P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
Chapter
3
EMI Filter Design
Nearly all power circuits contain an input electromagnetic interference
(EMI) filter. The main purpose of the EMI filter is to limit the inter-
ference that is conducted or radiated from the power circuit. Excessive
conducted or radiated interference can cause erratic behavior in other
systems that are in close proximity of, or that share an input source
with, the power circuit. If this interference affects the power circuit, it
can cause erratic operation, excessive ripple, or degraded regulation,
which can lead to system level problems. Input EMI filters may also
be used to limit inrush current, reduce conducted susceptibility, and
suppress spikes. The specifications for the allowable interference are
generally driven by the power circuit specification. The most common
specifications include MIL-STD-461 for military applications and FCC
for commercial applications. Many other EMI specifications also exist.
This chapter will deal with the design and analysis of EMI filters
that will reduce conducted interference and conducted susceptibility
and limit inrush current. The design of the input filter is slightly more
critical when the power circuit is a regulated switching circuit, rather
than a linear circuit, because a negative input resistance is created by
the regulated switching circuit.
Although it is possible to simulate the radiated interference of a
power circuit, it is beyond the scope of this book.
Basic Requirements
The design of an input EMI filter begins with the definition of two basic
requirements:
� The filter must provide the power converter with lower output impe-
dance than the negative input resistance of the power circuit.
63
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
64 Chapter Three
� The input filter attenuation must be sufficient to limit the resulting
interference to a level that is below the imposed specification.
The following flowchart provides a step-by-step approach that may be
used to design an input filter.
EMI filter design flowchart
Define
Impedance
YES
YES
YES
NO
Attenuation
Defined?
Harmonic
Content
Known
NO
Waveform
Known?
Estimate
Waveform
Calculate Fourier
components
Calculate
Attenuation
Calculate
component
Values
Defining the Negative Resistance
The negative resistance of the power circuit can be defined by looking
at the following conditions
Pin = Poutefficiency
Iin = PinVin
Rin = VinIin =
V2in
Pin
= V
2
in
∗efficiency
Pout
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
EMI Filter Design 65
The input resistance is negative because as the input voltage in-
creases, the input current decreases. As a simple example, we can use
PSpice to analyze the input resistance of the power circuit. PSpice can
analyze the input resistance in a number of ways. The simplest method
is the transfer function (.TF) analysis, which calculates the DC gain
and the small signal input and output impedance. The following exam-
ple uses the PSpice.TF analysis to measure the input resistance of a
switching power circuit.
Example 1—Input resistance analysis
Input File
RIN: INPUT RESISTANCE
.TF V(5) V1
V1 5 0 20
G1 5 0 Value = { 100/V(5) }
.END
Output File
RIN: INPUT RESISTANCE
.TF V(5) V1
V1 5 0 20
G1 5 0 Value = { 100/V(5) }
.END
.END
∗∗∗SMALL-SIGNAL CHARACTERISTICS
V(5)/V1 = 1.000E+00
INPUT RESISTANCE AT V1 = −4.004E+00
OUTPUT RESISTANCE AT V(5) = 0.000E+00
The G1 source simulates a power circuit, which has an input power of
100 W. V1 applies 20 VDC to the power circuit, and the .TF measures
the input impedance at node 5 and the output impedance at V1. The
results are placed in the output file. Note that PSpice calculated the
input impedance as a negative resistance of 4 �, which is in agreement
with the above derivation.
Defining the Harmonic Content
The next step in designing an input EMI filter is to determine the har-
monic content of the power circuit input current. If the input current
waveform is known, a Fourier analysis can be performed in order to es-
tablish the harmonic content of the waveform; however, even if the exact
waveform is not known, we can estimate the waveform with reasonable
accuracy. The design can be optimized later, if necessary.
Consider the pulsating waveform in Fig. 3.1. With a peak amplitude
of 1 and a base amplitude of 0, we can compute the Fourier series of
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
66 Chapter Three
Input Current
1
0
t
T
Duty Cycle = t/T
Figure 3.1 Pulsating waveform used in the Fourier series computation.
harmonic n as follows:
An = 2T
t∫
0
sin(nt)
Bn = 2T
t∫
0
cos(nt)
Cn =
√
A2n + B2n
If we assume that the input ripple current is pulsating and if we
know the duty cycle, we can proceed to the Fourier analysis. If the
duty cycle is not known, we will assume a value of 50%. This assump-
tion is the worst case, because the Fourier analysis of a pulsed wave-
form has a maxima at a value of 50%. In the next example, we will
use SPICE to calculate the Fourier coefficients of a 50% duty cycle
pulse.
Example 2—.FOUR analysis
The following example demonstrates the use of the .FOUR analysis. V1
is a pulsed voltage source, which has a 50% duty cycle and a 100-kHz
frequency. The .FOUR statement calculates the magnitude and phase
of the DC value and the first nine harmonics. The result is placed in
the output file as shown below.
EX2: DEMONSTRATING THE USE OF THE .FOUR ANALYSIS
.OPTIONS NUMDGT=3
.TRAN .01U 20U
.FOUR 100KHZ V(1)
V1 1 0 PULSE 0 1 0 0 0 5U 10U
.END
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
EMI Filter Design 67
FOURIER COMPONENTS OF TRANSIENT RESPONSE V(1)
DC COMPONENT = 5.010000E-01
HARMONIC FREQUENCY FOURIER NORMALIZED PHASE NORMALIZED
NO (Hz) COMPONENT COMPONENT (DEG) PHASE(DEG)
1 1.000E+05 6.366E-01 1.000E+00 -3.600E-01 0.000E+00
2 2.000E+05 2.000E-03 3.142E-03 8.928E+01 9.000E+01
3 3.000E+05 2.122E-01 3.333E-01 -1.080E+00 4.088E-09
4 4.000E+05 2.000E-03 3.142E-03 8.856E+01 9.000E+01
5 5.000E+05 1.273E-01 2.000E-01 -1.800E+00 2.044E-08
6 6.000E+05 2.000E-03 3.142E-03 8.784E+01 9.000E+01
7 7.000E+05 9.093E-02 1.428E-01 -2.520E+00 5.723E-08
8 8.000E+05 2.000E-03 3.142E-03 8.712E+01 9.000E+01
9 9.000E+05 7.072E-02 1.111E-01 -3.240E+00 1.226E-07
TOTAL HARMONIC DISTORTION = 4.288115E+01 PERCENT
As you can see from the output file, the fundamental harmonic has
a peak value that is 63.6% of the peak pulse amplitude. Although this
does provide the required information, it is far from elegant. A better
solution is to calculate the harmonics in Probe. The resulting plot is
shown in Fig. 3.2.
This is the worst case for a pulsed waveform and could be conserva-
tively used for the design of the input filter.
Example 3—Using the .STEP command to calculate harmonics
The next example uses the PSpice .STEP command to sweep the duty
cycle from 5% to 95% and look at the fundamental amplitude of the
resulting square wave. As in the previous example, V1 is a pulsed
Frequency
0Hz 0.5MHz 1.0MHz 1.5MHz 2.0MHz 2.5MHz 3.0MHz 3.5MHz 4.0MHz 4.5MHz 5.0MHz
V(1)
0V
400mV
800mV
SEL>>
Time
0s 2us 4us 6us 8us 10us 12us 14us 16us 18us 20us
V(1)
0V
0.5V
1.0V
Figure 3.2 The FFT feature of the Probe graphical waveform postprocessor is used to
calculate the harmonics of a square waveform.
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
68 Chapter Three
Frequency
0Hz 40KHz 80KHz 120KHz 160KHz 200KHz 240KHz 280KHz 320KHz 360KHz 400KHz 440KHz
... V(1)-V(0)
0V
0.2V
0.4V
0.6V
0.8V
1.0V
Figure 3.3 FFT of the .STEP analysis. The waveform with the largest amplitude at 100
kHz corresponds to the 50% duty cycle (TON= 5 µs).
voltage source. In this case, the pulse has an initial amplitude of
1 V and switches to 0 V after delay “TON.” “TON” is swept from 0.5 to
9.5 µs in 0.5-µs steps.
When the simulation is finished, you can use Probe to display the
X-Y data, or you may view the output file in a text editor. You will
have a graph of the fundamental harmonic versus “TON.” This confirms
the previous statement that the 50% duty cycle was the maxima and
provides a reference you may find helpful in the future.
X3: .STEP ANALYSIS
.PROBE
.PARAM TON=0.5u
.STEP PARAM TON 0.5u 9.5u 0.5u
.TRAN .1U 10U
.PRINT TRAN V(1)
V1 1 0 PULSE 1 0 {TON}
.END
The FFT results of the .STEP analysis are shown in the graphs of
Figs. 3.3 and 3.4.
Example 4 – EMI filter design
In order to design the EMI filter, we need to define a converter that
will operate with it. For the purpose of this example, let us assume that
we have a power converter that will operate with an input voltage of
18 to 32 V DC. The converter output power will be 75 W and will have
an operating efficiency of 75%. The converter will have a switching
frequency of 100 kHz. The conducted emissions requirement allows the
1-mA peak to be reflected back to the input lines. A second-order filter
will be used.
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
EMI Filter Design 69
1
1.00U 3.00U 5.00U 7.00U 9.00U
Time in Secs
700M
500M
300M
100M
-100.0M
Fu
nd
am
en
ta
l A
m
pl
itu
de
in
V
ol
ts
Fundamental Harmonic vs Ton for 10uSec Pulse Train
Figure 3.4 .STEP analysis result shows the 50% duty cycle as the maxima.
Let us follow the procedures that were defined in the EMI design
flowchart. Step 1 is to calculate the input impedance.
Calculating the input impedance. The input impedance was defined ear-
lier in this chapter as
V2in∗efficiency
Pout
It is obvious that the lowest impedance will occur at the minimum input
voltage. This value can be calculated as
182 × 0.75
75
= 3.24 �
Calculating the harmonic content. Because no detail is provided regard-
ing the pulse current waveforms, we will assume that the duty cycle is
50%. The average input current is
Iavg = PoutVin∗efficiency =
75
18 × 0.75 = 5.56A
At a duty cycle of 50%, the peak amplitude will be 11.12 A. In the
previous harmonic analysis, we defined the fundamental harmonic to
be 0.636Ipk=7.08 A.
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
70 Chapter Three
Calculating the required attenuation. With a maximum reflected ripple
current of 1-mA peak, we can define the attenuation required as
Attenuation = 7.08
0.001
= 7080 = 77 dB
Calculating the component values. The attenuation for a second-order
filter can be defined as
Attenuation =
(
fswitch
ffilter
)2
We can compute the filter frequency as
100 kHz√
Attenuation
= 100 kHz
84.14
= 1188 Hz.
The values of L and C can be defined by setting their impedances to
the input converter input impedance at the filter resonant frequency,
as defined above.
C = 1
2π
(
1188
) (
3.24
) = 41.35 µF
L = 3.24
2π
(
1188
)434 µH
Note that the characteristic impedance of the filter is defined by
Zo =
√
L
C
=
√
434 µH
41.35 µF
= 3.24 �
which is equal to the converter input impedance. In an actual design,
it is a good practice to provide a 6-dB margin for these characteristics.
Damping Elements
While this filter provides the proper impedance matching and the re-
quired attenuation, the impedance will be quite high at the resonant
frequency of the filter. The only damping elements in the circuit are the
DC resistance (DCR) of the inductor and the equivalent series resis-
tance (ESR) of the capacitor (which we have not defined). It is normally
necessary to provide damping of the L-C filter in order to restrict the
impedance of the filter at the resonant frequency. A shunt series R-C
network is used for this purpose. The value of the damping capacitor
is generally 3 to 5 times greater than that of the filter capacitor, and
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
EMI Filter Design 71
C1
41.35U
C2
CDAMP I1
AC
V(1)
R1
RDAMP
1
2
Frequency
100Hz 300Hz 1.0KHz 3.0KHz 10KHz 30KHz 100KHz 300KHz 1.0MHz
V(1)
0V
1.0V
2.0V
3.0V
4.0V
Figure 3.5 Schematic of the test circuit used to measure the impedance of the filter. The
waveform V(1) is equivalent to the impedance because the input is a current (I1 1 0 AC
1). The case for CDAMP = 120µ and RDAMP = 1.6 is shown.
the value of the damping resistor is generally close to the characteristic
impedance of the filter. The PSpice .Step command is ideal for defining
these elements.
The following circuit is designed to measure the impedance of the fil-
ter, while sweeping the damping capacitor from 120 to 200 µF in 40-µF
increments. For each value of the damping capacitor, the damping re-
sistor will be swept from 0.5 to 2 times the characteristic impedance
(1.6 to 6.4 �) in 0.6-� increments. The PSpice listing and schematic of
the test circuit (Fig. 3.5) are shown below.
The results are shown below.
EX4: TO MEASURE THE IMPEDANCE OF A FILTER
.AC DEC 10 100HZ 1MEGHZ
.PARAM CDAMP=120u
.PARAM RDAMP=1.6
.STEP PARAM CDAMP 120U 200U 40U
.STEP PARAM RDAMP 1.6 6.4 .6
.PROBE
C1 1 0 41.35U
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
72 Chapter Three
C2 1 2 {CDAMP}
R1 2 0 {RDAMP}
I1 0 1 AC 1
L1 0 1 434U
.END
The results are provided in the output file and are shown below.
Sweep Analysis of EX4.ckt
Count CDAMPRDAMPMaximum
1 1.20000e-004 1.60000e+000 3.891
2 1.20000e-004 2.20000e+000 3.440
3 1.20000e-004 2.80000e+000 3.557
4 1.20000e-004 3.40000e+000 3.916
5 1.20000e-004 4.00000e+000 4.395
6 1.20000e-004 4.60000e+000 4.840
7 1.20000e-004 5.20000e+000 5.248
8 1.20000e-004 5.80000e+000 5.619
9 1.20000e-004 6.40000e+000 6.104
10 1.60000e-004 1.60000e+000 2.994
11 1.60000e-004 2.20000e+000 2.869
12 1.60000e-004 2.80000e+000 3.153
13 1.60000e-004 3.40000e+000 3.672
14 1.60000e-004 4.00000e+000 4.161
15 1.60000e-004 4.60000e+000 4.614
16 1.60000e-004 5.20000e+000 5.033
17 1.60000e-004 5.80000e+000 5.580
18 1.60000e-004 6.40000e+000 6.121
19 2.00000e-004 1.60000e+000 2.489
20 2.00000e-004 2.20000e+000 2.593
21 2.00000e-004 2.80000e+000 3.024
22 2.00000e-004 3.40000e+000 3.547
23 2.00000e-004 4.00000e+000 4.038
24 2.00000e-004 4.60000e+000 4.494
25 2.00000e-004 5.20000e+000 5.040
26 2.00000e-004 5.80000e+000 5.591
27 2.00000e-004 6.40000e+000 6.137
The impedance was exceeded with the 120-µF damping capacitor
(Fig. 3.6). If we use a 160-µF capacitor, the impedance will be minimized
with a 2.2-� damping resistor. A lower impedance could be achieved
with a 200-µF damping capacitor and a 1.6-� damping resistor. We
will select the 160-µF capacitor and the 2.2-� resistor.
The following simulation shows the impedance characteristics and
the reflected ripple of the filter (see also Figs. 3.7 and 3.8).
EMI2: TO SHOW THE REFLECTED RIPPLE OF THE FILTER
.AC DEC 10 100HZ 100KHZ
.TRAN 1U 10M 9980U .1u UIC
.PROBE
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
EMI Filter Design 73
1
2.10 3.10 4.10 5.10 6.10
Rdamp
6.00
5.00
4.00
3.00
2.00
M
ax
im
um
in
pe
da
nc
e
in
O
hm
s
C=120 µF
C=200 µF
C=160 µF
Figure 3.6 Family of curves showing the maximum impedance versus the damping resis-
tor value. Each curve represents a different capacitor value.
C1 2 0 41.35U
C2 2 1 160U
R1 1 0 2.2
I1 0 2 AC 1 PULSE 0 11 0.1U 0.1U 0.1U 5U 10U
L1 0 2 434U IC=-5.5
.END
L1
434U
C1
41.35U
C2
160U
R1
2.2
I1
AC
V(2)
I(V1)
L1[I]4
2
1
Figure 3.7 Circuit used to show the impedance and the reflected ripple of the filter.
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
74 Chapter Three
9.988M
-5.611<
x
>
9.993M
-5.609<
x
>
1
9.982M 9.986M 9.990M 9.994M 9.998M
Time in Secs
-5.608
-5.609
-5.610
-5.611
-5.612
In
du
ct
or
C
ur
re
nt
in
Am
ps
∆x = 5.050U ∆y = 1.908M
1
200 500 1K 2K 5K 10K 20K 50K
Frequency in Hz
3.50
2.50
1.50
500M
-500M
Fi
lte
r I
m
pe
da
nc
e
V(
2)
in
A
m
ps
(a)
(b)
Figure 3.8 Current in the inductor (a) due to a current pulse input, and
impedance characteristics over frequency (b) for the filter circuit in Fig. 3.7.
Fourth-Order Filters
Because the physical size of power converters is continually shrinking,
higher order filters are being used more often than not. The filter is
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
EMI Filter Design 75
designed in much the same way as the second-order filter. The following
example demonstrates the design of a fourth-order filter using the same
design parameters as those that we used for the previous filter.
The “octave” rule basically states that resonances should be at least
an octave apart. In an effort to be conservative, let us use a factor of
2.5. The attenuation of the filter can be defined as
Attenuation =
(
fswitch
f1
)2
∗
(
fswitch
2.5 f1
)2
= f
4
switch
6.25 f 41
If we set the attenuation at 7080, as in the previous example, and
solve for f1 we obtain f1=6.895 kHz. The second pole is then at 2.5 f1 =
17.237 kHz.
The impedance of each section should be designed to be lower than the
impedance of the converter, which we had determined to be 3.24 � in
the previous example. The filter is loaded by the negative resistance of
the converter and produces a combined impedance of
Zloaded = Zin ∗ ZoZin + Zo
The loaded filter Q is defined as
Q = Zloaded
Zo
where Zo is the filter characteristic impedance defined by
Zo =
√
L
C
If we combine the above equations, we have
Q = Zin ∗ Zo
(Zin + Zo) Zo
Zo = −
(
Q − 1
Q
)
Zin
The filter Q is generally maintained below a value of 2. If we set Q = 2
and solve for Zo we obtain
Zo = −
(
2 − 1
2
) (−3.24) = 1.62 �
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
76 Chapter Three
If we use this impedance and the calculated resonant frequencies, we
can define both inductors and both capacitors.
L1 = 1.62
2π
(
6895
) = 37 µH
C1 = 1
2π
(
6895
) (
1.62
) = 14 µF
L2 = 1.62
2π
(
17, 237
) = 15 µH
C2 = 1
2π
(
17, 237
) (
1.62
) = 5.7 µF
As shown in the previous example, we can use the .Step command to
sweep the values of the damping capacitor and the damping resistor. If
we use a range of 3 to 5 times the value of the real capacitor, we will
sweep the damper capacitor from 42 to 70 µF in steps of 14 µF. We will
sweep the damper resistor from one-half to twice the Zo of the filter,
i.e., from 0.8 to 3.2 � in 0.2-� steps.
The schematic for the fourth-order filter and its impedance response
are shown in Fig. 3.9.
Note that two 10-M� resistors have been added. To aid circuit con-
vergence, the resistors were added to the nodes that are purely reactive.
The circuit listing and output file are shown below. A sweep of the maxi-
mum impedance as a function of the damping resistor and the damping
capacitor was also performed. The results of the sweep are shown in
Fig. 3.10. Each curve is for a different value of damping capacitor.
4THORD: A 4TH ORDER FILTER
.AC DEC 10 100HZ 1MEGHZ
.PROBE
.PARAM CDAMP=42u
.PARAM RDAMP=0.8
.STEP PARAM CDAMP 42U 70U 14U
∗.STEP PARAMRDAMP .8 3.2 .2
.PRINT AC V(4) VP(4)
1 1 0 5.7U
C2 4 2 {CDAMP}
R1 2 0 {RDAMP}
I1 0 4 AC 1
L2 1 4 37U
C3 4 0 14U
R2 1 0 10MEG
R3 4 0 10MEG
L1 0 1 15U
.END
P1: IML/OVY P2: IML/OVY QC: IML/OVY T1: IML
MHBD017-03 Sandler MHBD017-Sandler-v4.cls October 6, 2005 18:53
EMI Filter Design 77
L1 15U
C1
5.7U
C2
CDAMP
R1
RDAMP
I1
AC
V(4)