This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2795574, IEEE
Transactions on Industrial Electronics

A New Delay-compensation Scheme for Networked

Control Systems in Controller Area Networks
Hui Zhang, Yang Shi, Junmin Wang, and Hong Chen

Abstract—In this work, we aim to study a new delay- be treated for the performance and robustness of the closed-
compensation algorithm for networked control systems (NCSs) loop NCSs.
which are connected via the controller area network (CAN)
buses. Firstly, we analyze the property of CAN bus and find the Considering the network-induced delay in a feedback loop, a
main sources of CAN-bus-induced delays. The system controlled delay-free open-loop NCS usually becomes a system with de-
through a CAN bus is formulated into the typical framework layed states. In the literature, there are several main approaches
of NCSs. Different from the traditional state feedback or output
feedback control schemes, we propose a feedback control law dealing with the network-induced delays and the systems with
with the aid of a time-domain Smith predictor. By using the delayed states. The first one is to study the networked control
augmentation technique, a closed-loop system with a delayed systems under the discrete-time framework; see [8]–[10] and
state is obtained. Then, the asymptotical stability property of the references therein. Then, the systems with delayed states
the closed-loop system is investigated and a sufficient condition can be augmented into respective delay-free systems. The
with a bilinear matrix inequality (BMI) is derived. The obtained
BMI is converted into a linear matrix inequality (LMI) and the study of the original systems is reduced to study the augmented
feedback gain in the proposed control law is calculated by solving delay-free systems and the well-developed theory on the delay-
the LMI condition. Finally, two application examples are given free systems can be adopted. However, the augmentations
to illustrate the advantages and the effectiveness of the proposed would increase the dimensions of the systems. If the delay
controller design method. is small, the dimension increment may not affect the stability
Index Terms—Networked control system (NCS), controller validation and controller design. However, if the delay is large,
area network (CAN), time-domain Smith predictor, electronic the exponentially-increasing dimension would complicate the
throttle controller design. The second widely-used approach is to
construct Lyapunov functions and analyze the systems with
I. I NTRODUCTION the Lyapunov stability theory [11]–[14]. The second approach
does not require to augment the system. But selecting of an
Controller area network (CAN) was firstly developed for
appropriate Lyapunov function for a delayed system is not an
automotive industry [1]. Due to the superior performance, it
easy task. Here, an appropriate Lyapunov function means that
has been widely applied to many other areas such as man-
it would lead to less conservative stability condition compared
ufacturing, building automation, and aerospace applications.
to the existing ones. In order to obtain a relatively larger upper
When a control system consisting of sensors, actuators, and
bound of delay, several techniques can be seen in the literature.
controllers is connected together via a CAN bus, it becomes
Examples include the delay-partition [15] and the predictions
a networked control system (NCS). During the last decade,
in [16]–[18].
the increasing attention has been paid to NCSs; see [2]–[7]
and the references therein. One of the main reasons is that In this work, we consider uncertain discrete-time systems
the network control has some significant advantages over the over CAN networks. Different from the well-utilized ap-
traditional point-to-point control. However, aa the systems are proaches, we use an alternative approach to deal with the
becoming complex and the numbers of communication nodes CAN-bus-induced delays. A Smith predictor [19] is employed
are substantially increasing, the network-induced delays may to predict the non-delayed states. With the predicted states and
become comparable to the sampling periods and they should the delay information from the sensor nodes to the controller
node, the proposed control law is composed of both states
Hui Zhang’s work was partially supported by National Science Foundation from the plant and from the predictor. If the system model is
of China (NO. 61403252) and partially supported by China Automobile
Industry Innovation and Development Joint Fund (NO. U1664257). accurate without any uncertainty and the initial states of the
Hui Zhang is with the School of Transportation Science and Engineering, plant and the predictor are identical, the effect of the delay can
Beihang University, Beijing 100191, China; He is also with Science and Tech- be completely eliminated. Moreover, the Smith predictor can
nology on Reliability and Environmental Engineering Laboratory, Beihang
University, Beijing 100191, China. (e-mail: [email protected]). be combined with the mentioned second approach. It infers
Yang Shi is with the Department of Mechanical Engineering, University of from a given example that the maximal tolerable delay will
Victoria, Victoria, BC V8W 2Y2 Canada. (e-mail: [email protected]). drastically increase although the stability criterion is relatively
Junmin Wang is with the Department of Mechanical and Aerospace
Engineering, The Ohio State University, Columbus, OH 43210, USA (e-mail: conservative. The original contributions of this work can be
[email protected]). summarized as: 1) we propose a new delay compensation
Hong Chen is with the State Key Laboratory of Automotive Simulation scheme for NCSs with CAN buses; 2) with a relatively
and Control, Jilin University, Changchun 130025, China; She is also with
Department of Control Science and Engineering, Jilin University, Changchun simple Lyapunov function, the obtained result is much less
130025, China (e-mail: [email protected]). conservative.

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2795574, IEEE
Transactions on Industrial Electronics

II. F UNDAMENTALS AND P ROBLEM F ORMULATION simultaneously, the node with the lowest arbitration identifier
A. Fundamentals of CAN (ID) automatically occupies the bus and transmits the message.
However, nodes with a larger arbitration ID (lower priority)
The CAN bus was initially developed by Bosch and of-
must wait until the CAN bus becomes available before trying
ficially released in 1986 at a Society of Automotive Engi-
to transmit and compare the priorities again. Therefore, the
neers (SAE) conference [1]. After that, due to the successful
message from the larger arbitration ID is delayed on the bus. In
applications, Bosch published several versions of the CAN
the literature, there are many published results on the analysis
specifications. The latest one is CAN 2.0 published in 1991
of CAN-bus-induced delays. According to the work in [21],
[20]. Different from other traditional networks such as the
the upper bound of a CAN-bus-induced delay is calculated
universal serial bus (USB), a CAN bus shares the messages
with the following expression:
to the entire network such that the data in every node of
the system is consistent. As shown in Fig 1, according to (i + 2)l
τmax,i = ∑i−1 , (1)
ISO 11898 Standard, the CAN network is simply connected R − j=0 l
ci
with a single line of twisted-pair cable. It is terminated at
the two ends with 120-Ω resistors and the function of the where τmax,i is the upper bound of the delay for the i-th prior-
terminator resistors is to match the characteristic impedance ity CAN message, R denotes the message transmission rate on
of the cable line to prevent signal reflections. In an NCS the CAN bus, l represents the maximal frame length, and cj
communicated with a CAN bus, the nodes with different is the cycle length of the j-th priority CAN message. We can
functions such as sensor nodes, actuator nodes, and controller see from the calculation expression in Eqn. (1) that the upper
node are connected directly to the CAN bus line. bound of delay for the highest priority message is minimal.
Another fact is that the allocation of CAN message priorities
is designed by the corresponding system designer. Generally,
the priorities are proportional to the significance of messages.
Plant From the control system perspective, the controller node is
the most important one compared with the sensor nodes and
the actuator nodes. We make the following assumptions for an
NCS with a CAN bus:
Sensor node Sensor node Actuator Actuator 1) Assumption I: the controller node has the highest
#1 #n1 node #1 node #n2 priority such that the CAN-bus-induced delay from
CAN CAN CAN CAN the controller node to the actuator node is negligible
Transceiver Transceiver Transceiver Transceiver compared with the system sampling period;
2) Assumption II: the CAN-bus-induced delay from the
sensor node to the controller node is not negligible due
CAN High to the limited transmission rate and the relatively lower
priority.
120W CAN Low 120W
With the above two assumptions, an NCS with consideration of
CAN bus line the CAN-bus-induced delay is depicted by Fig. 2. The output
CAN
Transceiver

Controller
node
Actuator Plant Sensor

Fig. 1. Schematic diagram of an NCS with CAN bus.

CAN Network t

Since the CAN bus is a serial communications bus with

two wires, the bus can be only accessed by one node at each Controller
time. Generally, there are periodic and non-periodic messages
transmitted on the bus. Due to the existing of some random
non-periodic messages, there will be cases in which two or Fig. 2. Schematic diagram of an NCS with consideration of the CAN-bus-
more nodes trying to occupy the bus simultaneously. In such induced delay.
cases, the bus access is implemented with a nondestructive
and bit-wise arbitration process. The arbitration is in terms of of the plant is sampled by physical sensors and sent to the
the message priority and the allocation of priority to messages controller via a CAN bus. Due to the limited communication
is defined in the identifier (11 bits for CAN 2.0A and 29 bits capacity, the CAN-bus-induced delay is unavoidable. In Fig.
for CAN 2.0B). The main priority principle is: the lower the 2, we use τ to represent the delay from the sensor to the
binary message identifier number, the higher its priority. When controller, that is, the information to the controller is delayed
multiple nodes try to transmit a message onto the CAN bus τ by steps compared with the time at the plant side. In order to

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2795574, IEEE
Transactions on Industrial Electronics

have the delay information, the CAN nodes are synchronized Suppose that the dynamics of the approximate model is
and the time stamp at the plant side can be added to the
xs,k+1 = As xs,k + Bs uk ,
packet of sensor measurements. At the controller node, the (4)
delay on the sensor-to-controller node can be determined by ys,k = xs,k ,
comparing the time stamps. A controller is designed with where k is the sampling instant, xs,k ∈ Rn denotes the state
the delayed sensor measurements. To obtain a better control vector of the approximate model, uk ∈ Rm represents the
performance, the determined delay can also be employed in output of the controller , and the output ys,k is the state vector.
the controller design. It is noted that, due to the benefit of the The system matrix As and the input matrix Bs are constant
CAN properties, only one channel delay should be considered with appropriate dimensions. Since we discuss the linear time-
for the controller design. invariant system in this work, As can be chosen as A and Bs
can be chosen as B. The output of the controller is
uk = −K(xs,k−τ − xk−τ − xs,k ). (5)
B. Problem formulation
Substituting the control signal into the discrete-time plant and
Suppose that the plant in Fig. 2 is represented by a discrete- augmenting the predictor dynamics, one has:
time state-space model with system uncertainty as:
ξk+1 = Āξk + B̄ξk−τ
xk+1 = (A + ∆A)xk + Buk , = (Ā0 + ∆Ā0 )ξk + B̄ξk−τ (6)
(2)
yk = xk , = (Ā01 + B̄0 KC1 + ĒM F̄ )ξk + B̄0 KC2 ξk−τ ,
where k is the sampling instant, xk ∈ Rn denotes the state where
vector of the system, uk ∈ Rm represents the control input, [ ] [ ]
and the output yk is the state vector. The system matrix A and A BK BK −BK
Ā0 = , B̄ = ,
0 A + BK BK −BK
the input matrix B are constant with appropriate dimensions. [ ] [ ]
In addition, the system uncertainty is bounded and expressed A 0 B
Ā01 = , B̄0 = ,
as: 0 A B
[ ] [ ]
∆A = EM F, (3) C1 = 0 I , C2 = I −I ,
[ ] [ ]
with constant matrices E, F , and norm-bounded matrix |M | < E [ ] xk
∆Ā0 = M F 0 , ξk =
I. It is assumed that the input matrix B is of full-column rank. 0 xs,k
[ ]
Different from the traditional NCS analysis approaches such E [ ]
Ē = , F̄ = F 0 .
as the delay analysis in [14], [22], the delay prediction in 0
[23], [24], and the augmentation approach in [9], [10], [25],
It infers from (5) that, if the system models are accurate
in this work, we propose a time-domain Smith predictor to
and the initial states are identical, the state vector of the
compensate for the CAN-bus-induced delay and benefit the
approximate model should be the same with that of the plant.
feedback controller design. As shown in Fig. 3, an approximate
Then, the control law in (5) reduces into a state-feedback
model is adopted at the controller node. The approximate
control without any delay and the delay effect is eliminated.
model can be the nominal model of the plant or the reduced-
However, due to the existence of the system uncertainties, the
order model of the plant.
predicted state vector cannot be the same with the system
state vector. Then, the CAN-bus-induced delay still affect the
system stability. The objectives of this work arise from two-
fold:
Actuator Plant Sensor 1) analyze the stability of the closed-loop system in (6) and
derive the stability condition;
uk 2) based on the stability condition, develop a method to
determine the value for the gain K.
CAN Network t
III. M AIN RESULTS

Controller
- + In this section, we study the stability of the closed-loop
- - system in (6) and offer theorems to achieve the prescribed
design objectives. To deal with the uncertainty, we introduce
x s ,k
Model t the following widely applied lemma.
Lemma 1. [26] If there exist real matrices Ω̄ = Ω̄T , L̄ and H̄
Fig. 3. NCS with a Smith predictor and CAN-bus-induced delays. which are compatible with each other and M̄ satisfies |M̄ | ≤
I, then, the following condition
Ω̄ + L̄M̄ H̄ + H̄ T M̄ L̄T < 0, (7)

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2795574, IEEE
Transactions on Industrial Electronics

holds if and only if there is a positive scalar ζ existing such satisfied:

 
that   Λ1 ∗ ∗ ∗ ∗
Ω̄ ∗ ∗  
 L̄T −ζI Λ2 Λ5 ∗ ∗ ∗
∗ <0 (8) 


 < 0,
 S1T S2T −Z/τ̄ ∗ ∗  (11)
ζ H̄ 0 −ζI  
Λ̄3 N B̄ 0 Λ6 ∗
is satisfied.  Λ̄4 M B̄ 0 0 Λ7
where
Λ̄3 = N (Ā0 − ∆Ā0 − I),
Theorem 1. Suppose that the feedback gain K is given. The Λ̄4 = M (Ā0 − ∆Ā0 ).
closed-loop system in (6) is asymptotically stable if there exist
matrices P = P T > 0, Q = QT > 0, Z = Z T > 0, S1 , Note that the system uncertainty is involved in the condition in
S2 , M , N , and positive scalar ζ > 0 such that the following (11). In order to eliminate the varying matrix M̄ , we employ
condition holds: Lemma 1 and obtain the condition in (9).
 
Λ̄ ∗ ∗ In Theorem 1, by assuming that the feedback gain K
 L̄T −ζI ∗  < 0, (9) is given, we derive a sufficient condition guaranteeing the
ζ H̄ 0 −ζI stability. Since one objective of this work is to design the
where feedback gain, the following theorem provides the gain design
  approach.
0 0
 0 0  [ ]
 
L̄ =  0 0  , H̄ = 0 0 0 F̄ 0
, Theorem 2. Given scalars λ1 , λ2 , λ3 , and a constant matrix
  0 0 0 0 F̄
 N Ē 0  H, the closed-loop system in (6) is asymptotically stable
0 M Ē if there exist matrices P = P[T > 0, Q = ]QT > 0,
  M11 M12
Λ1 ∗ ∗ ∗ ∗ Z = Z T > 0, S1 , S2 , M = T, N =
[ ] λ M M22
 Λ2 Λ5 ∗ ∗ ∗  1 11
  λ2 M11 N12
Λ̄ = 
 S1T S2T −Z/τ̄ ∗ ∗ ,
 λ3 M11 N22
T , M̄11 , and positive scalar ζ > 0 such
 Λ3 N B̄ 0 Λ6 ∗  that the following condition holds:
Λ4 M B̄ 0 0 Λ7  
Λ̃ ∗ ∗
Λ1 = −P + (τ̂ + 1)Q + S1 + S1T ,  L̄T −ζI ∗  < 0, (12)
Λ2 = −S1T + S2 , ζ H̄ 0 −ζI
Λ3 = N (Ā0 − I), where
Λ4 = M Ā0 , Λ5 = −S2 − S2T − Q,  
Λ1 ∗ ∗ ∗ ∗
Λ6 = (Z − N − N T )/τ̄ ,  Λ2 Λ5 ∗ ∗ ∗
 
 S1T S]2T −Z/τ̄ ∗ ∗
Λ7 = (P − M − M T ),  [ 
Λ̃ = 
 Λ̃3 λ2 I ,
τ̂ = τ̄ − τ .  M̄11 C2 0 Λ6 ∗ 
 [ 3H ]
λ 
 I 
Λ̃4 M̄11 C2 0 0 Λ7
λ1 H
[ ]
Proof: Define two new variables as νk := ξk+1 − ξk λ2 I
Λ̃3 = N (Ā01 − I) + M̄11 C1 ,
and Ψk := [ξkT , ξk−1
T T
, ξk−2 , · · · , ξk−τ̄
T
]T , where ξk is the state λ3 I
vector of the system in (6). With an assumption of τ + 1 ≤ τ̄ , [ ]
I
consider a Lyapunov functional candidate for the system in Λ̃4 = M Ā01 + M̄11 C1 ,
λ1 I
(6) [ ]
I
Vk (Ψk ) :=V1k + V2k + V3k , T B̄0 = .
0
V1k :=ξkT P ξk , The feedback gain K can be calculated as
−1
∑ ∑
k−1
−1
V2k := νiT Zνi , K = M11 M̄11 . (13)
(10)
j=−τ̄ i=k+j
−τ Proof: Suppose that the matrices M and N in Theorem
∑
k−1 ∑ ∑
k−1
V3k := ξiT Qξi + ξiT Qξi , 1 have the structures as
j=−τ̄ +1 i=k+j−1
[ ]
i=k−ηk M11 M12
M= T,
where P = P > 0, Q = Q > 0 and Z = Z T > 0. By
T T λ1 HM11 M22
[ ] (14)
following similar steps in [27], [28], the closed-loop system λ2 M11 N12
N= T.
in (6) is asymptotically stable if the following condition is λ3 HM11 N22

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2795574, IEEE
Transactions on Industrial Electronics

Then, one gets the following equations: not hold, we can employ similar method in [32] to treat the
[ ] problem.
M11 KC1
M B̄0 KC1 = ,
λ1 HM11 KC1
[ ] IV. S IMULATION RESULTS
λ2 M11 KC1 In this section, we use two examples to show the effec-
N B̄0 KC1 = ,
λ3 HM11 KC1
[ ] (15) tiveness and the advantages of the proposed approach. The
M11 KC2 first example is a numerical one which is utilized to illustrate
M B̄0 KC2 = ,
λ1 HM11 KC2 the improvement on the delay compensation. The second
[ ]
λ2 M11 KC2 example is an electronic throttle valve for a vehicle engine.
N B̄0 KC2 = . It is controlled via a CAN bus and a Smith-predictor-based
λ3 HM11 KC2
controller is designed to compensate for the CAN-bus-induced
By defining a new variable M̄11 = M11 K, we can bridge delay.
the conditions in Theorem 1 and Theorem 2. Since M11 is Example 1: Consider the following discrete-time system
non-singular, the feedback gain K can be calculated with the [ ] [ ]
equation in (13). 0.8 0 −0.1 0
xk+1 = xk + uk . (16)
0.05 0.9 −0.2 −0.1
Remark 1. A part of this work was published in a conference
[29]. However, we have made significant and new contribu- Now, we assume that the feedback gain K in the systems (6)
tions from the following aspects. 1) The studied plant in [29] is a unit matrix that is K = I. In order to compare with other
is a linear time-invariant discrete-time system. However, in approaches, the widely used control law is the state-feedback
this work, we extend the plant to an uncertain system which as:
is more general than the one in [29]. 2) Due to the fact that uk = Kxk−τ . (17)
the plants are different, the obtained theorems are different. When the lower bound of delay τ = 2, the maximal tolerable
3) Compared to the iterative algorithm in [29], we propose delay τ̄ calculated from existing work in [30], [33], [34]
a non-iterative algorithm which makes the computation more and the Theorem 1 in this work is employed to compare
effective. In the literature, there are several methods to deal the conservativeness of different approaches. The sufficient
with the stabilization problem for an NCS such as the method conditions for different approaches are solved with MATLAB
in [30], the augmented one in [9], and the iterative method in LMI toolbox and the function ‘feasp’ is adopted.
[29]. Generally, comparing the computational loads of these
methods, the iterative method has the largest computational TABLE I
C ALCULATED UPPER DELAY BOUNDS τ̄ FOR DIFFERENT CASES .
effort and the method in [30] has the lowest one. The proposed
method in this work has less computational load than the one By Theorem 1 in [30] with control law in (17) 7
of the augmentation work in [9]. By Theorem 2 in [33] with control law in (17) 10
By Theorem 1 in [34] with control law in (17) 13
Remark 2. Different from the traditional state-feedback con-
trol or output-feedback control, in this work, a time-domain By Theorem 1 with control law in (5) 2997999
Smith predictor is introduced to predict the states with the
delayed measurements. If there are no system uncertainties or The calculated delay upper bounds τ̂ for different methods
initial state value variations, the predicted state values are the are listed in Table 1. With the same control law in (17), the
same with the system state values though the measurements authors selected a more complicated Lyapunov function and
are subject to CAN-network-induced delays. Then, the system the delay upper bound τ̄ increased from 7 to 13. However,
uses the predicted non-delayed states to construct the feedback we keep the same simple Lyapunov function and almost the
loop and the effect of delays is eliminated with the predictor. same control action, the upper delay bound jumps from 7 to
If the effect of system uncertainties does not exceed the effect 2997999. The increment of the upper bound is significant. We
of delayed measurements, the proposed method in this paper can conclude that the proposed delay compensation scenario
can achieve better results than the traditional state-feedback can lead to a much less conservative result.
control and output-feedback control. In order to verify the stability and compare the time-domain
Remark 3. In Theorem 2, there are some given scalars or trajectories, we assume that the initial state vector of the
matrix such as λ1 , λ2 , λ3 , and H. It is flexible to select the discrete-time system is [1, −1]T . Moreover, to compromise
prescribed constants. More tricks and experiences can be seen with a much smaller maximal tolerable delay from existing
in [28] and [31]. In addition, there is a matrix T in Theorem work, we select the CAN-bus-induced delay τ varying in the
2 as well. When the system matrix B is of full-column rank, range [2, 200]. The unit feedback gain K is kept as the same in
the matrix B̄0 is also of full-column rank. Then, the value of simulations driving by the control laws in (5) and in (17). With
T can be selected as: the simulation results, state trajectories in three different cases
[ ] are plotted. Fig. 4 depicts the state trajectories of the discrete-
(B̄0T B̄0 )−1 B̄0T
T = , time plant subject to the delays in Fig. 5 and the control law in
B̄0⊥
(5) when the Smith predictor in (4) has the same initial value
where B̄0⊥ represents an orthogonal basis for the null space . We can see that the states converge to zeros in less than 35
of B̄0 . If the rank assumption on the input matrix B does steps though the delay is relatively large.

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2795574, IEEE
Transactions on Industrial Electronics

1 1
First state First state
Second state Second state
0.5 0.5
Amplitude

Amplitude
0 0

-0.5 -0.5

-1 -1
0 20 40 60 80 100 0 100 200 300 400
Time in samples (k) Time in samples (k)
Fig. 4. Case I: State trajectories of the plant under the control law in (5) Fig. 6. Case II: State trajectories of the plant under the control law in (5)
when the Smith predictor has the same initial value. when the Smith predictor has zero initial value.

200 1
First state
Second state
150 0.5
Amplitude
delay τ

100 0

50 -0.5

0 -1
0 200 400 600 800 0 100 200 300 400
Time in samples (k) Time in samples (k)

Fig. 5. CAN-bus-induced delays in the simulation runs under both control Fig. 7. Case II: State trajectories of the Smith predictor under the control
laws. law in (5) when the Smith predictor has the zero initial value.

Since the Smith predictor may not always get the initial under the control law in (17). The state trajectories are given
value of the system to be controlled, in this case, its initial in Fig. 8. Compared with the trajectories in Fig. 4 and Fig.
value starts from zero. For this case, Fig. 6 and Fig. 7 show 6, the state convergence rate is much slower and the state
state trajectories of the plant and the Smith predictor under the trajectories under the control law in (17) will converge to zeros
control law in (5) when the Smith predictor has zero initial after more than 500 steps. In addition, compared with Fig.
value, respectively. It can be seen that both the states of the 4, the trajectories in Fig. 8 fluctuate considerably before the
plant and the Smith predictor converge to zeros around 200 convergence. In terms of the above comparison, the controller
steps which are larger than the value in Case I. In addition, with a Smith predictor has shown clear advantages over the
due to the unequal initial values, the states of the plant are other two control strategies.
different from the corresponding states of the Smith predictor Example 2: This example demonstrates the controller
during the first 50 steps. However, after 50 steps, the states of design for an electronic throttle valve in a CAN network
the Smith predictor are identical with the corresponding states environment. It has been reported that the electronic throttle
of the plant. We can claim that the Smith predictor has good valves have been widely used in vehicle engines [35], [36].
performance on the state prediction. One of the main reasons is that the fast response of the actuator
In the third case, the simulation is carried out for the plant would lead to less fuel consumption and exhaust emissions.

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2795574, IEEE
Transactions on Industrial Electronics

Note that the control law in (5) is a proportional-derivative-

1.5 like control since the second state of the system is the
First state difference of the first state. Therefore, we can evaluate the
1 Second state tracking performance of the designed controller. In the first
simulation, the predictor and the discrete-time model have the
0.5 same system matrix A. Fig. 10 and Fig. 11 depict the throttle
Amplitude

angle and angular velocity responses when the reference is a

unit pulse signal with the period of 2sec. Observe from the
0

-0.5 1.5
Throttle angle
-1
1
-1.5
0 200 400 600 800

θ (rad)
Time in samples (k) 0.5
Fig. 8. Case III: State trajectories of the plant under the control law in (17).

0
The schematic diagram of an electronic throttle value can be
shown in Fig. 9. The main components include a DC motor
-0.5
0 2 4 6
q Time (s)
Gear set
Fig. 10. Throttle angle response when the reference is a unit pulse signal
with the period of 2sec.

u figure that the response can quickly follow the unit reference
Throttle plate
Motor
and there is no steady error at all.

80
Fig. 9. Schematic diagram of an electronic throttle valve for vehicle engines. Angular velocity
60
and a throttle plate which are connected via a gear set. The 40
θ˙ (rad/sec)

throttle angle is a critical control variable and a fast response

would benefit the engine combustion and emission control. 20
According to the modeling work and parameter identifica-
tion in [37]–[39], we have the following discrete-time model 0
with the sampling period of 0.02s:
[ ] [ ] -20
0.9958 0.0144 0.1299
A= , B= . (18)
−0.3773 0.5003 11.6547
-40
The throttle angle is controlled by adjusting the voltage u
applied to the DC motor. To construct the feedback control, -60
the throttle angle is transmitted to the controller via the CAN 0 2 4 6
bus. The CAN-bus-induced delay is assumed to be within the Time (s)
range of [0, 0.06]s, that is, τ = 0 and τ̄ = 3. Moreover, the
system uncertainty does not exceed 1% of the system matrix Fig. 11. Angular velocity response when the reference is a unit pulse signal
with the period of 2sec.
A. In this example, the sufficient condition in Theorem 2 is
solved with the command “solvesdp” of YALMIP [40]. When
In the second simulation, the predictor has a different system
λ2 is selected as -0.01 and the other two scalars are zeros, the
matrix from the plant model. Fig. 12 illustrates the throttle
obtained feedback gain is:
[ ] angle responses of two difference cases when the predictor
K = −4.2298 −0.0996 . (19) is subject to uncertainties of 0.01A and −0.01A. We can see

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2795574, IEEE
Transactions on Industrial Electronics

a time-domain Smith predictor to forecast the future states.

1.5 With the prediction, we have proposed a feedback control law
Throttle angle with 0.01A
in which the delayed states of the plant, the delayed state
Throttle angle with -0.01A
of the predictor, and the non-delayed states of the predictor
1 are involved. By using the augmentation technique, a closed-
loop system with the non-delayed states and the delayed
states is obtained. The stability of the closed-loop system is
θ (rad)

0.5 analyzed and the sufficient condition guaranteeing the stability

is derived. Based on the stability condition, a method of
calculating the feedback gain is proposed. In the simulation
section, two examples are given. The first example has shown
0 that the proposed compensation algorithm has a significant
advantage compared to the existing state-feedback control. The
second example has demonstrated that the proposed approach
-0.5 is effective in the electronic throttle control over a CAN bus.
0 2 4 6 In the future research, we will exploit the problem with some
Time (s) other methods such as the sliding mode control [41], [42] and
T-S fuzzy control [43], [44].
Fig. 12. Throttle angle response when the reference is a unit pulse signal
with the period of 2sec and the predictor is subject to uncertainties of ±0.01A
. R EFERENCES
[1] U. Kiencke, S. Dais, and M. Litschel, “Automotive serial controller area
that both responses are fast as well. Though the steady error network,” in SAE Technical Paper 860391. SAE International, 1986.
[2] Y. Tipsuwan and M.-Y. Chow, “Control methodologies in networked
exists, the tracking performance is acceptable. In addition, control systems,” Control Eng. Pract., vol. 11, no. 10, pp. 1099–1111,
the predicted throttle angles for two different cases are given 2003.
in Fig. 13. Based on the performance of two simulation [3] L. Qiu, Y. Shi, J. Pan, and G. Xu, “Networked H∞ controller design for
a direct-drive linear motion control system,” IEEE Trans. Ind. Electron.,
cases, we can claim that the designed controller satisfies vol. 63, no. 10, pp. 6281–6291, 2016.
the prescribed requirements and the proposed compensation [4] L. Zhang, H. Gao, and O. Kaynak, “Network-induced constraints in
approach is effective. networked control systems – a survey,” IEEE Trans. Ind. Informat.,
vol. 9, no. 1, pp. 403–416, Feb 2013.
[5] H. Lin, H. Su, Z. Shu, Z. G. Wu, and Y. Xu, “Optimal estimation in
udp-like networked control systems with intermittent inputs: Stability
2.5 analysis and suboptimal filter design,” IEEE Trans. Autom. Contr.,
Predicted throttle angle with 0.01A vol. 61, no. 7, pp. 1794–1809, July 2016.
Predicted throttle angle with -0.01A [6] J. Qiu, H. Gao, and S. X. Ding, “Recent advances on fuzzy-model-
2 based nonlinear networked control systems: A survey,” IEEE Trans. Ind.
Electron., vol. 63, no. 2, pp. 1207–1217, 2016.
[7] D. Zhang, P. Shi, Q.-G. Wang, and L. Yu., “Analysis and synthesis of
1.5 networked control systems: A survey of recent advances and challenges,”
ISA Transactions, vol. 66, no. 1, pp. 376–392, 2017.
θ (rad)

[8] L. Xiao, A. H. Jonathan, and P. How, “Control with random communi-

1 cation delays via a discrete-time jump system approach,” in Proc. 2000
Amer. Contr. Conf., Chicago, Illinois, 2000, pp. 2199–2204.
[9] L. Zhang, Y. Shi, T. Chen, and B. Huang, “A new method for stabiliza-
0.5 tion of networked control systems with random delays,” IEEE Trans.
Autom. Contr., vol. 50, no. 8, pp. 1177–1181, 2005.
[10] Y. Shi and B. Yu, “Output feedback stabilization of networked control
0 systems with random delays modeled by Markov chains,” IEEE Trans.
Autom. Contr., vol. 54, no. 7, pp. 1668–1674, 2009.
[11] D. Yue, Q.-L. Han, and C. Peng, “State feedback controller design
-0.5 of networked control systems,” IEEE Trans. Circuits Syst. II, vol. 51,
0 2 4 6 no. 11, pp. 640–644, 2004.
[12] D. Yue, Q.-L. Han, and J. Lam, “Network-based robust H∞ control
Time (s) of systemswith uncertainty,” Automatica, vol. 41, no. 5, pp. 999–1007,
2005.
Fig. 13. Predicted throttle angle when the reference is a unit pulse signal with [13] X. Jiang, Q.-L. Han, S. Liu, and A. Xue, “A new H∞ stabilization
the period of 2sec and the predictor is subject to uncertainties of ±0.01A. criteria for networked control systems,” IEEE Trans. Autom. Contr.,
vol. 53, no. 4, pp. 1025–1032, 2008.
[14] H. Gao, X. Meng, and T. Chen, “Stabilization of networked control
systems with a new delay characterization,” IEEE Trans. Autom. Contr.,
V. C ONCLUSIONS vol. 53, no. 9, pp. 2142–2148, 2008.
[15] X. Meng, J. Lam, B. Du, and H. Gao, “A delay-partitioning approach
In this work, we have investigated a new delay compen- to the stability analysis of discrete-time systems,” Automatica, vol. 46,
sation algorithm for NCSs which are in the CAN network no. 3, pp. 610–614, 2010.
[16] R. Yang, G. ping Liu, P. Shi, C. Thomas, and M. Basin, “Predictive
environment. The system plant is subject to uncertainties. output feedback control for networked control systems,” IEEE Trans.
Different from some well-used approaches, we have employed Ind. Electron., vol. 61, no. 1, pp. 512–520, 2014.

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2795574, IEEE
Transactions on Industrial Electronics

[17] R. Wang, B. Wang, G.-P. Liu, W. Wang, and D. Rees, “H∞ controller [40] J. Lofberg, “YALMIP : a toolbox for modeling and optimization in
design for networked predictive control systems based on the average MATLAB,” in 2004 IEEE International Conference on Robotics and
dwell-time approach,” IEEE Trans. Circuits Syst. II, vol. 57, no. 4, pp. Automation, 2004, pp. 284–289.
310–314, 2010. [41] Y. Wang, H. Shen, H. R. Karimi, and D. Duan, “Dissipativity-based
[18] A. Gonzalez, P. Garcia, P. Albertos, P. Castillo, and R. Lozano, “Ro- fuzzy integral sliding mode control of continuous-time T-S fuzzy sys-
bustness of a discrete-time predictor-based controller for time-varying tems,” IEEE Trans. Fuzzy Syst., accepted and in press, 2017, DOI:
measurement delay,” Control Eng. Pract., vol. 20, no. 2, pp. 102–110, 10.1109/TFUZZ.2017.2710952.
2012. [42] Y. Wang, Y. Xia, H. Shen, and P. Zhou, “SMC design for robust stabiliza-
[19] O. J. M. Smith, “A controller to overcome dead-time,” ISA Transactions, tion of nonlinear Markovian jump singular systems,” IEEE Trans. Autom.
vol. 6, no. 2, pp. 28–33, 1959. Contr., accepted and in press, 2017, DOI: 10.1109/TAC.2017.2720970.
[20] R. Bosch., “CAN specification, version 2.0.” Germany Stuttgart,, 1991. [43] Y. Wei, J. Qiu, and H. R. Karimi, “Reliable output feedback control
[21] T. Herpel, K.-S. Hielscher, U. Klehmet, and R. German, “Stochastic of discrete-time fuzzy affine systems with actuator faults,” IEEE Trans.
and deterministic performance evaluation of automotive can communi- Circuits Syst. I, Reg. Paper, vol. 64, no. 1, pp. 170–181, 2017.
cation,” Comput. Netw., vol. 53, no. 8, pp. 1171–1185, 2009. [44] Y. Wei, J. Qiu, H. K. Lam, and L. Wu, “Approaches to T-S fuzzy-affine-
model-based reliable output feedback control for nonlinear Ito stochastic
[22] H. Gao and T. Chen, “New results on stability of discrete-time systems
systems,” IEEE Trans. Fuzzy Syst., vol. 25, no. 3, pp. 569–583, 2017.
with time-varying state delay,” IEEE Trans. Autom. Contr., vol. 52, no. 2,
pp. 328–334, 2007.
[23] G.-P. Liu, Y. Xia, J. Chen, D. Rees, and W. Hu, “Networked predictive
control of systems with random network delays in both forward and
feedback channels,” IEEE Trans. Ind. Electron., vol. 54, no. 3, pp. 1282–
1297, 2007.
[24] R. Wang, G.-P. Liu, W. Wang, D. Rees, and Y.-B. Zhao, “H∞ control for
networked predictive control systems based on the switched lyapunov
function method,” IEEE Trans. Ind. Electron., vol. 57, no. 10, pp. 3565–
3571, 2010.
[25] Y. Shi and B. Yu, “Robust mixed H2 /H∞ control of networked
control systems with random time delays in both forward and backward
communication links,” Automatica, vol. 47, no. 4, pp. 754–760, 2011.
[26] L. Xie and Y. C. Soh, “Robust control of linear systems with generalized
positive real uncertainty,” Automatica, vol. 33, no. 5, pp. 963–967, 1997.
[27] H. Zhang, Y. Shi, and A. Saadat Mehr, “Robust H∞ PID control
for multivariable networked control systems with disturbance/noise
attenuation,” Int. J. Robust Nonlinear Control, vol. 22, no. 2, pp. 183–
204, 2012.
[28] H. Zhang, Y. Shi, and A. Saadat Mehr, “Robust static output feedback
control and remote PID design for networked motor systems,” IEEE
Trans. Ind. Electron., vol. 58, no. 12, pp. 5396–5405, 2011.
[29] H. Zhang, Y. Chang, J. Wang, and Y. Shi, “Time-domain Smith predictor
in discrete-time networked control systems,” in Proceeding of the 11th
World Congress on Intelligent Control and Automation, 2014, pp. 1089–
1094.
[30] H. Gao, J. Lam, C. Wang, and Y. Wang, “Delay-dependent output-
feedback stabilisation of discrete-time systems with time-varying state
delay,” IEE Proc.-Control Theory Appl, vol. 151, no. 6, pp. 691–698,
2004.
[31] H. Zhang, A. Saadat Mehr, and Y. Shi, “Improved robust energy-to-peak
filtering for uncertain linear systems,” Signal Process., vol. 90, no. 9,
pp. 2667–2675, 2010.
[32] J. Dong and G.-H. Yang, “Robust static output feedback control Hui Zhang (M’15, SM’17) received the B.Sc. de-
synthesis for linear continuous systems with polytopic uncertainties,” gree in mechanical design manufacturing and au-
Automatica, vol. 49, no. 6, pp. 1821–1829, 2013. tomation from the Harbin Institute of Technology at
Weihai, Weihai, China in 2006, the M.Sc. degree
[33] X. G. Liu, R. R. Martin, M. Wu, and M. L. Tang, “Delay-dependent
in automotive engineering from Jilin University,
robust stabilisation of discrete-time systems with time-varying delay,”
Changchun, China in 2008, and the Ph.D. degree in
IEE Proc.-Control Theory Appl, vol. 153, no. 6, pp. 689–702, 2006.
mechanical engineering from University of Victoria,
[34] H. Gao and T. Chen, “New results on stability of discrete-time systems
Victoria, BC, Canada in 2012.
with time-varying state delay,” IEEE Trans. Autom. Contr., vol. 52, no. 2,
Dr. Zhang’s research interests include Diesel en-
pp. 328–334, 2007.
gine aftertreatment systems, vehicle dynamics and
[35] J. A. Cook, J. Sun, J. H. Buckland, I. V. Kolmanovsky, H. Peng, and control, mechatronics, robust control and filtering,
J. W. Grizzle, “Automotive powertrain control – a survey,” Asian J. networked control systems, and multi-agent systems. He is an author / co-
Control, vol. 8, no. 3, pp. 237–260, 2006. author of over 80 peer-reviewed papers on journals and conference proceed-
[36] S. Tuplin, M. C. Best, and M. A. Passmore, “Improvement of perceived ings.
vehicle performance through adaptive electronic throttle control,” Proc. Dr. Zhang has served on the IFAC Technical Committee on Automotive
Instn Mech. Engrs, Part D: J. Automobile Engineering, vol. 217, no. 2, Control, ASME Automotive and Transportation Systems Technical Commit-
pp. 97–106, 2003. tee, SAE Commercial Vehicle Committee, and International Program Commit-
[37] M. Corno, M. Tanelli, S. M. Savaresi, and L. Fabbri, “Design and tee for the IASTED International Conference on Control and Applications. Dr.
validation of a gain-scheduled controller for the electronic throttle body Zhang serves as an Associate Editor for Neurocomputing, SAE International
in ride-by-wire racing motorcycles,” IEEE Trans. Control Syst. Technol., Journal of Vehicle Dynamics, Stability, and NVH, SAE International Journal
vol. 19, no. 1, pp. 18–30, 2011. of Connected and Automated Vehicles; Board member of International Journal
[38] S. Zhang, J. J. Yang, and G. G. Zhu, “LPV modeling and mixed of Hybrid and Electric Vehicles, Mechanical Systems and Signal Processing;
constrained H2 /H∞ control of an electronic throttle,” IEEE/ASME Guest Editor of Mechatronics, IEEE Access, ISA Transactions, Mechanical
Trans. Mechatron., vol. 20, no. 5, pp. 2120–2132, 2015. Systems and Signal Processing, Journal of the Franklin Institute, and Inter-
[39] A. K. Al-Jiboory, A. White, S. Zhang, G. Zhu, and J. Choi, “Linear national Journal of Vehicle Design; Conference Editorial Board of ASME
matrix inequalities approach to input covariance constraint control with Dynamic Systems and Control Division and American Control Conference.
application to electronic throttle,” ASME. J. Dyn. Sys., Meas., Control, Dr. Zhang is a recipient of 2017 IEEE Transactions on Fuzzy Systems
vol. 137, no. 9, pp. 091 010–091 010–9, 2015. Outstanding Paper Award.

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIE.2018.2795574, IEEE
Transactions on Industrial Electronics

10

Yang Shi (SM’09, F’17) received the Ph.D. degree

in Electrical and Computer Engineering from the
University of Alberta, Edmonton, AB, Canada, in
2005. From 2005 to 2009, he was a Faculty Member
in the Department of Mechanical Engineering, U-
niversity of Saskatchewan, Saskatoon, SK, Canada.
Currently, he is a Professor in the Department of
Mechanical Engineering, University of Victoria, Vic-
toria, British Columbia, Canada. His current research
interests include networked and distributed systems,
model predictive control, industrial cyber-physical
systems, mechatronics, navigation and control of autonomous systems (AUV
and UAV), and energy systems.

Dr. Shi received the University of Saskatchewan Student Union Teach-

ing Excellence Award in 2007, and the Faculty of Engineering Teaching
Excellence Award at the University of Victoria in 2012. He received the
JSPS Invitation Fellowship (short-term), and was a Visiting Professor at
the University of Tokyo during Nov-Dec 2013 and the 2015 Craigdarroch
Silver Medal for Excellence in Research at the University of Victoria;
in 2017, he received the Humboldt Research Fellowship (for Experienced
researchers). Currently, he is a Co-Editor-in-Chief of IEEE Trans. Industrial
Electronics; also he serves as Associate Editor for Automatica, IEEE Trans.
Control Systems Technology, IEEE/ASME Trans. Mechatronics, IEEE Trans.
Cybernetics, ASME Journal of Dynamic Systems, Measurement, and Control.
He is a Fellow of IEEE, ASME and CSME, and a registered Professional
Engineer in British Columbia, Canada.

Junmin Wang (SM’14) received the B.E. degree in

Automotive Engineering and his first M.S. degree
in Power Machinery and Engineering from the Ts-
inghua University, Beijing, China in 1997 and 2000,
respectively. He received his second and third M.S.
degrees in Electrical Engineering and Mechanical
Engineering from the University of Minnesota, Twin
Cities in 2003, and the Ph.D. degree in Mechanical
Engineering from the University of Texas at Austin
in 2007.
Dr. Wang has 5-years of full-time industrial re-
search experience (May 2003 - August 2008) at Southwest Research Institute
(San Antonio, Texas). In September 2008, Dr. Wang joined Ohio State
University and founded the Vehicle Systems and Control Laboratory. Dr.
Wang is the author or co-author of more than 260 peer-reviewed publications
including 135 journal articles and 12 U.S. patents.
Prof. Wang is a recipient of the Ohio State University Harrison Faculty
Award for Excellence in 2015, IEEE Distinguished Lecturer in 2015, National
Science Foundation CAREER Award, SAE International Ralph R. Teetor
Educational Award in 2012, SAE International Vincent Bendix Automotive
Electronics Engineering Award in 2011, and Office of Naval Research
Young Investigator Program (ONR-YIP) Award in 2009. He is a Fellow of
SAE (Society of Automotive Engineers) and a Fellow of ASME (American
Society of Mechanical Engineers). He serves as Associate Editor for various
international journals.

Hong Chen (M’02, SM’12) received the B.S. and

M.S. degrees in process control from Zhejiang Uni-
versity, Zhejiang, China, in 1983 and 1986, respec-
tively, and the Ph.D. degree in system dynamics and
control engineering from the University of Stuttgart,
Stuttgart, Germany, in 1997. Since 1999, she has
been a Professor in Jilin University, Changchun,
China, where she currently serves as Tang Aoqing
Professor and as the Director of the State Key Lab-
oratory of Automotive Simulation and Control. Her
current research interests include model predictive
control, optimal and robust control, nonlinear control and applications in
mechatronic systems focusing on automotive systems.

0278-0046 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.