2017 14th International Multi-Conference on Systems, Signals & Devices (SSD)

A robust observer-based controller design for uncertain

discrete-time systems
Badreddine El haiek1 , Abdelaziz Hmamed1 , Ismail Er Rachid1 , Mohammed Alfidi2

Abstract— This paper is concerned with the design In comparison with existing results in this field,
problem of observer-based controllers for uncertain especially those in [7], in this paper, we propose,
linear discrete-time systems. The proposed approach new sufficient LMIs that guarantee the stability of
allows one to compute simultaneously the observer
and controller gains by solving a single bilinear matrix discrete-time uncertain linear systems under the action
inequalities(BMIs), which becomes an linear matrix of observer-based feedbacks. It will be shown that the
inequalities(LMIs) without using any equality con- computation of observer and controller gains can be
straint or an iterative program. Numerical examples done simultaneously, by solving only one LMI without
are given to show the superiority of the proposed any additional restrictive conditions on Lyapunov
approach.
matrix [7].
keywords: Linear matrix inequalities(LMIs)approach,
Observer-based control, Uncertain linear systems.
This note is organized as follows. The problem
I. Introduction formulation and preliminaries are presented in Section
The name of observer is alluded to as a dynamical In section II. in section III we provide the main
system that uses the information of the system inputs results concerning the design of robust observer based
and outputs to establish the unmeasured states of controllers via LMI approach for discrete−time linear
the system under consideration. For deterministic and systems with parameter uncertainties. Finally, in section
stochastic linear systems, the theory of observers is well IV The illustrative example is proposed to illustrate the
developed thanks to the pioneer works of Kalman [1] effectiveness of the proposed results.
and Luenberger [2]. However, a little attention has been Notaions:Throughout this note, we use the following
dedicated to observer-based control in the uncertain notations: Rn denotes the n−dimentional Euclidean
linear case [7], [8], [9], [10], [11], [22], [25] and still space. ∗ is used for the blocks induced by symmetry. I
an open problem [12]. It is worthwhile to mention is the identity matrix with appropriate dimensions. AT
that the main difficuly that encountered in adressing represents the transpose matrix of A. Q>0 means that
the observer-based controller for uncertain linear is Q is real symmetric and positive definite, and He(S) is
a non-convex problem [22]. Beside this obstacle, the defined as He(S)=S+S T .
observer-based feedback must be conceived not only to II. System description and problem statement
stabilize the system but also to deal with the impact
Consider the discrete-time uncertain linear systems
of eventual uncertainties. In such a framework, the
described by state-space equations:
available results contain matched uncertainties [13],
norm-bounded uncertainties [7], [10], [22], [3], [4], and x(k + 1) = (A + ΔA(k))x(k) + Bu(k)
(1)
uncertainties of dyadic types [5]. y(k) = (C + ΔC(k))x(k)
Our goal is to propose an LMI approach to adress where x(k)∈Rn is the state vector, y(k)∈Rp is the
the observer based problem for discrete-time uncertain output measurement, and u(k)∈Rm is the control input
linear systems. We will focus our study on papers vector.The nominal matrices A∈Rn×n , B∈Rm×n , and
that have examined the same problem with parameter C∈Rp×n are known and constant. First, we consider the
uncertainties, for recent works, we refer the readers to following assumptions:
[22], [23], [24], [25], [26], [27].
• The pairs (A,B) and (A,C) are stabilizable and
1 Badreddine El Haiek is with the Department of Physics Fac- detectable, respectively.
ulty of Sciences Dhar El Mehraz B.P. 1796 Fes-Atlas Morocco • There exist matrices Mi ,Ni ,Fi , i=1,2, of appropriate
[email protected] dimensions.
1 Abdelaziz Hmamed is with Department of Physics Fac-
ulty of Sciences Dhar El Mehraz B.P. 1796 Fes-Atlas Morocco
hammed− [email protected] The uncertain terms ΔA(k)∈Rn×n and ΔC(k)∈Rp×n
1 Ismail Errachid is with Department of Physics Faculty are unknown matrices that account for time-varying
of Sciences Dhar El Mehraz B.P. 1796 Fes-Atlas Morocco
[email protected] parameter uncertainties and are assumed to be
2 Mohammed Alfidi is with laboratory of engineering of systems structured under the form:
and applications, National school of applied sciences, Fes university
alfidi− [email protected] ΔA(k) = M1 F1 (k)N1 , ΔC(k) = M2 F2 (k)N2 (2)

978-1-5386-3175-1/17/$31.00 ©2017 IEEE 586

where the unknown matrices Fi (k) satisfy the condition
By developing product term PΠ into inequality (9) we
FiT (k)Fi (k) ≤ I, i = 1, 2. (3) get (7). 
The observer dynamics that we consider in this paper is
under the form: The difficulty to have an LMI stems from the off-
diagonal bilinear term P1 BK. Indeed, there is no
x̂(k + 1) = Ax̂(k) + Bu(k) + L(y(k) − C x̂(k)) (4) congruence transformation eliminating this bilinearity.
We propose a novel manner to overcome the obstacle
where x̂(k)∈Rn×n is the estimate of x(k) and
of the coupling P1 BK without equality constraint
L∈Rn×p is the observer gain to be determined.
or an a priori choice of the matrix P1 . The following
Let e(k) = x(k) − x̂(k) be the estimation error. Then,
preliminary lemmas are needed to derive the new results.
under the feedback

u(k) = −K x̂(k) (5)

Lemma 2 .[30](Schur Lemma): Let S1 , S2 and S3
where K∈Rm×n is the controller gain, the closed loop be three matrices of appropriate dimensions such that
−1 T
dynamics is described by S1 =S1T and S3 =S3T . Then,
 S1 <0 and S1 −S2 S3 S2 <0
  S 1 S2
(A − BK + ΔA(k)) BK if and only if < 0.
ξk+1 = ξk (6) S2T S3
ΔA(k) − LΔC(k) A − LC
  
Π Lemma 3 .[31] Let X, Y , and Δ be real matrices with
appropriate dimensions and ΔT Δ ≤ I the following
where ξkT =[x(k)T ,e(k)T ] is the augmented state vector.
inequality holds:
The objective consists in finding the parameters L and
K so that the closed-loop system (6) is asymptotically
stable. XΔY + Y T ΔT X T ≤ 1 XX T + Y T Y (10)

Lemma 1 System (1) is asymptotically stabilizable Lemma 4 .[29]: For matrices T, Q, U, and W with
by (4)−(5) if there exist two positive definite matrices appropriate dimensions and scalar ρ. The inequality
P1 ∈Rn×n , P2 ∈Rn×n , K, Y and L=P2−1 Y satisfying the
following matrix inequality: T + W T QT + QW < 0 (11)
⎡ ⎤
−P1 ∗ ∗ ∗
⎢ 0 −P2 ∗ ∗ ⎥ is fulfilled if the following condition holds:
Σ=⎢ ⎣ Σ31
⎥<0 (7)
P1 BK −P1 ∗ ⎦  
Σ41 P2 A − Y C −P2 T ∗
0 <0 (12)
ρQT + U W −ρU − ρU T
where
Σ31 = P1 (A + ΔA) − P1 BK Proof. Pre-and
 post-multiplying
 (12) by the full row
Σ41 = P2 ΔA − Y ΔC rank matrix I ρ1 W T and its transpose, respectively,
the inequality (11) can be obtained.
Proof. Now, let us associate to the dynamics (6) the
following Lyapunov function:
III. Main result
Vk = V (ξk ) = ξkT Pξk (8)
In this section, we will give the main result of this
  paper, namely to find the observer-based controller
P1 0 gains in such a way to stabilize the system in closed
Where P= > 0 be the Lyapunov function
0 P2 loop, i.e., to ensure that (6) is asymptotically stable. We
candidate. summarize the resulting LMI design conditions in the
Then, the difference between the Lyapunov functions at following theorem.
two consecutive time instants, ΔVk =Vk+1 −Vk is given
by ΔVk =ξkT (ΠT PΠ−P)ξk
Evidently, ΔVk <0, ∀ξk =0 if the matrix inequality
ξkT (ΠT PΠ−P)ξk <0. Theorem 1 System (1) is asymptotically stabilizable
The last matrix inequality is equivalent by the Schur by (4)−(5) if there exist a known scalar ρ, two positive
complement to definite matrices P1 ∈Rn×n , P2 ∈Rn×n and U ∈Rm×m ,
  N ∈Rm×n , Y =∈Rn×p satisfying the following linear
−P ∗ matrix inequality:
<0 (9)
PΠ −P

587
 ⎡ ⎤
−P1 ∗ ∗ ∗ ∗
uncertainties, the following theorem hold.
⎢ 0 −P2 ∗ ∗ ∗ ⎥
Ω =⎢
⎣ Ω31 BN −P1 ∗ ∗ ⎥<0
⎦
Ω41 P2 A − Y C 0 −P2 ∗
−N N ρ(P1 − BU )T 0 Υ Theorem 2 System (1) is asymptotically stabilizable
where by (4)−(5) if there exist a known scalar ρ, and
Υ = −ρU − ρU T (13) positive scalars 2 >0 and 3 >0, two positive definite
matrices P1 ∈Rn×n , P2 ∈Rn×n and U ∈Rm×m , N ∈Rm×n ,
Ω31 = P1 (A + ΔA) − BN Y ∈Rn×p satisfying the following linear matrix inequality:
Ω41 = P2 ΔA − Y ΔC ⎡ ⎤
Λ11 ∗ ∗ ∗ ∗ ∗ ∗
Proof. Suppose that inequality Ω holds. The feasible ⎢ 0 −P2 ∗ ∗ ∗ ∗ ∗ ⎥
⎢ ⎥
solution of this inequality satisfies −ρU − ρU T < 0 , ⎢ Λ31 BN −P1 ∗ ∗ ∗ ∗ ⎥
⎢ ⎥
⎢ 0 −P2 ∗ ∗ ∗ ⎥
which implies that the matrix
 U is nonsingular.
 Applying ⎢ Λ42 0 ⎥<0
−1
−N N 0 0 , ⎢ −N ∗ ⎥
⎡ 4 with W =U
Lemma ⎤ ⎢ N Λ53 0 Λ55 ∗ ⎥
0 ⎣ 0 0 M1T P1 M1T P2 0 −2 .I ∗ ⎦
⎢ 0 ⎥ 0 0 0 M2T Y T 0 0 −3 .I
Q=⎢ ⎥
⎣ (P1 B − BU ) ⎦, and (15)
⎡ 0 ⎤
−P1 ∗ ∗ ∗ Hence, The observer-based control gains given by
⎢ 0 −P2 ∗ ∗ ⎥ K=U −1 N and L=P2−1 Y .
T=⎢ ⎣ P1 (A + ΔA) − BN
⎥
BN −P1 ∗ ⎦ where
P2 ΔA − Y ΔC P2 A − Y C 0 −P2 Λ11 = −P1 + 2 N1T N1 + 3 N2T N2
the inequality in Ω leads to Λ31 = P1 A − BN
Λ42 = P2 A − Y C
⎡ ⎤
−P1 ∗ ∗ ∗ Λ53 = ρ(P1 B − BU )T
⎢ 0 −P ∗ ∗ ⎥ Λ55 = −ρU − ρU T
⎢ 2 ⎥
⎣ P1 (A + ΔA) − BN BN −P1 ∗ ⎦ Proof. We rewrite the matrix Ω as a sum of two
P2 ΔA − Y ΔC P2 A − Y C 0 −P2 matrices, one contains the uncertainties and another
⎧ ⎡ ⎤ ⎫ one without the uncertainties, that is:
⎪
⎪ 0 ⎪
⎪
⎨ ⎢ ⎥ −1   ⎬ Ω = Ω1 + ΩΔ
⎢ 0 ⎥U
+He ⎣ (P1 B − BU ) ⎦ −N N 0 0 where
⎪
⎪ ⎪
⎪
⎩ ⎭ Ω1 =
0 ⎡ ⎤
−P1 ∗ ∗ ∗ ∗
By defining K = U −1 N , we obtain
⎡ ⎤ −P2 ∗ ∗ ∗ ⎥
⎢ 0
−P1 ∗ ∗ ∗ ⎢ P1 A − BN BN −P1 ∗ ∗ ⎥
⎢ ⎣ ⎦
⎢ 0 −P 2 ∗ ∗ ⎥⎥ 0 P2 A − Y C 0 −P2 ∗
⎣ P1 (A + ΔA) − BN BN −P1 ∗ ⎦ −N N ρ(P1 B − BU )T 0 Υ
P2 ΔA − Y ΔC P2 A − Y C 0 −P2 where Υ define in (13).
⎧⎡ ⎤⎫ ⎡ ⎤
∗ ∗ ∗ ⎪ 0 ∗ ∗ ∗ ∗
⎪
⎪ 0 ⎪
⎨⎢ ⎬ ⎢ 0 0 ∗ ∗ ∗ ⎥
⎢ 0 0 ∗ ∗ ⎥ ⎥ ΩΔ = ⎢ P1 ΔA 0 0 ∗ ∗ ⎥
+He ⎣ ⎣ ⎦
⎪
⎪ −P1 BK + BN P1 BK − BN 0 ∗ ⎦⎪ ⎪ P2 ΔA − Y ΔC 0 0 0 ∗
⎩ ⎭
0 0 0 0 0 0 0 0 0
(14) By developing ΔA(k) and ΔC(k) we obtain:
From (14), it implies that the inequality (7) in lemma 1 ⎡ ⎤ ⎡ ⎤T ⎡ ⎤ ⎡ ⎤T
holds.  0 N1T N1T 0
⎢ 0 ⎥ ⎢ 0 ⎥ ⎢ 0 ⎥ T ⎢ 0 ⎥
ΩΔ =⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣ P1 M1 ⎦F1 (k)⎣ 0 ⎦ +⎣ 0 ⎦F1 (k)⎣ P1 M1 ⎦
⎢ ⎥
P2 M1 0 0 P2 M1
Remark 1 By using the matrix inequality technique, 0 0 0 0
⎡ ⎤ ⎡ T ⎤T ⎡ T ⎤ ⎡ ⎤T
theorem 1 introduces a new criterion for discrete-time 0 N2 N2 0
systems. In the new criterion, the appearance of crossing ⎢ 0 ⎥ ⎢ 0 ⎥ ⎢ 0 ⎥ ⎢ 0 ⎥
+⎢
⎣ 0 ⎥F2 (k)⎢ 0 ⎥ +⎢ 0 ⎥F2T (k)⎢
⎦ ⎣ ⎦ ⎣ ⎦ ⎣ 0 ⎥
⎦
terms P1 BK appearing in (7) has been avoided, it enables −Y M2 0 0 −Y M2
us to give LM I conditions for designing observer based 0 0 0 0
control. Consequently, using the lemma 3 and (3), we have

Notice that the theorem 1 contains the uncertainties, so Ω ≤ Ω1 + −1 T T −1 T T

2 X1 X1 + 2 X2 X2 + 3 X3 X3 + 3 X4 X4
now, we linearize the last theorem 1 with respect to the (16)

588
for any 2 >0,3 > 0, where
 
X1T = 0 0 M1T P1 M1T P2 0 ,
 
X2T = N1 0 0 0 0 ,
 
X3T = 0 0 0 −M2T Y T 0 ,
 
X4T = N2 0 0 0 0 .

Using the Schur complement lemma 2 in (16), we obtain that

Ω<0 holds if the LMI (15) is satisfied.

Remark 2 :The author in [7] inserted a free parameter α to

find a solution the problem of the observer-based controller de-
sign by changing the original BMI (7) into two dependent BMI
condition to design controller and observer. This decoupling
technique leads the apparition of the term αP1−1 . In order to Figure 1:State errors of the closed-loop system.
linearize this term, an additional parameter β is introduced.
Therefore, a supplementary condition on the matrix P1 is
α
required, namely, P1 > β 2 I . However, it is clear that our demonstrate that our theorem 2 is less conservative than
theorem is obtained without any constraint on P1 . theorem 1 in [7].

Remark 3 : The conditions in Theorem 2 is LMI for a given By employing Theorem 2 of this paper with ρ=5, the
scalar ρ. In order to avoid any optimization technique to search  controller and observer gainsare
resulting
for such an ρ, a logarithmically spaced search has been proposed −0.7212 −0.6539 0.5829
in [28]. They use a finite set of LMI constraint problems K= ,
1.1609 0.8941 1.2411
with: ρ ∈ E = [10−6 , 10−5 , ..., 106 ] . Thus, ρ is fixed and it is  
0.4039 −0.1515
not a decision variable. L= 1.6766 −0.6287
0.9874 −0.3703
IV. Numerical application.
with    
In this section, a numerical example is given to illustrate 10 −1
the effectiveness of the proposed method. x(0) = 7 ; x̂(0) = 4
−5  1.5 
A. Example sin(k) 0 0
Consider the following system in [7]: F1 (k)=F2 (k)= 0 sin(k) 0
    0 0 sin(k)
1 0.1 0.4 0.1 0.3 The state errors are shown in figure 1 from which we see that
A= 1 1 0.5 , B= −0.4 0.5 the closed−loop system is asymptotically stable. Thus, sim-
 −0.3 0 1 0.6 0.4 ulation results demonstrate the effectiveness of the proposed
1 1 1 observer-based controller.
C=
1 1 1
Methods theorem 1 [7] theorem 2
the uncertainties can be rewritten under the form (2) β1max 2.89 5
with   β2max 102 103
0 0 0
M1 = β1 0.1 0.3 1 table 1:Comparison between different LMI design methods.
 0 0.2 0
0 0 0
N1 = 0.2 0 0.4 V. Conclusion
0 0.1 0   In this paper, the observer-based controllers design problem
  0 0 0
0 0 0.3 for discrete-time linear systems has been tackled. Sufficient
M2 =β2 N2 = 0 0 0 condition for robust observer-based controllers design has
0 0 0.8
0 0 0.2 been established via linear matrix inequalities. Based on that,
new relaxed conditions have been proposed. Finally, a numer-
To show the superiority of the proposed design methodology ical example has been provided to illustrate the effectiveness
as compared with [7], We searched for the maximum values of the proposed method.
of β1 and β2 that satisfy the above described LMI conditions.
The simulation results of the observer-based controller are
References
given in table 1.
The LMIs theorem 2 remain solvable until the maximum [1] R.E. Kalman, A new approach to linear filtering and predic-
allowable β1max =5 and β2max =103 . In other word, it is tion problems, Transactions of the ASME, Journal of Basic
shown that the approaches proposed in [7] is not able to Engineering 82 (D) (1960)35−45.
achieve the maximum allowable values β1max and β2max [2] D.J. Luenberger, An introduction to observers, IEEE Trans-
obtained via the method proposed in this work. which actions on Automatic Control AC−16 (6) (1971) 596−602.

589
 [3] I.R. Petersen, C.V. Hollot, High-gain observers applied to [24] Heemels, W., Daafouz, J., Millerioux, G. (2010). Observer-
problems of stabilization of uncertain linear systems, dis- based control of discrete-time LPV systems with uncertain
turbance attenuation and H∞ optimization, International parameters. IEEE Transactions on Automatic Control, 55(9),
Journal of Adaptive Control and Signal Processing 2 (1988) 2130−2135.
347−369. [25] Ibrir, S. (2008). Design of static and dynamic output feed-
[4] Y.H. Chen, J.S. Chen, Combined controller-observer design for back controllers through Euler approximate models: uncertain
uncertain systems using necessary and sufficient conditions, in: systems with norm-bounded uncertainties. IMA Journal of
Proceedings of the 29th CDC, 1990, pp. 3452−3454. Mathematical Control and Information, 25(3), 281−296.
[5] I.R. Petersen, A Riccati equation approach to the design of [26] Kheloufi, H., Zemouche, A., Bedouhene, F., Boutayeb, M.
stabilizing controllers and observers for a class of uncertain (2013). On LMI conditions to design observer-based con-
systems, IEEE Transactions on Automatic Control 30 (9) trollers for linear systems with parameter uncertainties. Au-
(1985) 904−907. tomatica, 49, 3700−3704.
[6] B.R. Barmish, G. Leitmann, On ultimate boundness control [27] Shenquan Wanga, Yulian Jiang.Comment on ”On LMI con-
of uncertain systems in the absence of matching assumptions, ditions to design observer-based controllers for linear sys-
IEEE Transactions on Automatic Control AC−27 (1) (1982) tems with parameter uncertainties”. Automatica 49 (2013)
153−158. 3700−3704.
[7] S. Ibrir and S. Diopt (2008), Novel LMI conditions for [28] Oliveira, R.C.L.F., de Oliveira, M.C., Peres, P.L.D. (2011).
observer-based stabilization of Lipschitzian nonlinear systems Robust state feedback LMI methods for continuous−time
and uncertain linear systems in discrete-time, Applied Math- linear systems: discussions, extensions and numerical com-
ematics and Computation, vol. 206, no. 2, pp. 579-588. parisons. In 2011 IEEE int. symposium on computer−aided
[8] P. Khargonekar, I. Petersen, and K. Zhou, Robust stabiliza- control system design (pp. 1038−1043)
tion of uncertain linear systems: quadratic stabilizability and [29] Xiao−Heng Chang,Liang Zhang,Ju H.Park. Robust static
h infin; control theory, Automatic Control, IEEE Transactions output feedback H∞ control for uncertain fuzzy systems fuzzy
on, vol. 35, no. 3, pp. 356-361, mar 1990. set and systems,http://dx.doi.org/10.1016/j.fss.2014.10.023
[9] Z. Lin, X. Guan, Y. Liu, and P. Shi, Observer-based robust [30] Boyd S, El Ghaoui L, Feron E et al (1994) Linear matrix
control for uncertain systems with time-varying delay, IMA inequalities in system and control theory, studies in applied
Journal of Mathematical Control and Information, vol. 18, no. mathematics. SIAM, Philadelphia.
3, p. 439, 2001. [31] Petersen IR (1987) A stabilization algorithm for a class of
[10] S. Ibrir, Design of static and dynamic output feedback con- uncertain linear system. Syst Control Lett 8:351−357.
trollers through Euler approximate models: uncertain systems
with normbounded uncertainties, IMA J. Math. Control In-
form., vol. 25, no. 3, pp. 281-296, 2008.
[11] J. Daafouz and J. Bernussou, Parameter dependent lyapunov
functions for discrete time systems with time varying paramet-
ric uncertainties, Systems and Control Letters, vol. 43, no. 5,
pp. 355-359, 2001.
[12] M. Boutayeb and M. Darouach, Comments on ”a robust state
observer scheme”, Automatic Control, IEEE Transactions on,
vol. 48, no. 7, pp. 1292-1293, July.
[13] F. Jabbari and W. Schmitendorf, Effects of using observers on
stabilization of uncertain linear systems, Automatic Control,
IEEE Transactions on, vol. 38, no. 2, pp. 266-271, feb 1993.
[14] Atassi, A., Khalil, H. (2000). Separation results for the stabi-
lization of nonlinear systems using different high-gain observer
designs. Systems and Control Letters, 39(3), 183−191.
[15] Hendricks, E., Luther, J. (2001). Model and observer based
control of internal combustion engines. In Proceedings of the
international workshop on modeling,emissions and control in
automotive engines (MECA) (pp. 9−21).Fisciano,Italy.
[16] Karafyllis, I., Kravaris, C. (2005). Robust output feedback sta-
bilization and nonlinear observer design. Systems and Control
Letters, 54(10), 925−938.
[17] Pagilla, P., King, E., Dreinhoefer, L., Garimella, S. (2000).
Robust observer-based control of an aluminum strip process-
ing line. IEEE Transactions on Industry Applications, 36(3),
865−870.
[18] L. Xie, Output feedback h1 control of systems with parameter
uncertainty, International Journal of Control 63 (4) (1996)
741−750.
[19] S.K. Yang, Observer−based robust controller design for a
linear system with time−varying perturbations, Journal of
Mathematical Analysis and Applications 231 (1997) 641−642.
[20] H.H. Choi, M.J. Chung, Observer−based H∞ controller design
for state delayed linear systems, Automatica 32 (7) (1996)
1073−1075.
[21] C.−H. Lien, Robust observer−based control of systems with
state perturbations via LMI approach, IEEE Transactions on
Automatic Control 49 (8) (2004) 1365−1370.
[22] Arcak, M., Kokotovic, P. (2001). Observer-based control of
systems with slope-restricted nonlinearities. IEEE Transac-
tions on Automatic Control, 46(7), 1146−1150.
[23] Gao, H., Wang, C., Zhao, L. (2003). Comments on:  An LMI-
based approach for robust stabilization of uncertain stochastic
systems with time-varying delays . IEEE Transactions on Au-
tomatic Control, 48(11), 2073−2074.

590