Information Sciences 380 (2017) 117–131

Contents lists available at ScienceDirect

Information Sciences
journal homepage: www.elsevier.com/locate/ins

Distributed networked control systems: A brief overview

Xiaohua Ge, Fuwen Yang, Qing-Long Han∗
Griffith School of Engineering, Griffith University, Gold Coast Campus, QLD 4222, Australia

a r t i c l e i n f o a b s t r a c t

Article history: Distributed networked control systems have attracted intense attention from both academia
Received 14 January 2015 and industry due to the multidisciplinary nature among the areas of communication net-
Revised 11 July 2015
works, computer science and control. With ever-increasing research trends in these areas, it is
Accepted 18 July 2015
desirable to review recent advances and to identify methodologies for distributed networked
Available online 3 August 2015
control systems. This paper presents a brief overview of such systems regarding system con-
Keywords: figurations, challenging issues and methodologies. First, networked control systems are intro-
Distributed networked control system duced and their prevalent configurations including centralized, decentralized and distributed
Communication network structures are outlined. Second, an emphasis is laid on a number of challenging issues from the
Communication constraint analysis and synthesis of distributed networked control systems. More specifically, these chal-
Network topology lenging issues are identified through three integrated aspects: communication, computation
Time-triggered mechanism and control. Third, different methodologies in the literature for distributed networked control
Event-triggered mechanism systems are reviewed and categorized based on three pairs: undirected and directed graphs,
fixed and time-varying topologies, and time-triggered and event-triggered mechanisms. Fi-
nally, concluding remarks are drawn and some potential research directions are suggested.
© 2015 Elsevier Inc. All rights reserved.

1. Introduction

The world is connected through networks. The most prominent example is probably the Internet which consists of a huge
number of computers and users spread around the world. According to Wikipedia, the definition of the Internet is given as “a
global system of interconnected computer networks that use the standard Internet protocol suite (TCP/IP) to link several billion
devices worldwide”. With the advent of networks, traditional point-to-point control systems are being reshaped and redefined,
which gives birth to networked control systems (NCSs) wherein feedback control loops are closed via communication networks
[24,29,85]. The defining feature of an NCS is that system components, such as sensors, controllers and actuators, are connected
via some network medium, such as control area network (CAN), BACnet, Fieldbus, Ethernet and Internet. Therefore, information,
such as reference inputs, system outputs or control inputs, is exchanged among system components via networks [24]. Compared
with traditional point-to-point control systems, NCSs offer several advantages, such as low installation and maintenance costs,
high reliability, increased system flexibility, easy re-configurability and decreased volume of wiring.
Generally, there are three configurations of NCSs: centralized configuration, decentralized configuration, and distributed con-
figuration.
Centralized configuration. As shown in Fig. 1(a), sensed measurement (reference input, system state or system output) on each
sensor node should be sent to a centralized controller embedding a central processing unit for data fusion. The centralized con-
troller takes advantage of full knowledge of the system. Suitable data fusion methods are necessary to obtain fused data which

∗
Corresponding author. Tel.: +61 7 555 29390; fax: +61 7 555 28065.
E-mail address: q.han@griffith.edu.au (Q.-L. Han).

http://dx.doi.org/10.1016/j.ins.2015.07.047
0020-0255/© 2015 Elsevier Inc. All rights reserved.
118 X. Ge et al. / Information Sciences 380 (2017) 117–131

Fig. 1. Three general configurations of NCSs.

Table 1
References based on different configurations of NCSs.

Configurations References

Centralized configuration [25–30,50,52,54,57,67,68,71,73,75–78,81–83,88–90,93]

Decentralized configuration [1,3,10,43,44,53,62,70]
Distributed configuration [2,8,9,21–23,45,56,65,69,79,80,95,99]

are utilized for computing control executions. These control executions are then transmitted to actuators for actuating the sys-
tem. Even though multiple sensors and actuators may be present, this configuration, in certain situations, can be deemed as an
NCS with a single feedback loop where all measurement is required to be measured by a dedicated sensor node and then to be
sent together in a single packet to a remote controller node. A control execution is finally fed back to a dedicated actuator node.
Since information of the measurement is available at the controller side, the main advantage of this centralized configuration
is that the control performance may be generally optimal. Analysis and synthesis of NCSs by using this configuration have been
well studied and considerable results have been available in the literature, see for example, references in Table 1. The centralized
configuration, however, has several drawbacks: (i) the controller is susceptible to total failure of the central processing unit; (ii)
there is a high cost for the central processing unit collecting data from individual sensor node; (iii) there is an increased computa-
tional burden in the central processing unit; and (iv) due to scalability issue, it has a limited application scope in modern control
systems where a large number of distributed sensing and processing nodes are deployed such as airplanes and manufacturing
plants.
Decentralized configuration. Within this configuration, as shown in Fig. 1(b), each controller node may depend on only local
information that the controller possesses to make its local decision. Consequently, even though the entire system may have an
overall team objective, distributed controller nodes do not share information with neighboring nodes. Analysis and synthesis of
such NCSs may reduce to analyzing and synthesizing NCSs with multiple feedback loops, see for example, references presented
in Table 1. More specifically, each controller locally performs its local computation of a control execution by using measurement
through local communication link and then each control execution is sent back to an actuator node for actuating the system. Com-
pared with the centralized configuration, this decentralized configuration alleviates computation burden and design complexity
because local control executions are computed within local controllers. A major concern of the decentralized configuration lies
in that sub-optimal control performance may be achieved due to individual controller node not having as much information re-
garding the state or output of the system as in the centralized configuration. In this sense, a decentralized configuration may be
more scalable if local control decisions are close to an optimal global control decision. However, the absence of communication
and cooperation between decentralized controllers may lead to deteriorated system performance and limit the application scope
of the decentralized configuration in wireless sensor and actuator networks (WSANs) where a large number of nodes coordinate
their activities through communication networks.
Distributed configuration. An NCS within this configuration is known as a distributed networked control system (DNCS), which
possesses two notable characteristics: (i) information of each subsystem is exchanged among system components (e.g., sensors,
controllers and actuators) using a shared communication network; and (ii) the plant usually consists of a large number of simple
interacting units (e.g., subsystems and agents) that can be physically distributed and interconnected to others to coordinate their
 X. Ge et al. / Information Sciences 380 (2017) 117–131 119

tasks for achieving a desired overall objective. These two characteristics mainly lead to the so-called cooperative control. Within
such configuration, as shown in Fig. 1(c), each controller may be allowed to share its local information with its all underlying
neighboring controllers for gaining additional information about the plant dynamics. Thus, distributed controllers are capable to
coordinate their behavior by transmitting/receiving information to/from other controllers within a certain neighboring area. The
advantages imposed by this configuration include modularity, scalability and robustness.
With the rapid advancement and use of Internet, embedded systems, wireless communication technologies and novel control
strategies, DNCSs have been increasingly more paramount and at the core of critical infrastructures, service systems and indus-
trial control systems, such as distributed transportation networks, distributed electrical power systems and smart grids, remote
surgery, cellular phone networks, industrial and manufacturing systems, biological networks and flocking. Even though some
of the advantages and applications of DNCSs are clearly recognizable from systems and applications aforementioned, there is
still no comprehensive theory in completely understanding the analysis and synthesis of such systems. It becomes clear that any
progress made along this research line would then necessarily expand knowledge about broad theory and wide applications of
DNCSs.
In this paper, we provide a brief overview of DNCSs regarding recent advances, challenges, methodologies and future work.
The rest of the paper is organized as follows. In Section 2, a number of challenging issues in DNCSs are briefly surveyed and
outlined from three aspects: communication, computation and control. In Section 3, control methodologies to deal with some
challenges of DNCSs in the existing literature are classified and reviewed. In Section 4, conclusions are made and potential
research topics are discussed.

2. Challenging issues in DNCSs

The analysis and synthesis of DNCSs require an integration and coordination of communication, computation and control, as
illustrated in Fig. 2. Therefore, DNCSs may pose a number of challenging issues that may not be suitably addressed by traditional
control schemes and/or existing distributed control paradigms. Over the last decade, although system and control researchers
have developed several system science and control engineering methodologies, such as frequency domain methodology, time
domain methodology, state space methodology, robust control methodology and stochastic control methodology; and computer
science and networks researchers have made promising advancements such as in system software, high-speed computing and
new networking techniques, some fundamental work is still needed to address real-world challenges integrating communication,
computation and control in a uniform and systematic way.

2.1. The communication challenging issue

Traditional control systems deal with control issues in which the data communication link between sensors and controllers
(or, between controllers and actuators) is considered to be perfect. As a result, information sensed by a sensor is completely
and exactly transmitted to a controller (or, control input is completely and exactly fed back to an actuator) due to point-to-point
wired connections. This is in contrast to DNCSs because of the existence of communication networks. One of motivations for using
communication networks in DNCSs is that the system components such as sensors and/or controllers are spatially distributed.
Although DNCSs can replace traditionally point-to-point wired control systems while providing a number of advantages, the
insertion of communication networks in feedback control loops makes the analysis and design of DNCSs more complicated [72].
Several issues, as shown in Fig. 3, may be induced. Generally, they are regarded as the main factors that cause system performance
degradation and even prevent stable operation. Note that these issues induced by networks may have different terminologies in
the literature, such as network channel limitations, incomplete information, network-induced imperfections, network-induced
phenomena and network-induced constraints. In this paper, they are denoted as communication constraints induced by networks
and may include, but not limited to:

• network-induced delays;
• data packet dropouts (data losses/packet losses);
• data packet disorder;
• quantization error (data quantization, quantization effect);

Fig. 2. Interplay of communication, computation and control in DNCSs.

120 X. Ge et al. / Information Sciences 380 (2017) 117–131

Fig. 3. A schematic diagram of a DNCS through communication channel i, ∀ i = 1, 2, . . . , N.

• time-varying network topology;

• network channel fading;
• time-varying network throughput.

Network-induced delays. Time delays in a DNCS are essentially composed of three kinds of delays: (i) computational delays in
system components, such as sensors, controllers and actuators, due to a limited processing speed or capacity of digital devices;
(ii) network access delays in network queues due to the fact that data are transmitted through a network in a packet form and a
queued network packet may need to wait some time before being sent out. This is particularly the case when a communication
network is shared by other processes; and (iii) transmission delays in the communication network. This kind of delay may rely
on variable network conditions such as network channel quality and network traffic congestion due to a limited bandwidth of a
communication network. In practice, the computational delays in system components are usually small and may be negligible.
The network access delays in network queues and transmission delays in the communication network are commonly referred to
as network-induced delays. Note that there are some researchers who may only consider the transmission delays when modeling
delays in DNCSs and denote them as network-induced delays since network access delays are sometimes small compared with
the transmission delays. In principle, depending on diverse applications using different communication networks, networked-
induced delays may possess random or deterministic characteristics, constant or time-varying characteristics and small or large
characteristics (based on a prescribed sampling period). Under certain circumstance, communication delays among interacting
components or subsystems may pose heterogeneous coupling characteristics, as shown in [17,51]. It is worth noting that the
vast majority of the existing results investigate the negative effect of network-induced delays as they may lead to deteriorated
system performance or even make the closed-loop system unstable. However, for some systems, such as an offshore platform
[86,91,94,96] and a Duffing-Van der Pol’s oscillator [92], network-induced delays may have the positive effect on system perfor-
mance. For example, in [94], by artificially introducing a proper time delay into the control channel, a novel sliding mode control
scheme for an offshore steel jacket platform is proposed. It is shown through simulation results that the scheme is more effective
in improving the control performance and reducing control force of the offshore platform than some existing delay-free sliding
mode control schemes. In [92], by intentionally inserting a communication network that produces proper network-induced de-
lays in the feedback control loop, a stable and satisfactory control can be ensured for a class of T-S fuzzy systems that cannot
be stabilized by nondelayed output feedback controllers. It should be mentioned that different types of delays require different
control methodologies. Based on different types of network behaviors and configurations, different scenarios in practice and dif-
ferent ways to handle the delay problems, many results on analysis and synthesis of NCSs have been reported, see for example,
[29,55,72,85] and references therein.
Data packet dropouts. Data packet dropouts inevitably occur due to network traffic congestion and system components or
network links failures. Generally, the characteristic of data packet dropouts can be categorized into random and deterministic
types. Random data packet dropouts are commonly modeled as an independent and identically distributed process, such as a
Bernoulli process [75–77,97], and a Markov process [71]. A major concern under this modeling framework is how to establish
the relationship between an allowable probability of packet dropouts or an allowable data transmission rate and stability and/or
system performance. Regarding deterministic data packet dropouts, a fundamental problem is to find a maximum allowable
number of consecutive packet dropouts. Sometimes an excessively large network-induced delay may also be deemed as a packet
dropout. Therefore, modeling data packet dropouts as prolonged delays has also received considerable research attention, see for
example, [19,81,82].
Data packet disorder. The phenomenon of data packet disorder may occur when time stamped data packets transmitted over
networks are arrived at destination nodes of different temporal orders. This situation normally occurs in the presence of network-
induced delays and especially if the sizes of the network-induced delays are larger than a prescribed sampling period/interval.
A commonly used technique in most of studies to deal with disordered data packets is to discard them actively if the latest
data packets are arrived at the destination nodes because these out-of-order packets are outdated, see for example, a logical
zero-order-holder (ZOH) technique [16,87]. This actively discarding technique can be easily implemented in practical DNCSs by
embedding some intelligence at the controllers and/or actuators for comparison, checking and selection [85]. However, a key
problem lying in data packet disorder is how to decide whether or not those disordered data packets contain useful information
of the system for control design. If there is some useful information of the system in disordered data packets, how to use these
information for control design is another critical issue.
Quantization error. Due to the rapid growth of digital computing technology, digital control is widely applied in DNCSs, which
leads to the natural existence of quantization. The control execution on each distributed controller is usually processed by a
 X. Ge et al. / Information Sciences 380 (2017) 117–131 121

quantizer incorporating a discontinuous mapping from a continuous space to a finite set. Therefore, information sensed by a
sensor is required to be encoded by transforming analog signals to digital signals so as to be transmitted over a network before it
is sent to a remote controller. On the other side, control input needs to be decoded before it is fed into an actuator for actuating
the system. The classical digital control theory shows that quantization resulting from the finite word length may have a negative
effect on the stability and/or performance of the closed-loop control system as quantization may cause limit cycles and chaotic
behavior [60]. Therefore, traditional control methodologies of NCSs should be reevaluated before they are applied to DNCSs with
quantization error. We refer to Part I of the book [72] with regard to quantization over networks.
Time-varying network topology. Topology refers to the layout of various nodes connected with each other. Therefore, network
topology has an impact on the performance of distributed decisions making in DNCSs. In most cases, network topology may vary
with time and possess time-varying characteristics due to node mobility, node failure and link creations according to commu-
nication power and network access capacity. The existing literature on studying the effect of network topology is based on the
following two types: fixed network topology and time-varying network topology.
Network channel fading. In a communication network, especially a wireless network, if the source node (transmitter) or the
destination node (receiver) is moving actively, such as moving machines, vehicles and people; or moving passively, such as
node mobility, wireless channels are prone to multipath propagation loss and subject to shadow and fading. According to various
fading statistics, network channel fading can be roughly classified into fast fading and slow fading, and stationary fading and non-
stationary fading. One of the main problems of DNCSs is to establish the relationship between the loss of information caused by
channel fading and stability and/or control performance.
Time-varying network throughput. Throughput of a communication network denotes the rate of data successfully transmit-
ted over a communication channel and is usually measured in bits per second (bit/s or bps). In a communication network, the
throughput may be affected by various factors, such as limited processing capacity and/or power of network nodes, current net-
work load and network traffic congestion. Generally, the throughput is time-varying. Therefore, how to determine the maximum
data throughput over a communication network is a major concern by system designers and researchers in communication
theory.

2.2. The computation challenging issue

Computation in DNCSs is closely related to communication and control. Computational complexity differs from network
modeling and control algorithms. Generally, computation challenging issue in DNCSs arises from the following five aspects:

• The modeling of communication networks. For example, to address the analysis and synthesis problems of DNCSs, some ex-
isting work is based on an assumption that network-induced constraints, such as delays and data packet dropouts, occur only
through the sensor-to-controller communication link, and the communication network is modeled as a unilateral transmis-
sion link. In fact, it may be the effects of network-induced constraints induced through the controller-to-actuator channel
that affect the control performance as they are usually unknown and difficult to compensate in practice at the controller side.
As a result, a more general modeling framework of communication networks for these bilateral transmission links is required
for DNCSs, which, however, inevitably increases the computation complexity and burden.
• The complexity caused by the system structure, such as the increase of interaction between subsystems, the increase of
subsystem dimension and the increase of system uncertainty.
• The increase of numbers of system components, such as sensors, controllers and actuators.
• The increase of network connectivity. One key feature of DNCSs is that system components are interconnected and commu-
nicated with others. For example, distributed controllers cooperate with neighboring ones and share local control updates to
neighboring controllers to coordinate their control behavior for achieving an overall goal. Therefore, more interconnection
links among nodes over communication networks will result in more computation complexity.
• The complexity of control algorithms. Different ways and frameworks to model various network-induced constraints over
communication networks also call for different control algorithms to address system analysis and control design issues.
Hence, the computation complexity varies with the adoption of control algorithms.

2.3. The control challenging issue

Traditionally, control algorithms are usually designed based on an implicit assumption of unlimited computation resources,
non-delayed sensing and actuation, unlimited bandwidth and perfect communication environments, and control design mainly
focuses on establishing the relationship between control performance and computation complexity. In DNCSs, however, compu-
tation and communication resources are limited and often shared between multiple applications (such as subsystems, agents,
nodes and other processes). Thus, the development of real-time distributed control algorithms in DNCSs should be realized and
reevaluated by integrating communication, computation and control so as to achieve the desired control performance through
local, asynchronous, distributed and cooperative actions. Improving communication architectures and protocols to increase re-
liability seems an option. However, new control algorithms should be developed to deal with integrated computation and com-
munication challenges. It is a multidisciplinary research topic in the areas of communication networks, computer science and
control.
122 X. Ge et al. / Information Sciences 380 (2017) 117–131

3. Methodologies for DNCSs

In this section, methodologies for modeling, analysis and synthesis of DNCSs in the existing literature are briefly surveyed.
More specifically, we summarize and classify methodologies according to the following three pairs:
• Undirected graph versus directed graph.
• Fixed topology versus time-varying topology.
• Time-triggered mechanism versus event-triggered mechanism.

3.1. Notions of graph theory

Graph theory has been widely used to analyze and synthesize DNCSs. A graph is a mathematical structure which consists of
a set of vertices and a set of edges connecting the vertices. In the sequel, we recall some notions of graph theory.
Let V = {1, 2, . . . , N} denote an index set of N nodes (agents), E ⊆ V × V denote an edge set of paired nodes and A = [ai j ] ∈
RN×N denote a weighted adjacency matrix. A weighted graph G = (V, E, A) is utilized to model the interactions among nodes
and represent a fixed topology of nodes.
An edge of G is denoted by (i, j). The adjacency elements aij associated with the edges of the graph are positive, that is,
ai j > 0 ⇔ (i, j) ∈ E. We assume that self-loops are excluded in the graph, that is, aii = 0, i ∈ V. The neighborhood of node i is
denoted by Ni = { j ∈ V : (i, j) ∈ E }. An element of Ni is called a neighbor of node i. A node is called a source node if it has no
neighbors but is a neighbor of another node. A node is called an isolated node if it has no neighbor and it is not a neighbor
 
of any other node. The in-degree and out-degree of node i are denoted by degin i
= Nj=1 ai j and degout
i
= Nj=1 a ji , respectively.
G is called a balanced graph if degin i
= degout
i
, ∀ i ∈ V. Denoting by D = diag{degin 1
, degin
2
, . . . , degin
N
}, the Laplacian matrix of the
directed weighted graph G is defined as L = D − A.
Denote by a directed path the sequence (i1 , i2 ), (i2 , i3 ), . . . , (i p−1 , i p ). G is called a connected graph if for any i, j ∈ V, there
is a directed path from i to j. A directed tree is a graph where every node except the source node has exactly one neighbor. A
spanning tree of G is a directed tree whose node set is V and whose edge set is a subset of E. Therefore, if G is a connected graph,
it is equivalent to a balanced graph and contains a spanning tree.
G is called an undirected graph if A is a symmetric matrix with ai j = a ji . An edge of an undirected graph is denoted by an
unordered pair (i, j) (or (j, i)) and (i, j) ∈ E means that there is a data transmission link between node i and node j. Therefore, if G
is an undirected graph, it must be a balanced graph.
G is called a directed graph if A is a non-symmetric matrix with aij = aji . An edge of a directed graph is denoted by an ordered
pair (i, j) and (i, j) ∈ E if and only if node j can send information to node i directly, or node i can collect information from node j
directly.
Denote by G (t ) = (V, E (t ), A(t )) an interaction topology of nodes at time t, where the edge set E (t ) and the adjacency matrix
A(t ) are dependent on time t. If G (t ) generally varies with time t, it is called a time-varying topology; if G (t ) is stochastically
time-varying and driven by a stochastic process, it is called a randomly switching topology.

3.2. Methodologies based on undirected/directed graph and fixed/time-varying topology

DNCSs are built in a distributed fashion using a large number of controllers. To achieve a common objective of the system,
controllers usually interconnect and communicate with others to share local information of observation and decision, which
leads to cooperative control. Multi-agent systems (MASs) are such systems that use local information gathered from neighboring
agents to control a global behavior of the system. The majority of results in the area of DNCSs is inclined to model them into
MASs. A fundamental problem of MASs is to design a suitable networked control protocol for reaching consensus, that is, all
agents converge to a common point or state value either asymptotically or in finite time. Consensus is a crucial task of cooperative
control for MASs and has a variety of applications in the areas of flocking, distributed sensor networks and especially formation
control. In this section, we will briefly review results focusing on MASs. However, it should be mentioned that many systematic
results in this regard have been reported in the literature. We may concentrate ourselves on some typical results on consensus
and cooperative control which account for network-induced communication constraints, as shown in Table 2.
To begin with, we mention four survey papers [48,49,59,66] on consensus and cooperation in networked MASs, which com-
plement this work. In [48], a theoretical framework for consensus analysis of networked MASs with an emphasis on directed

Table 2
References on distributed consensus/control based on different methodologies.

Methodologies References

Graph Undirected [4,5,8,31–33,35,37,45–47,59,84,95]

Directed [6,7,9,12,14,20,23,34,36,40,41,51,58,63,65,69,74,79,80,97–99]
Topology Fixed [4,6–9,20,23,31,33,35–37,40,41,45,47,51,59,65,69,79,84,95,97–99]
Time-varying [5,12,14,33,34,36,41,45–47,51,58,59,63,74,80,95]
Mechanism Time-triggered [4–6,9,14,20,26,28,29,32,37,40,46,58,65,67,68,71,74,75,77,78,80,84,97]
Event-triggered [10,21–23,27,43–45,52,54,56,69,70,79,83,88,95,99]
 X. Ge et al. / Information Sciences 380 (2017) 117–131 123

graph and robustness to switching network topologies and communication delays was presented. An overview of basic concepts
of consensus and methods to investigate convergence and performance analysis of consensus algorithms were provided. Applica-
tions such as synchronization of coupled oscillators, formation control, fast consensus in small-world networks and gossip-based
algorithms were also outlined. In [49], an overview of some main results and progress in distributed multi-agent coordination
since 2006 was presented. Results were mainly categorized into several research directions such as consensus, formation con-
trol, optimization and estimation. Some promising research directions along with some open problems were also pointed out.
In [59], a brief review of consensus problems on cooperative control for MASs was presented. Theoretical results under both di-
rected and undirected graphs, time-invariant (fixed) and dynamically changing (time-varying) information exchange topologies
were first summarized. Then, coupling communication delays during information exchanged among agents were considered.
In [66], recent research in consensus problems for constrained multi-agent coordination were discussed. By constrained multi-
agent coordination, it was referred to the study of consensus problems when the system dynamics were constrained, such as
saturated/bounded, or with constrained consensus algorithms. Theoretical developments along these two research lines were
classified and reviewed. Some applications of constrained multi-agent coordination, such as constrained formation control, atti-
tude coordination via constrained control and flocking via constrained control, were summarized.
We now devote ourselves to identifying explicit research directions on consensus and cooperation addressing various
network-induced communication constraints. In the case that information exchange among agents suffers from communica-
tion delays, the consensus problems were studied in [9,41,47,51,63,74,79,98] for either fixed or switching interaction topologies.
When agents interact with neighbors through digital communications, it is of significance to design consensus protocols to
account for the effects of quantization which is nonlinear in nature, as shown in [4,6,12,34,37,46]. Considering limited commu-
nication channel capacity, the distributed consensus problem was studied in [32] regarding how many bits of information each
pair of adjacent agents need to exchange to achieve consensus of the whole network at each time step. It is proved in [32] that
only one bit information exchange per transmission suffices for exponential average consensus. To attenuate communication
noises among agents’ measurement, distributed consensus conditions were proposed in [35,36] to account for the effects of
fixed and/or time-varying topologies. In [97], the consensus problem was considered for a system of double integrator agents
which communicate through a network subject to noise, random communication delays and packet dropouts.
Under certain circumstances, it is desirable to construct nonlinear synchronization protocol for nonlinear distributed and
networked systems as shown in [7]. An adaptive synchronization control design method was proposed for distributed systems
with non-identical unknown nonlinear dynamics and for a target to be tracked with unknown nonlinear dynamics. Furthermore,
adaptive dynamic consensus problems were studied in [33], which focused on designing new distributed adaptive dynamic
consensus protocols with time-varying coupling weights between neighboring agents. The idea of adaptive dynamic consen-
sus protocol was further extended in [31] to study a distributed consensus problem for linear MASs subject to heterogeneous
matching uncertainties.
Note that even though some relative consensus protocols aforementioned are reviewed, as much research on the problems
of consensus and cooperative control is ongoing, this work is by no means complete. It should also be pointed out that most
consensus problems in the current literature have been focused on first-order or second-order MASs (systems with single or
double integrator dynamics). In contrast, the problems of cooperative control from the DNCSs perspective with network-induced
constraints have not been properly investigated. The study of consensus and cooperative control problems for systems with more
complicated nonlinear and/or networked dynamics is a promising research topic for the future research.

3.3. Methodologies based on time-triggered/event-triggered mechanisms

A crucial feature of distributed and networked systems is that the physical plant evolves in continuous real time while the
network operates according to digital logic in discrete time. Consequently, new hybrid system modeling and control methodolo-
gies are desirable for the design of DNCSs. Note that one main difficulty in the analysis and design of DNCSs is to deal with the
timing issue which can be classified into two communication mechanisms: time-triggered mechanisms (TTMs) and event-triggered
mechanisms (ETMs).

3.3.1. Time-triggered mechanisms

Time-triggered mechanisms are primarily based on an assumption that information exchange in a control loop is performed
in a time-triggered manner between sensors, controllers and actuators over a communication network. In this sense, data sam-
pling and transmission through the network is carried out at instant times triggered by clocks, as shown in Fig. 4. Therefore, all
control tasks in DNCSs are executed independently of the state evolution of the plant. This type of communication mechanisms
is motivated by the celebrated sampled-data control systems theory. Generally, the TTM is implemented based on three different
sampling frameworks: periodic (constant) sampling, variable (nonuniform, time-varying) sampling and stochastic sampling.
A periodic sampling framework. In this framework, measurement on each sensor node is sampled periodically and syn-
chronously at equidistant instant times. To simplify subsequent analysis, we consider a DNCS consisting of N subsystems (agents).
The subsystem i is described by the linear state-space representation of the continuous-time form
x˙ i (t ) = Ai xi (t ) + Bi ui (t ) + Ei wi (t ), xi (0) = x0i , i ∈ V; t ∈ R+ , (1)
where xi (t ) ∈ Rnx is the state and is assumed to be measurable; ui (t ) ∈ Rnu is the control input; x0i is an initial condition; wi (t ) ∈
Rnw belonging to L2 [0, ∞) is the external disturbance input; Ai , Bi and Ei are known constant parameter matrices with appropriate
124 X. Ge et al. / Information Sciences 380 (2017) 117–131

Fig. 4. A TTM of a DNCS through communication link i ∈ V.

dimensions. The state feedback control law (consensus protocol) is constructed as


ui (t ) = Ki ai j (xi (kh) − x j (kh)), t ∈ [kh, (k + 1)h); k ∈ N; i ∈ V, (2)
i∈Ni

where h > 0 is a constant sampling period and Ki is the controller (consensus protocol) gain matrix to be determined. For the
system (1), its discrete-time counterpart can be given by
xi (k + 1) = Ai xi (k) + Bi ui (k) + Ei wi (k), xi (0) = x0i , i ∈ V; k ∈ N (3)
and the state feedback control law is designed as

ui (k) = Ki ai j (xi (k) − x j (k)), i ∈ V. (4)
i∈Ni

As one of the most traditional sampling technique, periodic sampling has attracted a great deal of attention regarding analysis
and synthesis of sampled-data systems. There are considerable contributions in the field of robust control, H∞ control and filter-
ing, stochastic control, distributed control and filtering to deal with periodic sampling. In the context of MASs, the sampled-data
second-order consensus problems were studied in continuous-time case [5,6,14,20,40,84,97] and in discrete-time case [58].
A variable sampling framework. Even though periodic sampling is convenient for its implementation in practical applications,
an adoption of the traditional periodic sampling framework to model the plant may not capture the inter-sample behavior of the
systems under consideration [18]. In some circumstance, it is natural to lower the sampling rate, for example, when the system
state is approaching to its equilibrium and no external disturbance is acting on the system. As a result, the traditional periodic
sampling framework may increase frequency with which sensors sample, thereby leading to increased energy consumption
while sensors in practice usually have a low battery [18]. It is, therefore, a vital research topic both in theory and in practice
to deal with the effect of variable sampling in the analysis and synthesis of DNCSs. Consider a DNCS described by (1). Within
the variable sampling framework, if one considers a synchronously sampling paradigm, i.e., all sensors synchronously sample
measurement at discrete sampling instants
0 = t0 < t1 < · · · < tk < · · · , k ∈ N
∞
with limk→∞ tk = ∞ and k=0 [tk , tk+1 ) = [t0 , ∞), the state feedback control law (2) can be replaced by

ui (t ) = Ki ai j (xi (tk ) − x j (tk )), t ∈ [tk , tk+1 ); k ∈ N; i ∈ V. (5)
i∈Ni

If one considers an asynchronously sampling paradigm among sensors, the state feedback control law is designed as

ui (t ) = Ki ai j (xi (tki ) − x j (t ˜j )), t ∈ [tki , tki +1 ); k ∈ N; i ∈ V (6)
k j (t )
i∈Ni

where k˜ j (t )  arg min{t − t p | t ≥ t p , p ∈ N} and

j j
p

0 = t0i < t1i < · · · < tki < · · · , k ∈ N; i ∈ V.

Therefore, sensors are equipped with samplers with separate sampling periods. It should be noted that such asynchronously
sampling poses significant difficulties in the controller design of DNCSs, especially when network-induced delays, data packet
dropouts and data packet disorder are taken into consideration.
A stochastic sampling framework. In practice, the sampling period may change not only dynamically but also randomly due
to random sampler malfunction or failures. Concerning the effect of this stochastic sampling, a sampled-data synchronization
 X. Ge et al. / Information Sciences 380 (2017) 117–131 125

control problem was studied in [61] for dynamical networks consisting of a large number of interacted nodes. More specifically,
the dynamical network consisting of N interacting nodes is of the following form [61]

N
x˙ i (t ) = f (xi (t )) + ai j  x j (t ) + ui (t ), xi (0) = x0i , i ∈ V; t ∈ R+ , (7)
j=1

where  = diag{r1 , r2 , . . . , rnx } represents a matrix linking the jth state variable if rj = 0. The coupling configuration matrix

(adjacency matrix) is assumed to be symmetric and satisfies N i=1 ai j = 0 for all i ∈ V. As shown in [61], if we denote by s(t) the
solution to the unforced isolate node s˙ (t ) = f (s(t )) and let the error vector be ei (t ) = xi (t ) − s(t ), the error dynamics of network
(7) can be rewritten as

N
e˙ i (t ) = f˜(xi (t ), ei (t )) + ai j  e j (t ) + ui (t ), i ∈ V; t ∈ R+ , (8)
j=1

where f˜(xi (t ), s(t )) = f (xi (t )) − f (s(t )). We are interested in constructing the following controller:
ui (t ) = Kei (tki ), t ∈ [tki , tki +1 ); k ∈ N; i ∈ V. (9)

To facilitate analysis, denote an artificial time-varying delay dki (t ) on each sampling interval [tki , tki +1 ) as follows:

dki (t ) = t − tki , t ∈ [tki , tki +1 ); k ∈ N; i ∈ V

and a function di (t) over [0, ∞) as di (t ) = dki (t ), t ∈ [tki , tki +1 ); k ∈ N. A set of random variables ϱi (t), i ∈ V are introduced to denote
the sampling periods on each sampler. Without loss of generality, it is assumed that the sampling periods randomly take two
values ha and hM (in this case, samplers possess two different sampling rates. It can be extended to the case of three or more
sampling rates as shown in [15]) satisfying 0 < ha < hM with probability P rob{i (t ) = ha } = ¯ i and P rob{i (t ) = hM } = 1 − ¯ i ,
where 0 ≤ ¯ i ≤ 1 is a known constant. Then, we have that 0 ≤ di (t) < hM and di (t) is time-varying and non-differentiable at
sampling instants, i.e., a sawtooth function. If one partitions the interval [0, hM ) into two subintervals [0, ha ] and (ha , hM ), it is
easily shown that di (t) stochastically changes its values on these two subintervals. To take this effect of stochastic sampling into
account, a series of new random variables α i (t) are introduced [15]

1, 0 ≤ di (t ) ≤ ha
αi (t ) = i∈V
0, ha < di (t ) < hM ,
with occurrence probability given by P rob{αi (t ) = 1} = ᾱi and P rob{αi (t ) = 0} = 1 − ᾱi , where ᾱi = ¯ i + (ha /hM )(1 − ¯ i ). By
defining new functions
 
di (t ), αi (t ) = 1 ha , αi (t ) = 1
βi (t ) = θ (t ) = i ∈ V,
ha , αi (t ) = 0, i di (t ), αi (t ) = 0,
the error dynamics (8) with (9) can be rewritten as [61]

N
e˙ i (t ) = f˜(xi (t ), s(t )) + ai j  e j (t ) + αi (t )Kei (t − βi (t )) + (1 − αi (t ))Kei (t − θi (t )), i ∈ V. (10)
j=1

With regard to variable sampling, an output delay approach [13] is widely utilized. By assuming that the distance between con-
secutive sampling instants does not exceed a given bound, sampled-data systems under discrete measurement are modeled as
the systems with delayed measurement outputs. In the framework of stochastic sampling, the whole sampling period is parti-
tioned into several small intervals and the occurrence probability of the sampling period within each interval is given constants
and satisfies a Bernoulli distribution. It should be mentioned that additional bound information of these small intervals is re-
quired. Then, combining the output delay approach, sampling intervals are converted into bounded time-varying delays and the
resultant closed-loop system is formulated as a system with delays and stochastic parameters, as demonstrated in (10). Subse-
quently, stochastic delay system theory can be applied to design the desired controllers.

3.3.2. Event-triggered mechanisms

In the setting of sampling, only samples of measurement at discrete instants are utilized. An important issue is how to de-
termine appropriate sampling instants. For TTMs, those sampling instants are invoked by a timer or a clock. For example, in
traditional control systems with periodic sampling, sampler usually samples measurement equidistantly and equips a clock to
determine the sampling period. TTMs may facilitate the analysis and design of NCSs due to its simplicity for implementation.
However, TTMs may be not a good choice for resource-constrained control systems as the triggering of samples is indepen-
dent with the state evolution, the computation and communication resources of the system. This gives rise to the study of
event-triggered sampling and control. Note that there are various terminologies in the literature when referring to this concept,
for example, event-based control, event-driven control, send-on-delta control, level-crossing control, dead-band control, state-
triggered control, self-triggered control and event-triggered control. In this study, we prefer to the term “event-triggered” as it
explicitly indicates that the sampling instants are triggered by events.
126 X. Ge et al. / Information Sciences 380 (2017) 117–131

ETMs are deemed as a promising alternative compared to time-triggered ones because they only sample when a specific event
occurs, that is, when “needed”. The motivations of ETMs generally come from the following facts:

• Network bandwidth resource in a DNCS is usually scarce especially when the network medium is a multipurpose shared com-
munication network which connects multiple distributed tasks or is accessed in a mutually exclusive way by other neighbor-
ing processes.
• Energy resource is in essential limited due to the fact that system nodes are usually digital devices powered by batteries, es-
pecially in wireless communication networks. For example, in WSANs, an energy efficiency issue is the most important one
during the design procedure as it is costly to replace or recharge the batteries during the operation of the WSANs. There-
fore, data transmission in WSANs may not occur all the time and it is desirable to determine how frequently or when data
transmission through network medium should be invoked.
• From the control perspective, it is a waste of communication and computation resources to implement consecutive sampling
when there is no need for a corrective feedback signal. For example, when there is a little fluctuation between two consecutive
sampled-data or the system’s state is approaching to its equilibrium and no external disturbance is acting on the system,
it is a waste of scarce resources to broadcast and transmit the current sampled-data to a remote controller to execute the
control task. Thus, communication and computation resources should only be utilized when relevant information needs
to be broadcast and transmitted from a sensor to a controller and/or from a controller to an actuator. In other words, one
important issue in the implementation of DNCSs is to develop efficient communication mechanisms to determine when or
how frequently each sensor should broadcast and transmit the current sampled-data to a remote controller and/or from a
controller to an actuator to reduce the unnecessary occupancy of scarce communication and computation resources.

Even though event-triggered sampling and control have attracted ever-increasing attention from the control and communi-
cation communities in the last decade, the concept of ETMs is not new and can date back to the late 50s in the 20th century
when Ellis (1959) [11] claimed that the most suitable sampling mechanism is by transmitting those data significantly changed
among consecutive samples. There has been a vast literature on event-triggered control and estimation that is available to date.
Reviews made with respect to this subject never cease. We refer the reader especially to two recent survey papers [42] and [38]
and the many references therein to review the state of the art and categorize the methodologies on event-triggered control and
estimation. Generally, existing ETMs can be classified into two frameworks: periodic triggering framework and aperiodic triggering
framework. More specifically, we note that the survey paper [42] mainly focused on the work with regard to aperiodic triggering
mechanisms. Within the setting of aperiodic triggering, two schemes were explicitly categorized and analyzed, that is, event-
triggered schemes and self-triggered schemes. A comparison between these two schemes was also presented. Some concluding
remarks were then given. The survey paper [38], however, presented an overview of recent advances on event-based strate-
gies with emphasis on event-based control and estimation. Differences between event-based control problems and event-based
estimation problems were clarified. Some related directions for the further research work were then pointed out.
In the sequel, the principle of ETMs is outlined. The most common framework to implement an ETM is to feed a new mea-
surement into a controller or a new control action into an actuator whenever a tolerable threshold is surpassed, as shown in
Fig. 5. For example, when a plant is controlled through measurement feedback in order to counteract the impact of disturbances,
an ETM (or called event monitor, event processer, event organizer, event generator) usually situates at the sensor side to check
whether or not the error (or difference) between the current measurement of the plant and the latest transmitted measurement
surpasses a threshold, and further decides when the current measurement should be broadcast or transmitted to a controller
over a resource-constrained network medium. According to different definitions of this error and this threshold, various results
regarding event-triggered control may be obtained.
Consider the plant in Fig. 5 which is described in the continuous-time system of the form (1). For ease of demonstration, we
only demonstrate the ETM located in the uplink, i.e., the link from sensor i to controller i. The ETM, which consists of a sampler, an
event generator and a storer (or a buffer), is introduced to decide when each sensor’s current measurement should be transmitted
to a remote controller. More specifically, the sampler is employed to sample sensed measurement at discrete instants either in

Fig. 5. An ETM of a DNCS through communication link i ∈ V.

 X. Ge et al. / Information Sciences 380 (2017) 117–131 127

a periodic or an aperiodic manner; the event generator is adopted to generate a series of events; and the store is configured
to reserve the latest data packets. From the perspective of DNCSs, we next review some typical references based on different
definitions of the error and the threshold in ETMs.
• If an error is defined as the difference between the current continuous state measurement and its latest transmitted state
measurement, i.e., ei (t ) = xi (t ) − xi (tki ) in an aperiodic triggering framework, and the threshold is a prescribed constant δ i ,
the event-triggered control scheme consisting of an ETM and a state feedback control law is described as

Control law:ui (t ) = Ki ai j (xi (tki ) − x j (tki )), t ∈ [tki , tki +1 ); k ∈ N; i ∈ V
i∈Ni
 
ETM:tki +1 = inf t > tki | ei (t ) > δi , (11)

where tki ∈ R+ denotes the instant of transmitted data in the aperiodic triggering framework. In the periodic triggering frame-
work, however, the instant of transmitted data tki is replaced by tki h with h > 0 the sampling period and tki ∈ N. This control
scheme is referred to as the dead-band control in the literature.
Note that in the framework of aperiodic triggering, an important issue, “Zeno” behavior (the execution time distance be-
tween two consecutive events tki +1 − tki is infinite small), that needs to be considered during the event-triggered control in
continuous-time case should be avoided as this phenomenon drives the system to sample extremely fast. It is shown in [64]
that a minimal inter-event time or a lower bound of the execution time distance between two consecutive events can be
ensured to exist only under some assumption.
• If an error is defined as the difference between the current continuous state measurement and its latest transmitted state
measurement, i.e., ei (t ) = xi (t ) − xi (tki ), and is bounded by the constant threshold δ i and the current state measurement of
xi (t) and/or its neighbors xj (t), the event-triggered control scheme is described as

Control law:ui (t ) = Ki ai j (xi (tki ) − x j (tki )), t ∈ [tki , tki +1 ); k ∈ N; i ∈ V
i∈Ni
 
ETM:tki +1 = inf t > tki | ei (t ) > δi xi (t ) , or


ETM:tki +1 = inf t > tki | ei (t ) > δi ai j (xi (t ) − x j (t )) . (12)
j∈Ni

Under this ETM, a control problem for continuous-time DNCSs subject packet loss and transmission delays was investi-
gated [69]. A distributed event-triggered scheme which can locally predict the maximal allowable number of successive
data dropouts and the state-based deadlines for transmission delays was proposed. Both linear and nonlinear subsystems
were considered. For nonlinear subsystems, local controllers were designed to ensure the input-to-state stability; for linear
subsystems, the design problem was transformed into a linear matrix inequality feasibility problem, and if there were no
transmission delays, the resulting system were guaranteed to be finite-gain Lp stable; if the delays were bounded by given
deadlines, the asymptotic stability of the resulting system were ensured.
Consensus problems for continuous-time MASs with general linear dynamics were studied [95,99]. Applying the variable
substitution method, the consensus was achieved under both fixed and switching topologies [95] and the proposed event-
triggered algorithms excluded “Zeno” phenomenon of the closed-loop systems. Two sufficient conditions with/without con-
tinuous communication among neighboring agents were established to ensure the consensus of agents [99]. In [79], a con-
sensus problem for discrete-time second-order MASs subject to Markovian communication delays was addressed. Employing
the Lyapunov functional method and the Kronecker product technique, the mean square stability of the resulting closed-
loop system was analyzed. Sufficient criteria in terms of linear matrix inequalities for designing desired controllers were
established to ensure the consensus of all agents. In [39], a consensus problem for general discrete-time time-varying MASs
subject to energy-bounded external disturbances was considered. Within the proposed framework, relative measurement
outputs among adjacent agents were utilized in the estimator-type consensus protocol. Then, an H∞ consensus control prob-
lem was posed over a finite horizon. A constrained recursive Riccati difference equation approach was developed to obtain
sufficient conditions which achieved consensus with a disturbance attenuation level over a finite horizon [0, T]. By resorting
to the Moore–Penrose pseudo inverse, the desired controller and estimator parameters were iteratively computed.
• If an error is defined as the difference between the current continuous state measurement and its latest transmitted state
measurement, i.e., ei (t ) = xi (t ) − xi (tki h), and is bounded by c0 + c1 exp( − αt ) [21,22], where c0 , c1 ≥ 0 and α > 0, the event-
triggered control scheme is described as

Control law:ui (t ) = Ki xi (tki h) + Li ai j x j (t ), t ∈ [tki h, tki +1 h); k ∈ N; i ∈ V
i∈Ni
 
ETM:tki +1 h = inf t > tki h | ei (t ) > (c0 + c1 exp( − αt )) . (13)
Under this ETM, a distributed control problem for a DNCS consisting of N interconnected subsystems was addressed [21] and
the distributed control problem was further extended [22] to deal with network-induced constraints such as communication
delays and packet losses. It was shown [21,22] that the ETM in (13) can guarantee a lower bound for the inter-event times.
Thus, the “Zeno” behavior can be excluded.
128 X. Ge et al. / Information Sciences 380 (2017) 117–131

For ETMs (11)–(13), a key feature in common is that a persistent monitoring of the current state measurement xi (t) is a
necessity for the implementation of the event-triggered control schemes. As a result, some extra hardware may need to be
deployed to check and monitor the successive measurement, which, however, may increase the cost in practical applications.
For this reason, some researchers suggest the idea of checking the event-triggering conditions periodically, which leads to a
periodic triggering framework. A notable advantage of this periodic triggering framework is that the low bound of the inter-
event time can be guaranteed by the sampling period h. In other words, the undesirable “Zeno” behavior can be ruled out in
the framework of periodic triggering.
• If an error is defined as the difference between the current sampled state measurement and its latest transmitted state
measurement, i.e., ei (tki h + li h) = xi (tki h + li h) − xi (tki h) in the periodic triggering framework, where li ∈ N, and is bounded
by the constant threshold δ i and the latest transmitted state measurement of xi (tki h) and its neighbors x j (t ˜ h), the event-
j
kj
triggered control scheme is described as

Control law:ui (t ) = Ki ai j (xi (tki h) − x j (t ˜j h)), where
k j (t )
i∈Ni

t ∈ [tki h, tki +1 h); k ∈ N; i ∈ V; k˜ j (t )  arg min{t − t ˜j | t > t ˜j , k˜ ∈ N}

k˜ k k

ETM:tki +1 h = tki h + min li h |  2 ei (tki h + li h) > δi  2 Xi (tki h)

1 1
, where
li ≥1

Xi (tki h) = ai j (xi (tki h) − x j (t ˜j h)) and
kj
j∈Ni

k˜ j  arg min{tki + li − t ˜j | tki + li > t ˜j , k˜ ∈ N}. (14)

k˜ k k

Under this ETM, an average consensus problem for continuous-time first-order MASs over fixed or switching undirected
networks was tackled [45]. The Lyapunov function method was employed by an appropriate choice of event conditions to
design event detectors such that the states of agents converge to their initial average. The event condition was in sampled-
data fashion and the event detector used only information at discrete sampling instants. In [23], a consensus problem for
continuous-time MASs with general linear dynamics over fixed directed networks was investigated. The event triggering
condition was intermittently examined only at constant sampling instants. A delay system method was developed to study
the consensus of distributed multi-agent systems. Employing Lyapunov function method, a sufficient condition for designing
desired controllers was derived. Furthermore, a co-design algorithm for obtaining both the parameters of the ETM and con-
troller was proposed. In [8], a consensus problem for continuous-time first-order MASs was considered within the framework
of aperiodic triggering . The distributed even-triggered control scheme depended on knowledge of the states of neighboring
agents. The main results were then extended to a self-triggered setup in which each agent determined its next update time
at the previous time and had no need to check the state error.

In summary, although a great deal of attention has been paid to the consensus problems of MASs under event-triggered
communications in the literature, the study of the effects of network-induced communication constraints over the consensus
control performance still requires mutual research effort from control and communication communities. Note that most of the
existing results on event-triggered control and estimation have concentrated on preserving practical convergence and stability
(for example in terms of uniform boundedness, asymptotic stability, exponential stability or input-to-state stability) of the system
when implementing ETMs. Nevertheless, an indispensable assumption made in these results is that the resources of network
bandwidth and/or computation energy should be adequate during the design procedures of control and estimation. The only
concern is then to design suitable control laws or estimators to ensure the stability of the original system while saving a certain
level of network bandwidth and/or computation energy. In other words, the communication and computation resources are
assumed to be “enough” before they are “saved” by desired event-triggered control or estimation algorithms. As a result, the
majority of the existing results may not explicitly establish the relationship between system performance and scare resources in
their design procedures. Research on this subject remain challenging and open.

4. Conclusions and potential research directions

4.1. Conclusions

A brief overview of recent developments on DNCSs has been presented. NCSs have been firstly reviewed and three general
configurations of NCSs, i.e., centralized, decentralized and distributed configurations, have been discussed, respectively. Then,
challenging issues from the study and application of DNCSs have been outlined from three aspects: communication, computa-
tion and control. A significant emphasis has been put on categorizing some existing methodologies that have been popularly
applied for studying DNCSs according to three pairs: undirected and directed graphs, fixed and time-varying topologies, and
time-triggered and event-triggered mechanisms. Several recent results on MASs have been reviewed and discussed.
 X. Ge et al. / Information Sciences 380 (2017) 117–131 129

4.2. Potential research directions

Since DNCSs are an ongoing research topic, new results and methodologies are required to meet various requirements in
analysis, design and applications. Some potential research directions are suggested and listed as follows:

• Most of existing results on DNCSs in the literature consider only some of types of network-induced communication con-
straints, while ignoring the others. The comprehensive modeling and study for such communication constraints are not suf-
ficient yet. In practical engineering, there may exist many more complex communication constraints not mentioned in this
work. Therefore, general and unified modeling frameworks for these comprehensive and/or new network-induced commu-
nication constraints remain to be established.
• Due to the integration nature of DNCSs among communication, computation and control, the problem of analysis and synthe-
sis of DNCSs should be reevaluated by developing more integrated and tolerant control algorithms that combine comprehen-
sive studies of communication, computation and control. The study along this research line is still challenging and deserves
deeper investigation.
• The study of WSANs has not been adequately explored, especially from the DNCS perspective. There are several points that
should be carefully considered in the analysis and design procedures, such as energy efficiency of nodes, and multiple inter-
action and coordination among sensors, controllers and actuators.
• Applications of existing theory and methodologies of DNCSs to some practical engineering systems, such as distributed solar
power generation systems and mobile robots, are of significant engineering importance.

Acknowledgment

This work was supported in part by the Australian Research Council Discovery Project under grant DP1096780.

References

[1] F. Abdollahi, K. Khorasani, A decentralized markovian jump H∞ control routing strategy for mobile multi-agent networked systems, IEEE Trans. Control Syst.
Technol. 19 (2) (2011) 269–273.
[2] M. Andreasson, D. Dimarogonas, H. Sandberg, K. Johansson, Distributed control of networked dynamical systems: static feedback, integral action and
consensus, IEEE Trans. Autom. Control 59 (7) (2014) 1750–1764.
[3] N. Bauer, M. Donkers, N. van de Wouw, W. Heemels, Decentralized observer-based control via networked communication, Automatica 49 (2013) 2074–2086.
[4] R. Carli, F. Bullo, S. Zampieri, Quantized average consensus via dynamic coding/decoding schemes, Int. J. Robust Nonlinear Control 20 (2) (2010) 156–175.
[5] L. Cheng, Y. Wang, Z.-G. Hou, M. Tan, Z. Cao, Sampled-data based average consensus of second-order integral multi-agent systems: switching topologies and
communication noises, Automatica 49 (5) (2013) 1458–1464.
[6] W. Chen, X. Li, L. Jiao, Quantized consensus of second-order continuous-time multi-agent systems with a directed topology via sampled data, Automatica
49 (2013) 2236–2242.
[7] A. Das, F. Lewis, Distributed adaptive control for synchronization of unknown nonlinear networked systems, Automatica 46 (2010) 2014–2021.
[8] D. Dimarogonas, E. Frazzoli, K. Johansson, Distributed event-triggered control for multi-agent systems, IEEE Trans. Autom. Control 57 (5) (2012) 1291–1297.
[9] L. Ding, Q.-L. Han, G. Guo, Network-based leader-following consensus for distributed multi-agent systems, Automatica 49 (7) (2013) 2281–2286.
[10] M. Donkers, W. Heemels, Output-based event-triggered control with guaranteed L∞ -gain and improved and decentralized event-triggering, IEEE Trans.
Autom. Control 57 (6) (2012) 1362–1376.
[11] P. Ellis, Extension of phase plane analysis to quantized systems, IRE Trans. Autom. Control 4 (2) (1959) 43–54.
[12] P. Frasca, Continuous-time quantized consensus: convergence of krasovskii solutions, Syst. Control Lett. 61 (2) (2012) 273–278.
[13] E. Fridman, U. Shaked, V. Suplin, Input/output delay approach to robust sampled-data H∞ control, Syst. Control Lett. 54 (3) (2005) 271–282.
[14] Y. Gao, L. Wang, Sampled-data based consensus of continuous-time multi-agent systems with time-varying topology, IEEE Trans. Autom. Control 56 (5)
(2011) 1226–1231.
[15] H. Gao, J. Wu, P. Shi, Robust sampled-data H∞ control with stochastic sampling, Automatica 45 (2009) 1729–1736.
[16] X. Ge, Q.-L. Han, Distributed event-triggered H∞ filtering over sensor networks with communication delays, Inf. Sci. 291 (2015) 128–142.
[17] X. Ge, Q.-L. Han, X. Jiang, Distributed H∞ filtering over sensor networks with heterogeneous markovian coupling intercommunication delays, IET Control
Theory Appl. 9 (1) (2014) 82–90.
[18] X. Ge, Q.-L. Han, X. Jiang, Sampled-data H∞ filtering of Takagi–Sugeno fuzzy systems with interval time-varying delays, J. Frankl. Inst. 351 (5) (2014) 2515–
2542.
[19] X. Ge, Q.-L. Han, X. Jiang, Distributed fault detection over sensor networks with markovian switching topologies, Int. J. Gen. Syst. 43 (3-4) (2014) 305–318.
[20] Z.-H. Guan, Z.-W. Liu, G. Feng, M. Jian, Impulsive consensus algorithms for second-order multi-agent networks with sampled information, Automatica 48
(7) (2012) 1397–1404.
[21] M. Guinaldo, D. Dimarogonas, K. Johansson, J. Sánchez, S. Dormido, Distributed event-based control strategies for interconnected linear systems, IET Control
Theory Appl. 7 (6) (2013) 877–886.
[22] M. Guinaldo, D. Lehmann, J. Sánchez, S. Dormido, K. Johansson, Distributed event-triggered control for non-reliable networks, J. Frankl. Inst. 351 (2014)
5250–5273.
[23] G. Guo, L. Ding, Q.-L. Han, A distributed event-triggered transmission strategy for sampled-data consensus of multi-agent systems, Automatica 50 (5) (2014)
1489–1496.
[24] R. Gupta, M.-Y. Chow, Network control system: overview and research trends, IEEE Trans. Ind. Electron. 57 (7) (2010) 2527–2535.
[25] G. Gu, L. Qiu, Networked control system for multi-input plants based on polar logarithmic quantization, Syst. Control Lett. 69 (2014) 16–22.
[26] G. Hui, H. Zhang, Z. Wu, Y. Wang, Control synthesis problem for networked linear sampled-data control systems with band-limited channels, Inf. Sci. 275
(2014) 385–399.
[27] X.-C. Jia, X.-B. Chi, Q.-L. Han, N.-N. Zheng, Event-triggered fuzzy H∞ control for a class of nonlinear networked control systems using the deviation bounds
of asynchronous normalized membership functions, Inf. Sci. 259 (2014) 100–117.
[28] X. Jiang, Q.-L. Han, On designing fuzzy controllers for a class of nonlinear networked control systems, IEEE Trans. Fuzzy Syst. 16 (4) (2008) 1050–1060.
[29] X. Jiang, Q.-L. Han, S. Liu, A. Xue, A new H∞ stabilization criterion for networked control systems, IEEE Trans. Autom. Control 53 (4) (2008) 1025–1032.
[30] X. Kang, H. Ishii, Coarsest quantization for networked control of uncertain linear systems, Automatica 51 (2014) 1–8.
[31] Z. Li, Z. Duan, F. Lewis, Distributed robust consensus control of multi-agent systems with heterogeneous matching uncertainties, Automatica 50 (2) (2014)
883–889.
[32] T. Li, M. Fu, L. Xie, J.-F. Zhang, Distributed consensus with limited communication data rate, IEEE Trans. Autom. Control 56 (2) (2011) 279–292.
130 X. Ge et al. / Information Sciences 380 (2017) 117–131

[33] Z. Li, X. Liu, W. Ren, L. Xie, Distributed tracking control for linear multiagent systems with a leader of bounded unknown input, IEEE Trans. Autom. Control
58 (2) (2013) 518–523.
[34] T. Li, L. Xie, Distributed consensus over digital networks with limited bandwidth and time-varying topologies, Automatica 47 (2011) 2006–2015.
[35] T. Li, J.-F. Zhang, Mean square average-consensus under measurement noises and fixed topologies: necessary and sufficient conditions, Automatica 45
(2009) 1929–1936.
[36] T. Li, J.-F. Zhang, Consensus conditions of multi-agent systems with time-varying topologies and stochastic communication noises, IEEE Trans. Autom.
Control 55 (9) (2010) 2043–2057.
[37] S. Liu, T. Li, L. Xie, M. Fu, J.-F. Zhang, Continuous-time and sampled-data-based average consensus with logarithmic quantizers, Automatica 49 (2013)
3329–3336.
[38] Q. Liu, Z. Wang, X. He, D. Zhou, A survey of event-based strategies on control and estimation, Syst. Sci. Control Eng.: Open Access J. 2 (2014) 90–97.
[39] Q. Liu, Z. Wang, X. He, D. Zhou, Event-based H∞ consensus control of multi-agent systems with relative output feedback: the finite-horizon case, IEEE Trans.
Autom. Control (2015). http://dx.doi.org/10.1109/TAC.2015.2394872
[40] H. Liu, G. Xie, L. Wang, Necessary and sufficient conditions for solving consensus problems of double-integrator dynamics via sampled control, Int. J. Robust
Nonlinear Control 20 (2010) 1706–1722.
[41] S. Liu, L. Xie, H. Zhang, Distributed consensus for multi-agent systems with delays and noises in transmission channels, Automatica 47 (2011) 920–934.
[42] M. Mahmoud, A. Memon, Aperiodic triggering mechanisms for networked control systems, Inf. Sci. 296 (2015) 282–306.
[43] M. Mazo, M. Gao, Asynchronous decentralized event-triggered control, Automatica 50 (12) (2014) 3197–3203.
[44] M. Mazo, P. Tabuada, Decentralized event-triggered control over wireless sensor/actuator networks, IEEE Trans. Autom. Control 56 (10) (2011) 2456–2461.
[45] X. Meng, T. Chen, Event based agreement protocols for multi-agent networks, Automatica 49 (7) (2013) 2125–2132.
[46] A. Nedic, A. Olshevsky, A. Ozdaglar, J. Tsitsiklis, On distributed averaging algorithms and quantization effects, IEEE Trans. Autom. Control 54 (11) (2009)
2506–2517.
[47] R. Olfati-Saber, R. Murray, Consensus problems in networks of agents with switching topology and time-delays, IEEE Trans. Autom. Control 49 (9) (2004)
1520–1533.
[48] R. Olfati-Saber, J. Fax, R. Murray, Consensus and cooperation in networked multi-agent systems, Proc. IEEE 95 (1) (2007) 215–233.
[49] Y. Cao, W. Yu, W. Ren, G. Chen, An overview of recent progress in the study of distributed multi-agent coordination, IEEE Trans. Ind. Inform. 9 (1) (2013)
427–438.
[50] Z.-H. Pang, G.-P. Liu, D. Zhou, M. Chen, Output tracking control for networked systems: a model-based prediction approach, IEEE Trans. Ind. Electron. 61 (9)
(2014) 4867–4877.
[51] P. Papachristodoulou, A. Jadbabaie, U. Münz, Effects of delay in multi-agent consensus and oscillator synchronization, IEEE Trans. Autom. Control 55 (6)
(2010) 1471–1477.
[52] C. Peng, Q.-L. Han, A novel event-triggered transmission scheme and L2 control co-design for sampled-data control systems, IEEE Trans. Autom. Control 58
(10) (2013) 2620–2626.
[53] C. Peng, Q.-L. Han, D. Yue, Communication-delay-distribution-dependent decentralized control for large-scale systems with IP-based communication net-
works, IEEE Trans. Control Syst. Technol. 21 (3) (2013) 820–830.
[54] C. Peng, Q.-L. Han, D. Yue, To transmit or not to transmit: a discrete event-triggered communication scheme for networked Takagi–Sugeno fuzzy systems,
IEEE Trans. Fuzzy Syst. 21 (1) (2013) 164–170.
[55] C. Peng, D. Yue, Q.-L. Han, Communication and Control for Networked Complex Systems, Springer-Verlag, Berlin Heidelberg, 2015, doi:10.1007/978-3-662-
46813-5.
[56] C. Persis, R. Sailer, F. Wirth, Parsimonious event-triggered distributed control: a zeno free approach, Automatica 49 (2013) 2116–2124.
[57] R. Postoyan, N. van de Wouw, D. Nešić, W. Heemels, Tracking control for nonlinear networked control systems, IEEE Trans. Cybern. 59 (6) (2014) 1539–1554.
[58] J. Qin, H. Gao, A sufficient condition for convergence of sampled-data consensus for double-integrator dynamics with nonuniform and time-varying com-
munication delays, IEEE Trans. Autom. Control 57 (9) (2012) 2417–2422.
[59] W. Ren, R. Beard, E. Atkins, A survey of consensus problems in multi-agent coordination, in: Proceedings of the 2005 American Control Conference, Portland,
OR, USA, June 8-10, 2005, pp. 1859–1864.
[60] H. Shao, Q.-L. Han, Z. Zhang, X. Zhu, Sampling-interval-dependent stability for sampled-data systems with state quantization, Int. J. Robust Nonlinear Control
24 (17) (2014) 2995–3008.
[61] B. Shen, Z. Wang, X. Liu, Sampled-data synchronization control of dynamical networks with stochastic sampling, IEEE Trans. Autom. Control 57 (10) (2012)
2644–2670.
[62] S. Shu, F. Lin, Decentralized control of networked discrete event systems with communication delays, Automatica 50 (2014) 2108–2112.
[63] Y. Sun, L. Wang, Consensus of multi-agent systems in directed networks with nonuiform time-varying delays, IEEE Trans. Autom. Control 54 (79) (2009)
1607–1613.
[64] P. Tabuada, Event-triggered real-time scheduling of stabilizing control tasks, IEEE Trans. Autom. Control 52 (9) (2007) 1680–1685.
[65] Z. Wang, D. Ding, H. Dong, H. Shu, H∞ consensus control for multi-agent systems with missing measurements: the finite-horizon case, Syst. Control Lett. 62
(10) (2013) 827–836.
[66] Q. Wang, H. Gao, F. Alsaadi, T. Hayat, An overview of consensus problems in constrained multi-agent coordination, Syst. Sci. Control Eng.: Open Access J. 2
(1) (2014) 275–284.
[67] Y.-L. Wang, Q.-L. Han, Quantitative analysis and synthesis for networked control systems with non-uniformly distributed packet dropouts and interval
time-varying sampling periods, Int. J. Robust Nonlinear Control 25 (2) (2015) 282–300.
[68] Y.-L. Wang, Q.-L. Han, Modelling and controller design for discrete-time networked control systems with limited channels and data drift, Inf. Sci. 269 (2014)
332–348.
[69] X. Wang, M. Lemmon, Event-triggering in distributed networked control systems, IEEE Trans. Autom. Control 56 (3) (2011) 586–601.
[70] X. Wang, Y. Sun, N. Hovakimyan, Asynchronous task execution in networked control systems using decentralized event-triggering, Syst. Control Lett. 61
(2012) 936–944.
[71] J. Wu, T. Chen, Design of networked control systems with packet dropouts, IEEE Trans. Autom. Control 52 (7) (2007) 1314–1319.
[72] Y. Xia, M. Fu, G.-P. Liu, Analysis and Synthesis of Networked Control Systems, Springer-Verlag, Berlin Heidelberg, 2011.
[73] Y. Xia, W. Xie, B. Liu, X. Wang, Data-driven predictive control for networked control systems, Inf. Sci. 235 (2013) 45–54.
[74] F. Xiao, L. Wang, Asynchronous consensus in continuous-time multi-agent systems with switching topology and time-varying delays, IEEE Trans. Autom.
Control 53 (8) (2008) 1804–1816.
[75] F. Yang, Q.-L. Han, H∞ control for networked systems with multiple packet dropouts, Inf. Sci. 252 (2013) 106–117.
[76] R. Yang, P. Shi, G.-P. Liu, H. Gao, Network-based feedback control for systems with mixed delays based on quantization and dropout compensation, Auto-
matica 47 (2011) 2805–2809.
[77] F. Yang, Z. Wang, D. Ho, M. Gani, Robust H∞ control with missing measurements and time delays, IEEE Trans. Autom. Control 52 (9) (2007) 1666–1672.
[78] F. Yang, Z. Wang, Y. Hung, M. Gani, H∞ control for networked systems with random communication delays, IEEE Trans. Autom. Control 51 (3) (2006) 511–518.
[79] X. Yin, D. Yue, Event-triggered tracking control for heterogeneous multi-agent systems with markov communication delays, J. Frankl. Inst. 350 (2013)
1312–1334.
[80] K. You, Z. Li, L. Xie, Consensus condition for linear multi-agent systems over randomly switching topologies, Automatica 49 (10) (2013) 3125–3132.
[81] D. Yue, Q.-L. Han, J. Lam, Network-based robust H∞ control of systems with uncertainty, Automatica 41 (6) (2005) 999–1007.
[82] D. Yue, Q.-L. Han, C. Peng, State feedback controller design of networked control systems, IEEE Trans. Circuits Syst.–II: Express Briefs 51 (11) (2004) 640–644.
[83] D. Yue, E. Tian, Q.-L. Han, A delay system method for designing event-triggered controllers of networked control systems, IEEE Trans. Autom. Control 58 (2)
(2013) 475–481.
 X. Ge et al. / Information Sciences 380 (2017) 117–131 131

[84] W. Yu, L. Zhou, X. Yu, J. Lü, R. Lu, Consensus in multi-agent systems with second-order dynamics and sampled data, IEEE Trans. Ind. Inform. 9 (4) (2013)
2137–2146.
[85] L. Zhang, H. Gao, O. Kaynak, Network-induced constraints in networked control systems: a survey, IEEE Trans. Ind. Inform. 9 (1) (2013) 403–416.
[86] B.-L. Zhang, Q.-L. Han, Network-based modelling and active control for offshore steel jacket platform with tmd mechanisms, J. Sound Vib. 333 (25) (2014)
6796–6814.
[87] X.-M. Zhang, Q.-L. Han, Network-based H∞ filtering using a logic jumping-like trigger, Automatica 49 (5) (2013) 1428–1435.
[88] X.-M. Zhang, Q.-L. Han, Event-triggered dynamic output feedback control for networked control systems, IET Control Theory Appl. 8 (4) (2014) 226–234.
[89] X.-M. Zhang, Q.-L. Han, On designing network-based H∞ controllers for stochastic systems, Int. J. Robust Nonlinear Control 25 (1) (2015) 52–57.
[90] Y. Zhang, Q.-L. Han, Network-based synchronization of delayed neural networks, IEEE Trans. Circuits Syst.–I: Regul. Pap. 6 (3) (2013) 676–689.
[91] X.-M. Zhang, Q.-L. Han, D. Han, Effects of small time-delays on dynamic output feedback control of offshore steel jacket structures, J. Sound Vib. 330 (2011)
3883–3900.
[92] D. Zhang, Q.-L. Han, X. Jia, Network-based output tracking control for a class of T-S fuzzy systems that can not be stabilized by non-delayed output feedback
controllers, IEEE Trans. Cybern. 45 (8) (2015) 1151–1524.
[93] D. Zhang, Q.-L. Han, X.-C. Jia, Observer-based H∞ output tracking control for networked control systems, Int. J. Robust Nonlinear Control 24 (17) (2014)
2741–2760.
[94] B.-L. Zhang, Q.-L. Han, X.-M. Zhang, X. Yu, Sliding mode control with mixed current and delayed states for offshore steel jacket platforms, IEEE Trans. Control
Syst. Technol. 22 (5) (2014) 1769–1783.
[95] Z. Zhang, F. Hao, L. Zhang, L. Wang, Consensus of linear multi-agent systems via event-triggered control, Int. J. Control 87 (6) (2014) 1243–1251.
[96] B.-L. Zhang, L. Ma, Q.-L. Han, Sliding mode H∞ control for offshore steel jacket platforms subject to nonlinear self-excited wave force and external distur-
bance, Nonlinear Anal.: Real World Appl. 14 (1) (2013) 163–178.
[97] Y. Zhang, Y.-P. Tian, Consensus of data-sampled multi-agent systems with random communication delay and packet loss, IEEE Trans. Autom. Control 55 (4)
(2010) 939–943.
[98] B. Zhou, Z. Lin, Consensus of high-order multi-agent systems with large input and communication delays, Automatica 50 (2) (2014) 452–464.
[99] W. Zhu, Z.-P. Jiang, G. Feng, Event-based consensus of multi-agent systems with general linear models, Automatica 50 (2) (2014) 552–558.