Professional Documents
Culture Documents
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCSII.2021.3137836, IEEE
Transactions on Circuits and Systems II: Express Briefs
Abstract—This brief is concerned with the induced L∞ output in many cases. In order to get over this difficulty, some
feedback control problem for continuous-time systems. And alternative methods have been drawn into, like for instance
the main task is to design a desired controller such that the H∞ (energy-to-energy) performance [6] – [9], generalized
closed-loop system is asymptotically stable and satisfies the given
induced L∞ performance index. An induced L∞ performance H2 (energy-to-peak) performance [10] – [12], and induced
analysis criterion is proposed for the considered system via the L∞ (peak-to-peak) performance [13] – [16]. The induced L∞
descriptor representation approach. Then, sufficient condition for control is to design a desired controller so that the induced
the induced L∞ controller is expressed in the form of the linear L∞ norm of the disturbance input and the controlled output
matrix inequality (LMI). Finally, an integrating circuit is used is designed to less than (or equal to) the given value or
to explain the effectiveness and feasibility for the given induced
L∞ controller design strategy. can be minimized. The induced L∞ performance was first
proposed in [17], the problem of induced L∞ control (filtering)
Index Terms: Continuous-time systems, induced L∞ per- got significant concern [13] – [16]. In [13], the induced L2
formance, output feedback control, linear matrix inequalities and L∞ output feedback control problem for discrete-time
(LMIs). systems was studied. The authors used the state feedback
control method to implement the design of output feedback
I. I NTRODUCTION controller. In [14], the method of designing peak-to-peak filter
Due to the economic and operational limitations of measur- for the considered system with control output and performance
ing equipment, it is unattainable to get the entire state variables output quantization was given. In [15], the problem of peak-
for actual systems in many cases. However, the whole state to-peak filtering for nonlinear Markov jump systems was
of the system is required to be measured in the state feed- considered. In [16], the design conditions of fuzzy filter with
back. Considering the cost, reliability and simplicity, output gain uncertainty were studied. Until now, few works have
feedback is always used to control the stability of the system. considered the induced L∞ output feedback control problem
Hence, the research of output feedback control for system has based on descriptor representation approach.
more practical significance. In recent decades, a large number In the mentioned results [13] – [16], the problem of induced
of researchers have focused on output feedback control and got L∞ control (filtering) was investigated via LMIs. It should
a great many results [1] – [5]. In [1], tracking control method be noted that if the above methods are applied to the output
of the adaptive fuzzy output-feedback was presented for un- feedback control system, nonlinear coupling terms between the
certain switched nonlinear systems with arbitrary switching system matrices and the controller gain matrix will be difficult
and given performance restrict. In [2], a new security control to deal with. The above discussions become the research
concept was proposed, and a new event-triggered scheme motive of this brief, the contributions are concluded as follows:
with decentralized output feedback control was designed by 1) The problem of the induced L∞ output feedback control
Lyapunov stability theory and stochastic analysis technique. for the continuous-time systems is studied based on the
In [3], the adaptive decentralized controller design problem descriptor representation approach;
of observer-based for nonlinear systems was investigated. The 2) The semi-positive definite Lyapunov matrix on the main
design problem of the dynamic output feedback control for diagonal is avoided, and the strictly controller design condition
discrete-time semi-Markov jump systems was considered in is given via LMIs.
[4]. In [5], a new event-triggered adaptive dynamic output N otations: The symmetry matrix A > 0 (or A ≥ 0) means
feedback sliding mode control with actuator failure and output, that A is positive definite (or semi-positive definite). He{A} is
input quantization was proposed. defined as He{A} = A + AT . diag{ · · · } denotes a diagonal
As we all know, external disturbances have attracted wide matrix. We use (∗) as an ellipsis for the terms that are intro-
attention in control theory and application, many popular algo- duced by symmetry. L∞ [0, ∞) indicates the space p of bound
rithms have been developed gradually. However, it is not easy vector functions over [0, ∞). kz(t)k∞ = supt z T (t)z(t)
to know the statistical characteristics of external disturbances means the L∞ norm where supt z(t) signifies the supremum
This work was supported by the National Nature Science Foundation of
of the vector z(t).
China under Grant 61773298 and Grant 62173261. (Corresponding author:
Xiao-Heng Chang.)
X.-H. Chang and R.-R. Liu are with the School of Information Science and II. P ROBLEM S TATEMENT A ND P RELIMINARIES
Engineering, Wuhan University of Science and Technology, Wuhan 430081,
China (e-mail: changxiaoheng@sina.com; qliurongrong@sina.com). Consider a continuous-time system described by the follow-
ing state-space equation:
1549-7747 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Nantes. Downloaded on January 03,2022 at 14:19:24 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCSII.2021.3137836, IEEE
Transactions on Circuits and Systems II: Express Briefs
where
ẋ(t) = Ax(t) + Bu(t) + F w(t) 1
He{P Â} + P E + 2Ĉ T Ĉ
Φ1 =
y(t) = C1 x(t) + Hw(t) (1) λ (7)
1
z(t) = C2 x(t) Ψ = 2 [ Â B̂ ]T Ĉ T Ĉ[ Â B̂ ].
λ
where x(t) ∈ Rn is the state variable, y(t) ∈ Rf is Proof: For the closed-loop system (4) obtained by de-
the measurement output, u(t) ∈ Rm is the control input, scriptor representation approach, the Lyapunov function with
z(t) ∈ Rq is the controlled output, and w(t) ∈ Rv is the semi-positive definite matrix is chosen as
disturbance belonging to L∞ [0, ∞). The matrices A ∈ Rn×n , V (ξ(t)) = ξ T (t)P Eξ(t), P E = E T P T ≥ 0 (8)
B ∈ Rn×m , F ∈ Rn×v , C1 ∈ Rf ×n , H ∈ Rf ×v , and
C2 ∈ Rq×n are system matrices. and then
The output feedback controller is described as V̇ (ξ(t)) = ξ T (t)P [Âξ(t) + B̂w(t)]
(9)
u(t) = Ky(t) (2) + [Âξ(t) + B̂w(t)]T P T ξ(t).
where K is controller gain matrix to be designed. Next, pre- and post-multiplying the (6) by [ ξ T (t) wT (t) ]
Substituting (2) into (1), it yields and its transposition, respectively, there exists
ẋ(t) = Ax(t) + BKy(t) + F w(t) 1
He{ξ T (t)P [Âξ(t) + B̂w(t)]} + ξ T (t)P Eξ(t)
(3) λ
0 · ẏ(t) = C1 x(t) − y(t) + Hw(t). 1
+ 2 [Âξ(t) + B̂w(t)]T Ĉ T Ĉ[Âξ(t) + B̂w(t)] (10)
Defining ξ(t) = [ x (t) T T
y (t) ] . Then, according to T λ
equations (1) and (3), the closed-loop system can be deduced + 2ξ T (t)Ĉ T Ĉξ(t) − γ 2 wT (t)w(t) < 0.
as follows: According to (8), (9), and (10), we know that if (6) holds,
˙ = Âξ(t) + B̂w(t)
E ξ(t)
(4) a conclusion can be easily identified as
z(t) = Ĉξ(t) 1 1
V̇ (ξ(t)) + V (ξ(t)) + 2 ż T (t)ż(t) + 2z T (t)z(t)
where λ λ (11)
In×n 0
A
BK
< γ 2 wT (t)w(t).
E= , Â =
0 0 C1 −I Then, we can know that V̇ (ξ(t)) < 0 if w(t) = 0. Hence,
F if the inequality (6) is true, we can deem that the closed-loop
B̂ = , Ĉ = [ C2 0 ].
H system (4) is asymptotically stable with w(t) = 0.
Next, we are in a position to establish the induced L∞
Definition 1. [14] The closed-loop system (4) is asymptoti-
performance γ. First, for a scalar λ > 0, the well-known
cally stable and preserves the induced L∞ performance when
inequality conclusion is recalled as
the following two conditions are satisfied: 1) The closed-loop
system (4) is asymptotically stable when w(t) = 0; 2) Under 1 T 1 1
ż (t)z(t) + z T (t)ż(t) ≤ z T (t)z(t) + 2 ż T (t)ż(t).
zero initial conditions, for any nonzero w(t) ∈ L∞ and given λ λ λ
(12)
γ > 0, the induced L∞ norm satisfies kz(t)k∞ ≤ γkw(t)k∞ ,
According to (11) and (12), one gives
where γ is a prescribed performance index, z(t) is the
controlled output, and w(t) is the external disturbance input. 1 1 1
V̇ (ξ(t)) + V (ξ(t)) + z T (t)ż(t) + ż T (t)z(t)
λ λ λ (13)
+ z T (t)z(t) < γ 2 wT (t)w(t).
III. M AIN R ESULTS
We will commit to solving the problem of the induced L∞ With eλt > 0 is multiplied to both sides of the inequality
control for the closed-loop system (4) in this section. Firstly, (13), it will result in
we give a new induced L∞ performance analysis condition for 1
eλt [ V̇ (ξ(t)) + λV (ξ(t)) + z T (t)ż(t) + ż T (t)z(t)
the closed-loop system (4). Then, a set of LMIs are used to λ (14)
give the corresponding controller design condition under the + λz T (t)z(t) ] < eλt γ 2 wT (t)w(t).
specified induced L∞ performance criterion.
The inequality (14) can be transformed to
Theorem 1. Let γ > 0 be a given performance index, the
d{eλt [ V (ξ(t)) + z T (t)z(t) ]}
closed-loop system (4) is asymptotically stable and preserves < λeλt γ 2 wT (t)w(t). (15)
the specified induced L∞ performance, if there exist scalar dt
λ > 0 and matrix P ∈ R(n+f )×(n+f ) make the following Further, integrating both sides of the matrix inequality (15)
matrix inequalities hold: from 0 to t will deduce
Z t
P E = ET P T ≥ 0 (5) eλt [V (ξ(t)) + z T (t)z(t)] < γ 2 λeλs wT (s)w(s)ds
0
Φ1 ∗ < eλt γ 2 k w(t) k2∞ .
1 T T +Ψ<0 (6)
λ B̂ P −γ 2 I (16)
1549-7747 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Nantes. Downloaded on January 03,2022 at 14:19:24 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCSII.2021.3137836, IEEE
Transactions on Circuits and Systems II: Express Briefs
Then, we can rewrite the inequality (16) as Substitutingthe structure matrices E, Â, B̂, Ĉ, and Q =
T 2
Q1 0
V (ξ(t)) + z (t)z(t) < γ k w(t) k2∞ . (17) Q2 R Q2
into the inequality (23), where R is a dimen-
Since V (ξ(t)) ≥ 0, from the inequality (17), we can obtain sion adjustment matrix, one gives
the following conclusion: Θ1 ∗ ∗ ∗ ∗
1
1 T − λ He{Q 2} ∗ ∗ ∗
Θ2
k z(t) k∞ < γ k w(t) k∞ (18) 1 T 2
F
λ λ H −γ I ∗ ∗ <0
which is the induced L∞ performance condition. The closed- Θ3 C2 BKQ2 C2 F −λ2 I ∗
loop system (4) satisfies the asymptotic stability and the C2 Q1 0 0 0 − 21 I
induced L∞ performance index is also guaranteed. Hence, (24)
the conditions given in Definition 1 are satisfied. Then, we where
complete the proof. 1
Θ1 = He{AQ1 + BKQ2 R} + Q1
λ
A criterion of the induced L∞ performance analysis has 1
Θ2 = (C1 Q1 − Q2 R + QT2 K T B T )
been formulated in Theorem 1. In fact, the above condition λ
is nonlinear matrix inequality, then we give the following Θ3 = C2 AQ1 + C2 BKQ2 R.
induced L∞ controller design result. By defining N = KQ2 , we can obtain (20). Then, we
Theorem 2. Let γ > 0 and λ > 0 be given performance complete the proof.
index and scalar, the closed-loop system (4) is asymptotically
stable and preserves the specified induced L∞ performance, Remark 1. In Theorem 2, we obtain the desired induced
if there exist the positive definite matrix Q1 ∈ Rn×n , matrices L∞ controller design condition, which meets the requirements
Q2 ∈ Rf ×f , and N ∈ Rm×f such that the following matrix of the Definition 1. It is worth noting that the generation
inequalities hold: of Q2 in (20) (the position of Ω2 ) is avoided by defining
EQ = QT E T ≥ 0 (19) Q1 0
Q= . Because the presence of Q2 will restrict
Q2 R Q2
Ω1 ∗ ∗ ∗ ∗ the dimension of the system matrices to square matrices, which
Ω2 − λ1 He{Q2 } ∗ ∗ ∗
is not acceptable in the design of controller. And the dimension
1 T 1 T
λF λH −γ 2 I ∗ ∗ <0
adjustment matrix R is to adjust the dimensions of system
Ω3 C2 BN C2 F −λ2 I ∗ matrices. Its further explanation is shown in Remark 2.
C2 Q1 0 0 0 − 21 I
(20) IV. S IMULATION
where
1
Ω1 = He{AQ1 + BN R} + Q1 L1 L2
λ R1
1 1
Ω2 = (C1 Q1 − Q2 R + N T B T ) i1 i2
λ ic
Ω3 = C2 AQ1 + C2 BN R
and the controller gain matrix K is chosen as us ĉ C uc Ċ R2 is
K = N Q−1
2 . (21)
Proof: By defining P = Q−T , pre- and post-multiplying 0
(5) by QT and Q, respectively, we can get (19).
Then, using Schur complement to the matrix inequality (6), Fig. 1. Integrating circuit.
one gives
1 In this section, we will introduce an integrating circuit
∗ ∗ ∗
λ He{P Â} + P E and structure its dynamic model. The research plant is the
1 T T
λ B̂ P −γ 2 I ∗ ∗ < 0. (22) integrating circuit which has been shown in Fig. 1. The two
Ĉ Â Ĉ B̂ −λ2 I ∗ meshes I and II have been marked with counter clockwise
Ĉ 0 0 − 12 I detour; R1 and R2 are resistance elements; us is controlled
Multiplying the two sides of the inequality (22) by voltage source and its output voltage is controlled by the
diag{QT , I, I, I } and diag{Q, I, I, I }, respectively, it voltage or current of another branch; uc is capacitor and C
yields is capacitance component with zero initial energies; L1 and
1 L2 are inductor elements; is is controlled current source and
∗ ∗ ∗
λ He{ÂQ} + EQ its output current is controlled by other branch currents or
1 T
λ B̂ −γ 2 I ∗ ∗ < 0. voltages; i1 , i2 , and ic represent the branch currents of the
(23)
Ĉ ÂQ Ĉ B̂ −λ2 I ∗ corresponding three branches, respectively; node 0 and 1 are
ĈQ 0 0 − 21 I reference node and independent node, respectively.
1549-7747 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Nantes. Downloaded on January 03,2022 at 14:19:24 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCSII.2021.3137836, IEEE
Transactions on Circuits and Systems II: Express Briefs
Next, we can get the following related current and voltage 0.06
equations from Fig. 1: x1(t)
0.04 x2(t)
di1 duc
VR1 = i1 R1 , VL1 = L1 , ic = C x3(t)
dt dt (25) 0.02
di2
VR2 = (i2 + is )R2 , VL2 = L2 .
dt 0
According to KCL equation for node 1, we can easily obtain
-0.02
0 = i1 − ic − i2 . (26)
-0.04
From loops I and II, the corresponding KVL equations can 0 10 20 30 40 50
be written as Time (sec)
0 = us − uc − VR1 − VL1 , 0 = uc − VL2 − VR2 . (27) Fig. 2. History of the system states x(t).
Substituting (25) into (26) and (27), we can know that 0.1
duc
C = i1 − i2 0.08
dt
di1
L1 = −uc − i1 R1 + us (28) 0.06
dt
di2
L2 = uc − i2 R2 − is R2 . 0.04
dt
In Table I, the relevant parameters of the integrating circuit 0.02
are given , the following mathematical model can be obtained:
0
duc 0 10 20 30 40 50
= 5i1 − 5i2 Time (sec)
dt
di1
= −0.5uc − 10i1 + 0.5us + w(t)
p
Fig. 3. History of z T (t)z(t)/supt (wT (t)w(t)).
dt (29)
di2
= uc − 15i2 − 15is
dt The Fig. 2 shows the responses of the system states
y(t) = i2 + 0.3w(t), z(t) = uc . and the Fig. 3 shows the ratio for the simulation results
In the state space modeling, x(t) = [ uc i1 i2 ]T is of
p the controlled output to the external disturbance, i.e.,
chosen as state variable, u(t) = [ us T
is ] is chosen as z T (t)z(t)/supt (wT (t)w(t)), is 0.0944, less than the spec-
controlled input; i2 is chosen as measurement output; uc is ified bound γ = 1.0. The result of calculation and simulation
chosen as controlled output. Then, the above system (29) can shows that the condition presented in Theorem 2 is suitable
be represented in the form of system (1) with for the controller design for the above system (30). In other
words, the induced L∞ performance index γ = 1.0 for the
0 5 −5 0 0 system (30) is guaranteed in the circumstances.
A = −0.5 −10 0 , B = 0.5 0
1 0 −15 0 −15 TABLE I
RELATED PARAMETERS OF THE INTEGRATING CIRCUIT.
T
C1 = [ 0 0 1 ], F = [ 0 1 0 ]
C2 = [ 1 0 0 ], H = 0.3. R1 Resistance of resistor 1 20 Ω
(30) R2 Resistance of resistor 2 15 Ω
L1 Inductance of inductor 1 2000 mH
Next, the controller gain matrix K will be designed for L2 Inductance of inductor 2 1000 mH
the continuous-time system (30) based on the design criterion C Capacitance of capacitor 1 200 mF
provided by Theorem 2.
Considering the induced L∞ performance for the system
(30), the controller gain matrix (21) is designed by the Remark 2. For the known dimension adjustment matrix
obtained result in Theorem 2. For the chosen scalars γ = 1.0, R ∈ Rf ×n , the inequality (20) which obtained by Theorem
λ = 0.6, and matrix R = [ 1 0 0 ], by using the LMI 2 can be used to settle the induced L∞ output feedback con-
Control Toolbox, it can calculate the induced L∞ controller troller design problem of continuous-time systems. It should
gain matrix K as be noticed that the form of dimension adjustment matrix is
R = [ If ×f 0f ×(n−f ) ] in this brief. Then, the inequality
K = [ −2.1978 − 0.0911 ]T . (31) (20) in Theorem 2 can be verified via the LMI Control Toolbox
in Matlab.
In the actual circuit system, we suppose the initial condition
of the states as x(0) = [ 0 0 0 ]T and the disturbance has the Remark 3. From the above we can see that the matrix
form of w(t) = 0.5e(−0.2t) cos t. inequality P E ≥ 0 is verified by the matrix inequality
1549-7747 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Nantes. Downloaded on January 03,2022 at 14:19:24 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TCSII.2021.3137836, IEEE
Transactions on Circuits and Systems II: Express Briefs
1549-7747 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Nantes. Downloaded on January 03,2022 at 14:19:24 UTC from IEEE Xplore. Restrictions apply.