www.ietdl.

org
Published in IET Power Electronics
Received on 13th August 2008
Revised on 11th November 2008
doi: 10.1049/iet-pel.2008.0271

ISSN 1755-4535

LMI robust control design for boost

PWM converters
C. Olalla1 R. Leyva1 A. El Aroudi1 P. Garces1 I. Queinnec2
1

Departament dEnginyeria Electro`nica, Ele`ctrica i Automa`tica, Escola Te`cnica Superior dEnginyeria, Universitat Rovira i
Virgili, Campus Sescelades 43007, Tarragona, Spain
2
CNRS, LAAS, University of Toulouse, 7 Avenue du Colonel Roche, Toulouse F-31077, France
E-mail: carlos.olalla@urv.cat

Abstract: This work presents an analytical study and an experimental verication of a robust control design based
on a linear matrix inequalities (LMI) framework for boost regulators. With the proposed LMI method, nonlinearities and uncertainties are modelled as a convex polytope. Thus, the LMI constraints permit to robustly
guarantee a certain perturbation rejection level and a region of pole location. With this approach, the
multiobjective robust controller is computed automatically by a standard optimisation algorithm. The
proposed method results in a state-feedback law efciently implementable by operational ampliers. PSIM
simulations and experimental results obtained from a prototype are used to validate this approach. The
results obtained are compared with a conventional PID controller.

Introduction

Dc dc switched-mode converters are power efcient devices

used to match the voltage level of an energy source to the
specications of the load. The dynamics of such power
converters is described by non-linear models. Despite the
non-linearities, dc dc converters are usually driven by
means of linear (state or output) feedback controllers, thus
reducing the complexity and cost of the control system.
The control objective of such devices is (i) to maintain
regulation of the output voltage at the desired value, (ii) to
maximise the bandwidth of the closed-loop system in order
to reject disturbances and (iii) to satisfy certain transient
characteristics (as e.g. to minimise output overshoot). Such
linear controllers are usually designed considering a
linearisation of the model at a certain operation point. In
this case, large-signal transients may deteriorate the output
signal or even make the system diverge from the desired
operation point.
Owing to the intrinsic non-linear nature of the switched
regulators, several authors have proposed non-linear
controllers in order to maintain stability over a certain range
of operating conditions. Some of the rst works on nonlinear control for power converters can be found in [1, 2],
IET Power Electron., 2010, Vol. 3, Iss. 1, pp. 75 85
doi: 10.1049/iet-pel.2008.0271

where the authors propose non-linear strategies based on

quadratic Lyapunov functions. More recently, Cortes et al.
[3], He and Luo [4] and Leyva et al. [5], derive robust nonlinear controllers for power converters. These last works seek
large-signal stability of the regulator dynamics when a
complex control law is applied. The main disadvantages of
the previous non-linear controllers are the difculty to
predict transient performances and the complexity of the
implementation.
Besides of non-linear control, other authors have adapted
linear robust control techniques to power converters with the
aim to assure stability under different operating conditions.
Linear control laws are, a priori, easily implementable as, for
example, PID controllers [6 8]. Furthermore, the robust
linear control techniques, unlike conventional and nonlinear control, allow to take into account parametric
uncertainty. A correct treatment of uncertainties is of major
importance in power converters, since some of the regulator
parameters such as the storage elements or the load are
usually time dependent or partially unknown. Some of the
robust methods successfully adapted to power electronics are
H1 [9 11], m-synthesis [12], quantitative feedback theory
(QFT) [13, 14] and approaches based on linear matrix
inequalities (LMI) [15 18]. It is worth to point out some of
75

& The Institution of Engineering and Technology 2009

www.ietdl.org
their characteristics in order to compare such techniques: H1
and m-synthesis methods [9 12] require that the designer
arbitrarily choose weighting functions to specify the required
performances and to approximate the transient properties.
On the contrary, LMI-based control does not necessarily
require weighting functions and it can also deal with
transient requirements with pole placement constraints.
Also, it can be pointed out that the controller expression is
optimised manually in the QFT technique [13, 14], whereas
LMI control permits to optimise the controller parameters
automatically.
Since LMI provides the above advantages, the goal of this
work is to adapt LMI robust control concepts to dc/dc
regulator design. The fact that LMIs can be solved
automatically by efcient standard numerical algorithms
[19 21] has prompted a great number of researchers to
describe different control problems in terms of LMIs.
Therefore this technique can be considered as mature in
the eld of Control Theory.
The controller design shown in this work consists of
describing in terms of LMIs the following restrictions: rst,
the converter stability (in Lyapunovs sense); second, a
minimum level of perturbation rejection; and third, several
constraints in pole location. In all the cases, the design takes
into account uncertainty in the operation point and load
value. Finally, the design is completed when the restrictions
are numerically solved. The result is an efcient state
feedback controller. The main contribution of this work is
the experimental implementation of a prototype which
validates the feasibility of an LMI approach, successfully
applied in many engineering domains, in the area of power
electronics.
It is worth to mention that LMI control has been previously
reported in [1517] to design a state-feedback controller for a
dc/dc converter. The approach shown here differs from the
previous works since we consider a step-up converter, whereas
the previous references consider a buck converter. Therefore
our work takes into account a non-linear system, whose
linearisation yields a non-minimum phase plant, whereas
[1517] deal with a minimum phase linear case. In addition,

[1517] do not describe any experimental prototype, whereas

we present accurate experimental measures.
In order to evaluate the performance achieved by the LMI
controller, we have compared the results with a conventional
output-feedback PID law. Both controls can be easily
implemented by operational ampliers, a PWM circuit and
standard analogue elements.
The remaining sections are organised as follows. Section 2
shows the uncertain model of the non-minimum phase
converter. Section 3 introduces the LMIs and the
controller constraints taken into account. The validity of
the design is veried with experimental results from a boost
prototype in Section 4, where we compare the performance
of the LMI controller with the conventional PID case
numerically using PSIM package. Section 5 summarises the
key aspects of the design method and presents some
conclusions.

2 Boost converter uncertainty

model
Fig. 1 shows the schematic circuit diagram of a dc dc step-up
(boost) converter and the relevant control signals. In Fig. 1, vo
is the output voltage, vg the line voltage and iload the output
disturbance. The output voltage must be kept at a given
constant value Vref . R models the converter load, while C
and L represent, respectively, capacitor and inductor values.
The binary signal (ub) that turns on and off the switches is
controlled by means of a xed-frequency pulse width
modulation (PWM) circuit (see Fig. 1b). The constant
switching frequency is 1=Ts , where Ts is the switching
period equal to the sum of Ton (when ub 1) and Toff
(when ub 0) where the ratio Ton =(Ton Toff ) is the duty
cycle dd . The duty cycle is compared with a sawtooth signal
vs of amplitude VM . We assume that the converter operates
in continuous conduction mode (CCM) and that the
inductor current is not saturated.
The following expressions show the state-space averaged
and linearised model of the boost converter. Averaged

Figure 1 Schematic diagram of a boost converter and its control circuit

a Electric circuit
b PWM waveforms

76

& The Institution of Engineering and Technology 2009

IET Power Electron., 2010, Vol. 3, Iss. 1, pp. 75 85

doi: 10.1049/iet-pel.2008.0271

www.ietdl.org
models of dc dc converters neglect the high-frequency
dynamics because of the switching action, since we consider
Ts much smaller than the converter time constants. The
model is then valid for a frequency range up to half the
switching frequency [22] in the vicinity of the operation
point, since it has been linearised. The state-space
representation is written as


x_ (t) Ax(t) Bw w(t) Bu u(t) Bref Vref

z(t) C z x(t) Dzw w(t) Dzu u(t)

(1)

where
A [ Rnn , Bw [ Rnr , Bu [ Rnm , C z [ Rpn ,
pr
Dzw [ R , Dzu [ Rpm and
2

the changes in parameters Dd0 and R, we express the system

matrices (1) as function of these parameters


x_ (t) A(p)x(t) Bw w(t) Bu (p)u(t) Bref Vref

z(t) C z x(t) Dzw w(t) Dzu u(t)

Only matrices A and Bu depend on the uncertain terms, which

have been grouped in a vector p. In a general case, the vector p
consists of N uncertain parameters p (p1 , . . . , pN ), where
each uncertain parameter pi is bounded between a minimum
and a maximum value pi and pi
h
i
pi [ pi , pi

i L (t)


6
7
x(t) 4 vo (t) 5, w(t) i load (t) ,
x3 (t)


z(t) vo (t)



u(t) dd (t) ,

The state variable x3 is the integral of the error signal obtained

from the difference between the reference Vref and the output
voltage vo . At the equilibrium state, the voltage error is zero.
Therefore x3 is constant. The line voltage is considered a dc
value vg Vg . The disturbance vector w has been dened as
an output current source in order to characterise the output
impedance of the converter. Such output impedance
describes the output voltage vo behaviour in presence of
changes in the output current iload . The output z is the
output voltage vo and represents the controlled output
whose response has to full the control requirements. The
matrices of the state-space representation are as follows
2

D0
 d
L
1

RC
1

(2)

where Dd is the operating point duty cycle and its

complementary corresponds to Dd0 1  Dd . The remaining
state-space matrices are

Cz 0


1 0 ,

Dzw [0],

Dzu [0]

(3)

We consider that the load R and the duty-cycle Dd0 at the

operating point are uncertain or time-varying parameters. We
also consider that all other parameters are well known. It is
worth to point out that the same procedure can be used to
take into account more uncertain terms. Nevertheless, the
more uncertainty is considered in the converter, the lower
performance level can be assured. Thus, in order to deal with
IET Power Electron., 2010, Vol. 3, Iss. 1, pp. 75 85
doi: 10.1049/iet-pel.2008.0271

(5)

The admissible values of vector p are constrained in an

hyperrectangle in the parameter space RN with L 2N
vertices {v1 , . . . , vL }. The images of the matrix [A(p), Bu (p)]
for each vertex vi corresponds to a set {G1 , . . . , GL }. The
components of the set {G1 , . . . , GL } are the extrema of a
convex polytope, noted Co{G1 , . . . , GL }, which contains the
images for all admissible values of p if the matrix
[A(p), Bu (p)] depends linearly on p, that is
[A(p), Bu (p)] [ Co{G1 , . . . , GL }
(
)
L
L
X
X
:
li Gi , li  0,
li 1
i1

(6)

i1

For an in-deep explanation of polytopic models of uncertainty

see [19 (Ch. 2), 23, 24 (Ch. 4)].
Since the boost converter matrices A and Bu do not depend
linearly on the uncertain parameters D d0 and R, we dene two
new uncertain variables d 1=Dd0 and b 1=(Dd02 R) in
order to meet with a linear dependence.

Thus, the
parameter vector is dened as p R, Dd0 , d, b . By using
this parameter vector we can bound the uncertainty inside a
convex polytope.

07
0
7
17
7, B w 6
4  5,
7
C
05
0
0
3
0
Vg
2 3
6 D0 L 7
0
6
7
d
6
6 7
7
Bu 6
Vg 7, Bref 4 0 5
6
7
4 (D02 R)C 5
1
d
0

6 0
6 0
A6
6 Dd
4C
0
2

(4)

Based on the uncertainty model described above, the

synthesis objective is to nd a state-feedback gain K
(u Kx), where uncertainty is restricted inside the
following intervals
R [ [Rmin , Rmax ]
0
0
, Ddmax
]
Dd0 [ [Ddmin
0
0
d [ [1=Ddmax
, 1=Ddmin
]

(7)

02
02
b [ [1=(Ddmax
Rmax ), 1=(Ddmin
Rmin )]

Note that the uncertain model is inside a polytopic domain

formed by L 24 vertices.
The introduction of these new parameters d and b implies
a relaxation in the uncertainty restrictions, since we assume
the independence between uncertain parameters in order to
have linear relations. Consequently, the new relaxed model
77

& The Institution of Engineering and Technology 2009

www.ietdl.org
considers dynamic responses that do not correspond to any
real case. Therefore we will obtain a potentially conservative
solution.
The objectives of the design are to guarantee the stability,
to assure a minimum level of perturbation rejection and to
constraint the transient performance, for all the possible
cases [A( p), Bu( p)]. In next section, we will establish the
LMI conditions which satisfy these objectives.

3.2 Quadratic stability LMIs

The following theorem [23] adapts the quadratic stability
inequality (11) for a closed-loop system with a state feedback
u Kx.

Theorem 3.1: System (1) is stabilisable by state feedback

u K x if and only if there exists a symmetric matrix
W [ Rnn and a matrix Y [ Rmn such that


LMI design constraints

This section introduces the concept of LMI and presents the

constraints used in the controller synthesis problem.

3.1 Lyapunov-based stability

The use of matrix inequalities to demonstrate certain
properties of dynamical systems can already be found in
about 1890, when Lyapunov establishes his well-known
stability method, whose linear case has been reproduced
here for completeness [24].
Given a linear time-invariant (LTI) system
(8)

x_ Ax
the existence of a quadratic function of the form
V (x) x0 Px . 0,

8x = 0

(9)

that satises V_ (x) , 0 is a necessary and sufcient condition

to assure that the system is stable (i.e. all trajectories converge
to zero) [24]. Since V(x) has quadratic form, this condition is
referred as quadratic stability. In this case, the condition
V_ (x) , 0 can be rewritten as follows
V_ (x) x0 (A 0 P PA)x , 0,

8x = 0

s.t. A 0 P PA , 0

(11)

In this case, P is the matrix variable to be found to prove the

stability. Ref. [25] showed that these inequalities that
presented linear dependence on the variables, afterwards
called LMIs, can be solved by convex optimisation methods.
We will take advantage of such convex optimisation
methods, that have been implemented in computer
algorithms [19 21], in order to solve the LMIs that arise in
the control of the boost converter. In the following
subsections we introduce the LMIs to ensure the quadratic
stability, to deal appropriately with disturbances and to
constraint the pole placement of the closed-loop system.
78

& The Institution of Engineering and Technology 2009

(12)

a controller for such state feedback is given by K YW 1 .

The decomposition of K in matrix variables Y and W
allows to satisfy the linearity condition of these inequalities.
A detailed proof can be obtained in [23]. The case with
polytopic uncertainty in matrices A and Bu directly extends
by computing (12) at all the vertices {G1 , . . . , GL } of the
convex polytope Co{G1 , . . . , GL } [24, Ch. 5]. This
extension allows us to assure the quadratic stability of a
uncertain plant.

3.3 H1 control LMIs

The H1 norm of a stable scalar transfer function f (s) is the
peak value of jf (j v)j as a function of frequency [26]. It is
used as a measure of the performance of a system, for
example, to evaluate the minimum attenuation level of an
external disturbance. Considering the transfer function H(s)
from disturbances w to outputs z, the H1 norm of such
system is equal to
kzk2
kwk
2
v(t)=0

kH (s)k1 W sup

(13)

(10)

Thus, the system is stable if and only if there exists a symmetric

matrix P that is positive denite (in the following, the notation
P . 0 means that the matrix P is positive denite) for which
V_ (x) , 0. Inequality (10) is satised if and only if the term
(A 0 P PA) is negative denite, that is
9P . 0

W .0
AW WA 0 Bu Y Y 0 B0u , 0

where k  k1 and k  k2 stand for the innity and the Euclidian

norms, respectively.
The following theorem, adapted from [27], guarantees a
maximum H1 norm g (i.e. a minimum level of disturbance
attenuation).

Theorem 3.2: System (1) is stabilisable by state-feedback

u K x and kzk2 =kwk2 , g if and only if there exists a
symmetric denite positive matrix W [ Rnn and a matrix
Y [ Rmn such that the following inequality holds
2
6
4

AW WA 0 Bu Y Y 0 B0u

Bw

W Cy0 Y 0 D0yu

B0w
C y W Dyu Y

g1
0

0
g 1

3
7
5,0
(14)

a controller for such state feedback is given by K YW21.

IET Power Electron., 2010, Vol. 3, Iss. 1, pp. 75 85
doi: 10.1049/iet-pel.2008.0271

www.ietdl.org
A proof is given in [27]. Note that satisfaction of inequality
(14) implies satisfaction of (12) and therefore this theorem
also ensures quadratic stability. The polytopic case is
directly
applicable
by satisfying (14) for all the vertices


G1 , . . . , GL of the polytopic domain.

3.4 Pole placement LMIs

It is a desirable property of the closed-loop system that its
poles are located in a certain region of the complex plane,
in order to assure some dynamical properties like overshoot
and settling time. In [28], a region of the complex plane
S(a, r, u), depicted in Fig. 2, is dened such that the
(complex) poles of the system in the form x + jy satisfy
x , a , 0,

jx + jyj , r,

y , cot (u)x

(15)

In such a case the poles of the system x + jy zvn + j vd

ensure a certain damping at the desired rate (see [29]). The
presented region is equivalent to a minimum decay rate a,
a minimum damping ratio z . sin u and a maximum
vd , r cos u,
where
damped p
natural
frequency

v d v n 1  z2 .
The following theorem allows to constraint the location of
the closed-loop poles in the region S(a, r, u).

Theorem 3.3: The closed-loop poles of the system (1) with

a state-feedback u Kx are inside the region S(a, r, u) if
there exists a symmetric denite positive matrix W and a
matrix Y such that
AW WA 0 Bu Y Y 0 B0u 2aW , 0


rW
WA 0 Y 0 B0u
,0
AW Bu Y
rW

cos u(AW WA 0 Bu Y Y 0 B0u )

sin u(AW WA 0  Bu Y Y 0 B0u )

sin u(AW  WA 0 Bu Y  Y 0 B0u )
,0
cos u(AW WA 0 Bu Y Y 0 B0u )
and K YW 21 is the state feedback gain.

Figure 2 LMI region S(a, r, u)

IET Power Electron., 2010, Vol. 3, Iss. 1, pp. 75 85
doi: 10.1049/iet-pel.2008.0271

(16)

A detailed proof is reported in [28], where the pole

placement in generic regions of the complex plane is
explored. A different point of view on robust pole placement
is given in [30]. Once again, the polytopic case directly

extends by satisfying (16) (18) for all G1 , . . . , GL .
Therefore by taking into account these
 conditions
 in each
vertex of the convex polytope Co G1 , . . . , GL , we will
ensure a proper performance despite uncertainty.

3.5 Remarks
We summarise the LMI synthesis method as follows. The
design of a robust control for a boost converter consists of
solving inequalities (14) and (16)(18) to nd the matrix
variables Y and W that minimise the H1 norm g and that
satisfy the pole placement constraints of the
 region S(a, r, u)
for all the extrema of the polytopic model G1 , . . . , GL
min g
Y ,W

under conditions
(14),
 (16), (17) and (18)
8 Gi , i 1, . . . , L

(19)

It is necessary to remark that the stability and the H1 bound of

the closed-loop system is guaranteed for arbitrarily fast changes
in Dd0 and R. However, the pole placement constraint is only
satised if the time-dependent parameters changes are slowly
enough to recover the steady state of the system. For a survey
on how the rate change of parameters can be taken into
account, see [17].
In the next section, we develop in detail the given
procedure for a specied boost converter.

(17)

4 Control implementation and

experimental results
(18)

In this section, we present the results of the LMI design

method proposed above. First, we identify the parameter
values of the boost converter. Next, the state-feedback gain
K is numerically found. The resulting controller has been
simulated with a switched model of the converter by using
PSIM software. We have compared the simulated response
of the proposed controller with a conventional PID
controller, in order to evaluate the robustness and
performance achieved with our approach. Finally, the
simulation results have been veried with an experimental
set-up that is explained in detail at the end of this section.
The parameters of the dc dc converter, that have been
taken from [13], can be found in Table 1. The nominal
values of the uncertain converter parameters are R 50 V
and D d0 0.5. Note that since the output voltage vo is
considered constant, an uncertain operation point Dd0 in the
interval [Dd0 min , Dd0 max ] introduces a range of possible line
0
0
voltages Vg [
 [vo  Ddmin ,vo  Ddmax ] that are included in the
polytope Co G1 , . . . , GL dened in (6).
79

& The Institution of Engineering and Technology 2009

www.ietdl.org
Table 1 Boost converter parameters
Parameter

Value

[10, 50] V

D d0

[0.3, 1]

vo , Vref

24 V

600 mF

310 mH

Ts

5 ms

Given the parameters of Table 1, the uncertain variables d

and b, dened in (7), belong to the intervals [1, 3.33] and
[0.02, 1.11], respectively.

4.1 Controller synthesis

As it has been stated before the synthesis objective is to
minimise g while a pole placement constraint S(a, r, u) is
satised. The values of (a, r, u) can be found in Table 2.
In order to have the poles of the system inside the valid
frequency range of the model, r has been set to 1/10 of
the switching frequency. For a minimum damping ratio of
Table 2 Pole placement parameters
Parameter Value

a
r

130
2p
10Ts
258

0.4, u has been set to 258. Finally, for a fast decay rate, we
have tested several different values of a. For the present
case, the parameter a has been set to 130; a higher value of
a results in an unfeasible problem for this parameter set.
Solving problem (19) using Matlabs standard LMI
toolbox [19], which yields a state-feedback controller K
K [1:95

2:00

725:22]

(20)

and a guaranteed H1 bound from disturbances to outputs of

g 2.89, which is equivalent to 9.21 dB. The corresponding
control law that yields the duty cycle is
dd (t) 1:95iL (t)  2:00vo (t) 725:22x3 (t)

(21)

4.2 Simulations
The switched model of Fig. 1a with the controller K has been
implemented using PSIM simulator [31], taking the nominal
values of the converter.
In order to show the robustness of the proposed controller,
we have simulated the transient behaviour in presence of an
abrupt load change under nominal conditions and out of
nominal conditions. Fig. 3a depicts the output voltage
when the converter operates at the nominal equilibrium
point and reacts in front of a load step change of 0.5 A. It
can be noted that the output voltage response presents a
time constant of approximately 10 ms, that corresponds to
a decay rate of 400, which as expected is larger than the
minimum guaranteed decay rate (a 130). Furthermore,
the output voltage exhibits a small overshoot that agrees
with the damping ratio restriction. Fig. 3b shows the
simulation out of the nominal condition, without any
change in the controller parameters. The duty cycle at the
new equilibrium point is D d0 0.3, which corresponds to a
Vg equal to 7.2 V. Again, we simulate the regulator

Figure 3 PSIM simulated response for a loading step of 0.5 A with the nominal load R 50 V
a Under nominal duty cycle D d0 0.5
b Out of nominal duty cycle D d0 0.3

80

& The Institution of Engineering and Technology 2009

IET Power Electron., 2010, Vol. 3, Iss. 1, pp. 75 85

doi: 10.1049/iet-pel.2008.0271

www.ietdl.org

Figure 4 PSIM simulated response for a loading step of 0.5 A with the nonnominal load R 10 V
a Under nominal duty cycle D d0 0.5
b Out of nominal duty cycle D d0 0.3

response in front of a step load perturbation of 0.5 A. It can

be observed that despite the new operation point, the
proposed controller maintains a small overshoot of the
output voltage. Although this overshoot is larger than in
the nominal case, it is yet inside of the specied damping
ratio restriction. Besides, the decay rate is approximately
200 which is smaller than in the nominal case but larger
than the specied a. Fig. 4 shows the same simulations of
Fig. 3 with a different non-nominal load of R 10 V. The
simulations show that the LMI controller satises the
damping ratio specication and also the decay rate is better
than the specied a.
In order to contrast the performance and robustness of the
proposed regulator, we have also simulated the boost
converter with a conventional PID controller, whose control
law at low frequencies is given in (22). This comparison has
been made with the aim to show the benets of using a robust
control method compared with a conventional approach.
Obviously, this comparison could have been done with
another state-feedback or current-mode controller.
Nevertheless, the robust method would keep, to a great
extent, its advantages over a design disregarding uncertainty.

130, which are similar to those of the LMI controller. To

select such PID controller we have used sisotool of Matlab.
Vg (R  (L=Dd02 )s)
vo (s)

u(s)
CLRs2 Ls Dd02 R
PID(s)

0:0000447531s2 0:0209432186s 31:25

s((s=1  105 ) 1)((s=1  105 ) 1)

(23)
(24)

Fig. 3a depicts the transient simulation of the converter with the

PID controller for a load step change of 0.5 A in nominal
conditions. When compared with the LMI simulation, it can
be observed that the response of the PID controllers is better
in terms of decay rate and perturbation rejection, but it is
slightly worse in terms of damping ratio, since they have been
built using the concepts of phase and gain margin. However it
is important to remark that all the controllers present
adequate damping properties at the nominal operation point.

dd (t) Ki (Vref  vo (t)) Kp (Vref  vo (t))

Kd

d(Vref  vo (t))
dt

(22)

Equation (23) shows the control-to-output transfer function of

the boost converter, whereas (24) is the transfer function of the
PID controller, which corresponds with a PID controller with
Ki 31.25, Kp 0.0209432186 and Kd 0.0000447531.
This PID controller has been designed at the nominal model
of the converter. We impose as PID specications a phase
margin greater than 608 and a gain margin greater than
20 dB, which are common specications in power electronics
control. We also impose that the closed-loop poles present a
damping ratio greater than 0.4 and a decay rate greater than
IET Power Electron., 2010, Vol. 3, Iss. 1, pp. 75 85
doi: 10.1049/iet-pel.2008.0271

Figure 5 PSIM simulated output impedance of the converter

with the LMI controller. The simulation was made with a
current sink of 200 mA at nominal duty cycle D d0 0.5
81

& The Institution of Engineering and Technology 2009

www.ietdl.org
Out of these nominal conditions, in Fig. 3b, it is shown the
time-response simulation in presence of the same
perturbation of 0.5 A when the converter operates at a duty
cycle equal to Dd0 0.3. Again, we have maintained the
previous controller parameters. Since the PID controller was
not designed taking into account the uncertainty of the
converter, the regulator with PID controller loses its damping

properties and exhibits poor response when the duty cycle Dd0
decreases far from the nominal value. The simulations with
the non-nominal load R 10 V with the PID controller are
shown in Fig. 4. While in the nominal duty cycle Dd0 0:5
the behaviour is close to the case with the nominal load
R 50 V, when the duty cycle is out of its nominal value
Dd0 0:3, the output voltage behaviour is worse than the

Figure 6 Implementation diagram of a boost converter with the proposed LMI state-feedback regulation

Figure 7 Detail of the circuit implementation of the LMI state-feedback controller and the PWM regulator
82

& The Institution of Engineering and Technology 2009

IET Power Electron., 2010, Vol. 3, Iss. 1, pp. 75 85

doi: 10.1049/iet-pel.2008.0271

www.ietdl.org
LMI controller in terms of perturbation rejection, decay rate and
damping ratio.
Besides the transient response characteristics, we have
simulated the steady-state output impedance of the regulator
with the LMI controller. The frequency response for a
perturbation iload 200 mA around the nominal duty cycle
D d0 0.5 has been depicted in Fig. 5 for the cases of the
nominal load R 50 V and a non-nominal load R 10 V.
The output impedance gain peak with the LMI controller is
approximately 7 dB, which agrees with the H1 norm
specication bound g 9.21 dB found in the LMI synthesis.

4.3 Experimental verication

To demonstrate the advantages of the proposed control
scheme, an experimental prototype of a boost converter has
been implemented. The structure of the converter with the
state-feedback controller is shown in Fig. 6. The statefeedback controller, whose electric diagram is also given in
Fig. 7, only requires a small number of operational
ampliers and discrete components.

Figure 9 Experimental response for a loading step of 0.5 A

out of nominal duty cycle D d0 0.3. Upper traces are output
voltage. Lower traces are output current

Fig. 8 shows the prototype transient response under a

loading step of 0.5 A, as it was carried out in the
simulations. The slew rate of the current step is greater than
150 A/ms. For this gure, the converter was working at the
nominal operation point. The transient closely resembles the
simulation in Fig. 3. The robustness of the controller is
veried in Fig. 9, where the loading step is applied out of
the nominal conditions (D d0 0.3). Once again, the
experimental waveform agrees with the simulation in Fig. 4.
Besides the transient experiments, the output impedance of
the regulator has been measured with the help of a frequency

Figure 10 Experimental output impedance of the converter

with the LMI controller. The measurement was made with a
current sink of 200 mA at nominal duty cycle D d0 0.5
analyser connected to a voltage-controlled current sink of
200 mA. The measurements have been carried out for
R 10 and 50 V, at the nominal duty cycle Dd0 0.5. The
measurements, depicted in Fig. 10, show a perfect agreement
with the simulation results and demonstrate that despite of
uncertainty, the LMI controller fulls the objectives marked
in the synthesis procedure.

Figure 8 Experimental response for a loading step of 0.5 A

under nominal duty cycle D d0 0.5. Upper traces are output
voltage. Lower traces are output current
IET Power Electron., 2010, Vol. 3, Iss. 1, pp. 75 85
doi: 10.1049/iet-pel.2008.0271

Conclusions

This work has presented a robust controller design framework

based on LMIs for switched-mode dc dc converters. The
83

& The Institution of Engineering and Technology 2009

www.ietdl.org
LMI design method has been applied to a boost converter and
the results have been compared with a conventional PID
controller by means of both numerical simulations. Also the
experimental measurements from a laboratory prototype of a
boost converter have veried the results of the LMI
controller. The LMI control circuit implementation consists
of a standard current sensor, some OPAMPs and a PWM
device. We have shown that the state-feedback LMI
approach is a valid synthesis method for non-minimum
phase converters. With the proposed method, transient and
steady-state performances can be taken into account and the
resulting design has, in this case, good performance
characteristics, despite the conservatism in the uncertainty
model. The main advantage of this approach is that the
state-feedback controller can be synthesised automatically,
differently from other robust control methods, for which the
controller must be synthesised manually or using CAD
tools. Since the synthesis is carried out by means of a
Matlabs LMI toolbox, the method can be readily extensible
in order to consider parasitic resistances in the inductor and
capacitor by modifying the state-space matrices of the
model. Finally, experimental verications show a perfect
agreement with the design constraints despite uncertainty.
Future work will deal with the application of the method to
more complex power converters such as high-order circuits,
multiphase and multilevel converters.

Acknowledgment

This work was partially supported by the Spanish Ministerio

de Educacion y Ciencia under grants TEC2004-05608C02-02 and TEC2007-67988-C02-02.

References

[1] SANDERS S.R., VERGHESE G.C.: Lyapunov-based control for

switched power converters, IEEE Trans. Power Electron.,
1992, 7, (1), pp. 17 24
[2] KAWASAKI N., NOMURA H., MASUHIRO M.: A new control law of
bilinear DC-DC converters developed by direct application of
Lyapunov, IEEE Trans. Power Electron., 1995, 10, (3),
pp. 318325
[3] CORTES D., ALVAREZ J., FRADKOV A.: Tracking control of the
boost converter, IEE Proc. Control Theory Appl., 2004,
151, (2), pp. 218 224
[4] HE Y. , LUO F.L.: Sliding-mode control for dc dc
converters with constant switching frequency, IEE Proc.
Control Theory Appl., 2006, 153, (1), pp. 37 45
[5]

LEYVA R., CID-PASTOR A., ALONSO C., QUEINNEC I., TARBOURIECH S.,

MARTINEZ-SALAMERO L.:

Passivity-based integral control of a

boost converter for large-signal stability, IEE Proc. Control
Theory Appl., 2006, 153, (2), pp. 139 146
84

& The Institution of Engineering and Technology 2009

[6] GARCERA G., FIGUERES E., PASCUAL M., BENAVENT J.M.A.: Analysis
and design of a robust average current mode control loop
for parallel buck DC DC converters to reduce line and load
disturbance, IEE Proc. Electr. Power Appl., 2004, 151, (4),
pp. 414 424
[7] SUNTIO T. , HANKANIEMI M., KARPPANEN M.: Analysing the
dynamics of regulated converters, IEE Proc. Electr. Power
Appl., 2006, 153, (6), pp. 905 910
[8]

MORALES-SALDANA J.A. , GALARZA-QUIRINO R. , LEYVA-RAMOS J. ,

E.E.A. , ORTIZ-LOPEZ M.G.A. : Multiloop

controller design for a quadratic boost converter, IET
Electr. Power Appl., 2007, 1, (3), pp. 362 367
CARBAJAL-GUTIERREZ

[9] NAIM R., WEISS G., BEN-YAAKOV S.: H1 control applied to

boost power converters, IEEE Trans. Power Electron.,
1997, 12, (4), pp. 677 683
[10] TZANN-SHIN L. , CHIANG S.J., JHY-MING C.: H 1 loopshaping controller designs for the single-phase UPS
inverters, IEEE Trans. Power Electron., 2001, 16, (4),
pp. 473 481
[11] VIDAL-IDIARTE E., MARTINEZ-SALAMERO L., CALVENTE J., ROMERO A.:
An H1 control strategy for switching converters in slidingmode current control, IEEE Trans. Power Electron., 2006,
21, (2), pp. 553 556
[12] WALLIS G.F. , TYMERSKI R.: Generalized approach for
m synthesis of robust switching regulators, IEEE
Trans. Aerosp. Electron. Syst., 2000, 36, (2), pp. 422 431
[13] OLALLA C., EL AROUDI A. , LEYVA R. : Ana lisis y diseno de
convertidores cc-cc elevadores mediante control QFT.
Seminario Anual de Electronica y Automatica Industrial,
SAAEI06, 2006
[14] OLALLA C. , LEYVA R., EL AROUDI A. : QFT design for
current-mode PWM buck converters operating in
continuous and discontinuous conduction modes. IEEE
Annual Conf. Industrial Electronics, IECON06, 2006,
pp. 1828 1833
[15] MONTAGNER V.F., PERES P.L.D.: H1 control with pole location
for a DC DC converter with a switched load. Industrial
Electronics, 2003, ISIE 03 2003 IEEE Int. Symp., 2003,
vol. 1, pp. 550 555
[16] MONTAGNER V.F., PERES P.L.D.: Robust pole location for a
DC DC converter through parameter dependent control.
Proc. 2003 Int. Symp. Circuits and Systems, ISCAS 03,
2003, vol. 3, pp. 351 354
[17] MONTAGNER V.F., OLIVEIRA R.C.L.F., LEITE V.J.S., PERES P.L.D.: LMI
approach for H1 linear parameter-varying state feedback
control, IEE Proc. Control Theory Appl., 2005, 152, (2),
pp. 195 201
IET Power Electron., 2010, Vol. 3, Iss. 1, pp. 75 85
doi: 10.1049/iet-pel.2008.0271

www.ietdl.org
[18] BIN H., MINGGAO Y.: Robust LPV control of diesel auxiliary
power unit for series hybrid electric vehicles, IEEE Trans.
Power Electron., 2006, 21, (3), pp. 791 798
[19] GAHINET P., NEMIROVSKI A., LAUB A.J., CHILALI M.: LMI control
toolbox for use with matlab (The MathWorks, Inc., 1995)
[20] STURM J.: Using SeDuMi 1.02, a MATLAB toolbox for
optimization over symmetric cones, Optim. Methods
Softw., 1999, 11 12, pp. 625 653
[21] LO FBERG J.: YALMIP : A toolbox for modeling and
optimization in MATLAB. Proc. CACSD Conf. 2004,
available at http://control.ee.ethz.ch/~joloef/yalmip.php
[22] MIN C., JIAN S.: Reduced-order averaged modeling of
active-clamp converters, IEEE Trans. Power Electron.,
2006, 21, (2), pp. 487 494

[25] PYATNITSKII E.S., SKORODINSKII V.I.: Numerical methods of

Lyapunov function construction and their application to
the absolute stability problem, Syst. Control Lett., 1982,
2, (2), pp. 130 135
[26] SKOGESTAD S., POSTLETHWAITE I.: Multivariable feedback
control: analysis and design (John Wiley and Sons, 1996)
[27] GAHINET P., APKARIAN P.: A linear matrix inequality
approach to H1 control, Int. J. Robust Nonlinear Control,
1994, 4, (4), pp. 421 448
[28] CHILALI M., GAHINET P.: H1 design with pole placement
constraints: an LMI approach, IEEE Trans. Autom. Control,
1996, 41, (3), pp. 358 367
[29] HADDAD W.M., BERNSTEIN D.S. : Controller design with
regional pole constraints, IEEE Trans. Autom. Control,
1992, 37, (1), pp. 54 69

[23] BERNUSSOU J., PERES P.L.D. , GEROMEL J.: A linear

programming oriented procedure for quadratic
stabilization of uncertain systems, Syst. Control Lett.,
1989, 13, pp. 65 72

[30] GARCIA G., DAAFOUZ J., BERNUSSOU J.: Output feedback disk
pole assignment for systems with positive real uncertainty,
IEEE Trans. Autom. Control, 1996, 41, (9), pp. 1385 1391

[24] BOYD S., EL GHAOUI L., FERON E., BALAKRISHNAN V.: Linear matrix
lnequalities in system and control theory (SIAM, 1994)

[31] Psim 6.0. 2003: 35 Farwood Drive, Andover, MA

01810, USA, available at http://www.powersimtech.com

IET Power Electron., 2010, Vol. 3, Iss. 1, pp. 75 85

doi: 10.1049/iet-pel.2008.0271

85

& The Institution of Engineering and Technology 2009