43rd IEEE Conference on Decision and Control FrC07.

4
December 14-17, 2004
Atlantis, Paradise Island, Bahamas

Robust Tracking Control of DC-to-DC Nonlinear Power

Converters
Enric Fossas, Josep M. Olm and Alan S. I. Zinober

Abstract— Output voltage control of boost and buck-boost approximation of the current reference. The paper is struc-
converters must be performed indirectly, through the input tured as follows. Section II contains the system model and
current, due to the nonminimum phase character of these a set of preliminary results about the tracking problem in
systems. The devices have sensitivity to load perturbations,
which can be removed with the introduction of an observer nonperturbed converters. Section III introduces the Galerkin
that identifies the disturbed parameter and allows the tracking method. In Section IV a control strategy that furnishes the
of periodic signals at the load resistance. performance of boost and buck-boost converters with load
perturbation robustness is developed. Section V exemplifies
I. INTRODUCTION the robust tracking of a sinusoidal signal. The simulation
results are presented in Section VI, while conclusions and
Considerable research effort is being addressed to the
further research are contained in Section VII.
possibility of using DC-to-DC switchmode power convert-
ers as source inverters. Linear converters of the buck type II. NONPERTURBED DYNAMICS
have been used [1], [2] in robust AC voltage generation. [3] Ideal nonlinear switched boost and buck-boost devices
proposes an inverter topology that can produce an output admit a general state-space representation in terms of a two
voltage larger than the dc input voltage. The key is in dimensional bilinear system with the inductor current iL
the connection of the load differentially across two DC- and the capacitor voltage vC as state variables, and a control
to-DC boost converters. However, boost and buck-boost action ν taking its values in the discrete set {0, 1}. Namely,
converters are shown to need indirect control of the output
voltage [4], which makes them very sensitive to external diL
L = −vC + νvC + Vg [1 + k(ν − 1)]
perturbations and parameter uncertainties, and even more dτ
dvC vC
in inversion-based control schemes. The proposal in [3] C = iL − − ν iL ,
may be improved with the design of robust control strate- dτ R
gies to prevent undesirable effects of load perturbations. where k = 0 for the boost converter and k = 1 for the
An algebraic method [5] for the on-line identification of buck-boost converter. C, L, R and Vg are, respectively, the
uncertain parameters is used in the trajectory tracking prob- capacitor, inductor, load resistance and source values. For a
lem for a double bridge buck converter. Dynamic sliding systematic study it is advisable to consider a dimensionless
manifolds have been successfully used for the rejection model obtained by the change of variables
of unmatched disturbances [6] in output voltage tracking

1 L 1 1
by nonlinear power converters. For the indirect current x= iL , y= vC , t= √ τ,
Vg C Vg LC
control scheme, the key is on-line updating of the current √
reference according to the variations of the perturbed pa- and the introduction of λ = R−1 LC −1 and u = 1 − ν.
rameter. This particular type of sliding control is known as The equations then become
dynamic compensation. Here we propose the introduction
of a perturbation observer with dynamics proportional to x
= 1 − u(k + y) (1)
the tracking error of the output voltage which shows a y
= −λy + ux. (2)
reasonably rapid speed of identification and good simulation
Since L, R and C are positive constants, λ is positive.
results. A similar law has been used in [7] for regulation
L and C are usually considered known parameters, while
tasks. Its design complexity is considerably lower than the
perturbations may affect R and Vg . The equation resulting
proposal in [8]. The on-line updating of the input current
after the elimination of the control action u in equations
is performed, as in [8], through a first harmonic Galerkin
(1), (2) contains a differential relation between the state
E. Fossas, partially supported by MCYT grant DPI2002-03279, is
variables which does not depend on any input,
with the Institute of Industrial and Control Engineering, Technical
University of Catalonia, Av. Diagonal, 647, 08028 Barcelona, Spain
x(1 − x
) = (k + y)(y
+ λy). (3)
[email protected]
J.M. Olm is with the Department of Applied Mathematics IV, Technical When y is tracking a T -periodic reference f (t), the state
University of Catalonia, Campus Nord - Edif. C3, C. Jordi Girona, 1-3, variable x satisfies
08034 Barcelona, Spain [email protected]
A.S.I. Zinober is with the Department of Applied Mathematics, The x(1 − x
) = (k + f )(f
+ λf ), x(0) = x0 . (4)
University of Sheffield, Sheffield, UK
[email protected] Let g(t) = (k + f )(f
+ λf ) be a C ∞ (R) mapping.
0-7803-8682-5/04/$20.00 ©2004 IEEE 5291
Theorem 1. [4] If g(t) > 0 (g(t) < 0), equation (4) is clearly sensitive to external perturbations. Alternative
has one and only one unstable (asymptotically stable) T - methodologies will be explored in the next section, search-
periodic solution in R+ (R− ). ing for the easier handling of such a solution and the
possibility of introducing a robust scheme. We now consider
Assume now an ideal steady state where x and y track a sliding control law that satisfies the control goal, that
φ(t) and f (t), respectively, φ(t) being a T -periodic solution is, the tracking of φ(t) by x, which will induce internal
of (4) and ū(t) being the ideal continuous control that dynamics that lead y to track the output voltage reference
allows it. From equations (1), (2) we have f (t). It is worth mentioning that other control techniques
φ
= 1 − ū(k + f ) may be also applicable in this case to achieve x = φ(t). In
fact, the observer based methodology developed in [9] for
f
= −λf + ūφ.
regulation purposes can be easily extended to the tracking
Furthermore, the characteristic restrictions on switched con- situation.
verters result in the input satisfying 0 ≤ ū(t) ≤ 1, so Proposition 4. [4] Let s(x, t) = x − φ(t) be the switching
1 − φ
f
+ λf surface. The control law u = 0 if (k + y)s < 0, u = 1 if
0≤ ≤1 or, equivalently, 0≤ ≤ 1,
k+f φ (k + y)s > 0, produces in system (1), (2) asymptotic motion
to s(x, t) = 0.
thus entailing constraints on f (t). These inequalities can
also be written as III. GALERKIN METHOD



0 < f + λf ≤ φ or φ ≤ f + λf < 0, (5) The Galerkin method may be considered as a generaliza-
tion of the harmonic balance method [11]. Let {wn }n be a
where g(t) = 0 has been taken into account. complete orthonormal system in the separable Hilbert space
Proposition 2. [4] (i) The fulfilment of (5) requires g(t) > X. Then, given x ∈ X, let us assign
0, ∀t. (ii) If inf{g(t)} ≥ sup{f
(t)+λf (t)}, ∀t ∈ [0, +∞), n

then Pn x = (x|wj )wj .
j=1
φ(t) ≥ f
(t)+λf (t), ∀t and [f > 0 ⇔ f
+λf > 0].
The mapping Pn : X −→ Xn is an orthogonal projection
(iii) If f (t) = A + B sin ωt, sufficient conditions for the operator into the set Xn = span{w1 , . . . , wn }. Let also F :
fulfillment of (i) and (ii) are X −→ X be an operator in X, and consider the problem

 ω 2 F x = 0. (8)
A > B 1+ > 0, (6)
λ
 2 {w1 , w2 , . . .} being a base of X, the Galerkin method [10]
A + B 1 + ωλ proposes to approximate the solution of (8) changing x ∈ X
k+A ≥ B+   2 . (7) by xn ∈ Xn :
A − B 1 + ωλ n
xn = cnj wj ,
j=1

The next step is to consider the tracking problem y = and searching for the coefficients {cnj }j that satisfy system
f (t) for the system described in equations (1), (2). F xn = 0 projected onto Xn , known as the Galerkin
equations. These equations can also be written as
Proposition 3. [4] System (1), (2) has: (i) unstable tracking
dynamics when y is taken as the output. (ii) asymptotically Pn F xn = 0 or, equivalently, (F xn |wj ) = 0, j ≤ n.
stable tracking dynamics when x is taken as the output and
Reference [11] contains a detailed study of the existence
the restrictions stated in Proposition 2 (ii) are fulfilled.
and convergence of a Galerkin sequence of approximate
Hence, direct control of the output voltage has to be solutions of (4). It also deals with the output voltage
discarded, the alternative being a current mode indirect behavior when an approximate current reference is used
control action. The problem then focuses on finding the in the tracking control of the system. Uniform convergence
appropriate signal, i.e. bounded and preferable periodic, results and error bounds are obtained.
to be followed by such a current in order to produce, in Remark III.1. (i) When a Galerkin approximation is used
steady state, an output voltage coincident with the desired instead of the exact solution, an error appears due to the
T -periodic reference f (t). Theorem 1 provides the solution: fact that, in general, F xn = 0. Properties of the projection
x must be forced to track φ, the positive, T -periodic, operators yield
unstable solution of (4). However, the periodic solution
F xn = Pn F xn + (I − Pn )F xn = (I − Pn )F xn . (9)
of (4) is not analytically obtainable. Although it can be
numerically approximated by the integration of the ODE, (ii) It is important to realize that the first Galerkin ap-
both in forward or in backward time, the control technique proximation of φ, denoted φ1 from now on, always exists
5292
 
for our system whatever the function f (t) may be, its φ(t; λN + λ̂p ) 1 − φ
(t; λN + λ̂p )
analytical expression being easily obtainable and reasonably
manageable. Writing g(t) = (k + f )f
+ λf (k + f ), we see which, using (15), can be expressed as
that its Fourier expansion (k + y) [y
+ (λN + λp )y] =
g(t) = C0 + C1 cos ωt + D1 sin ωt + 
 = [k + f (t)] f
(t) + (λN + λ̂p )f (t) .
+ Cn cos nωt + Dn sin nωt (10)
n≥2 Introducing the error ey = y − f , the former equation
is such that the coefficients Ci , Di , are affine functions of becomes
λ. Therefore, writing φ1 as f
+ λf f (k + f )
e
y = − λ + ey + (λ̂p − λp ).
φ1 (t) = E0 + E1 cos ωt + F1 sin ωt (11) k + f + ey k + f + ey

and using it in P1 [φ1 (1 − φ
1 ) − g] = 0 yields The estimator dynamics is defined as

C1 + C0 D1 ω D1 − C0 C1 ω λ̂
p = −βf (t)ey ,

E0 = C0 , E1 = , F1 = .
1 + C02 ω 2 1 + C02 ω 2 where β stands for a positive gain. Note that the observation
IV. ADAPTIVE SCHEME error can be written as eλ = λ̂p − λp and, for a constant
Devices that operate under indirect control are especially perturbation, e
λ = λ̂
p . Then, the adaptive system is given
sensitive to disturbances. This is an easily observable fact by
in our system because the input-output differential relation f
+ λf f (k + f )
(3) depends on the system parameter λ, which can suffer e
y = − λ+ ey + eλ(16)
k + f + ey k + f + ey
perturbations. Here we will try to eliminate the undesirable
e
λ = −βf (t)ey , (17)
effects that load changes induce on the converter dynamics
by means of adaptive control. We may assume an unknown with (ey , eλ ) = (0, 0) as an equilibrium point. Its asymp-
value R for the load resistance, due to the addition of a totic stability must be now demonstrated. Assigning ξ =
constant disturbance term Rp to its nominal value RN ; (ey , eλ ) and
that is, R = RN + Rp , where R > 0. Consequently, the Ψ (ξ, t) =
parameter λ may be written as λ = λN + λp with
f
+ λf f (k + f )
λN Rp = − λ+ ey + eλ , −βf (t)ey ,
λp = − , (12) k + f + ey k + f + ey
RN + Rp
(16), (17) can be written as the T -periodic differential
where λ > 0. Therefore, system (1), (2) can de described
system
as
ξ
= Ψ(ξ, t). (18)
x
= 1 − u(1 + y) (13)

The matrix
y = −(λN + λp )y + ux. (14) 


∂Ψ −λ − f (t)+λf (t)
f (t)
Let f (t) > 0, ∀t, be the C ∞ and T -periodic reference A(t) = = k+f (t)

to be tracked by y, while λ̂p represents an estimation of

∂ξ ξ=0 −βf (t) 0
the additive disturbance λp detailed below. Assume likewise is bounded if the hypotheses about f (t) stated at the
that a certain control forces the state variable x to follow a beginning of the section are fulfilled: f > 0 and T -periodic
certain signal φ(t; λN + λ̂p ) such that entails the existence of a positive infimum for f , which

leads to k + f ≥ inf{f } > 0. Moreover, let
φ(t; λN + λ̂p ) 1 − φ
(t; λN + λ̂p ) =
 Ψ1 (ξ, t) = Ψ(ξ, t) − A(t)ξ =
= [k + f (t)] f
(t) + (λN + λ̂p )f (t) ; (15) 
f
+ λf f
= e2 − ey eλ , 0 .
its existence is guaranteed by standard results of ODE (k + f )(k + f + ey ) y k + f + ey
theory. The quoted law can be, for example,
Thus, one may write (18) with its linear part in a neigh-
1
ū = (1 − φ
) bourhood of the origin satisfying
k+y
ξ˙ = A(t)ξ + Ψ1 (ξ, t). (19)
and, from (13), x = φ(t, λN + λ̂p ) in steady state with
appropriated initial conditions. The dynamic behavior of the Note that Ψ1 (0, t) = 0 and that
state variable y will be given by (3) with x = φ:
Ψ1 (ξ, t)
lim =
(k + y) [y
+ (λN + λp )y] = ξ→0 ξ
5293
  
1  (f
+ λf )e2 − f (k + f )e e 
 y y λ be the first Galerkin approximation of φ(t; λN ). Note that
= lim   =0
ξ→0
e2y + e2λ  (k + f )(k + f + ey )  its coefficients are λN -dependent (see Remark III.1 (ii)).
For the perturbed case we may consider (11) and replace
uniformly in t, because the two terms of the numerator of λN by λN + λ̂p (t). We will therefore make use of
the limit have degree two, while the smallest degree of the
φ1 (t; λN + λ̂p ) ≈ φ(t; λN + λ̂p ).
denominator is 1. Let
1 e2λ Remark IV.1. (i) The sliding control law proposed in section
V (ξ) = e2y + ; II will induce the desired behavior for the state variable
2 β
x. Note that in such a context the load perturbations of
be a positive definite, radially unbounded, decrescent func- the ideal boost and buck-boost converters are additive field
tion. The derivative of V calculated over the linearized disturbances that do not satisfy the matching condition:
system (19) trajectories, that is, over ξ
= A(t)ξ, is
(0, −λp y) ∈
/ span{(−y, x) }.
f
(t) + λf (t) 2
V
(ξ) = − λ + ey .
k + f (t) (ii) The observer-based control purpose developed in [9]
loses its advantage of eliminating the need to sense the
Reasons identical to the ones that allow the boundedness
output variable when a perturbation observer is introduced.
of A(t) to be established, together with the additional
(iii) When the system is physically implemented, the fulfil-
restriction (k+f )−1 (f
+λf ) > 0 (equivalent to the demand
ment of the restrictions derived from the fact that the system
g(t) > 0 in section II), entail the existence of ρ ∈ R such
must always remain in the unsaturated zone (see Section II)
that (k + f )−1 (f
+ λf ) ≥ ρ > 0. Therefore,
has to be guaranteed.
V
(ξ) ≤ −(λ + ρ)e2y ,
V. ROBUST TRACKING OF A SINUSOIDAL
is negative semidefinite. It is straightforward to see that REFERENCE
the subset of the phase plane where the former derivative Our interest is focused on obtaining
vanishes is S = {(0, eλ )}. Moreover, the greatest invariant
set inside S is {(0, 0)} since y = A + B sin ωt, A, B > 0

ey = 0 =⇒ e
λ = −β · f (t) · 0 =⇒ eλ = α, α∈R at the output resistance of a nonlinear boost or buck-boost

converter under load disturbances. The restrictions estab-
and, as e
y = 0 (otherwise, the system would abandon S lished in Section IV act on the signal parameters as follows:
immediately), we have (i) The existence of a periodic indirect current reference φ
f
(t) + λf (t) for the nonperturbed exact problem and the location of the
e
y = 0 = − λ + · 0 + f (t) · α, steady state system in the unsaturated zone are related to
k + f (t)
(6), (7). We will demand these restrictions to be satisfied for
which means that 0 = f (t)·α. By hypothesis it is f (t) = 0, λ −→ λN + λ̂p , ∀t. (ii) The approximate approach needs
∀t, which leads to α = 0. Hence, applying the invariance the Galerkin approximation φ1 (t; λN ), which always exists
principle of LaSalle and the indirect method of Lyapunov (see Remark III.1 (ii)). The expression of φ1 (t; λN ) may be
for nonautonomous systems [13], it can be established that: found in (11), taking into account that the coefficients for
g(t) (10) are
Proposition 5. The origin (0, 0) of system (16, 17) is locally
uniformly asymptotically stable. B2
C0 = λ A2 + kA + , C1 = (k + A)Bw,
2
Note that when such an equilibrium point is reached,
the output y coincides with the desired reference f and, D1 = (k + 2A)Bλ, 2C2 = −B 2 λ, 2D2 = B 2 ω,
moreover, the signal φ(t, λN + λ̂p ) tracked by the state
variable x can be periodic. This follows from the application and Cn , Dn = 0, ∀n ≥ 3. The construction of φ1 (; λN +λ̂p )
of Theorem 1 to equation (15), where λ̂p is now constant. must be made from φ1 (t; λN ). (iii) The sliding surface is
The on-line updating of the current reference φ(t; λN + λ̂p ) s(x, t) = x − φ1 (t; λN + λ̂p ),
according to the instant variations suffered by λ̂p is almost
impossible in real problems. In fact, we have already while the control law is given in Proposition 4.
commented about the difficulties that arise if we intend
to work with φ(t; λN ), λN being constant, which justified VI. SIMULATION RESULTS
the introduction of the Galerkin method in Section III. Our The robust procedure is tested for a buck-boost converter
proposal considers the use of the first harmonic Galerkin with parameters Vg = 50V , RN = 10Ω, L = 0.018H and
approximation of φ, rapidly obtainable for our system. Let C = 0.00022F . The output voltage reference is
φ1 (t; λN ) = E0 (λN ) + E1 (λN ) cos(ωt) + F1 (λN ) sin(ωt) vCr = 135 + 15 sin 2πντ V,
5294
 A=2.7 B=0.3 ω=0.6252 β=0.0125 A=2.7 B=0.3 ω=0.6252 β=0.0125
1 5.5
λ (Rp=RN/5) yr
λob y (R =R /5)
1 p N
λ (R =R /2) 5 y (R =R /2)
1 p N
p N
0.9
λ y (R =R )
1 p N
ob
λ (R =R )
p N
λ 4.5
ob
0.8

0.7

3.5

0.6
3

0.5
2.5

0.4 2
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
t t

Fig. 1. Estimation of several disturbances. Fig. 2. Output voltage behavior for several disturbances, the gain β being
constant.

with ν = 50Hz. In dimensionless variables this corresponds

to λN = 0.9045, w = 0.6252 and A=2.7 B=0.3 ω=0.6252 β=0.0125
9.5

yr = f (t) = 2.7 + 0.3 sin ωt.

9

The existence of a periodic current reference φ is ensured

for the buck-boost in the nonperturbed case as well as its 8.5
φ (R =0)
location in an unsaturated phase plane zone for that situation 1 p
φ (R =R /2)
1 p N
(6), (7). A SIMULINK model of the converter has been 8
x1r
x1
used in the simulations. The control actuator switches at
a maximum frequency of 20 kHz. The initial conditions 7.5

of the state variables are close to the nonperturbed current

7
reference for x and to the output reference for y. The results
show good identification of the perturbations, with values 6.5
up to 100% of the load resistance level. The optimal gains
of the observer dynamics are around β = 0.0125 for the 6
buck-boost, and obviously depend on the disturbance. For
these gain values, Figure 1 plots the observer behavior in the 5.5
0 5 10 15 20 25 30 35 40
presence of initial perturbations such as Rp = 2Ω, Rp = 5Ω t

and Rp = 10Ω, that is, 20%, 50% and 100% of the nominal
value. The overshoot increases with the disturbance. Figure Fig. 3. Current response to a disturbance occurring at t = 10.
2 portrays the output voltage tracking its reference. It can
be seen that it needs two periods to correct the disturbance.
This phenomenon is due to the indirect control that this
variable undergoes. For a certain value of λ a concrete identification velocity and greater overshoot which, in turn,
current reference is needed so that the internal dynamics increase the duration of the transient. Hence, a compromise
make the output track the desired reference. If λ changes, between these two features is taken into account when
a new current reference is needed to maintain the output choosing β. The boost estimation algorithm seems to show
tracking. Although the system is always under control, the a better relationship between velocity and overshoot. More-
unavoidable transition of the current from one reference over, from a certain threshold for β the observer does not
to another is responsible for the transient tracking error identify the disturbance and the system destabilizes. Such
at the output. This is more clearly observed in Figures 3 aspects can be observed in Figure 5, where the behavior of
and 4, where the perturbation occurs at instant t = 10. the observer when Rp = 5Ω (50% of the nominal value)
Note that x never leaves its own reference, from which is plotted for several values of β. Finally, the steady state
it is indistinguishable. It should also be pointed out that relative error in the output tracking is depicted in Figure 6.
larger gain values of the observer dynamics result in faster Note that the output relative error does not exceed 0.7%.
5295
 A=2.7 B=0.3 ω=0.6252 β=0.0125 A=2.7 B=0.3 ω=0.6252 β=0.0125
4.5 0.02
yr
y1
0.015

4
0.01

0.005
3.5

3 −0.005

−0.01

2.5
−0.015

−0.02
30 31 32 33 34 35 36 37 38 39 40
2
0 5 10 15 20 25 30 35 40 t
t

Fig. 6. Relative error of the output voltage tracking in stationary state.

Fig. 4. Voltage response to a disturbance occurring at t = 10.

A=2.7 B=0.3 ω=0.6252 Rp=RN/2

the idea developed in [8], is left for further research.
0.95
λ
λ (β=0.010)
ob
R EFERENCES
0.9 λob (β=0.0125)
λ (β=0.015) [1] F. Boudjema, M. Boscardin, P. Bidan, J.C. Marpinard, M. Valentin
ob
and J.L. Abatut, ”VSS Approach to a Full Bridge Buck Converter
0.85 Used for AC Sine Voltage Generation”, in Proc. of the IECON, 1989,
pp. 82-89.
0.8 [2] M. Carpita and M. Marchesoni, Experimental Study of a Power
Conditioning System Using Sliding Mode Control, IEEE Trans.
Power Electronics, vol. 11, no. 5, September 1996, pp. 731-742.
0.75
[3] R.O. Cáceres and I. Barbi, A boost DC-AC converter: analysis,
design and experimentation, IEEE Trans. Power Electronics, vol. 14,
0.7 no. 1, January 1999, pp. 134-141.
[4] E. Fossas and J.M. Olm, Asymptotic Tracking in DC-to-DC Nonlin-
ear Power Converters, Discrete and Continuous Dynamical Systems
0.65
- Series B, vol. 2, no. 2, May 2002, pp. 295-307.
[5] H. Sira-Ramı́rez, E. Fossas and M. Fliess, ”An algebraic, on-
0.6 line, parameter identification approach to uncertain dc-to-ac power
conversion”, in Proc. of the CDC, 2002, pp. 2462-2467.
0.55
[6] Y.B. Shtessel, A.S.I. Zinober and I.A. Shkolnikov, ”Boost and Buck-
0 5 10 15 20 25 30 35 40 Boost Power Converters Control via Sliding Modes Using Dynamic
t
Sliding Manifold”, in Proc. of the CDC, 2002, pp. 2456-2461.
[7] G. Escobar, On nonlinear control of switching power electronics
Fig. 5. Relative errors of the output voltage tracking in stationary state. systems, Ph. D. Thesis, Université de Paris-Sud, U.F.R. Scientifique
d’Orsay, 1999.
[8] E. Fossas-Colet and A.S.I. Zinober, ”Adaptive Tracking Control of
Nonlinear Power Converters”, in Proc. of the IFAC Workshop on
VII. CONCLUSIONS AND FUTURE WORK Adapt. and Learn. in Control and Signal Process., Como, 2001, pp.
261-266.
Approximate asymptotic output voltage tracking has been [9] V. Utkin, J. Guldner and J. Shi, Sliding Mode Control in Electrome-
achieved for load perturbed, basic, nonlinear power convert- chanical Systems, Taylor & Francis, Philadelphia, 1999.
ers by means of dynamic compensation. Within an indirect [10] E. Zeidler, Nonlinear Functional Analysis and its Applications - Part
IIA: Linear Monotone Operators, Springer-Verlag, New York, 1990.
control scheme, an adaptive control estimates the perturbed [11] J.M. Olm, Asymptotic tracking with DC-to-DC bilinear power con-
parameter, and this information is incorporated into a first verters, Ph. D. Thesis, Universitat Politècnica de Catalunya, 2004.
order Galerkin approximation of the non perturbed current [12] S. Sastry, Nonlinear Systems. Analysis, Stability and Control, Sprin-
ger-Verlag, 1999.
reference. The current on-line updating is finally used in an [13] J.J.E. Slotine and W. Li, Applied Nonlinear Control, Prentice-Hall,
appropriate sliding surface. Optimization of the transient 1991.
response for the asymptotic tracking control law, following

5296