Hierarchical EMC analysis approach for power electronics applications

Abstract

In this paper, a novel method for EMI (electromagnetic interference) level prediction is proposed. The method is based on the hierarchical structure of the generation of EMI. That is, the determination of EMI level can be divided into three levels, namely the functional level, the transient level and propagation level. The lower level provides parameter values for the higher level. That makes the analysis to be a single direction chain. In functional level, the working points are obtained. Through transient level analysis, the exact noise sources are determined. In propagation level, the high frequency characteristics in the propagation path are expressed including the variation of parasitic parameters. Frequency analysis can be used to get the EMI level measured in the receiver side rapidly. Because this approach is a straight forward method, the impact of any components can be evaluated immediately instead of doing the simulation from begin to end.

Full text

Page 1

1176
Hierarchical EMC Analysis Approach for
Power Electronics Applications
Dongsheng Zhao∗, Braham Ferreira∗, Anne Roc’h†, and Frank Leferink†‡
∗Faculty of EWI, Delft University of Technology, Mekelweg 4, 2628CD, Delft, the Netherlands
Email: d.zhao@tudelft.nl Tel: +31 15 2785744
†Faculty of EWI, University of Twente, P. O. Box 217, 7500AE, Enschede, the Netherlands
‡Thales Netherlands, P. O. Box 42, 7550 GD, Hengelo, The Netherlands
IndexTerms - EMI/EMC, Conducted emission, PWMVSI,
motor drive, voltage source inverter
Abstract - In this paper, a novel method for EMI (Electro-
magnetic Interference) level prediction is proposed. The method
is based on the hierarchical structure of the generation of
EMI. That is, the determination of EMI level can be divided
into three levels, namely the functional level, the transient
level and propagation level. The lower level provides parameter
values for the higher level. That makes the analysis to be a
single direction chain. In functional level, the working points
are obtained. Through transient level analysis, the exact noise
sources are determined. In propagation level, the high frequency
characteristics in the propagation path are expressed including
the variation of parasitic parameters. Frequency analysis can be
used to get the EMI level measured in the receiver side rapidly.
Because this approach is a straight forward method, the impact
of any components can be evaluated immediately instead of doing
the simulation from begin to end.
I. NOMENCLATURE
Vdc
Vg+
Vg−
Vth
Il
Irr
Rg
gm
Cies
Cge
Ls
trr
voltage in dc-bus
on-state gate voltage
off-state gate voltage
gate threshold voltage
inductive load current
diode peak reverse recovery current
IGBT gate resistor
IGBT transconductance
IGBT input capacitance
IGBT gate capacitance
IGBT stray inductance
diode reverse recovery time
[V]
[V]
[V]
[V]
[A]
[A]
[Ω]
[S]
[F]
[F]
[H]
[s]
II. INTRODUCTION
A power electronics application is usually accompanied
by high voltage and current amplitudes, steep voltage and
current transients. The EMI (Electromagnetic Interference)
issue is regard as a main side effect for power electronics
applications. Noise level prediction is a vital task for filter
design, but many difficulties are encountered. For instance,
the prediction asks many experiences to build equivalent
circuit, using simplified model frequently loses high frequency
details, individual approach must be developed for various
topologies, etc. A universal method for EMI prediction of
power electronics applications is desirable.
The universal method should overcome these common
characteristics of power electronics application: (a) the large
difference between the time constants [1], (b) the long time
before reaching steady state. Doing simulation in time domain
is very time-consuming. Using frequency domain approach
needs to assume the periodic noise source and fixed propaga-
tion path, which is not true. For instance, the slopes of voltage
and current transient depends on the operation conditions, the
assumption of periodic noise source does not exist. Because
the junction capacitors of grid rectifiers change with the
reverse voltage [2], the noise propagation path is not fixed
as well. A hierarchical approach combines the advantage of
the time domain and frequency domain approach to achieve a
fast, universal and accurate prediction is desirable.
In this paper, the approach is described with detail, then
a PWM voltage source inverter feeds an induction motor
is analyzed by this approach because it is a typical power
electronic application. The result is given and compared with
experiment result. The comparison verifies the feasible and
advantages of the approach. The comparison is also done to
compare this approach to conventional methods, in the aspect
like speed and accuracy.
III. HIERARCHICAL STRUCTURE
The determination of EMI level can be divided into three
levels, namely the functional level, the transient level and
propagation level. The lower level provides parameter values
for the higher level. In functional level, the working points are
obtained. Through transient level analysis, exact noise sources
are determined. In propagation level, the high frequency char-
acteristics in the propagation path are expressed including the
variation of parasitic parameters.
A. Functional Level
In functional level, the switches are idealized, which have
the following characteristics:
• transient time is zero for turning on and turning off;
• the conducted impedance is zero when the switch turns
on, and is infinite large when the switch turns off.
The circuit is further simplified by ignoring parasitic parame-
ters in the circuit. These simplifications make the mathmatical
modelling and analysis feasible. The operation conditions
can also be determined by analytic approach. In frequency
978-1-4244-1668-4/08/$25.00 ©2008 IEEE

Page 2

1177
domain, the calculation in this level gives the correct frequency
spectrum in the low frequency range part. In high frequency
range part, the levels are overestimated because the transients
are idealized by zero transient time. Therefore, a correction
factor is needed.
B. Transient Level
As we know, the real signal around the transition can be
described very close to the real behaviour by the convolution
of two waves in the time domain. The first one is the piecewise
linear waveform obtained in the functional level. The second
one is the derivative of the signal in transition.
The signal in transient is the current and voltage transient of
the switches. Their waveforms are determined by the operation
condition of the switches. That means the waveforms change
at each transient. In Figure 1, a measurement is done for
the voltage waveform during a commutation, the trigger is
set to different current level and in a small range. It can be
observed easily, that the dv/dt changes significantly with the
load current.
(a) set trigger load current to 6.6A (b) set trigger load current to 0.5A
Fig. 1.The voltage transient slope changes with load current
C. Propagation Level
Known from [2], the effect of the equivalent capacitance
when the grid rectifiers turn-off is notable in conducted
frequency band. Also, all the switching devices and load have
their inner passive parasitic components, they are included as
a part of propagation path. The transfer ratio is defined as the
ratio between the voltage drop over the LISN 50Ω resistor and
the noise source.
D. Flow Chart of Hierarchical EMC Analysis Approach
The procedures of hierarchical EMC analysis approach are
summaried in Figure 2 shown below. In the flowing sections,
hierarchical EMC analysis approach is applied to predict
the EMI level of a variable speed motor drive system. The
explanations are available in next sections.
IV. IMPLEMENT STEPS IN FUNCTIONAL LEVEL
The inverter cicuit is operated by a sinusoidally pulse width
modulated signal to produce an output. The purpose in this
level is to get the right voltage and current waveforms by
analytical approach. The steps are listed below.
1) The fundamental and harmonics of output voltage are
calculated.
#
1
2
...
Il
3
Vds
rising
falling
...
rising
falling
3.2
...
-2
-2.4
101
102
Switching events
#
1
2
...
type
turn on
turn off
...
turn off
turn on
active siwtch
upper
upper
...
lower
lower
101
102
Edge waveform
determine
t1,t2,t3,t4
t5,t6
t5,t6
t1,t2,t3,t4
Power devices and circuit parameters
Vg+, Vg-, Vth,Rg,gm,Cies,Cge,Ccg,Vdc
Function
Level
Low frequency
part of noise
source spectrum
Complete
Spectrum
Variable
noise
source
Variable
propagation
path
High Frequency
part of noise
source spectrum
Fig. 2.Flow chart of hierarchical EMC design procedures
2) The parameters of low frequency induction motor model
are achieved via experiment measurement.
3) The fundamental and harmonics of output current are
derived via the eqivalent circuit.
4) The waveform of input current is reconstructed.
5) The operation condition of the current and voltage in
each switching event are recorded.
A. PWM Output Voltage Spectrum
The procedure to calculate the harmonic components of
PWM output voltage in an analytical way can be found in
many literature, like [3], [4]. The benefit is that the relationship
between the amplitude of harmonic components and variables
can be expressed by close-form equations. This approach use
double fourier integral approach to express the PWM switched
waveform by an infinite series of two-dimensional sinusoidal
harmonics.
f(t)=A00+
∞
?
∞
?
n=1
[A0ncos(nω0t) + B0nsin(nω0t)]
+
∞
?
Bmnsin((mωc+ nω0)t)]
m=1
n=−∞
[Amncos((mωc+ nω0)t) + ...
The coefficients can be calculated by,
Amn
=
1
2π2
1
2π2
?π
?π
−π
?π
?π
−π
f(x,y)cos(mx + ny)dxdy
Bmn
=
−π
−π
f(x,y)sin(mx + ny)dxdy
Where, x = ωct, y = ω0t.
An alternative approach is using FFT method, which is
approved to be faster and more efficient. The waveform can be
calculated according to the PWM schemes. Then, the wave-
form is converted to frequency domain via FFT method. The
FFT module is very common and the algorithm has already

Page 3

1178
been optimized to support large scale and quick calculation.
Therefore, the output voltage is expressed by,
f(t) =
∞
?
n=−∞
Cne
2πjn
T
t
(1)
Via these two approaches, the frequency spectrum can be
obtained.
B. The parameters of low frequency induction motor model
For fundamental and harmonic frequency of output voltage,
the equivalent circuit is shown in Figure 3.
R1
X1
X2
R2
Rc
Xm
Vin
Iin
+
-
R2(1/s-1)
(a) fundamental frequency
R1
hX1
hX2
Rc
hXm
Vh
Ih
+
-
R2/s
(b) h-th harmonic voltage
Fig. 3. Per-phase equivalent circuit
Exact values of the parameters in the model of induction
model can be extracted using experimental method. A standard
test procedure for this measurement is suggested by [5]. The
fractional slip s changes with the mechanical load.
The torque-slip relationship can be expressed by [6]:
Tmech=
1
ωs
?
nphV2
1,eq(R2/s)
(R1,eq+ (R2/s))2+ (X1,eq+ X2)2
?
(2)
Here,
V1,eq= Vin
?
jXm
R1+ j(X1+ Xm)
?
(3)
R1,eqand X1,eqare the real and imaginary part of Z1,eq.
Z1,eq=
jXm(R1+ jX1)
R1+ j(X1+ Xm)
(4)
The curve is drawn with load characteristic. The cross point
is the working point when steady-state is reached. With the
known ωm, the s is known for certain load and the model for
the fundermental and harmonic frequency is complete.
C. PWM Load Current Waveform
The fundamental and harmonic component of PWM output
current can be derived from the complete model above. Here
the motor leakage reactances X1and X2are assumed to be
invariant with the frequency.
With the frequency spectrum and phase information of the
output current, the waveform of the output current can be
reconstructed via inverse FFT.
D. Switching Events
We get the current and voltage waveforms from the analysis
in previous level. In the voltage source inverter, the switching
events occur at each voltage transient. The informations of
operation points are recorded with each switching event.
Here gives partial examples inside a switching events file.
The load current is obtained from the current wavefom. The
source voltage at the moment when transient occurs are also
recorded. That is used to calculate the reverse biased voltage
of the rectifier diodes which have influence to the noise
propagation path.
TABLE I
EVENTS LIST EXAMPLE
Event
10
11
12
13
14
15
Il
Vsa
24.8
29.7
33.0
35.1
38.2
43.3
Vsb
-281.0
-283.1
-284.4
-285.3
-286.5
-288.5
Vsc
256.2
253.4
251.4
250.2
248.2
245.2
-22.0
8.1
22.8
18.5
-4.0
-24.7
When the commutation happens between a transistor and a
diode. The transistor is called active switch. The identification
follows these rules:
?
The switching events are classified into two types, that is
turn-on and turn-off. It is necessary to identify them becuase
the slopes of these transients are calculated respectively. The
identification criteria are list below:


The types of transients are identified.
Active switch is
upper switch
lower switch
when Il> 0
when Il< 0
sw. event





upper sw. turns on
upper sw. turns off
lower sw. turns off
lower sw. turns on
when Il> 0 and Vdsrises
when Il> 0 and Vdsfalls
when Il< 0 and Vdsrises
when Il< 0 and Vdsfalls
TABLE II
EVENTS LIST EXAMPLE (CONT.)
Event
10
11
12
13
14
15
Type
turn-on
turn-off
turn-off
turn-on
turn-off
turn-off
V in active leg
-253.8
-253.8
-253.8
253.8
253.8
253.8
V in other two legs
253.8
-253.8
-253.8
-253.8
-253.8
253.8
253.8
253.8
-253.8
-253.8
253.8
253.8
V. TRANSIENT LEVEL
The switching transient in a power converter has tradition-
ally been analyzed by modelling it as a fixed slope dv/dt
and di/dt transient. Actually, this switching transient is a
superposition of multiple slopes [7], further more, these slopes
changes with the load current. The operation condition of the
switches determines the transient edge.
To aviods the excessive approximation that noise signal is
periodic, for the high frequency noise source, the operation
points should be taken into account.

Page 4

1179
t5
t6
Vds(t)
Il
(b)Turn-off
Vds(t)
Id(t)
(a)Turn-on
t4
t1
t3
t2
Il+Irr
Vdc
Vdc
Id(t)
t
t
Il
Fig. 4. Switching waveform
A. Determine the Transient Waveforms
For each IGBT transient events, the waveforms are sketched
in Figure 4. The nonlinear switching transient model is based
on the discussion in [7]. The slow transients are ignored
because it has been proved by simulation that ignoring these
slow transients would not have significant influence to the
noise frequency spectrum. The turn-on transient which is
illustrated in the left figure is first analyzed.
During t1, the Id increases in a slope which can be
calculated by Equation 5.
(dId
dt)t1=gmVg−− Vth− Il/gm
RgCies+ gmLs
(5)
The gmcan be derived from the typical transfer character-
istic curve in datasheet by
gm=
∆Id
∆Vge
(6)
The duration of t1is calculated by,
t1= Il/(dId
dt)t1
(7)
In most datasheets of IGBT, the diode reverse recovery
time trr0and peak reverse current Irr0are measured under
a standard test configuration with IF and
following relationships exist,
dIF
dt. From [8], the
trr,Irr∝
?
(dId
Id
(8)
1/trr,Irr∝
?
dt)t1
(9)
We can calculate trrand Irrby,
Irr= Irr0
?
Il(dId
IF(dIF
dt)t1
dt)
(10)
trr= trr0
?
Il(dIF
IF(dId
dt)
dt)t1
(11)
During t2, the Idchanges from Ilto Il+Irrwith the same
slope.
(dId
dt)t2= (dId
dt)t1
(12)
The time duration for Idrushing to the top is
t2= Irr/(dId
dt)t2
(13)
The time for Idbacking to the Ilis
t3= trr− t2
(14)
During t4, the Vdsdecreases to zero in a slope of
(dVds
dt
)t4=−1
Ccg
?Vg+− Vth− Il/gm
Rg
+Cge
gm
(dId
dt)t1
?
(15)
The duration of t4is calculated by,
t4= Vdc/(dVds
dt
)t4
(16)
The turn-off transient which is illustrated in the right figure
can be analyzed below. During t5, the Vdsincreases from zero
to Vdc, the slope is,
(dVds
dt
)t5=gmVg−− Vth+ Il/gm
RgCcg
(17)
The duration of t5is calculated by,
t5= Vdc/(dVds
dt
)t5
(18)
The overvoltage of Vdsis ignored in the noise model. If the
Vdsreaches Vdc, then the switch current Iddecays to zero.
(dId
dt)t6=gmVg−− Vth− Il/gm
RgCies+ gmLs
(19)
The duration of t6is calculated by,
t6= Il/(dId
dt)t6
(20)
Also, the “tailing” of the drain current of IGBT is ignored
here due to its low transient slope.
B. Transient Noise Source in Frequency Domain
The derivatives of the noise sources are first calculated.
They are then transformed into s domain. The value is the
correction factor to correct the overestimated high frequency
part spectrum of the noise source.

Page 5

1180
The calculation is done after the types of transients are
identified.
1 + Irr/Il
t1+ t2
Irr/Il
t3
−1
t4
−1
t6
1
t5
s
Ion(s)=
1 − e(−(t1+t2)s)
s
1 − e(−t3s)
s
1 − e(−t4s)
s
1 − e−(t6s)
s
1 − e−(t5s)
−
e−((t1+t2)s)
Von(s)=
e−(t1s)
Ioff(s)=
e−(t5s)
Voff(s)=
In Figure 5, the correction factors in many switching events
are calculated and drawn. We do find the influence in different
switching events.
10
2
10
4
10
6
10
8
10
−4
10
−3
10
−2
10
−1
10
0
10
1
(a) current
10
2
10
4
10
6
10
8
10
−4
10
−3
10
−2
10
−1
10
0
10
1
(b) voltage
Fig. 5.Noise source model, consider the variation of parasitic
VI. PROPAGATION LEVEL
In this level, the noise cell can be created based on the wave-
forms obtained in previous level. Unlike simplified models in
other’s work, where the circuits are simplified into CM and
DM eqivalent circuit, the calculation is done on a complete
circuit with fixed topology. transfer ratio is defined as the
ratio between the voltage drop over the LISN 50Ω resistor
and the voltage source and current source. This transfer ratio
is calculated based on the operation condition information at
each switching event to get more exact EMI level. The value
of these parasitic elements is variable with operation points.
A. Basic Noise Cell
The equivalent circuits and noise models have been dis-
cussed for many times. Conventional approaches use voltage
source to derive the CM noise and current source to derive the
DM noise. This is a reasonable approximation for simplicity.
In recent research works, some exemptions are found that the
voltage source can generate DM noise, which is named MM
noise [9], [10]. Also, the current source can generate CM
current flowing through ground [11]. They are illustrated by
dashed lines in Figure 6(a). A basic noise cell is proposed
to get compromise between simplicity for calculation and
completeness of model. We use complete circuits instead of
simplified CM and DM equivalent circuits, shown in Fig-
ure 6(b).
di / dtDM noise path
dv / dtCM noise path
Measurement
Result
Propagation path
(a) Conventional model
Basic cell
di / dt
dv / dt
Measurement
Result
Propagation
path
(b) The new proposed model
Fig. 6.Noise source model,
The approach is described here,
• The switching active components are the noise sources.
The high di/dt and dv/dt associated with the switching
actions is the main reason generating emission. They are
modeled as voltage source and current source correspond-
ing to fast transient of voltage and current.
• The flywheel diodes connected to the active components
are controlled sources. The basic noise cell is shown in
Figure 7.
• All parasitic components are included to get exact pre-
diction. The diagram is shown in Figure 8.

Page 6

1181
n
U
V
W
shaft
L1
Protect Earth
O
U'
V'
W'
G
Line
Line
Cable
Rectifier DC link
Inverter
Motor
Cable
Motor
L2
L3
a
b
c
D1
D3
D5
D4
D6
D2
CL1
CL2
Ch2
Ch1
T1
DT1
T3
DT3
T5
DT5
T6
DT6
T4
DT4
T2
DT2
Converter Frame
Heat sink
Motor Frame
P
N
H
F
Fig. 8.Propagation path includes all parasitics
T+
DT+
T-
DT-
Le
1
Lc1
Lc2
Le2
Lbus1
Lbus2
Le1
Lc1
Lc2
Le2
IT+
IT-=IT+
ZS+
Le1
Lc1
Lc2
ZS-
Le2
ZS+
(a)
(b)
(c)
VT+
VT-= -VT+
ZS-
Lbus1
Lbus1
Lbus2
Lbus2
Fig. 7.
model, (c)voltage noise source model
Noise cell model; (a)diagram of inverter leg, (b)current noise source
Based on the circuit above, the transfer ratio is calculated
for current noise source and voltage noise source respectively.
Two switching events are selected when the active switch is
upper switch, and the switch turns on and turns off.
In each transfer ratio plot, the voltage source or current
source is set to 1V or 1A. The voltage drops in the three
50Ω resistors in LISN are calculated. It is observed, that the
conduct pattern of rectifier diodes does not have significant
impact above 500kHz. While the states in inverter do change
the transfer ratio even in high frequency range.
VII. VALIDATION
The experimental bench has been built to verify this method.
It is designed to represent a typical variable speed drive
system. The main elements are depicted here:
F
10KHz100KHz
V(R54:2)
1.0MHz10MHz100MHz
V(R55:2)V(R5:2)
1.0uV
100uV
10mV
1.0V
100V
100nV
(a) when active sw. turns on
F
10KHz100KHz
V(R54:2)
1.0MHz10MHz100MHz
V(R55:2)V(R5:2)
1.0uV
100uV
10mV
1.0V
100V
100nV
(b) when active sw. turns off
Fig. 9.Voltage source transfer ratio
• line impedance stabilized network (LISN): 50Ω/50µH V-
LISN used for compliant test of band B (150 kHz-30
MHz) defined in CISPR 16. LISN100A3P, 100A 3-Phase,
Cranage EMC Tech. Ltd.;
• variable speed drive: Danfoss VLT5016. It is constituted
by four IGBT modules (FUJI 2MBI 75S-120 2-Pack
IGBT). Three of them are used for inverter legs. The
remaining one is for motor braking;
• power bridge rectifier: SKD62/16, SEMIKRON;
• induction motor: Heemaf Holland BV, 3×380V/7.5kW,
nominal speed: 1445/min 50Hz;
The noise source spectrum, correction factors and transfer
ratio are combined together to predict the EMI level. It
is compared to experiment result. The preliminary result is

Page 7

1182
F
10KHz 100KHz
V(R54:2)
1.0MHz10MHz100MHz
V(R55:2) V(R5:2)
1.0uV
100uV
10mV
1.0V
100V
100nV
(a) when active sw. turns on
F
10KHz 100KHz
V(R54:2)
1.0MHz10MHz 100MHz
V(R55:2) V(R5:2)
1.0uV
100uV
10mV
1.0V
100V
100nV
(b) when active sw. turns off
Fig. 10.Current source transfer ratio
shown below.
Run
40
50
60
70
80
90
100
110
120
.01
.1
1
10
dBuV
Fig. 11.LISN measurement result
VIII. CONCLUSIONS AND FURTHER RESEARCH
This paper presents a short introduction to the necessity
of predicting conducted noise emission of power electronic
converter and its difficulty. It proposes an approach to rapidly
obtain the prediction of the EMI level with enough accuracy. It
is realized via hierachical way. In functional level, large time
constant is used to get the working points of each switching
devices. Then the switching devices are replaced by current
noise source and voltage noise source. In transient level, the
transient edges of the noise source are replaced by piece-wise
linear lines using nonlinear switching transient model. The
operation condition information achieved in the previous level
is used to calculate the transient slope. In the propagation
level, thanks to the established working points information,
the characterization of propagation path is improved.
This approach is used in a voltage source inverter feeding
a induction motor to predict the EMI level. Our research task
is now oriented towards apply this approach to predict EMI
level of a ZCS-CV inverter with variable switching frequency.
REFERENCES
[1] P. Mugur, J. Roudet, and J. Crebier, “Power electronic converter EMC
analysis through state variable approach techniques,” IEEE Trans. Elec-
tromagnetic Compatibility, vol. 43, no. 2, pp. 229–238, 2001.
[2] J. He, J. Jiang, J. Huang, and W. Chen, “Model of EMI coupling paths
for an off-line power converter,” Applied Power Electronics Conference
and Exposition, 2004. APEC’04. Nineteenth Annual IEEE, vol. 2, 2004.
[3] D. Holmes and T. Lipo, Pulse Width Modulation for Power Converters:
Principles and Practice. IEEE Press, 2003.
[4] D. Gonzalez, J. Gago, and J. Balcells, “New simplified method for the
simulation of conducted EMI generated by switched power converters,”
IEEE Trans. Industrial Electronics, vol. 50, no. 6, pp. 1078–1084, 2003.
[5] IEEE-SA Standards Board, “IEEE Standard Test Procedure for
Polyphase Induction Motors and Generators,” IEEE Std 112-2004 (Re-
vision of IEEE Std 112-1996), pp. 1–79, 2004.
[6] A. Fitzgerald, S. Umans, and C. Kingsley Jr, Electric Machinery.
McGraw-Hill Publishing Co., 2002.
[7] J. Meng and W. Ma, “Power converter EMI analysis including IGBT
nonlinear switching transient model,” IEEE Trans. Industrial Electron-
ics, vol. 53, no. 5, pp. 1577–1583, 2006.
[8] N. Mohan, T. Undeland, and W. Robbins, Power Electronics: Converters,
Applications and Design.John Wiley & Sons, Inc, 2003.
[9] D. Zhang, D. Chen, and D. Sable, “Non-intrinsic differential mode noise
caused by ground current in an off-line power supply,” in Proc. IEEE
Power Electronics Specialists Conf. (PESC), vol. 2, pp. 1131–1133,
1998.
[10] J. Meng and W. Ma, “A new technique for modeling and analysis of
mixed-mode conducted EMI noise,” in Proc. IEEE Power Electronics
Specialists Conf. (PESC), vol. 2, pp. 1679–1687, 2004.
[11] X. Huang, “Frequency Domain Conductive Electromagnetic Interference
Modeling and Prediction with Parasitics Extraction for Inverters,” Dis-
sertation, Virginia Polytechnic Institute and State University, Blacksburg,
Virginia, 2004.