
R2 + Xc22
R + JX
JXl1(R1 + JXc1)
c1 R2 + JXc2 JXl1(Rl + JXc1)(R2 + JXc2)
1
+ 2
+
1+
2
(R21 + X2c21)(R22 + X2c22)

R21 + X2c21
R2 + Xc22
R1 + Xc22
Typical LISN values with F0 at 4000 are given in Table 17.1.

XC1 =
1
1.768
=
2KF0C1
K
Xl = 2F0L1 = 1.407K
XC2 =
1
1.768
=
2KF0C2
K
(17.9)
205
TABLE 17.1 Typical LISN Values with F0 at 4000

Line section
Midsection
Load section
R1 = 1 ohm
C1 = 22.5 f
L1 = 56 H
R2 = 50 ohms
C2 = 22.5 F
Using the standard outline of the 2 2 matrix with eight terms, the equations of
the individual terms follow.
A + JB C + JD
E + JF G + JH
The terms are

A = 1
B =
XL1XC2
R22
+ XC1
JR2XL1
R22 + X2C2
= 1
1.407 1.768
1.497 1.768K2
= 1
2
2
2
502K2 + 1.7682
50 + 1.768 / K
J50 1.407 K
J 50 1.407 K3
=
502 + 1.7682 / K2
502K2 + 1.7682
C = 0
D = JXl1 = J1.407K
E =
(17.10)
R1R2(R1 + R2) + R1X2C2 + R2X2C1 XL1(XC2R2 + XC1R1)

(R21 + X2C1)(R2 + X2C2)
The two capacitors are often the same values as in Fig. 17.5. This reduces to
E =
(R1 + R2)R1R2 + X2c2 Xl1Xc2

(R21 + X2c2)(R22 + X2c0)
where R1 and Rin are the same, which also follows for C1 and Cin, R2 and Rout
and C2 and Cout. These values are interchangeable as in the LISN in Fig. 17.5.
From the preceding values,
E =
51K2(47.51K2 + 1.7682)
(K2 + 1.7682)(502K2 + 1.7682)
F = J
XC1(R22 + X2C2 XLXC2) + XC2(R21 + X2C1) + R1R2XL
This reduces to
(R21 + X2C1)(R22 + X2C2)
206
Chapter 17
L1
LINE
C1
22.5 F
56 H
R1
1 Ohm
LOAD
C2
22.5 F
50 Ohm term
R2
1 KOhm
FIG. 17.5 The common LISN with equal capacitors.
F = J
G =
|70.35K5 + 4417.34K3 + 11.053K|

(K2 + 3.126)(2500K2 + 3.126)
R21 XL1XC1 + X2C1

R21 + X2C1
3.126 1.4876K2
K2 + 3.126
JR1XL1
J 1.407K2
H = 2
= 2
2
R1 + XC1
K + 3.126
G =
In the computer program, these are already provided, but the component values
must be entered if they are different from these default values. This shows where
they come from.
Finally, the component values are entered. These are (A through H) based
on the following simpler equations:
P = 2F0K
A =
B =
P2C22R22 P2C2L1 + 1
1 + P2C22R22
JP3C22L1R2
1 + P2C22R2
C = 0
207
D = JPL
E = P2
R1C1[1 + P2C0(C0R0 L)] + R0C20[1 + P2C1(C1(C1R21 L)]
F = JP
G =
(1 + P2C21R21)(1 + P2C20R20)
C1(1 + P2C20R20) + C0(1 + P2C21R2) + P2C1C0L(P2R1R0C1C0 1)
(1 + P2C20R20)(1 + P2C21R2)
1 + P2C1(C1R21 L)
H = J
1 + P2C1R2
|P3C21R1L1|
(1 + P2C21R1)
With this, the only variable is P, and this term varies with K. Again, this allows
all these equations to be plotted as a function of K. K is the normalized frequency,
which has the value one at the cutoff frequency.
17.5. THE DIN AND DOUT MATRICES

This is the dissipative filter and gives a loss of 6 dB per octave. DIN faces the line
and DOUT faces the load if both filters are used, one at each end. Several terms
must be specified, including Rin and Rout. The value of K is the same as K in
the other filters. Rin in Fig. 17.6 (bottom), should be the wanted filter input
impedance, such as 50 ohms, for the DIN filter. Rl, the output impedance needed,
should match the load for the DOUT, which can also be 50 ohms. This would
be a nice feature for the 220A specification. Here, the term is Rout but R4 is
used in Fig. 17.6 (bottom). The opposite two resistors, Rout (R2 in Fig. 17.6,
bottom) for DIN and Rin (R3 in Fig. 17.6, bottom) for DOUT, can be used to
match any filter wantedButterworth, , or otherand these can also be, here,
designed for 50 ohms.
The delta is not always equal to one because of the conversion back and
forth between the Y and A matrices. This leads to minor errors, or round-off errors,
within the terms. The values of K and F0 could come from the Common Term
area and the values of Rd, Rin, and Ro should be listed with this matrix. Both
DIN and DOUT are the same, but Rin and Ro should switch values.
As F0 decreases in value, both L and C increase in value. If the line
frequency is 400 Hz, the central filter cutoff should be 4000 Hz and above, say
4000 Hz. Then if the K ratio is 0.5, or making its cutoff 8000 Hz and still having
the full filter give a good impedance match at 10,000 or 14,000 Hz, say 50 ohms,
the basic DIN and DOUT equations are
208
DISSIPATIVE FILTER
L1
R1
C1
R2
DIN & DOUT WITH ANY FILTER

L1
L2
ANY FILTER
R1
The DIN and DOUT with any filter.
R3
C2
R4
Chapter 17
FIG. 17.6
R2
C1
Kd = KKratio
Rd =
RinRo
RdKratio
2F0
L =
209
C =
Kratio
2F0Rd
(17.11)
where Kd is the equivalent of the value of K in the other filters, Rd is the design
impedance for the DIN and DOUT only, and F0 is the cutoff frequency for all the
filters, including the DIN and DOUT.
The format of the DIN or DOUT matrix follows the same as that of the others
given here:
A + JB C + JD
E + JF G + JH
(17.12)
If the design calls for all the Rs to be 50 ohms and Kd is equal to 1, K ratio = 1,
the various elements, A through H, equal the end values given after each element:
A =
4K2dR2d K2dRin(Rin + Ro) + (Rin + Ro)2

K2dR2d + (Rin + Ro)2
A = 0.75
B = J
2K3dRdRin
4K2dR2d + (Rin + Ro)2
B = 0.25J
C =
K2dR2d(Rin + Ro)
C = 12.5
D = J
2K3dR3d + KdRd(Rin + Ro)

D = 37.5J
E =
K2d(Rin + Ro)
E = 0.005
F = J
2K3dR3d + Kd(Rin + Ro)2

Rd(4K2dR2d + (Rin + Ro)2)
F = 0.015J
G =
4k2dR2d K2dRo(Rin + Ro) + (Rin + Ro)2

(17.13)
210
Chapter 17
G = 0.75
H = J
2K3dRdRo
H = 0.25J
Typical values are F0 = 4000, K ratio = 1, Rin = 50, Ro = 50. Then Kd = K, Rin +
Ro = 100, and
0.75 + J 0.25 12.5 + J 37.7
0.005 + J 0.015 0.75 + J 0.25
(17.14)
as shown by the preceding solutions of A through H. Note that the delta is equal
to 1. These are the values at 4000 Hz.
DIN and DOUT are the same but Rin and Ro, if the two values are different,
exchange values so that Rin of DIN faces the line and Ro of DOUT faces the
load. The K ratio is usually 1 but must be the same if both DIN and DOUT are
used. This means that two matrices are required: one matrix for DIN and one
matrix for DOUT.
17.6. THE RCSHU MATRIX
This matrix is used to lower the Q of the filter and correct for a resonant rise
and/or a problem related to frequency, such as insufficient loss at a particular
frequency. The design method is to calculate the needed capacitor value to equal
the design impedance at the problem frequency and then place this combination
in series across the line. These are normally mounted inboard between two other
filter sections, such as between two Ls, and would be tied across the first Ls
capacitor. Often, this filter section can cure several problems at once. If, for
example, a three-stage L is needed to obtain the required insertion loss at 10 kHz,
two resonant rises appear in the graph. As the load impedance deviates from the
design impedance, the peak value of each resonant rise changes. Also, the loss
near 10 kHz may be at, or somewhat over, the insertion loss limit. Determine the
lower resonant rise frequency and tie the resultant network across either the first
or the second Ls capacitor. The result should be that the first resonant rise is
reduced and the second resonant rise is gone. The loss around 10 kHz, assuming
that the first resonant rise is well below 10 kHz, is. Circuit Q is reduced to impede
any oscillations by the addition of an RC shunt (Figs. 17.7 and 17.8). This is easier
to design than the Cauer and is automatically balanced.
The matrix is simple to form: Rc is the series resistance and should be
equal to Rd, the design impedance from the Common values to all filters, if any
such area exists, or listed here. C is calculated as before, and Xc = Rd at the
problem frequency. The proper values are placed in the E + JF area of the matrix
211
C
R = Rd ?
FIG. 17.7 The RC shunt.
as listed here. Once the value of the capacitor is known, the capacitive reactance
for the matrix must be a function of K, so the matrix varies with frequency and
can be plotted.
Xc =
E =
1
2F0KC
Rc
R2c
X2c
(17.15)
F =
JXc
R2c
+ X2c
RCSHU:
1 + J0 0 + J0
(17.16)
E + F 1 + J0
An RCSHU example is as follows: A triple L was designed with two

resonant rises. The first fell near 3800 Hz and the second near 4600 Hz, and the
design impedance was 10 ohms. This filter was also out of limits at 14 kHz.
Design the capacitor impedance to equal the 10 ohms at 3800 Hz for the first,
lowest, resonant rise. This gives
C =
1
= 4.188 F = 4.2 F
23800 10
The self-resonant frequency (SRF) of this capacitor need not be as good as that
of the rest of the filter capacitors unless there are also higher frequency problems.
This is being used primarily to solve some low-frequency problems, and the
highest frequency is 14 kHz in the case just stated. If the SFR is better than the
fifth harmonic, 70 kHz here, of this upper frequency, the job will get done. This
statement is being made so that the filter designer does not select a capacitor that
costs many times more than one that will do the job. Note that Xc is a function
of K and not single valued as shown before. Tie the 10 ohm resistor and the 4.2
F in series across one of the inbound capacitors that make up the first or second
212
.5L
.5L
CF
.5L
Rc = Rd ?
C
Crc
.5L
.5L
FIG. 17.8
Multiple L with RC shunt.
.5L
CF
Chapter 17
213
L counting from the load end of the filter. The closer to the load, the better this
network works, because the load is the higher impedance.
17.7. THE SERIES INDUCTOR, LSER, AND THE SHUNT
CAPACITOR, CSHU
These should be understood from the various earlier descriptions but are reviewed
here anyway. There are many uses for each of these, and in some locations they
can be used together. For example, these two together can make up Ls, Ts, s,
and so forth that are not functions of Rd. CSHU, alone, can be used to duplicate
the value of a feed-through capacitor. The LSER, the series inductor, can be the
common mode inductor. LSER and CSHU can work together to form part of the
common mode filter (Fig. 17.9).
The LSER is
Xl = 2F0KL
(17.17)
It is obvious that the value of Xl is a function of K and is placed in the matrix at D.

LSER:
1 + J0 0 + JXl
0 + J0 1 + J0
(17.18)
CSHU:
Xc =
1
2F0KC
(17.19)
This is the same as RCSHU except that Rc is zero, making the E value zero and
F the reciprocal of Xc.
LSER
CSHU
FIG. 17.9 The LSER and CSHU.
214
Chapter 17
CSHU:
Xc =
1
2F0KC
(17.20)
17.8. THE L MATRIX

This is a review of Sec. 11.2 showing the origin of the values that make up the
L filter, where K is the normalized frequency and Rd is the design impedance
(Fig. 17.10).
F = KF0
L =
Rd
2F0
Xl =
2FRd
= KRd
2F0
C =
1
2F0Rd
Xc =
2F0Rd
Rd
=
2F
K
Here, the inductors and capacitors as treated as pure. The series element of
inductive reactance is placed in the JD term, and the reciprocal of this impedance
of the shunt capacitor is placed in the JF term. The inductor faces the line, and
the capacitor faces the load. The L matrix is formed as follows:
Vin
1 + J0 0 + JKRd 1 + J0 0 + J0 Vo

=
JK
1 + J0 Io
Iin
0 + J0 1 + J0 0 +
Rd
L1
LINE
LOAD
C1
FIG. 17.10
The L filter.
(17.21)
215
Get rid of the unnecessary zeros, which are here only because the eight positions
must be filled so that the computer program sees the full eight units. In this way
the same solution formula can be reused repeatedly like a template.
Vin
1 0 + JKRd 1 J0 Vo
=
JK

1
1 Io
Iin
J0
Rd

(17.22)
The new matrix becomes

1 K2 JKRd Vo
Vin

= JK
1 Io
Iin
R

d
(17.23)
Then, inserting the other zeros so that the L has the same form as the rest,
1 k2 + J0 0 + JKRd Vo
Vin

=
JK
1 + J0 Io
Iin
0+
Rd

(17.24)
17.9. THE MATRIX

In this matrix, the capacitor value is split. Half is placed in front of the inductor
in parallel with the line, and the other half is in parallel with the load (Figs. 17.11
and 17.12). The capacitive reactance of both of these halves is then doubled. Yet,
it is the reciprocal of this impedance placed in term JF.
L1
LINE
LOAD
.5C1
FIG. 17.11
The filter.
.5C1
216
Chapter 17
HOT
.5L1
.5C1
LINE
.5L1
C1
.5L1
.5C1
LOAD
.5L1
NEUT
GRND
FIG. 17.12
Multiple balanced filter.
Vin
1 0 1 JKRd 1 0 Vo
=
JK
JK

Iin
1 0 1
1 Io
2Rd

2Rd
(17.25)
or half the value then used to make up the L matrix in term JF.
The matrix is
K2
+ J0 0 + JKRd
1
Vin
2
Vo
I =

2
2
0 + JK(4 K ) 1 K Io
in
4Rd
2
with the same form as the rest.
(17.26)
17.10. THE T MATRIX

In this case, the inductor is split with half of the inductor in series with the line
and the other half in series with the load (Figs. 17.13 and 17.14). This makes the
JD term half the value of the or L. The full matrix of the T is
JKRd
JKRd

1 2 1 0 1 2 Vo
Vin
=
0 1 JK 1 0 1 I
I
Rd
o
in

The full T matrix becomes
(17.27)
217
.5L1
.5L1
LINE
LOAD
C1
FIG. 17.13
The T filter.
JKRd

K 2
K2
+ J0 0 +
2 Vo
1
Vin
2
2
2
=

2
JK
K
0+
I
+ J0 Io
1
in
Rd
2

(17.28)
and better
JKRd(4 K2)
2 K2

+ J0 0 +
Vo
Vin
4

= 2

2
JK
(2 K )

I
+
J0
0
+
in
Io
R
2
d

.5L1
L1
.5L1
LINE
LOAD
C1
FIG. 17.14
The multiple T.
C1
218
Chapter 17
17.11. THE CAUER MATRIX OR ELLIPTIC FILTER

The Cauer filter (Fig. 17.15) gets rid of a problem frequency. In the 461
specifications in which 100 dB or so is needed at 10 to 14 kHz, a Cauer is
sometimes used. Fm is the problem frequency and Rc is the resistance in series
with the capacitor. Above Fm, the Cauer looks like a capacitor and the resistor
limits the minimum impedance in this capacitor leg. The impedance of Rc is
usually assigned the value of Rd, the design impedance of the filter. This is
usually employed on the center inductor of a triple-L network. The Cauer is
often associated with a triple-L filter (Fig. 17.16). The inductor value is the
same as before:
L =
Rd
2F0
(17.29)
and the Cauer capacitor must resonate with L at the problem frequency, Fm.
C =
F0
2F2mRd
The impedances are

Xl = KRd
Xc =
2F2mRd
M2Rd
=
2F0F
K
(17.30)
where M is the multiplier to convert from F0 to Fm.

Fm = MF0
The C term is
C =
K4RcR2d
K2R2c +
R2d(K2 M2)
and the D term is

D = J
(K3R2c Rd K3M2R3d + KM4R3d

K2R2c + R2d(K2 M2)
Rc and Rd should be equal but are listed here separately if for any reason they
should be different for a special problem that any engineer may encounter. If the
two terms are equal, C and D reduce to
C =
K4Rd
K2
+ (K2 M2)
(17.31)
R = Rd ?
CAUER
BALANCED CAUER
R = Rd ?
C
2C
LINE
.5L
LOAD
LINE
LOAD
R = .5Rd?
2C
.5L
FIG. 17.15
The Cauer matrix.
219
220
CAUER
R = Rd ?
L
LOAD
LINE
C
The Cauer or elliptic filter.
Chapter 17
FIG. 17.16
221
which approaches Rd at the higher k values, and

D = J
(K3Rd K3M2Rd + KM4Rd)

K2R2c + R2d(K2 M2)
which approaches
J
RdM2
K
which is the capacitive reactance, and this approaches zero as the frequency
increases. The matrix for the Cauer is
1 + J0 C + JD
0 + J0 1 + J0
and without the zeros is
(17.32)
1 C + JD
0
1
The conclusion of this chapter is that the matrix terms are difficult to evaluate,
and these are needed if you wish to form these transfer functions on your own.
The hope here is that this chapter is clear enough for the reader to evaluate.
18
Matrix Applications
The first step is to find the filter type needed for the application. The listings in
Sec. 5.8 should help. Either through documentation from the EMI test house or
mathematically from sections of Chapter 11, the required insertion loss must be
known. Often, this is given as a list of frequencies and the required loss at each.
The filter must be checked at each frequency to make sure the loss is met at each
frequency. The load current input voltage and the line frequency must be known
to get the design impedance, Rd, and to find the minimum cutoff frequency. Figure
out the value of Rd by dividing the voltage by the expected highest load current,
and the cutoff frequency should be 10 times the line frequency.
The first matrix in the chain, or cascade, should be a matrix that represents
the line. This could be the Rs, Linesim, LISN, or any other line unit added by the
user. These are not part of the filter, or filter construction, but aid in the loss.
Therefore, these values should be as accurate as possible. The second, third,
fourth, and so on are parts of the filter. These are matrices multiplied together,
and the final matrix is solved for the full loss of the filter. As K varies, the loss
changes and these can be plotted. This continues into the next section.
One condition discussed here is the low-current applications in which
operational amplifiers (OpAmps) are used. Many design engineers make some
errors when using this application. These engineers substitute active filters to
replace passive components in the lower current applications. These filters must
have very clean power or they cross talk, generating the same noise, and in some
cases even more noise, that the active filter was meant to cure.
224
Chapter 18
18.1. SINGLE-PHASE AC FILTER

As an example, an L is chosen by the filter designer. The design impedance is
found to be 10 ohms, and the line frequency is 400 Hz. The loss needed is 40 dB
at 20 kHz. This is a low frequency, and the Rs matrix is chosen for the first matrix
to represent the line. Rs is thought to be around 4 ohms at 20 kHz. The second
matrix is the L matrix chosen by the EMI design engineer. If these are automatically multiplied as a system in the computer, all the other matrices in the chain
have unit matrices in them. The final matrix is the result of only the Rs and the L
filter because only they control the answer. The value of F0 is changed to meet
this 40 dB loss by varying the value of F0 or by using a program such as
GOAL-SEEKER (an old shareware program). In the newest Lotus program, this
feature is built in. The engineer should add 6 dB for headroom to the loss. This
gives the filter designer some latitude in component values without fear of
harming the performance of the filter. It also gives some additional attenuation to
some unknown frequency spike that could push the equipment using the filter
over the limit. The value of F0 here is 1200 Hz to get the desired loss of 46 dB
(40 + 6 headroom), and this is sitting on the third harmonic and is far too low.
Another L is added. Skip by one matrix, and leave it as a unit matrix. This allows
changing it later to an RCSHU, or whatever, if needed. To review:
Matrix 1: Rs
Matrix 2: L
Matrix 3: left as a unit
Matrix 4: changed to L
All remaining matrices: units
This is followed by the column matrix with elements Vo and Io as shown
in Chapter 17. Io can be replaced by Vo/Rd, where Rd is the design impedance and
is the lowest load resistance seen by the filter. Proceed by adding the next L filter.
This changes the cutoff frequency to 4600 Hz, which is well above the 4000 Hz
recommended limit. The initial inductance was 1.3 MH, and now the requirement
is two at 345 H. The same is true for the capacitor. It was 13.3 F and is now
two at 3.46 F. This can be balanced and look as in Fig. 18.1.
FIG. 18.1 The balanced L.
Matrix Applications
225
FIG. 18.2 The full balanced double L with Zorro, RCSHU, and feed-throughs.
The load impedance was changed to see how the filter responded in the
standby mode, and a resonant rise was found at 2760 Hz. This required an
RCSHU to be added to the third matrix, and the common mode components are
added in Fig. 18.2. The new listing for the matrix layout is
Matrix 1: Rs
Matrix 2: L
Matrix 3: now an RCSHU
Matrix 4: L
All remaining matrices: units
This matrix group is still followed by the column matrix with elements of
Vo and Vo/Rd. Check that all dB losses at the other frequencies listed in the
specification meet their requirements for this filter. Most expect all the losses at
the higher frequencies to pass if the lowest passes. This is usually true, but there
are exceptions, so all must be checked. This is treated later, but the frequency that
requires the lowest cutoff frequency determines the cutoff frequency used.
The remaining job is to determine the Z (Zorro). The two feed-through
capacitors are added, giving 0.04 F (Fig. 18.3). The line-to-line capacitors are
out of the circuit along with the RCSHU. The 175 H inductors are divided by
2, giving 87.5 and a total of only 175 H for the four, adding little to the common
mode insertion loss. Give all this up to headroom. This is reduced to a single L.
FIG. 18.3 The common mode part of the filter in Fig. 18.2.
226
Chapter 18
FIG. 18.4 The value of Zorro for Figs. 18.1 and 18.2.
Remove or replace these elements with the unit matrix, except the Rs and
the L matrix. The loss for the common mode is 46 dB at the switcher frequency
of 50 kHz. The 0.04 F capacitor must be maintained. This requires an impedance
of 1125 ohms and a cutoff frequency of 3550 Hz. The Zorro is 50.4 MH, which
is ridiculous. The filter has been designed to component level, but the Zorro
should be fixed. Some would substitute any reasonable lower value, hope for the
best, and pray for an exemption from the specification if the unit fails. The circuit
is shown in Fig. 18.4.
All that remains is to design the two different inductors and the capacitors
or choose a supplier. This filter may be out slightly when tested but should require
only minor changes. High-frequency problems? Add Capcon (see Chapter 4) or
ferrite beads. Differential problems at the switcher? Improving the quality of the
RCSHU capacitor or increasing its value and/or raising the inductors to 180 H,
or 200 if needed, may solve the problem. Common mode? Increase the Zorro
somewhat. Too high a value for the common mode inductor, as before? Change
the feed-through values to 0.01, and add a second section. This is now a double
L, as shown in Fig. 18.5.
The new value of each Zorro is 7.9 MH. Now, the case could still be
presented again, examining the contribution of the differential mode to the
FIG. 18.5 Calculation of the double Zorro.
Matrix Applications
227
FIG. 18.6 Balanced single-phase filter with two double Ls.
common mode filter. One set of parallel differential inductors aids each common
mode inductor. The parallel value of these inductors is 87.5 H. The Zorro is still
an order of magnitude greater than the 87.5 H and is still not subtracted to lower
the size of the Zorro. This difference is left to increase the headroom. Note that
this filter (Fig. 18.6) has an internal shield to ground (the case). Note also that the
size of the Zorro has dropped. This is also a much more practical inductor with
this much current.
18.2. THREE-PHASE FILTER
There are two types of three-phase filters: the wye (or Y) and the delta. These
two types are then divided into two groups. The first is the higher current type,
in which each leg is a separate insert, with each insert the same type and size.
This is very important because of the possibility of installing the smaller neutral
filter in any of the main legs.
Also, the third, sixth, ninth, and so on currents add in phase on the neutral
wire along with the unbalanced current of the three legs. This harmonic content
would be the odd multiples of 3 and some of the even multiples such as 6 or 12.
Therefore, the wire size and filter size are the same for the neutral as required for
the other three legs.
A sin(Nt + N0)
A sin(Nt + N120)
A sin(Nt + N240)
3A sin(Nt)
(18.1)
This assumes that the peak harmonic current on all three legs is the same. If not,
the term 3A is replaced by (A+B+C). The three phases add for all multiples of 3,
if that harmonic exists. The sixth should be a very low level, if it exists at all, but
the ninth is strong, especially where the off-line regulator is used as part of the
power supply. Beware of multiphase transformers that reduce these harmonics.
This is a function of the number of phases used, but most eliminate the lower
harmonics such as the third harmonic. This eliminates that problem but can form
228
Chapter 18
a inductive voltage divider. The following approximate equation for the off-line
regulator current is normalized to one. Refer to Eq. (4.1).
4T
N N 2N
1
Cos
Sin Sin
2 N22
2T
2
T
T
N=1,3,5
(18.2)
If the 3, 9, 15, 21, and 27 are added, the peak reaches almost to 0.7. This current,
plus the unbalanced current, should prove that the neutral filter must be the
same size as the other three legs. The last three-phase type is the lower current mode.
To design the three-phase filter, the voltage and current seen by each filter
leg must be known. Find the maximum power required by the load. Divide this
by 3 to obtain the power per leg. Divide this answer by the line-to-ground voltage
for a wye or the line-to-line voltage for a delta. Multiply, for delta only, the last
answer by the square root of 3.
12,000
= 4000
3
19.231
3 = 33.309
4000
= 19.231
208
33.309
2 = 47.106
If the total load power, taking all the inefficiencies into consideration, is 12 kVA,
the power supplied to each line is 4 kVA. If the line-to-line voltage is 208 V,
the current is 19.231 times the square root of 3. This equals 33.309 line A.
Assuming that no off-line regulators are used, the current peak is 47.106 A.
The inductors must not saturate at this current, here 47.1 A. The design impedance is 208 V divided by 33.309, giving 6.24 ohms. Use 6 ohms for the
design impedance.
18.2.1. Low-Current Wye
The power is from each leg to neutral, and the neutral current is zero if the
currents in the three legs are well balanced. But, as mentioned before, the odd
third-order harmonics are still present. This is true only for the fundamental of
the power line frequency as discussed earlier. In the low-current filter, all of the
components are in the same container and the capacitors are wired from the leg
to neutral. This makes for smaller capacitors because of the charge from 208 to
120 V. This saves money, weight, and volume.
These filters are often called five wire because of the three legs plus the
neutral and the ground. The ground must be left intact and unfiltered. I have seen
many filters with components in the ground leg from various filter manufacturers.
This is a violation of the electrical code. The ground lead must be intact and not
broken. The only exception would be if ferrite beads or toroids were slipped over
the wire, leaving the ground lead a solid wire.
Matrix Applications
229
The output feed-through capacitors in Fig. 18.7 are to case ground, and the
ground wire is wired to the case. Also, the Zorro common mode inductor can be
used for common mode rejection because all parts are within the same enclosure.
The common mode inductor cancellation of the magnetic field of the
differential mode power current works as shown in Fig. 18.8. Discussing the
balanced fundamental power frequency first and ignoring the harmonics, the A,
B, and C currents generate magnetic fields that cancel. Here, a magnetic field
cannot be generated with no current flow in the neutral common mode leg. If the
system is unbalanced, (we know that it is), the difference current flows in the
neutral leg of the common mode inductor and still brings the common mode back
to balance. As far as the harmonics are concerned, whatever harmonic current
flows in any leg also flows in the neutral leg and still cancels. So, as far as the
power delivered to the load is concerned, the magnetic field of the common mode
is neutralized. Any common mode pulse or signal, either from the load or the line,
is attenuated by upsetting this balance. The design technique is the same. Rd is
the lowest line voltage divided by the highest line current required by the
equipmentnot inrush. Find the cutoff frequency, and the values of the inductors
and capacitors follow.
FIG. 18.7 The low-current wye triple-L filter.
230
Chapter 18
FIG. 18.8 The low-current wye triple-L filter with Zorro.
18.2.2. High-Current Wye

These filters become so large and heavy that they must be built using the insert
technique. All three legs and the neutral are the same electrically and physically, and all inserts are installed in a larger enclosure. In this way, if any leg
blows, only the failed filter of that leg needs to be replaced. This technique
also eases the installation by allowing the outer enclosure to be installed and
then the inserts placed in this enclosure one at a time. This is better than hoisting
the entire enclosure and contents, which has been done by various contractors.
They installed all the inserts, hoisted the full weighted unit, and then dropped it.
The entire filter had to be replaced at a premium cost because of time constraints.
This was expensive, to say the least.
18.2.2.1. A Single Insert
All the capacitors in Fig. 18.9 are to case ground. In this arrangement, there are
no components from line to line, such as capacitors or MOVs. The leakage
specification, even with power factor correction coils, is rarely met. There really
should not be a leakage current specification for this high-current type of filter.
This should be known before the specification is written and often is not known
or heeded by some specification writers. It has been asked after the filter is
shipped that some leakage specification also be met that was not provided to the
manufacturer initially, and power factor correction coils had to be added. This
improved the situation, but the filter still did not pass.
Matrix Applications
231
FIG. 18.9 The single high-current insert.
There is some limited common mode in the differential inductors, which

are in parallel, and the capacitors, which are also in parallel. Each inductor is
divided by the number in parallel, and the capacitors multiply in parallel. Also,
this assembly lacks the ability to use the Zorro common mode inductor for
common mode rejection. In the following case, the equivalent circuit is composed
of L/4 and 4C as shown in Fig. 18.10.
The common mode equivalent is as shown in Fig. 18.11. The working
impedance is now one fourth the design impedance, but the cutoff frequency is
the same.
FIG. 18.10
The three-phase four-wire filter.
232
L/4
L/4
4C
FIG. 18.11
L/4
L/4
4C
4C
4C
The common mode portion of the four-wire filter.
Chapter 18
Matrix Applications
233
Follow the single-phase design method. Rd is the line voltage divided by the
maximum line current in any one leg.
18.2.3. Low-Current Delta
This lower current type is often specified to pass 220A with the stipulation to
test any one line with the other two legs grounded. This is a good spot for the
type because the line and load impedances are both 50 ohms. The line-to-line
capacitors can be shared, giving twice the capacitance to ground (Fig. 18.12). This
makes our job easier.
If the A leg is under test in the 220A system, legs B and C are grounded.
Then the capacitors from A to B and A to C are in parallel on both sides of the
differential filter inductor. The impedance is decided in the same way using the
FIG. 18.12
The four-wire delta showing case ground for feed-through operation.
234
Chapter 18
line-to-line voltage and current. Follow the same procedure for the single phase.
If the 220A specification is all that has to be met, the capacitor values can be
divided by 2, making for smaller capacitors and saving money, weight, and
volume. Pick the Rs for matrix 1 and place the in matrix 2; the rest should be
unit matrices.
Matrix 1: Rs
Matrix 2:
Matrix 3: unit
Matrix 4: second if needed
All remaining matrices: unit matrices
Remember, the column matrix follows this. If the single meets all the requirements, the design is finished. If not, add a second for matrix 4, leaving matrix
3 a unit matrix. This is done for the same reason, to be able to add another network
for matrix 3. Once the capacitor value is established, split the value between the
lines. If the is a multiple, check for a resonant rise and fix as before. Continue
with the design as in the single phase. Check that each frequency listed in the
specification meets the required loss of the list. If not, the filter may have to be
redesigned for more loss or any of the other solutions depending on the frequency
of the problem.
18.2.4. High-Current Delta
This is the same as the high current wye except that the capacitors are again tied
to ground. These should be from leg to leg as in the low-current delta. There is
no convenient way to do this using the insert method. The value of the required
capacitance should be doubled to give half the required reactance to ground and
the other half back to the other leg.
18.3. TELEPHONE AND DATA FILTERS
These are easy EMI filters to design, and the design program provided works
well. The reason is that the input and output impedance is known, typically 50,
75, 135, and 600 ohms, and the two impedances are always the same. The filter
is always balanced; the two inputs are called tip and ring, and so are the two
outputs. Most telephone filters are 300 ohms from line to ground and 600 ohms
line to lineactually tip to ring. These are typically s or Ts. The currents are
low, and the filter resistance is usually not critical because it is such a small part
of the 300 ohms. That is the way to design it. Use the 300 ohms for the source
and load, and this will give the filter for the tip and the same for the ring (Fig.
18.13). As an example, a filter requiring 60 dB of loss at 20 kHz matched to 300
ohms would consist of four filters (Fig. 18.14).
Matrix Applications
FIG. 18.13
235
Telco filter.
Data filters are not much different; different impedance and the amount of
loss required would be about all. However, remember that Fig. 18.13 is only half
the filter. If this is the filter for the tip input and output, there is another for the
ring side.
18.4. IMPEDANCE-MATCHED FILTERSWHAT IS THE
IMPEDANCE LIMIT?
This follows from Sec. 18.3. As the impedances of the source and load grow, so
does the inductance, and the capacitance drops in value. It comes to a point where
the inductor becomes too large and the total DC resistance (DCR) is now a good
portion of the specified design impedance. The capacitors become too small. For
example, in a 1000-ohm impedance filter, the inductor is 1 million times the
capacitor value. I would suggest this as the upper value.
18.5. PULSE REQUIREMENTSHOW TO PASS
THE PULSE
This is a continuation of Secs. 18.3 and 18.4. Pulses are made up primarily of odd
harmonics. The quickest way is to obtain the pulse width, take the reciprocal to
get the frequency, and multiply this frequency by 10. The EMI filter should then
have a cutoff frequency above this frequency. If this is impedance matched as in
Secs. 18.3 and 18.4, the band pass will be very flat. This will pass the fundamental, 3rd, 5th, 7th, and 9th along with some of the 11th. The pulse distortion should
be minimal.
18.6. THE DC-TO-DC FILTER
The DC-to-DC type is often a tubular type and is unbalanced. Figure 18.15 shows
the proper hardware for installing the tubular through a wall or equipment
236
Chapter 18
FIG. 18.14 The Telco loss.
Matrix Applications
FIG. 18.15
237
The unbalanced feed-through filter.
housing. Figure 18.16 shows the unit installed. The filter is often a single
feed-through capacitor giving just 6 dB per octave loss. In some other filters,
inductor(s) are included, making up the L or T type. These last two give 12 or 18
dB per octave. The filter requires a good ground, or the filter is out of business.
If this is an L or T, then only the inductor(s) is in the circuit. This reduces the loss
to 6 dB rather than the 12 or 18 that the L or T would give. But what if the filter
is a single feed-through type? This would net zero loss, all because the filter is
not grounded.
The construction is as follows (Fig. 18.17). The output pin is a snug fit
through the capacitor center arbor, and this pin is isolated from the outer wall.
The capacitor is soldered to this outer wall washer (ground). The inductor and
FIG. 18.16
The installed feed-through filter.
238
FIG. 18.17
Chapter 18
Construction of the single-phase feed-through.
capacitor are soldered to the output pin. The outer tube is placed over the inductor
and capacitor and soldered to the end plate. The inductor is soldered to the input
pin pressed through the threaded area and extends into the Id of the inductor. This
top capacitor and threads are soldered to the tube, completing the unit.
18.7. LOW-CURRENT FILTERS
The low-current filter suffers from conditions opposite to those that affect the
high-current filter. Here, the inductors get bigger while the capacitors get smaller.
One way is to employ RC filters in which the value of R should be less than 10%
of the minimum load resistance. The disadvantage is that the circuit gives only 6
dB per octave. A better method for low-current filters is to employ active filters
(Fig. 18.18). They are small and light and can be designed with many poles, but
sometimes the higher frequencies suffer because of the open-loop gain of the
OpAmp. It is often overlooked here that the DC feeding OpAmps must be very
clean. Therefore, the filter(s) needs a filter (Fig. 18.19). There have been cases in
which a number of high-impedance lines must be filtered. The plus and minus
U?
OPAMP
FIG. 18.18
The active filter.
Matrix Applications
FIG. 18.19
239
The filter required to filter the active filters.
voltage for each OpAmp must be filtered with an RC filter or a conventional

passive filter, with the capacitor facing the load. The combined voltage feed must
be filtered with a passive filter.
R >>
B+
10Imax
If the voltage is 12 V and the maximum current is 10 mA, the maximum value
of R is 120 ohms. Make R 100 ohms, and the capacitive reactance value is also
100 ohms at half the needed cutoff frequency. Note that this gives a 1 V drop,
and if this is excessive, either the resistance must be lowered or the voltage supply
must be increased to allow for it. If 14 kHz is the OpAmp cutoff frequency, then
at 7 kHz, C is equal to 0.22 F; and round this off to a standard value. The same
is true for the other B, supply. If there is a group of these, say 10, then the total
B+ current is 100 mA. This gives an impedance of 120 ohms.
The impedance of the inductor should be 10 times the 120 ohms just
calculated, and the impedance of the capacitor should be one tenth this impedance
of 120 ohms. A good quality 1 F capacitor should remove the noise so that one
OpAmp does not add noise to the next OpAmp. The inductor removes any of
these signals from the main supply.
240
Chapter 18
18.8. F0THE EASY WAY

The filter can almost be designed without any programtruly, on the beach of a
desert island with wet sand and a sharp stick! In other words, filters can almost
be designed without a calculator if the design is simple. This means that the
design should be without RCSHU or Cauer circuits. Let us look again at the
single-phase section. The goal was 46 dB, with headroom, at 20 kHz. If the
engineer selected the L, as before, this would give 12 dB per octave or 40 dB per
decade of attenuation. The beached engineer would use his sharp stick to write
the values in Fig. 18.20 in the wet sand.
The designer divides the frequency by 2 and subtracts 12 dB for each row,
opting for the single L. It is obvious at the 5 kHz at 22 dB level that a single L
would not work for this 400 Hz power system. This is because the next row is
half this value of 5 kHz. This figure is below the minimum of 4 kHz that should
be followed for better design that keeps the filter transparent to both the line and
load. Look how close this lonely filter engineer was to our reading of 1260 Hz
for 46 dB of loss!
The engineer has many miles of wet sand and starts over a few feet away
in new clean wet sand with a double L at 24 dB per octave (Fig. 18.21). The engineer
concludes that the double L will work and that the cutoff frequency is 5000 Hz. The
double-L value of F0 was 4600 Hz, which is close. The 10 ohm design impedance
must also be known, so the value of the inductors and capacitors can be calculated.
For the inductor, the engineer scratches the following:
10
= 318 H
25000
and obtains 318 H. For a balanced filter, this is divided by 2, giving 160
Hnot all that far away from the 175. Actually, the value of 3 for would be
FIG. 18.20
Frequency
dB
20 kHz
46
10 kHz
34
5 kHz
22
2.5 kHz
10
1.25 kHz
-2
F0, the easy way, chart 1.
Matrix Applications
FIG. 18.21
241
Frequency
dB
20 kHz
46
10 kHz
22
5 kHz
-2
close enough for beach work. This would have given the value of 333 H. The
capacitor followed on more fresh sand a few feet over (I hope the tide does not
come in)!
1
= 3.18 F
2500,010
The designer knows that dividing the inductance value by the square of the
impedance can solve the capacitor value. This is 10 ohms and squared is 100
(much easier to do in the wet sand); again, this is close to the 3.5 used before.
The designer would round this up to 3.2 F anyway. The Zorro remains to be
calculated, and the designer knows that the feed-through limit is 0.02 F, or
paralleled giving 0.04 F to ground (Fig. 18.22). Note how close this is to the
3550 earlier. The Zorro inductor value follows:
10
= 509.3 H
23125
FIG. 18.22
Frequency
dB
50 kHz
46
25 kHz
34
12.5 kHz
22
6.25 kHz
10
3125 Hz
-2
242
Chapter 18
and the capacitor follows but should be held to two times the maximum to ground,
or 2 0.02 or 0.04. Again, the value of , for beach work, could be rounded to
3, but the value is 5.09 F. Dividing the 5.093 by 0.04 tells the engineer that the
capacitor is 127.32 times too big! So the beached designer multiplies the inductor
by this figure, 127.32, and obtains 64.8 MH. This value is somewhat close to the
50.4 MH from the earlier design. This value of inductance is also too big as was
concluded in the prior design. This engineer, similarly, adds another L filter,
making the common mode a double L. The double L has 24 dB of loss per octave
(Fig. 18.23).
The calculation of the inductor follows:
10
= 127.3 H
212,500
and the capacitor:
1
= 1.273 F
21,250,010
which is 63.65 times too big as compared with the total of 0.02 F (four
feed-throughs of 0.01 each), so the inductor is multiplied by 63.65, giving 8.1
MH, which is close to the earlier 7.9. Note that the impedance of the common
mode filter section is no longer the 10 ohms.
and the cutoff frequency is still 12,500 Hz:
The final design is shown in Fig. 18.24.
FIG. 18.23
Frequency
dB
50
46
25
22
12.5
-2
Matrix Applications
160 H
8.1 MH
.01 F
.01
3.2 F
160 H
3.2 F
.01
160 H
.01 F
FIG. 18.24 The fully designed filter using F0, the easy way.
243
244
FIG. 18.25
Chapter 18
Frequency
dB
160 kHz
16 kHz
60
8 kHz
78
4 kHz
96
Using the dB loss per decade and also the dB octave loss.
This technique of trading inductance for capacitance can be used in

common mode filter design but not in differential, or normal, mode design
because the design impedance changes. There is no need to match the impedance
in common mode. Also, the two Zorros lowers the circuit Q because the
inductance is so much lower than with the single Zorro. The Q was already very
low in common mode. Thus, the increase does not enhance oscillation. This
technique, if used in the differential mode, would raise the Q in the differential
mode, which is already too close to the edge of oscillation. Also, the common
mode load impedance of the actual equipment, in most applications, is much
higher than the differential mode impedance. This fights against the common
mode Q increase and reduces the chances of oscillation. The same technique
works on Ts and s except that 18 dB is used, replacing 12 dB for the L with
.5L1
.5L1
LINE
LOAD
C1
FIG. 18.26
The T used to show the element count.
Matrix Applications
245
36 dB Loss @ 100 kHz
86 dB Loss @ 800 kHz
Frequency
Loss
Frequency
Loss
100 kHz
36
800 kHz
86
50 kHz
24
80 kHz
46
25 kHz
12
40 kHz
34
12.5 kHz
20 kHz
22
10 kHz
10
5 kHz
-2
FIG. 18.27
The dB loss at two different frequencies.
18 dB for the Ts and s as the frequency in the chart is divided by 2. Use 30 dB

per octave for two or multiple networks, adding 12 dB for each network added.
Three networks would give 42 dB per octave. Count the elements of the entire
network: the double T has five elements at 6 dB per octave each, giving 30 dB
each octave. If the frequency is high enough, the decade system can be used and
the decade and octave can be used together. As an example, a T has 18 dB per
octave and 60 dB per decade. If 90 dB is required at 160 kHz, will the filter be
okay for a 400 Hz line frequency? Add 6 dB headroom for a required loss of
96 dB. The first step was the decade followed by two steps of octaves. The filter
should be okay (Figs. 18.25 and 18.26).
Again, the filter must be checked at each frequency listed in the specification to see that the filter meets the assigned loss at that frequency. One specification was as follows. 30 dB at 100 kHz and 80 dB at 800 kHz. Adding 6 dB for
headroom, 36 dB loss is required at 100 kHz and 86 dB at 800 kHz. Using the
single L, the loss chart is shown in Fig. 18.27. Most would think that a filter
designed for the lowest problem frequency would be good enough to handle all
the rest of the listgood odds but not always true as in the preceding example.
The single Ls cutoff frequency must be 5 kHz, not 12.5 kHz
The easy method for the computer or calculator is as shown in Fig. 18.28.
246
Chapter 18
Required Loss
dB
The Loss per Octave for filter
Number of Octaves required
Frequency @ needed Loss
The Estimated Cutoff Frequency
FO
dB
X = ------L
F
FO = -----X
2
F
FO = -------2
FIG. 18.28
dB-----L
F0, the easy way with calculator.
18.9. REMOTE HIGH-VOLTAGE SUPPLY FED

FROM A LOCAL DC POWER SUPPLY
A very capable power supply company had a contract to design a power supply
with several output voltages. One was 60 V at 0.5 A. After the design, they built
and tested three and sent two to the customer for approval. After several weeks,
the approval was received and the well-designed EMI filter at the power input
also passed the 461 specification. Two days later, a panic call came from the
customer. The request was for them also to design a remote power supply utilizing
the 60 V at 0.5 A for a built-in remote monitor. This monitor power supply
requires 16 inches of cable. A key player had died, so the other group could not
finish the design on time. The output was 20 kV at 0.8 mA. The power supply
company designed a double-ended forward converter with each half conducting
with a duty cycle of 0.8 of its conduction time with a switching frequency of
50 kHz. This makes the current pulses 100 kHz at an on-time current of 0.625 A.
The questions are:
1. With a proper filter at the front end of the main supply, is another filter
needed?
2. Where would the filter, if needed, be located: at the output of the
original supply, somewhere in the cable, or at the input of the new
supply?
Matrix Applications
247
3. Besides the obvious snubbers etc., what else could be done to reduce
the size and cost or to eliminate the filter?
4. Do we care what the 60 V power supplys output impedance is at
100 kHz?
The answer to the first question is yes, and the reason will be apparent as the other
questions are answered. The second answer is, if needed, at the input of the
remote supply because the current on the cable between the two supplies needs
to be steady DC with as little ripple as possible. This is to eliminate radiation to
other wires within the harness and also to the surrounding areas. For number 3,
if the input section of the remote supply was isolated, requiring a return lead to
the first supply, a twisted pair would help eliminate the radiated emissions.
Although the filter is designed properly, there will be some ripple on the cable
after filtering. A shielded pair would also be good but ground the shield only at
one end. The best end would be the 60 V end, to reduce ground currents. Also,
Capcon could be used to cover the wires between the two supplies and dissipate
the energy.
Before the last question is answered, the following should be understood. What if the original designer knew that the remote switcher frequency
was 100 kHz and designed the 60 V output impedance to be milliohms
around this frequency? What would the results be? The path of least resistance for the 100 kHz current pulses is through the cable, creating high-level
radiation to the surrounding media. Discounting radiation, the rest of this circuit should work fine. A filter at the input to the remote should reduce this
radiation.
Now the opposite condition: the original engineering group were unaware
of the purpose of the 60 V, and the output impedance is 24 ohms in the 100 kHz
region. What would the results be now? The radiation would be reduced, but the
switcher may not function properly at the reduced voltage. The peak on current
is 0.625 A, reducing the supply voltage to 45 V. A filter at the input to the remote
should reduce this switcher drop or starvation.
Note that the solution to both problems is the same: a filter. A good filter
reduces the peak-to-peak ripple level, reducing the radiation. This filter also
reduces the output impedance to the switchers following the filter located in the
remote power supply. The answer to question number 4 is no. The system needs
a filter whatever the output impedance and will fix either problem.
To design the filter, all the processes covered in this book could be gone
throughthe dBuA/MHz, the conducted emissions, and the radiated emission
but this takes time and involves testing costs. The filter can be designed without
all this and may require some final adjusting, but so would the system after all
the time and expense.
248
Chapter 18
The easiest way is to divide the 60 V by the 0.625 A on current and get the
on impedance of 96 ohms. Divide this by 10 to get the filter capacitor impedance
at 100 kHz of 9.6 ohms. The value of the capacitor is
1
= 0.166 F = 0.2 F
2 100,000 9.6
or round this up to 0.2 F and pick a type having a self-resonant frequency (SRF)
at least 10 times the switcher frequency. This is 100 kHz, in this case, giving
1 MHz SRF. The inductor should have 10 times the 96 ohms, for this example,
or 960 at 100 kHz.
960
= 0.002 H = 2 MH
2100,000
This is split with half in the hot side and half in the return side of 1 MH each
(Fig. 18.29). This is a large value of inductance but easy to design for the DC
current of 0.5 A, although the SRF may fight. Make it as high as possible.
The filter impedance is
which is close to the 96 ohms, and the cutoff frequency is
The approximate loss is shown in Fig. 18.30.

This should knock the socks off the radiated emissions and give a lowimpedance source to the switcher. The pulse power is
1 MH
1 MH
FIG. 18.29
.2 F
The remote filter in the monitor power supply.
Matrix Applications
FIG. 18.30
249
Frequency
dB Loss
8 kHz
16 kHz
12
32 kHz
24
64 kHz
36
100 kHz
42
The monitor filter loss.
2Ea 106
20 log
which equals
2 0.625 8 106
20 log
= 120 dB
10
The 20 dB per decade starts at

1
1
=
= 39,789 Hz
a
8 106
which is 39,789 Hz, and the level at 100 kHz is
1
1
20 log
= 8 dB
= 20 log
5 8 106
Fa
10
where a is in seconds and F is in Hz.

With 42 approximate loss in the filter and 8 from the wave, the full loss
would be 50, and subtracting this from 120 would leave 70 dBua. This should
be adequate because this is not on the power line and what does radiate is reduced
by distance. However, this may still be out of specification, depending on the
requirements in the specification. If this is out of specification, add the shielded
balanced cable first, with the ground at the initial supply end, first. If the system
still fails, add another, identical filter. This should give another 42 dB, so now
the level would be around 30 dBua and must be adequate. Do not try to solve this
problem by going to a double L with smaller values. This could work, but the DC
voltage drop feeding the switcher could be excessive.
19
Applications Using Round or
Square Conducting Rods
Two similar conditions are treated in this section. The first is the very high current
filters based on toroids and round conducting rods through them. The capacitors
are the feed-through type, also threaded onto the same conductive rod. The second
method uses square conductive rods, but the capacitors are not centered on the
rods as in the first method. They are 90 degrees to the rod, and the feed-through
bolts are threaded through holes in the conductive bar. In some cases, the
conductive rod, or bar, is made from material other than copper because of
the strength needed to support the torque of the nuts. This is especially true in the
first case following requiring aluminum. Make sure to use a large enough bus or
bar so that the heat rise is low. In some, heat sinks have been inserted between
the cores making up the inductor. The capacitors either surround this bus or are
tightened against this bus, in the second system, and so the bus temperature must
be low. This section discusses their major disadvantage. The best solution is not
to require them! Several thoughts are discussed here, with some ideas of how to
design around them if the need comes up.
19.1. VERY HIGH CURRENT FILTERS
First, three ideas are reviewed here. The first is the number of filters in cascade
or tandem. If the Q is low enough, the number of resonant rises is one less than
the tandem number in most applications. The maximum is normally four multiple
filters in tandem, which should yield three resonant rises. This holds only if the
Q is low enough; otherwise, there would be four resonant rises. In high-current
252
Chapter 19
filters, the rule of a maximum number in cascade, or tandem, is broken for several
reasons. The new maximum number is seven filters in tandem. For example, there
could be seven L filters one after another. This gives six potential resonant rises
if the Q is low enough. One reason for the exception of a maximum of four in
cascade is the high capacitor values required for proper design. They become very
unrealistic, and the self-resonant frequency (SRF) drops to low values for the two,
three, or four cascaded filters. As the number in tandem increases, the capacitor
value decreases. The inductor values have the opposite problem. They shrink in
value into the low H range. To fix this low value problem of the inductors and
to reduce the size of the capacitors, the second idea must be discussed. This is
paralleling the filters. If the number in parallel is N, the impedance of each unit
in parallel is multiplied by N. The value of the inductors is N times larger, and
the value of each capacitor is divided by N. The impedance of each filter unit in
parallel is
(19.1)
and the parallel filter impedance is the same as the initial impedance because of
N in parallel.
Another solution to the high-current filter is not to filter it! Filter the lower
current areas. The Army Corps of Engineers consistently ignores this idea. They
wanted an entire power line, with currents of 600, 800, and 1000 A per phase,
filtered rather than to filter the distribution legs, which would be easier. It is likely
that the sum total of the branch filters would be smaller in volume and weight
than the one big filter. These branch filters would probably cost less, also. The
filters ability to operate at higher frequencies, because of the improved SRFs,
would also greatly increase.
Another costly solution is to use transformers to step the voltage up and the
current down (Fig. 19.1). This raises the design impedance, making the components much more realistic. The higher design impedance increases the value of
the inductors, making them more realistic. This also holds true for the capacitors
and makes for smaller, more realistic capacitor values. Both components would
have better SRFs. Notice the second word, cost. These filters are as expensive as
gold anyway, so two transformers on opposite ends may not affect the price too
much. Also, the transformer adds to the differential and common mode loss,
cutting the cost of the filter. See Chapter 10.
Another way may be to filter the power before the step-down transformer
to save the cost of one transformer. The input transformer is either a wye to delta
or a delta to delta. Another advantage of adding the two transformers is that the
neutral current caused by unbalanced currents on the neutral is eliminated. This
Applications Using Round or Square Conducting Rods
253
FIG. 19.1 The step-up and step-down transformer for lower current filtering.
is true of all the harmonics that are multiples of 3 and especially the third
harmonic current. These add in phase on the neutral along with the neutral
unbalanced current and are eliminated. Filtering this as a three-phase three-wire
DELTA would also end the requirement to filter the neutral, which must be as big
as the filters in the other legs. (See Sec. 18.2 for more information.) Carry the
ground wire through to these filters because they must be well grounded to
function properly. The cost saved by eliminating the fourth neutral filter helps
offset the transformer cost if this technique is used. The current in each leg is still
high and may require several filters in parallel besides using the transformer(s)
technique.
The following is based on approximations that are useful for designing the
filter quickly. The values can be moved later when they are at least on the ballpark
property if not on the ballpark playing field.
The cutoff frequency, F0, the value of the inductance in microhenrys, and
the value of capacitance in microfarads are listed in Table 19.1. This is for 440 V,
400 A at 400 Hz requiring 100 dB at 14 kHz. These values were calculated by
using the methods discussed earlier through matrix algebra. The design impedance is 440 divided by 400, or 1.1, and multiple Ls were chosen.
TABLE 19.1 High-Current Unparalleled Filter Values

Number of
elements
Cutoff
frequency F0
Inductance
Capacitance
3750
4500
46
39
38.5
32.5
5
6
254
Chapter 19
The cutoff frequency of the five L filters in tandem is a little low for
400 Hz. This makes the six-element filter the better choice. But look at the value
of the capacitors required for either the five- or six-element filters. This would
require five or six capacitors this size and the inductors to make up the filter.
These values for the capacitors are much too high. Even if two capacitors were
paralleled, 20 F for the five elements and 16 F for the six elements, the SRF
would be too low. However, many EMI filter manufacturers use the GE-type
capacitors but the droves to make these filters. In some filters a total of 220 F
is common. The SRF of each capacitor would be in the low-frequency area around
60 kHz or lower, depending on the type of capacitor chosen. Paralleling the entire
filter would aid in solving this problem, as suggested in Table 19.2. The design
impedance is now 4.4 ohms, with four in parallel, and the current is now 100 A.
This would form a two by two, giving four filters in parallel. With this low current
(100 A per filter in parallel), conventional design could be used.
Four elements are much too low to consider for 400 Hz; five elements are
okay. The advantage of six elements is that the capacitance has been reduced from
9.2 to 8 F, giving a little higher SRF, and the same is true for the inductance. It
is obvious that this solution makes for a large number of elements. If the filter
engineer opts for the six-element filter, this equates to six by four inductors at
152 H and six by four capacitors at 8 F, or a total of 48 components. Big? Yes!
Heavy? Yes! Costly? Yes! (I hope you have to build more than one to help offset
the design costs!)
This is the disadvantage of the high-current filters. The best solution would
be to bring this power into the service, then filter the individual feeds after each
breaker. The current falls, requiring less elements in each filter and ending the
possible need for the two transformers and the need to parallel the filters. Some
prefer the opposite technique, to filter feed the breaker so that the filter is always
alive. This means that the distribution bus feeds the filters and the filters feed the
breakers. The reason is that most designs have too large capacitors in the filters,
and this creates difficulty in turning the power on because of the added inrush of
filter current. Look at Table 19.2, where six 32.5-F capacitors are in parallel.
The inductor furnishes low inductive reactance at the power frequency, so these
TABLE 19.2 High-Current Four-in-Parallel Filter Values

Number of
elements
Cutoff
frequency F0
Inductance
Capacitance
3000
3950
4600
233
177
152
12.0
9.2
8.0
4
5
6
255
give little isolation between the capacitors. This gives a total of approximately
195 F. This capacitance plus the normal inrush may cause the breaker problem,
but many designs have more capacitance, adding to the problem.
If the engineer is stuck and has to design this system, some ideas follow.
The current-carrying element is of bronze, brass, or other material, rather than a
wire or wires. The conductor rod should be round. Copper comes to mind but may
not support the end nuts, putting too much tension on each threaded end. This
stronger conducting material is often be plated with silver. This slightly improves
the current-handling ability of the rod. The bar or rod has a diameter specified to
handle the current flow with little heat rise. The conductor is not insulated in most
designs. Both the inductor and capacitor inner diameters should be slip fits over
this conductor without being too loose. The capacitors should be the feed-through
type with the arbor inside diameter a tad larger than the rod. The inductor is also
a toroid with the inside diameter slightly larger than the rod. These are slipped
onto the rod to make the filter elements. N, the number of turn(s) of the inductor,
is equal to one. This is the copper or silver-plated rod, with a 95% or better fill
factor compared with the typical 40% for toroids. This bus runs through the inside
diameter of the toroid as well as the capacitor. Later in this section, the bus bar
type of copper rather than the rod is discussed.
The approximate design equations for these high-current filters follow. The
inductance is
L =
0.4UeAcN2 108
Mpl
(19.2)
where N is now just one turn, Ac is the cross section of the core in centimeters
squared (times a stacking factor, Sf), Ue is the effective permeability, and Mpl is
the magnetic path length, also in centimeters. The core is tape-wound silicon steel
that has a soft core for its BH curve (Figs. 19.2 and 19.3). This inductor should
not be made using a square loop type of material. The tape thickness should be a
FIG. 19.2 Tape-wound toroid showing Id, Od, and Ac.
256
Chapter 19
FIG. 19.3 Toroid inserted on the copper bar.
size smaller than the size required for the line frequency, especially if the inductor
must be tuned as part of a Cauer. This is to prevent core losses at the harmonics
of the line frequency and harmonics from any off-line regulators. Otherwise, the
high harmonic current heats the core, requiring a larger core. Trade-offs of this
type are discussed later. The core manufacturer should also gap this toroid so that
the peak current does not saturate the core. This can be determined from the BH
curve, where H is
H =
0.4NIp
0.4
2 Irms
1.7772 Irms
=
=
Mpl
Mpl
Mpl
(19.3)
because N is equal to one. The last two equations imply that the majority of the
current passing through this filter must be close to unity power factor and the
off-line regulators are in the minority. If this is not true, the value of H is higher.
This core should be insulated, but some houses use quality tape. The inside
diameter must be maintained so that the toroid slips over the rod easily without
being too loose.
Then Ac and Mpl are
Ac =
SfW(Od Id)
2
Mpl =
(Od + Id)
2
(19.4)
where the units are in centimeters. L is

L =
0.4Ue(Od Id)WSf 108

= 0.4UeWSf 108
(Od + Id)
where Od >> Id. However, the MPL in Eq. (19.4) is often given as
MPL = (Od Id)0.5
(19.5)
257
Watch this: the poorer the conductive material is, the larger Id must be to handle
the current so that the bus does not have excessive heat rise. This heat rise could
destroy the capacitors. The dimensions are still in centimeters. The Id is the
diameter + of the conductor, and Od is the maximum allowable diameter based
on the size of the insert or filter box. The amount of inductance needed is known
because it was previously calculated. The effective permeability, Ue, approaches
Ue =
UmMpl
Od
=
Mpl + gUm
2g
(19.6)
where g is the gap, also in centimeters, and Um is the core permeability. This
equation is based on Um being in the tens of thousands range. The right side of
the denominator of the center equation is much larger than the left-hand term.
Replace Mpl with Eq. (19.4) and substitute Ue into Eq. (19.5):
L =
0.4OdWSf 108
2g
(19.7)
This is the individual value of each inductor used to make up the filter (six of
them in the preceding case before paralleling). Changing Od, W, and g to inches,
L =
0.2474OdWSf 108
g
(19.8)
Solve for the width, in inches, of the inductor. For the toroid specification used
here, the toroid dimensions given are Od, Id, and height. The W used here is the
height dimension.
W =
4.043Lg108
OdSf
(19.9)
In this way, the width of the inductor core in inches (here, the width is the height
of the core; there is only the conductor rod through the Id) can be found to meet
the required inductance. Make g vary with Od so that g/Od is a constant for each
filter in parallel. In this way, W varies with L. The stacking factor, Sf, varies with
the tape thickness only.
A good source for the toroid inductor, and also the C core discussed later
in this section, is National-Arnold (Adelanto, CA 92301; Richard H. Wood,
Engineering). Richard Wood has spent many hours doing research on pulse
transformers. He has concluded that the width of the toroid and C core should be
1 inch maximum to handle pulses properly. Therefore, stack the toroids and C
cores on the round rod and the bus to reach the desired width if the main feed is
to be pulses, for example, the current pulses from off-line regulators or power
factor correction circuits. The pulse permeability is lower than that specified, and
258
Chapter 19
the 1 inch width supports the pulses better, so use this technique if a high
percentage of pulses are expected. Otherwise, National-Arnold makes toroids of
almost any size needed up to 3 feet in diameter and width. Their loss graph in
watts per pound is shown in Fig. 19.4. For better information, obtain their catalog
for their tape-wound toroids and C cores.
Each toroid, or C core, is insulated from the rod or bus and from each other.
In Fig. 19.5, the spacing between the cores could be filled with a heat sink.
National-Arnold gaps the toroid, as in Fig. 19.6, by slitting the toroid, filling the
gap, and coating the toroid so that the gap defies detection. In some cases, the
toroid is cut halfway and then again cut fully across the core.
The Id and Od of the feed-through capacitor (Figs. 19.7 and 19.8) are the
same as for the inductor just calculated. The width must also be found. (See Sec.
7.2, which is mostly repeated here.) Ds is the outside diameter, Da is the diameter
of the arbor, and t is the thickness, all in inches. Ds is the outside diameter, Od,
and Da is the inside diameter, Id. Make Da, or Id, slightly larger than the conductor
rod and Ds, or Od, the same diameter as the high-current inductor calculated
before.
Solve for the number of turns, N:
N =
Ds Da
2t
Solve for the mean diameter (Dm)
FIG. 19.4 The National-Arnold core losses in watts per pound.
(19.10)
259
FIG. 19.5 Stacking of tape-wound toroids on the copper rod.
Dm =
Ds + Da
2
(19.11)
Solve for the mean length of turn (Lm)

Lm = Dm
(19.12)
Solve for the active length (Lt)

Lt = NLm
Substituting Eqs. (19.10), (19.11), and (19.12) into Eq. (19.13),
FIG. 19.6 Tape-wound core gapped by National-Arnold.
(19.13)
260
Chapter 19
FIG. 19.7 The extended foil capacitor.
Lt =
D2s D2a
4t
(19.14)
To continue, we need to know the K value of the chosen material and find
the dielectric thickness in mils. Values of K are also a function of shape, and the
dielectric is often chosen on the basis of the dissipation factor and size. Mylar
gives the smallest dissipation factor within its normal temperature range.
Polyester (Mylar), 900
Polycarbonate, 840
Paper (Resin or PBT), 580 (wet)
Solve for the active width (Aw)
Aw =
CKT
Lt
(19.15)
FIG. 19.8 Blowup of capacitor plates and dielectrics. T is the dielectric thickness
in mils and t is the thickness of two dielectrics and two foil plate thicknesses in
inches.
261
where T is the thickness of the material in mils, C is the capacitor value in F,

and K is the dielectric constant from before.
Substituting Eq. (19.14) into Eq. (19.15),
Aw =
4CKT
(D2s D2a )
(19.16)
If Ds, the Od of the inductor, and Da, the Id of the inductor, have been
decided, the width to be allotted for this capacitor can be found. This equation
helps to find the width required for the capacitor. The final width is slightly larger
than Aw + 2M, where M is the margin for corona. The extra width is to allow for
the extended film foldover and sputtering. This conductive tab should be soldered
to the conductive bus bar, and the same style of tab should be soldered to the
ground end of the capacitor. This should be soldered to a ground lug or to the
enclosures wall. Some extra width should be allowed for these two tabs. The tabs
must carry the capacitor current, not the full current of the filter section.
Figure 19.9 shows how the capacitor was slipped onto the conductor
(following an inductor) and soldered to several ground points, first to the ground
lugs and then the other side soldered to the center conductor. A toroid is slipped
onto the conductor afterward. The inductor is coated, so voltage is not present on
the surface of the core.
This continues until all the components are slipped over the conductor.
Another way to mount the capacitor is to solder or sweat the capacitor to a
partition that helps to isolate the compartments and forms a screen and shield.
These are slipped down the conductor with the capacitor first, and the shield is
soldered in place to the container. Be careful not to heat this to the point where
FIG. 19.9 The stacked capacitor and inductor on the copper conductor rod.
262
Chapter 19
the capacitor solder melts, disconnecting the capacitor from the shield; spot
welding may be better (Fig. 19.10).
The conductor rod is threaded on both ends. The rod is slipped into its
container box with the top and front cover removed and slipped through a
centered hole in the back cover. A ceramic insulator is slipped on over the threads
and centered in the hole of the back cover. A washer, smaller in diameter than the
insulator, is slipped on with the nut. This forms the output terminal. The components are slipped on one at a time over the long exposed conductor via the open
front and top of the filter container. Each capacitor is soldered before feeding on
the next inductor (toroid) until all the components are installed. The front cover
is then installed and soldered, followed by the ceramic insulator and then the
washer and nut. This forms the input side of the filter. The filter is tested and
cured in an oven with a vacuum and then tested again. Often the feed-through
capacitor requires additional torque to ensure proper ground and terminal contact.
Plated brass rods are often used because they take the required torque. Then the
top cover is soldered in place. A potting compound is often used to hold the
elements in place. This unit should not require any if the components are snug to
the conductor (this assumes the rod is rigid). If the rod is loose or if the
specification calls for potting, use just enough in the bottom to hold the components in place or stationary. These high-current filters are heavy enough without
the potting compound. To keep the filter as light as possible, use microballoons
mixed with the potting compound.
FIG. 19.10
The capacitor with ground plate.
263
The advantages of filters in parallel are as follows (Fig. 19.11):

1. The current reduction drops the silver-plated rod diameter by
approximately
Od =
Id
N0.5
(19.17)
where N is the number of filters in parallel, not the turns. This reduces
the Id, giving more inductance and capacitance for the Od.
2. The inductor value becomes reasonable by being multiplied by N.
The capacitor value is divided by N, making the capacitor smaller and
more reasonable. This improves the capacitors SRF and the inductors
efficiency.
3. H is reduced to H/N, reducing the size of the gap needed, where N is
the number of the filters in parallel, not the number of turns. The value
for turns is still equal to one.
4. The lower current filters may now be conventionally designed filters in
parallel rather than the type discussed here.
The equation for the inductor width [Eq. (19.9)] repeated here is
Wl =
404.3gLuh
4.043Lg108
=
OdSf
OdSf
(19.18)
This is the initial full-sized filter, before paralleling. The value of the paralleled
inductance is equal to the individual inductor of Eq. (19.18) times N, the number
in parallel.
FIG. 19.11
The high-current filter showing three sections.
264
Chapter 19
Wln =
403.3NgLuh
OdSf
(19.19)
Returning to the initial full-sized filter, before paralleling, the value of

paralleled capacitance, then, is equal to the individual capacitor divided by N.
This is the capacitor width divided by N:
Aw =
4CKTt
(D2s D2a )
R =
4CKTt
Wcn =
R
N(O2d I2d)
(19.20)
Wcn is the full capacitor width, so this is again the initial value of the capacitor
before paralleling the filters. This means that Wc, must be divided by N. The total
length of the filterback cover to front coveris approximately
404.3gNLH
R
+
Wlc = 1.1S(Wcn + Wln) = 1.1S
2
2
OdSf
N(Od Id)
(19.21)
where S is the number of filter elements (the number of L filters in tandem here,
six) and N is the number of filters in parallel. The multiplier, 1.1, is a factor
allowing for the margins, the capacitor lead thickness, the shield (if used), and
some for soldering. R is found from the initial capacitance needed, in F, before
paralleling. LH is the inductance in H before paralleling. The inductance and
capacitance were determined initially from the number of filter elements in
tandem (S). S must be an even number for those folded back and even or odd for
those in line.
The volume for the full filter, including all those in parallel, is the full filter
length Wlc, from Eq. (19.21), times the Od squared times the number of filters in
parallel, N.
404.3gNLH 2
R
Vtot = WLCNO2d = 1.1S
+
NOd
2
2
OdSf
N(Od Id)
(19.22)
If Od is again >> Id, the denominator of the first term cancels, reducing the first
term to R. The second term is directly multiplied by N times Od squared.
404.3gN2OdLH
Vtot = 1.1SR +
Sf
(19.23)
Whatever the method, this volume remains the same for all the different
numbers in parallel. Pick a combination that gives the filter a nice form factor,
where the filter is not a cube or 60 feet long, 4 inches wide, and 8 inches deep.
These can be folded back on themselves, dividing the length by two and twice
the depth. Figure 19.12 is a nice form factor in which the length is twice the
diagonal.
FIG. 19.12
265
Diagonal-to-length ratio.
The easiest method is to calculate the diagonal of the end piece compared
with the length. Twice the diagonal is normal and about the minimum, and three
times the diagonal is about maximum. There are three equations. The first
equation is for straight, where the Od is the depth and N times Od the width. The
second has the same width, but the length is half and the depth is twice as large
or twice the Od. This is folded back, so that the input and output terminals are on
the same face (see Figs. 19.11 and 19.12). The last is the even count type, such
as 4, 6, 8, and 9, which gives 2 by 2, 3 by 2, and 4 by 2, which requires the full
length, and the width and depth are multiples of Od as before. These equations
cannot be differentiated to solve for a minimum because all that I have tried give
points of inflection.
The three equations for the length to equal twice the diagonal follow with
the assumption that Od >> Id. Any other ratios for the equations are easily
established. Again, watch for higher Id dimensions because other materials
require a higher cross-sectional area than copper and the Od may not be greater
than the Id:
R
+ 404.3gN2OdLH = 2O3dN
N2 + 1
1.1S
2
2
N(O
I
)
d
d
404.3gN2OdLH
R
3
0.55S
+
N2 + 4
= 2OdN
2
2
Sf
N(Od Id)
(19.24)
404.3gN2OdLH
R
3
1.1S
+
N2 + 16
= OdN
2 I2)
S
N(O
f
d
d
The first equation is for one filter deep, or the depth is Od, and N times the Od
wide. The second equation is for the folded-back type, so the depth is again twice
the Od, the width is still N times the Od, and the length is half. The last is for the
even count 4, 6, 8, and 9. Again, parallel filters are rarely more than six. Surely
266
Chapter 19
by this count the conventional-type filter design could handle this reduced
current. Here, four in parallel would give 100 A each for the case listed in
Table 19.3, and a conventional filter could handle this current. These form cubic
equations. For the current of 400 A, the values are as in Table 19.3 for the filters
in parallel. Each listing is for one through nine filters in parallel. Rarely are more
than four in parallel ever used.
The single filter giving an impedance of 1.1 ohms is ridiculous. Look at the
diameter of the inductors and capacitors. I am not sure that the capacitor can be
wound, because most extended foil winding machines can handle about 6 to 8
inches in diameter. The inductor can be as big as 3 feet in diameter and the same
width at National-Arnold in Adelanto, California. The volume is good, and so is
the width (this should be length now). Two in parallel, giving 2.2 ohms each, are
better, giving a smaller diameter and less width. All these are Od depth elements
and N times the Od wide.
Figures 19.13 and 19.14 are for the folded-back type.
The rear terminals are all shorted together, and the normal metal protective
shroud, which covers the entire rear, is not shown. The same is true for the front,
where there is also a division between the top and bottom halves. The same idea
is used for three, five, and seven (rarely used). These are all folded back as in
these figures. The number of stages (S) must be even to split between the top and
bottom halves of the two folded-back designs. The length is cut in half (plus some
for the shrouds covering the front and rear connections). Four is a two by two,
six is a three by two, and eight is a four by two, requiring the full length. Nine is
a three by three, also running the full distance. The approximate sizes for the five
folded-back designs are listed here. Again, long before this many filters are in
parallel, the design can revert to conventional filter types.
For the five in parallel and folded back for 400 A (now 80), R is 601.38 and
Q is equal to 0.3558 with each inductor = 195 H. Each capacitor is equal to
TABLE 19.3 Chart for One to Nine Filters in Parallel

Z
F0
Irms
dIA
Width
1.1
2.2
3.3
4.4
5.5
6.6
7.7
8.8
9.9
6
6
6
5
5
5
5
5
5
4400
4475
4530
3850
3900
3900
3930
3930
3960
39
78
115
181
224
269
311
356
397
32.8
16.2
10.6
9.4
7.4
6.2
5.3
4.6
4.0
400
200
133
100
80
66.6
57.2
50
44.5
11.5
8.0
6.4
5.8
4.9
4.4
4.0
3.6
3.4
32.5
35.8
40.0
52.7
57.5
62.0
66.3
70.3
74.1
FIG. 19.13
267
Side view of two filters in parallel folded back.
6.6 F. The required diameter is 4.5 inches. F0 is 4436, which is very good. The
folded length is 28.2 inches. Z is 5.5 and S is equal to 6, so the bottom and top
component count is even. The volume is 2.9 ft3. Granted, this is only an
approximation, but if a 400 A filter can be built in less than 3 ft3, this is very
good. The inductors would be 1.8 inches wide, and the capacitors would be 6.5
inches wide. The capacitors may have to be split, requiring 3.27 inches each.
Winding machine can handle various sizes. Extra room is allowed for soldering,
width of contacts, and margins. It should be realized that the use of more than
four in parallel is very rare but is shown here for comparison. Also, the filter
design could revert to the conventional design techniques with this lower current
of just 80 A. This high-current type of design really starts at 100 A or so,
depending on the line voltage.
The same technique can be used on rectangular buses using C cores of the
same material instead of toroids. The capacitors must be wound on a special
rectangular arbor and so require more care while winding. It takes four to five
268
FIG. 19.14
Chapter 19
Front of two filters in parallel folded back with rear view below.
times as long to wind them, making them much more expensive. Often the
dielectric is composed entirely, or partly, of reconstituted mica because of the
abrupt turns required for Mylar. This overstresses some of the dielectrics.
The strap in Fig. 19.15 is stainless steel, not strap iron as used by a designer
some years ago when he ran out of stainless steel. The strap iron carried almost
all of the flux around the two gaps and melted the strap in a few seconds. These
cores are also available from National-Arnold.
Mount the capacitor as in Fig. 19.16 with either a shield or leads. This
design technique is summarized in Sec. 19.3.
The approach just discussed should be avoided. It costs much more than
breaking the filters down into smaller currents after the service panel. In this way,
several filters of the same size can be used for different feeds rather than filtering
the entire line. If three or four EMI filters of the same size are used, a spare is
not unreasonable. The sum total of the filters plus spares may still be less than
the one single filter.
FIG. 19.15
The C core showing shims and clamp.
269
270
FIG. 19.16
Chapter 19
The square capacitor used for copper bar.
19.2. HIGH-CURRENT SECOND METHOD

Each inductor assembly (C core) surrounds the bus, each capacitor assembly has
its own rod or bus, and these are tied together at the main bus (Figs. 19.17
and 19.18).
The three capacitors are the feed-through type, but the bolt carries no
current as the normal feed-through would and as the last feed-through capacitor
in Fig. 19.17 would. The capacitors carry only the normal capacitor leakage
current really reactance current to groundbut this is not through the bolt.
These three feed-through capacitor bolts are hot. The plate that the feed-through
is bolted to is ground, and the bolt is isolated from the plate. The three capacitor
foils are pressed to the ground plate, giving good SRF, equivalent series inductance (ESL), and equivalent series resistance (ESR). The extra ground plate can
be avoided if these three bolts are nonconductive and do not need to be isolated
from the plate or the outside filter wall. The nonconducting bolts, here, would
FIG. 19.17
Second method for high-current design, drawing A.
FIG. 19.18
271
Second method for high-current design, drawing B.
protrude through the wall of the filter and could be touched without being
shocked. The advantage of this style over the style using the ground plate is that
the three capacitors, along with the last feed-through, could be torqued again after
the heat cure from outside the filter body. In the first style, the cover over these
three bolts must be left off through the heat cure for access to these bolts.
This torque is required for proper pressure for both capacitor foils of each
capacitor to function properly. There are several advantages of this style over the
type in Sec. 19.1:
1. In the first style, the Id of all the capacitors had to be the same as that
of the inductors. In the second style, the Id can be very small for the
three capacitors to ground. In other words, this capacitor can also be
crushed or flattened in some designs. This is not true for the output
feed-through capacitor. This output capacitor bolt carries the full filter
current to the outside world. The bolts through the capacitors to ground
carry no current (unless it is conductive and shorted) to the outside
world. The output feed-through capacitor bolt must be able to carry the
full load current to the outside world, but this is a much smaller
capacitor value.
2. The opposite is true of the inductor. The bus can be as big as needed to
carry the full filter current through the center of the C core because this
second style, the inductor bus, does not pass through the Id of the
capacitor.
3. With the second style it is easier to adjust the tension required for
proper capacitor action.
All the methods that can be developed to design these high-current filters
require iterations. As the Od grows, with inductance and capacitance values held
272
Chapter 19
constant for the proper design, the width of the capacitor decreases. This increase
of the value of Od increases the Mpl, which decreases the Hthe magnetomotive
forcewhich moves the flux density away from saturationand increases the Ac
value of the inductance. If the Ac and Mpl are growing at the same rate, this holds
the inductance constant. This increase of Mpl requires a larger gap to hold Ue
constant; otherwise, the inductance increases. So, around and around it goes.
The disadvantage is that the bus must be oversized to allow room for the
capacitor feed. A copper bus is not very big to handle the current, 400 A here.
Copper can be used here because no torque is required, but often poorer
conductors with very conservative current densities are used. This gives the
needed width to support the capacitors to ground but also increases the filter
volume. It is a trade-off. Cm is the circular mils per ampere required and Wa is
equal to F times G of the C core.
Wa = 7.854 CmIrms107 square inches
The question is, can the capacitor fit on the bus? A top view of the capacitor and
inductor assembly along with the equations is shown in Fig. 19.19.
The approximate value of Od can be calculated from Fig. 19.19 because
all the quantities are known. Aw should be at least the C core dimension E
FIG. 19.19
Top view of C core, bus, and capacitor with equations.
273
(Fig. 19.20). The Od should be, at maximum, the C core dimension G. If thin
shims or washers are added to the bus end of the capacitor, the Od can extend
beyond the length of G to a maximum of G plus twice E. If the Od of the capacitor
is greater than this, the volume of the filter increases some.
For the inductor, the D dimension can be calculated from the equations just
listed and knowing the gap, Mpl, Ac, and Wa:
Ac = 6.432EdSf
Mpl = 5.08(G + F + 2E)
Wa = 6.452FG
Ap= WaAc = 41.62DEFGSf
where D, E, F, and G are in inches and Ac, Mpl, Wa, and Ap are in centimeters.
Knowing the value of the inductor, the needed width D can be found. The core
manufacturer can make these cores to the needed width, D, or several cores can
be stacked to reach this dimension. If the capacitor Od dimension is less than G,
the capacitors can be paralleled (Fig. 19.21).
The advantage of this technique is that the capacitors can be paralleled
across the bus (Figs. 19.22 and 19.23). This reduces the volume required for the
capacitors, bringing the C cores closer together and cutting the volume of the
filter. The bus area of the cross section is constant for the current required to be
FIG. 19.20
C core with dimensions.
274
FIG. 19.21
Chapter 19
Second method showing capacitors in parallel.
handled by the bus bar, so as the width increases for the parallel capacitors, the
thickness drops. There is a limit to this because the bus must fill the C core
window area (Wa). Therefore, there is a maximum ratio of width to thickness,
which is limited to about 5:1. Otherwise, the C core cannot be manufactured. The
cross-sectional area is a constant that varies with the current that the bus must
carry. This main bus can be copper because the bus is not threaded and is not
torqued as before.
Therefore, the bus is not wide enough to hold more than two capacitors
across it. Otherwise, the capacitors across the bus will extend beyond the C core
G dimension (window length). Also, this could extend beyond the G plus twice
the E dimension (G + 2E = the full length of the C cores) at both edges, increasing
the volume if shims or washers are added between the capacitors and the bus. For
conditions when copper must be used, some other ideas are given later.
The only difference here is that C is half the value. There is also a practical
limit, and keep in mind that each hole in the bus restricts the current. Sometimes,
FIG. 19.22
275
Top view showing parallel capacitors and equations.
thin copper busing is placed across the bus to make up for the cross-sectional area
hogged out for the capacitor bolts. This technique adds to the length of the bolt.
It is done to maintain the cross-sectional area of the bus at the point where the
bolt holes are drilled, keeping the bus temperature lower to lengthen the life of
the capacitor.
FIG. 19.23
Second method for high-current filter showing added bus.
276
Chapter 19
19.3. HIGH-CURRENT METHOD THREE

This technique gets rid of the bolt through the center Id by using Y capacitors with
very short leads. The self-resonant frequency approaches that of the feed-through
type and has little added equivalent series resistance or inductance over the
feed-through type. The advantages are as follows:
1. Very little bus bar area is drilled out compared with the bolts in method
one or two.
2. The torque required for the feed-through type of capacitor is eliminated
because the capacitor is fully assembled and stressed prior to installation in the filter.
3. The leads are soldered, removing the necessary good contact with the
bus bar or ground plane.
The disadvantage is that the capacitors can rarely be reused because of the
difficulty of getting the short leads back in the holes in the ground plane.
The approximate equations needed to fit the capacitor are listed in
Fig. 19.24 and continue in Fig. 19.25. Aw is the length of the capacitor in the E
direction and is longer than E to keep the ground plane away from the C core
inductors. The C cores are insulated, but it is still not a good idea to be pressing
or touching ground. Thus, Aw is equal to E plus 1/8 inch or longer. This eats up
volume but reduces the Dx dimension. The values of C, K, T, and t are known,
FIG. 19.24
The pressed capacitor and equations.
FIG. 19.25
277
The pressed capacitor. Top view of C core and bus with equations.
so Lt can be solved from the equation in Fig. 19.24. Remember, t is the thickness
of two dielectrics and two foils in inches; see Fig. 19.8.
Also, T is the dielectric in mils. Then Dx can be solved from the quadratic
equation given in Fig. 19.25. The last half must be smaller than the first half under
the square root. This also ensures that the last half of the quadratic is smaller than
the first term. The negative solution is the only valid solution; otherwise, Dx
would be larger than Dy. Once Dx is known, Da, the arbor diameter, is solved and
Od also follows. In Fig. 19.25, the value is shown to be larger than G + 2E of the
C core. This is okay if Dy is smaller than the container with extra clearance.
Again, none of these Y capacitors should touch the container wall.
19.4. REVIEW OF HIGH-CURRENT FILTERS

Most high-current filters are made with pieces of these various technologies.
Some extend the width of the conductor bar for the capacitor(s). Others use the
vee technology for the capacitors with shunts across the vee legs to carry the main
current. Others mount the inductor sideways and reverse each section so that the
capacitors are on opposite ends (Fig. 19.26).
The two wires to and from the capacitor (Fig. 19.27) across the two buses
are in parallel with the single wire directly between the two buses. These wires
must carry all the current between the two buses. Otherwise, the diameter of the
wires must increase. This is veeing the capacitor. The inductance in both the
capacitor legs adds to the inductance in the C cores if the two vee wires can
handle the current. This keeps the SRF of the capacitor high. Otherwise, the shunt
wire inductance is in parallel with these two vee wires (Fig. 19.28). With the
278
FIG. 19.26
Chapter 19
High-current filter, alternative method.
parallel shunt bus wire, The SRF of the vee capacitor is still higher than in the
situation in which the capacitor has single long leads.
The main thought is to avoid these high-current filters if at all possible. If
the designer cannot avoid them, parallel the filters so that conventional lower
current filters can be used, avoiding the high-current design. This section is
intended to give the designer some thoughts on the design that will help if this
situation cannot be avoided. The equations listed earlier are marginal but should
get the designer close so that, with minor modifications, the filter will be
complete.
Another alternative method is to replace the feed-through capacitor with
pressed or flattened Y capacitors mounted on a bus connected between the two
buses running through the C cores (Fig. 19.29). The width of the capacitors should
be about the same as that of the bus, and the buses carry all the current except for
FIG. 19.27
279
High-current filter, alternative method two.
the leakage current through the capacitors. Several capacitors can be mounted on
the bus, giving each capacitor a higher self-resonant frequency. Other advantages
are that the overall volume is smaller, the total capacitance is the sum of those
along the bus, and the Y capacitors do not require torquing after the filter is heat
cured. The output feed-through requires torquing after the filter curing for best
capacitor results. If the capacitors along the bus can be spaced a little wider,
allowing ample room to solder the leads, capacitors may be on both sides on the
bus. The upper capacitor bus hole would be between two capacitors on the
bottom, for example. The Dx dimension of the capacitor would be narrower,
allowing the bus also to be narrower. The cross section of the bus must be the
same to carry the full current. The bus can be cut out of thick copper sheet or
bolted together. In this way, the bus between the C cores may be a different size
than the bus through the C cores. Various alternative methods may be developed
using many of these ideas in combination depending on the situation.
280
Chapter 19
FIG. 19.28
The vee capacitor in parallel with shunt wire.
FIG. 19.29
The last alternative high-current filter.
281
Using the preceding suggestion, the inductance along the bus adds to the
overall inductance and is similar to veeing. The capacitor lead length is low,
making ESL and ESR very low and keeping the SRF high. The disadvantage is
that all the capacitors are lost during repair or when adjusting for insertion loss
if components must be added.
19.5. THREE IN PARALLEL
In some cases, three filters have been paralleled. In one case, 1000 A was required
and three 333 A filters were paralleled to accomplish this (Fig. 19.30). High loss
was required, and four sections were needed for the loss. Several inductors were
tuned to meet the low-frequency loss (Cauers), and large capacitor values were
stacked between the inductorstuned or otherwise. Tuning works best when
low-impedance capacitors to ground are located on both sides of the tuned
inductors. Large copper buses shunted the three sections together on the input and
output terminals.
The capacitors were GE 20 F, 480 V AC, and eight of these were used
along with three feed-through capacitors at 5 F. These feed-through capacitors
were used to connect the input and output terminals along with the central
shielded sections. This was required because 100 + dB was required at 14 kHz
and above. To get 100 dB and maintain it through 1 GHz requires shielded, or
isolated, sections, and feed-through capacitors do this job well.
The input and output section is one 5 F feed-through and one 20 F GE.
The 40 F sections are two 20 F capacitors, and the 45 F is two 20 F plus a
5F feed-through capacitor. This last section is where the shield appears across
the enclosure. This separates the two filter sections. Dirt in one chamber cannot
easily be transferred into the next. This technique is necessary to maintain 100 dB
and below. This filter holds the 100 dB well through 1 gHz.
Sum the total capacitance to ground. Here, this is 175 F. At the line
frequency, the inductors are out of the picture. Divide the total capacitive
FIG. 19.30
One of the three filters.
282
Chapter 19
reactance into the line voltage to get the reactive current, which is 90 degrees out
of phase with the line current. The angle of lag can be determined, and the square
root of the sum of the squares gives the hypotenuse and it is surprising, even with
all this capacitance.
IC = 2 ERMSFC = 2 48060 0.000175 = 31.66
So the leakage current is 31.66 A. The full current is
IT = (3332 + 31.662) = 334.5
The angle of lag, in degrees, is
31.66
1
= tan1
= tan (0.095) = 5.431
333
The leakage current cannot be specified in this type of filter.
19.6. CONCLUSION
The conclusion is obvious! Try to avoid these high-current types if at all possible.
These types are most often required for secure rooms, vaults, and EMI test
laboratories. In the latter, high current may be required to drive equipment under
test so one cannot use the redistributing principle as described throughout this
chapter. For vaults and secure rooms, redistributing principle can be used. Bring
in the power, apply it to the breakers, and filter each leg following the circuit
breakers. Granted, this area must be isolated from the rest of the secured area via
copper walls similar to those in the screen room.
20
Packaging Information
Packaging is a very important subject and is as critical in the EMI world as

it is in the rest of the electronics arena. Component layout is important in
all electronics. Components that are susceptible to H fields must be moved
away from these fields, yet wiring or printed circuit board traces must be short
as possible.
20.1. THE LAYOUT

The physical layout of the EMI filter for best performance is long and thin. From
Fig. 20.1, it is easy to see that the full length is much greater than the height and
width. In the EMI filters shown in Figs. 20.4 and 20.5, the toroids are spaced apart
by the value of the capacitor. This removes the tendency toward cross talk or
mutual inductance by increasing the distance between the inductive components.
Also, note the distance between the dirty or clean end and the clean or dirty end.
The output terminals are on the far end, away from the input connector end.
In Fig. 20.2, the output terminals are not shown. They are on the blind side.
If the input and output must be on the same face, a shield should run almost
the full length of the filter as shown in Fig. 20.3. The components are divided
between the two halves. Half run from the front to the back half and then double
back to the front on the other side of the shield. The shield must be fully grounded
for the entire length.
In the lower current mode, these parameters still hold true. The filters
aspect ratio should be long in length compared with the height and width. The
components should run from one end to the other, as in a transmission line, rather
284
Chapter 20
FIG. 20.1 The high-current insert showing front cover.
than hop back and fourth within the filter body. This reduces the cross talk effect,
and Capcon should be used to cover lead wires to help attenuate the upper
frequencies. The inductors should be mounted in quadratureyes, even toroids
as shown in Fig. 20.4 and the alternative Fig. 20.5. The alternative requires more
room, but all the toroids are in quadrature; two toroids in Fig. 20.4 are not, but
they are separated by additional distance. The upper left and the lower right are
in the same plane, with a capacitor between, but the distance is farther, which
reduces the magnetic coupling.
The top two inductors would be wired directly to the line-to-line capacitor
and the bottom two also wired directly to the capacitor using the vee technique.
See Sec. 7.3. This continues to the next section, still maintaining the quadrature
of the inductors. These are not usually mounted to a printed circuit board because
of the current and the DC resistance of the boards. Most EMI houses do not design
their filters to utilize printed circuit boards.
Some have designed the EMI filter as part of the power input, and this
is proper if the filter is mounted in a container that is a good conductor and
is grounded to the third green wire so that the feed-through capacitors can
FIG. 20.2 Filter showing good length-to-width ratio.
285
FIG. 20.3 Filter showing shield to separate clean and dirty areas.
function properly. This works well if the needed loss is for Federal Communications Commission (FCC) standards or if the required loss is low. Other designers have placed the filter within the power supply using open, or exposed,
components. This technique rarely works. The filter must have shielded components to function properly or other magnetic fields either are influenced by
the filter or couple their magnetic field to the inductors of the filter. This way, a
60 dB filter is now only 24 dB and the filter designer does not know what
went wrong.
The case or container of the filter must be a good conductor. The better this
surface conducts, the lower the magnetic field is on the outside of the case. Even
though cold-rolled steel is often used for the filter body, the container is often
silver plated to enhance the conduction in filters for military or other groups
requiring severe loss. This attenuates the H field and improves the radiated
FIG. 20.4 Toroids in quadrature.
286
Chapter 20
FIG. 20.5 Alternative methodbalanced quadrature.
emissions. The H field establishes a current on the surface of the filter body
container. The better this surface conducts and the thicker the case material, the
weaker the current on the other side of the case. This reduces the H field departing
from the case wall and is true for H fields either out of the filter or into the filter.
The container must also be a good conductor so that the feed-through capacitors
can function. The same would be true if the good conducting case was not wired
to ground or if the ground wire was loose or missing.
In the circuit of Fig. 20.6, neither the feed-throughs, C2, nor the two
common mode arresters work properly if the case ground is resistive or if the
FIG. 20.6 Arresters and feed-throughs to ground.
287
container has a high resistance. The one thing that we can say about ground is
that it isnt ground, or as most say, ground isnt. So, grounding of these filters
cannot be marginal or overlooked.
20.2. ESTIMATED VOLUME
The approximate volume of the higher current filter was discussed in Chapter 19.
The equation is repeated here. S is the number of filter elements (the number of
Ls, Ts, or s). The equation was based on Ls. Od and Id are the outside and inside
diameters in inches.
L =
0.4Ue(Od Id)WSf108
= 0.4UeWSf108
(Od + Id)
where L is the inductance and Sf is the stacking factor. For the capacitor, R is
R =
4CKTt
where C is the capacitance in F, K is the dielectric constant, and T and t are both
thicknesses but T is in mils. This was calculated on the basis of Ls, but Ts and pi
filters should still be close. These are approximations anyway.
404.3gNLuh 2
R
Vtot = =
+
NO
2
2
OdSf d
N(Od Id)
If Od is much bigger than Id, Vtot approaches

404.3gN2OdLuh
Vtot = 1.1SR +
Sf
On the other hand, Col. W. T. McLyman has provided the following data,
combined in Table 20.1, to calculate many different magnetic properties from his
book Transformer and Inductor Design Handbook (Marcel Dekker, New York).
Our goal here is volume and in the next section volume to weight. See also
Chapter 14.
The main idea is to find the area product, Ap, in centimeters to the fourth
power; from this, and knowing the core, the approximate volume can be found
(Fig. 20.7). According to McLyman, in the method listed in his book, the energy
of the inductor must be determined. The equation is
ENG =
LI2p
2
288
Chapter 20
TABLE 20.1 McLyman Magnetic Factors

Core type
Pot core
Powder core
Laminations
C core
Single coil
Tape wound
Kj25C
Kj50C
KvCm3
Kw (g)
433
403
366
323
395
250
632
590
534
468
569
365
1.20
1.14
1.14
1.16
1.16
1.15
14.5
13.1
19.7
17.9
25.6
25.0
48.0
58.8
68.2
66.6
76.6
82.3
The type of core to be used must be known to use the Table 20.1. Kj and X come
from Table 20.1. Ku is the winding factor0.4 for a toroidand Bm (in teslas)
must be known for the core type. The Ap is
Ap =
|2(ENG) 104|x
BmKuKj
Find the Ap for the different sizes of inductors, and add the different Ap values
for all the inductors for the total Ap. For example, a three-stage balanced L filter
would require six inductors, all of the same value. Find the energy based on the
peak current, and from knowledge of the core type, oabtain the components for
the Ap.
If the peak current is 5 A and the inductors are 250 H, ENG is
ENG =
250 106 52
= 0.0031
2
FIG. 20.7 Using the McLyman parameters for size and weight.
289
If MPP powder cores are the choice, then Ku = 0.4, Kj at 25C = 403, Bm = 0.7
tesla, and X = 1.14. The Ap follows:
Ap =
|2 0.0031 104|1.14
= 0.9792
0.7 0.4 403
Round this up to 1 and find the total Ap and then the volume:
VLtot = 1 13.1 6 = 78.6C3m
where 13.1 comes from Table 20.1 for the MPP core for cubic centimeters. The
capacitors are not included, but the weight ratio of the capacitors is of the order
of that of the inductors. The total would be 160 cm3, and this volume is only 60%
utilized. This gives 267 cm3, but allow for the container, feed-throughs, and
wiring and round this up to 280 cm3. This equates to 17 inches3. These are rough
estimates that should get the design engineer in the ballpark.
20.3. VOLUME-TO-WEIGHT RATIO
Bob Hassett at RFI Corp. (now retired) has done some research on the size-toweight ratio for EMI filters. This was primarily done on the tubular types of filters
mentioned in an earlier chapter. The ratio is 1.5 ounces per cubic inch. This
equates to 1.6 pounds for the filter discussed in the preceding section, which
needed 17 inches3.
From McLyman, again, use his area product and use the same Ap as in the
preceding section for one inductor. Multiplying this value of 58.8 for each
inductor by the six inductors, as listed for the powder cores in grams, gives
58.8 6 = 352.8 g
Doubling this for the capacitors, wiring, feed-throughs, container, and input
terminals,
352.8 2 = 705.6 g
The weight in pounds is
705.6
= 1.535 pounds
459.53
This looks like the same ballpark that RFI Corp. attends, even though McLyman
is at NASA Jet Propulsion Laboratory in Southern California and Bob Hassett is
in Bay Shore on Long Island, New York. Anyway, their two methods are close to
agreement.
290
Chapter 20
20.4. POTTING COMPOUNDS

Potting compounds add substantial weight and so should be used sparingly. Most
think that these enhance cooling, but most potting compounds do not aid this
function. In fact, they often hinder heat transfer. This could be a desirable feature
to avoid heating sensitive components. If heat transfer is the goal, some compounds have this feature, but they are often too expensive or not readably
available. Often, fine granules of aluminum are added to the epoxy to enhance
the heat transfer. Most often, potting is added to the filter to tie down the
components. Do not fill the enclosure with the potting material. Cover the bottom
to a height necessary to support the components.
21
Design Examples
21.1. SOUTHEAST ASIA FILTER FOR THE NAVY

A company in the Philippines wrote a marginal specification for a 400 Hz EMI
filter for the navy. Then they started the typical requests for changes, expecting
all the rest of the specification to remain the same. Can you make it smaller and
lighter etcetera? I do not know how anyone can think that weight and size can be
cut and still keep the same features. So some of the initial things were trimmed.
The filter had capacitors tied from line to line, doubling the capacitor values
to those shown in Fig. 21.1. This gave a resonant rise (Fig. 21.2) of over 11 dB
at 1.56 kHz and a 0.58 dB rise at 400 Hz. A 0.58 dB rise does not sound like
much, but the equation is
100.58/20 1 = 1.069 1 = 0.069 = 6.9%
FIG. 21.1 The simplified 400 Hz Asian filter.
292
The resonant rise.
Chapter 21
FIG. 21.2
Design Examples
293
FIG. 21.3 The Asian navy fix.
This is a heavy percentage increase. The value would be larger in the real world.
The fix is shown in Fig. 21.3, but the capacitors had to remain the same. Note
that the center capacitors do not add because of being isolated by the inductor in
the other two legs. Both filters have 60 dB at 20 kHz using the 220A specification.
The resonant rize has increased to 18.9 (Fig. 21.4), but so has the frequency.
The dB increase at 400 Hz is now 0.14. Using the same equation,
100.14/20 1 = 1.0162 1 = 0.0162 = 1.62%
The 1.62% is ideal and the real rise percentage would be more but should be less
than 3%. This was more of a repair than a new design, but it could have been
improved greatly with a completely new design. This is the big story regarding
400 Hz filters, especially three-phase 400 Hz EMI filters. Note the two resonant
rises in Fig. 21.4. This shows that the Q is still too high.
21.2. THE FAULTY 400 HZ SOURCE
Some years ago, a company on Long Island required a 400 Hz three-phase EMI
filter. Some time later, they stated that the filters were overheating. The conclusion was that the capacitor current to groundleakage currentwas excessive.
The line frequency was 400 Hz, so power factor correction coils were suggested.
A doghouse was added to the filter body, and the three inductors were wired in
and returned to the customer. Some weeks later, the customer was livid because
these filters were heating with minimal load current. It was so bad the doghouse
cover bowed and broke the solder loose. The customers test people were
complaining, but there was no complaint from the field. These filters were
working fine. When we loaded the filter, it became hot but nothing like their
report. We were using a 400 Hz motor generator to drive our filter and a resistor
load bank The customer was using an electronic three-phase 400 Hz source to
power our filter and their load.
294
Improved frequency plot.
Chapter 21
FIG. 21.4
Design Examples
295
There were two problems, and neither group knew of either problem. Their
400 Hz power source was rich in 2400 Hz. This was used to derive their 400 Hz.
Our filter had a resonant rise at 2400 Hz. Who knows the current we were pushing
through our capacitors to ground?
This is the problem of a few components to make the filter, especially for
400 Hz and, again, especially for three phase. This filter was a single . Granted,
this filter met the 220A test specification, but it should have been at least a double
. The resonant rise of the previous section (21.1) is still not high enough at
3.31 kHz after the fix, but 2.4 kHz in this section is even worse. Both of these
sections must be pulled up to over 4 kHz to avoid these problems.
21.3. THE ROUND ROD FILTER IN CHAPTER 19
The filter (Fig. 21.5) was specified for 60 A with loss at 1 MHz of 80 dB. This
was a good spot for Chapter 19type technology. This was also a 400 Hz
three-phase four-wire design. It required only a single with two feed-throughs,
one on each end. A bolt ran through the two capacitors with space in the middle.
These were two MPP toroids where the threaded rod just fit through the Id.
The outer diameter of the MPP core had to be less than the two feed-through
capacitors outer diameter. The values needed are two 0.6 F feed-through
capacitors and two MPP cores totaling 5 H.
The 220A test results showed this filter to be flat lined at 80 dB well into
the GHz range (Fig. 21.6). These small feed-throughs and the MPPs are highfrequency devices, which accounts for flat lined loss. It remained flat because it
was single chambered.
FIG. 21.5 The high-frequency rod filter.
296
Loss of the rod filter.
Chapter 21
FIG. 21.6
22
Questionable Designs
The drawings in this chapter have been around for years, and I have no idea of
the initial group or company that designed them. The reason that they are here is
for your critique, and the hope is that you come to the same viewpoint that most
filter people agree with.
22.1. 28 VOLTS AT 35 AMPS

The high current in Fig. 22.1 does suggest low values of inductance and high
values of capacitance. C1 at 88 F is far too high and would be, at best, resistive,
inductive at any reasonable switcher frequency, and yet this filter has four of
them. They add weight, volume, and cost with little EMI benefit except at the
very low-end frequencies. No doubt this is electrolytic and resistive by 30 kHz.
The RC shunt, composed of the 10 ohm resistor and the 8.1 F capacitor,
kicks in at 2 kHz, showing that the switcher frequency (or the current pulses) is
well above this frequency, as are most nowadays. The resistor should be nearer
the design impedance of 0.8 ohm. This would drop the capacitor value to 1 F,
again reducing the weight, volume, and cost but also raising the self-resonant
frequency (SRF). The 10 ohm resistor and the 8.1 F capacitor along with the
0.8 ohm and the 1 F would give little help to the 6.8 F feed-throughs. Although
the 0.8 ohm and the 1.0 F would offer higher impedance, it is not that much
more. The 6.8 F almost stands alone.
The inductors add little common mode loss to the filter. In most applications, the Zorro would be at least an order of magnitude above the 15 H inductor
in the circuit in Fig. 22.1. In the figure, the two differential inductors contribute
298
Chapter 22
FIG. 22.1 DC filter, 28 V at 35 A.
12.5 H to the common modethe same order of magnitudegiving a total of

27.5 H. This makes for a rare design.
There is some differential, with the two balanced inductors totaling 50 H
and the two feed-throughs in series giving 3.4 F. The impedance is 3.8 ohms,
and the cutoff is 12.2 kHz.
22.2. 120 VOLTS, 60 HZ, WITH TRANSZORBS

Figure 22.2 shows another case in which the differential inductor is larger by two
orders of magnitude than the common mode inductor. The two differential
inductors give 1.2 MH of common mode, so the 15 H adds no real support for
this situation. The 15 H is smaller than the error of these two differential
components. It seams that the two inductors are misnamed. The two 2.4 MH
should have been the common mode and the two 15 H inductors the differential
mode inductors.
The two RC shunts are not correct, either. The 47 ohm may be high unless
the design impedance is about 47 ohms and the capacitor is too low. The
299
FIG. 22.2 60 Hz with 120 V and transzorbs.
attenuation of this RC shunt starts at 340 kHz, so it will not help with any switcher
noise and most of the parasitic noise.
The transzorbs are wired wrong. The third transzorb in the ground return
should be wired line to line, making two from line to ground for common mode
pulses. The third should be from line to line for differential mode pulses. These
should be moved to the input to protect the filter and placed outside the filter
container for ease of replacement.
The feed-throughs may be high for 60 Hz, giving a higher leakage current
to ground, more than what is specified.
22.3. THE 28 V DC FILTER
The question about the circuit of Fig. 22.3 is, can the equipment that this filter is
used in even be turned on? Discounting the 0.5 ohm leg capacitor and the other
smaller capacitors, the total capacitance is 2520 F across the line in a circuit that
draws only 6.2 A. This filter current plus the equipment inrush plus the other
capacitor current will be more than the circuit breaker can handle.
From an EMI standpoint, these three large capacitors are useless anyway.
They are inductive long before any EMI requirement must be met. This is the
disadvantage of using most analysis programs because most treat these as pure
capacitors well into the GHz realm. Some programs allow the designer to add this
information, but most enter the component information as pure. Those that would
do so are often not aware of the true property of the component.
The two RC shunts are almost useless. They do not come into play until
720 kHz, which is far above most switcher frequencies and their harmonics and
300
Chapter 22
FIG. 22.3 The 28 V DC filter.
above most parasitics of the switcher transformers and/or inductors. If there is a

high-frequency problem with this filter, the 47 ohm resistor does not help. The
value should be closer to the load impedance, which should be near the design
impedance of the filter. This should be closer to 4.5 ohms.
The 0.1 F capacitors to ground should be feed-throughs for better SRF and
should be located on the load side. These two capacitors are in series across the
line. This gives a line-to-line value of 0.05 F and adds to the input capacitor of
the balanced pi for a total of 0.15 F. If the leakage inductance of the Zorro is
small and if the line impedance is small, this renders the 0.15 F useless well into
the upper kHz range. This is not a VDE, FCC, or CSA filter because the
component count is too high for these types. If losses are required in the lower
kHz range, this technique is not valid. If this filter was tested using the 220A or
some of the better LISNs, creating a line impedance in the lower kHz range, the
filter could look good. This is, again, why the pi filter should be limited in use.
If the 0.5 ohm and 120 F are there to lower the DC supply impedance,
this network should be moved to the load sidewhy lower the impedance and
then raise it by the two 12.5 H inductors? However, the 120 F is too big to
reduce the switcher drop, and so is the 2400 F. The best solution would be to
remove these four components along with the two RC shunts. Move the capacitors
to ground, as stated earlier, and then double the size of the output pi capacitor.
This makes this an L, which is all that is there anyway at a cutoff frequency 1.414
times higher. The circuit now looks as shown in Fig. 22.4.
With all the space saved, the 0.1 F could become a 6.8 feed-through,
aiding the 0.4 MH Zorro and raising the 12.5 to 56 H to make this a balanced
T. Careful here, though, this could add to the switcher drop. Add another 0.2 F
and make this a double balanced L. More must be known to design this properly,
such as the switcher frequency, the on time of the pulse, and the current peak
301
FIG. 22.4 The modified 28 V DC filter.
during this time. In other words, Figs. 22.3 and 22.4 will not function properly.
This EMI filter needs to be fully redesigned.
22.4. 120 V AC 400 HZ
The main complaint is that there is no real common mode or differential mode in
the filter in Fig. 22.5. The RC shunts minimum impedance is 453 ohms, giving
little aid to the common mode or differential, and the same is true for the 4700 pF
to ground. This may have been a medical application requiring low capacitance
to ground for low patient current. The current of nearly 1 A agrees with this. This
FIG. 22.5 The 120 V 400 Hz filter.
302
Chapter 22
FIG. 22.6 The modified 115 V 400 Hz filter.
filter gives 6 dB per octave for both common and differential mode. If the
800 Hthe bottom and top 400 H to convert to unbalancedis correct, a
0.05 F tied line to line after the inductors would give over 40 dB of loss at
450 kHz for FCC.
If this is not for medical use, change the 4700 pF for the limit of 0.02 F,
the legal limit for 400 Hz. This is an order of magnitude larger. Now, with the
added 0.05 F from line to line, there are both common mode and differential
mode losses. If the two RC shunts are there for a problem frequency, the 453 ohm
resistor is too high. Change these two resistors to 130 ohms each. One of these
RC networks, wired from line to line, may replace both networks to ground.
Otherwise, leave the RC shunts out. The changes are shown in Fig. 22.6.
Now, the filter has 10 MH and 0.04 F making up the common mode half
of the filter and 800 H and 0.05 F making up the differential mode half.
The common mode design impedance is 500 ohms and the cutoff is 8 kHz; the
differential mode impedance is 130 ohms with a cutoff frequency of 25 kHz.
22.5. REVIEW
It is impossible to look at a filter after the fact and judge it fairly. We do not know
all the required parameters of the filter: medical, 461, 220A only, resonant rises
at some unknown frequencies. Information may have been provided to the filter
designer either from the EMI test laboratory or related to some other aspect that
we are not aware of. I hope the reader understands that I am not trying to find
fault with any group.
23
Review of Filter Design
This chapter is a catchall for discussing some thoughts mentioned in the book but
not clarified. First, the full design technique is reviewed. If no other chapter of
this book is read, at least read this chapter!
23.1. FILTER DESIGN REVIEW
This is an overview of all the design information in this book with references back
to the chapter that discusses the subject. Much additional information has also
been stated repeatedly without being documented.
23.1.1. Steps in Designing a Filter
Before designing the filter, the following should be known: the power line
frequency if AC, minimal RMS line voltage, maximum RMS load current, type
of load, and lowest switcher frequency if more than one. Some of this can be
guessed at. If this is a single-phase 60 Hz line, a minimum of 100 V RMS should
do and the RMS line current could be raised by 10%. Attempt to determine the
type of load this filter must feed power to: off-line regulator, power factor
correction circuit, inductive input filter, or resistive. For a DC system, this may
well be a switcher. If the filter is being designed for multiple use for off-the-shelf
filter sales, worst cases from previous chapters should be accounted for in
the design.
304
Chapter 23
1. Find the best filter type from Chapter 15 by knowing the type or need
of the load. Also, the specification will help here. Knowledge of the specification
and losses required will help in selecting the line impedance, also from Chapter 15. Federal Communications Commission (FCC) requirements give a higher
line impedance because the loss requirements are at frequencies that give a higher
line impedance. The 50 ohm line impedance is valid at these frequencies.
2. Calculate the design impedance from the lowest voltage divided by the
highest current.
3. Determine the filter cutoff frequency, both differential and common
mode. Use either the matrix equations of Chapter 16 or 17 or Sec. 18.8 on F0 the
easy way. The second approach will get the filter designer onto the proper
ballpark property and the first will get the designer onto the ballpark playing field.
This can be a computer program written by the reader based on Chapters 16 and
17 or from the provided disk. Equations for the cutoff frequency are
L = loss for or T = (2N + 1) 6
L = loss for an L = 12N
The first equation is for the T or filter and the second equation is for the L.
Either of these equations provides the value of L for Fig. 23.1. N is the number
of filters in tandem.
Loss required
dB
Loss per Octave (Table 23.1)
Number of Octaves
Frequency requiring the loss
Cutoff frequency
Fo
FO =
F
2
FIG. 23.1 F0 the easy way equations.
dB
L
dB/L
f/2X
305
Another way to select the value for L is as follows. Choose the loss based
on the filter type. The number on top in Table 23.1 is the number of filters in
tandem. Table 23.1 gives the loss per octave for the filter type listed on the left,
the value of L in Fig. 23.1.
Once 10 times the power line frequency or more is reached, the filter type
and the cutoff frequency are known. As an example, three frequencies are
provided by the EMI test house. They list the approximate dB losses required to
meet the specification and the associated peak frequencies. Table 23.2 is given.
Table 23.2 gives the test house readings followed by the loss added by
the filter designer for the required headroom. The last two rows show the
cutoff frequency obtained by using the preceding methods and equations for
the single L filter and the readings for a double L filter obtained by the same
techniques.
If the power line frequency is 60 Hz, the single L will work and the
cutoff frequency is 1.4 kHz, but if the power line frequency is 400 Hz, the
double L must replace the single L at 12.98 kHz. Round either of these to a
convenient lower frequency. Often, the filter cutoff frequency is determined
by the lowest problem frequency. Be careful, because exceptions do occur, as
in the preceding example, so all listed trouble frequencies must be checked. Do
not assume that it is only the lowest frequency listed that will give the proper
cutoff frequency.
4. Equate the component values from steps 2 and 3 using the following
equations. Rd is the design impedance. This is the proper equation for all filters,
that is, , T, and L. The difference is that the value of L is divided or split for the
T and the value of C is divided or split for the .
L =
Rd
2F0
C =
1
2F0Rd
In a multiple T filter, the first and the last L are half the preceding value but the
values of the central inductors are the full value. Remember also that the value
TABLE 23.1 F0 the easy way chart

Filter
12
18
18
24
30
30
36
42
42
48
54
54
60
66
66
72
80
80
84
92
92
306
Chapter 23
TABLE 23.2 Chart of F0 for L and Double L Example

Test house trouble frequencies
Frequencies
Test house measured loss
Add 6 dB headroom
Single L F0
Double L F0
F1
36 kHz
36 dB
42 dB
4.5 kHz
17.2 kHz
F2
80 kHz
58 dB
64 dB
2.8 kHz
14.98 kHz
F3
120 kHz
71 dB
77 dB
1.4 kHz
12.98 kHz
of C above will be divided for the . Again, the central values of C for a multiple
will be the whole value and the first and last C values will be split. The T and
the are exactly opposite each other. Choose the filter layout based on Chapter
6 and also some information in Chapter 3. Balance the filter if possible.
5. Find the size of the components. Use McLymans data in Chapter 20.
The component volume comes from the outside diameter of the toroid,
connector, or round capacitor and Od is squared times the length. The pressed
capacitor volume is Dx times Dy times Aw. Total all the components, and divide
by 0.6. The aspect ratio should be a minimum of 2 to a maximum of 5 for the
length to end diagonal, but this is also a function of component size. For example,
a 9 3 2 may give the calculated needed volume and a good aspect ratio, but
the wound inductors may be 3 inches in diameter. Either all the inductors will be
aligned along the 9 3 inch direction or the height must change. This example
may work, but quadrature is violated and cross talk may be a problem even though
these are toroids.
6. Design and build the case, or container, on the basis of the components
and volume just determined, and make sure there are no solder or weld voids,
which can allow radiation or destroy the potential of passing any environmental
tests, such as for humidity and salt spray. Have the case silver plated for better
surface conduction if the loss is substantial. Also, aluminum is often used for
better conduction giving lower radiation. This is plated for still better conduction
for lower radiation and for ease of soldering.
7. Install the components, and test the filter in the open container. Tack
the lid down only for easy alteration or adjustment.
8. Adjust the filter for the desired loss, if needed, by the following steps:
Add lossy components (Sec. 5.4).
Add small line-to-line (X) capacitors in parallel with the existing capacitors.
Keep the lead length as short as possible.
Add ferrite beads if the current is low enough (typically 5-Amp limit).
307
Add several turns on the inductors (watch for saturation). If this filter is a
T, whatever turns added to the central inductors of the multiple T,
increase the turns by half on the two outer inductors. They are half the
value of the inner inductors.
Add an RCSHU (a resistor and capacitor in series across the line). See
Sec. 5.7.
Adjustments will move the designer from either the ballpark property or the
playing field to home plate with one of the approaches in step 3 above.
9. Make sure that the end product is buildable and repeatable for production.
23.2. FILTERS IN TANDEM
This section discusses the filter growing from one section to two sections to three
sections and so on. This book states repeatedly (as when going from a single to
a double L in Table 23.2) that as the sections grow, the size of the components
falls. The test problem is as follows:
Line frequency, 400 Hz
Line voltage, 120 V
Maximum line current, 6 A
Needed loss, 60 dB at 20 kHz
This will be solved first for the L then the T and followed by the , only to prove
the same holds true for all three types (Tables 23.3, 23.4, and 23.5). The program
is the one used in Chapters 16 and 17. This is a low current, and the maximum
should be four sections but six are given. The value of L is in MH and C in H.
The first L and C are the individual component values, and the final values of L
and C are the total component values for the filter.
TABLE 23.3
The L Filter Showing the Change as Sections Are Added

Extension
Number
in tandem
1
2
3
4
5
6
F0
Inductance
Capacitance
Inductance
Capacitance
405
2890
5170
6760
7760
8400
7.860
1.130
0.615
0.470
0.410
0.378
19.6
2.82
1.54
1.18
1.02
0.95
7.860
2.26
1.85
1.88
2.05
2.26
19.6
5.64
4.62
4.72
5.10
5.70
308
TABLE 23.4
Chapter 23
The T Filter Showing the Change as Sections Are Added
Extension
Number
in tandem
1
2
3
4
5
6
F0
Inductance
Capacitance
Inductance
Capacitance
940
3130
5075
6460
7500
8435
1.70
.508
0.314
0.246
0.215
0.198
8.64
2.54
1.56
1.23
1.07
0.99
3.40
2.032
1.884
1.968
2.15
2.37
8.64
5.08
4.68
4.92
5.35
5.94
Note how the total L and C drop in value from one to three sections, but
little is gained past three sections. Little self-resonant frequency (SRF) is gained
in moving the capacitor value from 1.54 (1.6 F) to 1.181 (1.2 F). The circuit
can be balanced, with better SRF, moving the 620 H to 310 H.
The T filter in Table 23.4 is the same as the L in Table 23.3. It is ridiculous
to go beyond three sections or four at the most. The inductance 314 H is the
value in each arm, or the total inductance for this one T is 628 H. The central
inductors are twice this initial value or 628 H each. Round these to even values,
and the three-section T circuit is shown in Fig. 23.2. The balanced circuit is shown
in Fig. 23.3.
Note that the filter values of the cutoff frequency, F0, are so much lower
for the various sections as compared with the T filter. Both the T and the are
said to have 18 dB of loss per octave. But the T has this loss, which is missing
in the . This is due to the low line impedance at 20 kHz, estimated at 4 ohms.
TABLE 23.5
The Filter Showing the Changes as Sections Are Added

Extension
Number
in tandem
1
2
3
4
5
6
F0
Inductance
Capacitance
Inductance
Capacitance
555
2520
4730
6360
7450
8155
5.735
1.263
0.672
0.500
0.427
0.390
7.164
1.578
.841
.626
.534
0.488
5.735
2.526
2.016
2.000
2.135
2.340
14.32
6.312
5.064
5.008
5.340
5.850
309
FIG. 23.2 The multiple T.
If this had been equated with 220A, the cutoff frequencies would have been
similar to those for the T. Under these conditions, three sections may be too low
and a fourth section may be required. This is caused by the value of F0. F0 is
cutting into the band pass, which is decided by the 10th harmonic of the power
line frequency. The value listed for C is the value of the outside capacitors, and
all inboard capacitors are twice this value. The four-section is shown in
Fig. 23.4.
In any of these applications, using 6 dB headroom, the values are not
critical. Round them to reasonable values. For the two end capacitors in the
balanced in Fig. 23.5, use 0.6 F rather than 0.63 F.
Comparing the L, T, and sections, the total inductance and capacitance
drop greatly from one section to two. A reasonable drop repeats going from two
sections to three sections. A small drop occurs from three to four sections. This
is followed by a modest increase from four to five sections and again from five
sections to six. If the problem is being caused by a low SRF, adding a section
may help. Going to a balanced arrangement almost doubles the inductor SRF, and
FIG. 23.3 The balanced multiple T.
310
Chapter 23
FIG. 23.4 The multiple .
this should help. If the problem is the filter size, going up a section also helps
because the component size is reduced, giving more room.
Looking again at the three types discussed, the is the least reasonable and
should be used only when a 220A test specification is called for. The reason this
happens is that the front-end capacitor is out of the circuit at the low frequencies
because of the low line impedance. This capacitor aids the filtering when the line
impedance rises to a reasonable value. An easy way to calculate this improvement
of lower total inductance and capacitance is through the equations listed earlier.
These are repeated in Fig. 23.6; they cannot be differentiated because they are not
continuous functions.
As the value of L doubles and triples because of a higher number of filters
in tandem, the denominator decreases and the cutoff frequency rises. The value
of the loss, L, can be calculated from
L = loss for or T = (2N + 1) 6
L = loss for an L = 12N
The upper value of L is for the and T types, and the lower value of L is for the
L filter (see Table 23.1).
FIG. 23.5 The balanced multiple .
311
Loss required
dB
Loss per Octave (Table 23.1)
Number of Octaves
Frequency requiring the loss
Cutoff frequency
Fo
FO =
dB/L
f/2X
F
2
dB
L
FIG. 23.6 F0, the easy way equations.
23.3. Q
The value of the circuit Q of the filter must be low, as stated, without explaining
the reason, often in this book. If the Q is low enough, the number of resonant
rises is one less than the filter sections. The resonant rise level is also a function
of Q. The input voltage level of the resonant rise frequency times Q of the circuit
at that frequency equals the voltage output at this frequency. If high-quality
capacitors are used, the Q of the circuit will fall to the inductors because
capacitors normally carry a much higher Q rating, typically 10 times greater than
the inductors. For a single L circuit, the equation for Q is
L
Rd
2F0
Xl =
2FRd
FRd
=
= KRd
2F0
F0
Q =
Xl
= K
Rd
This value of Q is for one inductor. The value of the Q will be much lower than
K because the design impedance, Rd, is lower than the operating impedance of
the load. These simple equations do not figure in the DC resistance of the
inductor, the line impedance, and the equivalent series resistance of the capacitors. This is another reason the design criterion sets the cutoff at least 10 times
312
Chapter 23
the line frequency and the headroom of 6 dB. The line harmonic content should
be low enough by the time the circuit Q is high enough to cause ringing or a
resonant rise. In multiple circuits, the Q is a function of the total inductance. The
value of Q must be multiplied by the number of inductors, but as the cascaded
sections increase, the value of F0 rises, lowering the individual inductance, and
also the total inductance (up to four) decreases.
For the T and filters, the equation for the value of L in the loss equation
is (2N+1) and the total loss per octave is 6(2N+1) dB. For the L, the loss per
octave is (12N) dB. Either of these two values replaces the value of L in the
previous section. The first equation is for either T or and the last is for L. N is
the number in tandem (Fig. 23.7).
The needed loss at the trouble frequency (dB), the tandem number (N), the
trouble frequency (Ft), and the needed Q are known. Substitute 1 for Q and solve
for F. This is the approximate frequency where Q is one (because of losses not
accounted for, the actual value of F will be higher). Is this near any power line
harmonic, especially the odd harmonics, or near any known problem frequency?
If yes, or just to be safe, add an RC shunt to this filter where Xc equals Rd at this
frequency F. If another RC shunt is required for other reasons, only one is needed.
FT
FO =
2
L=
XL =
dB
( 2 N +1) 6
dB
( 2 N +1) 6
Rd
R 2
= d
2F0
2FT
dB
( 2 N +1) 6
2FRd 2
2FT
Q=
NX L
Rd
dB
( 2 N +1) 6
FRd 2
FT
dB
( 2 N +1) 6
FN 2
FT
FIG. 23.7 Equations for Q.
313
Opt for the lower frequency RC shunt. If there are higher frequency Q problems,
opt for a higher quality capacitor that has a higher Q and a higher SRF. Also, in
this case, make sure the resistor used in the RC shunt is noninductive.
23.4. TESTING THE FILTERS
Throughout this book, the test specifications have been mentioned, but not the
testing itself. The test equipment input and output impedance is 50 ohms
unbalanced. The connectors are BNC and the cable is a coaxial cable (coax). This
means one side is ground. So, how are the various EMI filters checked?
In Fig. 23.8, the first ground to the left is the tracking generator coax ground
at the filter. The central ground represents the filter ground, and the last ground
on the right is the spectrum analyzer ground. The filter is pressed to ground with
clamps, vices, or other means. As stated throughout the book the, this filter must
be grounded to work properly. If it is not grounded, the capacitors within the tube
are off ground and cannot function. Here, common mode and differential mode
are not really treated differently.
In the balanced filter in Fig. 23.9, the hookup is the same. Will this check
the filter? Note that the bottom row of inductors is out of the picture, shorted out.
The inner capacitors are also out of the picture because of being in series with
the inductors. The loss pattern would not meet any specification. The question is,
what will fix the problem?
The answer is that the unbalanced output and input of the test system must
be converted to balanced. Matching transformers must be used on both ends
(Fig. 23.10). These transformers are 1:1, 50 ohms in and out high-frequency
devices. They are small high-frequency balun wound in ferrite balun cores. Not
all the internal ground components are shown in Fig. 23.10. This is measuring the
FIG. 23.8 Test setup for an unbalanced filter.
314
Chapter 23
FIG. 23.9 The balanced filter.
differential mode only. All capacitors and inductors add loss to the filter. What
about the common mode loss? How is it measured?
In Fig. 23.11, the differential mode components are removed. The balanced
inputs are shorted together and the output is also shorted together. The figure
shows the capacitance to ground. The Zorro, or common mode choke, is included.
The test equipment is grounded at the filter. A common mode loss pattern should
be obtained. If you test the filter as in Fig. 23.9, only the common mode inductor
and the one feed-through are being tested.
FIG. 23.10
The balanced filter with matching transformers.
FIG. 23.11
The common mode part of Fig. 23.10.
315

EMI Filter Design

Uploaded by

Copyright:

Available Formats

EMI Filter Design

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

EMI Filter Design

Uploaded by

Copyright:

Available Formats

EMI Filter

Richard Lee Ozenbaugh

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

PRINTED IN THE UNITED STATES OF AMERICA

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

where L is the inductance, C is the capacitance, Rd is the design impedance, and

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

I wish to thank: Colonel W. T. McLyman, NASA Jet Propulsion Laboratory, now

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

1. Why Call EMI Filters Black Magic?

2. Common and Differential ModeDefinition, Cause,

Common and Differential Mode Definitions

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

What Creates Differential Mode Noise?

3. EMI Filter Source Impedance of Various Power Lines

4. The Various AC Load Impedances

5. DC CircuitLoad and Source

Various Source Impedances

6. Typical EMI FiltersPros and Cons

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

7. Filter ComponentsThe Capacitor

8. Filter ComponentsThe Inductor

Inductor Styles and Specifications

9. Common Mode Components

10. The Transformers Addition to the Filter

11. Electromagnetic Pulse and Voltage Transients

The Three Theories

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

The Gas Tube

12. What Will Compromise the Filter?

13. Waves as Noise Sources

14. Study of the Off-line Regulator

With or Without Critical Value of Inductance

15. Initial Filter Design Requirements

Differential Mode Design Goals

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

Methods of Reducing the Size of the Inductor Due to

16. MatricesReview of A Matrices

Chain Matrix A: Transfer Functions

17. The Filter Design Technique

The Unit Matrix

19. Applications Using Round or Square Conducting Rods

Very High Current Filters

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

High-Current Method Three

20. Packaging Information

21. Design Examples

Southeast Asia Filter for the Navy

22. Questionable Designs

23. Review of Filter Design

Filter Design Review

Copyright 2001 by Marcel Dekker, Inc. All Rights Reserved.

Most engineers, both designers of electromagnetic interference (EMI) devices and

1.1. WHAT IS EMI?