Sliding mode- modulation Control of a Buck converter

H. Sira-Ramrez
Departamento de Ing. El ectrica
Secci on de Mecatr onica
Cinvestav-IPN
M exico D.F., M exico
AbstractA recently proposed approach that sets sliding
mode control of linear systems within a classical input-output
control framework (Sira-Ramrez [5]), is here used for the
input-output sliding mode stabilization of a buck DC-to-
DC Power Converter. The approach evades state measurements
and circumvents the explicit use of asymptotic state observers
in the sliding surface synthesis. The proposed scheme is also
robust with respect to unmatched state perturbation inputs. A
connection between sliding modes and classical -modulators
and -modulators is also brought to attention as a tool
for the realization of sliding mode control schemes in systems
commanded by a switch position where a desirable average
feedback controller design needs to be implemented.
I. INTRODUCTION
The many advantages of sliding mode control are well
reported, founded, and illustrated, in the existing literature.
The sliding mode control technique is, fundamentally, a state
space-based discontinuous feedback control technique. The
lack of complete knowledge of the state vector components
forces the designer to use asymptotic state observers, of the
Luenberger, or of the sliding mode type, or perhaps to resort
to direct output feedback control schemes. Unfortunately,
the rst approach is not robust with respect to unforseen
exogenous perturbation inputs, even if they happen to be
of the classical type (by this we mean: steps, ramps,
parabolas, etc). The second approach is quite limited in
nature and it is not applicable in a host of non-minimum
phase systems. Generally speaking, state space based sliding
mode techniques fail in the unmatched perturbation input
case. For general background on sliding mode control, we
refer the reader to the seminal book by Utkin, [7], the
recent books by Utkin, Guldner and Shi [8] and that by
Edwards and Spurgeon [1]. Recent developments, advances
and applications, of the sliding mode control area are found
in the book by Perruquetti and Barbot [3].
In this article, we propose a new approach for the synthesis
of sliding mode feedback control schemes for linear, time
invariant, controllable and observable Single Input Single
Output (SISO) systems, for which an average controller
design is assumed to be available. We show that the use of
classical -modulators
1
allows for the switched synthesis
of a feedback controller which has been synthesized within
an average context (i.e. assuming that the control input
1
A complete account of -modulators, extensively used in analog signal
encoding, which never beneted from the theoretical basis of sliding mode
control, is found in the classical book by Steele [6].
continuously takes values on a closed subset of the real
line, usually restricted to be the closed interval [0, 1]). We
show that a -modulator can be used to translate such a
continuous design into a discontinuous one with the property
that the equivalent output signal of the modulator, in an
ideal sliding mode sense, precisely matches the modulators
input signal.
When we combine -modulation with integral recon-
structors [2] of the system state vector and Generalized
Proportional Integral (GPI) control, the result is that the
required sliding motion is dynamically synthesized using
only the input and the output of the system while retaining
the essential robustness features of the average devised GPI
controller.
Hence, our approach is, fundamentally, an input-output
approach which emphasizes the synthesis of an adequate
continuous (average) feedback law, rather than the synthesis
of a sliding surface. The use of GPI controllers in the average
feedback controller design naturally leads to dynamic input-
output feedback schemes for the synthesis of the sliding
motion. As an outcome, the scheme here presented requires
no matching conditions whatsoever.
Section 2 presents a review of the simplest analog -
modulator and its connection with sliding mode control
schemes when the actual system input signal takes values
in a discrete set of the ON-OFF form, i.e. in the discrete
set {0, 1}. Section 3 deals with some generalities on how
to synthesize a sliding mode controller on the basis of
a given continuous feedback controller design. Section 4
concentrates on a direct application of -modulators in the
implementation of a continuous Generalized PI controller
for a buck converter model. Section 5 deals with the
conclusions of the article.
II. -MODULATORS
Consider the basic block diagram of Figure 1 reminiscent
of a -modulator block but with a binary valued forward
nonlinearity, taking values in the discrete set {0, 1}. For
ease of reference, we address such a block simply as a -
modulator. The following theorem summarizes the relation
of the depicted -modulator with sliding mode control while
establishing the basic features of its input output perfor-
mance.
Theorem 2.1: Consider the -modulator of Figure 1.
Given a continuously differentiable bounded, C
1
[t
0
, ),
Proceedings of the 42nd IEEE
Conference on Decision and Control
Maui, Hawaii USA, December 2003 WeP14-6
0-7803-7924-1/03/$17.00 2003 IEEE 2999
signal (t), then the locally decoded feedback signal x(t)
satises the convergence property: x(t) (t), in a nite
amount of time t
h
> t
0
, from any arbitrary initial value of the
tracking, or encoding, error e(t
0
) = x(t
0
) (t
0
), provided
the following encoding condition is satised for all t,
0 <

(t) < 1 (2.1)
Moreover a sliding motion exists on the perfect encoding
condition e = 0 for all t > t
h
, where the quantity t
h
is
satises
t
h
t
0
+
2| e(t
0
) |
m(t)
(2.2)
m(t) = (1 sup{

(t)})(1 + sign(e(t
0
)))
+(inf

(t))(1 sign(e(t
0
)))
Proof. From the gure, the variables in the -modulator
satisfy the following relations:
x = u, u =
1
2
[1 + sign( x)]
e = x (2.3)
Clearly, e =

(t)
1
2
[1 + sign (e)] and since

(t) is assumed
to be bounded within the interval [0, 1], we have:
e e = e

(t)
1
2
e
1
2
| e |
=
1
2
| e |
_
1
_
2

(t) 1
_
sign(e)
_
< 0
A sliding regime exists on e = 0 for all time t after the
hitting time t
h
(see [7]). Under ideal sliding, or encoding,
conditions, e = 0, e = 0, we have that x = (t) and
the equivalent (average) value of the coded output signal
u is given by u
eq
=

(t) for all t t
h
. The estimation of
the hitting time t
h
is based on the integrable error equation
obtained for the most unfavorable case of the input

in the
error equation for each possible sign of the initial condition.
Thus, if e(t
0
) > 0 then we consider as the error dynamics:
e = sup{

} 1, which yields at the hitting time t

h
,
e(t
h
) = e(t
0
) (1 sup{

(t)})(t
h
t
0
) = 0
t
h
= t
0
+
e(t
0
)
1 sup{

(t)}
While if e(t
0
) < 0, then we consider as the error dynamics
e = inf{

}. This yields, at the hitting time,

e(t
h
) = e(t
0
) + inf{

}(t
h
t
0
) = 0
t
h
= t
0

e(t
0
)
inf{

(t)}
= t
0
+
|e(t
0
)|
inf{

(t)}
The result (2.2) follows by combining these two worst case
estimates for the hitting time into a single formula which
takes into account the sign of the initial condition e(t
0
).

The -modulator output u ideally yields the time differ-

entiation of the modulator input signal (t) in an equivalent
control sense ([7]), i.e. in an average sense. The role of
the above described -modulator in sliding mode control
schemes, avoiding full state measurements, and using average
based designed controllers will be clear from the develop-
ments presented below.
III. IMPLEMENTING AN AVERAGE CONTROLLER
THROUGH -MODULATION
Suppose we have a smooth nonlinear system of the form
x = f(x) + ug(x) with u being a (continuous) control
input signal that, for some physical limitations, requires to
be bounded by the closed interval [0, 1]. Suppose, further-
more, that we have been able to specify a state feedback
controller of the form u = k(x), or a dynamic output
feedback controller of the form u = (y, ),

= (y, ),
with desirable closed loop performance features. Assume,
furthermore, that for any initial state of the system (and of
the dynamic controller, if such is the case), the values of the
feedback signal function, u(t), are uniformly strictly bounded
by the closed interval [0, 1].
If an additional implementation requirement entitles now
that the control input u of the system is no longer allowed
to continuously take values within the interval [0, 1], but
that it may only take values in the discrete set, {0, 1}, the
natural question is: how can we now implement the derived
continuous controller, so that we can recover, possibly in an
average sense, the desirable features of the derived static,
or dynamic, feedback controller design in view of the last
imposed actuator restriction?
The answer is clearly given by the key feature, implying
average differentiation of the input signal, in the previously
considered -modulator. Recall, incidentally, that the output
signal of such a modulator is restricted to take values,
precisely, in the discrete set {0, 1}. Thus, if the time integral
of the output of the designed continuous controller, call it
u
av
(t), is feed into the proposed -modulator, the output
signal of the -modulator reproduces, on the average, the
required control signal u
av
but delivers it in the required
form of a switched signal. Figure 3 shows the switch based
implementation of an output feedback controller, through a
-modulator, which reproduces, in an average sense, the
features of a designed continuous controller.
Note that the presence of an integral in the feedback
loop of the -modulator and the requirement of the integral
of the designed average control signal at the input of the
modulator, produces an equivalent modulator entitling a
single integration placed before the sign function. See Figure
4. This modulator is known as the --modulator.
We have the following equations for the - modulator
when the input signal is to be reproduced at the output of
the modulator in an average, equivalent control, sense.
e = u, u =
1
2
[1 + sign z]
z = e (3.1)
3000
We have the following proposition.
Theorem 3.1: Consider the --modulator of Figure 4.
Given a bounded signal (t), then the integral tracking error
z converges to zero in a nite amount of time t
h
, from any
initial value z(t
0
) at time t
0
, provided the following encoding
condition is satised for all t,
0 < (t) < 1 (3.2)
Moreover, a sliding motion exists on the perfect encoding
condition z = 0 for all t > t
h
, where t
h
satises,
t
h
t
0
+
2| z(t
0
) |
m(t)
(3.3)
m(t) = (1 sup{ (t)})(1 + sign(z(t
0
)))
+(inf (t))(1 sign(z(t
0
)))
Proof
From the equations (3.1), we have
dz
dt
= (t)
1
2
[1 + sign (z)] (3.4)
and therefore
z z =
1
2
| z | [1 (2(t) 1)sign (z)] (3.5)
Clearly, for (t) being a bounded signal in the open interval
(0, 1), it follows that z z < 0, and a sliding regime exists
on z = 0. Under ideal sliding motions, the conditions:
z = 0, dz/dt = 0 are valid. This implies that, on the average,
(t) = u(t), i.e. the output of the modied modulator
reproduces, in an equivalent control sense, the input to the
modulator. The nite time reachability of the sliding surface
z = 0 is established in a similar manner as in the previous
theorem.

In view of the previous result, we have the following

general result concerning the control of nonlinear systems
through sliding modes synthesized on the basis of an average
feedback controller and a --modulator. We only deal with
the dynamic output feedback controller case.
Theorem 3.2: Consider the following smooth nonlinear
single input, n-dimensional system: x = f(x) +ug(x), with
the smooth scalar output map, y = h(x). Assume the dy-
namic smooth output feedback controller u = (y, ),

=
(y, ), with R
p
, locally (globally) asymptotically
stabilizes the system to a desired constant equilibrium state,
denoted by X. Assume, furthermore, that the control signal,
u, is uniformly strictly bounded by the closed interval [0, 1]
of the real line. Then the closed loop system:
x = f(x) + ug(x), y = h(x)
u
av
(y, ) = (y, )

= (y, ), u =
1
2
[1 + sign z]
z = u
av
(y, ) u
exhibits an ideal sliding dynamics which is locally (globally)
asymptotically stable to the same constant state equilibrium
point, X, of the system.
The proof of this theorem is immediate upon realizing
that under the hypothesis on the average control input,
u
av
, theorem 3.1 establishes that a sliding regime exists
on the manifold z = 0. Under the invariance conditions,
z = 0, z = 0, which characterize ideal sliding motions, the
corresponding equivalent control, u
eq
, associated, with the
system satises: u
eq
(t) = u
av
(t). The ideal sliding dynamics
is then represented by
x = f(x) + u
av
g(x), y = h(x)
u
av
(y, ) = (y, ),

= (y, )
which is assumed to be locally (globally) asymptotically
stable towards the desired state equilibrium point.

IV. CONTROL OF A BUCK CIRCUIT

A. The buck converter model and a PD controller
Consider the buck converter circuit shown in Figure 2.
The system is described by the set of equations
L

i = v + uE, C v = i
v
R
(4.1)
where i represents the inductor current and v is the output ca-
pacitor voltage. The control input u, representing the switch
position function, is a discrete-valued signal taking values in
the set {0, 1}. The system parameters are constituted by: L,
which is the inductance of the input circuit; C the capacitance
of the output lter and R, the output load resistance. The
external voltage source has the constant value E. We assume
that the circuit is in continuous conduction mode, i.e. the
average value of the inductor current never drops to zero,
due to load variations.
We introduce the following state normalization and time
scale transformation:
x
1
=
i
E
_
L
C
, x
2
=
v
E
, =
t

LC
(4.2)
The normalized model is thus given by:
x
1
= x
2
+ u, x
2
= x
1

x
2
Q
(4.3)
where now, with an abuse of notation, the represents
derivation with respect to the normalized time, . The
variable x
1
is the normalized inductor current, x
2
is the
normalized output voltage and u, still represents the switch
position function. The constant system parameters are all
comprised now in the circuit quality parameter, denoted by
Q and given by the strictly positive quantity, R
_
C/L. We
assume that only the (normalized) output capacitor voltage,
y = x
2
, is available for measurement.
3001
In order to obtain a suitable average controller, assume for
a moment that the normalized buck converter equations ac-
tually represent a continuous system (i.e. an average system)
where u may take values in the closed interval [0, 1]. Take
the normalized average output capacitor voltage, x
2
, as the
system output, i.e. y = x
2
. Elimination of the normalized
average inductor current variable x
1
leads to the following
input-output differential relation,
y +
1
Q
y + y = u (4.4)
Clearly, the system, aside from being a stable system, it also
has no zero dynamics associated with y. Suppose we want
to devise a controller that asymptotically regulates the output
voltage to the desired average value y. Corresponding to
this desired normalized constant equilibrium value, we have
from (4.4) u = y.
Consider then, the following PD stabilizing feedback con-
troller with nominal input compensation,
u = u +
_
1
Q
k
2
_
y + (1 k
1
)(y y) (4.5)
The tracking error, e = y y, is clearly seen to satisfy the
following closed loop dynamics
e
2
+ k
2
e + k
1
e = 0 (4.6)
By appropriate choice of the design parameters k
2
, k
1
, the
origin of the error space coordinate, e, can be made into a
locally asymptotically exponentially stable equilibrium point
determined by the non-saturation condition.
0 < u +
_
1
Q
k
2
_
y + (1 k
1
)(y y) < 1 (4.7)
B. AN AVERAGE GENERALIZED PI (GPI) CONTROL
The PD controller (4.5) requires the time derivative of
the output signal y. A GPI controller can then be proposed
which substitutes the unmeasured signal y by its integral
reconstructor, denoted by ( y)
e
. Such a reconstructor, is
obtained by direct integration of the input output relation
(4.4), while neglecting the unknown initial condition y
0
. We
have,
( y)
e
=
1
Q
y
_
t
0
(y() u) d (4.8)
The use of ( y)
e
in the PD controller, instead of y, produces a
constant error due to the unaccounted relation, y = ( y)
e
+ y
0
.
An integral control corrective action is then added to the
proposed controller (4.5), in order to counteract the consci-
entious neglect of the unknown initial condition. We propose
then the following average dynamic GPI feedback controller,
u = u +
_
1
Q
k
2
_
( y)
e
+ (1 k
1
)(y y)
+k
0
_
t
0
(y() y)d
( y)
e
=
1
Q
y
_
t
0
(y() u) d (4.9)
The closed loop system stabilization error dynamics satises
the following linear integro-differential equation excited by
an unknown constant.
e + k
1
e + k
0
_
t
0
e()d =
_
1
Q
k
2
_
y
0
(4.10)
It is not difcult to verify that such a closed loop dynamics
has the origin as an asymptotically exponentially stable
equilibrium point for a suitable set of design parameters k
2
,
k
1
, k
0
. Indeed, dene a new state by means of
=
_
t
0
e()d
1
k
0
_
1
Q
k
2
_
y
0
(4.11)
The closed loop system is then equivalent to the following
composite linear system,
e + k
1
e + k
0
= 0

= e(), (0) =
1
k
0
_
1
Q
k
2
_
y
0
(4.12)
Clearly, the characteristic polynomial of the closed loop
system is given by
p(s) = s
3
+ k
2
s
2
+ k
1
s + k
0
(4.13)
Figure 5 shows some computer simulations of the closed
loop response of the non normalized converter system vari-
ables v, i, to the actions of the proposed average GPI
controller. The controller design parameters k
2
, k
1
, k
0
, were
chosen so that the average control input u
av
never saturates
to values outside the interval [0, 1]. For this, we set: k
2
= 3p,
k
1
= 3p
2
, k
0
= p
3
with p = 0.4, in the normalized system
simulations. The converter parameters were taken to be
L = 10
3
[H], C = 10
6
[Farad] R = 30 [Ohm] E = 15 [V ]
For these parameters, the time normalization factor is found
to be

LC = 3.1622 10
5
.
C. GPI-SLIDING MODE CONTROL VIA A --
MODULATOR
Based on the presented justications, we propose to imple-
ment the GPI controller (4.9) on the buck power converter
3002
by means of a --modulator, as previously discussed.
u =
1
2
(1 + sign (z)), z = e = u
av
u
u
av
= u +
_
1
Q
k
2
_
( y)
e
+ (1 k
1
)(y y)
+k
0
_
t
0
(y() y)d
( y)
e
=
1
Q
y
_
t
0
(y() u) d (4.14)
Under ideal sliding conditions on the sliding surface, z =
0, the corresponding dynamics is precisely represented by
the condition u = u
av
. The stability analysis carried out for
the closed loop behavior of the average system under a GPI
controller thus becomes valid. The sliding mode controller
results in the origin of the tracking error, y y, to be an
exponentially asymptotically stable equilibrium point for all
motions that do not saturate the control input u
av
(t) beyond
the interval [0, 1].
Figure 6 shows computer simulations depicting the closed
loop response of the system to the actions of the GPI con-
troller implemented through a --modulator. The controller
design parameters and the system parameters were chosen to
be exactly the same as those used in the previous simulation
of the average GPI feedback controlled responses.
In order to test the robustness of the --modulator imple-
mentation of the proposed average GPI controller, we tested
the system with an unmatched sudden constant perturbation,
denoted by 1(t ), appearing at time = 0.004 [s] of
value = 0.6667 [A]. i.e. we used the model:
L

i = v + uE
C v = i
v
R
+ 1(t ) (4.15)
Figure 7 shows computer simulations depicting the closed
loop response and the recovery features of the implemented
GPI---modulation controller. Ideal sliding motions in
traditional sliding mode control do not inherit the robustness
features nor are they robust with respect to unmatched
perturbations.
V. CONCLUSIONS
Average feedback controller designs usually represent the
desirable equivalent control in sliding mode control imple-
mentations. The exact synthesis of the equivalent control is
not physically possible in systems commanded by switches,
and sign non-linearities, such as in traditional, and double
bridge, DC-to-DC power converters. Knowledge of the feed-
back law dening the equivalent control leads to consider
a linear partial differential equation, for the sliding surface,
stating that the closed loop vector eld should be orthogonal
to the sliding surface gradient. However, it is still not obvious
how to synthesize a sliding surface, that corresponds to
a given equivalent control, due to the indeterminacy, and
arbitrariness, of the boundary conditions in the dening linear
partial differential equation that needs to be solved.
In this article, we have demonstrated that the use of
classical -modulators, and the closely associated --
modulators, can solve the sliding mode implementation prob-
lem of average feedback controllers in a rather efcient
manner. The proposed approach retains, in an average sense,
the desirable features of the designed average feedback
controller. When the proposed controllers are synthesized
using only inputs and outputs, as in GPI control, the explicit
asymptotic estimation of the state becomes unnecessary and,
moreover, the matching conditions, intimately related to
the state space representation of the system, are no longer
needed.
We have used a -modulator implementation of a
sliding mode controller for a given average GPI continuous
controller in a buck DC-to-DC power converter. Other non-
linear switched controlled systems may immediately benet
form the sliding mode feedback controller design framework
based on -modulators and nonlinear feedback con-
trollers arising from current nonlinear systems theory (for
instance, geometric, differential algebraic, atness, passivity,
energy methods, H

, etc.). For DC-to-DC power converters,

in particular, average passivity based controllers, such as
those developed in [4] may be readily implemented via
sliding mode controllers in a direct fashion.
ACKNOWLEDGMENT
Research supported by Conacyt-M exico Grant 42231-Y.
VI. REFERENCES
[1] C. Edwards and S. K. Spurgeon, Sliding Mode Control,
Talylor and Francis, London, 1998.
[2] M. Fliess, R. Marquez, and E. Delaleau, State Feed-
backs without Asymptotic Observers and Generalized
PID regulators Nonlinear Control in the Year 2000,
A. Isidori, F. Lamnabhi-Lagarrigue, W. Respondek,
Lecture Notes in Control and Information Sciences,
Springer, London 2000.
[3] W. Perruquetti and J. P. Barbot, Sliding mode control
in Engineering Marcel Dekker Inc., New York, 2002.
[4] H. Sira-Ramrez, R. Ortega, R. P erezMoreno and M.
GarcaEsteban, PassivityBased Controllers for the
Stabilization of DCtoDC Power Converters, Auto-
matica, Vol. 33, No. 4, pp. 499-513, 1997.
[5] H. Sira-Ramrez, Sliding modes, -modulators, and
Generalized Proportional Integral Control of linear
systems Asian Journal of Control, Special Issue on
Variable Structure Systems (to appear).
[6] R. Steele, Delta Modulation Systems, Pentech Press,
London, 1975.
[7] V.I. Utkin, Sliding modes and their applications in
Variable Structure Systems, Mir Publishers, Moscow,
1978.
3003
[8] V. I. Utkin, J. Guldner and J. Shi, Sliding Mode Con-
trol in Electromechanical Systems, Talylor and Francis,
London, 1999.
FIGURES
Fig. 1. Switch position based -modulator
Fig. 2. Switch regulated DC-to-DC buck power converter
Fig. 3. Sliding mode implementation of a designed continuous output
feedback controller through an integrator plus a -modulator
Fig. 4. Equivalence of integral plus -modulator scheme with a
modulator for average control law implementation
Fig. 5. Average system response to continuous (average) GPI feedback
controller design
Fig. 6. Closed loop response of a buck converter to a -modulator
implementation of a GPI stabilizing controller
Fig. 7. Robustness of a Sliding mode- modulator implementation of
a GPI stabilizing controller acting on a buck converter undergoing an
unmatched step perturbation input
3004