IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 7, NO.

1, JANUARY 2020 1

Networked Control Systems:

A Survey of Trends and Techniques
Xian-Ming Zhang, Senior Member, IEEE, Qing-Long Han, Fellow, IEEE, Xiaohua Ge, Member, IEEE, Derui Ding,
Member, IEEE, Lei Ding, Member, IEEE, Dong Yue, Senior Member, IEEE, and Chen Peng, Senior Member, IEEE

Abstract — Networked control systems are spatially distributed automation, robots, aircraft, automobiles, manufacturing plant
systems in which the communication between sensors, actuators, monitoring, remote diagnostics and troubleshooting, and tele-
and controllers occurs through a shared band-limited digital
communication network. Several advantages of the network ar- operations [1]–[6].
chitectures include reduced system wiring, plug and play devices, An NCS inherits both advantages and disadvantages from
increased system agility, and ease of system diagnosis and main- communication networks. On the one hand, communication
tenance. Consequently, networked control is the current trend for networks allow data packets to be shared among control
industrial automation and has ever-increasing applications in a components, which means that some traditional point-to-point
wide range of areas, such as smart grids, manufacturing systems,
process control, automobiles, automated highway systems, and wiring in the installation of a control system may be avoided.
unmanned aerial vehicles. The modelling, analysis, and control of As a result, control cost incurred from installation and
networked control systems have received considerable attention in maintenance can be reduced significantly. Besides, the well-
the last two decades. The ‘control over networks’ is one of the key used communication protocols can ensure data packets to be
research directions for networked control systems. This paper successfully transmitted between control components, which
aims at presenting a survey of trends and techniques in net-
worked control systems from the perspective of ‘control over net- renders an NCS of high reliability. On the other hand,
works’, providing a snapshot of five control issues: sampled-data communication networks transmit data packets only in the
control, quantization control, networked control, event-triggered form of digital signals rather than continuous signals. This
control, and security control. Some challenging issues are sugges- means that signals from physical plants should be sampled
ted to direct the future research. and quantized before being transmitted. Moreover, since the
Index Terms—Event-triggered control, networked control systems, network bandwidth is limited, network traffic congestion is
quantization control, sampled-data control, security control. unavoidable, usually leading to some constraints such as
network-induced delays and packet dropouts. Notice that
I. Introduction signal sampling, quantization errors, network-induced delays
HE rapid development of network technologies has
T brought much change to people’s life. Modern communic-
ation networks can provide swift and reliable communication
and packet dropouts may have the effects on control
performance of an NCS. Therefore, how to mitigate the
effects of those network imperfections on control performance
between any two or more physical plants located in different of NCSs has received considerable attention, see, e.g., survey
places. Such prominent characteristics make communication papers [4], [7]–[10] and references therein.
networks extensively used to connect control components Two of the key research directions for NCSs are ‘control of
within a control loop, leading to the so-called networked con- networks’ and ‘control over networks’ [11]. ‘Control of
trol systems (NCSs). NCSs have been applied in various networks’ aims at providing pretty good quality of service
areas, such as space exploration, environments, industrial (QoS) of communication networks such that NCSs can
Manuscript received May 31, 2019; accepted July 10, 2019. This work was achieve satisfactory control performance. In this scenario,
supported in part by the Australian Research Council Discovery Project network data flows perform very well and network resources
(DP160103567). Recommended by Associate Editor Mengchu Zhou. (Cor-
responding author: Qing-Long Han.) can be utilized efficiently and fairly. ‘Control over networks’
Citation: X.-M. Zhang, Q.-L. Han, X. H. Ge, D. Ding, L. Ding, D. Yue, and targets at devising suitable control strategies to reduce the
C. Peng, “Networked control systems: a survey of trends and techniques,” effects of network imperfections on desired control
IEEE/CAA J. Autom. Sinica, vol. 7, no. 1, pp. 1–17, Jan. 2020.
performance. Clearly, ‘control of networks’ focuses on how to
X.-M. Zhang, Q.-L. Han, X. Ge, and D. Ding are with the School of Soft-
ware and Electrical Engineering, Swinburne University of Technology, Mel- improve the QoS of networks while ‘control over networks’
bourne, VIC 3122, Australia (e-mail: [email protected]; on how to enhance control robustness against network
[email protected]; [email protected]; [email protected]). constraints. The former is within the field of computer science
L. Ding and D. Yue are with the Institute of Advanced Technology,
Nanjing University of Posts and Telecommunications, Nanjing 210023, China
while the latter falls into the scope of control systems. In this
(e-mail: [email protected]; [email protected]). paper, we place emphasis on the latter, i.e., ‘control over
C. Peng is with the Shanghai Key Laboratory of Power Station Automa- networks’.
tion Technology, School of Mechatronic Engineering and Automation, Recalling the existing literature, a rather large number of
Shanghai University, Shanghai 200072, China (e-mail: [email protected]).
Color versions of one or more of the figures in this paper are available on-
results have been reported on ‘control over networks’. Among
line at http://ieeexplore.ieee.org. those results, it is found that more attention is paid to the
Digital Object Identifier 10.1109/JAS.2019.1911651 following five issues:
2 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 7, NO. 1, JANUARY 2020

Sampled-data control, which is to investigate how sampling industrial control systems [19]. Thus, how to detect and cope
affects control performance of a control system. Since digital with attacks such that an NCS works in a safe environment is
techniques are widely implemented in real-world control an important issue, and has been a hot topic in the past decade
systems of continuous-time dynamics, it is more practical that [20]–[25].
the controller is designed based on sampled data. In a The five control issues above are fundamental issues for
sampled-data control system, one of the fundamental issues to NCSs. A great number of notable results have been reported
be addressed is how to select a proper sampling period in the literature. This paper presents a survey of trends and
because it directly affects stability of the closed-loop control techniques in NCSs, focusing on these five issues, and
systems. discusses some challenging issues for future research
Quantization control, which is to study the quantization directions.
effects on control systems. The process of converting signals Notation: P ≥ 0 (respectively, P > 0), where P is a
from analog to digital is an example of quantization, which symmetric matrix, means that P is positive semi-definite
maps input values from a larger set to output values in a (respectively, positive definite). col{· · · } represents a block-
smaller set. Quantization can be regarded as a lossy column vector. N and R are the sets of positive integers and
compression procedure. The resultant quantization errors real numbers, respectively.
between an input value and its quantized value definitely may
deteriorate system performance or destroy system stability. II. Sampled-data Control
Networked control, which admits network-induced delays In a sampled-data system, a continuous-time physical plant
and packet dropouts to exist. Its objective is to investigate the is controlled by a digital controller which generates a
effects of network-induced delays and packet dropouts on sequence of control inputs based on the sampled data packets
control systems. For example, how large is the network-induced [26]. In the case that the sampling is periodic, there are fruitful
delay (or the number of consecutive dropouts) endurable for a and mature results on analysis and synthesis of sampled-data
control system to retain certain performance? The core issue on systems, see, e.g., [27], [28]. However, those results may lose
such investigations is how to establish a relevant mathematical their effectiveness when the sampling is aperiodic. Therefore,
model for those network-induced constraints. By employing a during the past two decades, aperiodic sampled-data systems
time-delay system model, an interesting finding is that proper have been an active topic in the field of control, and a number
network-induced delays possibly play a positive role in of results have been reported in the literature, see, e.g., survey
improving system performance [13]. papers [29], [30] and references therein.
Event-triggered control, which has become a hot topic since Consider the following linear sampled-data system
2007. Under an event-triggered sampling scheme, whether or described by
{
not to sample system signals is determined by an event rather ẋ(t) = Ax(t) + Bu(t)
(1)
than the elapse of a fixed period [14]. As a result, computation u(t) = F x(tk ), t ∈ [tk , tk+1 )
resources, energy resources and network resources of an NCS where x(t) is the system state and u(t) is the control input; A
can be saved significantly. However, when external and B are known real matrices of compatible dimensions
disturbances are imposed on the physical plant, the positive while F is the control gain to be designed. An input delay
minimum inter-event time between any two consecutive method is well used to analyze stability of the system (1) and
sampling instants may not be ensured [15], leading to the so- design the control gain F . The basic idea is to introduce an ar-
called Zeno behaviour. In order to elude such Zeno behaviour, tificial piecewise-continuous delay τ(t) = t − tk . Then the con-
a sampled-data-based event-triggered transmission scheme is trol input u(t) can be rewritten as
proposed in [16], which allows system signals to be sampled
periodically while an event-generator is used to determine if u(t) = F x(t − τ(t)), t ∈ [tk , tk+1 ) (2)
the sampled signals should be released to the communication which leads to the following time-delay system model of the
network. closed-loop system
Security control, which addresses the security of an NCS in ẋ(t) = Ax(t) + BF x(t − τ(t)). (3)
the presence of malicious attacks. Although communication
networks provide a convenient platform for wide applications Suppose that max{tk+1 − tk k = 0, 1, 2, · · · } ≤ h̄ , where h̄ is a
of modern control systems, they are currently vulnerable to known positive constant. Then the input delay τ(t) satisfies
malicious attacks due to their openness and insufficient 0 ≤ τ(t) ≤ h̄, τ̇(t) = 1, a.e. (4)
protection. For example, by injecting malicious computer
By choosing a Lyapunov-Krasovskii functional candidate as
malware, hackers may falsify data packets or change the
V1 (t) = xT (t)Px(t) + VR (t) with
transmission paths of data packets; and by stealing the
w0 wt
encryption key, cyber attackers may illegally get access to the
VR (t) = ẋT (s)R ẋ(s)dsdθ (5)
remote control center, resulting in collapse of the control system. −h̄ t+θ
Several typical examples about security threats to modern where P > 0 and R > 0, some criteria on stability and the exist-
control systems can be referred to Slammer worm on ence of suitable control gain F are formulated based on Ly-
DavisBesse power plant in Ohio, U.S. (2003) [17], Maroochy punov-Krasovskii stability theorem. However, those criteria
Shire Council’s sewage control system in Queensland, are conservative [31] since they should be satisfied for any
Australia (2000) [18], and Stuxnet worm which targeted many continuous delay τ(t) ∈ [0, h̄], which is a constraint stricter
ZHANG et al.: NETWORKED CONTROL SYSTEMS: A SURVEY OF TRENDS AND TECHNIQUES 3

than (4). Inspired by [32], an improved input delay method for (k = 1, 2, · · ·). Then χk (σ) = Γ(σ)χk (0), and χk (σ) satisfies
sampled-data systems is proposed in [33], where a time-de- 


 χk+1 (0) = χk (T k ) = Γ(T k )χk (0), T k ∈ [h̄m , h̄ M ]
pendent Lyapunov functional is constructed as V2(t) = 



d (9)
xT (t)Px(t) + VU (t) with  χk (σ) = Aχk (σ) + BFχk (0), σ ∈ [0, T k ]
wt dσ
VU (t) = (h̄ − τ(t)) ẋT (s)U ẋ(s)dsdθ (6) It is clear that the function χk (σ) (σ ∈ [0, T k ]) is the state tra-
t−τ(t)
jectory of the sampled-data system on [tk , tk+1 ). The first equa-
where U > 0 . A characteristic of V2 (t) is that it does not grow tion in (9) presents the relationship between x(tk ) and x(tk+1 ) ,
after the sampling times due to the fact that VU (t) > 0 while it which permits the use of the discrete-time stability analysis
vanishes after the sampling times tk (k = 1, 2, · · · ). A stability method. The second equation in (9) reflects the motion of the
theorem tailored for such kinds of time-dependent Lyapunov state trajectory over the period [tk , tk+1 ). Therefore, the con-
functionals is also established in [33]. The stability theorem tinuous-time method can be employed to investigate the sta-
also allows one to construct a discontinuous Lyapunov func- bility of the sampled-data system. The equivalence between
tional as V3 (t) = xT (t)Px(t) + VW (t) with [34] discrete-time and continuous-time stability analysis methods
wt
is disclosed in [35]. The significance of the equivalence theor-
VW (t) = h̄2 ẋT (s)W ẋ(s)dsdθ
tk em is that the positive definiteness of the time-dependent Lya-
π wt
2 punov functionals and discontinuous Lyapunov functionals
− [x (s) − x (tk )]T W [x (s) − x (tk )] ds (7)
4 tk can be relaxed by a looping condition. According to this equi-
where W > 0 and tk ≤ t < tk+1 (k = 0, 1, 2, · · · ). The extended valence theorem, a Lyapunov functional candidate can be con-
Wirtinger inequality ensures that VW (t) ≥ 0 . structed as V4 (t) = xT (t)Px(t) + V0 (σ, χk ) , where V0 : [0, h̄ M ]×
Stability analysis based on the time-independent Lyapunov K → R is a continuous and differentiable functional satisfying
functional VR (t) , the time-dependent Lyapunov functional V0 (T, z(·)) = V0 (0, z(·)), ∀T ∈ [h̄m , h̄ M ], ∀z ∈ K, (10)
VU (t) and the discontinuous Lyapunov functional VW (t) can
be referred to [34] in detail. It can be found through which is a looping condition. In this sense, V0 (σ, χk ) is called
simulation that a stability criterion based on VU (t) produces a a looped functional. A looped functional suggested in [35] is
larger upper bound of h̄ than a stability criterion based on given as
{
either VW (t) or VR (t) . A natural question is why the V0 (σ, χk ) =(T k − σ) ξkT (σ)[S 1 ξk (σ) + 2S 2 χk (0)]
discontinuous Lyapunov functional VW (t) is introduced. To wσ }
answer this question, in [34], a sampled-data system is + χ̇Tk (s)Rχ̇k (s)ds + σχTk (0)Xχk (0) (11)
0
considered by introducing a constant transmission delay into
where ξk (σ) = χk (σ) − χk (0). Since V0 (0, χk ) = V0 (T k , χk )
the control input u(t) , i.e., u(t) = F x(tk − η) , where η > 0 is a
constant. For such a sampled-data system with a positive holds for ∀T k ∈ [h̄m , h̄ M ], the functional V0 (σ, χk ) satisfies the
constant transmission delay, it is shown that the corresponding looping condition (10) while the real matrices S 1 , S 2 , X and R
time-dependent Lyapunov functional VU (t) brings no clear are not necessarily positive definite. Revert V0 (σ, χk ) to a
advantages over the corresponding VR (t) (see Remark 1 in function on time t to obtain
[34]), while the discontinuous Lyapunov functional VW (t) is e =V0 (σ, χk )
V(t)
preferable since it can deliver less conservative and simpler  rt T   r t 
results than the other Lyapunov functionals VR (t) and VU (t).  t ẋ(s)ds   S 1 S2   t ẋ(s)ds 
= (tk+1 − t)  k     k 
However, from Example 1 in [34], one can see that the x (tk ) S 2T (t − tk ) X x (tk )
obtained results using the discontinuous Lyapunov functional wt
are still conservative. Moreover, it is also illustrated that, by + (tk+1 −t) ẋT (s)R ẋ(s)ds, t ∈ [tk , tk+1 ).
tk
introducing a constant transmission delay into the control (12)
input, the upper bound of h̄ is decreased significantly, which
reveals the negative effects of transmission delays on stability By taking such a looped functional, some less conservative
of sampled-data systems. stability conditions are derived in [35]. Borrowing the above
In 2012, a framework was proposed for stability analysis of idea, a two-sided looped functional is introduced in [36] in the
sampled-data systems [35]. In this framework, the closed-loop sense that information on both states at two vertices of the in-
system of the sampled-data system (1) is re-formulated, terval [tk , tk+1 ), i.e. x(tk ) and x(tk+1 ) , is exploited. It reads as
inspired by the lifting modeling technique. Denote V5 (t) = xT (t)Px(t) + V(t), where
 T  
T k := tk+1 − tk ∈ [h̄m , h̄ M ]  η1 (t)   DQ1 DQ2 + X  η1 (t) 
V(t) =    
  T  
 
where h̄m and h̄ M are two positive constants satisfying η2 Q2 D + X T (tk+1 − t)(t − tk )Z η2
h̄m ≤ h̄ M . Then for ∀t ∈ [tk , tk+1 ) , one has that wt
x(t) = Γ(t − tk )x(tk ), where + (tk+1 − t) ẋT (s)R1 ẋ(s)ds
tk
w t−tk w tk+1
Γ(t − tk ) := eA(t−tk ) + eAs dsBF (8) − (t − tk ) ẋT (s)R2 ẋ(s)ds (13)
0 t
Let σ := t − tk ∈ [0, T k ]. Define a function χk (σ) := x(tk + σ) with D := diag{(tk+1 − t)I, (t − tk )I} and
4 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 7, NO. 1, JANUARY 2020

 wt  f (y) = y + qk
 ẋ(s)ds    (14)
   x(tk ) 
η1 (t) :=  w t
tk
 , η2 := 
  . where f (y) is the quantized signal and qk is the quantization
 k+1  x(tk+1 ) error process. The quantization error process qk is usually
ẋ(s)ds
t
modeled as an unknown but bounded disturbance or an addit-
Compared with V(t) e , the two-sided looped functional V(t)
ive white uniform distributed noise with each element being
includes such an integral from t to tk+1. Moreover, the state uniformly distributed over [−0.5M, 0.5M]. For a stochastic
x(tk+1 ) is also included in V(t) . By employing the second- model, the variance of such a quantization error process is
order Bessel-Legendre inequality [37] to bound the integral (M 2 /12)I .
rt rt
terms − tk ẋT (s)R1 ẋ(s)ds and − t k+1 ẋT (s)R2 ẋ(s)ds appearing The quantization effect is investigated in [42] with the
in the derivative of V(t) , the obtained stability criteria are of model above for systems with time-varying delays and input
less conservatism than some existing ones as those in [35]. saturation. Both the domain of attraction and the reachable
Nevertheless, it is not an easy task to design suitable control ellipsoid are derived by using a Lyapunov-Krasovskii
gains F due to the fact that the system matrices are coupled functional method. It should be pointed out that the obtained
with a number of matrix variables in the stability criteria [36]. reachable ellipsoid is dependent on the quantization error
bound 0.5M . Recently, some quantized control issues are
III. Quantization Control addressed by using different control approaches. For example,
Signals should be coded via analog-to-digital coders before networked nonlinear systems subject to the try-once-discard
the signals are released to a communication network. On the protocol scheduling are analyzed in the framework of
one hand, an analog-to-digital coder is usually employed to exponentially ultimate boundedness [43]. Uncertain nonlinear
convert a continuous-time analog signal to its corresponding systems subject to input constraints are investigated in the
discrete-time digital one [38]. The error between the actual framework of approximate dynamic programming [44].
analog value and the converted digital one is usually Moreover, a robust analysis method is developed in the case
unavoidable due mainly to the operation of rounding or of imprecise quantizer level length [45]. For consensus control
truncation. On the other hand, data compression of sampled of multi-agent systems, there are a number of preliminary
signals is quite common in order to reduce network loads such results available in the literature, see, e.g., [46], [47] and
that limited channel bandwidth can be utilized effectively. As references therein.
typical coders, quantizers have gained ever-increasing interest Dynamic quantizers can be employed to guarantee desired
of researchers. Essentially, a quantizer is a symmetrical closed-loop system performance. A dynamic quantier is based
piecewise function with respect to the origin, mapping input on a dynamic quantizer function fµ (y) = µ f (y/µ), where µ is a
values to output values with an infinite countable number of dynamic variable referred to as a “zoom” variable, and f is a
levels [39]. Generally, there are uniform quantizers and certain quantizer. The zoom variable µ is used to adjust both
logarithmic quantizers. the quantizer range and the quantizer error bound based on the
Denote a quantizer as f (·) and the number of quantization value of output measurement y . The simplified form of a
levels as ♯g(ε) over the interval [ε, 1/ε]. Its density is defined dynamic quantizer is described by [48]–[50]
as [39], [40] ∥ fµ (y) − y∥ ≤ δµ, if ∥y∥ < Lµ (15)
♯g(ε) where δµ is the maximum quantization error for all y in the
η f = lim sup .
ε→0 − ln ε range of the quantizer. Based on the dynamic quantizer above,
With this definition, a logarithmical increase of the number of event-triggered input-to-state stability with a size-adjustable
quantization levels occurs if the length of interval [ε, 1/ε] gets set is investigated in [50]. Moreover, by introducing an ad-
larger. A small value of η f corresponds to a coarse quantizer. justable “center” and some “zoom” parameters, a coarse
Especially, η f becomes zero if the number of quantization quantizer is studied in [51] to achieve the input-to-state stabil-
levels is finite, while η f equals infinity if the quantizer is lin- ity with respect to external disturbance. However, when com-
ear. A classical approach to investigating quantization effects munication protocols are involved, investigating the quantiza-
is to treat the quantization errors as unknown but bounded dis- tion effects on NCSs becomes more complex and deserves
turbance or norm bounded uncertainties. In doing so, classical further effort [52]–[54].
robustness analysis tools or bounded stability theory can be
B. Logarithmic Quantizer
employed to address various quantization-induced issues. Fur-
thermore, the desired system performance can be guaranteed For the predetermined set of quantization levels
by adjusting the error bound or the quantization density, and { } { }
U = ± ui ui = ρ j u0 , j = 0, ±1, ±2 · · · ∪ 0
the desired controller or filter parameters can be obtained sim-
with 0 < ρ ≤ 1 and u0 > 0, a logarithmic quantizer f (·) is de-
ultaneously.
scribed by [40], [55]
A. Uniform Quantizer  ui ui


 ui , <y≤ ,
Suppose that the overall quantizer range is [−L, L] with 

 1 + κ 1 −κ


L > 0 . Denote by M the quantizer level length. Let f (y) = 
 0, y = 0, (16)



M = 2L/(2b − 1), where b is the number of bits for digital 

 − f (−y)
sensors. Then the uniform quantizer is given by [41] y<0
ZHANG et al.: NETWORKED CONTROL SYSTEMS: A SURVEY OF TRENDS AND TECHNIQUES 5

with κ = (1 − ρ)/(1 + ρ). For such a logarithmic quantizer, it is packet dropouts has been a fundamental issue, and a great
not difficult to verify that its quantization density satisfies number of results on this topic have been published in the
η f = 2/ ln(1/ρ). In this sense, the quantized measurement is literature, see some monographs [64], [65] and survey papers
modeled as [4], [7]–[9], [66], [67].
f (y) = (1 + △(k))y (17) In this section, we present a review of two common models
of network-induced delays and packet dropouts, namely
where △(k) ∈ [−κ, κ]. random model and input delay model, depending on the
There are three kinds of methods to analyze the system random and time-varying characteristics of network-induced
performance under a logarithmic quantizer. The first method delays and packet dropouts.
is based on a quantization-dependent Lyapunov function [55],
[56]. Using this method, the obtained conditions depend A. Random Model
closely on the parameter κ , leading to some less conservative A random model of network-induced delays and packet
results on control performance. The second one is called a dropouts is quite often and natural. This model is usually
robust analysis method. If one defines F(k) = κ−1 △(k), then based on the discrete-time description of the physical plant.
F(k) is an unknown real-valued time-varying matrix F(k) Suppose that the S-C and C-A channels have different random
satisfying F T (k)F(k) ≤ I . Thus, the quantization control attributions. A random model for network-induced delays is
problem can be transformed into the well-known robust established in [68]. Let τk ∈ [0, τ] and dk ∈ [0, d] be the random
control/filtering problem [57], [58]. Based on the robust time delays in the S-C channel and the C-A channel,
analysis method, a scheme with the form of coupled recursive respectively, where τ and d are two known positive integers.
Riccati difference equations is developed in [58] to deal with A reasonable assumption is made in [68] that two random
the H∞ control of time-varying systems over finite-horizon. It sequences {τk } and {dk } are homogeneous Markov chains that
is worth mentioning that this method is similar to the direct take values in {0, 1, · · · , τ} and {0, 1, · · · , d} with transition
utilization of the constraint ∥△(k)∥ ≤ κ , as is done in [59] for probability matrices [λi j ] and [πrs ], respectively. Under such
recursive filtering. The third method is related to the sector an assumption, the controller is of the form
representation of the quantization error of a logarithmic
u(k) = F(τk , dk−1 )x(k − τk − dk ). (20)
quantizer. The quantization error qk (y) can also be
characterized by the following sector condition It is claimed that the model above of network-induced delays
takes packet dropouts into account if λi j = 0 or πrs = 0 occurs
qTk (y)(qk (y) − 2κy) ≤ 0. (18) for j > i + 1 or s > r + 1. The advantage of the model above is
Then the quantized measurement is rewritten as that the control gain F is not a constant real matrix but a vari-
able one dependent on network-induced delays τk and dk−1.
f (y) = y + qk (y). (19)
In [69], [70], random delays in both S-C and C-A channels
Some well-known control methods using the above sector are modeled as a linear function of a stochastic variable
condition can be employed to cope with the problem of quant- satisfying Bernoulli random binary distribution. Suppose that
ization control, see, e.g., [60] and [61] in detail. δ ∈ R is a stochastic variable being a Bernoulli distributed
The analysis above just takes a snapshot for quantization white sequence with Prob{δ = 1} = δ̄ and Prob{δ = 0} = 1 − δ̄.
control, including two quantizer models and research The following description of sensor measurement yc,k
methods. Some other complex quantization models are not {
involved in this paper. For example, a logarithmic-hysteretic yk = Cxk
(21)
quantizer is proposed in [62] to describe both the quantization yc,k = (1 − δ)yk + δyk−1
and the deadzone. A logarithmic-uniform quantizer and a provides a random model of network-induced delays, where
hysteresis-uniform quantizer are designed in [63], where a xk and yk are the system state and the system output, respect-
stability analysis approach is developed by constructing a ively; and C is a constant real matrix. The output yk is sent to
compensation scheme for the effects of the state quantization, an observer subject to the network-induced delay τd . If τd is
and the system performance can be improved by appropriately less than one sampling period, yc,k = yk . If τd is longer than
adjusting design parameters. one sampling period but less than two sampling periods,
yc,k = yk−1. If δ takes the value of 1 during more than two con-
IV. Networked Control
secutive sampling periods, it is true that τd is longer than two
In an NCS, sensor data and control signals are transmitted sampling periods.
via communication channels from Sensor to Controller (S-C) The idea above is used to model random packet dropouts
and from Controller to Actuator (C-A). Due to the limitation [71]. In the S-C channel, let the stochastic variable αk ∈ R be a
of network bandwidth, there exist inevitable network traffic Bernoulli distributed white sequence. Then the measurement
and congestion, thus causing some network-induced output yk is described by
phenomena during data transmission. Network-induced delays
and packet dropouts are two common phenomena that an NCS yk = αk Cxk + Dwk (22)
encounters, and they are usually regarded as a factor that may where wk is the disturbance input, and C and D are constant
degrade system performance or jeopardize the system real matrices. αk = 0 means that a packet dropout occurs. This
stability. Thus, how to deal with network-induced delays and measurement output yk is sent to an observer-based controller
6 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 7, NO. 1, JANUARY 2020

to generate control signal ûk . Then the control signal ûk is j 1 2 3 4 5 6 7 8 9 10 11 12

transmitted to actuator via the C-A channel. The random pack- τi5

et dropout phenomenon in the C-A channel is modeled as a τi1 τi2 τi3 τi4 τi6
stochastic variable βk ∈ R , which is mutually independent of ik
i1 i5 i2 i3 i4 i6
αk . The stochastic variable βk is also a Bernoulli distributed
white sequence. Then the control input uk to actuate the phys-
Fig. 1. The diagram of time scheduling.
ical plant is given as uk = βk ûk. If βk = 0, then uk = 0 , which
means that the packet is lost during transmission. to τq for q , j, the next new packet transmitted successfully
Note that the model (21) is available just for random may be not ( j + 1, x(( j + 1)T s )). Denote by {i1 , i2 , i3 , · · · } the
network-induced delays while the model (22) just for random time-stamp sequence indicating that those sampled packets ar-
packet dropouts. A unified model to describe both random rive at the actuator in order. Then it is possible that ik > ik+1,
network-induced delays and packet dropouts is presented in but it must be true that ik T s + τik < ik+1 T s + τik+1. Fig. 1 is an
[72]. The idea is to introduce a number of indicator functions example to show the time-stamp sequence {i1 , i2 , i3 , · · · }. That
I{sk =0} and I{sk =d j } ( j = 1, 2, · · · , q), where sk is a stochastic is, i1 = 1, i2 = 3, i3 = 4, i4 = 6, i5 = 2 and i6 = 7 . Clearly, i4 > i5
variable to determine how large the network-induced delay while i4 T s + τi4 < i5 T s + τi5. Therefore, the control input can
and the possibility of data missing could be at the time instant be described by
k . The stochastic variable sk is no longer a Bernoulli distributed
white sequence, while its mathematical expectation satisfies u(t) = K x(ik T s ), ik T s +τik ≤ t < ik+1 T s +τik+1 . (25)
E{I{sk =0} } = Prob{sk = 0} = p0 and E{I{sk =d j } } = Prob{sk = d j } = Define a piecewise continuous function
∑q
p j with j=0 p j ≤ 1, where p j ( j = 0, 1, · · · , q) are known non- τ(t) = t − ik T s , t ∈ [ik T s + τik , ik+1 T s + τik+1 ). (26)
negative constants. In this situation, the system measurement Then the control input u(t) can be rewritten as
is given by
q
u(t) = K x(t − τ(t)), ik T s +τik ≤ t < ik+1 T s +τik+1 . (27)
yk = I{sk =0}C0 xk + Σ j=1 I{sk =d j }C j xk−d j + Dwk . (23)
It is clear that τ(t) includes information on network-induced
∑q
Based on this model, if j=0 p j < 1, the probability of packet delays. Moreover, information on packet dropouts is also in-
∑q cluded in τ(t). In fact, if {i1 , i2 , i3 , · · · } , N, then those packets
dropout is 1 − j=0 p j ; otherwise, no packet dropout occurs.
whose time-stamps belong to the complement set of
The analysis above provides several typical stochastic
{i1 , i2 , i3 , · · · } are dropped out; otherwise, no packet dropouts
methods to model random network-induced delays and packet
happen. This model τ(t) has been widely used in NCSs, which
dropouts. Although there are different approaches reported in
is referred to as an input delay method [73], [74].
the literature to dealing with the same problem, the basic ideas
In the model above, the packet disorder phenomenon is non-
of them are similar to the ones mentioned above. In the recent
negligible. A packet disorder occurs if the packet sent earlier
years, these methods have stirred up much interest of
arrives at the actuator later. That is, if ik+1 < ik, the packet with
researchers in the field and have been widely used to handle a
stamp ik+1 is transmitted in disorder, see for example the
number of issues, such as input-to-state stability, input-output
packet (i5 , x(i5 T s )) in Fig. 1. As a result, the network-induced
stability, H∞ control, Kalman filtering, H∞ filtering, fault-
delay is very big (e.g. τi5 is much bigger than others). Let us
tolerant control and so on.
consider τ(t) on the time interval t ∈ [i5 T s + τi5 , i6 T s + τi6 ). It is
B. Input Delay Model clear that τi5 ≤ τ(t) < (i6 − i5 )T s + τi6, which leads to large
An input delay model is well-used to deal with network- lower and upper bounds of τ(t), definitely making the
induced delays and packet dropouts. Suppose that the obtained stability conditions infeasible. Keeping this in mind,
controller and actuator are event-driven while the sensor is a logic zero-order-hold mechanism is introduced to actively
time-driven with a period T s > 0 . Each sampled signal from drop out those disordered packets [75]. It can be done if the
the sensor is stamped by time, and encapsulated together with ZOH chooses the latest packet to store, see [75] in detail. The
its time-stamp into a packet, say ( j, x( jT s )) ( j = 1, 2, · · · ). The introduction of a logic ZOH provides a feasible way for active
packet ( j, x( jT s )) is transmitted to the controller via the S-C dropouts. Such an active dropout can ensure ik+1 > ik, which
channel and ultimately to the actuator via the C-A channel. If makes the input delay method more useful in the stability
the transmission is successful, the time-consuming denoted by analysis and synthesis of NCSs.
τ j from the sensor to the actuator is called a network-induced Another model can be referred to [76]. The basic idea
delay, which is lumped together from both channels. Thus, the behind this model is based on the assumption that the
time-instant when the packet arrives at the actuator can be properties of both two delays from the S-C channel and the C-
calculated as jT s + τ j. At this time, if state feedback is A channel may not be identical. Denote by d s (t) and da (t),
concerned, the control input that actuates the physical plant respectively, the time-varying delays induced from the S-C
can be given by channel and the C-A channel. Two network-induced delays
d s (t) and da (t) may not be combined together since they are of
u(t)|t→( jT s +τ j )+ = K x( jT s ). (24)
different properties (for instance, one is differentiable while
This control input will be changed only when the next new one is non-differentiable). In this situation, the control input
packet arrives at the actuator. Since τ j is not necessarily equal (for state feedback) can be described as
ZHANG et al.: NETWORKED CONTROL SYSTEMS: A SURVEY OF TRENDS AND TECHNIQUES 7

u(t) = K x(t − d s (t) − da (t)). (28) compared to time-triggered control [82]. However, the study
Then the closed-loop NCS is modeled as a system with two on event-triggered control is still sporadic until a general
additive time-varying delay components. Although plenty of solution to the problem of stability is derived for event-
results on NCSs have been obtained based on this model, how triggered control systems [14]. Boosted by the results, event-
these two delays reflect some exact features of communica- triggered control has been a quite hot topic in the field of
tion networks has not been discussed yet. NCSs even up until now, see, e.g., survey papers [83]–[85]
The input delay model above has an implicit assumption and references therein.
that there are only one sensor, one controller and one activator Event-triggered control is a natural way in the sense that a
in the NCS with no competition of them when gaining access control execution depends on an event rather than the elapse
to networks. If there are more network nodes and only one of a fixed time period. An event-generator is used to detect if
node per transmission is allowed to transmit its packet at any an event is triggered before control executes. In an NCS, the
time, proper communication scheduling (or communication event-generator is usually embedded into the sensor to decide
protocol) among the nodes is necessary. Arguably, there are whether or not the current signal should be sampled, leading
two communication protocols available: static and dynamic. A to event-triggered sampling (Fig. 2(a)). Alternatively, an
static communication protocol, like Round-Robin (RR), event-generator is located behind a sensor to determine if
periodically assigns one node in a prefixed order to transmit sampled signals should be released to the communication
its packet, while a dynamic communication protocol, like Try- network to transmit, resulting in event-triggered transmission
Once-Discard (TOD), dynamically assigns one node to (Fig. 2(b)). Event-triggered sampling is an ideal approach to
transmit its packet based on the network-induced errors [77]. choosing only necessarily sampled signals to carry out some
Under either an RR protocol or a TOD protocol, it is still desired control tasks such that communication resources,
challenging to establish such an input delay model as above to energy resources for battery-based devices, and network
handle network-induced delays and packet dropouts resources can be saved to a maximum extent. Nevertheless,
simultaneously. For this reason, in [77], it is assumed that no event-triggered sampling is liable to fall into Zeno behaviour,
network-induced delays and packet dropouts occur, and in which means that infinite sampling times occur within a
[78], [79], network-induced delays are allowed while packet certain finite time-interval, especially when the system is
dropouts are not expected when RR protocols or TOD disturbed by external noises. Event-triggered transmission,
protocols are involved. however, is based on either periodically or aperiodically
The random and input delay models aforementioned are sampled data, and thus the Zeno behaviour is not involved. In
commonly used to describe both network-induced delays and this sense, such an event-triggered transmission scheme is also
packet dropouts. Based on these two models, the related known as a sampled-data-based event-triggered transmission
closed-loop NCS is modeled as a number of systems either in scheme [86]–[91], which aims at reducing network loads to
the continuous-time domain or in the discrete-time domain, save precious network communication resources. It is
such as stochastic systems, time-delay systems, switched noteworthy that the sampled-data-based event-triggered
systems, Markov jumping systems, impulsive systems, and transmission scheme can save precious network resources
hybrid systems, and thus leading to a large number of results while computation burdens and energy consuming of sensors
on NCSs in the open literature. Although those results may may not be reduced [30]. In the recent years, event-triggered
reveal more or less effects of network imperfections on NCSs, sampling and event-triggered transmission both have received
there is much room for improvement. How to build an exact much attention. The key point is how to devise a suitable
mathematical model for network imperfections, especially for condition to generate an event. In this section, we provide a
packet dropouts, such that the effects of them on NCSs can be brief summary of recent developments on devising event-
disclosed explicitly is of much significance and importance triggering schemes (ETSs) for both event-triggered sampling
both in theory and in practice. and event-triggered transmission.

V. Event-triggered Control Plant Plant

Event- Sensor
Event-triggered control has been long in existence, which Actuator
Generator
Actuator Event-
can date back to 1960 for the research of an accelerometer Sensor Generator

with pulse feedback [80], where pulse generators are used to Communication network Communication network
move a pendulum towards the center position once a deviation Controller Controller
is detected. Such pulse generators embrace the basic idea of
(a) Event-triggered sampling scheme (b) Event-triggered transmission scheme
event-triggered control. In 1993, it was pointed out that
‘control is not executed unless it is required’ when answering Fig. 2. Two schemes of event-triggered control.
the question ‘Should responsive systems be event-triggered or
time-triggered?’ [81]. Unfortunately, this idea did not attract A. ETSs for Event-triggered Sampling
much research interest at that time. Seemingly, it is pretty An ETS is originally motivated by Lyapunov stability.
simple and natural, but its theoretical analysis is quite Consider the control system described by
complicated. An encouraging result contributes to the
experimental analysis on a first-order stochastic system, which ẋ(t) = f (x(t), u(t)) (29)
shows that event-triggered control has a number of advantages where x(t) and u(t) , respectively, are the system state and the
8 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 7, NO. 1, JANUARY 2020

control input, respectively. A suitable controller u(t) = µ(x(t)) γ(||e(t)||)

is designed a priori such that the resultant closed-loop system σα(||x(tk+1)||)

σα(||x(tk+2)||)
ẋ(t) = f (x(t), µ(x(t) + e(t))) (30)
σα(||x(tk+3)||)
is input-to-state stable (ISS) with respect to e , where e is the
t
error vector. Then there exists an ISS Lyapunov function V(x) tk tk+1 tk+2 tk+3
satisfying
Fig. 3. Two schemes of event-triggered control.
ρ1 (∥x∥) ≤ V(x) ≤ ρ2 (∥x∥) (31)
[ ]
V̇(x)|(30) ≤ −α(∥x∥) + γ(∥e∥) (32) η(t) + θ σα(∥x(t)∥) − γ(∥e(t)∥) ≤ 0 (38)

where ρ1 (·), ρ2 (·), α(·) and γ(·) are class K∞ functions1. where θ > 0 is a parameter to be designed, and η(t) satisfies
Event-triggered sampling aims to provide a scheme to η̇ = −β(η) + σα(∥x∥) − γ(∥e∥), η(0) = η0 (39)
sample the system state such that V̇(x) < −ᾱ(∥x∥) ≤ 0 for all
where β(·) is a class K∞ function and η0 > 0.
t ≥ 0 , where ᾱ(·) is a class K∞ function. We denote the Under the dynamic ETS (38), it is proven that η(t) ≥ 0 for
sampling sequence by t ≥ 0 . At tk , σα(∥x(tk )∥) > γ(∥e(tk )∥) = 0 . Let t increase slowly
0 = t0 , t1 , t2 , · · · , tk , tk+1 , · · · (33) from tk . Then σα(∥x(t)∥) > γ(∥e(t)∥). Once σα(∥x(t)∥) = γ(∥e(t)∥)
Then x(t) = x(tk ) for tk ≤ t < tk+1. Set e(t) = x(t) − x(tk ) , is satisfied, the dynamic ETS (38) allows t to keep going until
t ∈ [tk , tk+1 ) . Suppose that the communication network has (38) holds, which means that the inequality σα(∥x(t)∥) ≥
ideal quality of service. Then the control input can be given by γ(∥e(t)∥) does not need to be always satisfied for t ≥ 0 . In this
sense, the minimum inter-event time min{tk+1 − tk |k = 0, 1,
u(t) = u(tk ) = µ(x(tk )) = µ(x(t)+e(t)), t ∈ [tk , tk+1 ) (34) 2, · · · } is larger than that of the static ETS (35). On the other
In order to maintain the system stability, an event-triggering hand, if rewriting (38) as
condition can be devised as
η(t)
σα(∥x(t)∥) ≤ γ(∥e(t)∥) σα(∥x(t)∥) − γ(∥e(t)∥) ≤ −
(35) θ
where the threshold σ ∈ (0, 1) is a constant. Once the condi- the dynamic ETS reduces to the static ETS (35) if θ → ∞.
tion (35) is satisfied, an event, i.e., sampling the system state, Such a dynamic ETS is refined in [93] to deal with the distrib-
is immediately invoked. In this sense, the sampling instant can uted set-membership estimation issue for a class of discrete-
be determined by time linear time-varying systems over a wireless multi-sensor
{ network. It is analytically and numerically shown that such a
t0 = 0
(36) dynamic ETS can lead to larger average inter-event time and
tk+1 = inf{t > tk |σα(∥x(t)∥) ≤ γ(∥e(t− )∥)}. thus less totally released data packets.
Such an ETS can deduce the following inequality Note that both the static ETS (36) and the dynamic ETS
(38) cannot guarantee a positive minimum inter-event time to
V̇(x)|(30) ≤ −α(∥x∥) + γ(∥e∥) ≤ −(1 − σ)α(∥x(t)∥) (37)
avoid Zeno behaviour, especially for perturbed physical plants
which ensures the stability of the closed-loop system under [15]. In this situation, a mixed time- and event-triggered
the sampling scheme due to σ ∈ (0, 1). scheme is proposed in [94], where the sampling instant is
The event-triggering condition (35) depends on the system calculated by
state x(t) and its error e(t) rather than a fixed time period. {
t0 = 0
Thus, signals are sampled only when necessary. In what (40)
follows, we give a detailed explanation about the condition. tk+1 = inf{t > tk + T |F (σ, x, e) ≤ 0}
First, note that α(·) and γ(·) are continuous and strictly with T > 0 denoting a constant and F (σ, x, e) representing a
increasing functions, and that e(tk ) = 0 . Hence, at tk , one has certain function with respect to σ, x and e . Clearly, with this
σα(∥x(tk )∥) > γ(∥e(tk )∥) = 0 . According to continuity, when t scheme, a positive minimum inter-event time greater than T is
is increasing from tk , the inequality σα(∥x(t)∥) > γ(∥e(t)∥) guaranteed even though external disturbance is imposed on
holds. Second, if the closed-loop system is deviating from its the system. This scheme is also studied in [95] by presenting a
stability, x(t) will be getting larger, leading to the increase of switching model for the closed-loop system. Mixed time- and
both σα(∥x(t)∥) and γ(∥e(t)∥). Once σα(∥x(t)∥) = γ(∥e(t)∥) is event-triggered schemes have strong application background.
satisfied, a new event time tk+1 is generated such that e(t) In fact, in non-rush hours, such a mixed triggering scheme can
returns to zero, i.e. e(t)|t=tk+1 = 0, which turns the closed-loop be used to control the traffic lights’ change. Once a car ar-
system back to stability, see Fig. 3. rives just before a stopping line, an event should be triggered
The ETS above is known as a static ETS. To relax the to turn the light from red to green. Such a task can be com-
event-triggering condition (35), a dynamic ETS is proposed in pleted within a fixed time T after the event is triggered.
[92]. The basic idea is to introduce an internal dynamic A common feature of the ETSs above is that the system
variable η(t) in (35) such that state x(t) should be monitored at any time. To carry out such a
task, extra hardware should be embedded into the system such
1A function α(·) is said to be of class K∞ if it is continuous, strictly increas-
that an ETS works well. This definitely incurs an increase of
ing and satisfies lim s→∞ α(s) = ∞ and α(0) = 0 . cost. In this situation, self-triggered control comes to the fore
ZHANG et al.: NETWORKED CONTROL SYSTEMS: A SURVEY OF TRENDS AND TECHNIQUES 9

{
[96]. The concept of self-trigger means that the next sampling t0 = 0
(44)
time can be predicted by the current sampled signal, which tk+1 h = tk h + min{ jh (43) holds, j = 1, 2, · · · }.
can be generally described by Analysis in depth on the ETS (44) can be found in [84]. As
tk+1 = tk + F (x(tk )) (41) a complementary, we present one more insight to the
where F (x(tk )) is a prediction function with respect to x(tk ). It condition (43) compared with (35) and (38). A clear
is clear that a self triggering scheme allows us to compute the observation on the event-triggering conditions (35) and (38) is
next time tk+1 offline based on the available state x(tk ). that they are closely dependent on the derivative of a certain
However, how to define an exact prediction function is a chal- Lyapunov function. This is because they should be designed
lenging issue. Based on an ISS Lypunov function and L2 gain to passively cater for some preset closed-loop performance.
performance, a prediction function is defined as [96] However, the condition (43) is independent of any Lyapunov
function and its derivative. The objective of (43) is to reduce
1 ( ρ(x(tk )) ) the number of sampled data packets. According to the real-
F (x(tk )) = ln 1 + δα (42)
α µ0 (x(tk )) time status of network traffic, one can adjust the threshold to
where α, δ, α, ρ(·) and µ0 (·) are referred to [96]. For exponen- release more or less data packets to the network. Then, based
tial ISS, a different prediction function is designed in [97]. on the ETS, we actively design suitable controllers such that
ETSs for event-triggered sampling have gained considerable the closed-loop system can achieve some expected
attention during the past decade. A large number of results performance. In this sense, the condition (43) is more practical
have been reported on ETSs for linear or nonlinear or than those (35) and (38). Thus, it has been widely used to deal
stochastic systems in either continuous-time or discrete-time with a number of control issues over networks, such as H∞
domain, see some recent results, e.g., [98]–[106]. However, control, L2 control, sliding mode control, fuzzy control,
most results are derived from the digital implementation adaptive control, stochastic control, distributed control, fault
perspective. First, design a suitable controller such that the diagnosis, state estimation and filtering, consensus and
closed-loop system has some proper performance. Then, formation control.
design an ETS to implement the predesigned controller in a Recently, the event-triggering condition (43) is modified as
digital platform such that the closed-loop performance can be
ψT (tk , j)Ω1 ψ(tk , j) ≥ εxT (tk h)Ω2 x(tk h) (45)
retained while the number of sampling can be reduced as
much as possible. This represents a kind of passive event- where Ω1 and Ω2 are two different weighting symmetric posit-
triggered control schemes since the ETS should cater for the ive definite matrices [107], [108]. The introduction of the
predesigned controller. Active event-triggered control is to weighting matrix Ω in (43) is just to enhance the feasibility of
actively design a suitable controller for the physical plant related linear matrix inequalities [109]. If Ω1 and Ω2 are two
under a certain ETS. That is, the controller is designed to cater design parameters, then for any threshold ε ∈ (0, 1), the ETS
for the ETS. As such, some closed-loop performance can be may be reduced to a time-triggering scheme since there may
ensured while the number of sampling and/or transmission can exist Ω1 and Ω2 such that no event will be triggered. Thus, the
be reduced to the minimum extent. Nevertheless, such an event-triggering condition (45) works well unless Ω1 and Ω2
active event-triggered control scheme remains challenging are given a priori. Moreover, from some simulation results,
although is of great practical significance. Another point is we find that the ETS indeed can be regarded as a time-trigger-
that event-triggered control should take communication ing scheme if we set Ω1 = Ω−12 . However, this finding has not
constraints like network-induced delays and packet dropouts been proven from theoretical points.
into account, while just few results on them are reported, see, The ETSs for both event-triggered sampling and event-
e.g., [99], [100]. triggered transmission mentioned above are designed based on
a constant threshold parameter σ or ε. As a matter a fact, the
B. ETSs for Event-triggered Transmission choice of such a threshold parameter greatly affects a data
An ETS for event-triggered transmission aims at saving transmission rate, which is in general time-varying, through
precious network resources by reducing the number of data practical communication channels. For example, raw
packets before being released to a communication network. It transmission rates over an IEEE 802.11b wireless local area
works at every sampling instant to choose necessary data network may vary over time at 1, 2, 5.5, and 11 Mb/s due to
packets to transmit through networks. Suppose that the time-varying interference and random wireless fading [110]. It
sequence of sampling instants is {ih|i = 1, 2, · · · }, where h > 0 is thus reasonable to consider an essentially time-varying
is the sampling period. The event-triggering condition for an threshold parameter within ETSs. In [110], by assuming that
ETS is usually devised as follows [16] the threshold parameter in the ETS is time-varying with
attainable upper and lower bounds, the problem of distributed
ψT (tk , j)Ωψ(tk , j) ≥ εxT (tk h)Ωx(tk h) event-triggered H∞ consensus filtering for a discrete-time
ψ(tk , j) = x((tk + j)h) − x(tk h) (43) linear system over a wireless sensor network is addressed
where ε ∈ (0, 1) is a threshold and Ω a weighting symmetric through a threshold-parameter-dependent approach. Although
positive definite matrix; and tk h is the latest time instant when such an ETS provides certain freedom to adjust the inter-event
a data-packet is triggered. Based on this condition, the next times, it does not fully explore the relationship between the
time instant tk+1 h can be determined by time-varying threshold parameter and the variant transmission
10 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 7, NO. 1, JANUARY 2020

rates over wireless transmission channels, which leaves much Actuator 1 Sensor 1
room for improvement. Recently, in the context of multi-agent A1

…
Plant
coordination control, for the ith agent, a dynamical ETS is
proposed as [111] Actuator M Sensor N

ψTi (tki , υ)Ω1 ψi (tki , υ) ≥ ε((tki +υ)h)xiT (tki h)Ω2 xi (tki h) (46) Physical layer
Communication network
where ψi (tki , υ) = xi ((tki + υ)h) − xi (tki h) and ε(·) is a time-vary-
ing threshold parameter whose value is adjusted according to A3 Cyber layer A2
a dynamic rule. A key property of the dynamic threshold para- -State Estimator/Monitor
meter lies in that ε(·) is a monotone nonincreasing sequence, -Anomaly/Attack Detector

…
…
which is shown to be beneficial in ensuring a better tradeoff -Controller
-Other functions
between reducing inter-agent communication frequency and
Remote control center
preserving an expected control performance compared with
the static ETSs. This is because the static ETSs may not trig-
Fig. 4. A general architecture of an NCS under data availability and integ-
ger and transmit those data packets that contain useful inform-
rity attacks. A1: data integrity attacks on the system state; A2: data availabil-
ation to preserve the desired control performance. However,
ity and/or integrity attacks on sensor data; A3: data availability and/or integ-
how to evaluate which data packet possesses useful informa-
rity attacks on control data.
tion for guaranteeing system performance is not discussed in
[111] and deserves further investigation. properties of sensor and control data, i.e., confidentiality,
VI. Security Control availability, and integrity. Notice that information-theoretic
studies on attacks focus mainly on employing some
The ever-increasing demand for guaranteeing reliable and
information assurance techniques (e.g., data authentication,
safe monitoring and operation of modern critical
access control, cryptographic algorithms) to preserve data
infrastructure, such as smart grids, water and gas supply
confidentiality, while control-theoretic strategies against
distribution systems, intelligent transportation systems and so
attacks are devised primarily from evaluating and maintaining
on, has stimulated intensive research and industry interest in
the availability and trustworthiness of sensor and control data.
developing security control strategies for NCSs.
This is because information security techniques do not exploit
In this section, we first describe a general security control
the compatibility of data with the dynamics of the underlying
problem for an NCS with a discrete-time linear plant under
physical plant and they are ineffective against insider or
various attacks. Then, we classify and discuss some recent
sophisticated attackers targeting dynamics of the physical
results on security control based on two broad categories of
data availability and integrity attacks. plant [112]. Moreover, data availability and data integrity are
two common security goals in the context of system
A. Security Control Problem monitoring and control. From Fig. 4, one can see that the
A typical architecture of an NCS is composed of a physical adversary may inject malicious attack signals a p (k) and a s (k)
layer and a cyber layer. The physical layer consists of a on the sensor measurement and/or control commands by
physical plant and a number of physical components like compromising their availability and/or integrity.
sensors and actuators, while the cyber layer is comprised of An attacker’s strategy generally relies on three factors: the
several communication devices, monitors, detectors, attack duration Ta := {k0 , k0 + 1, · · · , kL }, the attack profiles
controllers and other functions, as shown in Fig. 4. Fa := {Θ1 , Θ2 , · · · , ΘL } and the stealthiness, where k0 denotes
Suppose that the physical plant is modeled as a discrete- the attack start time; kL = k0 + L − 1 represents the attack stop
time linear time-invariant system described by time, and Θ1 , Θ2 , · · · , ΘL are some compatible non-zero
{ vectors. The attack duration and profiles characterize the
x(k + 1) = Ax(k) + Bu(k) + Bw w(k) + B p a p (k) adversary’s destructiveness while the stealthiness reflects the
(47)
y(k) = Cx(k) + Dv(k) + D s a s (k) detectability of the adversary by any anomaly/attack detector.
where x(k), u(k), w(k) and y(k) are the system state, control in- It is noteworthy that the stealthiness of an adversary often
put, external disturbance input (or process noise), and system depends on a priori model knowledge of the plant, disclosure
measurement output, respectively; v(k) is the sensor measure- resources and disruption capabilities [22], based on which, the
ment/communication noise; a p (k) = col{a x (k), au (k)} denotes adversary can perform his/her malicious actions by carefully
the system attack vector with a x (k) and au (k) related to the designing the attack vector a(k) = col{a p (k), a s (k)} as
system state and the control input, respectively; a s (k) is the {
Θk−k0 +1 , k ∈ T a
sensor attack vector; and A, B, Bw , B p ,C, D, and D s are real a(k) = (48)
matrices of compatible dimensions. The system (47) is con- 0, k < Ta.
sidered to be in normal operation if a p (k) ≡ 0 and a s (k) ≡ 0 for The attack vector a(k) can be used to reveal the accessibility
∀k ∈ N . and integrity of control and sensor data under attack. For
Sophisticated attackers usually impose various types of example, by properly choosing B p a p (k) = −Bu(k) and
attacks upon remote control systems. The malicious impact of D s a s (k) = −Cx(k), control and sensor data can be blocked on
an attack on NCSs may be eventually revealed by three key the receiver’s side, leading to data interruption; by letting
ZHANG et al.: NETWORKED CONTROL SYSTEMS: A SURVEY OF TRENDS AND TECHNIQUES 11

a p (k) and a s (k) be arbitrary non-zero vectors, the integrity of load frequency control problem for multi-area power systems
control and sensor data can be compromised. Note that a under energy-limited DoS attacks is studied, where the
sophisticated adversary will coordinate the system attack maximum DoS attack duration is given. In [21], the optimal
vector a p (k) and the sensor attack vector a s (k) to completely control issue on DoS attacks under a maximum number of
manipulate the remote control system without being jamming actions within a prescribed control horizon is
detected/recovered or with vastly increased time and cost for considered. The problem of event-triggered control is
detection/recovery. addressed [115] for NCSs under some “well-structured”
From the analysis above, a security control problem of the pulse-width modulated DoS jamming attacks. To remove the
system (47) can be stated as: Given the available sensor assumption on the “structure” of the DoS attack signal, an
measurement {y(k), k ∈ N} of system (47) subject to the attack output-feedback dynamic event-triggered control framework
vector a(k), develop a desirable security control scheme is presented [23] to cope with a general class of DoS attacks
u(k) = C(y(k)) such that: 1) the true system state can be characterized by frequency and duration. It is shown that both
estimated accurately and reliably; and 2) the manipulated resource-awareness and resilience can be investigated based
system can be recovered from attacked operation back to on this unified framework. The above secure control methods
normal. In what follows, some results addressing such a are devised just for some certain categories of DoS attacks.
security control issue of NCSs under data availability and data Another alternative method is to use the game theory, where
integrity attacks are discussed. the controller and adversary are deemed as two players with
competing goals in a game. Taking [116] for example, a game
B. Security Control Under Data Available Attacks between the defender and the attacker is described as: the
Data availability refers to the ability of sensor and control defender compensates for the attack-induced performance loss
data of being accessible at their destinations. A lack of while the attacker drives the control system out of the safety
availability results in an interruption, also called as a denial of zone using the lowest attacking intensity. Based on the game
service (DoS), of the data. To launch a data availability attack theory, a number of optimal algorithms for both cyber
(often occurs at the attack points A2 and A3 in Fig. 4), the defenders and DoS attackers are well designed.
adversary can jam the communication channels, attack the Although a data availability attack is not stealthy to system
routing protocols, flood with network traffic some devices, operators, its detection is generally nontrivial because the
compromise system components and prevent them from absence of sensor and control data may be misdiagnosed as
sending/receiving data, etc. Thus, data availability attacks are network-induced constraints like packet dropouts, which
also referred to as DoS attacks [21], [23], [113]–[116] or deserves further investigation. In fact, an adversary may
jamming attacks [118]. purposely take full advantage of randomness to imitate packet
A direct consequence of a data availability attack on the dropouts that occur due to congestion or channel constraints
NCS is that the sensor and control data cannot arrive at their [123]. It should be also noted that some existing approaches
destinations on time, and thus are deemed as “dropped” or treat DoS/jamming attacks as packet dropouts when
“lost”. A common model for such an attack is to choose establishing mathematical attack models such that quantitative
B p a p (k) = −BBu u(k) and D s a s (k) = −ByCx(k), where Bu and analysis can be made on performance degradation caused by
By are two binary incidence matrices indicating which attacks. How to expose the active behaviors of attackers and
channels suffer the data interruption. On the other hand, given the passive impacts of packet dropouts should be further
limited energy budget and/or evitable exposure of exploited.
himself/herself, an adversary will not persistently but
randomly decide when and where to launch the attacks. C. Security Control Under Data Integrity Attacks
Furthermore, the attacked data launched on the Data integrity refers to the trustworthiness of sensor and
communication channels may also experience random control data. A lack of integrity results in deception, false
network congestion and transmission rate constraints, etc. data, or misleading actions. To launch a data integrity attack
Thus, it seems natural to mathematically model a data (often occurs at the attack points A2 and A3 in Fig. 4), an
availability attack as random packet dropouts, e.g., a adversary can modify the contents of sensor measurement and
Bernoulli-type adversary B p a p (k) = −α(k)BBu u(k) and control commands by inserting erroneous or misleading data
D s a s (k) = −β(k)ByCx(k), where α(k) and β(k) are Bernoulli when they are transmitted via communication channels. Note
random variables with certain probabilities. Several methods that a data integrity attack is generally more cunning and
to deal with Bernoulli packet dropouts [119], [120] and difficult to tackle as opposed to a data availability attack due
Markovian packet dropouts [121], [122] can be employed to to the fact that the tampered data are disseminated in an
evaluate the impact of such attacks on the closed-loop system epidemic manner over the network and may cause the
performance, see, e.g., [21], [22] on Bernoulli-type correlated negative impact on the system performance and
adversaries and [113], [123] on Markovian-type adversaries. It operation. From the controller's perspective, intensive research
should be also mentioned that packet dropouts caused by data efforts have been made on a number of issues: (i) How to
availability attacks need not follow some specific probability accurately and resiliently estimate the true system state when
distributions. In this sense, new methods on analysis and the system is manipulated by data integrity attacks; (ii) How
control design against attacks are proposed at the right to detect data integrity attacks; and (iii) How to safely control
moment. To mention a few, in [114], the event-triggered H∞ the system when it is under data integrity attacks. Up until
12 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 7, NO. 1, JANUARY 2020

now, various types of data integrity attacks have been adversaries may possess sufficient system model knowledge,
identified in the control-theoretic literature, such as deception disclosure and disruption resources to perform covert
attacks [38], [117], [124]–[126]; false data injection attacks (stealthy) attacks [135]–[137] that can completely bypass
[127], [128]; replay attacks [20], [22], [129]; sparse sensor traditional anomaly detectors. For example, a typical covert
attacks [130]. attack model introduced in [135] has the following form [138]
The issue (i) is related to the resilient and secure estimation. ● (System attack vector). Depending on the available
With the assumption that the data integrity attack signals are resources of the adversary, the vector a p (k) in (47) can be
bounded [131]–[133] or follow specific probabilistic arbitrarily selected;
distributions [38], [126], one of approaches to solving the ● (Sensor attack vector). The vector a s (k) in (47) is properly
problem is to model the data integrity attack signals as system designed as
external disturbance, process or measurement noise, as done 

 xa (k + 1) = Axa (k) + Ba a p (k),
in the conventional literature on robust control and filtering. It 


a s (k) = −Cxa (k) − B p a p (k) (49)
should be pointed out that if a naive adversary only makes 

small perturbations in system dynamics and/or data, such an xa (k) = 0, k ≤ k0
ineffective attack strategy can be readily combated by the where xa (k) = x(k) − xh (k) denotes the system state compon-
inherently robust and resilient estimators and controllers. On ent induced by the attack and xh (k) is the system state com-
the other hand, significant perturbations caused by the ponent under no attack.
adversary can be easily diagnosed by some anomaly detectors. The principle of the above covert attack strategy lies in the
There is no doubt that a cunning adversary will carefully use following two aspects. First, the adversary applies some pre-
the available resources to generate proper perturbations while defined data integrity attacks modeled by a p (k) to corrupt the
maximizing the damage to the system. system state. Second, he or she adds some spurious or
Notice that the ultimate goal of malicious attacks is to misleading data modeled by a s (k) to cover the changes of
deteriorate the desired system performance or destabilize the system measurement so as to render the attacks completely
underlying system. Designing robust and resilient state undetectable and stealthy to the remote detector. A covert
estimators is an effective way to mitigate the impact of these attack is completely stealthy to traditional anomaly detectors
attacks on the system performance and operation. if system measurement from all sensors are compromised (i.e.,
Nevertheless, it is essentially desirable to develop suitable D s = I ) [135], [138], and can be implemented by using
detection mechanisms that can locate and identify the attacks different strategies, such as replaying, false data injecting,
in a timely fashion [25], and further warn the system operator zero-dynamics and so on, see [22] in detail.
to allow corrective actions, thus mapping the issue (ii) into the
attack detection. Generally, the following two attack detection VII. Conclusion and Some Challenging Issues
approaches have been widely employed in the context of NCSs have been a hot topic for over two decades. Boosted
secure estimation and control: 1) χ2-detector based on Kalman by the rapid development of network communication
filters; and 2) Fault detection and isolation techniques. technologies, NCSs have found ever-increasing applications
However, such a χ2-detector approach requires that the attack in a wide range of areas, such as smart grids, intelligent
signals obey Gaussian distribution [132] and may fail to detect transportation, unmanned space (surface/marine) vehicles,
sophisticated deception attacks with good knowledge of the robotic networks, and so on [139]. Those applications in turn
system. The widely-used fault detection and isolation tools, on have stimulated theoretical research of NCSs to higher levels
the other hand, treat attacks and faults in the same manner, by bringing a large number of practical issues from two
which greatly simplifies the detection procedures [134]. When primary topics ‘control of networks’ and ‘control over
system faults and malicious attacks are both present, this networks’. This paper has surveyed recent theoretical
approach might be efficient if one could ascertain the developments of five control issues from the perspective of
similarities between them. Nevertheless, faults and attacks ‘control over network’, including sampled-data control,
also possess inherently distinct features in terms of quantization control, networked control, event-triggered
stealthiness, destructiveness, and occurrence places, which control, and security control. In what follows, we present a
make these tools invalid to tackle a smart and cunning number of challenging issues worthy of future research.
attacker. To the best of the authors' knowledge, it remains 1) A co-design scheme is expected for an NCS by
challenging to establish a rigorous theoretical secure simultaneously taking ‘control of networks’ and ‘control over
estimation and control framework to effectively identify the networks’ into account. With such a scheme, on the one hand,
effects caused by malicious attacks from faults. suitable communication protocols [53], [140] can be designed
Both issues (i) and (ii) provide the first step towards the to ensure good QoS of communication networks, and on the
security guarantee of the closed-loop system, while issue (iii) other hand, notable control strategies can be provided such
represents the ultimate goal of the system designers, that the NCS achieves desirable control performance to carry
preserving the security of the closed-loop system. It is noted out expected control tasks;
that most researches along this line focus on investigating 2) In most of event-triggered control results, it is assumed
system vulnerabilities or preserving system resilience and that all event-triggered packets have been successfully
safety against a particular type of attacks or some detectable transmitted. However, in the open network environments,
attacks. Nevertheless, it is not uncommon that sophisticated there may exist cyber attacks, some triggered packets may be
ZHANG et al.: NETWORKED CONTROL SYSTEMS: A SURVEY OF TRENDS AND TECHNIQUES 13

not transmitted to the following points safely and timely. How May 2013.
to develop resilient and secure event-triggered communication [7] R. A. Gupta and M. Y. Chow, “Networked control system: overview
and research trends,” IEEE Trans. Ind. Electron., vol. 57, no. 7,
schemes in the presence of malicious attacks is an attractive pp. 2527–535, Jul. 2010.
topic; [8] X.-M. Zhang, Q.-L. Han, and X. Yu, “Survey on recent advances in
3) Although a large number of results have been reported on networked control systems,” IEEE Trans. Ind. Inform., vol. 12, no. 5,
coordination of networked MASs from various perspectives, pp. 1740–1752, Oct. 2016.
there remains much room to explore secure coordination of [9] X. Ge, F. Yang, and Q.-L. Han, “Distributed networked control
networked MASs subject to various network-induced systems: a brief overview,” Inf. Sci., vol. 380, pp. 117–131, Feb. 2017.
constraints and attacks simultaneously; [10] M. S. Mahmoud and M. M. Hamdan, “Fundamental issues in
networked control systems,” IEEE/CAA J. Autom. Sinica, vol. 5, no. 5,
4) A smart grid which incorporates physical power grids, pp. 902–922, Sep. 2018.
information and communication systems, and human societies [11] R. M. Murray, Control in an Information Rich World: Report of the
is a typical example of networked industrial control systems. Panel on Future Directions in Control, Dynamics, and Systems.
Due to the use of communication networks, reliability and Philadelphia, PA: SIAM, 2003.
security of smart grids may inevitably be threaten and [12] Q. Song, F. Liu, G. H. Wen, J. D. Cao, and Y. Tang, “Synchronization
of coupled harmonic oscillators via sampled position data control,”
compromised by network-induced constraints and attacks. As IEEE Trans. Circuits Syst. I: Reg. Pap., vol. 63, no. 7, pp. 1079–1088,
a result, it is essential and critical to develop an advanced Jul. 2016.
control method for smart grids that can deliver reliability and [13] D. Zhang, Q.-L. Han, and X. Jia, “Network-based output tracking
security of energy supply, and efficiency of operation by control for a class of T-S fuzzy systems that can not be stabilized by
nondelayed output feedback controllers,” IEEE Trans. Cybern.,
taking network-induced constraints and attacks into account; vol. 45, no. 8, pp. 1511–1524, Aug. 2015.
5) The majority of existing control literature focuses on
[14] P. Tabuada, “Event-triggered real-time scheduling of stabilizing
ensuring asymptotic or exponential convergence of NCSs control tasks,” IEEE Trans. Autom. Control, vol. 52, no. 9,
under network-induced constraints. In real control pp. 1680–1685, Sep. 2007.
implementation, however, it is more desirable and practical [15] D. P. Borgers and W. P. M. H. Heemels, “Event-separation properties
for an NCS to achieve fast convergence while preserving good of event-triggered control systems,” IEEE Trans. Autom. Control,
vol. 59, no. 10, pp. 2644–2656, Oct. 2014.
robustness and control performance. It is well acknowledged
[16] D. Yue, E. Tian, and Q.-L. Han, “A delay system method for designing
that adopting fixed-time control can guarantee that the closed- event-triggered controllers of networked control systems,” IEEE
loop systems achieve stability or stabilization in a fixed or Trans. Autom. Control, vol. 58, no. 2, pp. 475–481, Feb. 2013.
prescribed time [141]–[143]. However, it becomes more [17] S. Kuvshinkova, SQL slammer worm lessons learned for consideration
difficult and challenging for NCSs to ensure the fixed-time by the electricity sector, North Amer. Elect. Rel. Council, Atlanta, GA,
USA, NIST SP800-53, Jun. 2003.
convergence due to the presence of aforementioned various
[18] J. Slay and M. Miller, “Lessons learned from the Maroochy water
network-induced constraints and security threats, which breach, ” in Critical Infrastructure Protection, E. Goetz and S. Shenoi,
deserves deep investigation; and Eds. Boston, MA, USA: Springer, 2008, pp. 73–82.
6) Besides data availability and data integrity, another [19] J. Farwell and R. Rohozinski, “Stuxnet and the future of cyber war,”
concern on security control in NCSs is data confidentiality, Survival, vol. 53, no. 1, pp. 23–40, Jan. 2011.
which is closely related to privacy protection. How to devise [20] Y. L. Mo and B. Sinopoli, “Secure control against replay attacks, ” in
effective privacy-preserving security control schemes for Proc. 47th Annu. Allerton Conf. Communication, Control, and
Computing, Monticello, USA, 2009, pp. 911–918.
NCSs in the presence of network-induced constraints will be a
[21] S. Amin, A. A. Cárdenas, and S. S. Sastry, “Safe and secure networked
significant and challenging issue. Furthermore, it will be control systems under denial-of-service attacks, ” in Proc. 12th Int.
promising to deeply investigate the impacts of combinational Conf. Hybrid Systems: Computation and Control, San Francisco, CA,
data availability, integrity and confidentiality attacks on USA, 2009, pp. 31–45.
control performance of NCSs. [22] A. Teixeira, I. Shames, H. Sandberg, and K. H. Johansson, “A secure
control framework for resource-limited adversaries,” Automatica,
vol. 51, pp. 135–148, Jan. 2015.
References [23] V. S. Dolk, P. Tesi, C. De Persis, and W. P. M. H. Heemels, “Event-
triggered control systems under denial-of-service attacks,” IEEE
[1] K. D. Kim and P. R. Kumar, “Cyber-physical systems: a perspective at
Trans. Control Netw. Syst., vol. 4, no. 1, pp. 93–105, Mar. 2017.
the centennial,” Proc. IEEE, vol. 100, no. S, pp. 1287–1308, May 2012.
[2] L. Ding, Q.-L. Han, L. Y. Wang, and E. Sindi, “Distributed [24] E. Mousavinejad, F. Yang, Q.-L. Han, and L. Vlacic, “A novel cyber
cooperative optimal control of DC microgrids with communication attack detection method in networked control systems,” IEEE Trans.
delays,” IEEE Trans. Ind. Inform., vol. 14, no. 9, pp. 3924–3935, Sep. Cybern., vol. 48, no. 11, pp. 3254–3264, Nov. 2018.
2018. [25] D. Ding, Q.-L. Han, Y. Xiang, X. Ge, and X.-M. Zhang, “A survey on
[3] H. Sandberg, S. Amin, and K. H. Johansson, “Cyberphysical security security control and attack detection for industrial cyber-physical
in networked control systems: an introduction to the issue,” IEEE systems,” Neurocomputing, vol. 275, pp. 1674–1683, Jan. 1674.
Control Syst. Mag., vol. 35, no. 1, pp. 20–23, Feb. 2015. [26] X. Ge, Q.-L. Han, and X.-M. Zhang, “Achieving cluster formation of
[4] J. P. Hespanha, P. Naghshtabrizi, and Y. G. Xu, “A survey of recent multi-agent systems under aperiodic sampling and communication
results in networked control systems,” Proc. IEEE, vol. 95, no. 1, delays,” IEEE Trans. Ind. Electron., vol. 65, no. 4, pp. 3417–3426, Apr.
pp. 138–162, Jan. 2007. 2018.
[5] Y.-L. Wang and Q.-L. Han, “Network-based modelling and dynamic [27] T. W. Chen, and B. A. Francis, Optimal Sampled-Data Control
output feedback control for unmanned marine vehicles in network Systems, Berlin, Heidelberg: Springer, 1993.
environments,” Automatica, vol. 91, pp. 43–53, May 2018. [28] K. J. Åström and B. Wittenmark, Computer-Controlled Systems:
[6] X.-M. Zhang and Q.-L. Han, “Network-based H∞ filtering using a theory and Design, 3rd ed. Englewood Cliffs, NJ: Prentice-Hall, 1997.
logic jumping-like trigger,” Automatica, vol. 49, no. 5, pp. 1428–1435, [29] L. Hetel, C. Fiter, H. Omran, A. Seuret, E. Fridman, J.-P. Richard, and
14 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 7, NO. 1, JANUARY 2020

S. I. Niculescu, “Recent developments on the stability of systems with triggered quantized control for input-to-state stabilization of linear
aperiodic sampling: an overview,” Automatica, vol. 76, pp. 309–335, systems with distributed output sensors, ” IEEE Trans. Autom.
Feb. 2017. Control, to be published. doi: 10.1109/TAC.2019.2900338
[30] X. Ge, Q.-L. Han, D. Ding, X.-M. Zhang, and B. Ning, “A survey on [51] Y. Sharon and D. Liberzon, “Input to state stabilizing controller for
recent advances in distributed sampled-data cooperative control of systems with coarse quantization,” IEEE Trans. Autom. Control,
multi-agent systems,” Neurocomputing, vol. 275, pp. 1684–1701, Jan. vol. 57, no. 4, pp. 830–844, Apr. 2012.
1684. [52] D. Ding, Q.-L. Han, Z. Wang, and X. Ge, “A survey on model-based
[31] L. Mirkin, “Some remarks on the use of time-varying delay to model distributed control and filtering for industrial cyber-physical systems,”
sample-and-hold circuits,” IEEE Trans. Autom. Control, vol. 52, no. 6, IEEE Trans. Ind. Inform., vol. 15, no. 5, pp. 2483–2499, May 2019.
pp. 1109–1112, Jun. 2007. [53] D. Ding, Z. Wang, Q.-L. Han, and G. Wei, “Neural-network-based
[32] P. Naghshtabrizi, J. P. Hespanha, and A. R. Teel, “Exponential output-feedback control under Round-Robin scheduling protocols,”
stability of impulsive systems with application to uncertain sampled- IEEE Trans. Cybern., vol. 49, no. 6, pp. 2372–2384, Jun. 2019.
data systems,” Syst. Control Lett., vol. 57, no. 5, pp. 378–385, May [54] J. F. Wang and C. F. Liu, “Stabilization of uncertain systems with
2008. Markovian modes of time delay and quantization density,” IEEE/CAA
[33] E. Fridman, “A refined input delay approach to sampled-data control,” J. of Autom. Sinica, vol. 5, no. 2, pp. 463–470, Mar. 2018.
Automatica, vol. 46, no. 2, pp. 421–427, Feb. 2010. [55] H. J. Gao and T. W. Chen, “A new approach to quantized feedback
[34] K. Liu and E. Fridman, “Wirtinger's inequality and Lyapunov-based control systems,” Automatica, vol. 44, no. 2, pp. 534–542, Feb. 2008.
sampled-data stabilization,” Automatica, vol. 48, no. 1, pp. 102–108, [56] L. X. Zhang, Z. P. Ning, and W. X. Zheng, “Observer-based control
Jan. 2012. for piecewise-affine systems with both input and output quantization,”
[35] A. Seuret, “A novel stability analysis of linear systems under IEEE Trans. Autom. Control, vol. 62, no. 11, pp. 5858–5865, Nov.
asynchronous samplings,” Automatica, vol. 48, no. 1, pp. 177–182, Jan. 2017.
2012. [57] G. Feng, “Stability analysis of piecewise discrete-time linear systems,”
[36] H.-B. Zeng, K. L. Teo, and Y. He, “A new looped-functional for IEEE Trans. Autom. Control, vol. 47, no. 7, pp. 1108–1112, Jul. 2002.
stability analysis of sampled-data systems,” Automatica, vol. 82, [58] Z. D. Wang, H. L. Dong, B. Shen, and H. J. Gao, “Finite-horizon H∞
pp. 328–331, Aug. 2017. filtering with missing measurements and quantization effects,” IEEE
[37] A. Seuret and F. Gouaisbaut, “Hierarchy of LMI conditions for the Trans. Autom. Control, vol. 58, no. 7, pp. 1707–1718, Jul. 2013.
stability analysis of time-delay systems,” Syst. Control Lett., vol. 81, [59] C. B. Wen, Z. D. Wang, Q. Y. Liu, and F. E. Alsaadi, “Recursive
pp. 1–7, Jul. 2015. distributed filtering for a class of state-saturated systems with fading
[38] D. Ding, Z. Wang, D. Ho, and G. Wei, “Distributed recursive filtering measurements and quantization effects,” IEEE Trans. Syst. Man,
for stochastic systems under uniform quantizations and deception Cybern.: syst., vol. 48, no. 6, pp. 930–941, Jun. 2018.
attacks through sensor networks,” Automatica, vol. 78, pp. 231–240, [60] M. Liu, D. W. C. Ho, and Y. G. Niu, “Robust filtering design for
Apr. 2017. stochastic system with mode-dependent output quantization,” IEEE
[39] N. Elia and S. K. Mitter, “Stabilization of linear systems with limited Trans. Signal Process., vol. 58, no. 12, pp. 6410–6416, Dec. 2010.
information,” IEEE Trans. Autom. Control, vol. 46, no. 9, [61] B. Zhou, G. R. Duan, and J. Lam, “On the absolute stability approach
pp. 1384–1400, Sep. 2001. to quantized feedback control,” Automatica, vol. 46, no. 2,
[40] M. Y. Fu and L. H. Xie, “The sector bound approach to quantized pp. 337–346, Feb. 2010.
feedback control,” IEEE Trans. Autom. Control, vol. 50, no. 11, [62] C. L. Wang, C. Y. Wen, Y. Lin, and W. Wang, “Decentralized
pp. 1698–1711, Nov. 2005. adaptive tracking control for a class of interconnected nonlinear
[41] A. J. Rojas and F. Lotero, “Signal-to-noise ratio limited output systems with input quantization,” Automatica, vol. 81, pp. 359–368,
feedback control subject to channel input quantization,” IEEE Trans. Jul. 2017.
Autom. Control, vol. 60, no. 2, pp. 475–479, Feb. 2015. [63] J. Zhou, C. Y. Wen, W. Wang, and F. Yang, “Adaptive backstepping
[42] E. Fridman and M. Dambrine, “Control under quantization, saturation control of nonlinear uncertain systems with quantized states, ” IEEE
and delay: an LMI approach,” Automatica, vol. 45, no. 10, Trans. Autom. Control, to be published. doi: 10.1109/TAC.2019.
pp. 2258–2264, Oct. 2009. 2906931
[43] L. Zou, Z. Wang, Q.-L. Han, and D. Zhou, “Ultimate boundedness [64] F. Y. Wang and D. R. Liu, Networked Control Systems: Theory and
control for networked systems with try-once-discard protocol and Applications. London, U.K.: Springer-Verlag, 2008.
uniform quantization effects,” IEEE Trans. Autom. Control, vol. 62, [65] J. Lunze, Control Theory of Digitally Networked Dynamic Systems.
no. 12, pp. 6582–6588, Dec. 2017. New York, USA: Springer, 2014.
[44] H. Xu, Q. M. Zhao, and S. Jagannathan, “Finite-horizon near-optimal [66] J. Baillieul and P. J. Antsaklis, “Control and communication
output feedback neural network control of quantized nonlinear challenges in networked real-time systems,” Proc. IEEE, vol. 95, no. 1,
discrete-time systems with input constraint,” IEEE Trans. Neural pp. 9–28, Jan. 2007.
Netw. Learn. Syst., vol. 26, no. 8, pp. 1776–1788, Aug. 2015. [67] L. X. Zhang, H. J. Gao, and O. Kaynak, “Network-induced constraints
[45] M. Q. Shen, S. K. Nguang, C. K. Ahn, and Q.-G. Wang, “Robust H2 in networked control systems-a survey,” IEEE Trans. Ind. Inform.,
control of linear systems with mismatched quantization,” IEEE Trans. vol. 9, no. 1, pp. 403–416, Feb. 2013.
Autom. Control, vol. 64, no. 4, pp. 1702–1709, Apr. 2019. [68] L. Q. Zhang, Y. Shi, T. W. Chen, and B. Huang, “A new method for
[46] Y. Meng, T. Li, and J.-F. Zhang, “Coordination over multi-agent stabilization of networked control systems with random delays,” IEEE
networks with unmeasurable states and finite-level quantization,” Trans. Autom. Control, vol. 50, no. 8, pp. 1177–1181, Aug. 2005.
IEEE Trans. Autom. Control, vol. 62, no. 9, pp. 4647–4653, Sep. 2017. [69] A. Ray, “Output feedback control under randomly varying distributed
[47] D. Thanou, E. Kokiopoulou, Y. Pu, and P. Frossard, “Distributed delays,” J. Guid., Control, Dyna., vol. 17, no. 4, pp. 701–711, Jul.-Aug.
average consensus with quantization refinement,” IEEE Trans. Signal 1994.
Process., vol. 61, no. 1, pp. 194–205, Jan. 2013. [70] F. W. Yang, Z. D. Wang, Y. S. Hung, and M. Gani, “H∞ control for
[48] T. F. Liu, Z.-P. Jiang, and D. J. Hill, “Small-gain based output- networked systems with random communication delays,” IEEE Trans.
feedback controller design for a class of nonlinear systems with Autom. Control, vol. 51, no. 3, pp. 511–518, Mar. 2006.
actuator dynamic quantization,” IEEE Trans. Autom. Control, vol. 57, [71] Z. D. Wang, F. W. Yang, D. W. C. Ho, and X. H. Liu, “Robust H∞
no. 5, pp. 1326–1332, May 2012. control for networked systems with random packet losses,” IEEE
[49] T. F. Liu and Z.-P. Jiang, “Event-triggered control of nonlinear Trans. Syst. Man, Cybern. Part B (Cybern.), vol. 37, no. 4,
systems with state quantization,” IEEE Trans. Autom. Control, vol. 64, pp. 916–924, Aug. 2007.
no. 2, pp. 797–803, Feb. 2019. [72] X. He, Z. Wang, and D. Zhou, “Robust H∞ filtering for networked
[50] M. Abdelrahim, V. S. Dolk, and W. P. M. H. Heemels, “Event- systems with multiple state delays,” Int. J. Control, vol. 80, no. 8,
ZHANG et al.: NETWORKED CONTROL SYSTEMS: A SURVEY OF TRENDS AND TECHNIQUES 15

pp. 121711232, Aug. 1217. feedback control for LTI systems, ” in Proc. 51st IEEE Conf. Decision
[73] D. Yue, Q.-L. Han, and J. Lam, “Network-based robust H∞ control of and Control, Maui, HI, USA, 2012, pp. 6597–6602.
systems with uncertainty,” Automatica, vol. 41, no. 6, pp. 999–1007, [95] A. Selivanov and E. Fridman, “Event-triggered H∞ control: a switching
Jun. 2005. approach,” IEEE Trans. Autom. Control, vol. 61, no. 10,
[74] L. Ding, Q.-L. Han, and G. Guo, “Network-based leader-following pp. 3221–3226, Oct. 2016.
consensus for distributed multi-agent systems,” Automatica, vol. 49, [96] X. F. Wang and M. D. Lemmon, “Self-triggered feedback control
no. 7, pp. 2281–2286, Jul. 2013. systems with finite-gain L2 stability,” IEEE Trans. Autom. Control,
[75] J. L. Xiong and J. Lam, “Stabilization of networked control systems vol. 54, no. 3, pp. 452–467, Mar. 2009.
with a logic ZOH,” IEEE Trans. Autom. Control, vol. 54, no. 2, [97] M. Mazo Jr., A. Anta, and P. Tabuada, “An ISS self-triggered
pp. 358–363, Feb. 2009. implementation of linear controllers,” Automatica, vol. 46, no. 8,
[76] J. Lam, H. J. Gao, and C. H. Wang, “Stability analysis for continuous pp. 1310–1314, Aug. 2010.
systems with two additive time-varying delay components,” Syst. [98] D. Zhang, Q.-L. Han, and X. Jia, “Network-based output tracking
Control Lett., vol. 56, no. 1, pp. 16–24, Jan. 2007. control for T-S fuzzy systems using an event-triggered communication
[77] D. Dačić and D. Nešić, “Quadratic stabilization of linear networked scheme,” Fuzzy Sets Syst., vol. 273, pp. 26–48, Aug. 2015.
control systems via simultaneous protocol and controller design,” [99] C. Peng and Q.-L. Han, “On designing a novel self-triggered sampling
Automatica, vol. 43, no. 7, pp. 1145–1155, Jul. 2007. scheme for networked control systems with data losses and
[78] W. P. M. H. Heemels, A. R. Teel, N. van de Wouw, and D. Nesic, communication delays,” IEEE Trans. Ind. Electron., vol. 63, no. 2,
“Networked control systems with communication constraints: tradeoffs pp. 1239–1248, Feb. 2016.
between transmission intervals, delays and performance,” IEEE Trans. [100] D. P. Borgers, R. Postoyan, A. Anta, P. Tabuada, D. Nešić, and W. P.
Autom. Control, vol. 55, no. 8, pp. 1781–1796, Aug. 2010. M. H. Heemels, “Periodic event-triggered control of nonlinear systems
[79] K. Liu, E. Fridman, and L. Hetel, “Networked control systems in the using over approximation techniques,” Automatica, vol. 94, pp. 81–87,
presence of scheduling protocols and communication delays,” SIAM J. Aug. 2018.
Control Optim., vol. 53, no. 4, pp. 1768–1788, Jul. 2015. [101] X. X. Mi and S. Y. Li, “Event-triggered MPC design for distributed
[80] C. S. Draper, Inertial Guidance. Oxford: Pergamon Press, 1960. systems with network communications,” IEEE/CAA J. Autom. Sinica,
vol. 5, no. 1, pp. 240–250, Jan. 2018.
[81] H. Kopetz, “Should responsive systems be event-triggered or time-
triggered?” IEICE Trans. Inform. Syst. Jpn., vol. E76-D, no. 11, [102] M. Ghodrat and H. J. Marquez, “On the local input-output stability of
pp. 1325–1332, 1993. event-triggered control systems,” IEEE Trans. Autom. Control, vol. 64,
no. 1, pp. 174–189, Jan. 2019.
[82] K. J. Åström and B. Bernhardsson, “Comparison of periodic and event
based sampling for first-order stochastic systems,” IFAC Proc. Vol., [103] H. Yu, F. Hao, and T. W. Chen, “A uniform analysis on input-to-state
vol. 32, no. 2, pp. 5006–5011, Jul. 1999. stability of decentralized event-triggered control systems, ” IEEE
Trans. Autom. Control, to be published. doi: 10.1109/TAC.2018.
[83] C. Peng and F. Q. Li, “A survey on recent advances in event-triggered
2879764
communication and control,” Inf. Sci., vol. 457-458, pp. 113–125, Aug.
2018. [104] F. D. Brunner, W. P. M. H. Heemels, and F. Allgöwer, “Event-
triggered and self-triggered control for linear systems based on
[84] X.-M. Zhang, Q.-L. Han, and B.-L. Zhang, “An overview and deep
reachable sets,” Automatica, vol. 101, pp. 15–26, Mar. 2019.
investigation on sampled-data-based event-triggered control and
filtering for networked systems,” IEEE Trans. Ind. Inform., vol. 13, [105] Y. F. Gao, X. M. Sun, C. Y. Wen, and W. Wang, “Event-triggered
no. 1, pp. 4–16, Feb. 2017. control for stochastic nonlinear systems,” Automatica, vol. 95,
pp. 534–538, Sep. 2018.
[85] L. Ding, Q.-L. Han, X. Ge, and X.-M. Zhang, “An overview of recent
advances in event-triggered consensus of multiagent systems,” IEEE [106] W. Chen, D. Ding, X. Ge, Q.-L. Han, and G. Wei, “H∞ containment
Trans. Cybern., vol. 48, no. 4, pp. 1110–1123, Apr. 2018. control of multiagent systems under event-triggered communication
scheduling: the finite-horizon case, ” IEEE Trans. Cybern., to be
[86] X.-M. Zhang and Q.-L. Han, “A decentralized event-triggered
published. doi: 10.1109/TCYB.2018.2885567
dissipative control scheme for systems with multiple sensors to sample
the system outputs,” IEEE Trans. Cybern., vol. 46, no. 12, [107] J. Wang, X.-M. Zhang, Y. Lin, X. Ge, and Q.-L. Han, “Event-triggered
pp. 2745–2757, Dec. 2016. dissipative control for networked stochastic systems under non-
uniform sampling,” Inf. Sci., vol. 447, pp. 216–228, Jun. 2018.
[87] X.-M. Zhang and Q.-L. Han, “Event-triggered dynamic output
feedback control for networked control systems,” IET Control Theory [108] S. L. Hu, D. Yue, X. P. Xie, and Z. P. Du, “Event-triggered H∞
Appl., vol. 8, no. 4, pp. 226–234, Mar. 2014. stabilization for networked stochastic systems with multiplicative
[88] B.-L. Zhang, Q.-L. Han, and X.-M. Zhang, “Event-triggered H∞ noise and network-induced delays,” Inf. Sci., vol. 299, pp. 178–197,
reliable control for offshore structures in network environments,” J. Apr. 2015.
Sound Vib., vol. 368, pp. 1–21, Apr. 2016. [109] X.-M. Zhang and Q.-L. Han, “Event-based H∞ filtering for sampled-
[89] B.-L. Zhang, Q.-L. Han, and X.-M. Zhang, “Recent advances in data systems,” Automatica, vol. 51, pp. 55–69, Jan. 2015.
vibration control of offshore platforms,” Nonlinear Dyn., vol. 89, no. 2, [110] X. Ge, Q.-L. Han, and Z. Wang, “A threshold-parameter-dependent
pp. 755–771, Jul. 2017. approach to designing distributed event-triggered H∞ consensus filters
[90] J. Wang, X.-M. Zhang, and Q.-L. Han, “Event-triggered generalized over sensor networks,” IEEE Trans. Cybern., vol. 49, no. 4,
dissipativity filtering for neural networks with time-varying delays,” pp. 1148–1159, Apr. 2019.
IEEE Trans. Neural Netw. Learn. Syst., vol. 27, no. 1, pp. 77–88, Jan. [111] X. Ge and Q.-L. Han, “Distributed formation control of networked
2016. multi-agent systems using a dynamic event-triggered communication
[91] L. Ding, Q.-L. Han, and X.-M. Zhang, “Distributed secondary control mechanism,” IEEE Trans. Ind. Electron., vol. 64, no. 10,
for active power sharing and frequency regulation in islanded pp. 8118–8127, Oct. 2017.
microgrids using an event-triggered communication mechanism,” [112] F. Pasqualetti, F. Dörfler, and F. Bullo, “Attack detection and
IEEE Trans. Ind. Inform., vol. 15, no. 7, pp. 3910–3922, Jul. 2019. identification in cyber-physical systems,” IEEE Trans. Autom.
[92] A. Girard, “Dynamic triggering mechanisms for event-triggered Control, vol. 58, no. 11, pp. 2715–2729, Nov. 2013.
control,” IEEE Trans. Autom. Control, vol. 60, no. 7, pp. 1992–1997, [113] C. De Persis and P. Tesi, “Resilient control under denial-of-service,”
Jul. 2015. IFAC Proc. Vol., vol. 47, no. 3, pp. 134–139, Aug. 2014.
[93] X. Ge, Q.-L. Han, and Z. Wang, “A dynamic event-triggered [114] C. Peng, J. C. Li, and M. R. Fei, “Resilient event-triggering H∞ load
transmission scheme for distributed set-membership estimation over frequency control for multi-area power systems with energy-limited
wireless sensor networks,” IEEE Trans. Cybern., vol. 49, no. 1, DoS attacks,” IEEE Trans. Power Syst., vol. 32, no. 5, pp. 4110–4118,
pp. 171–183, Jan. 2019. Sep. 2017.
[94] P. Tallapragada and N. Chopra, “Event-triggered dynamic output [115] H. S. Foroush and S. Martínez, “On triggering control of single-input
16 IEEE/CAA JOURNAL OF AUTOMATICA SINICA, VOL. 7, NO. 1, JANUARY 2020

linear systems under pulse-width modulated DoS signals,” SIAM J. stealthy attacks in control systems, ” in Proc. 50th Annu. Allerton
Control Optim., vol. 54, no. 6, pp. 3084–3105, Nov. 2016. Conf. Communication, Control, and Computing, Monticello, IL, USA,
[116] Y. Yuan, H. H. Yuan, L. Guo, H. J. Yang, and S. L. Sun, “Resilient 2012, pp. 1806–1813.
control of networked control system under DoS attacks: a unified game [136] C. Kwon, W. Y. Liu, and I. Hwang, “Security analysis for cyber-
approach,” IEEE Trans. Ind. Inform., vol. 12, no. 5, pp. 1786–1794, physical systems against stealthy deception attacks, ” in Proc. 2013
Oct. 2016. Amer. Control Conf., Washington, DC, USA, 2013, pp. 3344–3349.
[117] S. Hu, D. Yue, Q.-L. Han, X. Xie, X. Chen, and C. Dou, “Observer- [137] Y. L. Mo and B. Sinopoli, “On the performance degradation of cyber-
based event-triggered control for networked linear systems subject to physical systems under stealthy integrity attacks,” IEEE Trans. Autom.
denial-of-service attacks, ” IEEE Trans. Cybern., to be published. doi: Control, vol. 61, no. 9, pp. 2618–2624, Sep. 2016.
10.1109/TCYB.2019.2903817 [138] V. Do, L. Fillatre, and I. Nikiforov, “A statistical method for detecting
[118] H. Foroush and S. Martínez, “On event-triggered control of linear cyber/physical attacks on SCADA systems, ” in Proc. 2014 IEEE
systems under periodic denial-of-service jamming attacks, ” in Proc. Conf. Control Applications, Juan Les Antibes, France, pp. 364-369.
51st IEEE Conf. Decision and Control, Maui, Hawaii, USA, 2012, pp. [139] Y.-L. Wang, Q.-L. Han, M.-R. Fei, and C. Peng, “Network-based T-S
2551-2556. fuzzy dynamic positioning controller design for unmanned marine
[119] W. Zhang, M. S. Branicky, and S. M. Phillips, “Stability of networked vehicles,” IEEE Trans. Cybern., vol. 48, no. 9, pp. 2750–2763, Sep.
control systems,” IEEE Control Syst. Mag., vol. 21, no. 1, pp. 84–99, 2018.
Feb. 2001. [140] L. Zou, Z. Wang, Q.-L. Han, and D. Zhou, “Recursive filtering for
[120] L. Schenato, B. Sinopoli, M. Franceschetti, K. Poolla, and S. S. Sastry, time-varying systems with random access protocol,” IEEE Trans.
“Foundations of control and estimation over lossy networks,” Proc. Autom. Control, vol. 64, no. 2, pp. 720–727, Feb. 2019.
IEEE, vol. 95, no. 1, pp. 163–187, Jan. 2007. [141] Z. Zuo, Q.-L. Han, B. Ning, X. Ge, and X.-M., Zhang, “An overview
[121] V. Gupta, N. C. Martins, and J. S. Baras, “Optimal output feedback of recent advances in fixed-time cooperative control of multiagent
control using two remote sensors over erasure channels,” IEEE Trans. systems,” IEEE Trans. Ind. Inform., vol. 14, no. 6, pp. 2322–2334, Jun.
Autom. Control, vol. 54, no. 7, pp. 1463–1476, Jul. 2009. 2018.
[122] K. Okano and H. Ishii, “Stabilization of uncertain systems with finite [142] B. Ning and Q.-L. Han, “Prescribed finite-time consensus tracking for
data rates and Markovian packet losses,” IEEE Trans. Control Netw. multiagent systems with nonholonomic chained-form dynamics,”
Syst., vol. 1, no. 4, pp. 298–307, Dec. 2014. IEEE Trans. Autom. Control, vol. 64, no. 4, pp. 1686–1693, Apr. 2019.
[123] A. Cetinkaya, H. Ishii, and T. Hayakawa, “Networked control under [143] B. Ning, Q.-L. Han, and Z. Zuo, “Practical fixed-time consensus for
random and malicious packet losses,” IEEE Trans. Autom. Control, integrator-type multi-agent systems: a time base generator approach,”
vol. 62, no. 5, pp. 2434–2449, May 2017. Automatica, vol. 105, pp. 406–414, Jul. 2019.
[124] Y. L. Mo and B. Sinopoli, “Secure estimation in the presence of
integrity attacks,” IEEE Trans. Autom. Control, vol. 60, no. 4,
Xian-Ming Zhang (M’16-SM’18) received the
pp. 1145–1151, Apr. 2015. M.Sc. degree in applied mathematics and the Ph.D.
[125] D. Ding, Z. Wang, Q.-L. Han, and G. Wei, “Security control for degree in control theory and engineering from Cent-
discrete-time stochastic nonlinear systems subject to deception ral South University, Changsha, China, in 1992 and
attacks,” IEEE Trans. Syst., Man, Cybern.: Syst., vol. 48, no. 5, 2006, respectively.
pp. 779–789, May 2018. In 1992, he joined Central South University,
[126] S. Xiao, Q.-L. Han, X. Ge, and Y. Zhang, “Secure distributed finite- where he was an Associate Professor with the School
time filtering for positive systems over sensor networks under of Mathematics and Statistics. From 2007 to 2014, he
deception attacks, ” IEEE Trans. Cybern., to be published. doi: was a Post-Doctoral Research Fellow and a Lecturer
10.1109/TCYB.2019.2900478 with the School of Engineering and Technology,
Central Queensland University, Rockhampton, QLD, Australia. From 2014 to
[127] Y. L. Mo, E. Garone, A. Casavola, and B. Sinopoli, “False data 2016, he was a Lecturer with the Griffith School of Engineering, Griffith Uni-
injection attacks against state estimation in wireless sensor networks, ” versity, Gold Coast, QLD, Australia. In 2016, he joined the Swinburne Uni-
in Proc. 49th IEEE Conf. Decision and Control, Atlanta, GA, USA, versity of Technology, Melbourne, VIC, Australia, where he is currently a
2010, pp. 5967–5972. Senior Lecturer with the School of Software and Electrical Engineering. His
[128] Y. Guan and X. Ge, “Distributed attack detection and secure current research interests include H-infinity filtering, event-triggered control
estimation of networked cyber-physical systems against false data systems, networked control systems, neural networks, distributed systems,
injection attacks and jamming attacks,” IEEE Trans. Signal Inf. and time-delay systems.
Process. Netw., vol. 4, no. 1, pp. 48–59, Mar. 2018. Dr. Zhang was a recipient of second National Natural Science Award in
China in 2013, and first Hunan Provincial Natural Science Award in Hunan
[129] M. H. Zhu and S. Martínez, “On the performance analysis of resilient
Province in China in 2011, both jointly with Prof. M. Wu and Prof. Y. He,
networked control systems under replay attacks,” IEEE Trans. Autom.
and the IET Premium Award in 2016, jointly with Prof. Q.-L. Han. He is an
Control, vol. 59, no. 3, pp. 804–808, Mar. 2014.
Associate Editor of the Journal of the Franklin Institute, International Journ-
[130] Y. Shoukry and P. Tabuada, “Event-triggered state observers for al of Control, Automation, and Systems and the Neurocomputing, and he is an
sparse sensor noise/attacks,” IEEE Trans. Autom. Control, vol. 61, Editorial Board Member of Neural Computing and Applications.
no. 8, pp. 2079–2091, Aug. 2016.
[131] M. Pajic, J. Weimer, N. Bezzo, O. Sokolsky, G. J. Pappas, and I. Lee,
“Design and implementation of attack-resilient cyberphysical systems: Qing-Long Han (M’09–SM’13-F’19) received the
B.Sc. degree in Mathematics from Shandong Nor-
with a focus on attack-resilient state estimators,” IEEE Control Syst.
mal University, Jinan, China, in 1983, and the M.Sc.
Mag., vol. 37, no. 2, pp. 66–81, Apr. 2017.
and Ph.D. degrees in Control Engineering and Elec-
[132] L. Ma, Z. Wang, Q.-L. Han, and H.-K. Lam, “Variance-constrained trical Engineering from East China University of Sci-
distributed filtering for time-varying systems with multiplicative ence and Technology, Shanghai, China, in 1992 and
noises and deception attacks over sensor networks,” IEEE Sens. J., 1997, respectively.
vol. 17, no. 7, pp. 2279–2288, Apr. 2017. From September 1997 to December 1998, he was
[133] Y. Nakahira and Y. L. Mo, “Attack-resilient H2, H∞, and ?1 state a Post-doctoral Researcher Fellow with the Labor-
estimator” IEEE Trans. Autom. Control, vol. 63, no. 12, pp. 4353–4360, atoire d’Automatique et d’Informatique Industielle
Dec. 2018. (currently, Laboratoire d’Informatique et d’Automatique pour les Systémes),
École Supérieure d’Ing’enieurs de Poitiers (currently, École Nationale
[134] X. Ge, Q.-L. Han, X.-M. Zhang, L. Ding, and F. Yang, “Distributed Supérieure d’Ingénieurs de Poitiers), Université de Poitiers, France. From
event-triggered estimation over sensor networks: a survey, ” IEEE January 1999 to August 2001, he was a Research Assistant Professor with the
Trans. Cybern., to be published. doi: 10.1109/TCYB.2019.2917179 Department of Mechanical and Industrial Engineering at Southern Illinois
[135] A. Teixeira, I. Shames, H. Sandberg, and K. Johansson, “Revealing University at Edwardsville, USA. From September 2001 to December 2014,
ZHANG et al.: NETWORKED CONTROL SYSTEMS: A SURVEY OF TRENDS AND TECHNIQUES 17

he was Laureate Professor, an Associate Dean (Research and Innovation) with for Neurocomputing. He is also a very active reviewer for many international
the Higher Education Division, and the Founding Director of the Centre for journals.
Intelligent and Networked Systems at Central Queensland University, Aus-
tralia. From December 2014 to May 2016, he was Deputy Dean (Research),
with the Griffith Sciences, and a Professor with the Griffith School of Engin- Lei Ding (M’18) received the B.Sc. degree in auto-
eering, Griffith University, Australia. In May 2016, he joined Swinburne Uni- mation, and the M.Sc. and Ph.D. degrees in control
versity of Technology, Australia, where he is currently Pro Vice-Chancellor theory and control engineering from Dalian Mari-
(Research Quality) and a Distinguished Professor. In March 2010, he was ap- time University, Dalian, China, in 2007, 2009 and
pointed Chang Jiang (Yangtze River) Scholar Chair Professor by Ministry of 2014, respectively. From November 2010 to Novem-
Education, China. His research interests include networked control systems, ber 2012, he was a visiting Ph.D. student sponsored
multi-agent systems, time-delay systems, complex dynamical systems and by China Scholarship Council at the Centre for Intel-
neural networks. ligent and Networked Systems, Central Queensland
Professor Han is one of The World’s Most Influential Scientific Minds: University, Australia. From January 2015 to January
2016, he was a visiting Research Fellow at Western
2014-2016, and 2018. He is a Highly Cited Researcher according to Clarivate
Sydney University, Australia. From May 2016 to January 2017, he was a
Analytics (formerly Thomson Reuters). He is a Fellow of The Institution of
Postdoctoral Associate at Wayne State University, USA. From February 2017
Engineers Australia. He is an Associate Editor of several international journ-
to March 2019, He worked as a Research Fellow at Swinburne University of
als, including the IEEE Transactions on Cybernetics, the IEEE Transactions
Technology, Australia. In March 2019, he joined Nanjing University of Posts
on Industrial Electronics, the IEEE Transactions on Industrial Informatics,
and Telecommunications, where he is currently a Professor with the Institute
IEEE Industrial Electronics Magazine, the IEEE/CAA Journal of Automatica of Advanced Technology. His research interests include event-triggered con-
Sinica, Control Engineering Practice, and Information Sciences. trol and distributed cooperative control of multi-agent systems and their ap-
plications to smart grids.
Xiaohua Ge (M’18) received the B.Eng. degree in
electronics and information engineering from Nan- Dong Yue (SM’08) received the Ph.D. degree in
chang Hangkong University, Nanchang, China, in Control Science and Engineering from the South
2008, the M.Eng. degree in control theory and con- China University of Technology, Guangzhou, China,
trol engineering from Hangzhou Dianzi University, in 1995. He is currently a Changjiang Professor and
Hangzhou, China, in 2011, and the Ph.D. degree in Dean of Institute of Advanced Technology, Nanjing
computer engineering from Central Queensland Uni- University of Posts and Telecommunications. His re-
versity, Rockhampton, QLD, Australia, in 2014. search interests include analysis and synthesis of net-
From 2011 to 2013, he was a Research Assistant worked control systems, multi-agent systems, optim-
with the Centre for Intelligent and Networked Sys- al control of power systems, and internet of things.
tems, Central Queensland University, Rockhampton, QLD, Australia, where He is currently an Associate Editor of the IEEE
he was a Research Fellow in 2014. From 2015 to 2017, he was a Research Transactions on Industrial Informatics, the IEEE Transactions on Neural Net-
Fellow with the Griffith School of Engineering, Griffith University, Gold works and Learning Systems, Journal of The Franklin Institute, International
Coast, QLD, Australia. He is currently a Lecturer with the School of Soft- Journal of Systems Science and the IEEE Control Systems Society Confer-
ware and Electrical Engineering, Swinburne University of Technology, Mel- ence Editorial Board.
bourne, VIC, Australia. His research interests include distributed estimation
over sensor networks, distributed coordination in multi-agent systems, secur-
ity and privacy preserving in cyber-physical systems. Chen Peng (M’13–SM’15) received the Ph.D. de-
gree in control theory and control engineering from
the Chinese University of Mining Technology,
Derui Ding (M’16) received both the B.Sc. degree in Xuzhou, China, in 2002. From 2004 to 2005, he was
Industry Engineering in 2004 and the M.Sc. degree a Research Associate with the University of Hong
in Detection Technology and Automation Equip- Kong, Hong Kong. From 2006 to 2007, he was a
ment in 2007 from Anhui Polytechnic University, Visiting Scholar with the Queensland University of
Wuhu, China, and the Ph.D. degree in Control The- Technology, Brisbane, QLD, Australia. From 2011
ory and Control Engineering in 2014 from Donghua to 2012, he was a Post-Doctoral Research Fellow
University, Shanghai, China. From July 2007 to with the Central Queensland University, Rockhamp-
December 2014, he was a teaching assistant and then ton, QLD, Australia. In 2012, he was appointed as an Eastern Scholar by the
a lecturer in the Department of Mathematics, Anhui Municipal Commission of Education, Shanghai, China, and joined Shanghai
Polytechnic University, Wuhu, China. He is cur- University, Shanghai, where he is currently the Director of Centre of Net-
rently a senior research fellow with the School of Software and Electrical En- worked Control Systems and a Distinguished Professor. In 2018, he was ap-
gineering, Swinburne University of Technology, Melbourne, Australia. From pointed as an Outstanding Academic Leader by the Municipal Commission of
June 2012 to September 2012, he was a research assistant in the Department Science and Technology, Shanghai. His current research interests include net-
of Mechanical Engineering, the University of Hong Kong, Hong Kong. From worked control systems, distributed control systems, security control, and in-
March 2013 to March 2014, he was a visiting scholar in the Department of In- telligent control systems. Prof. Peng was a recipient of the one of the Most
formation Systems and Computing, Brunel University London, UK. His re- Cited Chinese Researchers Award in Computer Science by Elsevier from
search interests include nonlinear stochastic control and filtering, as well as 2014 to 2018. He is an Associate Editor of a number of international journals,
multi-agent systems and sensor networks. He has published around 40 papers including Information Sciences and Transactions of the Institute of Measure-
in refereed international journals. Dr. Ding is serving as an Associate Editor ment and Control.