Вы находитесь на странице: 1из 15

2776 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO.

12, DECEMBER 2009


Simple Modeling and Identication Procedures for
Black-Box Behavioral Modeling of Power
Converters Based on Transient Response Analysis
Virgilio Valdivia, Student Member, IEEE, Andr es Barrado, Member, IEEE, Antonio L azaro, Member, IEEE,
Pablo Zumel, Member, IEEE, Carmen Raga, Student Member, IEEE, and Cristina Fern andez, Member, IEEE
AbstractToday, black-box behavioral models of power con-
verters are becoming interesting for system level simulation of
power electronics systems. These models can be used to evaluate
the response of systems that are composed of commercial convert-
ers, since they can be fully parameterized by analyzing the actual
converter response. To optimize the required computational re-
sources, these models should be as simple as possible. Furthermore,
the identication of the parameters should be carried out easily,
looking for simple experiments and straightforward adjusting al-
gorithms. Easy modeling and identication procedures, based on a
transient response analysis, are proposed in this paper. Using this
method, a large-signal black-box behavioral model of a power
converter is composed of reduced-order transfer functions, which
are identied by analyzing the step response of the converter. Both
an actual commercial dcdc converter and a line-commutated rec-
tier have been modeled and identied by means of this approach,
in order to validate the proposed procedures.
Index TermsBehavioral modeling, black-box modeling,
reduced-order model, system identication.
I. INTRODUCTION
T
ODAY, power electronics systems with multiple converters
and loads are growing dramatically (e.g., onboard power
electronics-based systems for aircrafts) [1][7]. The system
level analysis is complex [8][10], and simulation tools are
needed to carry it out.
Therefore, models of the power converters oriented to system
level simulation (also known as behavioral models) are needed.
These models should be as simple as possible in order to op-
timize the simulation time. This need was rst discussed two
decades ago [1], and several works have been published about
modeling of converters for system level simulation and their ap-
Manuscript received March 6, 2009; revised June 12, 2009. Current version
published December 28, 2009. This work was supported in part by the Span-
ish Ministry of Education and Science under the Research Project Dise no y
Modelado de Sistemas Electr onicos Aeroespaciales. Nivel Subsistema (Code:
DPI2006-14866-C02-02) and by the private contract with the European Aero-
nautic Defense and Space CompanyConstrucciones Aeron auticas Sociedad
An onima under the Research Project High-Voltage Direct-Current Load Distri-
bution System (Code: 04-AEC0527-000050/2005) nanced by The European
Regional Development Fund via the Aerospace Sector Plan of the Community
of Madrid. This paper was presented in part at the IEEE Applied Power Elec-
tronics Conference 2009. Recommended for publication by Associate Editor J.
Jatskevich.
The authors are with the Power Electronics Systems Group, Electronics Tech-
nology Department, University Carlos III of Madrid, Legan es 28911, Spain
(e-mail: vvaldivi@ing.uc3m.es; barrado@ing.uc3m.es; alazaro@ing.uc3m.es;
pzumel@ing.uc3m.es; mraga@ing.uc3m.es; cfernand@ing.uc3m.es).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TPEL.2009.2030957
plication [1], [11][18]. However, the system designer does not
usually have access to all necessary data to parameterize these
models (e.g., a systemcomposed of commercial converters), so a
black-box approach is required, and identication procedures
of the models are also needed, in order to parameterize them.
A black-box behavioral model of a power converter is a
mathematical model oriented to system level analysis. These
models only compute the needed signals to analyze the response
of the power system. They are dened by a set of elements that
could have no physical interpretation, and they are identied by
analyzing the response of the modeled converter.
First works on behavioral models and their identication,
focused on power electronics, have been recently proposed.
1) References [19][21] propose a model for dcdc convert-
ers composed of one nonlinear static network and two
linear dynamic networks (a WienerHammerstein struc-
ture). The static network models the converter steady-state
behavior in a nonlinear way (efciency, static load and line
regulations, and constant power source behavior). The dy-
namic networks are linear time invariant (LTI), composed
of passive elements (R, L, and C), and model the inrush
current, high-frequency input impedance, and the output
impedance. In addition, models of the protections, start-
up, and other features are also proposed. The model is
entirely parameterized by means of the datasheet data, or
equivalent tests.
2) References [22][25] propose the model shown in Fig. 1.
It is an unterminated two-port-network model composed
of four dynamic systems: the output impedance Z
o
, the
audiosusceptibility G
o
, the back current gain H
i
, and the
input admittance Y
i
. V
ss
is the steady-state output volt-
age. Each dynamic system can be composed of either a
linear model or a nonlinear model, so the dynamic behav-
ior of dcdc converters with nonlinear dynamics can be
accurately modeled by means of this approach. The identi-
cation of each dynamic system is based on the converter
frequency response, which is obtained using a network
analyzer, a linear amplier, and an isolation transformer.
However, the required hardware equipment is relatively
expensive and may not always be available.
3) Other references propose behavioral models related to
power systems. In [26], behavioral models of protection
devices (solid-state power controllers) are described and
in [27], a model of the electromagnetic interference gen-
eration in power converters is proposed.
0885-8993/$26.00 2009 IEEE
VALDIVIA et al.: SIMPLE MODELING AND IDENTIFICATION PROCEDURES FOR BLACK-BOX BEHAVIORAL MODELING 2777
Fig. 1. Black-box behavioral model of a dcdc converter.
These approaches are oriented to model both the input and
the output signals average behavior, like conventional average
models.
In this paper, easy modeling and identication procedures for
black-box models of power converters are proposed. These
models are composed of reduced-order dynamic systems, which
allow optimizing the required computational resources for sim-
ulation [1], [23]. On the other hand, the proposed identication
procedure is based on analyzing the converter step response,
which is obtained from simple experiments, carried out with in-
expensive equipment and straightforward adjusting algorithms
(see Fig. 2).
The proposed procedures allow modeling and identifying
large-signal models of these dynamic systems, either with a
linear or nonlinear dynamic response.
The paper is organized as follows.
1) Section II explains the proposed modeling method. It is
explained for the output network of power converters with
controlled dc output voltage, and it can also be applied
similarly to model other networks.
2) Section III deals with the identication of the models.
Existing methods are reviewed and discussed, and identi-
cation methods are proposed.
3) Section IV shows experimental results. The proposed pro-
cedures are used to identify a black-box behavioral
model of two converters output: a commercial step-down
dcdc converter and an actual line-commutated rectier,
which provides a 270 V
dc
bus for an experimental test
bench of an aircraft power system. Some step tests are
also shown and discussed.
II. MODELING METHOD
A. Structures of the Dynamic Systems that Compose
a Black-Box Model
As explained before, a black-box dynamic model of a power
converter can be composed of several dynamic systems (see
Fig. 1). However, these dynamic systems can exhibit either a
linear or a nonlinear behavior. The next structures can be used
to model them.
1) Linear Models: If the dynamic system can be consid-
ered linear, or nearly linear, it can be modeled as a single
LTI model. An LTI model can be dened by a single transfer
function.
For example, this approach is useful to model the output
impedance of current-mode controlled (CMC) converters since
it can be assumed linear in many practical cases (see Fig. 3).
2) Nonlinear Models: If the dynamic systems exhibit a non-
linear behavior in the converter range of operation, a nonlinear
approach should be considered.
1) Polytopic Structures: Reference [24] shows that polytopic
structures are a suitable choice in modeling the nonlinear
dynamics of a power converter. A polytopic structure is
composed of a set of local small-signal LTI models, and
each one is parameterized for a certain operating condi-
tion. Each local model is weighted as a function of the con-
verter operating point by means of weighting functions,
and the response of all of themis combined to compute the
model response. For example, this approach is useful to
model the output impedance of voltage-mode-controlled
buck-derived converters, since it depends on the input volt-
age in many cases (e.g., converters without feedforward
compensations). Fig. 4 shows an example, in which Z
o
is composed of three small-signals local models obtained
on three different input voltages (V
iQ1
, V
iQ2
, and V
iQ3
).
A deeper explanation of polytopic structures applied on
power electronics modeling can be found in [24] and [28].
2) Structures Based on Nonlinear Static Networks Connected
with Linear Dynamic Networks: References [19][21],
[25] have shown that nonlinear static networks connected
with linear dynamic networks can be applied in model-
ing converters, which exhibit a nonlinear static behavior.
They can be used to model the static load and line regula-
tions, the constant power source behavior of the regulated
converter (if the input voltage increases, the input current
decreases), and the efciency. For example, [22] proposes
a Wiener structure to model the back current gain H
i
(see
Fig. 1) in order to consider the constant power source be-
havior and the efciency. Fig. 5 shows the use of this kind
of structures to model Z
o
. It is composed of a static non-
linear network R
c
(i
o
), whose output is a function of i
o
,
and an LTI model LTI
Z
o
, dened by a transfer function.
This structure allows modeling the static load regulation
in a nonlinear way.
Though these structures have been shown in modeling the
output impedance (see Figs. 35), they can also be applied in
modeling the other dynamic systems that compose a black-
box behavioral model.
Therefore, LTI models are suitable to compose both linear and
nonlinear dynamic black-box models, using the explained ap-
proaches. This paper is focused on modeling and identication
of these LTI models.
B. Model Characterization Through the Converter Response
The converter response has to be analyzed in order to choose
the proper LTI parameters of the models.
The two-port network model shown in Fig. 1 is an unter-
minated model [22], so the dynamic systems that compose it
only should model the internal converter dynamics. The experi-
ments should be chosen in such a way that the inuences of the
2778 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 12, DECEMBER 2009
Fig. 2. Proposed procedures applied over the identication of the output impedance of a black-box behavioral model.
Fig. 3. Output impedance of a power converter dened by a single LTI model.
external equipments (source and load) are minimized [22], [23].
Furthermore, the experiment has to be informative enough,
meaning that the obtained data allow an accurate tting of the
LTI model in the frequency bandwidth of interest. As a conse-
quence, the choice of a proper input signal is a critical task [29].
For example, reference [22] proposes a set of experiments to
measure the frequency response of the dynamic systems that
compose the unterminated model. A sinusoidal input signal is
applied by means of a network analyzer and a linear amplier.
However, this paper deals with modeling and identication of
the two-port network based on the step response of the converter.
It focuses on the output network of converters with dc output
voltage, and it is worth noticing that the discussed methods can
also be applied to model other networks in a similar way.
In this proposal, the step input signal has been selected, since
step tests can be easily carried out in power electronics labo-
ratories with low-cost equipment, and the measured transient
Fig. 4. Output impedance of a power converter dened by a polytopic
structure.
Fig. 5. Output impedance of a power converter dened by a nonlinear static
network connected with a linear dynamic network.
waveforms can be used to t accurately reduced-order models
in the frequency bandwidth of interest.
The necessary tests to parameterize the output network are
the following. Notice that the step slew rate limits the frequency
bandwidth to be identied.
VALDIVIA et al.: SIMPLE MODELING AND IDENTIFICATION PROCEDURES FOR BLACK-BOX BEHAVIORAL MODELING 2779
1) An output current step is applied while the input volt-
age is kept constant in order to parameterize the output
impedance Z
o
.
2) An input voltage step is applied while the output current
is kept constant in order to parameterize the audiosuscep-
tibility G
o
.
In practice, it is not possible to keep these quantities abso-
lutely constant. The input voltage will be perturbed under a load
step test because of the output impedance of the voltage source.
On the other hand, the output current will be perturbed under
an input voltage step because of the load. These effects could
affect the measurements, but their inuence can be assumed
negligible if both the voltage source and the load are properly
selected, and the perturbation of the input voltage and the output
current, respectively, is low enough. If these perturbations are
found to be signicant, methods proposed in [23] can be applied
to remove this inuence. Some actual implementations of these
tests are discussed in Section IV.
C. Proposed Reduced-Order Models
Once the converter response has been measured, the number
of parameters that compose each LTI model can be found. As
explained earlier, an LTI model can be dened by means of a
transfer function (1), where m n.
LTI(s) =
b
m
s
m
+ b
m1
s
m1
+ + b
o
a
n
s
n
+ a
n1
s
n1
+ + a
o
. (1)
Power converters usually have a high number of poles and
zeros in their closed-loop dynamic transfer functions. This fact
is a handicap: selecting the optimal number of parameters of
each dynamic system would be difcult and high-order models
could be derived, requiring a high computational cost. However,
this procedure can be simplied if the following fact is taken
into account: the response of a dynamic LTI system can be
analyzed as the sum of rst-order and second-order subsystems
response [30]. The rst-order subsystems allow modeling the
real poles of the system, and the second-order subsystems allow
modeling the complex poles. Therefore, using partial fraction
expansion (1) can be rewritten as (2).
LTI(s) = K
0
+

i
K
1i
s +
1i
+

j
K
21j
s + K
22j
s
2
+ 2
2j

2j
s +
2
2j
.
(2)
This approach is simple and suitable when reduced-order
models of the dynamic systems are required. The main rst-
order and second-order evolutions can be identied in the step
response, and the LTI model can be composed of the sum of the
corresponding rst-order and second-order transfer functions
(K
0
= 0, when m < n).
Regarding regulated converters with dc output, a zero at the
origin appears in the transfer functions of Z
o
and G
o
, because
these converters have a pole at the origin in their control stages
to minimize steady-state error. Therefore, (1) can be rewritten
Fig. 6. Overdamped transient response of a CMC buck converter under a load
current step (switching model).
as (3), (b
0
= 0).
Z
o
(s), G
o
(s) =

i
K
1i
s
s +
1i
+

j
K
21j
s
2
+ K
22j
s
s
2
+ 2
2j

2j
s +
2
2j
.
(3)
This zero means that measuring the step response is equivalent
to measure the impulse response of a system with the same
parameters but without that zero. As a consequence, the step
response can be analyzed as the sum of the impulse response of
rst-order and second-order subsystems, and the model can be
fully characterized (an LTI model is completely dened by its
impulse response).
Two kinds of step responses can generally be found: over-
damped (real poles) and underdamped (both real and complex
poles).
1) Overdamped: A typical overdamped response of a power
converter is shown in Fig. 6. For example, this kind of response
can be found at the output voltage of CMC dcdc converters,
when load steps are applied.
In most cases, overdamped responses can be properly approx-
imated by the sumof the responses of two rst-order subsystems
(4).
LTI
1
(s) =
K
11
s
s +
11
+
K
12
s
s +
12
. (4)
As a consequence, the converter output impedance of this
example can be approximated by means of (4). Fig. 7 shows
how this response is approximated by the sum of two rst-
order subsystems responses: the slowest exponential evolution
is mainly modeled by the slowest subsystem fos1, and the
initial time interval of the response is mainly modeled by the
fastest subsystem fos2.
The approximated model and the true system are compared
in Fig. 8. It can be concluded that the response of the reduced-
order model is quite close to the converter response. Moreover,
the order of the adjusted model is signicantly lower: the true
system is a sixth-order system, and the adjusted model is a
second-order system.
2780 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 12, DECEMBER 2009
Fig. 7. Overdamped response approximated as the sum of two rst-order
subsystems responses.
Fig. 8. Output impedance of a CMC buck converter: true system (average
small signal model, sixth-order, black line) and reduced-order model (second-
order, gray line).
Fig. 9. Underdamped transient response of a VMC buck converter under a
load current step (switching model).
If both real poles are very close, the systemcan be modeled by
means of (5), which corresponds to a critically damped second-
order system.
LTI

1
(s) =
K
21
s
(s +
21
)
2
. (5)
2) Underdamped: A typical underdamped response of a
power converter is shown in Fig. 9. For example, this kind
of responses can be commonly found at the output voltage of
voltage-mode controlled (VMC) power converters, when load
steps are applied.
Most underdamped responses can be properly approximated
by the sum of the responses of rst-order subsystems and a sec-
ond order subsystem. In this example, the sum of two rst-order
subsystems, fos1, and fos2, and a second-order subsystem
sos allows obtaining a good t (6).
LTI
2
(s) =
K
11
s
s +
11
+
K
12
s
s +
12
+
K
21
s
2
+ K
22
s
s
2
+ 2
2

21
s +
2
21
. (6)
The added second-order subsystem corresponds to a damped
sinusoidal response, which could present a certain phase delay,
which is related to term K
21
. Regarding the example shown
in Fig. 9, no delay is found, so K
21
= 0, and the response is
approximated, as shown in Fig. 10.
The output impedance of the power converter and the ap-
proximated model are compared in Fig. 11. The model matches
properly the true system, though the order reduction is less than
in the previous example: the true system is a fth-order system,
and the adjusted model is a fourth-order system.
Fig. 12 shows an example of an underdamped response in
which a phase delay is found. It corresponds to the output volt-
age of a VMC boost converter, when an input voltage v
i
step is
applied (v
i
v
i
). The reduced-order model and the true sys-
tem match properly in the bandwidth of interest (approximately
up to 5 kHz in this particular example, Fig. 13), and the time
waveforms are very close (see Fig. 12).
In this case, only one rst-order subsystem was necessary to
obtain a proper adjustment, so a third-order model is used (7),
and a signicant model order reduction is achieved.
LTI

2
(s) =
K
11
s
s +
11
+
K
21
s
2
+ K
22
s
s
2
+ 2
21

21
s +
2
21
. (7)
VALDIVIA et al.: SIMPLE MODELING AND IDENTIFICATION PROCEDURES FOR BLACK-BOX BEHAVIORAL MODELING 2781
Fig. 10. Underdamped response approximated as the sum of two rst-order
and one second-order subsystems responses.
Fig. 11. Output impedance of a VMC buck converter: true system (average
small-signal model, fth-order, black line) and reduced-order model (fourth-
order, gray line).
Fig. 12. Underdamped transient response of a VMC boost converter, under
input voltage step. Responses of both the switching model (black line) and the
adjusted model (gray line).
Fig. 13. Audiosusceptibility of a VMC boost converter: true system (average
small-signal model, fth-order, black line) and reduced-order model (third-
order, gray line).
In some cases, a transient response close enough to a pure
second-order system response can be found. Then, both rst-
order subsystems can be avoided in order to adjust the reduced-
order behavioral model (8).
LTI

2
(s) =
K
21
s
2
+ K
22
s
s
2
+ 2
21

21
s +
2
21
. (8)
Table I summarizes the proposed reduced-order models. They
can be applied to model the dynamic systems, which dene a
power converter dc output in most practical cases. Only in some
cases, if the proposed reduced-order models were not suitable
to obtain an accurate enough tting, higher order models would
be necessary.
2782 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 12, DECEMBER 2009
TABLE I
PROPOSED REDUCED-ORDER MODELS FOR G
o
AND Z
o
III. IDENTIFICATION
Reduced-order models have been proposed in Section II.
However, an identication method is needed to parameterize
them. In this section, the main LTI systems identication meth-
ods are reviewed and discussed. Finally, an alternative one is
proposed.
A. Brief Review of LTI Systems Identication Methods
A high number of LTI models identication methods have
been reported in literature. There are two main methods: non-
parametric and parametric [29], [31]:
1) Nonparametric Methods: Nonparametric methods are
used to identify models that are not dened by a nite number
of parameters. Nonparametric models are curves or functions,
such as the frequency response (Bode, Nyquist or Nichols plots)
or the impulse response.
These methods can be classied in two groups.
a) Impulse response estimation methods: These methods
are based on the transient response. The identication
is performed by analyzing the impulse response, the
step response, or by means of correlation techniques.
b) Frequency response estimation methods: These meth-
ods are based on the frequency response analysis and
the spectral analysis.
Nonparametric methods are easy to use, and any assump-
tion about the identied system structure is not required. They
are used in some power electronics applications, e.g., autotun-
ing in digitally controlled converters [33], [34]. However, the
resulting models are not suitable for simulation.
2) Parametric Methods: The parametric methods are applied
to identify models, which are dened by a set of parameters.
They are generally based on the least-squares method and the
extensions of it, and they can be applied to t models using both
transient response and frequency response data [29], [31], [32].
The parametric models can be dened by means of several
structures.
1) Transfer Function Models: Common structures of transfer
function models are autoregressive with external input,
autoregressive moving average with external input, output
error (OE), and Box-Jenkins.
2) State Space Models: These methods allow adjusting LTI
models accurately, which can be applied for simulation.
The selection of a proper input signal is a critical task,
in order to reach a ne adjustment of the model in the
frequency bandwidth of interest. Moreover, proper model
structure and number of parameters have to be selected.
Parametric methods are used to identify the power converters
black-box models in [22][25] from frequency response data.
B. Proposed Identication Method
Regarding system level simulation, parametric identication
methods seem to be the proper ones, since the identied models
can be directly applied for simulation.
1) Transfer function models can be adjusted using these
methods, from step transient response.
2) Section II proposes the parameters that should compose
them.
In this section, two identication methods are proposed: the
rst one is based on split the step response as a sumof subsystem
responses, and a sequential identication of them. The second
one is based on the least-squares method and the extensions of
it.
1) Method-Based on Sequential Identication of Subsystems:
This method is suitable to identify those structures proposed
in Section II. If an input signal close to an ideal step is ap-
plied to the converter, the measured response can be split as
a sum of rst-order and second-order responses, and the cor-
responding subsystems can be easily identied following a se-
quential procedure. The main advantage of this approach is the
ease of implementation, since the identication can be carried
out, following a simple procedure and no data preprocessing is
required.
The proposed procedure is dened by a set of steps described
shortly. Fig. 14 shows its application to identify the buck sim-
ulated response, as shown in Fig. 9, and can be applied in a
similar way to model other kind of responses.
VALDIVIA et al.: SIMPLE MODELING AND IDENTIFICATION PROCEDURES FOR BLACK-BOX BEHAVIORAL MODELING 2783
Fig. 14. Identication iterative procedure applied on the response shown in Fig. 9.
In addition, the detailed application of the method to identify
a commercial converter is shown in Section IV.
1) Step 1: Subtract the steady-state value ssfromthe actual
response ar.
2) Step 2: Find a time interval at the end of the transient
response in which only the response of a single subsystem
is dominant and identify this subsystem. In this case, an
exponential evolution is found, so a rst-order subsystem
fos1 is tted.
3) Step 3: Subtract from ar-ss the effect of the identied
subsystem fos1. Identify a new dominant subsystem at
the end of the new response. An underdamped response
is clearly found in this case, so a second-order subsystem
sos is tted.
4) Step 4: Subtract from ar-ss-fos1 the effect of the
identied subsystem sos and identify the remaining
subsystems (in this case, another rst-order subsystem
fos2).
Once the subsystems have been identied, the LTI model can
be composed of the sum of the identied subsystems.
The identication of rst-order and second-order subsystems
can be carried out in many ways (e.g., by means of normalized
plots and analytical expressions) [30]. In this paper, a set of
simple analytical expressions is given.
1) Identication of First-Order Subsystems: First-order sub-
systems are dened by two parameters: a gain K
1
and
a time constant
1
= 1/
1
. Hence, both parameters can
be identied by selecting two points P
11
(y
11
, t
11
) and
2784 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 12, DECEMBER 2009
P
12
(y
12
, t
12
) into the transient response of the subsys-
tem, and applying (9) (u is the height of the step).
It is recommended to select these points separately
enough, in the interval in which this subsystem is dom-
inant, in order to achieve a proper t. Both Step 2 and
Step 4 (Fig. 14) illustrate the identication of rst-order
subsystems.
y(s) = u(s)LTI
1
(s) =
u
s
K
1
s
s +
1
y(t) =uK
1
e
t
1

1
=
ln(y
11
/y
12
)
t
12
t
11
K
1
=
y
12
u
e
t
1 2

1
. (9)
2) Identication of Second-Order Critically Damped Subsys-
tem: Critically damped systems are dened by a constant
K
2
and an angular frequency
2
. This subsystem should
be identied when an overdamped response, in which both
real poles are very close, is found. The identication can
be carried out by selecting the peak point of the transient
response P
2
(y
2
, t
2
), and applying (10).
y(s) = u(s)LTI

1
(s) =
u
s
K
2
s
(s +
2
)
2

y(t) = uK
2
te
t
2

2
=
1
t
2
K
2
=
y
2
u
e
t
2

2
t
2
. (10)
3) Identication of Second-Order Underdamped Subsys-
tems: The second-order subsystems transfer functions are
parameterized by identifying their gains K
21
and K
22
,
oscillation
2
, and damping factor
2
. The identication
can be carried out by measuring the damped oscillation
frequency
d
, selecting two maximum and/or minimum
points (either local or absolute), into the transient response
of the subsystem, P
21
(y
21
, t
21
), P
22
(y
22
, t
22
), and apply-
ing (11). Step 3 (see Fig. 14) illustrates the identication
of second-order underdamped subsystems.
y(s) = u(s)LTI
2
(s) =
u
s
K
21
s
2
+ K
22
s
s
2
+ 2
2

2
s +
2
2

y(t) = uGe
t
2

2
sin
_

2
_
1
2
2
t +
_

2
=
ln
_

y
2 1
y
2 2

_
1
(t
2 2
t
2 1
)
d
_
1 +
_
ln
_

y
2 1
y
2 2

_
1
(t
2 2
t
2 1
)
d

2
= atan
_
_
1
2
2

2
_

d
t
21

2
=

d
_
1
2
2
G =
y
21
u
e
t
2 1

2
sin (
d
t
21
+ )
K
21
=
tan ()
_
1 + tan
2
()
G
K
22
= K
21

2
+
G
_
1 + tan
2
()

d
. (11)
Therefore, the main advantage of this method is the ease of
use. It can be used when an input signal close to an ideal step
can be applied.
Fig. 15. OE model.
If the response of each subsystemis separate enough fromthe
response of the others, an accurate t can be performed using
the proposed method. However, if the subsystems responses are
not separate enough, nding the best t could be more difcult
and the following methods may be used instead of it.
2) Least-Squares Method and the Extensions of It: Paramet-
ric identication based on the least-squares method and the ex-
tensions of it can be generally applied to t the transfer function
models, which compose the black-box model, from measured
step responses. There are several structures of transfer func-
tion models, and they differ on how they model the dynamic
relation between input, output, and disturbance. A disturbance
model is not considered in this paper, so the OE model is sug-
gested (see Fig. 15), in which B(Z)/F(z) is the LTI system to be
identied [29].
Identication algorithms of OE models have been widely dis-
cussed in literature [29], [31], [32], and there are commercial
tools that allowidentifying them(e.g., MATLABSystemIdenti-
cation Toolbox [35]). The OE model identication algorithms
are useful even when the step signal is not too close to an ideal
step (the identied frequency bandwidth will be limited by the
step slew rate). However, the required tting algorithms may be
complex and specic software tools may be required, as well as
data preprocessing.
Second-order to fourth-order models are suitable to t
most step responses accurately, according to that exposed in
Section II.
C. Linear and Nonlinear Dynamic Systems Identication
The LTI models can be used to model both linear and
nonlinear dynamics of power converters, as explained in
Section II.
1) Linear Dynamic Systems Identication: if the identied
dynamic system can be supposed linear, the correspond-
ing LTI model can be identied by means of conven-
tional parametric methods, or the method proposed in this
section.
2) Nonlinear Dynamic Systems Identication: If the dynamic
system is nonlinear, the nonlinear structures can be iden-
tied as follows.
a) Polytopic structure: Small step tests applied on dif-
ferent operating conditions should be performed.
For example, the identication of a VMC buck con-
verter audiosusceptibility is shown in Fig. 16. It de-
pends on the input voltage operating point, so small
input voltage steps are applied on three operating
input voltage values V
iQ1
, V
iQ2
, and V
iQ3
.
VALDIVIA et al.: SIMPLE MODELING AND IDENTIFICATION PROCEDURES FOR BLACK-BOX BEHAVIORAL MODELING 2785
Fig. 16. Polytopic structure of VMC buck audiosusceptibility.
b) Structures based on a nonlinear static model con-
nected with a linear dynamic model: The static non-
linear model can be identied by means of the anal-
ysis of the steady-state response of the converter at
different operating points. After that, the effect of
the static network is subtracted from the measured
response, and nally the LTI model is identied.
IV. EXPERIMENTAL RESULTS
The proposed procedures have been applied to obtain a
black-box behavioral model of the output of two kinds of
converters: Astep-down commercial dcdc converter and a line-
commutated rectier.
A. DCDC Switching Converter
The Texas Instruments PTN78020WAZ switching converter
[36] has been modeled. The converter has been set to provide
a regulated output voltage V
o
= 2.5 V, and the considered op-
erating ranges during the tests have been V
i
= [815 V] and
P
o
= [413 W], respectively. Both an output capacitor of 470F
and an input capacitor of 2.2 F have been externally placed.
In this case, both the audiosusceptibility G
o
and the output
impedance Z
o
have been considered, and their dependence on
the operating point has been evaluated. The applied tests are
shown in Fig. 17.
1) The load step test [see Fig. 17(a)] allows identifying the
output impedance of the model. A MOSFET IRF740 has
been used to apply the step. A certain oscillation appears
at the end of the step because of using resistors. However,
if the output current waveform is close to an ideal step,
the proposed identication method III.B.1 can be used.
A step-down test is preferred because the oscillation is
smaller, so the output current waveform is closer to an
ideal step. Several loads and input voltage values have
been set in order to evaluate the Z
o
dependence on both
V
i
and I
o
operating points. The input voltage source should
have lowenough output impedance to ensure that the input
voltage perturbation is low enough, and its inuence on
the output voltage transient response is negligible. In this
Fig. 17. Step tests applied to identify the black-box model of the dcdc
converter: (a) load step and (b) input voltage step.
Fig. 18. Behavioral black-box model of the actual dcdc converter: (a)
model structure, (b) static load regulation, (c) weighting functions of the poly-
topic structure, and (d) LTI models parameters.
case, a voltage source HP 6012B has been used. On the
other hand, output current tests can be also carried out
by means of an electronic load with a high slew rate, if
available.
2786 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 12, DECEMBER 2009
Fig. 19. LTI model identication using the proposed procedure.
Fig. 20. Comparison between the actual measured response and the identied model response. Output current perturbations are applied while the input voltage
is kept constant. Dark lines: actual measured response. Color lines: Model response: v
o
yellow. i
o
blue. (a) Load step-down response used to identify Z
o
.
V
i
= 8 V. (b) Load step-up response used for validation. V
i
= 15 V. (c) Exponential load current response used for validation V
i
= 10 V.
2) The input voltage step test [see Fig. 17(b)] allows identify-
ing the audiosusceptibility of the model. An input voltage
step with high enough slew rate is applied by closing the
switch connected in parallel with the diode. An electronic
load has been connected at the output, in a way that the
output current can be assumed constant during each test.
The HP 6050 A has been used in this case and the i
o
perturbation has been found to be negligible. The G
o
de-
pendence on V
i
and I
o
operating points has been evaluated
by setting different values of them.
Following results were obtained.
1) Output Impedance Z
o
: It has been veried that the response
dependence on I
o
and V
i
operating points is not signicant.
However, static load regulation has been measured.
Therefore, Z
o
has been modeled by means of an LTI model
connected with a nonlinear static network [see Fig. 18(a)], which
have been parameterized as follows.
a) First, the static load regulation has been characterized, by
measuring the steady-state output voltage along the output
power range, in order to parameterize Rc(i
o
). Detailed
data are shown in Fig. 18(b).
b) After that the LTI model is parameterized. A load step-
down test has been applied, in which the output current
waveform is very close to an ideal step. First, the effect of
Rc(i
o
) has been subtracted from the measured waveform.
Second, a fourth-order model corresponding to (6) was ad-
justed using the proposed identication method 3.2.1 (see
Fig. 19). The points in the transient response have been
selected according to the criteria exposed in Section III. It
is worth noticing that the difference between the gains of
both rst-order subsystems allow modeling accurately the
initial sudden rise due to the output capacitor equivalent
series resistance. Models data are given in Fig. 18(d).
Fig. 20 shows a comparison between the adjusted model re-
sponse and the actual converter response when the output cur-
rent is perturbed while input voltage is kept constant. Both load
step-down and load step-up responses have been compared [see
Fig. 20(a) test has been used for identication and Fig. 20(b)
has been used for validation]. In addition, an exponential out-
put current has been set up by means of the electronic load, in
order to validate that the model works properly not only when
load step tests are tested, but also when other current transient
waveforms are applied [see Fig. 20(c)].
2) Audiosusceptibility G
o
: It has been veried that the G
o
dependence on I
o
operating point can be assumed negligible.
However, a signicant dependence on the input voltage was
VALDIVIA et al.: SIMPLE MODELING AND IDENTIFICATION PROCEDURES FOR BLACK-BOX BEHAVIORAL MODELING 2787
Fig. 21. Comparison between the actual measured response and the identi-
ed model response. Input voltage perturbations are applied while the output
current is kept constant. Dark lines: actual measured response. Color lines:
model response: v
o
yellow. v
i
green. (a) V
i
step-up response used to iden-
tify G
oV
i
=8
. I
o
= 2 A. (b) V
i
step-up response used to identify G
oV
i
=11. 5
.
I
o
= 2 A.
found (the higher the input voltage, the lower G
o
). On the other
hand, no signicant static line regulation has been measured.
Therefore, a polytopic structure has been selected to model
it [see Fig. 18(a)]. The range partitioning and the weighting
functions have to be properly selected [24], [28]. Generally,
more partitions should mean higher model accuracy. On the
other hand, the sum of the weighting functions has to be always
equal to one.
As a simple example, an equally spaced partition of the input
voltage range in two regions and triangular weighting functions
were chosen [see Fig. 18(c)]. The input voltage steps have been
applied on each weighting function maximum point. Since the
Fig. 22. Comparison between the actual measured response and the identied
model response. Both input voltage and output current perturbations are applied.
Dark lines: actual measured response. Color lines: model response: v
o
yellow.
v
i
green. i
o
blue (note that both measured and simulated currents are totally
overlapped).
Fig. 23. Load step tests applied on the actual line-commutated rectier.
magnitude of the steps is low enough, the identication of each
small signal LTI model is accurate.
A second-order model is suitable to get a good t in this case
(8). The absolute maximum and minimum points have been
chosen as P
21
and P
22
, respectively, in order to apply (11) in
all cases. Fig. 21(a) and (b) shows the transient responses used
to identify two of the three LTI local models. Detailed data of
them are given in Fig. 18(d).
Finally, both v
i
and i
o
are simultaneously perturbed in order
to validate the behavior of the whole model (see Fig. 22). It can
be seen that both responses are very close, so it is concluded that
the output of the dcdc converter has been properly modeled.
B. Line-Commutated Rectier
This converter generates the dc bus of a distributed power
system for an aircraft test bench. The rated power is 40 kW. The
input ac voltage of the converter is obtained from the ac grid, so
its magnitude is assumed to be constant, and G
o
is neglected.
The proposed procedures are signicantly useful in this case.
1) Areduced-order model of the converter is suitable in order
to optimize the required computational cost, because the
power distribution system is composed of a high number
of loads and converters.
2) Several resistive loads are implemented in the test bench,
so load steps can be easily carried out (see Fig. 23). The
2788 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 12, DECEMBER 2009
Fig. 24. Behavioral black-box model of the actual rectier: (a) model struc-
ture, (b) static load regulation, and (c) LTI model parameters.
Fig. 25. Comparison between the responses of the actual rectier and the iden-
tied model when load steps are applied. Blue lines: actual converter response;
red and brown lines: adjusted model response. (a) 20%40% of rated load step.
(b) 30%70% of rated load step.
i
o
waveform is close to an ideal step, so the proposed
identication method III.B.1 can be used.
3) The output voltage waveform has a signicant ripple,
in which the harmonic content is close to the main fre-
quency oscillation of the transient response [interval of
time (0.050.15 s)]. Using the proposed identication
method, a signal preprocessing, which could be complex,
is avoided.
The load steps have been applied using different resistive
loads, in order to evaluate the Z
o
dependence on I
o
operating
point. It has been veried that it is negligible (the converter oper-
ates in continuous conduction mode in the considered operating
range). However, static load regulation has been measured, so
Z
o
was modeled by means of a nonlinear static model connected
with an LTI model structure [see Fig. 24(a)].
Afourth-order LTI model has been chosen again (6), although
a second-order model (8) would also have been a proper choice
because the computational cost is optimized at the expense of a
low accuracy loss. The model data are given in Fig. 24(c). The
second-order model is adjusted assuming that it is dominant at
time 0.1 s, and both rst-order subsystems are added to obtain
a ner t. A comparison between the transient response of the
actual rectier and the identied model is shown in Fig. 25,
showing that both of them are close enough.
V. CONCLUSION
Simple modeling and identication methods for black-box
behavioral models of power converters, based on the transient
response analysis, have been proposed.
1) Some concepts about black-box behavioral modeling
have been discussed, showing that LTI models can be used
to compose large-signal black-box models of power
converters. When the behavior of the converter is linear,
a single LTI model is suitable. When the behavior of the
converter is nonlinear, nonlinear structures that combine
LTI models and nonlinear functions can be used.
2) A set of reduced-order LTI models has been proposed.
Reduced-order models are appropriate when large power
systems are analyzed, since the simulation time can be
optimized. Those models can be selected by simply an-
alyzing the step response of the converter, which can be
obtained in an easy way.
3) The main identication procedures of LTI models have
been reviewed, and their applicability to the identication
of black-box models of power converters, based on the
step response, has been discussed. An alternative and sim-
ple method has been proposed. This method allows iden-
tifying the proposed reduced-order models, without need
of data preprocessing and complex-tting algorithms.
4) Finally, black-box behavioral models of the output net-
work of both an actual commercial dcdc converter and
an actual line-commutated rectier, which is applied in a
test bench of an aircraft power system, have been mod-
eled and identied by means of the proposed approaches.
The experimental results show a proper response of the
identied models.
REFERENCES
[1] B. H. Cho and F. C. Y. Lee, Modeling and analysis of spacecraft power
systems, IEEE Trans. Power Electron., vol. 3, no. 1, pp. 4454, May
1988.
[2] M. D. Kankum and M. E. Elbuluk, A survey of power electronics appli-
cations in aerospace technologies, presented at the 36th Intersoc. Energy
Convers. Eng. Conf., Akron, OH, Nov. 2001.
[3] A. Emadi and M. Ehsani, Aircraft power systems: Technology, state of
the art and future trends, IEEE Aerosp. Electron. Syst. Mag., vol. 15,
no. 1, pp. 2832, Jan. 2000.
VALDIVIA et al.: SIMPLE MODELING AND IDENTIFICATION PROCEDURES FOR BLACK-BOX BEHAVIORAL MODELING 2789
[4] J. A. Rosero, J. A. Ortega, E. Aldabas, and L. Romeral, Moving towards
a more electric aircraft, IEEE Aerosp. Electron. Syst. Mag., vol. 22, no. 3,
pp. 39, Mar. 2007.
[5] A. Emadi and M. Ehsani, Multi-converter power electronic systems:
Denition and applications, in Proc. IEEE Power Electron. Spec. Conf.,
Vancouver, BC, Canada, 2001, pp. 12301236.
[6] R. Jayabalan, B. Fahimi, A. Koenig, and S. Pekarek, Applications of
power electronics-based systems in vehicular technology: State of the art
and future trends, in Proc. IEEE Power Electron. Spec. Conf. (PESC
2004), Jun. 2025, pp. 18871894.
[7] D. Boroyevich, R. Burgos, L. Arnedo, and F. Wang, Synthesis and inte-
gration of future electronic power distribution systems, in Proc. Power
Convers. Conf. (PCC 2007), Apr. 25, pp. 18.
[8] W. Zhu, S. Pekarek, J. Jatskevich, O. Wasynczuk, and D. Delisle, A
model-in-the-loop interface to emulate source dynamics in a zonal DC
distribution system, IEEETrans. Power Electron., vol. 20, no. 2, pp. 438
445, Mar. 2005.
[9] X. Feng, J. Liu, and F. C. Lee, Impedance specications for stable DC
distributed power systems, IEEE Trans. Power Electron., vol. 17, no. 2,
pp. 157162, Mar. 2002.
[10] J. Liu, X. Feng, F. C. Lee, and D. Boroyevich, Stability margin monitoring
for DC distributed power systems via perturbation approaches, IEEE
Trans. Power Electron., vol. 18, no. 6, pp. 12541261, Nov. 2003.
[11] G. A. Franz, G. W. Ludwig, and R. L. Steigerwald, Modeling and simu-
lation of distributed power systems, in Proc. IEEE Power Electron. Spec.
Conf., San Antonio, TX, 1990, pp. 606610.
[12] K.-S. Tam and L. Yang, Functional models for space power electronic
circuits, IEEE Trans. Aerosp. Electron. Syst., vol. 31, no. 1, pp. 288296,
Jan. 1995.
[13] K. J. Karimi, A. Booker, and A. Mong, Modeling, simulation, and ver-
ication of large DC power electronics systems, in Proc. IEEE Power
Electron. Spec. Conf., Baveno, Italy, Jun. 1996, pp. 17311737.
[14] T. Suntio and D. Gadoura, Use of unterminated two-port modeling tech-
nique in analysis of input lter interactions in telecom DPS systems,
in Proc. Telecommun. Energy Conf. (INTELEC 2002), Sep. 29Oct. 3,
pp. 560565.
[15] M. Karppanen, M. Hankaniemi, T. Suntio, and M. Sippola, Dynami-
cal Characterization of peak-current-mode-controlled buck converter with
output current feedforward, IEEE Trans. Power Electron., vol. 22, no. 2,
pp. 444451, Mar. 2007.
[16] R. Wang, J. Liu, and H. Wang, Universal approach to modeling current
mode controlled converters in distributed power systems for large-signal
subsystem interactions investigation, in Proc. IEEE Appl. Power Elec-
tron. Conf. (APEC 2007), Feb. 25Mar. 1, pp. 442448.
[17] A. Emadi, Modeling and analysis of multiconverter DC power electronic
systems using the generalized state-space averaging method, IEEETrans.
Ind. Electron., vol. 51, no. 3, pp. 661668, Jun. 2004.
[18] P. G. Maranesi, V. Tavazzi, and V. Varoli, Two-port characterization of
PWM voltage regulators at low frequencies, IEEE Trans. Ind. Electron.,
vol. 35, no. 3, pp. 444450, Aug. 1988.
[19] J. A. Oliver, R. Prieto, V. Romero, and J. A. Cobos, Behavioral model-
ing of dc-dc converters for large-signal simulation of distributed power
systems, in Proc. IEEE Appl. Power Electron. Conf. (APEC 2006), Mar.
1923, pp. 10541060.
[20] J. A. Oliver, R. Prieto, V. Romero, and J. A. Cobos, Behavioral mod-
eling of multi-output DC-DC converters for large signal simulation of
distributed power systems, in Proc. IEEE Power Electron. Spec. Conf.
(PESC 2006), Jun. 1822, pp. 16.
[21] J. Oliver, R. Prieto, J. Cobos, P. Alou, and O. Garcia, Hybrid Wiener
Hammerstein structure for grey-box modeling of DC DC converters,
in Proc. IEEE Appl. Power Electron. Conf. (APEC 2009), Feb. 1519,
pp. 280285.
[22] L. Arnedo, R. Burgos, F. Wang, and D. Boroyevich, Black-box terminal
characterization modeling of DC-to-DC converters, in Proc. IEEE Appl.
Power Electron. Conf. (APEC 2007), Feb. 25Mar. 1, pp. 457463.
[23] L. Arnedo, D. Boroyevich, R. Burgos, and F. Wang, Un-terminated fre-
quency response measurements and model order reduction for black-box
terminal characterization models, in Proc. IEEE Appl. Power Electron.
Conf. (APEC 2008), Feb. 2428, pp. 10541060.
[24] L. Arnedo, D. Boroyevich, R. Burgos, and F. Wang, Polytopic black-box
modeling of DC-DC converters, in Proc. IEEE Power Electron. Spec.
Conf. (PESC 2008), Jun. 1519, pp. 10151021.
[25] L. Arnedo, R. Burgos, D. Boroyevich, and F. Wang, System-level black-
box Dc to Dc converter models, in Proc. IEEE Appl. Power Electron.
Conf. (APEC 2009), Feb. 1519, pp. 14761481.
[26] A. Barrado, D. Izquierdo, C. Raga, A. Lazaro, and M. Sanz, SSPC
model with variable reset time, environmental temperature compensation
and thermal memory effect, in Proc. IEEE Appl. Power Electron. Conf.
(APEC 2008), Feb. 2428, pp. 17161721.
[27] Q. Liu, F. Wang, and D. Boroyevich, Modular-terminal-behavioral
(MTB) model for characterizing switching module conducted EMI gener-
ation in converter systems, IEEE Trans. Power Electron., vol. 21, no. 6,
pp. 18041814, Nov. 2006.
[28] S. Glover, Modeling and stability analysis of power electronics based
systems, Ph.D. Dissertation, Purdue Univ., Indianapolis, IN, 2003, p.
179.
[29] L. Ljung, SystemIdentication: Theory for the User, 2nd ed. Englewood
Cliffs, NJ: Prentice-Hall, 1999.
[30] K. Ogata, Modern Control Engineering, 4th ed. Englewood Cliffs, NJ:
Prentice-Hall, 2002.
[31] T. S oderstr om and P. Stoica, System Identication. Englewood Cliffs,
NJ: Prentice-Hall, 1989.
[32] Y. Zhu, Multivariable System Identication for Process Control. Ams-
terdam, The Netherlands: Elsevier, 2001.
[33] B. Miao, R. Zane, and D. Maksimovic, System identication of power
converters with digital control through cross-correlation methods, IEEE
Trans. Power Electron., vol. 20, no. 5, pp. 10931099, Sep. 2005.
[34] M. Shirazi, J. Morroni, A. Dolgov, R. Zane, and D. Maksimovic, Integra-
tion of frequency response measurement capabilities in digital controllers
for DC-DC converters, IEEE Trans. Power Electron., vol. 23, no. 5,
pp. 25242535, Sep. 2008.
[35] MATLAB R2008b. (2008). [Online]. Available: http://www.mathworks.
com/
[36] TI PTN78020WAZ. (2008). [Online]. Available: http://focus.ti.com/lit/ds/
symlink/ptn78020w.pdf
Virgilio Valdivia (S09) was born in Barcelona in
1983. He received the M.Sc. degree in electrical en-
gineering in 2006 from the University Carlos III of
Madrid, Legan es, Spain, where he is currently work-
ing toward the Ph.D. degree in electrical engineering.
Since 2005, he has been a Member of the Power
Electronics Systems Group, University Carlos III of
Madrid. His research interests include modeling and
simulation of power electronics systems and model-
ing and design of integrated magnetic components
for power electronics converters.
Andr es Barrado (M04) was born in Badajoz, Spain,
in 1968. He received the M.Sc. degree in electrical en-
gineering fromthe Polytechnic University of Madrid,
Madrid, Spain, in 1994, and the Ph.D. degree from
the University Carlos III of Madrid, Legan es, Spain,
in 2000.
Since 1994, he has been an Associate Professor at
the University Carlos III of Madrid, and the Head of
the Power Electronics Systems Group since 2001. His
current research interests include switching-mode
power supply, dcdc and acdc converters, modeling
and control of systems, low-voltage fast transient response dcdc converters,
and energy harvesting.
Antonio L azaro (M03) was born in Madrid, Spain,
in 1968. He received the M.Sc. degree in electri-
cal engineering from the Universidad Polit ecnica de
Madrid, Madrid, Spain, in 1995, and the Ph.D. degree
in electronic engineering from the University Carlos
III of Madrid, Legan es, Spain, in 2003.
Since 1995, he has been an Assistant Professor at
the University Carlos III of Madrid. He has been in-
volved in power electronics since 1994, participating
in more than 30 research and development projects
for industry. He is the author or coauthor of more than
100 papers published in IEEE journals and conferences. He is the holder of three
patents. His current research interests include switched-mode power supplies,
power factor correction circuits, inverters (uninterruptible power system and
grid-connected applications), and modeling and control of switching converters
and digital control techniques.
2790 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 24, NO. 12, DECEMBER 2009
Pablo Zumel (M06) received the B.S. degree in
electrical engineering from the University of Burgos,
Burgos, Spain, in 1995, the rst M.S. degree in elec-
trical engineering from the Universidad Polit ecnica
de Madrid, Madrid, Spain, in 1999, the second M.S.
degree from the Ecole Centrale Paris, Paris, France,
in 2000, and the Ph.D. degree from the Universidad
Polit ecnica de Madrid, in 2005.
From 1999 to 2003, he was a Researcher in
the Divisi on de Ingeniera Electr onica, Universidad
Polit ecnica de Madrid. Since 2003, he has been with
the Department of Electronics Technology, University Carlos III of Madrid,
Legan es, Spain, where he is currently an Assistant Professor. His research in-
terests include digital control in power electronics, power electronics system
modeling, and educational issues on power electronics.
Carmen Raga (S06) was born in Madrid, Spain, in
1976. She received the M.Sc. degree in industrial en-
gineering from the University Carlos III of Madrid,
Legan es, Spain, in 2005.
Since 2005, she has been an Assistant Professor at
the University Carlos III of Madrid. Her current re-
search interests include switching-mode power sup-
plies, modeling of dcdc and acdc converters, and
power distribution systems.
Cristina Fern andez (M05) was born in Leon, Spain,
in 1972. She received the M.S. and Ph.D. degrees in
electrical engineering from the Universidad Politec-
nica de Madrid (UPM), Madrid, Spain, in 1998 and
2004, respectively.
From 1997 to 2003, she was a Researcher at the
UPM. Since 2003, she has been an Assistant Pro-
fessor in the Department of Electronic Technology,
University Carlos III of Madrid, Legan es, Spain. Her
current research interests include contact-less power
supplies, modeling of power electronics systems, and
educational issues on power electronics.

Вам также может понравиться