Abstract
This paper addresses the problem of global asymptotic stability of a class of discrete uncertain state-delayed systems described by the Fornasini-Marchesini second local state-space (FMSLSS) model using generalized overflow nonlinearities. The uncertainties are assumed to be norm bounded. A computationally tractable, that is, linear-matrix-inequality-(LMI-) based new criterion for the global asymptotic stability of such system is proposed. It is demonstrated that several previously reported stability criteria for two-dimensional (2D) systems are recovered from the presented approach as special cases. Numerical examples are given to illustrate the usefulness of the presented approach.
1. Introduction
Two-dimensional (2D) systems play an important role in filtering, image data processing and transmission, water stream heating, seismographic data processing, thermal processes, biomedical imaging, gas absorption [1, 2], river pollution modeling [3], process of gas filtration [4], grid-based wireless sensor networks [5, 6], and many other areas. The study of such systems has received considerable attention in the last two decades.
When designing a linear discrete system with fixed point arithmetic using a digital computer or special-purpose hardware, several kinds of nonlinearities, such as overflow and quantization, are produced. These nonlinearities may lead to instability in the designed system. The study of nonlinear systems has received much attention in the literature [6–21].
Physical systems usually suffer from uncertainties that arise because of variations in system parameters, modeling errors, or some ignored factors. The subject of uncertain systems has received a considerable amount of interest [13–16, 21–29].
Delays are frequently encountered in many dynamical systems due to finite speed of information propagation and computational lags. Time delays are another source of instability in the designed system. A significant amount of work concerning the stability of time-delay systems has been done [18, 19, 24–32].
The design of a 2D system so as to ensure the stability of the designed system is an interesting and challenging problem. During the past few decades, the stability properties of 2D discrete systems described by the Fornasini-Marchesini second local state-space (FMSLSS) model [33] have been investigated extensively [6–10, 12–16, 18, 19, 21–26, 28, 30–32, 34–36]. Several publications [7–10, 12, 18, 19] relating to the issue of the global asymptotic stability of the FMSLSS model with overflow nonlinearities have appeared. The stability analysis of uncertain FMSLSS model with overflow nonlinearities has been carried out in [13–16, 21]. In [24], the problem of robust stability and stabilization of 2D state-delayed systems has been addressed. Reference [37] deals with the problem of stabilization of nonlinear 2D discrete Takagi-Sugeno fuzzy systems. The problem of global asymptotic stability of 2D state-delayed FMSLSS model with saturation nonlinearities has been studied in [18, 19]. A 2D filtering approach with performance measure for state-delayed FMSLSS model has been developed in [25]. The guaranteed cost control problem for 2D state-delayed systems has been studied in [28].
The stability analysis of 2D discrete systems described by the FMSLSS model in the simultaneous presence of nonlinearity, state delay, and parameter uncertainty in their physical models is an important and realistic problem. Since the characterization of the evolution of nonlinear uncertain dynamical state-delayed systems as a deterministic set of state equations is a formidable task, the stability analysis of such systems is generally difficult. To the best of authors’ knowledge, such problem has not been addressed so far in the literature.
This paper, therefore, deals with the problem of global asymptotic stability of a class of 2D discrete uncertain state-delayed systems described by the FMSLSS model employing generalized overflow nonlinearities. Parametric uncertainties involved in the system are assumed to be norm bounded. The paper is organized as follows. Section 2 presents a description of the system under consideration. In Section 3, the main result and its corollaries are presented in the form of linear matrix inequality (LMI) which can be effectively solved using Matlab LMI Toolbox [38, 39]. It is demonstrated in Section 4 that several previously reported results are recovered from the presented approach as special cases. Numerical examples highlighting the usefulness of the presented criteria are discussed in Section 5.
2. System Description
The following notations are used throughout the paper:: set of real matrices,: set of real vectors,: set of integers,: identity matrix of appropriate dimension,: null matrix or null vector of appropriate dimension,: transpose of the matrix (or vector) ,: is positive definite symmetric matrix,: is negative definite symmetric matrix,: any vector or matrix norm,diag : diagonal matrix with diagonal elements ,max : maximum value of scalars and .
The 2D discrete uncertain state-delayed system to be studied presently is described by the FMSLSS model [33] employing generalized overflow arithmetic. The system under consideration is given bywhere are horizontal and vertical coordinates, respectively; is the local state vector; , , , are the known real constant matrices; are the unknown real matrices representing parametric uncertainties in the state matrices; , are constant positive integers representing delays along vertical direction and horizontal direction, respectively. The generalized overflow characteristic is given by where With the appropriate selection of and, (2a) characterizes the common types of overflow arithmetics employed in applications such as saturation (), zeroing (), triangular (), and two’s complement ().
The uncertain matrix is defined in the norm-bounded form [13–16, 21–24, 27–29] as where are known real constant matrices with appropriate dimensions and is an unknown real matrix satisfying
In the uncertainty structure given by (3a)-(3b), the matrices characterize how the uncertain parameters in enter the state matrices. The matrix can always be restricted as (3b) by appropriately selecting . In other words, there is no loss of generality in choosing as in (3b).
It is assumed [32] that the system has a finite set of initial conditions, that is, there exist two positive integers and such that
Equations (1a)–(4) represent a class of 2D discrete dynamical systems which include 2D discrete systems implemented in a finite register length, 2D digital control systems with overflow nonlinearities, models of various physical phenomena (e.g., compartmental systems, single carriageway traffic flow [40], grid-based wireless sensor networks modeling [6], etc.), and various dynamical processes represented by the Darboux equation [41–43].
3. Main Result and Its Corollaries
Consider two ranges for
These two ranges together constitute (2b). In what follows, a criterion applicable to (5) and a different criterion applicable to (6) for the global asymptotic stability of the system described by (1a)-(1b), (2a), (3a)-(3b), and (4) are presented.
Before presenting our main result, we recall the following lemma.
Lemma 1 (see Xie et al. [44], Boyd et al. [38]). Letand be real matrices of appropriate dimensions with satisfying then for all if and only if there exists a positive scalar such that
Next, suppose is a matrix characterized bywhere it is understood that, for , corresponds to a scalar. Thus, corresponding to , the matrix takes the formwhere .
Now, we have the following result.
Theorem 1. The null solution of the system characterized by (1a)-(1b), (2a), (3a)-(3b)–(5) is globally asymptotically stable if there exist positive scalars, , and positive definite symmetric matrices such that where and is defined by (9a)–(9c).
Proof. Consider a 2D quadratic Lyapunov-Krasovskii function [24]
where
Application of (12a)–(12c) to system (1a)-(1b) yields
It follows from (1a), (12a)–(12c), and (13)
Now consider the quantity “” given by
It is clear that, owing to (2a), (5), and (9c), is nonnegative. (14) can be rearranged as
whereObserve that, if, then From (13) and condition , it follows that
where the equality sign holds only when . Let denote the set defined by
For any nonnegative integer max , it follows from (19) and the initial condition (4) that
where the equality sign holds only when
In the above derivation, the fact that , , , , and the positive definiteness of the function have been used.
From (21), we obtain
Consequently,
Thus condition is a sufficient condition for the global asymptotic stability of the system described by (1a)-(1b), (2a), (3a)-(3b)–(5). Using (3a) the condition can be rewritten in the following form:
where
By Lemma 1, (25a)–(25d) is equivalent to
where . The equivalence of (26) and (11a)–(11d) follows trivially from Schur complement. This completes the proof of Theorem 1.
Remark 1. The matrix inequality (11a)–(11d) is linear in the unknown parameters, and . Therefore, it can be solved using the Matlab LMI Toolbox [38, 39].
Next, we prove the following result.
Theorem 2. The null solution of the system characterized by (1a)-(1b), (2a), (3a)-(3b), (4), and (6) is globally asymptotically stable if there exist positive definite symmetric matrices , an positive definite diagonal matrix , and a positive scalar , satisfying where
Proof. Consider the quantity given by
Observe that, for , is nonnegative [10]. Equation (14) takes the form where
Note that, if, then for . Thus condition is a sufficient condition for the global asymptotic stability of system (1a)-(1b), (2a), (3a)-(3b), (4), and (6). By the Schur complement, the condition is equivalent to
where denotes .
Applying (3a), (30) can be rewritten in the following form:
whereBy Lemma 1, (31a)–(31d) is equivalent to
where . Using Schur complement, (32) can equivalently be expressed as (27a)-(27b). The rest of the proof is similar to that of Theorem 1.
Next, consider a class of nonlinearities characterized by
Clearly, the quantity defined by (28) is nonnegative under the restriction (33a)-(33b). Therefore, using the steps of the proof of Theorem 2, one arrives at the following.
Theorem 3. The null solution of the system described by (1a)-(1b), (3a)-(3b), (4), and (33a)-(33b) is globally asymptotically stable provided there exist positive definite symmetric matrices , , , , an positive definite diagonal matrix , and a positive scalar such that (27a)-(27b) holds true.
Remark 2. Observe that, for triangular overflow arithmetic, Theorem 3 will always lead to less conservative conditions than Theorem 2. As far as triangular overflow arithmetic is concerned, one is required to choose in (33a)-(33b), rather than . By contrast, (2a)-(2b) is overly restrictive for the characterization of triangular overflow arithmetic. As a matter of fact, (2a)-(2b) fails to make any distinction between two’s complement and triangular nonlinearities.
Next, consider the system (1a)-(1b), (3a)-(3b), and (4) in absence of state delay; therefore, the system becomes
Now, as a consequence of the presented approach, we have the following corollaries.
Corollary 1. The null solution of the system described by (34a)–(34d), (2a), and (5) is globally asymptotically stable if there exist positive scalars , and positive definite symmetric matrices andsuch that where is defined by (9a)–(9c).
Proof. Pertaining to the system described by (34a)–(34d), (2a), and (5), choosing and , and in (11a)–(11d), we obtain (35). This completes the proof of Corollary 1.
The following result can easily be obtained as a direct consequence of Theorem 2.
Corollary 2. The null solution of the system described by (34a)–(34d), (2a), and (6) is globally asymptotically stable if there exist scalar > 0, positive definite symmetric matrices , , and an positive definite diagonal matrix satisfying
Proof. For the system described by (34a)–(34d), (2a), and (6), substituting and and in (27a)-(27b), we obtain (36). This completes the proof of Corollary 2.
Remark 3. In [13], a criterion (see [13, Theorem 5]) for the global asymptotic stability of 2D discrete uncertain systems described by the FMSLSS model employing (2a) and (5) has been established. However, as indicated in [13, Remark 4], [13, Theorem 5] is computationally demanding. On the other hand, (35) is linear in the unknown parameters , , ,, and, hence, is computationally tractable [38, 39]. Further, pertaining to the 2D system without state-delay, Corollary 2 is identical to [13, Theorem 4]. Therefore, the present work may be treated as an extension of [13] from the case of nonlinear uncertain discrete systems to a general class of systems in the simultaneous presence of nonlinearity, uncertainty, and state delays.
4. Comparison
In this section, we will compare the main results of this paper with the results stated in [14, 15, 21, 24].
Theorem 4 (Du and Xie [15]). The null solution of the 2D system described by (34a)–(34d), employing saturation nonlinearities (i.e., (2a)-(2b) with ) is globally asymptotically stable if there exist positive definite symmetric matrices and a positive definite diagonal matrix , and a positive scalar such that
Pertaining to the saturation nonlinearity [21, Theorem 2] leads to the following result.
Theorem 5 (Dey and Kar [21]). The null solution of the system described by (34a)–(34d) employing saturation nonlinearities (i.e., (2a)-(2b) with ) is globally asymptotically stable if there exist a positive scalar and positive definite symmetric matrices, such that
Theorem 6 (Singh [14]). Suppose there exists a matrix , positive definite diagonal matrices , , positive definite symmetric matrices and , a scalar > 0 satisfyingwhere Then the null solution of the system (34a)–(34d) using saturation nonlinearities (i.e., (2a)-(2b) with ) is globally asymptotically stable.
Theorem 7 (Paszke et al. [24]). Consider the system (1a)–(4) with no nonlinearity and no uncertainty; then system becomes The null solution of the system described by (40) is globally asymptotically stable if there exist n×n positive definite symmetric matrices , and such that
Now, we have the following propositions.
Proposition 1. Corollary 1 implies Theorem 4.
Proof. With and (saturation nonlinearities), matrix in Corollary 1 reduces to a positive definite diagonal matrix and, consequently, Corollary 1 becomes Theorem 4. Thus, Theorem 4 is recovered from Corollary 1 as a special case.
Proposition 2. Corollary 1 implies Theorem 5.
Proof. It is not difficult to show that, with (35) reduces to (38b). Condition (42) along with (9a)–(9c) and yieldswhere . Using (43a)-(43b), we obtain Thus, condition (42) leads to (38a).
Proposition 3. Corollary 1 implies Theorem 6.
Proof. Using Schur complement, (39a) can equivalently be expressed as
Now, consider the case where . In this case, the matrix in Corollary 1 corresponds to a positive scalar. On the other hand, for , the matrices and in Theorem 6 reduce to a scalar equal to zero and a positive scalar, respectively. Now, from (35) and (45), it is clear that, for , Corollary 1 is equivalent to Theorem 6.
Next, consider the case where . In this case, let us choose the positive scalars and given by (9c) aswhere . In view of (46a)-(46b) and , one obtains from (9a)–(9c) With (47a)-(47b), the matrix in Corollary 1 reduces to () and, consequently, (35) reduces to (45). Thus, Theorem 6 is recovered from Corollary 1 as a special case.
Remark 4. Pertaining to saturation nonlinearities, the matrix in Corollary 1 is more general than the matrix () in Theorem 6. For a given row, the off-diagonal elements of the matrix () are equal. However, for , such restrictions are not required for the matrix in Corollary 1. In other words, Corollary 1 offers a greater flexibility than Theorem 6 for testing the saturation overflow stability of system (34a)–(34d).
Remark 5. To establish global asymptotic stability via Theorem 6, one is required to test (39a) (possibly) for all possible combinations of ’s belonging to (1, −1, 0). For an th-order system, the number of such possible combinations of ’s is . Therefore, (39a)–(39d) is computationally more complex than (35).
Proposition 4. Theorem 1 implies Theorem 7.
Proof. The system described by (1a)–(4) employing “no nonlinearity” and “no uncertainty” implies and . In this case, (14) can be rearranged as which satisfies if (41) holds true and only when . Thus, Theorem 7 is recovered from Theorem 1 as a special case.
5. Illustrative Examples
To illustrate the effectiveness of the presented results, we now consider the following examples.
Example 1. Consider the 2D discrete system represented by (1a)-(1b), (2a), (3a)-(3b), (4) withUsing the Matlab LMI Toolbox [38, 39], it has been found that LMI (11a)–(11d) is feasible for the following values of unknown parameters.
Thus, according to Theorem 1, the system under consideration is globally asymptotically stable.
Figure 1 shows the trajectories of the two state variables for the present example with
(a)
(b)
Likewise, the trajectory trace of the present system has been carried out for a number of arbitrary (randomly generated) initial conditions and it supports the fact (which has been arrived at via Theorem 1) that the system is asymptotically stable.
Example 2. Consider the 2D discrete system (34a)–(34d), (2a) where and are given by (49a) and Using the Matlab LMI Toolbox [38, 39], it turns out that [15, 21] fail to determine the global asymptotic stability of the present system. On the other hand, LMI (35) is feasible for the following values of unknown parameters. Therefore, Corollary 1 affirms the global asymptotic stability for the present system.
It may be noted that, in the absence of state-delay, the system of Example 1 reduces to that of Example 2. Whereas [15, 21] fail to determine the global asymptotic stability of the system of Example 2, the present approach succeeds to establish the global asymptotic stability of the same system even in the presence of state-delay.
6. Conclusion
Sufficient conditions in terms of LMIs are established for the global asymptotic stability of a class of 2D discrete uncertain state-delayed systems using generalized overflow nonlinearities. The presented approach turns out to be a generalization over the results reported in [14, 15, 21, 24]. Pertaining to the saturation nonlinearities, the presented criteria turn out to be less restrictive than previously reported criteria [14, 15, 21].
Acknowledgments
The authors wish to thank Professor Derong Liu and the anonymous reviewers for their constructive comments and suggestions.