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/TII.2018.2805707, IEEE
Transactions on Industrial Informatics

Peak-to-Peak Filtering for Networked Nonlinear DC Motor Systems with Quantization

Xiao-Heng Chang, Member, IEEE, and Yi-Ming Wang

Abstract—This paper investigates the peak-to-peak filtering problem. Up until now, few research results have taken into
problem for a class of networked nonlinear DC motor systems account the quantization problem of uncertain systems, espe-
with quantization. The nonlinear DC motor system is modeled cially the uncertainty is included in system outputs.
by a Takagi-Sugeno (T-S) fuzzy model. Consider that the mea-
surement output signal and the performance output signal of The concerned object of this paper is a class of networked
the system are quantized by two static quantizers before being DC motor systems. DC motor system is the most common
transmitted by the digital communication channel, respectively. kind of motor system, which plays a vital role in the pro-
Attention is focused on the design of a peak-to-peak filter duction of industry and people’s daily life. Over the past
such that the filtering error system is asymptotically stable and few years, the problem of DC motor system stability analysis
satisfies the prescribed peak-to-peak filtering performance index.
Sufficient conditions for such a peak-to-peak filter are expressed and performance design has been extensively explored [16] –
in the form of linear matrix inequalities. Finally, an illustrative [18]. However, most of the research results regard DC motor
simulation is given to show the effectiveness of the proposed system as a linear system. In fact, in practical engineering, the
approach. DC motor system is actually a complex plant that contains
Index Terms: Networked systems, Quantization, Nonlinear a variety of nonlinearities, such as armature resistor nonlin-
DC motor systems, T-S fuzzy systems, Peak-to-peak filtering. earity, excitation nonlinearity, and saturation nonlinearity, etc.
Therefore, from the perspective of practical application, the
networked nonlinear DC motor system is selected as the object
I. I NTRODUCTION of this paper. In order to handle the nonlinear problem of
In recent years, with the rapid development of communica- networked DC motor system, the T-S fuzzy model is used in
tion and computer technology and the continuous improvement this paper. One of the important advantages of the T-S fuzzy
of control system performance, more and more sensors, con- model is its universal approximation of any smooth nonlinear
trollers, actuators, and other components are contained in the function by a “blending” of some local linear system models.
control system, which leads to the increasing scale of control Within the general framework of T-S fuzzy system, there have
system. Therefore, a growing number of control systems select been a lot of researches on stability analysis and performance
the network to connect various components. This kind of design [19] – [27].
system that transmits information through a real-time network On the other hand, due to the fact that the state variables of
is called networked system [1] – [7]. Compared with the systems are not always available, the signal estimation problem
traditional point-to-point system, the networked system has of system is a significant problem in the fields of control and
many advantages, such as resource sharing, remote operation, signal processing. In the past few decades, the filtering theory
low cost, and easy maintenance. However, the introduction has been greatly developed, and some new filtering methods
of the network as a communication channel to the system have been proposed, such as energy-to-energy filtering [28]
has also brought us some new problems, such as quantization – [36], energy-to-peak filtering [37] – [41], and peak-to-
of signals, data packet losses, signals transmission delay, etc. peak filtering [42] – [45]. For energy-to-energy filtering and
Because there often exist some quantization errors after the energy-to-peak filtering strategies, the disturbance signals are
signals are quantized, which may lead to system performance considered as L2 -type (energy bounded disturbance). How-
degradation or even system instability. Hence, the quantization ever, in practical engineering application, the signals are often
of signals is an important issue to be addressed. So far, there not L2 -type and they are only bound or equivalent to L∞ -
have been a lot of research results on signals quantization type. Under the circumstances, the minimization of the peak
problem in networked systems [8] – [15]. It is noted that value of the estimation error for bounded disturbances ought
most of these results do not consider the uncertain system. to be considered, this kind of performance for systems is
But, from a practical point of view, the system always has named peak-to-peak filtering. The peak-to-peak performance
some uncertainties, which are caused by many factors, such was first demonstrated by Vidyasagar in [42]. Then, many
as measurement error, approximate discretization, parameter important results involving peak-to-peak performance have
identification error and so on. Therefore, it is necessary to been proposed for linear systems [43] – [45]. To the best of the
consider the uncertain system when studying the quantization authors’ knowledge, there are few researches on peak-to-peak
filtering problem of nonlinear system.
Manuscript received **; revised **; accepted **. The work was supported This paper explores the peak-to-peak filtering problem for
in part by the National Nature Science Foundation of China under Grants
61773298 and 61471275. Paper no. TII-17-3036. (Corresponding author: X.- a class of networked nonlinear DC motor systems with quan-
H. Chang.) tization. By contrasting with the existing results on quantized
Xiao-Heng Chang and Yi-Ming Wang are with the School of Information filtering problem of networked systems, the main contributions
Science and Engineering, Wuhan University of Science and Technology,
Wuhan, Hubei 430081, China. Email: changxiaoheng@sina.com; kwangyim- of this paper can be summarized as: 1) The T-S fuzzy method
ing@sina.com. is used to describe the nonlinear DC motor system and the

1551-3203 (c) 2018 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.
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/TII.2018.2805707, IEEE
Transactions on Industrial Informatics

peak-to-peak filter is designed for the T-S fuzzy system based is a very important nonlinear resistor, which is common in
on the fuzzy Lyapunov function approach; 2) The networked practical applications. Therefore, it is very meaningful to
system with two quantized signals is investigated, namely, the consider armature resistor as a current controlled nonlinear
measurement output signal and the performance output signal; resistor in a DC motor system.
3) The system parameter uncertainties are also considered. In this paper, we study the analysis and synthesis of the DC
Notations: Rn and diag{} are utilized to denote n- motor system with a nonlinear armature resistor which com-
dimensional Euclidean space and block-diagonal matrix, re- poses a nonlinear DC motor system. The schematic diagram
spectively. ||·|| denotes the Euclidean vector norm. The symbol of the nonlinear DC motor system is shown in Fig. 2, where a
P > 0 means that P is a matrix that is positive definite and complicated structure of the actual nonlinear DC motor system
symmetric. For the signal w(k) : [ 0, ∞) ∈ Rm , the l∞ norm is simplified to a simple schematic, and the parameters of this
is defined as ||w||∞ = supk≥0 ||w(k)||. The space of signals graph are described in Table I.
m
l∞ is defined to be the set of all m-dimensional signals, which
m
belongs to l∞ , i.e. l∞ = {w(k) ∈ Rm : ||w(k)|| < ∞}. When 





the dimension of the signal is either clear from the context or

immaterial, we drop the superscript and just write l∞ . More-
 

     

 

over, the notations will be adopted for simplicity N (υ(k)) = 0

Pr Pr
w
 

υm (h(k))Nm and N (υ(k + 1)) = υl (h(k + 1))Nl .



m=1 l=1

II. P ROBLEM FORMULATION Fig. 2. Nonlinear DC motor system.

The framework of the networked filtering system considered
in this paper is depicted in Fig. 1. In this figure, the nonlinear
DC motor system is represented by a T-S fuzzy model. The TABLE I
PARAMETERS OF NONLINEAR DC MOTOR SYSTEM
quantized outputs of the system are transmitted to the filter part
via unreliable communication channel. In what follows, we J Moment of inertia of the mechanical rotating part
formulate the DC motor system, T-S fuzzy system, quantizer, B Viscous friction coefficient of the rotating part
R Armature resistor of the DC motor system
filter, and filtering error system, respectively. L Armature inductor of the DC motor system
ω(t) Rotational speed of the DC motor system
u(t) Voltage at both ends of the DC motor system
i(t) Armature current of the DC motor system
( )
 
  

uR (t) The voltage across the armature resistor


uL (t) The voltage across the armature inductor

In the DC motor system described in Fig. 2, a class of cur-

rent controlled nonlinear armature resistors are considered and
( ) assumed that its volt-ampere characteristic can be expressed


by function relation

Fig. 1. Networked filtering problem of nonlinear DC motor system. uR (t) = f (i(t)), (1)

and it is easy to draw the volt-ampere characteristic curve

of the nonlinear resistor through this function relation. For
A. Nonlinear DC Motor System example, when
In the existing literature on research of DC motor system,
most of the research results regard the armature resistor of uR (t) = 4i(t) + i3 (t), (2)
DC motor system as a linear resistor. However, in the actual
the volt-ampere characteristic curve, as shown in Fig. 3, can
DC motor system, the parameters of armature resistor always
be obtained.
change more or less with voltage or current, so generally
According to the positive direction of the armature circuit
speaking, armature resistors in all DC motor systems are
shown in Fig. 2, the electromotive force balance equation can
nonlinear resistors. Nonlinear resistor is mainly divided into
be listed by Kirchhoff’s Voltage Law (KVL):
three categories: (i) Current controlled nonlinear resistor. The
terminal voltage of such nonlinear resistor is a single value uL (t) + f (i(t)) + e(t) = u(t), (3)
function of its current. (ii) Voltage controlled nonlinear re-
sistor. The current of such nonlinear resistor is a single value where uL (t) = L di(t) dt is the inductor voltage which is
function of the voltage. (iii) Monotonic type nonlinear resistor, derived from the voltage and current relation of the inductor;
the volt ampere characteristic of such nonlinear resistor is e(t) = Kb ω(t) is the back electromotive force, and Kb is back
monotonically increasing or monotonically decreasing, and it electromotive force constant, where e(t) is generated by the
is controlled by current and voltage at the same time. Among armature winding cutting off the main flux when the armature
all the nonlinear resistors, current controlled nonlinear resistor rotates; u(t) is applied voltage to nonlinear DC motor system.

1551-3203 (c) 2018 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.
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/TII.2018.2805707, IEEE
Transactions on Industrial Informatics

[ h1 (k), . . . , hθ (k) ], hλ (k), λ = 1, . . . , θ are premise

variables; Nλm are the fuzzy sets with m = 1, . . . , r, λ =
1, . . . , θ and r is the number of fuzzy rules; Am , Bm , Cm ,
Dm , Lm , and Em , m = 1, . . . , r are known constant ma-
trices; ∆Am , ∆Cm , and ∆Lm , m = 1, . . . , r are parameter
 

uncertain matrices. In this paper, we consider these uncertain

matrices of the form
   
∆Am FAm


   
 ∆Cm  =  FCm  ∆x (k)Nx , (7)
∆Lm FLm
   

Fig. 3. Volt-ampere characteristic of nonlinear resistor R. where ∆x (k) is an uncertainty satisfying ∆Tx (k)∆x (k) ≤ I;
FAm , FCm , FLm , m = 1, . . . , r, and Nx are known constant
matrices.
By the torque balance equation of the nonlinear DC motor Denote
system, we have θ
Y
T (t) − TL (t) = J dω(t) (4) sm (h(k)) = Nλm (hλ (k)), m = 1, . . . , r,
dt ,
λ=1
where T (t) = Ka i(t) is the electromagnetic torque, and the ¡ ¢
where Nλm hλ (k) is the grade of membership function of
Ka is the torque constant, in which T (t) is the rotation
hλ (k) in Nλm .
torque formed on the rotor by the interaction between the
One is assumed that
magnetic flux of the rotating magnetic field and the rotor r
X
current; TL (t) = Bω(t) is the torque required when the motor sm (h(k)) > 0, sm (h(k)) > 0, m = 1, . . . , r.
drives the rotating load. It is called the load torque. m=1
In this paper, we consider a factual situation that there are
Let
the parameter perturbation and disturbance signal in the DC
sm (h(k))
motor system. Then, by combining (3) and (4), a dynamic υm (h(k)) = P
r , m = 1, . . . , r, (8)
model of the nonlinear DC motor system can be obtained as si (h(k))
follows: i=1

L di(t) then
dt = −f (i(t)) − (Kb + ∆b (t))ω(t) + u(t), (5) r
X
J dω(t)
dt = Ka i(t) − Bω(t) + w(t), υm (h(k)) ≥ 0, υm (h(k)) = 1, m = 1, . . . , r. (9)
where ∆b (t) is an uncertainty satisfying |∆b (t)| < Kb ; w(t) is m=1

the torque disturbance caused by the variations of the practical The T-S fuzzy model (6) is deduced as follows:
working conditions [46]. By selecting armature current i(t) x(k + 1) = (A(υ(k)) + ∆A (υ(k)))x(k) + B(υ(k))w(k),
and rotation speed ω(t) as the system states and u(t) as y(k) = (C(υ(k)) + ∆C (υ(k)))x(k) + D(υ(k))w(k),
the control input, we can get the state space equation for z(k) = (L(υ(k)) + ∆L (υ(k)))x(k) + E(υ(k))w(k),
the nonlinear DC motor system. Furthermore, since R is (10)
a passive resistance, without w(t) the autonomous system where ∆χ (υ(k)) = Fχ (υ(k))∆x (k)Nx , χ = A, C, L.
is asymptotically stable [47], then one always can choose
u(t) = 0 for studying the filtering problem. C. Quantizer
The networked filtering problem investigated in this paper
B. T-S Fuzzy Model is shown in Fig. 1, where the measurement output signal y(k)
A T-S fuzzy model can be used to approximate the nonlinear and the performance output signal z(k) from the discrete-time
DC motor system (5). Of course, the T-S model should T-S fuzzy model are quantized by static quantiers Q(y(k))
be discretized by taking into account the transmission of and Q(z(k)), respectively, before transmission through the
signals over the communication channel. Then, considering digital communication channel. In this paper, we consider the
uncertainties in system parameters, the following uncertain static logarithmic and time-invariant quantizer [48] to quantize
discrete-time T-S fuzzy model is given to study the networked signals. The quantization levels have the following form:
filtering: U = {±ui , ui = ρi u0 , i = 0, ±1, ±2, ...} ∪ {±u0 } ∪ {0},
m (11)
R : if h1 (k) is N1m and . . . hθ (k) is Nθm , then
x(k + 1) = (Am + ∆Am )x(k) + Bm w(k), where 0 < ρ < 1 is quantization density and u0 > 0. The
(6)
y(k) = (Cm + ∆Cm )x(k) + Dm w(k), quantizer function Q(·) is defined as
z(k) = (Lm + ∆Lm )x(k) + Em w(k),  1 1

 ui , ui < ϑ ≤ ui ,
 1+ξ 1−ξ
where x(k) ∈ Rn are the state variable, y(k) ∈ Rp is
Q(ϑ) = (12)
the measurement output, z(k) ∈ Rs is the performance 
 0, ϑ = 0,

output, and w(k) ∈ Rq is the noise signal in l∞ ; h(k) = − Q(−ϑ), ϑ < 0,

1551-3203 (c) 2018 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.
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/TII.2018.2805707, IEEE
Transactions on Industrial Informatics

with At this point the purpose of this paper can be summarized

1−ρ
. ξ= (13) as the design of a filter like (19) for any uncertainties ∆x (k),
1+ρ ∆y (k), and ∆z (k) satisfying (7) and (18), respectively, such
To this end, the following two quantizers are employed in this that the filtering error system (20) satisfies the following two
paper: conditions:
h iT 1) When w(k) = 0, the filtering error system (20) is
qy (k) = Q(y(k)) = Q(y1 (k)) · · · Q(yp (k)) , asymptotically stable;
h iT (14)
qz (k) = Q(z(k)) = Q(z1 (k)) · · · Q(zs (k)) . 2) Under zero initial conditions, for any nonzero w(k) ∈ l∞
and given γ > 0, the error output satisfies
For the quantization effects, we resort to the sector bound
method proposed in [48] to deal with the quantization errors, ||e||∞ ≤ γ||w||∞ . (21)
the two quantizers described in (14) can be remodeled as Before carrying out the following work, we first introduce
qy (k) − y(k) = ∆y (k)y(k), the following lemma that is useful for deriving peak-to-peak
(15) filter design conditions.
qz (k) − z(k) = ∆z (k)z(k),
where Lemma 1. [49] For the given symmetry matrix Ω and the
∆y (k) = diag{ ∆1y (k), · · · , ∆py (k) }, ∆∗y (k) ∈ [ −ξy∗ ξy∗ ], matrices Ω1 , Ω2 , which have appropriate dimensions, the
∆z (k) = diag{ ∆1z (k), · · · , ∆sz (k) }, ∆∗z (k) ∈ [ −ξz∗ ξz∗ ]. following inequality holds
(16) Ω + Ω1 F Ω2 + ΩT2 F T ΩT1 < 0, (22)
By (15), we can obtain that
T
for all F satisfying F F ≤ I, if and only if there exists a
qy (k) = (I + ∆y (k))y(k),
(17) scalar ε > 0 to satisfy
qz (k) = (I + ∆z (k))z(k),
where Ω + ε−1 Ω1 ΩT1 + εΩT2 Ω2 < 0. (23)
|∆y (k)| ≤ Jy , Jy = diag{ ξy1 , ..., ξyp }, Remark 2. In this paper, the transmitted signals y(k) and
(18)
|∆z (k)| ≤ Jz , Jz = diag{ ξz1 , ..., ξzs }. z(k) through the communication channel are quantized when
Remark 1. From [48], the quantization errors ∆y (k) and studying the networked filtering problem. It is an important
∆z (k) are modeled as a kind of norm-bounded uncertainties in contribution relative to the existing results on filtering with
(17), it implies that the quantization effects considered in this quantization in which only the measurement output is quan-
paper can be effectively handled by using the robust technique. tized. On the other hand, parameter uncertainties have been
taken into consideration in the fuzzy system (10), which implies
D. Filter that the proposed design is of more universal importance for
As the system measurement output y(k) and the perfor- the practical applications.
mance output z(k), the premise variables or the membership
functions also will be quantized before they arrive at the filter III. P EAK - TO - PEAK FILTERING PERFORMANCE ANALYSIS
even though they are measurable [41]. So, this paper uses the Theorem 1. For a given scalar γ > 0 and consider the
following filter to estimate z(k): filtering error system in (20). The filtering error system is
xf (k + 1) = Af xf (k) + Bf qy (k), asymptotically stable with the peak-to-peak performance index
(19) γ, if there exist matrices P (υ(k)) > 0 and P (υ(k + 1)) > 0,
zf (k) = Cf xf (k) + Df qy (k),
scalars α ∈ (0, 1), ζ(υ(k)) ∈ (0, 1), and ζ(υ(k + 1)) ∈ (0, 1)
where xf (k) and zf (k) are the state and the output of the filter, meeting the following two matrix inequalities
respectively; Af , Bf , Cf , and Df are filter gain matrices to · ¸
be designed. −αP (υ(k)) + AT P (υ(k + 1))A AT P (υ(k + 1))B
< 0,
∗ Ψ1
E. Filtering Error System · ¸ (24)
−αP (υ(k)) + γ −2 C T C γ −2 C T D
Combine (10), (17) and (19), the filtering error system is < 0, (25)
∗ Ψ2
given by
ψ(k + 1) = Aψ(k) + Bw(k), where Ψ1 = −(ζ(υ(k + 1)) − αζ(υ(k)))I + B T P (υ(k + 1))B
(20)
e(k) = Cψ(k) + Dw(k), and Ψ2 = −(1 − αζ(υ(k)))I + γ −2 DT D.
where Proof. Suppose that inequality (24) holds. Pre-multiplying ΞT
ψ(k) ·= [ xT (k) xTf (k) ]T , e(k) = qz (k) − zf (k), ¸ and post-multiplying Ξ to (24) with Ξ = [ ψ T (k) wT (k)]T 6=
A(υ(k)) + ∆A (υ(k)) 0 0, one obtains
A= ,
· B f (I + ∆ y (k))(C(υ(k))¸ + ∆ C (υ(k))) A f (Aψ(k) + Bw(k))T P (υ(k + 1))(Aψ(k) + Bw(k))
B(υ(k)) −αψ T (k)P (υ(k))ψ(k) < (ζ(υ(k + 1)) (26)
B= ,
Bf (I + ∆y (k))D(υ(k)) − αζ(υ(k)))wT (k)w(k).
C = [ (I + ∆z (k))(L(υ(k)) + ∆L (υ(k))) − F − Cf ],
F = Df (I + ∆y (k))(C(υ(k)) + ∆C (υ(k))), By picking a fuzzy Lyapunov function as
D = (I + ∆z (k))E(υ(k)) − Df (I + ∆y (k))D(υ(k)). V (ψ(k)) = ψ T (k)P (υ(k))ψ(k) with P (υ(k)) > 0,

1551-3203 (c) 2018 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.
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/TII.2018.2805707, IEEE
Transactions on Industrial Informatics

and recalling (20), one has V (ψ(k + 1)) = IV. P EAK - TO - PEAK FILTER DESIGN
(Aψ(k) + Bw(k))T P (υ(k + 1))(Aψ(k) + Bw(k)), then (26)
can be rewritten as Note that the both matrices A and C contain two types
of uncertainties in the matrix inequalities (24) and (25),
V (ψ(k + 1)) − αV (ψ(k)) < (ζ(υ(k + 1))
(27) respectively, which need be eliminated to get LMI-based filter
− αζ(υ(k)))wT (k)w(k). design conditions.
Because of α < 1, so we can get that V (ψ(k + 1)) − By applying Schur complement to the matrix inequality
V (ψ(k)) < V (ψ(k + 1)) − αV (ψ(k)), it with (27) gives (24), one has
 
V (ψ(k+1))−V (ψ(k)) < (ζ(υ(k+1))−αζ(υ(k)))wT (k)w(k). −αP (υ(k)) 0 AT
(28)  ∗ Ψ3 BT  < 0, (37)
−1
From (24), we can see that ζ(υ(k + 1)) − αζ(υ(k)) > 0, then ∗ ∗ −P (υ(k + 1))
(28) makes V (ψ(k + 1)) − V (ψ(k)) < 0 when w(k) = 0,
which implies that the filtering error system (20) is asymptot- where Ψ3 = −(ζ(υ(k + 1)) − αζ(υ(k)))I. Let us consider a
ically stable. nonsingular matrix G with appropriate dimension and denote
Next, we will discuss the peak-to-peak performance of the Λ = diag{I, I, G}. Use the congruence with Λ to (37), one
filtering error system (20) with w(k) 6= 0. By multiplying yields
α−k > 0 on both sides of (27), we have  
−αP (υ(k)) 0 A T GT
α−k V (ψ(k + 1)) − α−k+1 V (ψ(k))  ∗ Ψ3 B T GT  < 0. (38)
−k −k+1 T (29) −1 T
< (α ζ(υ(k + 1)) − α ζ(υ(k)))w (k)w(k). ∗ ∗ −GP (υ(k + 1))G
Take the sum on both sides of the matrix inequality (29) So as to remove uncertainties, we need to isolate the
from 0 to k − 1, we give uncertainties in the matrix inequality (38) before they are
P ³ −d
k−1 ´ eliminated. According to the definitions in (13) and (18), we
α V (ψ(d + 1)) − α−d+1 V (ψ(d)) affirm that the matrix Jy is nonsingular, then (38) can be
d=0
P ³ −d
k−1 ´ rewritten as follows:
< (α ζ(υ(d + 1)) − α−d+1 ζ(υ(d)))wT (d)w(d) n ∆ (k) o
d=0 y
P ³ −d
k−1 ´ Γ + He X Y < 0, (39)
Jy
= (α ζ(υ(d + 1)) − α−d+1 ζ(υ(d)))||w(d)||2
d=0
P ³ −d
k−1 ´ where
< supk ||w(k)||2 α ζ(υ(d + 1)) − α−d+1 ζ(υ(d)) .  
d=0 −αP (υ(k)) 0 AT11 GT
(30) Γ= ∗ Ψ3 T
B11 GT ,
After summing up for (30), one can be obtained as T
· ∗ ∗ −GP (υ(k + 1))G ¸
α−k+1 V (ψ(k)) − A(υ(k)) + ∆A (υ(k)) 0
³ αV (ψ(0)) ´ A11 =
Bf (C(υ(k)) + ∆C (υ(k))) Af
,
(31)
< supk ||w(k)||2 α−k+1 ζ(υ(k)) − αζ(υ(0)) .
B11 = [ B T (υ(k)) DT (υ(k))BfT ]T ,
Under zero initial conditions, from (31) the following con- X = [ 0 0 JTy [ 0 BfT ]GT ]T ,
clusion can be obtained: Y = [ [ C(υ(k)) + ∆C (υ(k)) 0 ] D(υ(k)) 0 ].
α−k+1 V (ψ(k)) < supk ||w(k)||2 α−k+1 ζ(υ(k)), (32) ³ ´T
∆y (k) ∆y (k)
From (18), one has Jy Jy ≤ I. By using Lemma
that is 1 to (39), then which holds if and only if exists a scalar ε(υ(k+
V (ψ(k)) < ζ(υ(k))supk ||w(k)||2 . (33) 1)) > 0 such that
On the other hand, rewriting (25) leads to 1 T T
· ¸ · T¸ Γ+ ε(υ(k+1)) X X + ε(υ(k + 1))Y Y = Γ+
−αP (υ(k)) 0 −2 C
£ ¤ 1 T T T
+γ C D < 0. ε(υ(k+1)) [ X ε(υ(k + 1))Y ][ X ε(υ(k + 1))Y ] <0
0 −(1 − αζ(υ(k)))I DT (40)
(34) is satisfied. Applying Schur complement to the above inequal-
On the left of (34), multiplying ΞT and Ξ, from left and right, ity gives
respectively, we have · ¸
Γ [ X ε(υ(k + 1))Y T ]
< 0. (41)
γ −2 ||e(k)||2 < αV (ψ(k))+(1−αζ(υ(k)))||w(k)||2 , ∀k ∈ Z+ . ∗ −ε(υ(k + 1))I
(35)
Bringing (33) into (35) yields Notice that the uncertainty ∆y (k) dose not exist in (41) by
using Lemma 1, and only the uncertainty ∆x (k) is left. Next,
γ −2 ||e(k)||2 < αζ(υ(k))supk ||w(k)||2 the same method as eliminating ∆y (k) is used to eliminate
(36)
+ (1 − αζ(υ(k)))supk ||w(k)||2 , ∀k ∈ Z+ . ∆x (k), we gain
In (36), taking the supremum over k ∈ Z+ yields ||e||∞ < · ¸
T [ T1 η(υ(k + 1))T2T ]
γ||w||∞ . The proof is completed. < 0, (42)
∗ −η(υ(k + 1))I

1551-3203 (c) 2018 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.
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/TII.2018.2805707, IEEE
Transactions on Industrial Informatics

where is satisfied, where

· ¸
· ¸ Π [ R βLT ]
J = ,
Γ [ X ε(υ(k + 1))Y1T ] ∗ −βI
T = 1 , 
∗ −ε(υ(k + 1))I  −αP (υ(k)) 0 VT
−αP (υ(k)) 0 AT12 GT Π= ∗ −(1 − αζ(υ(k)))I D11 T 
,
Γ1 =  ∗ Ψ3 T
B12 G T  , ∗ ∗ −γ 2 I
· ∗ ∗ ¸ −GP −1 (υ(k + 1))GT V = [ L(υ(k)) − D̄f C(υ(k)) − C̄f ],
A 0 · ¸T
A12 = , 0 0 JTz
Bf C(υ(k)) Af R = ,
· 0 0 −JTy D̄fT ¸
Y1 = [ [ C(υ(k)) 0 ] D(υ(k)) 0 ], [ L(υ(k)) 0 ] E(υ(k)) 0
T T
B12 = [ B (υ(k)) D (υ(k))Bf ] , T T L = ,
T T T T
[ C(υ(k)) 0 ] D(υ(k)) 0
T1 = [ 0 0 [ FA (υ(k)) FC (υ(k))Bf ]G D11 = E(υ(k)) − D̄f D(υ(k)),
[ 0 ε(υ(k + 1))FCT (υ(k)) ] ]T , J2 = [ [ Nx 0 ] 0 0 0 ],
T2 = [ [ Nx 0 ] 0 0 0 ]. J1 = [ Φ [ 0 0 βFLT (υ(k)) βFCT (υ(k)) ] ]T ,
Φ = [ 0 0 FLT (υ(k)) − FCT (υ(k))D̄fT ].
In order to make the design of peak-to-peak filter easier From (44) and (45), we have
to achieve, we need to eliminate the coupling items in (42). Pr Pr
Constructing two matrix variables as υm (h(k))υl (h(k + 1))Θml < 0, (46)
m=1 l=1

· ¸ · ¸ and
G1 G2 P ∗ P
r
G= , Pm = m1 >0 (43) υm (h(k))Υm < 0, (47)
G4 G2 Pm2 Pm3 m=1

where
· ¸
and defining two new matrix variables Āf = G2 Af , B̄f = Ψml [ T1ml ηl T2T ]
Θml = ,
G2 Bf . From the fact that −(G − P (υ(k + 1)))T P −1 (υ(k + · ∗ −ηl I ¸
1))(G − P (υ(k + 1))) ≤ 0 with P (υ(k + 1)) > 0 means T
Γ [ X1 εl Y1m ]
Ψml = 2ml ,
−GP −1 (υ(k + 1))GT ≤ −G − GT + P (υ(k + 1)), we know ∗ −εl I
that (42) can be verified by Y1m = [ [ Cm 0 ] Dm 0 ],
T1ml = [·0 0 MTm [ 0 εl¸FCm T
] ]T ,
· ¸ G1 FAm + B̄f FCm
U [ T1 η(υ(k + 1))T2T ] Mm = ,
< 0, (44) G4 FAm + B̄f FCm 
∗ −η(υ(k + 1))I
−αPm 0 ATGm
Γ2ml =  ∗ −(ζl − αζm )I T
BGm ,
T
where · ∗ ∗ ¸ −G − G + Pl
G1 Am + B̄f Cm Āf
· ¸ AGm = ,
Γ2 [ X1 ε(υ(k + 1))Y1T ] ·G4 Am + B̄f Cm ¸ Āf
U= , G1 Bm + B̄f Dm
∗ −ε(υ(k + 1))I  BGm = ,
−αP (υ(k)) 0 ATG · G4 Bm + B̄f DmT ¸ · ¸
Γ2 =  ∗ Ψ3 BGT , Jm [ J1m δJ2 ] Πm [ R βLTm ]
Υm = , Jm = ,
· ∗ T
∗ −G − G +¸P (υ(k + 1)) · ∗ −δI ¸ ∗ −βI
G1 A(υ(k)) + B̄f C(υ(k)) Āf [ Lm 0 ] Em 0
AG = , Lm = ,
·G4 A(υ(k)) + B̄f C(υ(k)) ¸ Āf  Cm 0 ] Dm 0
[
T

G1 B(υ(k)) + B̄f D(υ(k)) −αPm 0 Vm
BG =
G4 B(υ(k)) + B̄f D(υ(k))
, Πm =  ∗ −(1 − αζm )I D11m T ,
X1 = [ 0 0 JTy [ B̄fT B̄fT ] ]T , ∗ ∗ −γ 2 I
Vm = [ Lm − D̄f Cm − C̄f ],
T1 = [·0 0 MT [ 0 ε(υ(k + 1))F T T
¸ C (υ(k)) ] ] , D11m = Em − D̄f Dm ,
G1 FA (υ(k)) + B̄f FC (υ(k)) T T
M= . J1m = [ Φm [ 0 0 βFLm βFCm ] ]T ,
G4 FA (υ(k)) + B̄f FC (υ(k)) T T T
Φm = [ 0 0 FLm − FCm D̄f ],
and T2 , Pm (Pl ), X1 , J2 , and R have been defined in (42),
Analogous to the derivation process for the condition (44), (43), (44), and (45), respectively.
with defining C̄f = Cf and D̄f = Df the matrix inequality By recalling the property of the membership functions
(25) holds if the following condition described in (9), we can know that if (48) and (49) are
established, the matrix inequalities (46) and (47) are true.
· ¸
J [ J1 δJ2T ] Through the above derivation, we can get the following
<0 (45) condition to design the peak-to-peak filter (19):
∗ −δI

1551-3203 (c) 2018 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.
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/TII.2018.2805707, IEEE
Transactions on Industrial Informatics

Theorem 2. Consider the filtering error system in (20). For TABLE III
given constants γ > 0 and α ∈ (0, 1), if there exist scalars M INIMUM PEAK - TO - PEAK PERFORMANCES γmin UNDER DIFFERENT
VALUES OF SCALAR α
ζm (ζl ) ∈ (0, 1), εl > 0, ηl > 0, β > 0, and δ > 0, matrices
Pm1 (Pl1 ) > 0, Pm2 (Pl2 ), Pm3 (Pl3 ) > 0, G1 , G2 , G4 , Āf , α 0.88 0.89 0.90 0.91 0.92 0.93
B̄f , C̄f , and D̄f satisfying the following LMIs: γmin 8.3722 8.0521 7.8838 7.8382 7.8951 8.0536

Θml < 0, m, l = 1, ..., r, (48)

TABLE IV
Υm < 0, m = 1, ..., r, (49) M INIMUM PEAK - TO - PEAK PERFORMANCES γmin UNDER DIFFERENT
QUANTIZATION DENSITY ρy AND ρz
then the system is asymptotically stable and satisfies the peak-
ρy 0.2 0.3 0.4 0.5 0.6 0.7
to-peak filtering performance γ. In addition, the peak-to-peak
ρz 0.2 0.3 0.4 0.5 0.6 0.7
filter gain matrices in (19) can be obtained by the following ξy 0.6667 0.5385 0.4286 0.3333 0.2500 0.1765
equations ξz 0.6667 0.5385 0.4286 0.3333 0.2500 0.1765
γmin 14.3302 12.5009 10.9910 9.7425 8.7041 7.8382
Af = G−1 −1
2 Āf , Bf = G2 B̄f , Cf = C̄f , Df = D̄f .
(50)
Remark 3. In Theorem 2, for a given α ∈ (0, 1), the peak- The response of the system states x1 (t) and x2 (t) in (52), with
to-peak filter design conditions (48) and (49) are strict LMIs. w(t) = 0 and initial conditions x1 (0) = 3, x2 (0) = 0.5, is
Therefore, the gain matrices and performance γ of the peak- illustrated in Fig. 4, which shows that the response converges
to-peak filter can be obtained by Matlab LMI control toolbox bear off the stable equilibrium point x1 (t) = x2 (t) = 0, which
[50]. implies that the system is stable.

Remark 4. Theorem 2 has proposed a novel peak-to-peak fil-   

ter design criterion for the filtering error system in (20), which
# $ % & '

( ) * + ,

allows that the Lyapunov matrix P , auxiliary variables ζ, ε,   

and η are multiple depending on the membership function of 

the fuzzy system. The multiplicity might cause less conservative   

design for peak-to-peak filtering. 

Remark 5. As the auxiliary variables ε and η in (44), δ and β



in (45) are also allowed to be multiple, however, since P (υ(k+

1)) is not included in (25), we choose them to be common in  

    

order not to increase the number of matrix inequalities.

      ! "

V. S IMULATION Fig. 4. Response of the states for the nonlinear system (52).

In this simulation section, we consider the volt-ampere

Let us rewrite (52) as
characteristic of the armature resistor of the nonlinear DC · ¸
motor system (5) as (2), and the related parameters of the −4 − x21 (t) −(0.612 + 0.06b(t))
ẋ(t) = x(t)
nonlinear DC motor system are given in Table II. 0.576/0.082 −0.3/0.082 · ¸
0
+ w(t).
TABLE II 1/0.082
N ONLINEAR DC M OTOR SYSTEM PARAMETERS (53)
parameter value unit For the nonlinear system (53), based on the sector nonlinearity
J 0.082 kg · m2 approach presented in [51], one can assume that −3 ≤ x1 (t) ≤
B 0.3 N · m · s/rad 3, that is, −13 ≤ −4−x21 (t) ≤ −4. By fuzzy modeling for the
L 1000 mH nonlinear system, the state equation (53) can be represented
Ka 0.576 N · m/A
Kb 0.612 V · s/rad by the following T-S model with two fuzzy rules:
2
X
Let x1 (t) = i(t) and x2 (t) = ω(t) be the state variable, then ¡ ¢
ẋ(t) = υm (h(t)) Amb x(t) + Bmb w(t) , (54)
the system (5) with u(t) = 0 is governed by the following state m=1
equations:
where x(t) = [ x1 (t) x2 (t) ]T and
Lẋ1 (t) = −uR (t) − (Kb + ∆b (t))x2 (t), · ¸
(51)
J ẋ2 (t) = Ka x1 (t) − Bx2 (t) + w(t). −4 −(0.612 + 0.06b(t))
A1b = ,
With these parameters in Table II and ∆b (t) = 0.06b(t) (−1 ≤ ·0.576/0.082 −0.3/0.082 ¸
b(t) ≤ 1), we have −13 −(0.612 + 0.06b(t))
A2b = ,
0.576/0.082
· ¸ −0.3/0.082
ẋ1 (t) = −4x1 (t) − x31 (t) − (0.612 + 0.06b(t))x2 (t), 0
ẋ2 (t) = (0.576/0.082)x1 (t) − 0.3/0.082x2 (t) B1b = B2b = .
1/0.082
+ 1/0.082w(t).
(52) By applying the maximum and minimum values of x21 (t),

1551-3203 (c) 2018 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.
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/TII.2018.2805707, IEEE
Transactions on Industrial Informatics

υ1 (h1 (t)) and υ2 (h1 (t)) with h1 (t) = x1 (t) can be described of z(k) and zf (k). Fig. 8 displays the response of error signal
as e(k), it can be perceived
p that the error curve tends to 0. The
x21 (t) = υ1 (h1 (t)) × 9 + υ2 (h1 (t)) × 0, simulation ratio of eT (k)e(k)/||w||2∞ is shown in Fig. 9. It
(55)
υ1 (h1 (t)) + υ2 (h1 (t)) = 1. can be known that the maximum value of the ratio is 0.9984
Then, the membership function can be given by that is smaller than 7.8382. From the simulation results of this
example, we can gain the conclusion that the design method
υ1 (h1 (t)) = x21 (t)/9, υ2 (h1 (t)) = 1 − x21 (t)/9. (56) of the peak-to-peak filter proposed in this paper is effective.
In the following, we will study the networked peak-to-peak 

filtering problem for the system (54). By taking a sampling     

time T = 0.05s, we get a discrete-time T-S fuzz system as 

    

(10) and the corresponding system parameter matrices are

· ¸ · ¸ 

0.8 −0.0306 0
A1 = , B1 = ,
·0.3512 0.8171 ¸ ·0.05/0.082¸
 

0.35 −0.0306 0
A2 = , B2 = ,
0.3512 0.8171 0.05/0.082 (57)
 

C1 = [ −3 0.5 ], D1 = 2.5, E1 = 1,
C2 = [ 1.5 2 ], D2 = −2.3, E2 = −0.5,

      

L1 = [ 0.2 1 ], L2 = [ 0.3 1 ],
Fig. 5. Response of system states.
and the following parameters are also considered in (7)
FA1 = FA2 = [ −0.6 0 ]T , 

FC1 = FC2 = 0.01,      

(58)
FL1 = FL2 = 3.5, 
    

Nx = [ 0 0.1 ].


Table III shows the minimum peak-to-peak attenuation level

obtained by solving LMIs (48) and (49) with α1 = 0.88, α2 = 

0.89, α3 = 0.90, α4 = 0.91, α5 = 0.92, and α6 = 0.93 under

the same quantization density ρy = ρz = 0.7. From Table  

III, we can get that γmin decreases first and then increases as
the scalar α increases. The minimum peak-to-peak attenuation

      

levels obtained by solving (48) and (49) with ρy = ρz = 0.2,

ρy = ρz = 0.3, ρy = ρz = 0.4, ρy = ρz = 0.5, ρy = ρz = Fig. 6. Response of filter states.
0.6, and ρy = ρz = 0.7 are shown in Table IV under the
α = 0.91. From this Table, we can see that γmin increases as 

the quantization density ρy and ρz decrease when the α is fixed    

to the constant value 0.91. Moreover, when γmin = 7.8382,

    

the related matrix variables of Theorem 2 can be obtained as: 

· ¸ · ¸
0.1013 −0.0267 0.0049
Āf = , B̄f = ,


−0.0231 0.0077 −0.0020

C̄f = [·0.4922 − 0.1798 ],¸ D̄f = 0.4569,
 

0.1268 −0.0325
G2 = .
 

−0.0323 0.0102

      

Bring (50) into the above formulas, we have 

· ¸ · ¸
1.1671 −0.0851 −0.0633 Fig. 7. Response of z(k) and zf (k).
Af = , Bf = ,
1.4381 0.4899 −0.3976
Cf = [ 0.4922 − 0.1798 ], Df = 0.4569. Comparative Explanations: The proposed method in this
paper gives effective design for quantized peak-to-peak filter-
Assume that the initial conditions x(0) = xf (0) = [ 0 0 ]T ,
ing for a class of nonlinear DC motor systems based on the
the uncertainties ∆x (k) = 0.6cos(0.8k), ∆y (k) = 0.1sin(k),
T-S fuzzy model. Compared with some existing results, the
and ∆z (k) = 0.1sin(0.8k), and the load disturbance is
 main advantages of the proposed method can be summarized
 0.5,
 0 ≤ k ≤ 5, as: 1) The proposed fuzzy peak-to-peak filtering result is based
w(k) = −1, 5 < k ≤ 10, (59) on the multiple Lyapunov matrix and auxiliary variables that

 provides less conservative design. As a comparison, if the
0, otherwise.
condition in [44] with the common Lyapunov matrix and
Finally, the response results of x(k) and xf (k) are displayed auxiliary variables is applied to design the peak-to-peak filter
in Fig. 5 and Fig. 6, respectively. Fig. 7 displays the responses for the fuzzy system (10) with (57) and (58), the minimum

1551-3203 (c) 2018 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.
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/TII.2018.2805707, IEEE
Transactions on Industrial Informatics

R EFERENCES


  

[1] X.-M. Zhang, Q.-L. Han, and B.-L. Zhang, “An overview and deep
investigation on sampled-data-based event-triggered control and filtering
for networked systems,” IEEE Transactions on Industrial Informatics, vol.


13, no. 1, pp. 4–16, Feb. 2017.

   

[2] X.-M. Zhang and Q.-L. Han, “Event-triggered H∞ control for a class
of nonlinear networked control systems using novel integral inequalities,”
 
International Journal of Robust and Nonlinear Control, vol. 27, no. 4,
pp. 679–700, Mar. 2017.
[3] Y.-L. Wang and Q.-L. Han, “Network-based fault detection filter and
controller coordinated design for unmanned surface vehicles in network
 

      

environments,” IEEE Transactions on Industrial Informatics, vol. 12, no.



5, pp. 1753–1765, Oct. 2016.

Fig. 8. Filtering error e(k). [4] C. Peng, D. Yue, and M.-R. Fei, “A higher energy-efficient sampling
scheme for networked control systems over IEEE 802.15.4 wireless
  

networks,” IEEE Transactions on Industrial Informatics, vol. 12, no. 5,

pp. 1766–1774, Oct. 2016.
[5] X. Zhao, C. Ma, X. Xing, and X. Zheng, “A stochastic sampling
consensus protocol of networked euler-lagrange systems with application


to two-link manipulator,” IEEE Transactions on Industrial Informatics,

vol. 11, no. 4, pp. 907–914, Aug. 2015.
[6] H. Zhang, Q. Hong, H. Yan, F. Yang, and G. Guo, “Event-based
distributed H∞ filtering networks of 2-DOF quarter-car suspension
systems,” IEEE Transactions on Industrial Informatics, vol. 13, no. 1,
 

pp. 312–321, Feb. 2017.

[7] H. Yan, F. Qian, H. Zhang, F. Yang, and G. Guo, “H∞ fault detection for
networked mechanical spring-mass systems with incomplete information,”

      

IEEE Transactions on Industrial Electronics, vol. 63, no. 9, pp. 5622–

5631, Sep. 2016.


p [8] X.-H. Chang, Z.-M. Li, J. Xiong, and Y.-M. Wang, “LMI approaches
Fig. 9. Ratio of eT (k)e(k)/||w||2∞ . to input and output quantized feedback stabilization of linear systems,”
Applied Mathematics and Computation, vol. 315, pp. 162–175, Dec. 2017.
[9] H. Gao and T. Chen, “A new approach to quantized feedback control
systems,” Automatica, vol. 44, no. 2, pp. 534–542, Feb. 2008.
attenuation level γ is 8.0443 that is larger than 7.8382; 2) The [10] C. Zhang, G. Feng, H. Gao, and J. Qiu, “H∞ filtering for nonlinear
proposed quantized filtering design is effective for uncertain discrete-time systems subject to quantization and packet dropouts,” IEEE
T-S fuzzy systems with both the measurement output signal Transactions on Fuzzy Systems, vol. 19, no. 2, pp. 353–365, Apr. 2011.
[11] D. Zhang, Q.-G. Wang, D. Srinivasan, H. Li, and L. Yu, “Asynchronous
and the performance output signal. Obviously, since the system state estimation for discrete-time switched complex networks with com-
outputs y(k) and z(k) contain parameter uncertainties in the munication constraints,” IEEE Transactions on Neural Networks and
fuzzy system (20), the results in [10] – [13] are infeasible for Learning Systems, doi: 10.1109/TNNLS.2017.2678681.
[12] D. Zhang, S. K. Nguang, D. Srinivasan, and L. Yu, “Dis-
quantized filtering design of the system. tributed filtering for discrete-time T-S fuzzy systems with incom-
plete measurements,” IEEE Transactions on Fuzzy Systems, doi:
10.1109/TFUZZ.2017.2725228.
VI. CONCLUSION [13] Z. Wang, H. Dong, B. Shen, and H. Gao, “Finite-horizon H∞ filtering
with missing measurements and quantization effects,” IEEE Transactions
on Automatic Control, vol. 58, no. 7, pp. 1707–1718, Jul. 2013.
In this paper, the peak-to-peak filter design problem of a [14] L. Zhang, Z. Ning, and W. X. Zheng, “Observer-based control for
class of networked nonlinear DC motor systems under the piecewise-affine systems with both input and output quantization,” IEEE
transmission of digital communication channel is studied. The Transactions on Automatic Control, vol. 62, no. 11, pp. 5858–5865, Nov.
2017.
nonlinear DC motor system is represented by a T-S fuzzy [15] M. S. Mahmoud and N. B. Almutairi, “Feedback fuzzy control for
model and two static quantizers are used to quantize the quantized networked systems with random delays,” Applied Mathematics
measurement output signal and the performance output signal, and Computation, vol. 290, pp. 80–97, Nov. 2016.
respectively. Useful method has been applied to remove the [16] K. Cheon, J. Kim, M. Hamadache, and D. Lee, “On replacing PID
controller with deep learning controller for DC motor system,” Journal
uncertainties and obtain the conditions for devising a peak-to- of Automation and Control Engineering, vol. 3, no. 6, pp. 452–456, Dec.
peak filter such that the filtering error system is asymptotically 2015.
stable and satisfies a given peak-to-peak performance. A [17] C. Yuan and F. Wu, “Delay scheduled impulsive control for networked
control systems,” IEEE Transactions on Control of Network Systems, vol.
simulation has been provided to prove the validity of the 4, no. 3, pp. 587–597, Sep. 2017.
proposed peak-to-peak filtering method. [18] H. Ahmed and A. Rajoriya, “A hybrid of sliding mode control and fuzzy
gain scheduling PID control using fuzzy supervisory switched system
Note that the phenomena of packet dropout and transmis- for DC motor speed control system,” International Journal of Grid and
sion delay are also common in networked control systems Distributed Computing, vol. 9, no. 5, pp. 41–54, 2016.
in addition to transmission signal quantization. Considering [19] X. Zhao, Y. Yin, L. Zhang, and H. Yang, “Control of switched nonlinear
the packet loss, and transmission delay in peak-to-peak filter systems via T-S fuzzy modeling,” IEEE Transactions on Fuzzy Systems,
vol. 24, no. 1, pp. 235–241, Feb. 2016.
design problem of the networked system also has practical [20] C. Peng, S. Ma, and X. Xie, “Observer-based non-PDC control for
significance. The developed results in this paper can also networked T-S fuzzy systems with an event-triggered communication,”
be used to handle networked filtering problems with packet IEEE Transactions on Cybernetics, vol. 47, no. 8, pp. 2279–2287, Aug.
2017.
dropout and delay from the point of view of LMI, it is true [21] J. Qiu, S. X. Ding, H. Gao, and S. Yin, “Fuzzy-model-based reliable
that detailed work requires further study. static output feedback H∞ control of nonlinear hyperbolic PDE systems,”

1551-3203 (c) 2018 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.
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/TII.2018.2805707, IEEE
Transactions on Industrial Informatics

IEEE Transactions on Fuzzy Systems, vol. 24, no. 2, pp. 388–400, Apr. [46] A. E. Fitzgerald, C. Kingsley, S. D. Umans, and B. James, Electric
2016. machinery. New York: McGraw-Hill, 2003.
[22] X. Xie, D. Yue, T. Ma, and X. Zhu, “Further studies on control synthesis [47] M. Namazov, “DC motor position control using fuzzy proportional-
of discrete-time T-S fuzzy systems via augmented multi-indexed matrix derivative controllers with different defuzzification methods,” Turkish
approach,” IEEE Transactions on Cybernetics, vol. 44, no. 12, pp. 2784– Journal of Fuzzy Systems vol. 1, no. 1, pp. 36–54, 2010.
2791, Dec. 2014. [48] M. Fu and L. Xie, “The sector bound approach to quantized feedback
[23] X. Su, L. Wu, and P. Shi, “Sensor networks with random link failures: control,” IEEE Transactions on Automatic Control, vol. 50, no. 11, pp.
Distributed filtering for T-S fuzzy systems,” IEEE Transactions on Indus- 1698–1711, Nov. 2005.
trial Informatics, vol. 9, no. 3, pp. 1739–1750, Aug. 2013. [49] S. Xu, P. V. Dooren, R. Stefan, and J. Lam, “Robust stability and stabi-
[24] X. Su, H. Zhou, and Y.-D. Song, “An optimal divisioning technique to lization for singular systems with state delay and parameter uncertainty,”
stabilization synthesis of T-S fuzzy delayed systems,” IEEE Transactions IEEE Transactions on Automatic Control, vol. 47, no. 7, pp. 1122–1128,
on Cybernetics, vol. 47, no. 5, pp. 1147–1156, May 2017. Jul. 2002.
[25] Y. Xue, B.-C. Zheng, and X. Yu, “Robust sliding mode control for T-S [50] P. Gahinet, A. Nemirovski, A. J. Laub, and M. Chilali, LMI Control
fuzzy systems via quantized state feedback,” IEEE Transactions on Fuzzy Toolbox. The MathWorks, Inc, Natick, MA, 1995.
Systems, doi: 10.1109/TFUZZ.2017.2771467. [51] K. Tanaka and H. O. Wang, Fuzzy Control Systems Design and Analysis:
[26] X. Xie, D. Yang, and H. Ma, “Observer design of discrete-time T-S A Linear Matrix Inequality Approach. New York: Wiley, 2001.
fuzzy systems via multi-instant homogenous matrix polynomials,” IEEE
Transactions on Fuzzy Systems, vol. 22, no. 6, pp. 1714–1719, Dec. 2014.
[27] B. Ding and H. Pan, “Dynamic output feedback-predictive control of a
Takagi-Sugeno model with bounded disturbance,” IEEE Transactions on
Fuzzy Systems, vol. 25, no. 3, pp. 653–667, Jun. 2017.
[28] X.-H. Chang and G.-H. Yang, “Nonfragile H∞ filtering of continuous-
time fuzzy systems,” IEEE Transactions on Signal Processing, vol. 59,
no. 4, pp. 1528–1538, Apr. 2011.
[29] D. Zhang, Q.-G. Wang, L. Yu, and Q.-K. Shao, “H∞ filtering for
networked systems with multiple time-varying transmissions and random
packet dropouts,” IEEE Transactions on Industrial Informatics, vol. 9,
no. 3, pp. 1705–1716, Aug. 2013.
[30] Z. Wang, F. Yang, D. W. C. Ho, and X. Liu, “Robust H∞ filtering
for stochastic time-delay systems with missing measurements,” IEEE
Transactions on Signal Processing, vol. 54, no. 7, pp. 2579–2587, Jul.
2006. Xiao-Heng Chang (M-11) received the B.E. and
[31] Z. Duan, J. Zhang, C. Zhang, and E. Mosca, “Robust H2 and H∞ M.S. degrees from Liaoning Technical University,
filtering for uncertain linear systems,” Automatica, vol. 42, no. 11, pp. Fuxin, China, in 1998 and 2004, respectively, and
1919–1926, Nov. 2006. the Ph.D. degree from Northeastern University,
[32] J. Zhang and Y. Xia, “New LMI approach to fuzzy H∞ filter designs,” Shenyang, China, in 2007. He is currently a Pro-
IEEE Transactions on Circuits and Systems II: Express Briefs, vol. 56, fessor with the School of Information Science and
no. 9, pp. 739–743, Sept. 2009. Engineering, Wuhan University of Science and Tech-
[33] X.-M. Zhang and Q.-L. Han, “Robust H∞ filtering for a class of nology, Wuhan, China. He is also a Chutian Scholars
uncertain linear systems with time-varying delay,” Automatica, vol. 44, Distinguished Professor of Hubei Province, China.
no. 1, pp. 157–166, Jan. 2008. His research interests include fuzzy control systems,
[34] X.-M. Zhang and Q.-L. Han, “Network-based H∞ filtering using a logic networked control systems and robust control.
jumping-like trigger,” Automatica, vol. 49, no. 5, pp. 1428–1435, May
2013.
[35] D. Zhang, P. Shi, W.-A. Zhang, and L. Yu, “Energy-efficient distributed
filtering in sensor networks: A unified switched system approach,” IEEE
Transactions on Cybernetics, vol. 47, no. 7, pp. 1618–1629, Jul. 2017.
[36] X.-H. Chang, “Robust nonfragile H∞ filtering of fuzzy systems with
linear fractional parametric uncertainties,” IEEE Transactions on Fuzzy
Systems, vol. 20, no. 6, pp. 1001–1011, Dec. 2012.
[37] X.-H. Chang, J. H. Park, and P. Shi, “Fuzzy resilient energy-to-peak
filtering for continuous-time nonlinear systems,” IEEE Transactions on
Fuzzy Systems, vol. 25, no. 6, pp. 1576–1588, Dec. 2017.
[38] Z.-M. Li, X.-H. Chang, K. Mathiyalagan, and J. Xiong, “Robust energy-
to-peak filtering for discrete-time nonlinear systems with measurement
quantization,” Signal Processing, vol. 139, pp. 102–109, Oct. 2017.
[39] J. Feng and K. Han, “Robust full- and reduced-order energy-to-peak Yi-Ming Wang received the B.E. degree from Col-
filtering for discrete-time uncertain linear systems,” Signal Processing, lege of Science & Arts of Jianghan University,
vol. 108, pp. 183–194, Mar. 2015. Wuhan, China, in 2016. He is currently working
[40] S. He, “Energy-to-peak filtering for T-S fuzzy systems with Markovian toward the M.S. degree in the School of Informa-
jumping: The finite-time case,” Neurocomputing, vol. 168, pp. 348–355, tion Science and Engineering, Wuhan University
Nov. 2015. of Science and Technology, Wuhan, China. His
[41] X.-H. Chang, Z.-M. Li, and J. H. Park, “Fuzzy generalized H2 filtering research interests include networked control systems
for nonlinear discrete-time systems with measurement quantization,” and fuzzy control systems.
IEEE Transactions on Systems, Man, and Cybernetics: Systems, doi:
10.1109/TSMC.2017.2743012.
[42] M. Vidyasagar, “Optimal rejection of persistent bounded disturbances,”
IEEE Transactions on Automatic Control, vol. 31, no. 6, pp. 527–534,
Jun. 1986.
[43] K. Nagpal, J. Abedor, and K. Poolla, “An LMI approach to peak-to-peak
gain minimisation: filtering and control,” American Control Conference,
1994.
[44] U. Shaked and I. Yaesh, “Robust servo synthesis by minimization of
induced l2 and l∞ norms,” IEEE International Symposium on Industrial
Electronics, pp. 24–29.
[45] S. He and F. Liu, “Robust peak-to-peak filtering for Markov jump
systems,” Signal Processing, vol. 90, no. 2, pp. 513–522, Feb. 2010.

1551-3203 (c) 2018 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.