Int. J. Appl. Math. Comput. Sci., 2020, Vol. 30, No.

3, 517–528
DOI: 10.34768/amcs-2020-0038

DISCRETE–TIME SLIDING MODE CONTROL OF LINEAR SYSTEMS WITH

INPUT SATURATION

B OBAN VESELI Ć a , Č EDOMIR MILOSAVLJEVI Ć b , B RANISLAVA PERUNI ČI Ć-DRA ŽENOVI Ć c ,

S ENAD HUSEINBEGOVI Ć c , M ILUTIN PETRONIJEVI Ć a,∗

a
Faculty of Electronic Engineering
University of Niš
A. Medvedeva 14, Niš 18000, Serbia
e-mail: [email protected]
b
Faculty of Electrical Engineering
University of Istočno Sarajevo
Vuka Karadžića 30, 71123 Istočno Sarajevo, Bosnia and Herzegovina
c
Faculty of Electrical Engineering
University of Sarajevo
Zmaja od Bosne bb, 71000 Sarajevo, Bosnia and Herzegovina

The paper proposes a discrete-time sliding mode controller for single input linear dynamical systems, under requirements
of the fast response without overshoot and strong robustness to matched disturbances. The system input saturation is
imposed during the design due to inevitable limitations of most actuators. The system disturbances are compensated by
employing nonlinear estimation by integrating the signum of the sliding variable. Hence, the proposed control structure
may be regarded as a super-twisting-like algorithm. The designed system stability is analyzed as well as the sliding man-
ifold convergence conditions are derived using a discrete-time model of the system in the δ-domain. The results obtained
theoretically have been verified by computer simulations.

Keywords: discrete-time sliding mode control, super-twisting controller, input saturation, disturbance compensation.

1. Introduction manifold under the action of the discontinuous control. In

addition to these remarkable features, some shortcomings
Variable structure control systems (VSCSs) (Emelyanov,
have also been noticed. The first one is related to the
1957) operating in sliding mode (SM) (Utkin, 1992), as a
discontinuous nature of control. Namely, every real
robust nonlinear control technique, have attracted a lot of
control system has small transport/inertial delays usually
attention in the literature. This is due to the following
present in the control plant and/or in actuators and sensors
important features: order reduction of the system
that are not captured by its mathematical model used
dynamics in SM, theoretical invariance to parameter
in controller design. Such unmodeled dynamics can be
variations and disturbances acting through the control
excited by discontinuous control inducing high-frequency
channels (Draženović, 1969) and a simple two-step design
oscillations (chattering) around the sliding manifold (and
procedure (Utkin, 1992). First, the desired SM dynamics
the equilibrium state). Chattering cannot be tolerated
is selected by defining the appropriate sliding manifold
in most applications, especially in electromechanical
in the state space, and then a discontinuous control is
systems, where it results in unpleasant sound effects, wear
designed that brings the system state to the manifold in a
out of mechanical parts, or drive overheating.
finite time (Ackermann and Utkin, 1998). Further motion
towards the equilibrium takes place along the sliding In order to mitigate the effects of chattering, the
following methods were proposed: the boundary layer
∗ Corresponding author method (Slotine, 1984), application of an observer
 518 B. Veselić et al.

(Bondarev et al., 1985) as a bypass for chattering, Euler discretization method, this property is lost (Yan
the implicit method for realization of the signum et al., 2015). The system state arrives into a vicinity of the
function (Golo et al., 2000; Huber et al., 2016) and sliding manifold in a finite time, and the further motion
application of higher order sliding modes (HOSMs) takes place in a quasi-sliding domain around the manifold
(Levant, 1993). In the first method, the discontinuous even in the absence of disturbances.
(signum) function is replaced by a high-gain saturation This paper discusses DT SMC design for single
or a sigmoid function that eliminates or alleviates input linear time-invariant (LTI) systems that provides
chattering, but the system robustness is reduced because a fast response without overshoot, but with reduced
the system gain becomes limited. In the second chattering and strong robustness to matched disturbances.
approach, the controller governs the observer and not Furthermore, input saturation is imposed on the system
the real plant, thus excluding the plant from the considered, which is a realistic situation due to
chattering-contaminated loop. The disadvantage is that inevitable construction limitations of actuators. Hence,
the plant parameter variations deteriorate robustness and the controller output is saturated, which has required
system performance. In the implicit control method, the additional stability analysis to determine sliding manifold
signum function is treated as a multi-valued function, convergence conditions under saturation. This problem
which reduces chattering. This approach is more was also discussed by Bartolini et al. (1995) and Corradini
complicated for practical implementation and does not et al. (2014) for classic DT SMC systems, and Golkani
result in significantly better results than a method that is et al. (2018), Shtessel et al. (2012), Castillo et al. (2016)
the subject of this paper. for CT STC.
HOSMs have become very popular in the past As in the work of Golo and Milosavljević (2000), the
decade. Since the introduction of HOSMs, the former analysis and design of the DT SM controller in this paper
first-order SMs have been called classic SMs. In classic have been also carried out using a DT model of the system
SM, the sliding variable is a continuous function of time, in the δ-domain, which in this case allows separation of
whereas its time derivative is a discontinuous function. In the reaching control component that achieves reaching the
the r-th-order SM, r − 1 time-derivatives of the sliding sliding manifold and the sliding control component that
variable are continuous, whereas the r-th time derivative secures sliding along that manifold. The reaching control
is a discontinuous function. The second-order sliding of the proposed controller is selected to be linear in order
mode certainly attracted most attention because of its not to excite chattering. However, to provide robustness
practical feasibility (Bartolini et al., 1998). According it is necessary to introduce disturbance compensation.
to Utkin (2016), all HOSMs are ultimately reduced to Some methods for it were suggested by Su et al. (2000)
classical SMs in the sliding variable subspace, and only and Milosavljević et al. (2007) for DTSMC systems,
the super-twisting control (STC) algorithm (Levant, 1993) both applying linear estimation laws. In this paper, an
achieves a true second-order SM. The STC algorithm has additional compensation control component is introduced,
become very popular in SM theory and practice. STC which is formed by applying nonlinear estimation by
was originally developed for single input continuous-time integrating the signum of the sliding variable. Thus,
(CT) systems (Levant, 1993). The relative order of the the proposed control structure resembles the structure
system must be one with respect to the sliding variable, of discretized STC. Finally, in order to avoid integrator
taken as an output. wind-up because of the saturation in the system, an
Modern control systems are implemented using anti-wind-up mechanism is embedded into the controller.
computers or microprocessors, i.e., in the discrete-time The rest of the paper is organized as follows.
(DT) domain. DTSM analysis and development began Section 2 summarizes some basic design aspects of DT
in the 1980s and 1990s (Milosavljević, 1985; Drakunov SMC and briefly describes the DT realization of an STC.
and Utkin, 1989; Gao et al., 1995; Bartolini et al., 1995; Section 3 carries the main results and proposes a DT
Bartoszewicz, 1998; Bartoszewicz and Leśniewski, 2014; SM controller that meets the predefined design tasks.
Bartoszewicz and Adamiak, 2019) in the case of the Stability analysis and convergence conditions are also
first-order SM. The DT development of HOSMs began given. Section 4 supports the designed control system
with the paper (Bartolini et al., 2001) and became one of by presenting simulation results of an illustrative example.
the research fields by well-known authors (Bartoszewicz The paper ends with conclusions and the used literature.
and Latosinski, 2017; Chakrabarty et al., 2017). However,
there are relatively few papers in the field of DT
realization of STC (Salgado et al., 2016; 2011; Yan
2. Preliminaries
et al., 2015; Koch and Reichhartinger, 2019). An Consider a linear time invariant continuous time dynamic
essential feature of CT STC is that the system state system described as
reaches exactly the sliding manifold in a finite time.
By simple discretization of CT STC by applying the ẋ(t) = Ax (t) + b (u (t) + d (t)) , (1)
Discrete-time sliding mode control of linear systems with input saturation 519

where x ∈ Rn is the state vector, u ∈ R is the control the system (2). Also, the operator † in (6) denotes the
signal, and A ∈ Rn×n and b ∈ Rn×1 are the state and matrix pseudo-inversion. The control that establishes the
input matrices, respectively. Also, the system is subjected SM along the manifold (4) is obtained by solving equation
to a bounded matched disturbance d, |d (t)| ≤ d0 < ∞. sk+1 = 0, using the model (2), as
To implement SMC it is necessary (i) to design a sliding
manifold in the state space and (ii) to choose a control that ueq,k = −cd Ad xk − dk . (7)
establishes a stable SM. The system motion can be divided
into two phases: reaching phase and SM. Consequently, This linear control is the equivalent control, which is
there are two control components. The reaching control in this case both the reaching control and the SM
should bring the system state from any initial state to the control. Under this control, the system reaches the sliding
sliding manifold, and the SM control should drive the manifold in one step and slides along it afterwards. It
system state along the sliding surface. In CT systems requires the knowledge of disturbance dk , which is usually
it is possible to have a unique reaching and SM control unknown. Therefore, only the equivalent control for
but it must have a discontinuous nature, which can excite nominal system (dk = 0) is feasible in practice as
unmodeled dynamics and induce chattering.
uk = −cd Ad xk = −kde xk . (8)
As the scope of the paper is DTSMC, some
elementary design principles of these systems are This means that the disturbance should be estimated and
summarized below. compensated as best as possible by an additional control
component. Also, there is no possibility of shaping the
2.1. Discrete-time sliding mode control design. reaching phase.
Consider a linear time invariant continuous time dynamic
system described as
2.2. Discrete-time sliding mode control design in
xk+1 = Ad xk + bd (uk + dk ) , (2) the δ-domain. Another design approach is to use a
 T DT model of a system in the δ-domain (Golo and
AT
Ad = e , bd = eAt b dt, (3) Milosavljević, 2000). A mathematical model of the
0 system (1) in the δ-domain can be obtained using (2) in
under the assumption that the sampling period T is the following manner:
sufficiently small and the disturbance is slowly varying,
xk+1 − xk
due to which the disturbance can be considered constant δxk = = Aδ xk + bδ (uk + dk ) , (9)
T
over the sampling interval. Note that the time
discretization disrupts the matching property (Draženović, 1 bd
1969) from CT. It was shown by Abidi et al. (2007) Aδ = (Ad − In ) , bδ = . (10)
T T
that
 the unmatched part of disturbance in DT is of order Now, the sliding variable in the δ-domain is defined as
O T 3 . However, for constant disturbances the matching
property is preserved, which is the ground for the main sδ,k = cδ xk , cδ bδ = 1. (11)
assumption.
The sliding motion should be organized along the The desired eigenvalues (5) can be mapped into the
sliding manifold δ-domain according to (10) as λδ,i = (λd,i − 1) T −1 ,
(i = 1, . . . , n − 1), which gives the spectrum in the
sk = cd xk = 0, (4)
δ-domain
where the constant vector cd should provide the desired  
reduced order SM dynamics, defined by an eigenvalue λδ = λδ,1 λδ,2 · · · λδ,n−1 0 . (12)
spectrum
Equation (6) for finding the sliding manifold
 
λd = λd,1 λd,2 · · · λd,n−1 0 . (5) parameters in this case becomes
  †
According to the comprehensive approach to the sliding cδ = kδe 1 Aδ bδ , (13)
manifold design (Draženović et al., 2013), vector cd
that provides the desired dynamics (5) and common where kδe is the state feedback gain vector that provides
requirement cd bd = 1 can be easily found using the desired spectrum (12) in the system (9). According to
  † (11) and (9) it follows that
cd = kde 1 Ad bd . (6)
sδ,k+1 − sδ,k
δsδ,k =
A gain vector kde in (6) is the one that provides the T (14)
desired spectrum (5) by conventional state feedback in = cδ δxk = cδ Aδ xk + uk + dk .
 520 B. Veselić et al.

The equivalent control for the δ-domain compensation, is the STC algorithm (Levant, 1993). It
representation can be determined from the condition belongs to the group of second-order SMC algorithms,
sδ,k+1 = 0 by solving the previous equation with respect developed for CT control systems with relative degree one
to uk . This yields the equivalent control in the following with respect to the sliding variable. STC is defined by the
form: following equations:
sδ,k
ueq,k = −cδ Aδ xk − − dk . (15)
T u = −kp |s|1/2 sgn(s) + w, (19a)
Unlike control (7), in this control form there is the
term sδ,k T −1 that plays a role of the reaching control, ẇ = −ki sgn (s) . (19b)
which becomes zero when the sliding surface is reached. Here w is the disturbance estimate that is used in
Hence, this approach offers the possibility to influence (19a) for the compensation to obtain robustness of the
the reaching phase of the system motion, which will ST controller. A DT version of the STC (Koch and
be exploited in this paper for obtaining a suitable Reichhartinger, 2019) is usually obtained by applying the
control structure that minimizes chattering and increases explicit Euler discretization method, yielding
dynamical properties in real systems. The feasible part of
the equivalent control (15) is given by uk = −kp |sk |1/2 sgn(sk ) + wk , (20a)
  wk = wk−1 − ki T sgn(sk−1 ). (20b)
sδ,k 1
uk = −cδ Aδ xk − = − kδe + cδ xk . (16) Comparison of the controllers (18) and (20) shows that
T T
  they have identical structure. However, the control
By comparing (8) and (16), the equality kδe + T1 cδ = components in (18) are linear, whereas in (20) they are
kde holds. Also, from the design condition cδ bδ = cd bd = realized using nonlinear functions.
1 and bd = T bδ , it follows that cδ = T cd. It has been noticed by Milosavljević et al. (2019)
1/2
that the factor |s| present in (19a), which provides
2.3. Disturbance compensation via the sliding vari- finite time reaching in CT, loses its importance in the DT
able. If the control (8) is applied in the system (2), realization (20a) since then it is only possible to achieve
the future value of the sliding variable sk , according to quasi-SM in real systems. By neglecting the square root,
(4) and (2), would be sk+1 = dk . Hence, dk−1 = the control (20a) becomes linear. Hence, a combination
sk , which means that the past value of the disturbance of (18a) and (20b) was proposed as the controller by
can be estimated according to the current value of the Milosavljević et al. (2019), whose performance was
sliding variable. This knowledge can be employed for thoroughly analysed in the control of first order plants
disturbance compensation. with/without unmodeled dynamics.
A compensation control for slowly varying
disturbances was proposed by Milosavljević et al. 3. DTSM controller design
(2007) and Lješnjanin et al. (2011) in the form
The control task is to design a digital controller
uc,k = uc,k−1 − hsk , (17) that provides a fast response without overshoot with
significant robustness to disturbances. The controller
with stability condition 0 < hT < 1. This control is, should also take into account the actuator saturation
in fact, an integration of the sliding variable. Due to property that limits the magnitude of the control signal.
this integrating property, the compensation control fully To meet the given requirements, the designed DTSM
rejects stepwise disturbances and significantly decreases controller generally combines in stages two control
the impact of other slowly time-varying disturbances. The principles with the proposed modifications in order to
overall control can now be rewritten as eliminate their well-known shortcomings. The first
control phase is linear providing a near deadbeat response
uk = −cd Ad xk + uc,k , (18a)
while a nonlinear control phase, based on a super twisting
uc,k = uc,k−1 − hsk . (18b) like algorithm, is activated in the vicinity of the sliding
manifold. In this way a quick arrival into the quasi-sliding
The described method belongs to the class of linear domain is ensured, avoiding the response overshoot due to
compensators. Further improvements in system integrator windup during the inevitable control saturation
compensation can be achieved by employing nonlinear arising in the first stage of motion.
estimation of disturbance. The proposed control law is given by the following
set of equations:
2.4. Discrete-time realization of the STC. A
very popular SMC algorithm, which incorporates a U0 sgn(uΣ,k ) if |uΣ,k | > U0 ,
uk = (21a)
sliding variable based part for disturbance estimation and uΣ,k if |uΣ,k | ≤ U0 ,
Discrete-time sliding mode control of linear systems with input saturation 521

uΣ,k = ul,k − p2,k uc,k , (21b) 3.1. Stability of the system with saturated control.
sδ,k Since the saturation is invoked by the linear control (22)
ul,k = −cδ Aδ xk − [ks1 + (1 − p2,k ) ks2 ] , (21c) and the compensational control is activated only in the
T
small vicinity of the sliding manifold, stability analysis
uc,k = uc,k−1 + kint T sgn (sδ,k−1 ) , (21d) during saturation will be analyzed with respect to linear
control. From (14), the following equation defines the
0 if |uΣ,k | > U0 ,
p1,k = (21e) sliding manifold dynamics:
1 if |uΣ,k | ≤ U0 ,
sδ,k+1 = sδ,k + T cδ Aδ xk + T (uk + dk ) . (25)
p2,k = p1,k−1 , (21f)
The control signal is calculated according to (22). If
ks1 , ks2 > 0, ks1 + ks2 ≤ 1. (21g) the calculated control is greater than the allowed actuator
The given control strategy can be summarized as limitation, i.e., ul,k > U0 , the applied control would be
follows. The control signal consists of two parts: linear ul,k
uk = U0 sgn (ul,k ) = U0 . (26)
ul,k and nonlinear uc,k . For the system state far away from |ul,k |
the sliding manifold, the control will be saturated due to
the linear control term that tends to bring the system state The following proposition imposes a condition on the
onto the sliding surface in a nominal system. Hence, a limit value U0 to get the system out of saturation.
high control magnitude will be generated so the control
signal must be limited to the U0 value, which is acceptable Proposition 1. The DT system (9) with the controller
by an actuator. In the sampling instant when the controller (21) operating in the saturation will leave this mode in a
output exits saturation, the linear control finite number of sampling periods if
sδ,k U0 > |cδ Aδ xk | + d0 , ∀k ≥ 0. (27)
ul,k = −cδ Aδ xk − (ks1 + ks2 ) (22)
T Proof. Under the applied saturated control (26), the
is applied during only one sampling period. For the limit sliding dynamics becomes
case ks1 + ks2 = 1, the control (22) is a deadbeat control
sδ,k+1 = sδ,k + T cδ Aδ xk + T dk
that ensures sδ,k+1 = 0 in the nominal system (dk = 0).
In real cases ks1 + ks2 ≤ 1 since dk = 0 (including T cδ Aδ xk + (ks1 + ks2 ) sδ,k (28)
− U0 .
unmodeled dynamics) in order to adjust the width of a |ul,k |
quasi-sliding domain. Such control brings the system Let ks1 + ks2 + ks3 = 1, where 0 ≤ ks3 < 1. Then the
state to a vicinity of the sliding manifold. In the next sliding dynamics can be rewritten as
sampling period, the gain of the linear part is reduced and
the nonlinear component is activated, so the control signal sδ,k+1 = [(ks1 + ks2 ) sδ,k + T cδ Aδ xk ]
 
becomes U0 (29)
× 1− + ks3 sδ,k + T dk .
|ul,k |
uk = −cδ Aδ xk − T −1 ks1 sδ,k − uc,k , (23)
Since |ul,k | > U0 we have
uc,k = uc,k−1 + kint T sgn (sδ,k−1 ) . (24)  
U0
The control (23), (24) represents an ST like structure, 0< 1− < 1.
|ul,k |
which can be obtained from the original ST algorithm
by replacing the factor |s|
1/2
with |s|. Neglecting the Also, having in mind that 0 ≤ ks3 < 1 and |dk | ≤ d0 , the
square root function in the DT realization of the STC following inequality can be obtained:
is quite acceptable, since only a quasi-sliding mode  
U0
can be attained by a digital controller. The control |sδ,k+1 | ≤ |(ks1 + ks2 ) sδ,k + T cδ Aδ xk | 1 −
|ul,k |
component uc,k , as an output of the DT integrator, can
be understood as a compensational control that tends to + ks3 |sδ,k | + T d0
 
cancel disturbance effects on the system behavior, which U0
= T |ul,k | 1 − + ks3 |sδ,k | + T d0
will be displayed later in the text. |ul,k |
In the light of the given explanation of the controller = T |ul,k | − T U0 + ks3 |sδ,k | + T d0
operation, to prove the stability of the proposed control
≤ (ks1 + ks2 ) |sδ,k | + T |cδ Aδ xk | − T U0
system means to prove the state trajectory convergence
to the sliding manifold for both saturated and unsaturated + ks3 |sδ,k | + T d0
control phases. The required stability analysis is presented = |sδ,k | + T (|cδ Aδ xk | + d0 − U0 ) .
in the subsequent subsections. (30)
 522 B. Veselić et al.

Hence 3.2. Stability of the system with unsaturated control.

After the control signal exits saturation, by virtue of (21)
|sδ,k+1 | ≤ |sδ,k | + T (|cδ Aδ xk | + d0 − U0 ) . (31)
in the next discretization period the following control acts:
If the condition (27) holds, then
|sδ,k+1 | < |sδ,k | , (32) uk = −cδ Aδ xk − T −1 (ks1 + ks2 ) sδ,k , (37)

which shows that |sδ,k | decreases monotonically and the so δsδ,k becomes
system trajectories will be directed towards the sliding
manifold. This means that both sδ,k and xk will be δsδ,k = −T −1 (ks1 + ks2 ) sδ,k + dk . (38)
decreasing, making ul,k decreasing as well. Therefore, it
is inevitable that the control signal becomes smaller than Since δsδ,k = T −1 (sδ,k+1 − sδ,k ), we get
the saturation limit and the system exits saturation. 
Usually, in practice the value U0 cannot be chosen sδ,k+1 = (1 − ks1 − ks2 ) sδ,k + T dk . (39)
greater than the actual physical limitation of the actuator.
Also, the maximum disturbance magnitude d0 acting on It is interesting to notice that the limit case ks1 +
the system is determined by the system construction and ks2 = 1 gives sδ,k+1 = T dk , which indicates
cannot be affected. Therefore, an area in the state space that this control brings the system state into an O (T )
should be determined where prerequisite (27) is valid vicinity of the sliding manifold whose dimensions depend
for the predefined U0 and d0 . The following remark on the disturbance magnitude. In the nominal case
conservatively finds that area. (dk = 0) , sδ,k+1 = 0 is achieved, so this is actually the
Remark 1. The condition (27) is satisfied for the given equivalent control providing the deadbeat response.
U0 and d0 within a ball around the origin defined by After the first discretization period of the unsaturated
control (37), the controller gain is then reduced (ks2 = 0)
U0 − d0
x < r, r= . (33) and the compensational control uc,k is activated,
cδ Aδ  according to (21). Hence, the controller is now described
where x defines Euclidean vector norm. by
Proof. The product cδ Aδ xk is a scalar, so that uk = −cδ Aδ xk − T −1 ks1 sδ,k − uc,k , (40)
|cδ Aδ xk | = cδ Aδ xk . Then, inequality (31) can be
rewritten as uc,k = uc,k−1 + kint T sgn (sδ,k−1 ) . (41)

|sδ,k+1 | ≤ |sδ,k | + T (cδ Aδ xk  + d0 − U0 ), (34) Then

Since cδ Aδ xk  ≤ cδ Aδ  · xk , the above inequality
sδ,k+1 = (1 − ks1 ) sδ,k + T (dk − uc,k ) , (42)
becomes
|sδ,k+1 | ≤ |sδ,k | + T (cδ Aδ  · xk  + d0 − U0 ). (35) uc,k+1 = uc,k + kint T sgn (sδ,k ) . (43)
To fulfil the sliding manifold convergence condition (32),
The compensational control uc,k actually represents
it is necessary to have cδ Aδ  · xk  + d0 − U0 < 0, ∀k ≥
the disturbance estimated value so let a new variable zk =
0, which gives
dk − uc,k be introduced, denoting the estimation error.
U0 − d0 Then the system dynamics can be described as
xk  < , ∀k ≥ 0. (36)
cδ Aδ 
This inequality defines a ball around the origin with the sδ,k+1 = (1 − ks1 ) sδ,k + T zk , (44)
radius r given in (33). 

It is not always possible in any system to fulfil (27) zk+1 = −kint T sgn (sδ,k ) + zk + Δk , (45)
for a given initial state. Besides the initial state, the where Δk = dk+1 − dk . The stability of the
fulfilment of (27) depends on the plant dynamics as well above described system is established by the following
as the required sliding mode dynamics. It may happen proposition.
that for demanding a fast sliding mode dynamics, the
condition (27) cannot be fulfilled for a given initial state Proposition 2. The SM convergence condition for the
for the maximum admissible U0 . Then the system slips DT system described by (44) and (45), with the controller
into instability. It should also be emphasized that the parameters 0 < ks1 ≤ 1 and kint > 0, is satisfied within
occurrence of an external disturbance further worsens the area defined by
the fulfilment of this condition. Time derivatives of the
reference signal must be bounded as well since they are kint T 2
components of the disturbance d(t). |sδ,k | > . (46)
ks1
Discrete-time sliding mode control of linear systems with input saturation 523

Proof. By using sgn (sδ,k ) = sδ,k / |sδ,k |, the Suppose that the system state reached the convergence
above nonlinear system (44), (45) can be rewritten in boundary |sδ,k | = kint T 2 /ks1 . Let the compensational
a pseudo-linear form (Koch and Reichhartinger, 2019; control be uc,k = 0 (the integrator is empty or reset)
Ghane and Menhaj, 2015) and let the system be affected by a disturbance having
a maximal magnitude of |dk | = d0 . Then the sliding
σk+1 = Λ (sδ,k ) σk + pk , (47) variable dynamics (42) is described by

 sδ,k+1 = (1 − ks1 ) sδ,k + T dk . (51)
sδ,k
σk = , Hence
zk

 |sδ,k+1 | ≤ (1 − ks1 ) |sδ,k | + T |dk |
1 − ks1 T
Λ (sδ,k ) = −kint T ,
|sδ,k | 1 kint T 2 (52)

 = (1 − ks1 ) + T d0 .
0 ks1
pk = .
Δk Thus, according to (46) the quasi-sliding domain is
obtained as
The characteristic equation of the pseudo-linear

system (45) can be found as kint T 2 (1 − ks1 ) kint T
< |sδ,k+1 | ≤ T + d0 , (53)
ks1 ks1
F (z) = det (zI − Λ) = 0, (48)
and the system accuracy is of order
 
which yields O T 2 < |sδ,k+1 | ≤ O (T ) . (54)
2
F (z) = a2 z + a1 z + a0 (sδ,k ) = 0, a2 > 0, (49) The obtained upper bound (53) of |sδ | is valid
only when disturbance affects the system (d0 = 0) and
represents the maximal possible deviation from the
a2 = 1, a1 = ks1 − 2,
sliding manifold. However, the effective width of the
kint T 2 quasi-sliding domain is much smaller, since the integrator
a0 = 1 − ks1 + .
|sδ,k | output uc has been accumulated after some time to
compensate the disturbance.
The convergence area can be determined using the
Jury stability test that examines the system characteristic
equation. Stability conditions for the second order system 4. Design example
(47) are given through the following inequalities F (1) > The proposed control structure has been examined in an
0, F (−1) > 0 and |a0 | < a2 . In the case of (49) these illustrative example and verified through simulation tests.
conditions are expressed by the following requirements For that purpose, let an arbitrary (academic) fifth order
unstable controllable linear plant model (1) be defined by
kint T 2 ⎡ ⎤ ⎡ ⎤
> 0,
|sδ,k | 1 2 3 −5 6 1
⎢−2 6 −3 −4 −7 ⎥ ⎢−2⎥
kint T 2 ⎢ ⎥ ⎢ ⎥
4 − 2ks1 + > 0, (50) A=⎢ ⎥
⎢ 2 −4 6 −10 12 ⎥ , b = ⎢ 3 ⎥ ,
⎢ ⎥
|sδ,k | ⎣−8 −6 −4 ⎦ ⎣−1⎦
3 1
kint T 2 4 12 −6 −8 −14 2
< ks1 .
|sδ,k |  T
with the initial state x (0) = 10 −5 0 −10 5 .
Since kint > 0 and ks1 ≤ 1, the first two conditions are The corresponding discrete-time δ-model (9) for the
always fulfilled. The convergence area can be determined sampling time T = 0.001 s has the following matrices:
from the third condition as (46).  ⎡
1.0097 2.0521 2.9996
Therefore, the system (47) is stable within the space ⎢ −1.9801 5.9919 −2.9920
⎢
defined by the condition (46), i.e., the system trajectories Aδ = ⎢⎢ 2.0272 −3.9198 6.0112
are directed towards the sliding manifold sδ = 0. It is ⎣ −8.0161 −6.0213 −4.024
important to notice that in the case of constant and slowly −3.9601 11.9839 −5.984
varying disturbances, for which can be assumed that dk = ⎤
dk−1 , the system (47) is autonomous since the external −5.0531 5.9695
input Δk = 0. Then the system trajectories will reach the −3.97 −6.9979 ⎥ ⎥
convergence boundary |sδ,k | = kint T 2 /ks1 . −10.0902 11.967 ⎥ ⎥,
To determine a quasi-sliding domain in the case of 3.0528 0.9676 ⎦
arbitrary disturbances, a worst-case analysis is applied. −7.94 −13.9958
 524 B. Veselić et al.
⎡ ⎤
1.0115
⎢−2.0165⎥ 400
⎢ ⎥
bδ = ⎢ ⎥
⎢ 3.031 ⎥ .
200
⎣−1.0046⎦ 0
a)
-200
1.967 -400
The control task is to bring the system state -600
0 2 4 6 8 10
from the initial state into the equilibrium (the origin)
with the prescribed dynamics by organizing SM along 150
the appropriate manifold. Let the desired SM 100
dynamics
 be defined by  the eigenvalue spectrum λ = b)
50
−1 −2  λT −3 −4
 0 in the CT domain. Using
λδ,i = e i
− 1 /T, i = 1, . . . , n − 1, the δ-domain 0
0 2 4 6 8 10
spectrum is obtained as t [s]
 
λδ = −0.9995 −1.998 −2.9955 −3.992 0
Fig. 1. System response: state coordinates (a), the fulfilment of
To provide the SM mode dynamics, the required sliding the condition (27) (b).
manifold vector cδ is calculated using (13), which yields
 
cδ = 0.4437 0.5794 0.3072 −0.6614 0.063 100

In order to accomplish the control task, the proposed 0

a)
controller (21) is employed, whose parameters are set as
-100
follows: ks1 = 0.9, ks2 = 0.1 and kint = 100. Also, let
the actuator saturation be defined by U0 = 150. 0 2 4 6 8 10
To check the analytically predicted system
behaviour, let the system first be subjected to an action 100
of step external disturbance d (t) = 100h(t − 3), where
50
h(t) is the Heaviside function. The next four figures show b)
the simulation results of this case. Evolution of the state
0
variables is presented in Fig. 1(a), which shows that the 0 2 4 6 8 10
state variables asymptotically reach the origin despite t [s]
the action of the disturbance. The controller output is
given in Fig. 2(a). It can be noted that the control signal Fig. 2. Output of controller (21): the overall control signal (a),
is initially saturated, but exits saturation after a short compensation control component uc and disturbance d
period of time. Figure 1(b) indicates that the condition (b).
(27), which ensures convergence during saturation and
consequently provides saturation termination, is satisfied
during the whole system motion. The dashed line in constant disturbances, such as in this case, it is shown
Fig. 1(a) denotes U0 . Also, Fig. 2(b) shows that the that the system trajectory should reach the convergence
compensation control component uc accurately estimates boundary |sδ,k | = kint T 2 /ks1 . This claim is confirmed
the disturbance, and thus eliminates its impact. by the zoomed details of the sliding variable, given in
According to the sliding variable, which is presented Fig. 4, where the dashed lines denote the convergence
in Fig. 3(a) and enlarged in Fig. 3(b), it can be concluded boundaries and the circles represent the sliding variable in
that the DTSM occurs in a very short time, providing discrete-time instants. It is evident that for all the circles in
desired dynamics and robustness to matched disturbances. the convergence area, the consequent system trajectory is
It can be noticed in Fig. 3(b) that the disturbance directed toward the sliding manifold. This property does
initial action forces the system state to leave the sliding not exist in the space between the two dashed lines, which
manifold, but the controller compensation part restores confirms the correctness of Proposition 2.
sliding motion. In constructing the proposed control structure, one
It is important to emphasize that the sliding motion of the goals was to eliminate the main shortcoming of the
takes place in some vicinity of the sliding manifold due STC algorithm (the appearance of an overshoot), while
to the discrete nature of the controller. Proposition 2 reducing the chattering. Therefore, in the next simulation
finds convergence area defined by (46). This means that test, a comparison of the performances of the proposed
there is no convergence in the immediate vicinity of the and the discretized ST algorithm is performed. Parameters
sliding manifold up to the bounds given by (46). For of the discretized ST controller (20) are set as kp = 100
Discrete-time sliding mode control of linear systems with input saturation 525

10 400
200
0
5 a) a)
-200
-400
0 -600
0 2 4 6 8 10 0 2 4 6 8 10
0.2 400
200
0.1 0
0 b) -200 b)
-400
-0.1 -600
0 2 4 6 8 10 0 2 4 6 8 10
t [s] t [s]

Fig. 3. Sliding variable (enlarged scale in (b)). Fig. 5. State variables: the proposed controller (21) (a), the ST
controller (20) (b).
-4
10 6
8
5
6
4

4
3

2 2

0 1

-2 0

3.99 4 4.01 4.02 4.03 0 2 4 6 8 10

t [s] t [s]

Fig. 4. Sliding variable (zoomed details). Fig. 6. Sliding variables: solid line—the proposed controller
(21), dashed line—the ST controller (20).

and ki = 200. Since there are no specific methods of presented in Fig. 7, although a more pronounced
parameter tuning of the ST controller, these values were chattering can be observed in the ST controller output.
adopted by the criterion to obtain dynamics as close as This advantage of the proposed controller is more visible
possible to the sliding manifold reaching ones compared in the enlarged plot of the sliding variables in Fig. 8.
with the proposed controller. Now, the systems are The proposed controller creates a narrower quasi-sliding
subjected to a more complex disturbance, consisting of domain around the sliding manifold, and the chattering is
sinusoidal and constant parts, i.e., d (t) = 10 sin (4πt) + hardly visible. The width of the resulting quasi-sliding
100h (t − 3). The other conditions and settings in the domain is greater than the convergence boundary, since
simulation are unchanged from the previous case. the external disturbance contains the varying component.
Figure 5 shows evolution of the state variables of the
both controllers. The proposed controller provides the
desired dynamics, while a deviation can be observed in the
5. Conclusion
response of the STC. The reason for such behaviour can Based on the conducted simulation tests, it can be
be identified according to the sliding variables, which are concluded that the designed DT SMC system has
presented in Fig. 6. It is obvious that the STC produces completely satisfied the predefined design requirements.
an overshoot in reaching the sliding manifold and is more A fast response and good elimination of disturbances were
sensitive to abrupt changes in disturbance, which occurred achieved. The overshoot, observed in the response of
in t = 3 s. The performance of the proposed controller is the DT ST controller, was successfully avoided by the
consistent with the previous case. proposed controller. The integrator wind-up did not occur,
Both controllers give similar control signals, either. Moreover, the proposed system induced a much
 526 B. Veselić et al.

Ackermann, J. and Utkin, V. (1998). Sliding mode control design

100 based on Ackermann’s formula, IEEE Transactions on Au-
tomatic Control 43(2): 234–237.
0
a) Bartolini, G., Ferrara, A., Pisano, A. and Usai, E. (1998).
-100 Adaptive reduction of the control effort in chattering-free
sliding-mode control of uncertain nonlinear systems, In-
0 2 4 6 8 10 ternational Journal of Applied Mathematics and Computer
100 Science 8(1): 51–71.
Bartolini, G., Ferrara, A. and Utkin, V.I. (1995). Adaptive
0
sliding mode control in discrete-time systems, Automatica
b)
-100
31(5): 769–773.
Bartolini, G., Pisano, A. and Usai, E. (2001). Digital
0 2 4 6 8 10 second-order sliding mode control for uncertain nonlinear
t [s]
systems, Automatica 37(9): 1371–1377.

Fig. 7. Control signals: the proposed controller (21) (a), the ST Bartoszewicz, A. (1998). Discrete-time quasi-sliding mode
controller (20) (b). control strategies, IEEE Transactions on Industrial Elec-
tronics 45(4): 633–637.
Bartoszewicz, A. and Adamiak, K. (2019). A reference
10-3
trajectory based discrete time sliding mode control
5
strategy, International Journal of Applied Mathemat-
ics and Computer Science 29(3): 517–525, DOI:
0 a)
10.2478/amcs-2019-0038.
-5 Bartoszewicz, A. and Latosinski, P. (2017). Reaching law for
0 2 4 6 8 10 DSMC systems with relative degree 2 switching variable,
10-3 International Journal of Control 90(8): 1626–1638.
5
Bartoszewicz, A. and Leśniewski, P. (2014). An optimal
0
sliding mode congestion controller for connection-oriented
b) communication networks with lossy links, International
-5
Journal of Applied Mathematics and Computer Science
0 2 4 6 8 10
24(1): 87–97, DOI: 10.2478/amcs-2014-0007.
t [s] Bondarev, A.G., Bondarev, S.A., Kostyleva, N.E. and Utkin, V.I.
(1985). Sliding modes in systems with asymptotic state
Fig. 8. Sliding variables (enlarged scale): the proposed con- observers, Avtomatika i Telemekhanika 46(6): 5–11.
troller (21) (a), the ST controller (20) (b).
Castillo, I., Steinberger, M., Fridman, L., Moreno, J.A. and
Horn, M. (2016). Saturated super-twisting algorithm:
Lyapunov based approach, IEEE 55th Conference on
smaller chattering than in the case of DT STC. Decision and Control (CDC), Las Vegas, NV, USA,
The proposed DT SM controller left the saturation pp. 269–273.
mode in finite time as predicted. The theoretically Chakrabarty, S., Bandyopadhyay, B. and Bartoszewicz, A.
proved stability and the obtained convergence bounds (2017). Discrete-time sliding mode control with outputs
are confirmed by the presented simulation results, which of relative degree more than one, in A. Bartoszewicz (Ed.),
speak in favor to the proposed controller. Recent Developments in Sliding Mode Control Theory and
Applications, InTech, London, pp. 21–44.
Corradini, M.L., Cristofaro, A. and Orlando, G. (2014).
Acknowledgment Sliding-mode control of discrete time linear plants with
input saturation: Application to a twin-rotor system, In-
This research was supported by the Ministry of Education, ternational Journal of Control 87(8): 1523–1535.
Science and Technological Development of the Republic
Draženović, B. (1969). The invariance conditions in variable
of Serbia.
structure systems, Automatica 5(3): 287–295.
Draženović, B., Milosavljević, Č. and Veselić, B. (2013).
References Comprehensive approach to sliding mode design and
analysis in linear systems, in B. Bandyopadhyay et al.
Abidi, K., Xu, J. and Yu, X. (2007). On the discrete-time (Eds), Advances in Sliding Mode Control: Concept,
integral sliding-mode control, IEEE Transactions on Au- Theory and Implementation, Springer, Berlin/Heidelberg,
tomatic Control 52(4): 709–715. pp. 1–19.
Discrete-time sliding mode control of linear systems with input saturation 527

Drakunov, S.V. and Utkin, V.I. (1989). On discrete-time sliding Salgado, I., Kamal, S., Bandyopadhyay, B., Chairez, I. and
mode, Proceedings of IFAC Symposium on Nonlinear Con- Fridman, L. (2016). Control of discrete time systems based
trol Systems Design, Capri, Italy, pp. 484–489. on recurrent super-twisting-like algorithm, ISA Transac-
Emelyanov, S.V. (1957). A method to obtain complex regulation tions 64: 47–55.
laws using only the error signal or the regulated coordinate Salgado, I., Kamal, S., Chairez, I., Bandyopadhyay, B. and
and its first derivatives, Avtomatika i Telemekhanika Fridman, L. (2011). Super-twisting-like algorithm in
18(10): 873–885. discrete time nonlinear systems, Proceedings of the 2011
Gao, W., Wang, Y. and Homaifa, A. (1995). Discrete-time International Conference on Advanced Mechatronic Sys-
variable structure control systems, IEEE Transactions on tems, Zhengzhou, China, pp. 497–502.
Industrial Electronics 42(2): 117–122. Shtessel, Y., Taleb, M. and Plestan, F. (2012). A
Ghane, H. and Menhaj, M. B. (2015). Eigenstructure-based novel adaptive-gain supertwisting sliding mode controller:
analysis for non-linear autonomous systems, IMA Journal methodology and application, Automatica 48(5): 759–769.
of Mathematical Control and Information 32(1): 21–40. Slotine, J.J.E. (1984). Sliding controller design for non-linear
Golkani, M.A., Koch, S., Reichhartinger, M. and Horn, M. systems, International Journal of Control 40(2): 421–434.
(2018). A novel saturated super-twisting algorithm, Sys- Su, W.C., Drakunov, S.V. and Ozguner, U. (2000). An O(T 2 )
tems and Control Letters 119: 52–56. boundary layer in sliding mode for sampled-data systems,
Golo, G. and Milosavljević, Č. (2000). Robust discrete-time IEEE Transactions on Automatic Control 45(3): 482–485.
chattering free sliding mode control, Systems and Control Utkin, V. (2016). Discussion aspects of higher order sliding
Letters 41(1): 19–28. mode control, IEEE Transactions on Automatic Control
Golo, G., Schaft, A. and Milosavljević, Č. (2000). Discretization 61(3): 829–833.
of control law for a class of variable structure control Utkin, V.I. (1992). Sliding Modes in Control and Optimization,
systems, Technical Report 1551, University of Twente, Springer, Heidelberg.
Enschede. Yan, Y., Yu, X. and Sun, C. (2015). Discretization behaviors
Huber, O., Brogliato, B., Acary, V., Boubakir, A., Plestan, F. of a super-twisting algorithm based sliding mode control
and Wang, B. (2016). Experimental results on implicit and system, 2015 International Workshop on Recent Advances
explicit time-discretization of equivalent-control-based in Sliding Modes (RASM), Istanbul, Turkey, pp. 1–5.
sliding-mode control, in L. Fridman et al. (Eds), Re-
cent Trends in Sliding Mode Control, IET, London,
pp. 207–235. Boban Veselić received his PhD degree in au-
tomatic control from the Faculty of Electronic
Koch, S. and Reichhartinger, M. (2019). Discrete-time Engineering, University of Niš, Serbia, in 2006.
equivalents of the super-twisting algorithm, Automatica Since 1995, he has been with the Department
107: 190–199. of Automatic Control of the University of Niš,
where he is currently an associate professor. His
Levant, A. (1993). Sliding order and sliding accuracy in major field of study are automatic control sys-
sliding mode control, International Journal of Control tems with special expertise in sliding mode con-
58(6): 1247–1263. trol, on which he has published over 100 scien-
tific papers. His current research interests include
Lješnjanin, M., Peruničić, B., Milosavljević, Č. and Veselić, B. continuous- and discrete-time sliding mode control systems, disturbance
(2011). Disturbance compensation in digital sliding mode, estimation and compensation, servo systems, as well as control of elec-
2011 IEEE EUROCON, International Conference on Com- tric drives.
puter as a Tool, Lisboa, Portugal, pp. 1–4.
Milosavljević, Č. (1985). General conditions for the existence Čedomir Milosavljević (1940) received, re-
of quasi-sliding mode on the switching hyper-plane in spectively, his BSc, MSc and PhD degrees from
discrete variable structure systems, Automatic and Remote Moscow Power Institute (1966), the University of
Control 46(3): 307–314. Niš (1975) and the University of Sarajevo (1982).
From 1967 to 1977 he was with the Electronic In-
Milosavljević, Č., Peruničić-Draženović, B., Veselić, B. and dustry Corporation, Niš, and from 1978 to 2005
Mitić, D. (2007). A new design of servomechanisms with the Faculty of Electronic Engineering, Uni-
with digital sliding mode, Electrical Engineering versity of Niš, where he was a founder of control
89(3): 233–244. engineering studies. He has authored over 250
papers, eight textbooks and 50 devices. He is a
Milosavljević, Č., Petronijević, M., Veselić, B., pioneer in investigations of discrete-time sliding-mode control. His re-
Peruničić-Draženović, B. and Huseinbegović, S. (2019). search interests include sliding modes, motion control systems, and in-
Robust discrete-time quasi-sliding mode based nonlinear dustrial electronics.
PI controller design for control of plants with input
saturation, Journal of Control Engineering and Applied
Informatics 21(3): 31–41.
 528 B. Veselić et al.

Branislava Peruničić-Draženović received her Milutin P. Petronijević received his PhD de-
BSc degree in electrical engineering from the gree in electrical engineering from the University
University of Belgrade in 1960, her candidate of Niš, Serbia (2012). He is currently an asso-
of technical sciences degree from the Institute ciate professor at the Department of Power Engi-
of Control Problems, Moscow, in 1969, and her neering there. He has published numerous papers
PhD degree in electrical engineering from the in conference proceedings and journals. His re-
University of Sarajevo in 1971. She worked for search interests are in control of power electronic
Energoinvest in Sarajevo as a project leader from converters and electric drives. His current re-
1960 to 1970. She was a professor at the Uni- search interests concern real-time application of
versity of Sarajevo from 1976 to 2000, when she sliding mode control in servo drives and power
became a professor emeritus. Her research interests include automatic electronic converters in micro-grid.
control and power systems measurement, control, analysis and planning.
Dr. Peruničić-Draženović was a Fulbright Visiting Scholar in 1984. She Received: 19 March 2020
established the IEEE Section in Bosnia and Herzegovina. She has been Revised: 12 May 2020
an elected member of the Bosnian Academy of Sciences since 1986.
Accepted: 14 May 2020

Senad Huseinbegović is an assistant profes-

sor with the Department of Automatic Control
and Electronics, Faculty of Electrical Engineer-
ing, University of Sarajevo, Bosnia and Herze-
govina. Since 2019, he has been the head of the
Department of Automatic Control and Electron-
ics, University of Sarajevo. He received his BSc,
MSc and PhD degrees in electrical engineering
from the University of Sarajevo, in 2006, 2009,
and 2015, respectively. His research interests in-
clude power electronics, electrical machines and drives, power system
control and protection.