This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2845467, IEEE
Transactions on Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1

A Transformer-Isolated Common-Mode Active

EMI Filter without Additional Components on
Power Lines

Sangyeong Jeong, Dongil Shin, and Jingook Kim, Senior Member, IEEE

 [4]-[20]. The performance of the AEFs using an OP-amp has

Abstract— This paper proposes a new structure of been explained with the concept of impedance multiplication
transformer-isolated active EMI filter (AEF) without additional produced by the AEFs [4], [5]. The AEFs employing a
components on power lines for the reduction of common-mode push–pull amplifier for reduction of the CM noises have been
(CM) conducted emissions (CE). The noise attenuation
presented in [6]-[9]. Another kind of AEF, which generates
performance and loop gain are rigorously analyzed, and the
design guides for the proposed AEF are developed. The AEF is digitally controlled compensation signals, has been also
designed in a compact package for the application to a switching proposed in [10] and [11]. The topology of the AEFs can be
mode power supply (SMPS). The performance, stability, and classified according to the noise sensing and compensation
leakage currents of the AEF are experimentally validated by methods [12]-[14]. The AEFs of current-sense
various measurements. The CM EMI filter, including the AEF, current-compensation (CSCC) and voltage-sense
achieves the additional CE reduction by 10 to 20 dB at a
voltage-compensation (VSVC) employ additional transformers
frequency range lower than 1MHz. Also, the noise peaks due to
the resonances at several MHz are reduced by 7dB. The weak as well as capacitors for noise sensing and compensation
points of the conventional CM EMI filter at both low and high [15]-[17]. The current-sense voltage-compensation (CSVC)
frequencies are effectively improved by employing the proposed AEFs in [18], [19] utilize only transformers at both sensing and
AEF. compensating sides, while the voltage-sense
Index Terms— Electromagnetic interference (EMI), conducted current-compensation (VSCC) AEFs in [9] and [20] are
emission (CE), active EMI filter (AEF), common mode (CM),
implemented with only capacitive coupling.
transformer-isolated, current-sense current-compensation
(CSCC), current-sense voltage-compensation (CSVC), leakage The capacitive coupling has the advantage of compact-size
current over the transformer coupling. Because the transformers for
noise sensing and compensation are usually installed on the
I. INTRODUCTION power lines, the wires wound at the transformer should be

T HE electromagnetic interference (EMI) problems are sufficiently thick against a high current flow and heat problems,
critical issues in a switching mode power supply (SMPS) which increases the size and cost of the transformer. Also, the
with a fast switching frequency [1], [2]. The conducted transformer can be affected by the magnetic saturation
emissions (CE) are regulated from 10kHz to 30MHz, which degrading the performance of AEF [17]. Therefore, the AEF of
enforces a careful design of the SMPS and EMI filters. An EMI the VSCC topology employing only capacitive coupling is the
filter should be installed at the side of AC utility lines to reduce most desirable for a high-power application in terms of the size
the CE noises from the power converters. The common-mode and cost [9]. However, the capacitor-coupling method has a
(CM) CE noises are usually suppressed by using CM chokes significant drawback in reliability and safety. According to how
and Y-capacitors. To suppress the CM noises at low-frequency the AEF supply voltage is provided, the reference voltage of the
range, a large CM choke or large Y-capacitor in a filter stage is AEF power supply can be different from the earth ground,
required. However, the value of the Y-capacitor is limited by which may cause electrical overstress (EOS) problems at the
the safety regulation on leakage currents [2], and a large CM coupling capacitors or other active components of the AEF [21].
choke is bulky and costly [3]. Also, when the capacitor for sensing or compensation is
Active EMI filters (AEFs) have been suggested to effectively accidently broken due to the EOS or electromagnetic
suppress the CE without significant increase of the filter size susceptibility (EMS) problems such as lightning, it can be a
short or low-impedance path [22]. In this case, active
components in the AEF are directly exposed to the main power
Manuscript received xxxx, 2018. lines, and the semiconductor devices can be critically damaged
This work was supported in part by the Basic Science Research Program
through the National Research Foundation of Korea (NRF and burnt-out.
-2016R1D1A1B03934382). (Corresponding author: Jingook Kim) The galvanic isolation by using transformers is usually much
S. Jeong, D. Shin, and J. Kim are with the school of ECE, Ulsan National safer than the capacitive coupling against faults of the sensing
Institute of Science and Technology, Ulsan 44919, South Korea (e-mail: ;
[email protected]; [email protected]; [email protected]). or compensation components [23]. Moreover, if an AEF is fully

0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2845467, IEEE
Transactions on Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 2

transformer-isolated from the power lines, active circuit product SMPS board. Furthermore, the amount of leakage
components in the AEF can operate at any different reference current to the earth ground is measured to confirm the safety of
voltages from the earth ground, which provides an important using the AEF.
design freedom in the dc-supply voltage for the AEF. The AEFs
in the CSVC topology are well transformer-isolated; however, II. ANALYSIS OF THE PROPOSED AEF
they become too bulky and costly because the transformers are
located along the main power lines requiring thick wires, which
also cause high risks of the magnetic saturation and heat
problems. Thus, there have not been many researches on the
CSVC-type AEFs, and only a few are previously published [18],
[19].

Fig. 1. Configuration of the proposed transformer-isolated AEF installed as an Fig. 2. Circuit model of the proposed AEF.
add-on to the CM L-C EMI filter.
The circuit model of the proposed AEF is described in Fig. 2.
This paper proposes a new structure of the fully CY represents the capacitance of the Y-capacitors. The CM
transformer-isolated AEF. Fig. 1 depicts the configuration of choke is modeled as Lcm and Mcm, which represent the
the proposed AEF. The AEF is installed as an add-on to the self-inductance and mutual-inductance of the windings on
existing conventional CM L-C EMI filter consisting of a CM power lines.
choke and Y-capacitors. The proposed AEF structure is similar The AEF is mainly composed of the sensing winding wound
to the conventional CSCC AEF topology, but an injection on the CM choke, the amplifier part, and the injection
transformer is added between the amplifier output and the transformer. The turn ratio of the winding on a power line to the
sensing winding is set as 1:Nsen, and the self-inductance of the
compensation Y-capacitors. Because the injection transformer
sensing winding is approximately given as Nsen2Lcm. Msen
is not installed on the main power lines, only the small
represents the mutual inductance between the winding on a
compensation signal current flows at the transformer. The
power line and the sensing winding for the AEF input. Likewise,
injection transformer can be realized in a compact size due to its Minj represents the mutual inductance in the injection
small current regardless of the operating current of the transformer, and the turn ratio of the primary winding to the
application, resulting in a low risk of the magnetic saturation secondary winding is set as 1:Ninj. The self-inductance at each
and heat problem. Also, the sensing part of the AEF does not side is then given as Linj and Ninj2Linj, respectively. Note that
require any additional transformer, but a thin noise-sensing Mcm, Msen, and Minj are calculated as kcmLcm, ksenNsenLcm, and
wire is additionally wound on the already existing commercial kinjNinjLinj, respectively, where kcm, ksen, and kinj represent each
CM choke. There have been attempts to add a sensing winding coupling coefficient. The values of kcm, ksen, and kinj are usually
directly on a commercial CM choke [8], [24], but the adverse distributed from 0.99 to 1 in a practical design. Last, the
effects and the maximally allowable turn number of the sensing amplifier part is implemented with the non-inverting OP-amp
winding have not been investigated. In summary, the major configuration with the resistors, R1 and R2.
novel feature of the proposed AEF can be characterized as 1) Considering the feedback stability of the AEF, several
fully transformer-isolated and 2) compact-sized design without additional components, such as the low-pass filter, bypass
additional components on the main power lines. By virtue of branch, and phase compensator, are required, as shown in Fig. 2.
these features, the proposed AEF has a smaller size and better Rf and Cf comprise the low-pass filter at OP-amp input to
performance than the other transformer-isolated CSVC AEFs prevent the stability problem caused by the sensing winding
in [18] and [19]. resonance at the high-frequency range. Rd1, Cd, and Rd2 also
Many useful explicit design guidelines for complete design work for stability as bypass and damping circuits to avoid the
resonance at the injection transformer, which additionally
of the proposed AEF are derived in this paper. In Section II, the
mitigate the performance degradation due to the resonance
proposed transformer-isolated AEF is analyzed to estimate the
between noise source impedance and the Y-capacitors. Rc and
noise attenuation performance. Based on the analysis, proper
Cc are the phase compensator for the stability at the
design guidelines for the performance and stability of the AEF low-frequency range.
is developed in Section III. In Section IV, the filter insertion Although the ground reference voltage of the AEF is set
loss and loop gain of the AEF are validated by measurements differently from the earth ground, the AEF is symmetric with
using a vector network analyzer (VNA). The reduction of the regard to the ac zero potential and can be analyzed by bisecting
CM CE noise by the AEF is also demonstrated with a real the circuits. The equivalent circuit model of a half portion is

0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2845467, IEEE
Transactions on Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3

developed, as shown in Fig. 3. For more accuracy, the parasitic

circuit parameters of the CM choke, sensing winding, and
injection transformers are also modeled and included as Rcm,
Ccm, Rsen, Csen, Rinj1, Cinj1, Rinj2, and Cinj2, respectively. The CM
noise source of the equipment under test (EUT) is modeled as
the Thevenin equivalent circuit, Vn and Zn, which represent the
CM noise source voltage and impedance, respectively. ZLISN
represents the impedance of the line impedance stabilization
network (LISN). Zline, Zin,AEF, and ZY,eff represent the
impedances looking into each direction with regard to the ac
zero potential, respectively.
In the following subsections, the operation principle of the
Fig. 4. Equivalent circuit model of the CM choke at the power line.
AEF is analyzed based on the expressions of the effective
inductance of the CM choke and the effective capacitance of The effect of the sensing winding on the CM choke
the Y-capacitor branch. inductance is summarized in Fig.4. The right gray box
represents the equivalent circuit model of the CM choke at the
power line with considering the voltage induced by 2Msen. The
induced voltage of 2sMsenIsen has the opposite polarity to the
voltage drop of s(Lcm+Mcm)Icm. By defining X(s) as
(2MsenIsen)/(LcmIcm), the total voltage across the choke
inductance can be simplified as s(1+kcm-X(s))LcmIcm. Thus, the
effective inductance of the CM choke, Lcm,eff, is expressed as
(1+kcm-X(s))Lcm, as given in (2).
In the case without any sensing winding, ksen=X(s)=0; thus
Lcm,eff is simply given as (1+ kcm)Lcm, as expected. However,
when sNsen2Lcm in (3) is much higher than (Zin,AEF||Zsen,para),
Fig. 3. Equivalent half-portion circuit model including parasitic components.
X(s)≈2ksen2 and Lcm,eff ≈Lcm(1+ kcm-2ksen2)≈0 because both kcm
and ksen are very close to 1. This implies that the choke
A. Effective Inductance of CM choke inductance can be critically affected by the current flowing at
The impedance looking into the power line in front of the the sensing winding. Therefore, the turn number of the sensing
CM choke, Zline, is solved using Kirchhoff’s laws as winding, Nsen, should be limited to maintain the choke
1 inductance.
Z line  (sLcm,eff ) || Rcm ||  Z LISN (1)
sC cm B. Effective Capacitance of Y-capacitors
where The impedance looking into the Y-capacitor branch, ZY,eff, is
Lcm,eff (s)  1  kcm  X (s)Lcm (2) also solved as
2M sen I sen 2
2sksen 2
Nsen Lcm 1 1   ( s)
X ( s)   2 (3) Z Y , eff (s)   (6)
Lcm Icm sNsen Lcm  Zin, AEF || Z sen, para sC Y 1   (s)
1 where
Z sen, para  Rsen || (4)
sC sen  ( s)  skinj Ninj Linj G1 ( s)Gamp ( s)
1 1 1 1
Z in , AEF  R f  . (5) ( Rd 2 
)( Rinj1 || )( Rinj 2 || ) (7)
sCd sCinj1 sCinj 2
sC f 
1 1 1 1
Lcm,eff represents the effective inductance of the CM choke, {Rd1  Rd 2   Rinj 2 || }( Ro   Rinj1 || ) A( s)
sCd sCinj 2 sCo sCinj1
where the inductance cancelling term, X(s), appears. X(s) is
defined as (2MsenIsen)/(LcmIcm), where Icm and Isen represent the 1 1
 ( s )  sCY ( Rd 1  Rinj 2 || ) || ( Rd 2  )
current flowing through the inductance branches of (Lcm+Mcm) sCinj 2 sCd
and 2Nsen2Lcm in Fig. 3, respectively. Zsen,para represents the 1 2 1
( Rinj 2 || ) Rd 2 
impedance of the parasitic R-C components of the sensing sCinj 2 sCd (8)
winding. Zin,AEF denotes the impedance looking into the [1  
1 1 1
low-pass filter at the amplifier input. The OP-amp input ( Rd 1  Rinj 2 || ) A( s ) Rd 1  Rd 2   Rinj 2 ||
sCinj 2 sCd sCinj 2
impedance is assumed to be large in the interested frequency
1 2
range and ignored in (5). ( Rinj1 || )
sCinj1 1
{  Rinj1 ||  sLinj }]
1 1 sCinj1
Ro   Rinj1 ||
sCo sCinj1

0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2845467, IEEE
Transactions on Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4

Vin,amp 2sksen N sen Lcm Z sen, para boosted to (1+ NsenNinj(1+R2/R1))CY, as shown in Fig 4(c). As
G1 ( s)   the frequency increases near fop,max, (s) starts to decreases,
Vin sN sen 2 Lcm  Z sen, para || Zin, AEF
representing that the compensation voltage from the AEF
1 1 (9) decreases. At the same time, (s) become comparable with 1,
Rcm ||  Zline  Z LISN
sC f sCcm which implies that the impedance of the bypass branch,

1 1 (Rd2+1/sCd), becomes lower than that of the injection
Rf   Z sen, para ( Rcm || ) Zline
sC f sCcm transformer path. Therefore, the CM noise current mainly flows
through the bypass branch, and the impedance of ZY,eff is
1
R2  Rc || approximated to (1/sCY+2(Rd2+1/sCd)).
Vout ,amp sC c Fig. 5(c) shows that the magnitude of ZY,eff increases larger
Gamp ( s )   1 (10)
Vin ,amp R1 than that of 1/sCY after fop,max, because Cd and Rd2 are added to
the current path at the frequency range. However, the damping
1 2
( Rinj1 || ) resistor Rd2 plays an important positive role in mitigating the
sCinj1 1
A(s)  {  Rinj1 ||  sLinj } resonance between the Y-capacitors and CM noise source
1 1 sCinj1 (11)
Ro   Rinj1 || impedance. The critical performance degradation of the overall
sCo sCinj1
CM filter is often caused by the resonance and should be
1 1 avoided, which will be demonstrated in detail in Section IV.
 {sNinj 2 Linj  ( Rd1  Rd 2  ) || Rinj 2 || }  ( skinj Ninj Linj )2
sCd sCinj 2
where  (s) and  (s) can be physically understood as a boosting
factor and a bypassing factor, respectively, as will be explained
later.
G1(s) is the voltage gain from Vin to Vin,amp, while Gamp(s) is
the amplifier gain from Vin,amp to Vout,amp. The frequency
bandwidth of the OP-amp is assumed to be sufficiently higher
than the interested frequency range. The expression of ZY,eff in
(6) can be understood as the impedance of an effective
capacitance, CY,eff(s), which is defined as
(a)
1   (s)
CY ,eff ( s )  CY . (12)
1   (s)
The change of CY,eff(s) with varying the frequency is
summarized in Fig. 5. The effects of the AEF and resultant
current paths at the Y-capacitor branch are described in Fig.
5(a). The gray dotted box represents the effects of the AEF on
the CY,eff. fop,min and fop,max are the target minimum and maximum
operating frequency of the AEF, respectively, which can be
designed by the circuit parameters of the AEF.
As an example, the magnitudes of (s) and (s) of a properly
(b)
designed AEF are plotted with varying frequency in Fig. 5(b).
The impedance of the Y-capacitor branch, ZY,eff, is also plotted
in Fig. 5(c). At frequencies sufficiently lower than fop,min, both
(s) and (s) are much smaller than 1, and ZY,eff in (6) is simply
approximated to 1/sCY. This implies that the bypass circuits and
injection transformer are negligible compared with the
impedance of CY, and the compensated noise voltage from the
AEF is also very small. In the operating frequency range of the
AEF from fop,min to fop,max, the (s) become larger than 1, while
the  (s) is still much smaller than 1. That is, the AEF provides
the compensation voltage to the Y-capacitor branch, as (c)
depicted as -Vin in Fig. 5(a), while the bypass circuits are still Fig. 5. (a) Operation of the AEF depending on frequency range, (b) plot of the
negligible. The magnitude of (s) is predominantly maintained (s) and the (s), and (c) impedance curve of ZY,eff.
at NsenNinj(1+R2/R1) in the operation frequency range, which is C. Insertion Loss of the Total Filter
the multiplication of voltage gains from the sensing winding, The noise attenuation performance of a filter is commonly
amplifier, and injection transformer. Therefore, the CM current quantified by the insertion loss (IL), which is defined as a ratio
flowing through the Y-capacitor branch is amplified by of the noise voltage received at LISN without the filter to that
(1+NsenNinj(1+R2/R1)), resulting that the effective capacitance is with the filter installed. The IL of the total EMI filter in Fig. 3 is

0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2845467, IEEE
Transactions on Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 5

derived as Csen as
Zline || ZY ,eff  2Z n Zline f r , s en 
1 . (17)
IL  . (13)
2N s en LcmC s en
Zline || ZY ,eff 2Z n  Z LISN
As the frequency increases, the IL of (13) starts to increase Next, because the fr,sen should be higher than fr,cm as
predominantly from the frequency point where the ZY,eff  1   1 
f  f , (18)
r , s en   r ,cm  
becomes smaller than Zline. The low-frequency boundary of the  
2N s en LcmC s en   2 Lcm,eff C cm
 
filter operation can be approximated to 1 / 2 Lcm,eff CY ,eff . The a design guideline on the turn number of sensing winding is
proposed AEF greatly decreases the ZY,eff in the target extracted as
frequency range, as shown in Fig. 5(c), which increases the IL Lcm,eff Ccm 2Ccm ,
of the total filter. Also, the AEF makes the total filter start to N sen   (19)
LcmC sen Csen
operate from a lower frequency.
where Lcm,eff is approximated to (1+kcm)Lcm ≈ 2Lcm. The design
III. DESIGN GUIDELINES OF THE PROPOSED AEF guideline of (19), which guarantees the maintained Zline, is
derived as the maximally allowable turn ratio of the sensing
In this section, the practical design guidelines for the winding. Although the exact values of the parasitic capacitance
proposed AEF are developed with consideration of in the CM chokes, Ccm and Csen in (19), are actually unknown
performance and stability. prior to the design, (19) can still provide useful guideline on the
A. Design of the Sensing Winding and Input Low-Pass Filter turn number of the sensing winding.
The sensing winding is directly wound on the CM choke
herein, which is different from [18] and [19], where a separate
additional sensing transformer was utilized. Not using a
separate sensing transformer is desired in terms of size and
cost; however, the CM choke inductance and power line
impedance, Lcm,eff and Zline, can be decreased by the additional
sensing winding, as aforementioned. Even if the proposed AEF
well increases the Y-capacitance, CY,eff, the decreased power
line impedance, Zline, can cause the degradation of the noise
attenuation performance of the total CM EMI filter. Therefore,
a proper design guideline for the sensing winding is required to
prevent the decrease of Zline.
Zline is predominantly determined by the parasitic
capacitance after the first self-resonance frequency of the CM
(a)
choke, fr,cm, which is given as 1/ 2 Lcm,eff Ccm . Thus, after the
frequency fr,cm, the inductance cancelling term X(s) has a minor
effect on the magnitude of Zline. Also, the X(s) in (3) becomes
significant only if sNsen2Lcm is higher than (Zin,AEF||Zsen,para).
Therefore, if the frequency point where sNsen2Lcm starts to be
larger than (Zin,AEF||Zsen,para) is higher than fr,cm, the Zline is rarely
affected by the sensing winding. This condition can be
conveniently satisfied by the following design procedure. First,
Zin,AEF is designed to be higher than Zsen,para for all interested
frequency ranges, such as 10kHz to 30MHz. That is,
   
 Z in , AEF  R f  1    Z sen, para  Rsen || 1  . (14)
 sC   sC 
 f   sen 

More simply, Rf and Cf are chosen to be

(b)
1
Rf  , (15) Fig. 6. Impedance curves of Zline: (a) when Nsen violates (19) and (b) when Nsen
satisfies (19)
sC sen
C f  C sen. (16) Numerical examples are demonstrated in Fig. 6. Two
If the inequality condition of (14) sufficiently holds, different values of Nsen are tested with a designed AEF, where
(Zin,AEF||Zsen,para) ≈ Zsen,para. Then, the frequency point where both Ccm and Csen are set as the same fixed value for simple
sNsen2Lcm starts to be larger than (Zin,AEF||Zsen,para) is analysis. The magnitudes of Zline without and with the AEF,
approximated to the resonance frequency between Nsen2Lcm and Zline,w/o AEF , and Zline,w/ AEF are plotted and compared. In Fig. 6(a),
the condition (19) is violated as Nsen=2, and the fr,sen is lower

0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2845467, IEEE
Transactions on Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 6

than fr,cm. Oppositely, in Fig. 6(b), (19) is satisfied as Nsen=0.5, 1

and the fr,sen is higher than fr,cm. As a result, it is clearly shown f op,max  . (23)
2 2
that Zline,w/AEF in Fig. 6 (a) is significantly decreased compared 2 (1  k inj ) N inj Linj C d
with Zline,w/oAEF, whereas that in Fig. 6 (b) is hardly changed. By substituting (23) into (22), the relation between Cd and
It is also shown that another resonance is created at a high Cinj2 is extracted as
frequency in the Zline after employing the AEF. The resonance C d  Cinj 2 . (24)
occurs due to the sensing winding at the frequency of
Some values of damping resistors, Rd1 and Rd2, are also
1 / 2 (1  k sen ) N sen LcmCsen , as indicated in Fig. 6(a) and (b). This
2 2
required for stability at high frequencies. A few tens of ohm for
resonance provides an adverse effect on the system feedback Rd2 is recommended to mitigate the resonance between the
stability at the high-frequency range, so the low-pass filter Y-capacitors and CM noise source impedance at
consisting of Rf and Cf is necessary at the OP-amp input to high-frequency range, which will be demonstrated by
suppress the resonance. For the low-pass filter not to affect the experiments in Section IV.
AEF performance in the operating frequency range, the cut-off Assuming the effects of the resistors Rd1, Rd2, and Rinj2 are
frequency of the filter should be larger than the maximum negligible in the AEF operation except for the resonances, the
operation frequency, fop,max, but smaller than the resonance condition of (24) allows the expressions (7)–(11) to be
frequency as approximated to
f op,max 
1

1 . (20)  (s)  N senN inj Gamp (s) ,  (s)  0 , (25)
2R f C f 2 (1  k 2 ) N 2 L C in the frequency range until fop,max. The effective capacitance of
sen sen cm sen
The expressions (15), (16), and (20) are the guidelines for the the Y-cap, CY,eff (s), in (12) is then simplified to
low-pass filter design. CY ,eff (s)  (1  NsenNinj Gamp (s))CY . (26)

B. Design of the Injection Transformer and Amplifier The phase compensation elements, Rc and Cc, in Gamp(s)
should have little effects on the AEF operation, and (26) is
The design of injection transformer and amplifier parts is
further simplified to a frequency-independent value, as also
predominantly responsible for the major performance
denoted in Fig. 5, as
parameters of the AEF in Fig. 5: fop,min, fop,max, and CY,eff .
CY ,eff  (1  N senNinj (1  R2 / R1 ))CY . (27)
The capacitor Co at the amplifier output is utilized to block
unwanted signals at a frequency below the target operating Finally, several useful design guidelines for the AEF can be
frequency range. The Co in series with Linj constitute a derived as the following. Even though Nsen is restricted by (19),
high-pass filter, and its cut-off frequency determines the the CY,eff in (27) can be designed to be the multiple times of CY
minimum operating frequency of the AEF, which is derived as by increasing the Ninj and the amplifier gain, (1+R2/R1). As
1 increasing Ninj, however, the maximum operating frequency
f op,min  . (21) fop,max is decreased by (23). Also, a high amplifier gain requires
2 Linj C o
a large output voltage swing and a large gain-bandwidth for the
At the frequency of fop,min, the impedance of Co in series with OP-amp. Therefore, the proper values of Ninj and (1+R2/R1)
Linj critically decreases, which increases the output current of should be selected with consideration for the cost of OP-amp
OP-amp. Therefore, Ro is also added at the OP-amp output to and fop,max of the AEF.
limit the impedance at the resonance frequency but should be Also, the condition of (22) implies that the fop,max of the AEF
sufficiently smaller than sLinj at all the operating frequency can be adjusted by Ninj, Linj, and Cinj2 of the injection
ranges. transformer. Because the parasitic capacitance Cinj2 is not an
On the other hand, as explained in Fig. 5(a), the maximum independent design parameter, Ninj and Linj should be designed
operating frequency of the AEF, fop,max, is determined by the to be small to achieve a high fop,max. However, a small Ninj
frequency boundary where the impedance of the bypass branch results in a small CY,eff, and lowering Linj increases the fop,min
becomes lower than that of the injection transformer path. from (21). Consequently, for the optimized performance of the
Similarly to the resonance due to the sensing winding shown in AEF, the following design process is suggested. First, Co is
Fig. 6, the resonance at the secondary winding of the injection designed to be as large as possible in the given size of physical
transformer also can cause the feedback instability, and the package, and Linj is lowered up to the limit of (21) for the target
bypass branch should start to work at a frequency lower than fop,min. Next, in order to achieve the largest CY,eff, Ninj is
the resonance frequency. The resonance at the secondary increased up to the limit of (22) for the target fop,max.
winding is created at the frequency of 1 / 2 (1  k inj 2 ) N inj 2 Linj C inj 2 , C. Stability Check
which should then be higher than fop,max, as The AEF is basically a feedback system with an analog-input
1 . (22) and an analog-output, and its stability should be carefully
f 
op , max
2 2
2 (1  k inj ) N inj Linj C inj 2 designed and guaranteed. If the system is unstable, the system
can oscillate even when the EUT noise source is not applied.
The fop,max is determined by the resonance between the The feedback stability can be confirmed by the phase and gain
inductance part of the injection transformer, (1-kinj2)Ninj2Linj, margins of the loop gain. To derive the loop gain from the
and the capacitance in the bypass branch, Cd, as

0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2845467, IEEE
Transactions on Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 7

circuit model in Fig. 3, the feedback loop is disconnected at the

1  Rc 
OP-amp output, and a test voltage source Vt is applied to the Cc  Lcm,eff CY ,eff 1   (32)
injection transformer from the disconnected node, while the Rc  R1  R2 

noise source voltage Vn is not applied. The ratio of the voltage The expressions of (31) and (32) deliver the design
at the node in front of CM choke, Vin, to the test voltage, Vt, can guidelines for the Rc and Cc.
then be calculated as It is worth noting that the G2(s) of (28) depends on the noise
1 1 source impedance of the EUT, Zn, so does the loop gain of (30).
skinj Ninj Linj ( Rinj1 ||
)( Rinj 2 || )
V sCinj1 sCinj 2 From the expressions, it is found that a larger magnitude of Zn
G2 (s)  in 
Vt ( R  1  R || 1 ){sL  ( R  1 ) || R || 1 } tends to result in a smaller gain margin by increasing the
o
sCo
inj1
sCinj1
inj o
sCo
inj1
sCinj1 (28) magnitude of the loop gain. Thus, the infinite value of Zn
1
usually gives rise to the worst stability condition in the AEF
Rd 2  applications, which has been validated by simulations and
sCd
 measurements. Therefore, in this paper, the loop gain of the
1 1 1 1
(  )( Rd1  Rd 2   Rinj 2 || ) B( s ) designed AEF is calculated or measured with the condition of
Z n Zline sCd sCinj 2
the infinite Zn in order to guarantee the stability for any EUT
where applications.
1 1 1 1 1
B( s)  {(  )  ZCy  2( Rd 2  ) || ( Rd1  Rinj 2 || )}
Z n Zline sCd sCinj 2

1 1 ( sM inj )2
{sNinj 2 Linj  ( Rinj 2 || ) || ( Rd1  Rd 2  ) }
sCinj 2 sCd sL  ( R  1 ) || ( R || 1 )
inj o inj1
sCo sCinj1
1 2 1 2
2( Rinj 2 || ) ( Rd 2  )
sCinj 2 sCd

1 1 2
( Rd1  Rinj 2 ||  Rd 2  )
sCinj 2 sCd
(29)
The voltage gains of (Vin,amp/Vin) and (Vout,amp/Vin,amp) are
already derived as G1(s) and Gamp(s) in (9) and (10),
respectively. Therefore, the loop gain of the system is written as (a)
Vout,amp Vin Vin,amp Vout,amp
Gloop (s)     G1 (s)G2 (s)Gamp (s) . (30)
Vt Vt Vin Vin,amp
The purpose of employing the Rc and Cc is to increase the
phase margin of Gloop(s) for stability at the low-frequency range
because the resonance between the effective inductance of the
choke, Lcm,eff, and the effective capacitance of the Y-cap branch,
CY,eff, also cause a risk of instability. The resonance frequency
of 1 / 2 Lcm,eff CY ,eff determines the low-frequency boundary of
the filter operation, and should be lower than the low-frequency
limit of the CE regulation in a proper EMI filter design. The
maximum amount of the phase compensation due to the Rc and
Cc is calculated as (b)
Fig. 7. Comparison of loop gain: (a) an unstable situation without damping
max(| Gloop, w/ comp ( f )  Gloop, w/o comp ( f ) |) components, Rd1, Cd, Rd2, and phase compensator, Rc, Cc, (b) a stable situation
with those components.
 Rc 2  , (31)
 tan 1  
 4( R1  R2 )( R1  R2  Rc )  As an example, the Gloop(s) of a designed filter with the AEF
 
is plotted using (30), as shown in Fig. 7. The bypass branch and
which arises at the frequency of 1 Rc . the phase compensator, Rd1, Cd, Rd2, Rc, Cc, are not installed in
1
2Rc C c R1  R2 Fig. 7(a) but are installed in Fig. 7(b). Their effects on the
Gloop, w/ comp (s) and Gloop, w/o comp (s) represent the phases stability is clearly demonstrated. The instability due to a drastic
phase shift near 10MHz in Fig. 7(a), which is attributed to the
of Gloop(s) with and without the phase compensator,
secondary winding of injection transformer, is relieved by the
respectively. By setting the maximum phase-compensation
bypass branch in the case of Fig. 7(b). The drastic phase shift at
frequency to be the resonance frequency of 1/ 2 Lcm,eff CY ,eff ,
low frequencies below 100 kHz in Fig. 7(a), which is attributed
another expression on the Rc and Cc is extracted as to the resonance between Lcm,eff and CY,eff, also causes the risk of
instability. As shown in Fig. 7(b), the gain margin is greatly
increased by using the phase compensator, Rc and Cc.

0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2845467, IEEE
Transactions on Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 8

D. OP-amp Selection and Overall Design Procedure

The operating high-frequency limit of the OP-amp in the
non-inverting amplifier, fOPamp, should be higher than the
high-frequency limit of the CE regulation, fCE,max, as
fOPamp  fCE,max (33)
Also, the voltage and current capacity of the OP-amp should
be sufficient for noise compensation. To estimate the required
OP-amp capacity, the voltage and current at the OP-amp output,
Vout,amp(s) and Iout,amp(s), are calculated from the circuit model in
Fig. 3, respectively, as
Vout,amp (s)  Vin (s)G1(s)Gamp (s) (34)
Vin ( s)
I out ,amp ( s )   [G1 ( s )Gamp ( s)
1 1
2(  Ro  sLinj1 || Rinj1 || )
sCo sCinj1
1 1
Rd 1  Rd 2   Rinj1 ||
1 sCd sCinj1
 kinj Ninj ( sLinj1 || Rinj1 || )
sCinj1 1 1
( Rd 2  )( Rinj 2 || )
sCd sCinj 2
1 2 2
 (2 Rd 2  ) || (2 Rd 1  2 Rinj1 || )
sCY sCd sCinj1
 {1  }]
ZY ,eff

(35)
where Fig. 8. Design flow of the proposed AEF.
(Zline  Z LISN ) || ZY ,eff
Vin (s)  Vn (s) . (36)
2Zn  (Zline  Z LISN ) || ZY ,eff Fig. 8 summarizes the design flow for the proposed AEF
Because Vin(s) is determined by both Zn and Vn(s) as well as derived in the previous sections. Assume that the maximum
the filter impedance including the AEF, the information of allowable value of the passive Y-capacitor, CY, is given from
noise source model is necessary to estimate Vout,amp(s) and the safety limit on the earth leakage current, prior to the overall
Iout,amp(s). The noise source model of an operating SMPS, Zn filter design. The required CM choke inductance and effective
and Vn(s), can be extracted by various measurement methods Y-capacitance in the total CM EMI filter, Lcm,eff, and CY,eff,, are
developed in [25]-[28]. Once Zn and Vn(s) are found, the then determined from the extracted CM CE noise model of the
time-domain waveforms of the OP-amp output voltage and EUT, with considering that the overall filter starts to work from
current, vout,amp(t) and iout,amp(t), can be calculated from the the resonance frequency between them. Next, the target
spectra of Vout,amp(s) and Iout,amp(s) given in (34)-(36). Therefore, operation frequency range of the AEF, fop,min, fop,max, are also
the voltage and current capacity of the OP-amp output, decided. From the target performance of the AEF, fop,min, fop,max,
vOPamp,max and iOPamp,max, should be sufficient to provide the and CY,eff, the sensing and input parts of the AEF are designed at
calculated vout,amp(t) and iout,amp(t), respectively, as, Step 1. The turn ratio of sensing winding, Nsen, is designed by


vOPamp,max  max vout ,amp (t )  (37)
the conditions of (19), and the input low-pass filter, Rf and Cf, is
designed by the condition of (15), (16), and (20). In Step 2, the
iOPamp,max  max  2i out ,amp (t )  . (38) amplifier and injection transformer parts are designed. The
values of Co and Linj are chosen by (21) from the target fop,min.
Because Iout,amp(s) is defined in the half-circuit model, the The values of Ninj and (1+R2/R1) are then designed by using (23)
actual current flowing at the OP-amp is twice the calculated and (27) with consideration of the target fop,max and CY,eff. From
current, as shown (38). the designed values of Linj and Ninj, the injection transformer is
Among the various design factors affecting Iout,amp(s), as manufactured, and the Cinj1 and Cinj2 are extracted by
shown in (35), increasing Ninj predominantly increases Iout,amp(s) measurements. The capacitance of bypass branch, Cd, is also
in the operation frequency range of AEF. The voltage gain of determined by (24). If necessary for the low frequency stability,
injection transformer due to Ninj can reduce the budget of the the phase compensator, Rc and Cc, is also added in the amplifier
OP-amp output voltage by trade of output current, as mentioned part based on the design guides of (31) and (32). In Step 3, the
in Section III-B. The injection transformer not only isolates the feedback stability of the overall filter with the AEF is
AEF ground from the SMPS ground, but also provides the gain confirmed by plotting the loop gain using (30). If the stability
and another design flexibility for the OP-amp circuit. margin at any frequency point is not sufficient, the
corresponding circuit components among Rf, Cf, Rd1, Rd2, Cd, Rc,

0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2845467, IEEE
Transactions on Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 9

and Cc are properly tuned. In the final Step 4, a proper OP-amp to be 470pF with considering the safety standard of the
is selected based on the required frequency-bandwidth, voltage application. However, the insertion loss of the filter without the
and current capacity for the noise compensation by using AEF is not sufficient at the low-frequency range below 1MHz.
(33)-(38). Also, another problem in the designed filter is that the inductive
component in the noise source model and the Y-capacitance in
IV. IMPLEMENTATION AND MEASUREMENTS OF THE the filter comprise a resonance circuit, and the insertion loss of
PROPOSED AEF the filter is significantly reduced at the resonant frequency of
In this section, the proposed AEF is designed in a compact about 4.2MHz. The resonance effects between the source
package for the application to a SMPS board, and its noise impedance and a component in the EMI filter often arise, which
reduction performances are demonstrated by various can critically degrade the filter performance in a narrow
measurements. frequency band [29], [32]. In order to increase the insertion loss
at low-frequency range as well as to mitigate the resonance
A. Implementation of the CM EMI Filter with the Proposed effect, the proposed AEF is designed and installed as an add-on
AEF to the L-C filter instead of increasing the stage of the filter.
The target SMPS board includes various power circuits, such It is desired to enhance the effective Y-capacitance by more
as the rectifier, power factor correction (PFC) converter, than seven times to sufficiently suppress the low-frequency CM
flyback converter, and LLC resonance converter. The SMPS noises, and the target CY,eff of the AEF is selected as 3.4nF. The
has the power rating of 200W with 64kHz and 110kHz target fop,min and fop,max are then set as about 2.4kHz and 1MHz,
switching frequencies. To efficiently design and analyze the respectively. Also, the values of Rd1 and Rd2 in the AEF are
total EMI filter including the AEF, the information of noise designed to sufficiently mitigate the resonance between the
source impedance of the EUT is required because the proper source impedance and the Y-capacitor branch.
impedance mismatching at the CE frequency range should be
obtained by the filter [29]. As a simple method, the amounts of TABLE I
CE noises from the SMPS are measured without and with a COMPONENT VALUES OF THE PROPOSED ACTIVE EMI FILTER
reference CM choke installed at the power lines, respectively.
Components Values
The CM noise source impedance of the SMPS can then be
calculated from the choke impedance and the insertion loss of Lcm, Rcm, Ccm, kcm 13.5mH, 112k5 pF, 0.997
the choke [25]. Nsen, Rsen, Csen, ksen 0.3, 80k20 pF, 0.997
Linj, Rinj1, Cinj1, kinj 45H, 10k5 pF, 0.994
Ninj, Rinj2, Cinj2 6, 112k6 pF
R1, R2, Rc, Cc, Rf, Cf 4001k6.8k5.6nF, 1k10pF
CY, Ro, Co, Rd1, Rd2, Cd 470pF, 0.2294F, 220301.5nF

Fig. 9. Extracted CM noise impedance of the target SMPS and impedance curve
of the series C-L circuit model.

Fig. 9 shows the magnitude of the extracted CM noise source

impedance of the target SMPS, |2Zn|, which represents the
impedance of the half-portion circuit model, as shown in Fig. 3. Fig. 10. Photograph of the proposed AEF installed at the CM L-C filter.
The impedance curve can be approximately modeled as a series
C-L circuit of 500pF-capacitance and 6H-inductance. The Fig. 10 shows the proposed AEF installed at the CM L-C
capacitive component is attributed to the parasitic capacitance filter in the target SMPS board. The proposed AEF has been
of heat sinks, winding of transformers, and internal capacitors, designed according to the design flow of Fig. 8 and
while the inductive component is predominantly due to the implemented in a compact package with the size of 22 mm  20
leakage inductance of transformers and the PFC inductor in the mm. The wires in the sensing winding and the injection
target SMPS board [30], [31]. Prior to the design of the AEF, a transformer can be thin because they carry only small noise
conventional CM L-C filter consisting of a commercial CM signals and are free from heat problems. Especially, the sensing
choke and Y-capacitors has been designed for the target SMPS winding at CM choke was realized in a small area on the
board. The inductance of the choke in the designed filter is already existing commercial CM choke. The circuit model
13.5mH at 100kHz, and the value of a Y-capacitance is chosen parameters for the CM choke with sensing winding and

0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2845467, IEEE
Transactions on Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 10

injection transformer have been extracted using the VNA, as product application. In the experiments herein, the supply
explained in [17]. All circuit parameters of the designed AEF voltages for the AEF have been provided by an external power
are summarized in Table I. supply to verify that the AEF works well at any reference
As explained in Section III-D, the information of both Vn and voltage, which can be far different from the earth ground.
Zn is necessary to estimate the required voltage and current
B. Experimental Validations
capacity of the OP-amp. In addition to the |Zn| in Fig. 9, the
magnitude of the CM noise source voltage, |Vn(s)|, can be also
extracted using the same measurement method using a
spectrum analyzer and a reference CM choke [26]. The
magnitude of the spectra of Vout,amp(s) and Iout,amp(s) are then
calculated from (34)-(36) with the extracted |Zn| and |Vn(s)|.
Although the phases of the voltage and current spectra are still (a)
unknown, the worst-case time-domain waveforms can be
approximately estimated by considering all the frequency
spectra to be in-phase. The actual voltage and current
waveforms at the OP-amp output should be different from the
worst-case waveforms, however; the peak and RMS values of
the worst-case waveforms can provide an initial guideline to
identify the required OP-amp capacity. After the calculation,
the maximum absolute peaks of the worst-case waveforms
were obtained as 4.7V and 420mA, respectively. Accordingly,
a commercial OP-amp (OPA2677) with the maximum
available output voltage and current of 5V and 500mA at 6V
dc supply voltages was employed in the amplifier part. After
implementing the AEF at the target SMPS, the actual voltage
and current waveforms at the OP-amp output were also (b)
measured by using a differential probe and a current probe, Fig. 12. (a) VNA port condition for the measurement of loop gain (b) calculated
respectively, as shown in Fig. 11. The dotted horizontal lines versus measured results.
represent the maximum values of the calculated worst-case Prior to the CE measurements with the operating SMPS, the
waveforms. The peak values of the actual waveforms are a bit characteristics and modeling expressions of the AEF are
smaller than those of the worst-case waveforms, which rigorously validated by measurement using the VNA. The loop
validates the worst-case estimation method for the OP-amp gain of the total filter with the AEF, Gloop(s), can be directly
selection. When the injection transformer is not employed, a measured by disconnecting the OP-amp output node and
much larger maximum output voltage of the OP-amp is applying the VNA ports, as shown in Fig. 12 (a). The Gloop(s) is
necessary, which requires a more expensive OP-amp and a then calculated as
higher DC power supply. By virtue of the injection transformer
V Z
gain, a lower-voltage OP-amp can be used in the AEF by trade Gloop (s)   2   21 (39)
of OP-amp current. V1 Z11
where Zij is the measured impedance parameters in the test
configuration. The magnitude and phase of Gloop(s) extracted
from the measurements by (39) are plotted in Fig. 12 (b), which
show good agreement with the analytical results calculated by
(30). After the careful design, according to the extracted
guidelines, the gain and phase margins are sufficient at both
low and high-frequency ranges.

Fig. 11. Measured OP-amp output voltage and current waveform. (a)

The supply voltage for the OP-amp can be provided by

converting a SMPS DC output [9], [17]. A DC voltage is
usually available from the SMPS or control board in a real

0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2845467, IEEE
Transactions on Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 11

frequency correlate well with the IL of each filter in Fig. 13.

The AEF achieves the additional CE reduction of 15dB at
64kHz. Also, in the filter without the AEF, the critical noise
peaks occur at a few MHz due to the poor IL at the resonance
frequencies, which is damped by 7dB after using the AEF.

(b)
Fig. 13. (a) VNA port condition for the measurement of insertion loss and (b)
calculated versus measured results

Also, the IL of the total filter can be directly extracted from

the VNA measurements using the port condition shown in Fig.
13 (a), as Fig. 14. Measurement results of the CM conducted emissions.
Z LISN / 2   Z n  Z11  Z 22  Z12   (40)
IL(s)   1    
Z n  Z LISN / 2 
 Z 21  Z LISN / 2  Z LISN / 2 

which is derived with including the effect of source impedance,
Zn. The IL extracted from the measurement by (40) is plotted in
Fig. 13(b), which shows good agreement with the analytical
result calculated by (13). Also, the ILs of the filter with the AEF
are compared with those without the AEF. In any case, the Zn is
modeled as a series circuit of 500pF-capacitance and
6H-inductance in both plots, as extracted in Fig. 9.
In the filter without the AEF, the IL critically decreases
below 0dB at 31kHz due to the resonance between the CM
choke and Y-capacitor, which determines the low-frequency
boundary of the filter operation, as aforementioned. Also,
another significant degradation of the IL occurs at 4.2MHz due Fig. 15. Measured waveform and RMS values of the leakage currents.
to the resonance between the noise source impedance and
Y-capacitors. In the design of a conventional CM filter, the CE noises at
After applying the AEF, the low-frequency resonance moves low frequencies are the main culprits of increasing the filter
to a much lower frequency by the increased CY,eff, and the IL size, weight, and cost. At high frequencies, the noise peaks
degradation at the resonance is also much mitigated. The AEF caused by unwanted resonances are often the most troublesome.
achieves the additional IL by 10 to 20 dB at a frequency range Thus, the weak points of the conventional CM filter in both low
lower than 1MHz. The additional IL at 64kHz, which is the and high frequencies are effectively improved by employing
lowest switching frequency of the target SMPS, is 15dB. Also, the AEF with the size smaller than a half of a CM choke, and
the IL degradation due to the resonance at 4.2MHz is mitigated the CE level is managed under 57dB V in all the frequency
by the damping resistors implemented in the AEF. Meanwhile, ranges of the CE regulation.
the IL at the frequency range higher than 5MHz is a bit reduced Leakage currents flowing to earth ground through the
compared with the filter without the AEF. In the AEF design Y-capacitors were also measured to verify the safety of using
for the target application, however, the reduction of CE noises the proposed AEF. The leakage current measurement setup of a
at the low-frequency range and the resonance frequencies are single-phase power system was used, which is defined in the
most required. IEC60990 [33]. The measuring network of ‘let-go’ was used
Next, the CM CE noises from the target SMPS are measured herein because the ‘let-go’ measurement condition imposes the
from 10kHz to 30MHz using the CE test standard setup. The worst leakage current for the AEF. Fig. 15 shows the measured
AEF consumes 0.31W root-mean-square (RMS) power when waveforms and RMS values of the leakage current for the case
applied to the SMPS. In addition to the bare SMPS noise with and without the AEF, respectively. The leakage current is
without any filter, the noises after employing the filters with predominantly caused by the ac utility voltage with 220V and
and without the AEF, respectively, have been tested, as shown 60Hz on a main power line. Even though the effective
in Fig. 14. The amounts of noise reduction depending on the Y-capacitance for the CE noises is greatly increased by the

0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2845467, IEEE
Transactions on Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 12

AEF, the leakage current is rarely affected. It is attributed to the reduction of the CM CE noises by applying the AEF has been
measuring network having low-pass filter characteristics with also experimentally validated. Leakage currents were also
the cut-off frequency below 1kHz, while the AEF does not measured, and it was verified that the proposed AEF hardly
change the Y-capacitance at the low-frequency. increase the leakage current through Y-capacitors.
The AEF achieves the additional CE reduction by 10 to 20
C. Comparison between Transformer-Isolated AEFs
dB at the frequency range lower than 1MHz. The critical noise
The performance and characteristics of the proposed AEF peaks at the resonance frequencies are also reduced by 7dB. As
are compared with those of other transformer-isolated AEFs a result, the weak points of the conventional CM filter in both
published in previous literatures, as summarized in Table II. low and high frequencies are effectively improved by
The AEFs in [18] and [19] are of the CSVC topology and employing the AEF, and the CE level is managed under
require two transformers installed on the main power lines. The 57dBV in all the frequency range of the CE regulation.
AEF in [18] have a maximum IL of 20dB but poor performance
in the low-frequency range under 500kHz. The AEF in [19] REFERENCES
have good performance in the low-frequency range from
[1] K. Mainali, R. Oruganti., "Conducted EMI mitigation techniques for
150kHz but have relatively low performance at the switch-mode power converters: a survey," IEEE Trans. Power Electron.,
high-frequency range over 2MHz. Both AEFs have drawbacks vol.25, no.9, pp.2344-2356, Sept. 2010
in size and performance due to the transformers at power lines, [2] L. Tihanyi, Electromagnetic Compatibility in Power Electronics. New
York: IEEE Press, 1995.
as aforementioned. The proposed AEF has good performance [3] O. M. Timothy, P. L. Richard, EMI filter Design, 3rd ed, CRC Press,
in the very low-frequency range from a few tens of kHz. The 2011.
injection transformer is the smallest and rarely affected by [4] W. Chen, X. Yang, and Z. Wang, "An active EMI filtering technique for
improving passive filter low-frequency performance," IEEE Trans.
power rating of EUT because it is installed on the small signal Electromagn. Compat., vol. 48, no. 1, pp. 172-177, Feb. 2006.
path. Also, for the sensing winding, thin wire is directly wound [5] W. J. Chen; X. Yang; Z. Wang, "A novel hybrid common-mode EMI
on the existing commercial CM choke instead of using the filter with active impedance multiplication," IEEE Trans. Ind. Electron.,
vol.58, no.5, pp.1826-1834, May 2011
additional current transformer (CT) on power lines. Therefore, [6] M. C. Di Piazza, G. Tinè, and G. Vitale, “An improved active
the proposed AEF can be the most easily embedded inside a common-mode voltage compensation device for induction motor drives,”
real product without increasing the size of filter, regardless of IEEE Trans. Ind. Electron., vol. 55, no. 4, pp. 1823–1834, Apr. 2008.
the power rating of the EUT. [7] M. C. Di Piazza, A. Ragusa, and G. Vitale, “Design of grid-side
electromagnetic interference filters in AC motor drives with motor-side
common mode active compensation,” IEEE Trans. Electromagn.
V. CONCLUSION Compat., vol. 51, pt. 2, no. 3, pp. 673–682, Aug. 2009.
[8] Y. Son, S. Sul, "A new active common-mode EMI filter for PWM
A new structure of transformer-isolated AEF without inverter,"IEEE Trans. Power Electron., vol.18, no.6, pp.1309-1314, Nov.
additional components on power lines has been proposed. The 2003
proposed AEF has been rigorously analyzed to estimate the [9] D. Shin, S. Jeong and J. Kim, "Quantified Design Guidelines of Compact
Transformerless Active EMI Filter for Performance, Stability, and High
noise attenuation performance and loop gain. The proper design Voltage Immunity," IEEE Trans. Power Electron., vol. PP, no. 99, pp.
guideline has been developed based on the analytical results. 1-1.
The proposed AEF has been designed in a compact package [10] D. Hamza, M. Pahlevaninezhad, and P. Jain, “Implementation of a novel
digital active EMI technique in a DSP-based DC–DC digital controller
and optimized for the target SMPS board with consideration of used in electric vehicle (EV),” IEEE Trans. Power Electron., vol. 28, no.
noise source impedance. The calculated insertion loss and loop 7, pp. 3126–3137, Jul. 2013.
gain have been validated with the VNA measurements. The [11] J. Huang and H. Shi, “A hybrid filter for the suppression of
common-mode voltage and differential-mode harmonics in three-phase

TABLE II
COMPARISON BETWEEN TRANSFORMER-ISOLATED AEFS
Diameter of
Max. IL of
injection
Total Filter Noise sensing method Benefits Drawbacks
transformer
[dB]
[mm]
- Low performance in low frequencies
under 500kHz.
[18] 20 30 Additional CT - Fully transformer-isolated.
- Both CT and injection transformer are
on power lines.
- Low performance in high frequencies
- Fully transformer-isolated.
over 2MHz.
[19] 30 23 Additional CT - High performance in low frequencies from
- Both CT and Injection transformer are
150kHz.
on power lines.
- Fully transformer-isolated.
- High performance in very low frequencies
This Sensing winding on - Low performance in high frequencies
55 16 from a few tens of kHz.
work existing CM choke over 1MHz.
- No additional components on power lines.
- Reduced high-frequency resonances

0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2018.2845467, IEEE
Transactions on Power Electronics
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 13

inverters with CPPM,” IEEE Trans. Ind. Electron., vol. 62, no. 7, pp.
3991–4000, Jul. 2015.
[12] L. LaWhite and M. F. Schlecht, “Design of active ripple filters for power Sangyeong Jeong received the B.S. degree in
circuits operating in the 1–10 MHz range,” IEEE Trans. Power Electron., electrical engineering from the Ulsan National
vol. 3, no. 3, pp. 310–317, Jul. 1988. Institute of Science and Technology, Ulsan, South
[13] N.K. Poon, J. C. P. Liu, C. K. Tse, and M. H. Pong, "Techniques for input Korea, in 2015, where he is currently pursuing the Ph.
ripple current cancellation: classification and implementation," IEEE D. degree. His current research interests include
Trans. Power Electron., vol.15, no.6, pp.1144-1152, Nov 2000 electromagnetic compatibility (EMC) problems in the
[14] Y. Son, S. Sul, "Generalization of active filters for EMI reduction and power system and the noise reduction techniques.
harmonics compensation," IEEE Trans. Ind. Appl., vol.42, no.2,
pp.545-551, March./April 2006
[15] S. Wang, Y. Y. Maillet, F. Wang, D. Boroyevich, and R. Burgos,
“Investigation of hybrid EMI filters for common-mode EMI suppression
in a motor drive system,” IEEE Trans. Power Electron., vol. 25, no. 4, pp.
1034–1045, Apr. 2010.
[16] Y. Chu, S. Wang, and Q. Wang, “Modeling and stability analysis of Dongil Shin received his B.S. degree in electronic
active/hybrid common-mode EMI filters for DC/DC power converters,” engineering from the Kumoh National Institute of
IEEE Trans. Power Electron., vol. 31, no. 9, pp. 6254–6263, Sep. 2016. Technology, Gumi, Korea, in 2012, and the Ph.D.
[17] D. Shin, S. Kim, G. Jeong, et al. “Analysis and design guide of active EMI degree with School of Electrical and Computer
filter in a compact package for reduction of common-mode conducted Engineering, Ulsan National Institute of Science and
emissions,” IEEE Trans. Electromagn. Compat., vol. 57, no. 4, pp. Technology (UNIST), Ulsan, Korea in 2018. He is
6670-671, Aug. 2015. currently working for VC PE Division in LG
[18] W. Chen; W. Zhang; X. Yang; Z. Sheng; Z. Wang, "An experimental Electronics Company, Incheon, South Korea. His
Study of Common- and Differential-Mode Active EMI Filter current research interests include electromagnetic
Compensation Characteristics," IEEE Trans. Electromagn. Compat, compatibility problems in the electric vehicle and the
vol.51, no.3, pp.683-691, Aug. 2009 noise reduction technic.
[19] K. Mainali and R. Oruganti, “Design of a current-sense voltage-feedback
common mode EMI filter for an off-line power converter,” in Proc. IEEE
Power Electron. Spec. Conf., 2008, pp. 1632–1638. Jingook Kim (M’09-SM’15) received his B.S., M.S.,
[20] M. L. Heldwein, H. Ertl; J. Biela, J.W Kolar, "Implementation of a and Ph.D. degrees in electrical engineering from
transformerless common-mode active filter for offline converter Korea Advanced Institute of Science and
systems," IEEE Trans. Ind. Electron., vol.57, no.5, pp.1772-1786, May Technology, Daejon, Korea, in 2000, 2002, and 2006,
2010 respectively. From 2006 to 2008, he was with
[21] S. H. Voldman, "Electrical Overstress (EOS): Challenges for component DRAM design team in Memory Division of
and system-level co-design," 2015 IEEE 11th International Conference Samsung Electronics, Hwasung, Korea, as a senior
on ASIC (ASICON), Chengdu, 2015, pp. 1-4. engineer. From January 2009 to July 2011, he
[22] P. F. Keebler, "Chokeless power-line filters with integral surge worked for the EMC Laboratory at the Missouri
protection," 2011 IEEE International Symposium on Electromagnetic University of Science and Technology, Missouri,
Compatibility, Long Beach, CA, USA, 2011, pp. 184-189. USA, as a postdoc fellow. In July 2011, he joined the
[23] J. Huntington, Show Networks and Control Systems: Formerly Control Ulsan National Institute of Science and Technology (UNIST), Ulsan, Korea,
Systems, Zircon Designs Press, 2012 where he is currently an associate professor. His current research interests
[24] N. Mortensen and G. Venkataramanan, "An Active Common Mode EMI include EMC, EMC filter, power electronics EMC, electrostatic discharge
Filter for Switching Converters," 2008 IEEE Industry Applications (ESD), RF interference.
Society Annual Meeting, Edmonton, Alta., 2008, pp. 1-7.
[25] D. Zhang, D. Y. Chen, M. J. Nave and D. Sable, "Measurement of noise
source impedance of off-line converters," IEEE Trans. Power Electron.,
vol. 15, no. 5, pp. 820-825, Sep 2000.
[26] J. Jeong, S. Kim and C. Kim and J. Kim, "Estimation of the worst-case
conducted emission noise due to multiple parallel power converters," in
Proc. 2016 URSI Asia-Pacific Radio Science Conference (URSI
AP-RASC), Seoul, 2016, pp. 931-934.
[27] J. Meng, W. Ma, Q. Pan, Z. Zhao and L. Zhang, "Noise Source Lumped
Circuit Modeling and Identification for Power Converters," IEEE Trans.
Ind. Electron., vol. 53, no. 6, pp. 1853-1861, Dec. 2006.
[28] K. See and J. Deng, "Measurement of noise source impedance of SMPS
using a two probes approach," IEEE Trans. Power Electron., vol. 19, no.
3, pp. 862-868, May 2004.
[29] F. Luo, D. Dong, D. Boroyevich, P. Mattavelli and S. Wang, "Improving
High-Frequency Performance of an Input Common Mode EMI Filter
Using an Impedance-Mismatching Filter," IEEE Transactions on Power
Electronics., vol. 29, no. 10, pp. 5111-5115, Oct. 1 2014.
[30] J. Lou, Q. Sun, W. You, Y. Shu, A. Bhobe and J. Yu, "Analytical
calculation of conducted EMI in flyback converters," 2016 Asia-Pacific
International Symposium on Electromagnetic Compatibility (APEMC),
Shenzhen, 2016, pp. 395-397.
[31] H. Ye, Z. Yang, J. Dai, C. Yan, X. Xin and J. Ying, "Common mode
noise modeling and analysis of dual boost PFC circuit," INTELEC 2004.
26th Annual International Telecommunications Energy Conference, 2004,
pp. 575-582.
[32] F. Luo, D. Boroyevich and P. Mattavelli, "Improving EMI filter design
with in circuit impedance mismatching," 2012 Twenty-Seventh Annual
IEEE Applied Power Electronics Conference and Exposition (APEC),
Orlando, FL, 2012, pp. 1652-1658.
[33] IEC 60990:1999 ed. 2.0; “International standard –Methods of
measurement of touch current and protective conductor current”, 1999

0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.