IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO.

3, MARCH 2013

1157

A High Frequency Equivalent Circuit and Parameter

Extraction Procedure for Common Mode
Choke in the EMI Filter
Wenhua Tan, Carlos Cuellar, Member, IEEE, Xavier Margueron, Member, IEEE, and Nadir Idir, Member, IEEE

AbstractPower converters with high switching frequency generate conducted electromagnetic interference (EMI) noise. EMI
filters are thus widely used to reduce these conducted noises for the
compliance with electromagnetic compatibility standards. In this
paper, a high-frequency (HF) equivalent circuit model for common
mode (CM) chokes used in EMI filters is proposed together with
its parameter extraction procedure. This procedure is based on
impedance measurements and it incorporates an iterative rational
function approximation fitting algorithm to extract the parameters in the model. The proposed model and procedure is applied
to a planar CM choke which is used to realize an EMI filter. The
simulated results of the filter show good agreement with the experimental ones. This extraction procedure is quite general and
it can also be extended to identify the HF model of other passive
components.
Index TermsCommon mode (CM) choke, electromagnetic
interference (EMI) filter, equivalent circuit, rational function
approximation.

I. INTRODUCTION
LECTROMAGNETIC interference (EMI) filters are commonly used solutions for mitigating the conducted emissions produced by power converters [1]. A typical topology of
EMI filter with ideal components is shown in Fig. 1. However, a
real EMI filter is much more complicated and its performances
depend on many factors such as filter topology, magnetic material, grounding method, stray elements of components, and
parasitic coupling between components etc. [2][4]. Complete
physic-based models of an EMI filter or their components require extensive electromagnetic analysis, so they are very complex and specific to limited applications [5], [6]. Modeling methods based on measurements are more general, direct, and accurate for describing the characteristics of the device under study.
The insertion loss (IL) of an EMI filter is usually measured
with 50 /50 convention. However, the IL of the filter under
working condition depends on the impedances of the source and

Manuscript received February 6, 2012; revised March 30, 2012 and May
25, 2012; accepted July 2, 2012. Date of current version October 12, 2012.
Recommended for publication by Associate Editor K. Ngo.
W. Tan and X. Margueron are with the Laboratory of Electrical Engineering
and Power Electronics of Lille, Ecole Centrale de Lille, Villeneuve dAscq
59655, France (e-mail: [email protected]; [email protected]).
C. Cuellar is with the Universite des Sciences et Technologies de Lille,
Villeneuve dAscq 59655, France (e-mail: [email protected]).
N. Idir is with Instituts Universitaires de Technologie A, University of Lille
1, Villeneuve dAscq 59655, France (e-mail: [email protected]).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TPEL.2012.2209206

Fig. 1.

Typical topology of the EMI filter.

the load, which vary with frequency [7], [8]. In order to obtain
the IL of the filter, many solutions have been reported. In [9], a
four-port measurement method using a vector network analyzer
is presented. This method consists in a black-box modeling of
the whole filter with mixed-mode S-parameters. By postprocessing the obtained S-parameter data, the IL of the filter can hence
be calculated with any source and load impedances. In [10], a
modal model of common mode (CM) chokes based on four-port
S-parameter measurements is presented. Though modal models
enable to analyze the conversion of the noise between differential mode (DM) and CM, they also require compatible modal
model of noise sources for simulations, which complicates the
modeling process. An alternative is to use equivalent circuit
models, which are physic based and are compatible with most
of the simulation tools. In [11], the equivalent circuit of a DM
EMI filter is identified by S-parameters measurements. This approach can correctly extract the parasitic couplings in the filter,
resulting in good modeling precision. Impedance measurement
is a more frequently used technique, which has long been studied
for identifying the equivalent circuits of passive magnetic components [12], [13]. Recently, a lumped-element high-frequency
(HF) model for CM chokes has been proposed [14], [15]. This
model can be easily built by extracting the parameters from
impedance measurement results and can effectively describe
the HF characteristics of the studied the CM choke and EMI
filter. However, the topology of the HF model is chosen heuristically, which requires lots of experience and tests. Moreover, the
extraction procedure derives the parameters of the model by observing the impedance curves, leading to quite time-consuming
trial/error iterations.
To simplify the extraction process, numerical methods can be
applied. In [16], a genetic algorithm is used to extract the behavior model of chokes, yielding an accurate and reliable broadband
equivalent circuit. Nevertheless, the convergence time of such
algorithm is usually long when the problem has a large search
space. In [17], a physics-based equivalent circuit of CM chokes

0885-8993/$31.00 2012 IEEE

1158

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 3, MARCH 2013

Fig. 3. Impedance measurement configurations for the extraction. (a) T 0 .

(b) T 1 . (c) T 2 . (d) T 3 . (e) T 4 .

With these elements, the HF behavior of the magnetizing

impedance Z2 is accurately modeled.
Fig. 2. HF circuit model for CM chokes proposed in [15]. (a) Toroidal CM
choke. (b) HF equivalent circuit. (c) Equivalent circuit of leakage impedance
Z 1 . (d) Equivalent circuit of magnetizing impedance Z 2 .

is built with a rational function approximation (RFA) method

based on linear optimization [18]. However, the topology of the
equivalent circuit is predefined so prior knowledge about the
component is needed.
In this study, an HF model of CM chokes is proposed together
with a systematic extraction procedure. This procedure is aided
by the RFA method incorporating an iterative scheme (referred
as IRFA) and it allows a fast and accurate extraction of the
parameters in the model. This paper is organized as follows.
In Section II, the HF model introduced in [15] is reviewed.
Next, the RFA and the proposed IRFA methods are presented in
Section III. In Section IV, the proposed HF model of CM chokes
and its extraction procedure are introduced and then applied to
a planar CM choke. Finally, experimental verifications of the
procedure as well as some discussion of the results are given in
Section V. Section VI concludes this paper.
II. HF MODEL OF CM CHOKES
A. Review of the Existing HF Model for CM Chokes
The equivalent circuit for toroidal CM chokes proposed in
[15] [see Fig. 2(a)] is illustrated in Fig. 2(b). As seen, it is a
symmetrical structure and the transformation ratio is assumed
to be unity. This model can describe the behavior of many types
of CM chokes as long as the winding symmetry is satisfied. The
leakage impedance Z1 includes the winding resistance, the leakage inductance, and their variations due to skin and proximity
effects of the winding conductor [see Fig. 2(c)]. The magnetizing impedance Z2 describes the behavior of magnetic core,
including the frequency-dependent inductance and losses [see
Fig. 2(d)]. In addition, Ce stands for the self-parasitic capacitance of each winding whereas the sum Cs + Cp + Cc is related
to the interwinding capacitances of the component. Examining
the admittance Y2 = (Z2 )1 , it is found that Y2 contains:
1) one pole at origin: the branch of L2 ;
2) one stable real pole: the branch of R2 , RC 2 , C2 , and R2 ;
3) one pure imaginary pole-pair: the branch of L2 and C2 .

B. Method of Parameter Extraction

To identify the parameters of the model shown in Fig. 2(b),
five selected impedance measurements are carried out, as presented in Fig. 3. All the parameters of the model are obtained
through these impedance measurements summarized as follows:
1) T0 , T1 , T2 , and T3 : parasitic capacitances Cc , Ce , Cp , and
Cs ;
2) T0 and T1 : leakage impedance Z1 ;
3) T2 and T4 : magnetizing impedance Z2 .
The extraction procedure is based on observations of the
impedance curves with some typical frequency responses, e.g.,
20 dB/Dec as inductance and 20 dB/Dec as capacitance. The
values of the parasitic capacitances are calculated using the
resonance frequencies. To fit the model with the measurement
results, some manual adjustments of the parameters are needed
for Z1 and Z2 .
Though good accuracy is achieved in [15], the HF model
shown in Fig. 2(b) can be improved for the following reasons:
1) The extraction procedure requires manual adjustments on
the parameters of the model to achieve a desired accuracy.
As a result, the procedure is complex and time consuming.
Therefore, a computer-aided procedure is preferable.
2) The topology of the equivalent circuit is determined according to experimental observations. For example, the
admittance Y2 = 1/Z2 is chosen to have one stable pole
and one imaginary pole-pair from experience. In consequence, a systematic equivalent circuit synthesis method
is needed to choose the topology.
In view of these necessities, an HF equivalent circuit model
for CM chokes and an IRFA-based extraction procedure are
proposed to facilitate the process. The IRFA method is first
presented in the next section.
III. PARAMETER EXTRACTION USING ITERATIVE RATIONAL
FUNCTION APPROXIMATION
A. Brief Review of Rational Function Approximation Method
The impedance Z(s) (or admittance Y (s)) of a linear circuit
network can be represented by a rational function
Z(s) =

bm sm + bm 1 sm 1 + + b1 s + b0
N (s)
=
D(s)
an sn + an 1 sn 1 + + a1 s + 1

(1)

TAN et al.: HIGH FREQUENCY EQUIVALENT CIRCUIT AND PARAMETER EXTRACTION PROCEDURE

1159

with |m n| 1 and s = j. Without loss of generality, only

the impedance Z(s) is considered in this section. The same
analysis is valid for the admittance Y (s). The goal of the RFA
is to search the values of ai and bi to minimize the error between the rational function Z(s) and measured data Zm eas (s),
as expressed by


N (sk )
D(sk )
k
(2)
where sk = jk . To handle this nonlinear optimization problem,
Levy used a linearization technique [19]. Stemming from Levys
method, the authors introduced in [18] the RFA method for
generating the macromodel of HF interconnects. Reformulating
(2) by multiplying D(sk ) gives
arg min
a i ,b i

|Zm eas (sk ) Z(sk )|2 Zm eas (sk ) =

N (sk ) Zm eas (sk )D(sk ) = 0.

(3)

Separating the real parts and imaginary parts of (3) yields a

linear system of ai and bi

Re[N (sk )] Re[Zm eas (sk )D(sk )] = 0
(4)
Im[N (sk )] Im[Zm eas (sk )D(sk )] = 0.
This system is overdetermined due to the large number of measured points. In order to calculate the unknowns ai and bi , a
least-square method using QR factorization is applied to solve
the overdetermined system (4). Once the values of bi are known,
the poles of (1) can be derived by solving the roots of the denominator D(s). The impedance Z(s) can then be written into
a pole-residue form
Z(s) = d + e s +

n

i=1

ri
.
s pi

(5)

It should be noted that ri and pi can be complex-valued. Again,

the terms d, e, and ri are solved from this linear overdetermined
system (5) by the least-square method [18].
B. Iterative Rational Function Approximation Method
The RFA method is proven to be efficient with noise-free
simulated data. However, it cannot be directly employed for our
application, since Levys linearization technique is known to
have a frequency bias problem [20]. In (2), Zm eas (sk ) is multiplied by D(sk ) that increases rapidly with sk (as frequency
increases). This means the errors of (2) are biased by D(sk ),
making HF errors more important than low frequency (LF) errors. As a result, this frequency bias due to D(sk ) causes poor
fitting precision at LF. However, to correctly identify elements
like dc winding resistance of CM chokes, this lack of fitting
precision at LF is undesirable. To handle the frequency bias, the
RFA method is improved in this study using the Sanathanan
Koerner (SK) iteration [20]. The denominator of the iteration
t 1 D[t1] (sk ) is divided on both sides of (3) for the iteration
t, giving
N [t] (sk )
Zm eas (sk )D[t] (sk )

= 0.
D[t1] (sk )
D[t1] (sk )

(6)

Fig. 4.

IRFA algorithm.

It can be seen that the bias due to D[t] (sk ) is alleviated by

dividing D[t1] (sk ), resulting in better fitting accuracy at LF.
Based on (4) and (6), the proposed IRFA formulation used in
this study is given by

[t]
[t]
Re[PRe (sk )N [t] (sk )] Re[PRe (sk )Zm eas (sk )D[t] (sk )] = 0
[t]

[t]

Im[PIm (sk )N [t] (sk )] Im[PIm (sk )Zm eas (sk )D[t] (sk )] = 0
(7)
[t]
[t]
with the weighting factors PRe (sk ) and PIm (sk ) given by
[t]
PRe (sk) =

[D[t1](s k ) ]

Re[D[t1](s k ) ]Re[N [t1](s k ) ] + Im[D[t1](s k ) ]Im[N [t1](s k ) ]

[t]

P (s ) =

Im k

[D[t1] (sk)]

.
[t1]
[t1]
Re[D
(sk)]Im[N
(sk)]Im[D[t1] (sk)]Re[N [t1] (sk)]
(8)
[t]
[t]
It should be mentioned that in PRe (sk ) and PIm (sk ), a balanced
weighting is used to achieve comparable fitting precision for
both real parts and imaginary parts. The derivation of (8) is
detailed in Appendix. The algorithm of IRFA is given in Fig. 4,
where the iter is defined as the maximum error
iter = max {|Zt (jk ) Zm eas (jk )| / |Zm eas (jk )|} .
k

(9)
The RFA method is first performed to provide a starting
point, the IRFA is then applied until the stopping condition(s)
(in this study: m ax = 10%, Nm ax = 15) are fulfilled. To show
the efficiency of the proposed IRFA algorithm, different methods are applied to fit the measured impedance (from 40 Hz to
70 MHz) of an 8-turn planar inductor with Ferroxcube 3F3Planar E38 ferrite core. A rational function like (1) with m =
n = 6 is used for the fitting. It can be seen in Fig. 5(a) that
the RFA fitting results are wrong on LF due to the frequency
bias problem and the errors decrease when frequency increases.
In Fig. 5(b) and (c), the IRFA methods alleviate the frequency
bias problem in Fig. 5(a). As seen, better fitting accuracies are
achieved at LF. The difference between Fig. 5(c) and (b) is the

1160

Fig. 5.

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 3, MARCH 2013

Comparison of the fitting results. (a) RFA with frequency bias problem. (b) IRFA without balanced weighting. (c) IRFA with balanced weighting.

implementation of balanced weighting. It can be seen from the

error curve shown in Fig. 5(c) that with balanced weighting, the
fitting accuracy of the real part is further improved in comparison with Fig. 5(b). The precision of IRFA fitting is very good
on wide frequency band because the impedance curve is quite
smooth (very few resonances). However, the error may exceed
the specified error level m ax on the two extremities of fitting
range and the resonance frequencies. To achieve a better precision, one can increase the order of the rational function at the
expense of obtaining a more bulky equivalent circuit. In order to
obtain a simple equivalent circuit, the choice of the order turns
out to be a tradeoff between the accuracy and the complexity of
the circuit.
C. Equivalent Circuit Synthesis
Based on the results obtained by the IRFA method, a systematic approach is used for synthesizing the equivalent circuit [21].
Representing (5) by the Foster expansion
Z(s) = d + e s +

Nr

i=1

Nc


ri
+
s pi
i=1

real p oles

s2

ai s + bi
.
+ mi s + ni

com plex conjugate p ole pairs

(10)
Note that the ai , bi , mi , and ni are different variables from
those used in (1). As seen, the Z(s) can be expressed by the
sum of a constant term d, an s-proportional term e, real poleresidue terms, and complex conjugate pole-residue pair terms.
The same expression holds for the admittance Y (s). The constant d corresponds to a resistance and the es term corresponds
to a capacitance for Y (s) or an inductance for Z(s), respectively.
In Table I, the detailed equivalent circuit synthesis methods for
real-pole terms and complex-pole pair terms are summarized.
1) Real pole: with a positive residue ri , the equivalent circuit is simply an RL in series for Y (s) or RC in parallel for
Z(s), as shown in Table I. However, negative ri may appear in

the IRFA results, resulting in negative valued elements. Strictly

speaking, negative elements are not feasible in real circuit design so that they should be avoided. However, for EMI modeling and simulation purposes, these elements can be accepted
for synthesizing an accurate equivalent circuit. Nonetheless, the
negative elements are difficult to handle during time-domain
simulations. To circumvent this difficulty, we suggest another
topology: a positive-valued cell together with a negative resistance, as illustrated in Table I. Suppose that a negative residue ri
is extracted with a stable real pole pi (pi < 0) for the admittance
Y (s). Normally, a negative RL series cell is obtained. However,
according to Table I, the circuit composed of a positive RC
series cell in parallel with a negative resistance R can be used,
where the expression for R, C, and R are given by

positive

R = pi /ri
(11)
C = 1/(pi R) positive

negative.
R = pi /ri
This circuit is equivalent to the negative RL series cell since
they share the same rational function. The negative resistance
R are subsequently merged into the constant term d in (10),
resulting in a new constant term d = d 1/R = d ri /pi .
Similar analysis can be done for the case of impedance Z(s).
With these circuits, negative elements for real pole terms can be
reduced to minimum.
2) Complex-pole pair: to treat complex-pole pair, a fourelement circuit can be used [21]. It is the minimal type realization since there are four free variables ai , bi , mi , and ni in
the rational function (see Table I). Another equivalent circuit
referred as extended type, uses a six-element circuit but with
easier parameter calculation formulas [22]. In Fig. 6, the equivalent circuit corresponding to the admittance Y (s) in Foster
expansion is illustrated.
As seen, the proposed IRFA together with the equivalent
circuit synthesis method enables generating a highly accurate
HF model from a measured impedance (or admittance) with

TAN et al.: HIGH FREQUENCY EQUIVALENT CIRCUIT AND PARAMETER EXTRACTION PROCEDURE

1161

TABLE I
EQUIVALENT CIRCUIT SYNTHESIS FOR REAL POLES AND COMPLEX-POLE PAIRS

TABLE II
MEASUREMENT CONFIGURATIONS FOR THE PROPOSED
EXTRACTION PROCEDURE

Fig. 6.

Fig. 7.

Equivalent circuit synthesis for the Foster expansion of Y (s).

Proposed HF model for CM chokes.

flexibility. Based the IRFA and the circuit synthesis method, the
HF model for CM chokes will be introduced in the following
section.
IV. IRFA ADAPTED HF MODEL OF CM CHOKES
AND EXTRACTION PROCEDURE
A. Proposed HF Model of CM Chokes
The proposed HF model of CM chokes is shown in Fig. 7. It
should be noted that the parasitic capacitances are assumed to be
lossless. Besides, the capacitances Cs in Fig. 2(b) are discarded
in our model since Ce , Cp , and Cc are sufficient for describing
the electrostatic behavior of a strongly coupled two-winding
transformer [13]. Instead of using a heuristic method to identify

the leakage impedance Z1 and the magnetizing impedance Z2 ,

the topologies of Z1 and Z2 of the proposed model are represented by the equivalent circuit in Fig. 6, which are determined
by the results of the IRFA method. It is to be noted that the
choice of the topologies follows strict mathematical calculation
so that the final equivalent circuit is guaranteed to be accurate
over a wide frequency range.
B. Extraction Procedure
The extraction of the proposed HF model shown in Fig. 7
begins with three specific impedance measurements T3 , T0 , and
T2 , as shown in Table II. The IRFA method is then applied to
extract the equivalent circuit for the impedances ZT 3 , ZT 0 , and
ZT 2 . Note that the measurement T1 in Fig. 3 is discarded since
Cs is removed from the model. Moreover, the measurement T4
used in [15] is not necessary in the proposed procedure because
T2 can provide enough information to extract the magnetizing
impedance Z2 . In order to validate the proposed method and the

1162

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 3, MARCH 2013

illustrated in Fig. 8(b). The fitting results for YT 0 (s) are listed
in Table III.
It can be seen from Table III that the extracted eT 0 is negative,
resulting in a negative capacitance. The latter is not allowed
because it will cause simulation instabilities [24]. In fact, the
phase plot of ZT 0 [see Fig. 8(b)] shows that this impedance is
not capacitive beyond the resonance frequency fr 51 MHz.
Around this frequency, the measured impedance ZT 0 is affected
by distributed effect because the winding length of the studied
CM choke is about 1 m, which is comparable to the wave length
of the FR-4 based PCB structure at fr . Therefore, using the
simple parallel circuit shown in Table II to describe ZT 0 or
YT 0 gives rise to the negative capacitance. To obtain a positive
capacitance, an approximation is performed. First, the complex
pole-residue pair (pT 0 , rT 0 ) (pT 0 , rT 0 ), dT 0 = 3.0893E 3
and eT 0 = 3.2803E 12 terms of YT 0 (s) in Table III are
recombined together to form a new admittance called YC T 0 (s)
YC T 0 (s) = dT 0 + eT 0 s +

rT 0
rT 0
+
.
s pT 0
s pT 0

(13)

Indeed, the true parasitic capacitance is merged in YC T 0 (s).

The expression (13) is then simplified using YC T 0 (s) = dT 0 +
eT 0 s by equating the values of YC T 0 (s) and YC T 0 (s) at the
resonance frequency sr = j2fr , as given by
Fig. 8.

Measured impedances for the extraction. (a) Z T 3 . (b) Z T 0 . (c) Z T 2 .

extraction procedure, it is applied to model a planar CM choke

with eight turns on each winding. First, impedance measurements are carried out with an HP4294A precision impedance
analyzer. The configurations T3 , T0 , and T2 are treated successively, as described in the following parts. Configuration T3
represents a capacitive configuration which gives the value of
Cc + Cp . The admittance YT 3 = 1/ZT 3 is fitted with the expression
YT 3 = 1/ZT 3 = d + e s

(12)

where d denotes the dielectric losses and e denotes the capacitance 2(Cc + Cp ). The fitting starts from 100 Hz to guarantee
a good measurement precision of capacitances. As the capacitance model is an RC parallel cell, the fitting stops at 30 MHz
to exclude the resonance, as shown in Fig. 8(a). The loss term d
is neglected according to the lossless capacitance assumption.
The extracted result for 2(Cc + Cp ) is 29.98 pF.
Configuration T0 is a flux-subtracting configuration so only
the leakage impedance Z1 /2 and the capacitance 2(Ce + Cp )
are concerned. Again, to facilitate the circuit synthesis, the admittance YT 0 = 1/ZT 0 is treated in the IRFA. In fact, the capacitance 2(Ce + Cp ) corresponds the es term in Fig. 6, so the
rational function YT 0 (s) represented in the form of (1) must
fulfill: m n = 1. In order to obtain a simple and accurate
equivalent circuit, the values of m and n should be lowest possible but still guarantee an acceptable accuracy. In this case, we
choose m = 6 and n = 5 after a few trials. In order to extract the
dc resistance as well as the parasitic capacitance 2(Ce + Cp ),
the fitting frequency range covers from 40 Hz to 70 MHz, as

rT 0
rT 0
+
= dT 0 + eT 0 sr .
sr pT 0
sr pT 0
(14)
The obtained results are: eT 0 = 3.22E 11 and dT 0 = 1.48E
03, which can be further represented by an RC parallel
cell. With this simplification, the positive parasitic capacitance
2(Ce + Cp ) and its parallel resistance are obtained. As the complex pole pair is thrown away during the approximation, the
fitting precision of the real part of ZT 0 is impaired at high
frequencies, since the complex poles are indispensable for obtaining high fitting precision around the resonance frequency,
as shown in Fig. 9(a). However, tradeoffs have to be accepted
with this lumped-element equivalent circuit.
The final circuit representation for impedance ZT 0 is shown
in Fig. 10(b). The impedance Z1 /2 is contained in the dashed
box. The elements of the RL cells are calculated with the real
pole-residue pairs listed in Table III. These components can be
interpreted as the partial element equivalent circuits for modeling the eddy current effect.
Configuration T2 is used as the last step of the procedure
to identify the equivalent circuit of ZT 2 . A rough calculation
is first performed at 100 kHz to estimate the CM inductance
with Im(ZT 2 ) = LCM /2, giving LCM = 382.4 H. Similar
to ZT 0 , the admittance form YT 2 is processed for the fitting,
using a rational function with m = 6 and n = 5. As shown
in Fig. 8(c), the fitting stops at 30 MHz, just before the second
resonance that cannot be modeled by the lumped-element equivalent circuit in Fig. 7. The fitting results are shown in Table III.
According to the equivalent circuit T2 in Table II, eT 2 corresponds to the value of the parasitic capacitance 2(Ce + Cc )
whereas the remaining parts (i.e., dT 2 , real poles and complex poles) represent the admittance 2(Z1 + Z2 )1 . With Z1
dT 0 + eT 0 sr +

TAN et al.: HIGH FREQUENCY EQUIVALENT CIRCUIT AND PARAMETER EXTRACTION PROCEDURE

1163

TABLE III
FITTING RESULTS FOR THE PLANAR CM CHOKE

Fig. 9.

Equivalent circuit extraction using IRFA. (a) Z T 0 . (b) Z T 2 .

positive R(2.28 k)C(2.03 pF) series branch, which is interpreted as the nonmagnetic HF losses in the ferrite material due
to the capacitive isolation of the grains [23]. The final equivalent circuit of Z2 /2 and ZT 2 is shown in Fig. 10(c). In Z2 , the
L(190 H)R(4 m) branch presents the dominant pole according the definition: a pole pi with its residue ri is called dominant
pole if its Fi = |ri /Re(pi )| is much larger than the other poles.
The values Fi for the poles of ZT 0 and Z2 are listed in Table III
and the Fi of the L(190 H)R(4 m) branch is much larger
than the others. Therefore, it has the most important influence
in Z2 . The 190-H inductance is in fact the CM filtering inductance because it is half the CM inductance LCM estimated
from ZT 2 . The simulated impedance of ZT 2 is compared with
measured data in Fig. 9(b) in terms of real part and magnitude.
As seen, the simulated results agree well with the measurements
up to 30 MHz.
With the parasitic capacitances obtained from the three configurations, three linear equations expressed in (15) can be established, giving the values of the stray capacitances: Cc = 6.1 pF,
Ce = 7.2 pF, and Cp = 8.9 pF

2Cc + 2Cp = 30 pF
(15)
2Ce + 2Cp = 32.2 pF

2Ce + 2Cc = 26.7 pF.

Till now the HF model for the studied planar CM choke is established. The obtained results will be validated by experimental
tests in next section.
V. EXPERIMENTAL VERIFICATION AND DISCUSSION
A. Experimental Verification

Fig. 10.

Extracted circuits for the CM choke. (a) Z T 3 . (b) Z T 0 . (c) Z T 2 .

being identified [see Fig. 10(b)], the magnetizing impedance

Z2 is calculated by subtracting the contribution of Z1 . The
IRFA is applied again to get the pole-residue information of
the admittance Y2 (s) = 1/Z2 (s), as listed in Table III. Note
that a real pole p = 2.1616E + 08 with negative residue
r = 9.4903E + 04 is extracted. This negative pole-residue is
treated with the method presented in Table I, resulting in a

To validate the HF model and the procedure, the planar CM

choke is first examined by impedance measurements. Two measurement configurations are chosen: open circuit ZO and short
circuit ZSC . These two configurations correspond to extreme
loaded impedances for the component. Using SPICE, an ac
analysis from 40 Hz to 30 MHz (Log Sweep) is performed on
the identified HF model and the simulation results are compared
with measurements data, as shown in Fig. 11. It can be seen that
the simulated curves match closely with the measured ones from
40 Hz to 30 MHz.
As there are many small-valued elements in the model, the
sensitivity of the model should be examined. Some qualitative
results are given for this analysis.

1164

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 3, MARCH 2013

Fig. 12. IL measurements of the EMI filter. (a) EMI filter for test. (b) Equivalent circuit of capacitors. (c) Circuit of the measurement. (d) Measurement
setup.

Fig. 11. Comparison between the measured and simulated impedances of the
HF model. (a) Open circuit test-Z O . (b) Short circuit test-Z S C .

1) The three parasitic capacitances Ce , Cp , and Cc affect

the resonance frequency and the response beyond the
resonance.
2) In Z1 and Z2 , the branches of dominant poles determine
the behavior of the model on the inductive region. The
model is less sensitive to the parameters in the braches of
nondominant poles.
3) The impedance around the resonance varies rapidly so they
are sensitive to the dominant poles branches, the complex
pole-pairs branches and parasitic capacitances.
In a word, the model is quite robust because the behavior of
the model does not change abruptly due to small variations of
some parameters. However, cares should always be taken during
the measurements and the fitting process to guarantee a good
accuracy of the extracted model.
Next, an EMI filter using the topology shown in Fig. 1 is
realized with the planar CM choke, as shown in Fig. 12(a). The
leakage of the CM choke acts as DM inductors. The impedance
of the CM capacitors (4.7 pF) and the DM capacitor (68 nF) are
measured by an impedance analyzer. The equivalent circuit for
the capacitors is given in Fig. 12(b) and their impedance ZC (s)
can be expressed by
ZC (s) = RESR + sLESL +

RP
sCRP + 1

(16)

where RESR stands for equivalent series resistance and LESL

for equivalent series inductance. To get the values of the parameters, (16) can be fitted by the IRFA method using a rational
function (1) with m = 2 and n = 1. The IL of the filter is measured on an Agilent 5071C network analyzer using a four-port
S parameter measurement method [9] [see Fig. 12(c) and (d)].

The S parameter matrix [Sij , i, j = 1, 2, 3, 4] is measured from

100 kHz to 30 MHz with logarithmic sweep. Postprocessing of
this matrix is performed to obtain the mixed-mode S parameter
Scc21 and Sdd21 through the following equations:

Scc21 = (S21 + S23 + S41 + S43 )

2
(17)

1
S
(S
=

S
+
S
)
dd21
21
23
41
43
2
where Scc21 and Sdd21 give the CM and DM IL for the EMI
filter, respectively [9]. Moreover, a four-port S parameter simulation is performed over the same frequency range with the same
configuration as Fig. 12(c). The simulated CM IL Scc21 sim and
DM IL Sdd21 sim are deduced by (17) as well. The comparison
between the measured ILs and the simulated ones are presented
in Fig. 13. It can be seen that the simulated result closely matches
the measurement for CM. However, large difference at HF is observed for the DM. This is due to the parasitic coupling effects
among the choke, capacitors, and trace loop [4], [11], which are
not considered in this model. However, this is out of the scope
of this paper.
B. Discussion
Through the IRFA method, accurate models for CM chokes
can be built from impedance measurements. However, several
important aspects of the model and the procedure should be
addressed.
1) Passivity: as the whole extraction procedure treats passive
magnetic components, the passivity issues have to be considered. The passivity of a component requires that the
component be dissipative in energy. However, the IRFA
method cannot guarantee the passivity of the calculated
rational function. During the past ten years, many works
have been reported on the passivity verification and enforcement techniques [24], [25]. In this paper, all the extracted results are verified to be passive using the method
of [24]. In practice, as CM chokes are naturally dissipative

TAN et al.: HIGH FREQUENCY EQUIVALENT CIRCUIT AND PARAMETER EXTRACTION PROCEDURE

1165

plied on a planar CM choke and is validated by simulations and

measurements.
APPENDIX
According to the SK iteration, the denominator of the iteration
t 1 D[t1] (sk ) is divided at both sides of the (3) of the iteration
t, resulting in (6) and
N [t] (sk )(D[t1] (sk ))
Zm eas (sk )D[t] (sk )(D[t1] (sk ))
=
[t1]
2
|D
(sk )|
|D[t1] (sk )|2
(18)
where the superscript denotes the complex conjugate. For
magnetic components, the imaginary part of the impedance
(Z L) is much larger than its real part (losses) in the inductive region. This unbalance will lead to a low fitting quality
for the real part during the least square process. In order to
achieve a balanced fitting precision between the real part and
the imaginary part, the real and imaginary parts of (18) are normalized respectively by 1/Re[Z [t1] (sk )] and 1/Im[Z [t1] (sk )]
as weighting factors
Fig. 13.

Comparison of simulated and measured IL. (a) CM. (b) DM.

in HF due to core losses and copper losses, the passivity

is usually respected if fitting precision is good enough.
2) Distributed effect: due to the distributed effect, manual
adjustments are needed to avoid negative elements, which
require some experience about the equivalent circuit synthesis. This is the limit of using a lumped-element equivalent circuit model. The model presented in this paper is effective up to 30 MHz, which corresponds to the frequency
range of interest for the conducted emissions. Nonetheless, it is necessary to consider the distributed effect to
cover frequencies beyond 30 MHz, which requires implementing a distributed circuit model.
3) Saturation: CM chokes always work under flux bias which
will cause saturation effect [26]. The proposed model is
established on the basis of small signal measurements.
Therefore, the error of the model will increase as the current increases. However, the primary focus of this paper is
to propose a small signal HF model of CM chokes that can
be systematically synthesized. As the first step, the small
signal model is now available and it can then be coupled
with nonlinear cell that accounts for the saturation of the
magnetic material, which is the objective of a future study.
VI. CONCLUSION
In this paper, an HF equivalent circuit model of CM chokes
used in the EMI filter is proposed incorporating a systematic extraction procedure. The procedure uses an iterative rational function approximation method to extract pole-residue
form expressions from impedance measurements. Based on the
pole-residue information, a systematic synthesis method is applied to generate the equivalent circuit. The proposed procedure
gives accurate models in a short time. This procedure is ap-

1
Re[Z [t1] (sk )]

|D[t1] (sk )|2

Re[N [t1] (sk )(D[t1] (sk )) ]

|D[t1] (sk )|2

=
.
Im[Z [t1] (sk )]
Im[N [t1] (sk )(D[t1] (sk )) ]

(19)

Combining the SK iteration (18) with the weighting factors (19),

the final formulations (7) and (8) are derived.
REFERENCES
[1] K. Mainali and R. Oruganti, Conducted EMI mitigation techniques for
switch-mode power converters: A survey, IEEE Trans. Power Electron.,
vol. 25, no. 9, pp. 23442356, Sep. 2010.
[2] Y. Maillet, R. Lai, S. Wang, F. Wang, R. Burgos, and D. Boroyevich,
High-density EMI filter design for DC-Fed motor drives, IEEE Trans.
Power Electron., vol. 25, no. 5, pp. 11631172, May 2010.
[3] M. Hartmann, H. Ertl, and J. W. Kolar, EMI filter design for a 1 MHz,
10 kW three-phase/level PWM rectifier, IEEE Trans. Power Electron.,
vol. 26, no. 4, pp. 11921204, Apr. 2011.
[4] S. Wang, F. C. Lee, D. Y. Chen, and W. G. Odendaal, Effects of parasitic
parameters on EMI filter performance, IEEE Trans. Power Electron.,
vol. 19, no. 3, pp. 869877, May 2004.
[5] M. Kovacic, Z. Hanic, S. Stipetic, S. Krishnamurthy, and D. Zarko, Analytical wideband model of a common-mode choke, IEEE Trans. Power
Electron., vol. 27, no. 7, pp. 31733185, Jul. 2012.
[6] X. Wu, D. Xu, Z. Wen, Y. Okuma, and K. Mino, Design, modeling, and
improvement of integrated EMI filter with flexible multilayer foils, IEEE
Trans. Power Electron., vol. 26, no. 5, pp. 13441354, May 2011.
[7] V. Tarateeraseth, B. Hu, K. Y. See, and F. G. Canavero, Accurate extraction of noise source impedance of an SMPS under operating conditions,
IEEE Trans. Power Electron., vol. 25, no. 1, pp. 111117, Jan. 2010.
[8] P. Kong, Y. Jiang, and F. C. Lee, Common mode EMI noise characteristics
of low-power ACDC converters, IEEE Trans. Power Electron., vol. 27,
no. 2, pp. 731738, Feb. 2012.
[9] K. S. Kostov and J. J. Kyyra, Insertion loss in terms of four-port network
parameters, IET Sci., Meas. Technol., vol. 3, pp. 208216, May 2009.
[10] A.-M. Sanchez, A. Perez, J. R. Regue, M. Ribo, P. Rodrguez-Cepeda, and
F. J. Pajares, A modal model of common-mode chokes for conducted
interference prediction, IEEE Trans. Electromagn. Compat., vol. 52,
no. 3, pp. 749752, Aug. 2010.
[11] S. Wang, F. C. Lee, and W. G. Odendaal, Characterization and parasitic
extraction of EMI filters using scattering parameters, IEEE Trans. Power
Electron., vol. 20, no. 2, pp. 502510, Mar. 2005.

1166

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 28, NO. 3, MARCH 2013

[12] X. Margueron and J.-P. Keradec, Identifying the magnetic part of the
equivalent circuit of n-winding transformers, IEEE Trans. Instrum.
Meas., vol. 56, no. 1, pp. 146152, Feb. 2007.
[13] A. Schellmanns, K. Berrouche, and J.-P. Keradec, Multiwinding transformers: A successive refinement method to characterize a general equivalent circuit, IEEE Trans. Instrum. Meas., vol. 47, no. 5, pp. 13161321,
Oct. 1998.
[14] J.-L. Kotny, X. Margueron, and N. Idir, High frequency modeling method
of EMI filters, in Proc. IEEE Energy Convers. Congr. Expo.Conf., Sep.
2009, pp. 16711678.
[15] J.-L. Kotny, X. Margueron, and N. Idir, High frequency model of the
coupled inductors used in EMI Filters, IEEE Trans. Power Electron.,
vol. 27, no. 6, pp. 28052812, Jun. 2012.
[16] S. Skibin and I. Stevanovic, Behavioral circuit modeling of chokes with
multi-resonances using genetic algorithm, in Proc. IEEE Symp. Electromagn. Compat., Aug. 2011, pp. 454458.
[17] W. Tan, C. Cuellar, X. Margueron, and N. Idir, Automatic identification of
magnetic component equivalent circuits using impedance measurements,
in Proc. IEEE Instrum. Meas. Technol. Conf., May 2011, pp. 13171322.
[18] M. Elzinga, K. Virga, L. Zhao, and J. L. Prince, Pole-residue formulation
for transient simulation of high-frequency interconnects using householder
LS curve-fitting techniques, IEEE Trans. Adv. Packag., vol. 25, no. 2,
pp. 142147, May 2000.
[19] E. C. Levy, Complex-curve fitting, IRE Trans. Automat. Contr., vol. AC4, pp. 3744, 1959.
[20] C. K. Sanathanan and J. Koerner, Transfer function synthesis as a ratio
of two complex polynomials, IEEE Trans. Autom. Control, vol. AC-8,
no. 1, pp. 5658, Jan. 1963.
[21] G. Antonini, Spice equivalent circuits of frequency domain responses,
IEEE Trans. Electromagn. Compat., vol. 45, no. 3, pp. 502512, Aug.
2003.
[22] Z. Ye and Z. Yu, Passive-assured rational function approach for compact
modeling of on-chip passive components, in Proc. Int. Conf. Solid-State
Integr. Circuits Technol., 2004, pp. 217220.
[23] A. Schellmanns, P. Fouassier, J.-P. Keradec, and J.-L. Schanen, Equivalent circuits for transformers based on one-dimensional propagation:
Accounting for multilayer structure of windings and ferrite losses, IEEE
Trans. Magn., vol. 36, no. 5, pp. 37783784, Sep. 2000.
[24] B. Gustavsen and A. Semlyen, Enforcing passivity for admittance matrices approximated by rational functions, IEEE Trans. Power Syst., vol. 16,
no. 1, pp. 97104, Jan. 2001.
[25] S. Grivet-Talocia and A. Ubolli, A comparative study of passivity enforcement schemes for linear lumped macromodels, IEEE Trans. Adv.
Packag., vol. 31, no. 4, pp. 673683, Nov. 2008.
[26] N. Zhu, J. Kang, D. Xu, B. Wu, and Y. Xiao, An integrated AC choke
design for common-mode current suppression in neutral-connected power
converter systems, IEEE Trans. Power Electron., vol. 27, no. 3, pp. 1228
1236, Mar. 2012.

Carlos Cuellar (M12) was born in Peru, in 1982.

He received the Diploma degree in electronic engineering from the National University of San Antonio
Abad, Cusco, Peru, in 2004, and the M.S. degree in
electrical engineering from the Universite des Sciences et Technologies de Lille, Villeneuve dAscq,
France, in 2010, where he is currently working toward the Ph.D. degree.
In 2008, he was an Engineer in the Machupicchu
Hydroelectric plant, Peru. His current research interests include power conversion, passive EMI filter
design, and the characterization of magnetic materials used in power electronics.

Wenhua Tan was born in Jiangsu, China, in 1984.

He received B.E. and M.E. degrees from the Xian
Jiaotong University, Xian, China, in 2006 and 2009,
respectively, and the Eng. degree from the Ecole Centrale de Lyon, Ecully, France, in 2009. Since 2009, he
has been working toward the Ph.D. degree in electrical engineering in the Ecole Centrale de Lille, Universite Lille Nord de France, Villeneuve dAscq, France.
His main research interests include passive component integration, passive component modeling, and
EMI filters design.

Xavier Margueron (M09) was born in Chambery,

France, in 1980. He received the Diploma degree
in electrical engineering from the Ecole Nationale
Superieure dIngenieurs Electriciens de Grenoble,
Grenoble, France, in 2003, and the Doctorat de Genie
Electrique from the Universite Joseph Fourier, Grenoble, in 2006.
Since September 2007, he has been an Associate
Professor at the Ecole Centrale de Lille, Universite
Lille Nord de France, Lille, France, where he carries
out research in the Laboratoire dElectrotechnique et
dElectronique de Puissance de Lille. His research interests include HF power
electronics and passive component modeling.

Nadir Idir (M93) received the Ph.D. degree electrical engineering from the University of Lille 1, France,
in 1993.
He is currently a Full Professor with Instituts Universitaires de Technologie A of the University of Lille
1, where he teaches power electronics and electromagnetic compatibility. Since 1993, he has been with
the Laboratory of Electrical Engineering and Power
Electronics, University of Lille 1. His research interests include, design methodologies for HF switching
converters, power devices (SiC and GaN), electromagnetic interference (EMI) in static converters, HF modeling of the passive
components, EMI filter design methodologies for switching converters.