SCIENCE CHINA

Information Sciences
. RESEARCH PAPER . June 2022, Vol. 65 162203:1–162203:17
https://doi.org/10.1007/s11432-021-3375-5

Event-based sliding mode control under

denial-of-service attacks
Yingxin TIAN1,2 , Xiang LI1 , Bo DONG1, Yabin GAO1* & Ligang WU1
1
Department of Control Science and Engineering, Harbin Institute of Technology, Harbin 150001, China;
2
School of Software, Harbin Institute of Technology, Harbin 150001, China

Received 11 May 2021/Revised 24 July 2021/Accepted 8 October 2021/Published online 18 April 2022

Abstract This paper deals with the problem of estimator-based sliding mode control against denial-of-
service (DoS) attacks and discrete events via a time-delay approach. A networked system is considered an
uncertain dynamical system with matched and mismatched perturbations and exogenous disturbance in the
network environment. A network-resource-aware event-triggering mechanism is designed with aperiodically
releasing system measurements. Furthermore, to describe the DoS attack duration and inter-event time, a
time-delay modeling approach considers the DOS attack duration and inter-event time as a “time delay” of
the measurements between the sensor and controller over the network is proposed. Consequently, an interval-
time-delay system with uncertainties is formulated. A state-observer-based sliding mode controller, by which
the ideal sliding mode can be achieved, is proposed against the DoS attacks. The resulting sliding motion is
proved to be robust and stable with an L2 gain performance. Finally, the effectiveness and applicability of
the present sliding mode control are validated in a simulated pendulum system.
Keywords sliding mode control, time delay, uncertain system, state observer, event trigger

Citation Tian Y X, Li X, Dong B, et al. Event-based sliding mode control under denial-of-service attacks. Sci
China Inf Sci, 2022, 65(6): 162203, https://doi.org/10.1007/s11432-021-3375-5

1 Introduction
It is widely accepted that the emergence of networks brought significant development to traditional
control systems of point-to-point. The control loop forms a closed-loop control system through some
communication networks, which produce the so-called networked control systems (NCSs) [1–3]. Since the
1950s, great changes have taken place in NCSs. The accelerated integration of computing, communication,
and control has stimulated the interest of researchers from various disciplines in the emerging field of
NCSs [4, 5]. Generally, an NCS is composed of plants, sensors, controllers, and actuators [4]. Sometimes,
some filters or observers, as an estimator component, are also part of the components of the NCS. Among
these components, the sensor unit is responsible for obtaining information, the controller unit provides
commands or decisions, and the actuator unit executes the control commands or decisions. The estimator
is sometimes used as the auxiliary of the controller or sensor unit to provide the update of the controller
or sensor, and the communication network is responsible for exchanging information/data [6].
In signal transmission, compared with the analog implementation of signal, the advantage of NCS for
signal (data) transmission through the digital communication channel is that it increases system flex-
ibility and reliability, and reduces installation and maintenance cost. Consequently, significant studies
focus on the application of NCS in industrial control, transportation, aerospace, and other fields. Ad-
ditionally, there are outstanding achievements in applying complex control processes, such as advanced
aircraft, robots, building intelligence, telemedicine, remote teaching, military command, and manufac-
turing. However, due to the actual operation characteristics of these control processes, there are some
uncertainties in the operation of NCS, such as delay, multi-channel interference, congestion, timing disor-
der, data loss, and jitter. Even in the large-scale NCS, each system unit also faces complex uncertainties
and constraints, such as anonymous attacks, failures, and limitations of communication networks. For
* Corresponding author (email: [email protected])

c Science China Press and Springer-Verlag GmbH Germany, part of Springer Nature 2022 info.scichina.com link.springer.com
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:2

example, Liu et al. [7] investigated the state feedback controller design of nominal linear systems subject
to denial-of-service (DoS) attacks and hybrid-triggering communication strategy. By considering that
DoS attacks imposed by power-constrained pulse-width-modulated jammers are partially identified, Hu
et al. [8] studied the event-based state feedback control of networked linear systems under the resilient
event-triggering communication in a framework of a switched control system. Dong et al. [9] developed
a robust adaptive quantized control of networked Markov jump systems with parameter uncertainty,
matched uncertain nonlinearity, and disturbance.
Currently, there are many control theories and methods of the system with uncertainties. In addition
to the proportion-integral-differential (PID) control commonly used in practice, there are some advanced
control theories, such as fuzzy control, backstepping control, bang-bang control, sliding mode control
(SMC), internal model control, model predictive control, prescribed performance control, nonlinear con-
trol, and some of their combination control. As a powerful and robust control methodology dealing with
nonlinearity and uncertainties, i.e., matched system perturbations and exogenous disturbances, SMC
has been successfully applied to dynamical systems with complex uncertainties [10–13] in practical ap-
plications, such as robots, underwater robots, aircraft and spacecraft, flexible space structures, motors,
automotive engines, electronic systems, and fuel cell power systems. This successful application is due
to its insensitivity to parameter uncertainty, strong robustness to external disturbances, good transient
performance, and tracking ability [14, 15]. In SMC, a discontinuous control is utilized to force system
trajectories onto a predesigned sliding surface in finite time and then toward the system equilibrium
point asymptotically or boundedly. Thus, it possesses properties of fast response and excellent distur-
bance rejection [10, 16]. Benefiting from SMC for the networked systems, Liu et al. [17] dealt with the
observer-based SMC problem of linear Itô stochastic time-delay systems with a logarithmic quantization
between sensor and controller. Zhang et al. [18] solved the problem of SMC for linear NCSs with round
trip delays by using current and previous measurements and inputs for reconstructing state variables. Wu
et al. [19] studied the event-based SMC of networked stochastic systems based on state observer by using
a time-delay approach. Then, the problem of the dissipativity-based resilient SMC under DoS attacks
was addressed in [20,21]. Notably, few techniques were developed to secure SMC against the DoS attacks
when system states are unmeasurable in real NCSs. The main challenges dealing with such output feed-
back SMC against the DoS attacks are: how to construct a sliding surface that can be reached in finite
time due to the intermittency of the DoS attacks, and how to analyze the stability of the SMC-based
system in the complex network environment with mixed discrete events and network DoS attacks and
delays.
Based on the discussions above, this paper deals with secure SMC of the NCS with discrete events and
network DoS attacks via an output feedback approach. The network DoS attacks and delays between the
sensor and controller units are formulated. The plant is considered an uncertain dynamical system with
matched and mismatched perturbations and exogenous disturbance. The SMC designed in the paper is
integral-type based on a state observer. The contributions of this work are summarized as follows.
(i) The discrete event at the sensor unit and the DoS attacks on the network between the sensor and
the controller are modeled in a framework of an interval time-delay system, without using the constraints
of the DoS frequency. Transmission delay in the network is also allowed. The discrete event is designed
as an aperiodic event-triggering mechanism to regulate the stability of the closed-loop system subject to
the DoS attacks and discrete events.
(ii) The networked system is fully considered with matched and mismatched system perturbations
and external disturbances. The state observer is employed to estimate the unmeasurable states, and an
integral-sliding-mode controller based on the observer state is designed to enforce the convergence of the
estimates. Consequently, the ideal sliding mode can be ensured.
(iii) The stability of the resulting closed-loop system, including the resulting state estimation system,
is analyzed by the Lyapunov-Krasovskii functional method. Then, sufficient conditions of the robustness
and stability with an L2 gain performance are provided, by which the parameters of the controller and
observer can be determined for the secure control of the networked uncertain dynamical system subject
to the DoS attacks.
The remainder of this paper is organized as follows. The descriptions of the networked system with the
DoS attacks and event-triggering mechanisms are shown in Section 2. Section 3 presents the main results
for this work. Some discussions on the extensions of the presented secure control method are provided in
Section 4. An application of the proposed method is given in Section 5. Finally, Section 6 presents the
conclusion.
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:3

Defense controller Smart sensor

Controller Plant Sampler

u(t) y(t)

x(t) u(t) y(kT)

Estimator Detector
DoS attacks
y(t)
y(ti)
ZOH Sender
y(tj)
Network Event-
triggering
mechanism

Figure 1 (Color online) An illustration of the proposed structure of the secure estimator-based SMC under DoS attacks and
discrete events.

Notations. L2 ∈ [0, +∞) denotes the space of square-integrable vectors. The superscripts “T”
T
and “−1” denote the matrix transpose and inverse, respectively. [X]sym
 is used to denote X + X for
simplicity. A block diagonal matrix is denoted by the shorthand diag X1 , X2 , . . . , Xn with diagonal
matrices X1 , X2 , . . . , Xn .

2 System description and problem formulation

2.1 System description

Consider the following uncertain dynamical system:

(
ẋ(t) = Ax(t) + B(u(t) + v(x, t)) + f (x, t) + Ew(t),
(1)
y(t) = Cx(t) + F w(t),

where x(t) ∈ Rdx , u(t) ∈ Rdu , and y(t) ∈ Rdy are the vectors of system states, control inputs, and
system measurement outputs, respectively. v(x, t) ∈ Rdu and f (x, t) ∈ Rdx are respectively the matched
and mismatched system perturbations. w(t) ∈ Rdw denotes exogenous disturbance inputs belonging to
L2 [0, +∞). The matrices A ∈ Rdx ×dx , B ∈ Rdx ×du , C ∈ Rdy ×dx , E ∈ Rdx ×dw , and F ∈ Rdy ×dw are
known system parameters.
For the plant (1), when observer-based SMC strategy is applied integrating with the smart sensor, the
structural diagram of the overall control system is depicted in Figure 1. In this paper, we consider the
DoS attacks on the network between the sensor and the controller units. The smart sensor consisted of
the sampler and the event trigger is used to send the measurement (data) aperiodically for the mitigation
of the network communication. The network linking the sensor and the controller is considered with some
DoS attacks interrupting the data transmission. In the defense controller unit, an estimator is used to
estimate the system state and then to design a sliding mode controller, by taking the DoS attacks into
account. The specific formulations of the smart sensor and the defense controller will be presented in
Subsections 2.2–2.4.

2.2 Event-triggering mechanism

As for the measurement output y(t), we employ a sampler to sample y(t) with a sampling period T of
which the sequence T , {T1 , T2 , . . . , T∞ } is allowed to be float, that is the sampler can aperiodically
sample y(t). Meanwhile, the k-th sampled measurement y(kTk ) (T ∈ T, k = 1, 2, 3, . . .) is set to be sent
to the event detector of the event-triggering mechanism for calculating and detecting if the measurement
y(kT ) violates the event-triggering condition formulated in (2). For simplicity, we write Tk as T for
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:4

each sampling instant in the following context. We ignore the time to sample the measurement and
calculate/detect the event.

δ(ki , l) , eT T
y ey − ρy (ki T )y(ki T ) > 0, (2)

where ey , y((ki + l)T ) − y(ki T ). 0 < ρ < 1 denotes a custom coefficient of triggering error tolerance.
y(ki T ) (ki ∈ N, i = 1, 2, . . . , ∞) denotes the latest triggered measurement at the i-th triggering instant
ki T . y((ki + l)T ) (l = 1, 2, 3, . . .) denotes the current sampled/detected the measurement.
Hence, the next triggering time instant ki+1 T of the event trigger can be determined by

ki+1 T = ki T + arg min δ(ki , l) | δ(ki , l) > 0 . (3)
l>1

Evidently, it holds that

δ(ki , l) 6 0, ∀(ki + l)T ∈ [ki T, ki+1 T ). (4)

As a result, the sampled measurements will be released aperiodically to the next unit (the controller unit)
over the network with certain network resources saved.

2.3 Network and DoS attacks

In this work, the network communication between the sensor and the controller units is considered in
the whole control loop. When the triggered measurement y(ki T ) is transmitted over the network to
the controller unit, it may be confronted with data packet dropouts, transmission delay, and malicious
attacks, particularly for large-scale wireless networks. This paper is focused on the technical development
of the design of SMC against the DoS attacks. A DoS attack off and on interrupts the communication
service for a segment of time. This attack leads to that triggered measurement cannot be transmitted
and be received in time. It thus may detrimentally affect the system stability [22].
Without loss of generality, for the m-th DoS attack, i.e., the network is attacked initially at time instant
am , let us define the attack interval with Γm , {am } ∪ [am + µm ), where µm denotes the length of the
m-th DoS interval, and m ∈ M where M , {1, 2, 3, . . . , mmax } with mmax the maximum number of the
attack segments. Let A denote the set of times at which a DoS attack is active. For simplicity, define
ti , ki T . Then, the measurement y ′ (ti ) received at the zero-order holder (ZOH) can be formulated as
(
′ y(ti ), ti ∈
/ A,
y (ti ) =
Null, ti ∈ A,

where “Null” denotes no data received at the ZOH.

2.4 Estimator and controller

The estimator is employed to estimate the state of the plant, by using the measurement transformed
over the network. Considering the discrete event and the possible interrupt transmission caused by the
DoS attack, we employ a ZOH to receive the discrete signal. It keeps outputting a continuous signal
and updating the output when a new discrete signal y ′ (ti ) is received. Let us define the output of the
ZOH as ȳ(t). Notably, ȳ(t) should keep outputting the current signal if the newly received measurement
y ′ (ti ) = Null. Then, the following Luenberger state observer used as the estimator is constructed
(
˙
x̂(t) = Ax̂(t) + Bu(t) + L(ȳ(t) − ŷ(t)),
(5)
ŷ(t) = C x̂(t),

where x̂(t) ∈ Rdx and ŷ(t) ∈ Rdy are respectively the state estimate vector and output estimate vector.
The matrix L ∈ Rdx ×dy is the observation gain to be designed. The matrices {A, B, C} are the parameters
of the plant (1).
Our purpose in this paper is to design a sliding mode controller to stabilize the system (1). We may
express the sliding mode controller

u(t) = uo (t) + us (t) (6)

 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:5

ti ti+1 ti+2 ti+3 ti+13

Case A
tjh tj+1
h tj+9
h t

ti ti+1 ti+2 ti+3 ti+4

Case B
t
tjh am+3 tj+1
h
am am+1 am+2
µm µm+1 µm+2 µm+3
(a)

ti ti+1 ti+2 ti+3 ti+13

Case A′
dj tjh dj+2 tj+9
h t
tj+1
h
dj+4 dj+7

ti ti+1 ti+2 ti+3 ti+4

Case B′
tjh tj+1
h t
am am+1 am+2 am+3
µm µm+1 µm+2 µm+3
(b)

DoS attack segments [ti, ti+1) [tjh , tj+1

h)

Figure 2 (Color online) Illustrations of the triggered instants ti and the received instants th
j of the measurement y(ti ) under the
DoS attack. (a) Without the network transmission delay dj ; (b) with the network transmission delay dj .

with two terms, that is the nominal term uo (t) and the switching term us (t). The switching term us (t)
acts on a guarantee of the finite-time reachability of sliding modes and the compensation to the matched
perturbations. uo (t) and us (t) are to be designed. Meanwhile, based on the state estimate x̂(t) the sliding
variable s(t) ∈ Rdu is an integral-type one formulated as follows:
(
s(t) = K x̂(t) + z(t), z(0) = −K x̂(0),

(7)
ż(t) = −K Ax̂(t) + Buo (t) ,

where the matrix K ∈ Rdu ×dx is a given parameter of the (ideal) sliding surface s(t) = 0. z(t) ∈ Rdu is
an intermediate variable denoting the integral term of the sliding function.

2.5 Formulation of time-delay model

In this paper, by using the idea of the “time delay”, we formulate the measurement y(·) from the plant
to the ZOH over the sampling, event-detecting, and interrupting (attacking).
For the discrete (triggered) measurement y(ti ), it may suffer from the DoS attack segment before
the ZOH receiving the triggered measurement y(ti ). Therefore, the ZOH may not receive a triggered
measurement every triggering time instant. Define thj the (global) time instant of the j-th time ZOH
receiving (updating) the signal from the network, and j = 1, 2, 3, . . .. Figure 2 illustrates the instants
when the measurement y(ti ) was triggered at the event trigger and the instants when y(ti ) was received
at the ZOH.
One thing is certain that the time instant thj will not locate in a DoS attack segment if there are no
delays of the measurement transmission from the network to the ZOH, as illustrated in Figure 2. Hence,
without losing generality and strictness, we consider there are some momentary transmission delay

dj < min+ {Tk },

k∈N

when the ZOH receiving the measurement y(ti ) at time instant thj . The cases B and B′ what need to be
noticed are that if lots of triggered instants in succession occur all in DoS attack segments, the ZOH will
not receive new measurements for a long time. It leads to the feedback control losing its function and
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:6

significance and thus the stability of the control system will no longer be guaranteed. To this end, we
will require such an event-triggering condition fulfilling that the maximum

∆tmax = max{∆ti , ti+1 − ti }

i∈N

of the inter-event time ∆ti (i.e., the interval length of two adjacent event-triggering instants) will be less
than the minimum secure durations

νmin = min {νm , am+1 − (am + µm )}.

m∈M

In this requirement, the triggered instants ti will not miss the secure durations and thus the cases B and
B′ are avoided.
The following assumptions on the plant and the network DoS attack phenomenon are taken into account
in this paper, for theoretical analysis.
Assumption 1. Consider the system in (1). The matrix B is of full column rank, i.e., rank(B) = m,
and the pairs (A, B) and (A, C) are controllable and observable, respectively.
Assumption 2. Consider the system in (1). The matched uncertainty term v(x, t) is bounded by
v(x, t) 6 α with α > 0. The mismatched uncertainty term f (x, t) can be formulated with f (x, t) =
M L(t)N x(t), where M ∈ Rdx ×dp and N ∈ Rdq ×dx are constant matrices, and L(t) ∈ Rdp ×dq is an
unknown time-varying matrix satisfying Lebesgue measurable condition LT (t)L(t) 6 Iq , ∀t > 0.
Assumption 3. Consider the DoS attack in the network between the sensor and the controller units.
For any m ∈ M, there exist
(1) a maximum duration µmax of the DoS attack segments µm ,
(2) and a minimum duration νmin of the secure segments νm , am+1 − (am + µm ).

Assumption 4. The maximum sampling period Tmax = mink∈N+ Tk is fairly less than the minimum
duration νmin of the secure segments between any two adjacent DoS attack time instants am and am+1 ,
m ∈ M.
Remark 1. In fact, due to the limited energy of DoS attacks and its own anonymity, the DoS attack
duration cannot be very large. Assumptions 3 and 4 are reasonable and practical for a networked control
system. It is also consistent with the assumptions and design in [22, 23]. Besides, we will develop the
event-triggering condition (2) against the DoS attack segment, for avoiding too many triggering instants
in succession occurring in the attack segments to update new measurements timely. Therefore, the event-
triggering condition (2) should be mended and the next triggering time instant ki+1 T of the event trigger
is determined by

ki+1 T = ki T + arg min δ(ki , l) | δ(ki , l) > 0 ∪ ∆ti > νmin · T. (8)
l>1

Taking the requirement ∆tmax < νmin into account, we know that when a triggered measurement y(ti )
newly arrives at the ZOH, the ZOH will “wait” a span of time including the corresponding transmission
delay time and possible DoS attack duration. In addition, for the real-time system output y(t), it will
take the time of the sampling and repeatedly event-detections when a new triggered measurement y(ti ) is
generated. Then, for any y(ti ) (i ∈ N+. ) arrives at the ZOH at the instant thj , it is not difficult to obtain
the following conclusion.
• For the case of the network transmission without delay dj ,

thj+1 − thj = ti+l∗ − ti < 2∆tmax + µmax , (9)

l∗ ∈N+

where ti+l∗ denotes the time instant of a new triggered measurement y(ti+l∗ ) arriving at the ZOH at thj+1
without transmission delay.
• For the case of the network transmission with delay dj ,

thj+1 − dj+1 − (thj − dj ) = ti+l∗ − ti < 2∆tmax + µmax , (10)

l∗ ∈N+

where ti+l∗ denotes the time instant of y(ti+l∗ ) arriving at the ZOH at thj+1 with transmission delay dj+1 .
Apparently, the case of the network transmission without delay is a special case of the one with delay
dj = 0 for j ∈ N+ .
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:7

Consequently, for any t ∈ [thj , thj+1 ) (j = 1, 2, 3, . . .), we introduce the following auxiliary variable:

h(t) := t − thj = t − (ti + dj ), t ∈ [thj , thj+1 ), (11)

which satisfies that

0 6 h(t) < hM (12)

with hM = 2∆tmax + µmax + dmax − dmin . Then, the triggered measurement y(ti ) can be rewritten as

y(ti ) = y(t − h̃(t)), t ∈ [thj , thj+1 ), (13)

where h̃(t) = h(t) + dj ∈ [dj , hM + dj ) ⊆ [dmin , hM + dmax ). Then, it holds that

ȳ(t) = y(ti ) = y(t − h̃(t)), ∀t ∈ [thj , thj+1 ). (14)

Considering the synchronicity of the plant and the controller, define th0 = 0 and ȳ(th0 ) = 0, t ∈ [th0 , th1 ).
Then, the observer (5) can be further expressed as the following time-delay model:
(
˙
x̂(t) = Ax̂(t) + Bu(t) + L(y(t − h̃(t)) − ŷ(t)),
(15)
ŷ(t) = C x̂(t), t ∈ [thj , thj+1 ).

In Section 3, we are ready to propose a sliding mode controller u, based on the observer model (15), for
the uncertain system (1). Because part of the state is not measurable, in this paper, our design principle
is to employ the observer to obtain the state estimates, and then use the state estimates to construct a
sliding mode controller, for achieving stability of the system with the network DoS attacks. Definition 1
is for stability analysis of the resulting SMC system.
Definition 1. The nominal system (1) with v(t) ≡ 0, f (t) ≡ 0, and w(t) ≡ 0 is said to be stable if there
exists β > 0 fulfilling kx(t)k2 < β for t > 0. Furthermore, if limt→∞ kx(t)k2 = 0 for all initial x(0), then
the nominal system (1) is said to be asymptotically stable. Additionally, if the nominal system in (1) is
asymptotically stable, the uncertain system (1) with internal control u(t), external disturbance w(t), and
all admissible uncertainties {v(t), f (t)} described in Assumption 2, is said to be robustly stable.

3 Theoretical results
3.1 Controller design

By using the state estimates x̂(t), the following sliding mode controller is constructed as u(t) = uo (t) +
us (t), with

uo (t) = Gx̂(t) − η(t), (16)

us (t) = −(KB)−1 kKL(ȳ(t) − ŷ(t))ksign(s(t)) − κ(KB)−1 |s|λ sign(s(t)), (17)

where η(t) , (KB)−1 KL(ȳ(t)− ŷ(t)), κ > 0, and 0 < λ < 1. Matrix G ∈ Rdu ×dx is a controller parameter
to be designed.
As a result, based on the observer (15), the following sliding motion dynamics can be yielded under
the ideal sliding mode s(t) ≡ ṡ(t) ≡ 0 in the light of ṡ(t) = 0 and s(0) = 0:


˙
x̂(t) = (A + BK)x̂(t) + HL y(t − h̃(t)) − ŷ(t) (18)

with H , I − B(KB)−1 K. Considering the design objective of the convergence of the system state
estimation error x̃(t) , x(t) − x̂(t) and the system state x(t), define ξ(t) , [x̃T (t) x̂T (t)]T . Then, the
following dynamics of the system augmented with the plant and the observer in the sliding mode can be
extended as follows. For t ∈ [thj , thj+1 ),

˙
ξ(t) = (A + L̃(t))ξ(t) + Hξ(t − h̃(t)) + E w̃(t), (19)
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:8

where w(t) = w(0) and ξ(t) = ξ(0) for t ∈ [−hM − dmax , 0), w̃(t) , [wT (t) v T (t) wT (t − h̃(t))]T , and
" # " # " #
A −A + HLC E B −HLF −HLC 0dx ×dx
A, , E, , H, ,
0dx ×dx A + BG 0dx ×dw 0dx ×du HLF HLC 0dx ×dx
 T  
M , M T 0dp ×dx , N , N 0dq ×dx , L̃(t) , ML(t)N .

Remark 2. Since the system (19) is augmented with the dynamics of the plant and the observer is
derived in the sliding mode, the finite-time reachability of the ideal sliding mode s(t) = 0 should be
guaranteed. Then, some conditions of the stability of the augmented system (19) can be explored for the
stability of the whole closed-loop system, including the convergence of the resulting sliding motion (18)
which is relative to the observer dynamics.

3.2 Reachability analysis of the sliding mode

Theorem 1. Consider the sliding variable (7) and the SMC law (6). The sliding mode s(t) = 0 can be
achieved in finite-time under the SMC law (6) with the controller terms in (16) and (17).
Proof. Consider the following Lyapunov function Vs (s(t)) = 0.5sT (t)s(t). Recalling the dynamics of x̂(t)
in (15) or (5), we have
V̇s (s(t)) = sT (t)ṡ(t)



= sT (t) K Ax̂(t) + Bu(t) + L(ȳ(t) − ŷ(t)) − K Ax̂(t) + Buo (t)
= −kKL(ȳ(t) − ŷ(t))kks(t)k1 + sT (t)KL(ȳ(t) − ŷ(t)) − κ|s|λ sign(s(t))
6 −κ|s|λ sign(s(t)),
which can be concluded that Vs (s(t)) = 0, that is s(t) = 0, can be realized within finite time.
From the results of proof above we known that, the term −(KB)−1 kKL(ȳ(t) − ŷ(t))ksign(s(t)) in
us (t) acts on the compensation for the observation error ȳ(t) − ŷ(t) which is influenced by some factors
including the DoS attack duration. Meanwhile, one can apply some other reaching functions instead
of −κ|s|λ sign(s(t)) in us (t), for a good reachability performance including the fast reaching and the
chattering weakening.

3.3 Stability analysis of the sliding motion

The augmented system (19) is a kind of time-varying delay system with interval time delay dmin 6 h̃(t) <
hM + dmax , as discussed in (13). In this paper, we apply the Lyapunov-Krasovskii functional method
to explore the stability of this time-delay system. Let hd = hM + dmax in the following context, for
simplicity.
Theorem 2. Consider the uncertain time-varying delay system (19). For some scalars hd > dmin > 0,
the uncertain system (19) is robustly stable with an L2 gain less than or equal to γ if there exist a
scalar λ > 0, positive definite matrices Qi ∈ Rdx ×dx , Ri ∈ Rdx ×dx , matrices Sjk ∈ R2dx ×2dx , and
Sj5 ∈ R(2dw +dv )×2dx (i = 1, 2, . . . , 6 and j, k = 1, 2, . . . , 4) such that
Φ̄ + [Ψ]sym + ῩT
1 (dmin Q4 + h̄Q5 + hd Q6 )Ῡ1 + S < 0, (20)

where S , +S1 Q−1 T −1 T −1 T −1 T

4 S1 + S2 Q5 S2 + S3 Q5 S3 + S4 Q6 S4 and
 
Φ̄0 0 Q1 H 0 Q1 E
 
 ∗ −Q2 0 0 0  " #
  C̄ C̄  
 
Φ̄ ,  ∗ ∗ 0 0 0 , C , , Φ̄0 , Q1 (A + L̃(t)) sym + C + Q2 + Q3 ,

 ∗ ∗
 C̄ C̄
 ∗ −Q3 0  
∗ ∗ ∗ ∗ −γ 2 I
   
Ψ , (S1 + S4 ) −(S1 − S2 ) −(S2 − S3 ) −(S3 + S4 ) 0dw ×dw , Si , Si1 T T T T T T,
Si2 Si3 Si4 Si5
  
Ῡ1 , A + L̃(t) 0 H 0 E , C̄ , C T C, Qi , diag Qi , Ri ,
 
Λ1 , diag − Q4 , −Q5 , −Q6 , Λ2 , diag − Q4 , −Q5 , −Q5 , −Q6 . (21)
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:9

Proof. Choose the Lyapunov functional candidate:

6
X
V (t) = Vi (t), (22)
i=1

where
Z t
V1 (t) = ξ T (t)Q1 ξ(t), V2 (t) = ξ T (s)Q2 ξ(s)ds,
t−dmin
Z t Z 0 Z t
V3 (t) = ξ T (s)Q3 ξ(s)ds, V4 (t) = dmin ξ˙T (s)Q4 ξ(s)dsdv,
˙
t−hd −dmin t+v
Z −dmin Z t Z 0 Z t
V5 (t) = h̄ ξ˙T (s)Q5 ξ(s)dsdv,
˙ V6 (t) = hd ˙
ξ̇ T (s)Q6 ξ(s)dsdv,
−hd t+v −hd t+v

˙
where h̄ , hd − dmin . Consider that h̃(t) = 1, ∀t ∈ [thj , thj+1 ). Then, by calculating the derivatives of Vi (t),
we obtain that
˙
V̇1 = 2ξ T (t)Q1 ξ(t), V̇2 = ξ T (t)Q2 ξ(t) − ξ T (t − dmin )Q2 ξ(t − dmin ),
V̇3 = ξ T (t)Q3 ξ(t) − ξ T (t − hd )Q3 ξ(t − hd ),
Z t
V̇4 = d2min ξ˙T (t)Q4 ξ(t)
˙ − dmin ξ˙T (v)Q4 ξ(v)dv,
˙
t−dmin
Z t−dmin
V̇5 = h̄2 ξ˙T (t)Q5 ξ(t)
˙ − h̄ ˙T ˙
ξ (v)Q5 ξ(v)dv,
t−hd
Z t
V̇6 = h2d ξ˙T (t)Q6 ξ(t)
˙ − hd ξ˙T (v)Q6 ξ(v)dv.
˙
t−hd

T T T
Consider the slack matrices: Si , [Si1 Si2 Si3 S T S T ]T (i = 1, 2, 3, 4). By combining them with the
R t i4 i5
˙
Leibniz-Newton formula ξ(t − dmin ) − ξ(t) = t−dmin ξ(s)ds, we can get the following identities:
 Z t 
0 = 2ζ T (t)S1 ξ(t) − ξ(t − dmin ) − ˙
ξ(s)ds ,
t−dmin
 
t−dmin ˙
0 = 2ζ T (t)S2 ξ(t − dmin ) − ξ(t − h̃(t)) − ξt− h̃(t)
ξ(v)dv ,
Z t−h̃(t) !
0 = 2ζ T (t)S3 ξ(t − h̃(t)) − ξ(t − hd ) − ˙
ξ(v)dv ,
t−hd
 Z t 
T
0 = 2ζ (t)S4 ξ(t) − ξ(t − hd ) − ˙ξ(v)dv , (23)
t−hd

where ζ(t) = [ξ T (t) ξ T (t − dmin ) ξ T (t − h̃(t)) ξ T (t − hd ) w̃T (t)]T .

Before further, let us introduce the following lemma [24]: For any vectors a, b and appropriate dimension
matrix X > 0, it holds that aT b + bT a 6 aT X −1 a + bT Xb. Furthermore, consider the following lemma
from [25] that: For a matrix 0 < X ∈ Rn×n , and a continuous vector function b(t) in [t1 , t2 ] → Rn , it
holds that
Z t2  Z t2  Z t2
bT (s)ds X b(s)ds 6 (t2 − t1 ) bT (s)Xb(s)ds.
t1 t1 t1

Then, by considering that

Z t Z t
T
ζ (t)S1 ˙
ξ(s)ds = ˙ ds,
ζ T (t)S1 ξ(s)
t−dmin t−dmin | {z } |{z}
aT b

one can derive that

Z t Z t
T
−2ζ (t)S1 ˙
ξ(s)ds 6 ζ T (t)S1 Q−1 T
ξ˙T (v)Q4 ξ(v)dv,
˙
4 S1 ζ(t) + dmin (24)
t−dmin t−dmin
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:10

Z t−dmin Z t−dmin
T
−2ζ (t)S2 ˙
ξ(s)ds 6 ζ T (t)S2 Q−1 T
ξ˙T (v)Q5 ξ(v)dv,
˙
5 S2 ζ(t) + h1 (25)
t−h̃(t) t−h̃(t)
Z t−h̃(t) Z t−dmin
T
−2ζ (t)S3 ˙
ξ(s)ds 6 ζ T (t)S3 Q−1 T ˙
ξ̇ T (v)Q5 ξ(v)dv,
5 S3 ζ(t) + h2 (26)
t−hd t−h̃(t)
Z t Z t
−2ζ T (t)S4 ˙
ξ(v)ds 6 ζ T (t)S4 Q−1 T
ξ˙T (v)Q6 ξ̇(v)dv,
6 S4 ζ(t) + hd (27)
t−hd t−hd

where h1 , h̃(t) − dmin , h2 , hd − h̃(t) and h1 + h2 = h̄.

Firstly, for the nominal time-delay system (19) with w̃(t) ≡ 0, recalling h̃(t) − dmin < hd − dmin = h̄
and hd − h̃(t) 6 hd − dmin = h̄, from the calculated results of V̇ (t) and (23)–(27), we can directly obtain
that V̇ (t) < 0 for any ξ(t) 6= 0, according to the condition (20). This means that the asymptotic stability
of the system (19) can be guaranteed.
Secondly, for the time-delay system (19), let us consider a finite-gain dissipativity [26, 27] to evaluate
the system performance. Suppose γ is a given positive real number. The time-delay system (19) is said
to have an L2 gain less than or equal to γ if
Z th
N
Z th
N
T
y (t)y(t)dt 6 γ 2 wT (t)w(t)dt
0 0

for all t ∈ [thj , thj+1 ) and all w̃(t) ∈ L2 [0, +∞) with the output y(t) resulting by w̃(t) from initial state
ξ(thj ). Then, considering the uncertain time-delay system (19) with the disturbance w̃(t) ∈ L2 [0, ∞), we
further derive that

V̇ (t) + y T (t)y(t) − γ 2 w̃T (t)w̃(t)

 
< ξ T (t) Φ̄ + ῩT 2 2 2
1 (dmin Q4 + h̄ Q5 + hd Q6 )Ῡ1 + [Ψ]sym + S ξ(t). (28)

According to the condition (20), it holds that for ξ(t) 6= 0,

V̇ (t) + y T (t)y(t) − γ 2 w̃T (t)w̃(t) < 0,

from which by integrating the left-hand side terms from thj to thj+1 , it yields
Z th
j+1  T 
y (t)y(t)dt − γ 2 wT (t)w(t) dt < V (thj ) − V (thj+1 ).
th
j

SN h h
Furthermore, for j = 0, 1, 2, 3, . . ., that is t ∈ j=0 [tj , tj+1 ) ⊂ [0, ∞), we can further obtain

Z th
N   N Z
X th
j+1  T 
T 2 T
y (t)y(t)dt − γ w (t)w(t) dt = y (t)y(t)dt − γ 2 wT (t)w(t) dt
th
0 j=0 th
j

N
X
< V (thj ) − V (thj+1 ) = V (th0 ) − V (thN ).
j=0

Under the zero initial condition V (th0 ) = V (0) = 0, and V (thN ) > 0, it thus can be concluded that
Z th
N
Z th
N
y T (t)y(t)dt 6 γ 2 w̃T (t)w̃(t)dt.
0 0

Therefore, according to Definition 1, the uncertain time-delay system (19) is robustly stable with an L2
gain less than or equal to γ, with the system parameters fulfilling the condition (20). This completes the
proof.
The condition in Theorem 2 embeds the time-varying matrix L̃(t), which includes the unknown system
perturbation f (x, (t)). We develop this condition to another condition without time-varying terms for
the verifiability of the stability condition for the resulting system (19).
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:11

Corollary 1. Consider the uncertain time-varying delay system (19). For some scalars hd > dmin > 0,
the uncertain system (19) is robustly stable with an L2 gain less than or equal to γ if there exist a
scalar λ > 0, positive definite matrices Qi ∈ Rdx ×dx , Ri ∈ Rdx ×dx , matrices Sjk ∈ R2dx ×2dx , and
Sj5 ∈ R(2dw +dv )×2dx (i = 1, 2, . . . , 6 and j, k = 1, 2, . . . , 4) such that
 
Φ + [Ψ]sym Π1 Π2 ΠT
3
 
 ∗ Λ1 0 ΠT 
4 
Ξ,
 < 0, (29)
 ∗ ∗ Λ2 0 
∗ ∗ ∗ −λI

where
 
Φ0 0 Q1 H 0 Q1 E
 
 ∗ −Q2 0 0 0 
 
0  , Φ0 , [Q1 A]sym + Q2 + Q3 + C + λN T N ,
 
Φ, ∗ ∗ 0 0
 
 ∗ ∗ ∗ −Q3 0 
 
∗ ∗ ∗ ∗ −γ 2 I
     
1 Q4 h̄Υ1 Q5 hd Υ1 Q6 , Π2 , S1 S2 S3 S4 , Υ1 , A 0 H 0 E ,
Π1 , dmin ΥT T T

   
Π3 , MT Q1 0 0 0 0 , Π4 , dmin MT Q4 h̄MT Q5 hd MT Q6 . (30)

Proof. To get the same conclusion, we may prove that the condition (20) can be derived from the
condition (29). By using the Schur complement to Ξ < 0 in (29), it can be obtained that Ξ̃ + λN̄ T N̄ +
λ−1 ΠT
5 Π5 < 0, where
 
  Φ̃ 0 0 Q 1 H 0 Q 1 E
 
Φ̃ + [Ψ]sym Π1 Π2  ∗ −Q2 0 0 0 
   
 
Ξ̃ , 
 ∗ Λ 1 0 
 , Φ̃ ,  ∗ ∗ 0 0 0 ,
 
∗ ∗ Λ2  ∗
 ∗ ∗ −Q3 0  
∗ ∗ ∗ ∗ −γ 2 I
   
N̄ , N 0 0 0 0 0 0 0 0 0 0 0 , Φ̃0 , Q1 A sym + Q2 + Q3 ,
   
Π5 , Π51 dmin MT Q4 h̄MT Q5 hd MT Q6 0 0 0 0 , Π51 , MT Q1 0 0 0 0 .

Furthermore, in the light of Proposition 2.1 in [28], considering Ξ̃ + λN̄ T N̄ + λ−1 ΠT5 Π5 < 0, one can
obtain Ξ̄ < 0, where
 
Φ̄ + [Ψ]sym Π̄1 Π2
   
Ξ̄ , 
 ∗ Λ1 0   , Π̄1 , dmin Ῡ1 Q4 h̄Ῡ1 Q5 hd Ῡ1 Q6 ,
T T T

∗ ∗ Λ2

and Φ̄ and Ῡ1 are defined in (21). Then, using the Schur complement to Ξ̄ < 0, we obtain

Φ̄ + [Ψ]sym + ῩT 2 2 2
1 (dmin Q4 + h̄ Q5 + hd Q6 )Ῡ1 + S < 0,

which meets the condition (20) in Theorem 2. The proof is completed.

Remark 3. Theorem 2 and Corollary 1 provide the stability criterion of closed-loop system in sliding
mode state. In fact, the stability criterion also gives the stability condition of the sliding motion, that is
dynamics of the observer state x̂. Moreover, since the initial s(0) = 0 and the derivative ṡ(t) = 0, ∀t > 0,
it guarantees that the system trajectories will always move towards the system origin from the initial
condition s(0) = 0.
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:12

Remark 4. These stability criteria which are delay-dependent ones depend on the bounds dmin and
hd of the time delay. Since hd = 2∆tmax + µmax + 2dmax − dmin , large intervals of the sampling, inter-
event time, DoS attack duration and transmission delay may affect the stability of the control system.
Moreover, the sampling period T also affects the event-triggering instants because the event detecting
period is designed in accordance with T . Besides, for the triggering error tolerance 0 < ρ < 1, a small ρ
implies a faster triggering, which may reduce the value of the delay. Considering the allowed maximums
of the DoS attack durations and transmission delays, we may apply the event-triggering condition (8) for
releasing the measurements.
Based on the results designed and analyzed above, we summarize the algorithm of the presented time-
delay-based SMC against DoS attacks and discrete events as Algorithm 1.

Algorithm 1 Time-delay-based estimation and SMC against DoS attacks and discrete events
Require: Complete parameters of a physical system (1).
Ensure: Secure control of the system (1) with DoS attacks on the network between the controller and the sensor.
1: Initialize k = 0, sampling sequence T , {T1 , T2 , . . . , T∞ }, triggering error tolerance 0 < ρ < 1 of event-triggering condition
(8), sliding surface matrix K (7), hM = 2∆tmax + µmax + dmax − dmin , hd = hM + dmax , initial condition x(0) = x̂(0);
2: Solve the conditions (29) in Corollary 1 for the parameter {G, L} of the controller and the observer;
3: Update the controller according to the SMC law: u(t) = uo (t) + us (t) as given in (16), (17), and (7);
4: Update the state estimator under the updated controller: x̂(t) ˙ = Ax̂(t) + Bu(t) + L(ȳ(t) − ŷ(t)) (5);
5: Update the system (1) under the updated controller;
6: return x(t) and x̂(t).

4 Discussions on the extension of the presented secure control method

In fact, the designed SMC law is based on the integral-type model with the sliding surface s(t) = s(0) = 0.
Therefore, the switching term can be removed from the SMC law and some modifications on the estimator
can be made to design a continuous control law to secure control against the DoS attacks. Under the
same assumptions, let us consider the plant (1). The following sliding mode observer is employed to
estimate the system state.
(  
˙
x̂(t) = Ax̂(t) + Bu(t) − (KB)−1 kKL(ȳ(t) − ŷ(t))k + λIdu sign(s(t)) + L(ȳ(t) − ŷ(t)),
(31)
ŷ(t) = C x̂(t),

where λ is a positive scalar. The controller can be thus modified as

u(t) = Gx̂(t). (32)

By using the same sliding variable s(t) ∈ Rdu in (7), the finite-time reachability of the sliding surface
s(t) = 0 can be achieved.
Theorem 3. Consider the observer-based sliding variable (7) and control law (32). The sliding surface
s(t) = 0 can be achieved in finite-time under the control law (32).
Proof. The result can be easily obtained by considering the following Lyapunov function Vs (s(t)) =
0.5sT (t)s(t), and thus the proof is omitted.
Hence, in the ideal sliding mode, the resulting augmented system (19) consisting of the control system
and the estimation system is updated as
˙
ξ(t) = (A + L̃(t))ξ(t) + Hξ(t − h̃(t)) + E w̃(t), (33)

where w(t) = w(0) and ξ(t) = ξ(0) for t ∈ [−hM − dmax , 0), w̃(t) , [wT (t) v T (t) wT (t − h̃(t))]T , M, N ,
and L̃ are defined in (19), and
" # " # " #
A −A + LC E B −LF −LC 0dx ×dx
A :, , E :, , H :, .
0dx ×dx A + BG 0dx ×dw 0dx ×du LF LC 0dx ×dx

By following Theorem 2 and Corollary 1, the following conditions are directly obtained to judge the
stability of the control system and the estimation system.
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:13

Corollary 2. Consider the uncertain time-varying delay system (33). For some scalars hd > dmin > 0,
the uncertain system (33) is robustly stable with an L2 gain less than or equal to γ if there exist a
scalar λ > 0, positive definite matrices Qi ∈ Rdx ×dx , Ri ∈ Rdx ×dx , matrices Sjk ∈ R2dx ×2dx , and
Sj5 ∈ R(2dw +dv )×2dx (i = 1, 2, . . . , 6 and j, k = 1, 2, . . . , 4) such that the matrix inequality in (29) holds.
Remark 5. Since the control law includes no switching term, this sliding-mode-observer-based control
can be extended to the plant with the output y(t) = Cx(t) + Du(t) + F w(t). Some nonlinearities in the
system dynamics can be also allowed. The switching term in the sliding mode observer can be modified to
compensate for the nonlinearities. In this DoS attack case, the switching term plays a role to compensate
for the error caused by the DoS attacks and the external disturbances.
Remark 6. The presented event-triggering mechanism (2) is a common static one used in NCSs.
Recently, some advanced event-triggering conditions for improving the control and communication per-
formance have been reported in the literature. Readers can refer to [29, 30] for details. Since the errors
induced by the advanced event-triggering conditions can be described by the time-delay model, the secure
control design method can be integrated with the advanced event-triggering conditions for some advanced
event-based secure control in NCSs.

5 Application to networked pendulum systems

Let us consider a networked system of an inverted pendulum with a cart [31] to show the effectiveness of
the presented secure control method. By taking the mass of the pendulum m = 0.535 kg, the length of the
pendulum l = 0.365 m, the mass of the cart M = 3.2 kg, the gravitational acceleration g = 9.807 m/s2 ,
the cart position coordinate x1 , the pendulum angle from vertical x3 , the state vector x = [x1 ẋ1 x3 ẋ3 ] =
[x1 x2 x3 x4 ], and the control input force u, a linearized system dynamics of the inverted pendulum system
is obtained below.
   
  0 1 0 0   0
ẋ1 
−(I + ml2 )b m2 gl2
 x1  I + ml 
   0 0
   
 
 ẋ2   I(M + m) + M ml2 I(M + m) + M ml2   x2   I(M + m) + M ml2 


ẋ = 
 ẋ  = 
 x +
  u,
 3 0  0 0 1 3  
  0 

ẋ4
 −mlb mgl(M + m) 
x
 ml 
0 0 4
I(M + m) + M ml 2
I(M + m) + M ml2 I(M + m) + M ml2
 
where y = 10 00 01 00 x is considered the measurable system output. In the simulations, consider the
external disturbance w = 0.1(rand − 0.5) with E = [0.01 0 0.01 0]T and F = [0.01 0]T , the system
perturbations v(t) = 0.1 sin(u(t)) and f (t) = M L(t)N x(t) with M = [0.1 0.1 0 0]T , N = [0.1 0.1 0 0],
and L(t) = sin(x1 (t) + x3 (t)).
For the simulated networked pendulum system, we use a fixed sampling period of the sampler as
T = 0.02 s. Assume that the transmission delay dj on the j-th network transmission is in [0.004, 0.016] s
with dmin = 0.004 s and dmax = 0.016 s, and the constraints of the DoS attack duration µmax = 0.8 s,
νmin = 0.1 s. Then, according to the conditions in Corollary 1, we use the following parameters of the
controller and observer in the simulations to show the response of the controlled system:
 
27.7832 −0.0098  
  G = 12.2506 13.1018 −80.1023 −15.7809 ,
 160.5792 −119.7819 
L=  4.4431
,
 35.0350 
  
K = 0.10 −0.50 0.20 −0.30 ,
−93.6083 268.5555

and the L2 gain performance γmin = 0.0023. By using the event-triggering condition coefficient ρ = 0.05
and the initial condition x(0) = x̂(0) = [π/3 0 π/4 0]T , a DoS attack sequence is shown in Figure 3(a),
and the responses of the networked pendulum system are in Figures 4 and 5. The inner-event intervals
are depicted in Figure 3(b) based on the event-triggering mechanism. Figures 4(a) and (b) show the real
measurements at the sampler y(kT ), the event sender y(ti ) and the ZOH ȳ(t) of the system output y(t),
at different time instants. The control input force is depicted in Figure 4(c), under which the states of
the pendulum and its estimation states are shown in Figures 5(a) and (b), respectively. Besides, the
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:14

(a)
DoS attack durations

0 5 10 15 20 25 30 35 40
t (s)

0.10
(b)
Inter-event intervals

0.08

0.06

0.04

0.02

0
0 5 10 15 20 25 30 35 40
t (s)

Figure 3 (Color online) Illustrations of (a) the DoS attack durations and (b) the event-triggering instants.

1.5
(a)
0.4
1.0
0.2
0.5
0
0.3 0.4 0.5 0.6 0.7 0.8
0
y1(t) y1(kT) y1(ti) y1(t)
−0.5
0 5 10 15 20 25 30 35 40
t (s)

1.0 20
(b) 0.2 (c) u(t)
0.1
0
0
0.5
−0.1 5
−20 10
0.5 1.0 1.5 −5
0 −10
−40 −15
y2(t) y2(kT) y2(ti) y2(t) 0 0.5 1.0 1.5 2.0 2.5

0.5 −60
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
t (s) t (s)

Figure 4 (Color online) (a) and (b) are the measurements at the samplers y1 (kT ) and y2 (kT ), the event senders y1 (ti ) and
y2 (ti ), and the ZOHs ȳ1 (t) and ȳ2 (t) of the system outputs y1 (t) and y2 (t), at different time instants, while (c) is the control input.

estimation errors of the states are shown in Figure 5(c). Obviously, the simulation results illustrate that
the proposed estimator-based SMC is effective to secure control against the DoS attacks.
Additionally, let us check different DoS attack durations on the network affect the stability of the
networked pendulum system, by using several sets of the constraints of the DoS attack durations, that is
• Set 1: No DoS attacks µmax = 0,
• Set 2: DoS attack duration constraints µmax = 1.0 s, νmin = 0.1 s,
• Set 3: DoS attack duration constraints µmax = 3.0 s, νmin = 0.1 s,
• Set 4: DoS attack duration constraints µmax = 3.0 s, νmin = 1.0 s.
Then, the corresponding results are obtained as shown in Figures 6–9, from which one can observe
that the DoS attacks (with larger values of the DoS attack duration µmax ) degrade the system stability
(more serious). Figures 6–9 show the trajectories of the plant states x(t), the observer states x̂(t) and
their errors x̃(t), under Sets 1–4, respectively. Besides, from Figures 8 and 9, it can be concluded that
a large νmin can mitigate the impact of the DoS attacks on the system stability since more triggered
measurements can be transformed from the sensor to the controller.
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:15

2
(a) x1(t) x2(t) x3(t) x4(t)
0
1
0

x(t)
−2 −1
−2
−4 −3
0.5 1.0 1.5 2.0 2.5
−6
0 5 10 15 20 25 30 35 40
t (s)

2 0.5
(b) (c)
x1(t) x2(t) x3(t) x4(t) x1(t) x2(t) x3(t) x4(t)
0 0
0.4
1 0.2

x(t)
x(t)

−2 0 −0.5 0
−0.2
−1 −0.4
−2 −0.6
−4 −1.0 −0.8
0 0.5 1.0 1.5 2.0 2.5 0 0.5 1.0 1.5 2.0 2.5
−6 −1.5
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
t (s) t (s)

Figure 5 (Color online) (a) Trajectories of the plant states xi (t), (b) the observer states x̂i (t), and (c) their errors x̃i (t) (i =
1, 2, 3, 4).

2
0
x(t)

−2
−4 x1(t) x2(t) x3(t) x4(t)
(a)
−6
0 1 2 3 4 5 6 7 8 9 10
t (s)
2 0.5
0 0
x(t)

x(t)

−2 −0.5
−4 −1.0
(b) x1(t) x2(t) x3(t) x4(t) (c) x1(t) x2(t) x3(t) x4(t)
−6 −1.5
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
t (s) t (s)

Figure 6 (Color online) (a) Trajectories of the plant states x(t), (b) the observer states x̂(t), and (c) their errors x̃(t), under
Set 1.

(a)
0

−2
x(t)

−4 x1(t) x2(t) x3(t) x4(t)

DoS duration 1
0 1 2 3 4 5 6 7 8 9 10
t (s)

(b) (c)
0 0
x(t)

−2
x(t)

−0.5
DoS duration 1 x1(t) x2(t) x3(t) x4(t) DoS duration 1 x1(t) x2(t) x3(t) x4(t)
−4 −1.0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
t (s) t (s)

Figure 7 (Color online) (a) Trajectories of the plant states x(t), (b) the observer states x̂(t), and (c) their errors x̃(t), under
Set 2.

6 Conclusion
This paper solved the problem of SMC against the DoS attacks and discrete events, for a class of uncer-
tain dynamical systems in the network environment. The time-delay modeling approach was proposed to
describe the DoS attack duration and inter-event time as the “time delay”. The event-triggering mech-
anism was integrated into a smart sensor aperiodically releasing the measurement to the network with
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:16

(a)
0

−2

x(t)
−4 DoS duration 2 x1(t) x2(t) x3(t) x4(t)
0 1 2 3 4 5 6 7 8 9 10
t (s)

(b) (c)
0 0
x(t)

−2

x(t)
−0.5
DoS duration 2 x1(t) x2(t) x3(t) x4(t) DoS duration 2 x1(t) x2(t) x3(t) x4(t)
−4 −1.0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
t (s) t (s)

Figure 8 (Color online) (a) Trajectories of the plant states x(t), (b) the observer states x̂(t), and (c) their errors x̃(t), under
Set 3.

(a)
0
x(t)

−2

−4 x1(t) x2(t) x3(t) x4(t)

DoS duration 3
0 1 2 3 4 5 6 7 8 9 10
t (s)
(b) 0
0 (c)
x(t)

−2 −0.5
x(t)

DoS duration 3 x1(t) x2(t) x3(t) x4(t) DoS duration 3 x1(t) x2(t) x3(t) x4(t)
−4 −1.0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
t (s) t (s)

Figure 9 (Color online) (a) Trajectories of the plant states x(t), (b) the observer states x̂(t), and (c) their errors x̃(t), under
Set 4.

certain network resources saved. Then, an interval-time-delay system with uncertainties was represented.
A state observer was employed with the ZOH, and then an estimator-based controller was designed by
which the ideal sliding mode can be achieved. Moreover, the resulting sliding motion was proved to
be robust and stable with an L2 gain performance. The present SMC was finally validated by some
simulations. As a matter of fact, one can develop the present time-delay-based modeling and control to
nonlinear and stochastic systems in terms of the DoS attacks.
Acknowledgements This work was supported in part by National Key R&D Program of China (Grant No. 2019YFB1312001),
National Natural Science Foundation of China (Grant Nos. 62033005, 62103118, 62022030), China Postdoctoral Science Foundation
(Grant Nos. 2021T140160, 2021M700037), and Natural Science Foundation of Heilongjiang Province (Grant No. ZD2021F001).

References
1 Zhang W, Branicky M S, Phillips S M. Stability of networked control systems. IEEE Control Syst, 2001, 21: 84–99
2 Bemporad A, Heemels M, Johansson M, et al. Networked Control Systems. Berlin: Springer, 2010
3 Farivar F, Haghighi M S, Jolfaei A, et al. On the security of networked control systems in smart vehicle and its adaptive
cruise control. IEEE Trans Intell Transp Syst, 2021, 22: 3824–3831
4 Zhang X M, Han Q L, Ge X, et al. Networked control systems: a survey of trends and techniques. IEEE CAA J Autom Sin,
2019, 7: 1–17
5 Yang L, Gao Y B, Feng Z G, et al. Event-based output feedback control of periodic systems: a piecewise impulsive method.
Int J Robust Nonlinear Control, 2021, 31: 7569–7586
6 Li L Y, Cong P J, Cao K, et al. Feedback control of real-time EtherCAT networks for reliability enhancement in CPS.
In: Proceedings of Design, Automation & Test in Europe Conference & Exhibition (DATE), 2018. 688–693
7 Liu J L, Wu Z G, Yue D, et al. Stabilization of networked control systems with hybrid-driven mechanism and probabilistic
cyber attacks. IEEE Trans Syst Man Cybern Syst, 2021, 51: 943–953
8 Hu S L, Yue D, Xie X P, et al. Resilient event-triggered controller synthesis of networked control systems under periodic DoS
jamming attacks. IEEE Trans Cybern, 2019, 49: 4271–4281
9 Dong S L, Chen G R, Liu M Q, et al. Robust adaptive H∞ control for networked uncertain semi-Markov jump nonlinear
systems with input quantization. Sci China Inf Sci, 2022, 65: 189201
10 Edwards C, Colet E F, Fridman L. Advances in Variable Structure and Sliding Mode Control. Berlin: Springer, 2006
11 Utkin V, Poznyak A, Orlov Y, et al. Conventional and high order sliding mode control. J Franklin Inst, 2020, 357: 10244–10261
 Tian Y X, et al. Sci China Inf Sci June 2022 Vol. 65 162203:17

12 Huang X, Zhai D, Dong J X. Adaptive integral sliding-mode control strategy of data-driven cyber-physical systems against a
class of actuator attacks. IET Control Theory Appl, 2018, 12: 1440–1447
13 Gao Y B, Liu J X, Sun G H, et al. Fault deviation estimation and integral sliding mode control design for Lipschitz nonlinear
systems. Syst Control Lett, 2019, 123: 8–15
14 Sun G H, Wu L G, Kuang Z A, et al. Practical tracking control of linear motor via fractional-order sliding mode. Automatica,
2018, 94: 221–235
15 Liu J X, Gao Y B, Su X J, et al. Disturbance-observer-based control for air management of PEM fuel cell systems via sliding
mode technique. IEEE Trans Control Syst Technol, 2019, 27: 1129–1138
16 Utkin V I. Sliding Modes in Control and Optimization. Berlin: Springer, 2013
17 Liu M, Sun G H. Observer-based sliding mode control for Itô stochastic time-delay systems with limited capacity channel. J
Franklin Inst, 2012, 349: 1602–1616
18 Zhang J H, Lam J, Xia Y Q. Output feedback sliding mode control under networked environment. Int J Syst Sci, 2013, 44:
750–759
19 Wu L G, Gao Y B, Liu J X, et al. Event-triggered sliding mode control of stochastic systems via output feedback. Automatica,
2017, 82: 79–92
20 Ma R J, Shi P, Wu L G. Dissipativity-based sliding-mode control of cyber-physical systems under denial-of-service attacks.
IEEE Trans Cybern, 2021, 51: 2306–2318
21 Guo L N, Zheng B C, Liu X G. Event-triggered sliding-mode control of linear uncertain system under periodic DoS attacks.
In: Proceedings of Chinese Control And Decision Conference (CCDC), 2020. 1080–1085
22 Dolk V S, Tesi P, de Persis C, et al. Event-triggered control systems under denial-of-service attacks. IEEE Trans Control
Netw Syst, 2017, 4: 93–105
23 de Persis C, Tesi P. Input-to-state stabilizing control under denial-of-service. IEEE Trans Autom Control, 2015, 60: 2930–2944
24 Xie L, de Souza C E. Robust H∞ control for linear systems with norm-bounded time-varying uncertainty. IEEE Trans Autom
Control, 1992, 37: 1188–1191
25 Gu K Q, Chen J, Kharitonov V L. Stability of Time Delay Systems. Berlin: Springer, 2003
26 Hill D, Moylan P. Connections between finite-gain and asymptotic stability. IEEE Trans Autom Control, 1980, 25: 931–936
27 Lin W, Byrnes C I. H∞ -control of discrete-time nonlinear systems. IEEE Trans Autom Control, 1996, 41: 494–510
28 Li X, de Souza C E. Criteria for robust stability and stabilization of uncertain linear systems with state delay. Automatica,
1997, 33: 1657–1662
29 Ge X H, Han Q L, Ding L, et al. Dynamic event-triggered distributed coordination control and its applications: a survey of
trends and techniques. IEEE Trans Syst Man Cybern Syst, 2020, 50: 3112–3125
30 Ge X, Han Q L, Zhang X M, et al. Dynamic event-triggered control and estimation: a survey. Int J Autom Comput, 2021,
18: 857–886
31 Zhang J H, Feng G. Event-driven observer-based output feedback control for linear systems. Automatica, 2014, 50: 1852–1859