Wirtinger-like Lyapunov–Krasovskii functionals for discrete-time delay systems

Wirtinger-like Lyapunov–Krasovskii functionals for discrete-time delay systems Abstract Time-dependent Lyapunov functionals appeared to be very efficient for sampled-data systems. Recently, new Lyapunov functionals were constructed for sampled-data control in the presence of a constant input delay. The construction of these functionals was based on Wirtinger’s inequality leading to simplified and efficient stability conditions in terms of linear matrix inequalities. In this article, we extend the latter results to the discrete-time sampled-data systems. We show that the proposed approach is less conservative on some examples with a lower number of decision variables. 1. Introduction Sampled-data systems have been studied extensively over the past decades (see e.g. Chen & Francis, 1995; Fridman et al., 2004; Mirkin, 2007; Naghshtabrizi et al., 2008; Fujioka, 2009 and the references therein). Modeling of continuous-time systems with digital control in the form of continuous-time systems with time-varying delay (Mikheev et al., 1988) and the extension of Krasovskii method to systems with fast varying delays (without any constraints on the delay derivative as in Shaked (2003) and to discontinuous delays (Fridman et al., 2004) have allowed the development of the time-delay approach to sampled-data and to network-based control (see Section 7 of Fridman, 2014 for details). Till Fridman (2010), the conventional time-independent Lyapunov functionals $$V(x_t,\dot x_t)$$ for systems with fast-varying delays were applied to sampled-data systems (Fridman et al., 2004). These functionals did not take advantage of the sawtooth evolution of the delays induced by sampled-and-hold. The latter drawback was removed in Fridman (2010) and Seuret (2012), where time-dependent Lyapunov functionals (inspired by Naghshtabrizi et al., 2008) were constructed for sampled-data systems. A different time-dependent Lyapunov functional was suggested in Liu & Fridman (2012) which is based on Wirtinger’s inequality (see for instance Kammler, 2007, Liu et al., 2010): Let $$z(t):(a,b)\to \mathbb R^n$$ be absolutely continuous with $$\dot z\in L_2[a,b]$$ and $$z(a)=0$$. Then for any $$n\times n$$ matrix $$W>0$$ Wirtinger’s inequality holds: ∫abz˙T(ξ)Wz˙(ξ)dξ≥π24(b−a)2∫abzT(ξ)Wz(ξ)dξ. The Wirtinger-based linear matrix inequality (LMI) is a single LMI with fewer decision variables than the LMIs of Fridman (2010) and Seuret (2012). More important, differently from the Lyapunov functionals of Fridman (2010) and Seuret (2012), the extension of the Wirtinger-based Lyapunov functionals to a more general sampled-data system in the presence of a constant input/output delay leads to efficient stability conditions (see e.g. Liu & Fridman, 2012). In this article, we aim at extending the results of Liu & Fridman (2012) to discrete-time sampled-data systems. Unlike the continuous-time case, the discrete-time formulation has surprisingly attracted only few attention in the literature even if the formulation represents an efficient way to model the dynamics of discrete-time systems subject to control packet losses. The problem of packet losses indeed appears in many applications of networked control systems (see for instance Hespanha et al., 2007, Zampieri, 2008. As in the continuous-time case, the Wirtinger-based Lyapunov functionals essentially reduce the numerical complexity of the resulting LMIs leading in some examples to less restrictive conditions. Similarly to the continuous-time case, discrete-time sampled-data can be seen as a discrete-time system subject to a particular time-varying delay, for which there exist many stability conditions (see e.g. Gao & Chen, 2007, Shao & Han, 2011, Liu & Fridman, 2012). However, such approaches do not account accurately the particularities of the sawtooth delay. In the continuous-time case, the analysis of this class of functionals is made possible by considering that the functionals do not grow at the sampling instants. A translation of such analysis in the discrete-time framework is not easy and requires a dedicated analysis. The article is organized as follows. Section 2 describes the problem formulation. Section 3 shows some preliminary summation inequalities including a Wirtinger’s and Jensen’s inequality as well as a recent summation inequality that includes the Jensen’s inequality as a consequence. This last inequality is the counter part of the Wirtinger-based inequality provided in Seuret & Gouaisbaut (2013). Section 4 presents the main results on the stability analysis of discrete-time sampled-data systems. Section 6 shows the efficiency of the proposed method on some examples. Finally Section 7 draws some conclusions. Notations: Throughout the article, $$\mathbb Z$$ ($$\mathbb N$$) denotes the set of (positive) integers, $$\mathbb R^n$$ the $$n$$-dimensional Euclidean space with vector norm $$|\cdot|$$, $$\mathbb R^{n\times m}$$ the set of all $$n\times m$$ real matrices. For any symmetric matrix $$P \in \mathbb R^{n\times n}$$, the notation $$P> 0$$ (or $$P<0$$) means that $$P$$ is positive (or negative) definite. The set $$\mathbb S_{n}^+$$ refers to the set of symmetric positive definite matrices. For any matrices $$A$$, $$B$$ in $$\mathbb R^{n\times n}$$, the notation $$diag(A,B)$$ denotes the block diagonal matrix $\left[\begin{smallmatrix} A&0\\ {\ast}& B \end{smallmatrix}\right]$ . For any square matrix, $$\mathrm{He}(A)$$ stands for $$A+A^T$$. Along the paper, for any real number $$a<b$$, the notation $$[a,\ b]_ \mathbb Z$$ denotes $$[a,\ b]\cap \mathbb Z$$. The same notations will also hold for open intervals. 2 Problem formulation Consider a linear discrete-time time-delay system of the form: {x(t+1)=Ax(t)+Adx(tk−h)∀k∈[ki,ki+1)Zx(θ)=x0(θ)∀θ∈[−h, 0],  (2.1) where $$x(t)\in\mathbb{R}^{n}$$ is the state vector, $$x_0$$ is the initial condition and $$A$$, $$A_d\in\mathbb{R}^{n\times n}$$ are constant matrices. The delay $$h\in \mathbb N$$ is assumed to be constant and known and the sequence of integers $$\{t_k\}_{k\in \mathbb N}\subset\mathbb N$$ satisfies ∀k∈Ntk+1−tk∈[1, Tmax]Z,limk→+∞tk=+∞. (2.2) Following the principles of the input delay approach for discrete-time systems, we can define an input delay function $$\tau$$ given by τ(t)=t−tk,∀t∈[tk,tk+1)Z,τ(t+1)={τ(t)+1ift<tk+10ift=tk+1.  (2.3) The system (2.1) can then be rewritten as a discrete-time system subject to a time-varying delay. The delay function can be seen as the discrete-time version of the sawtooth delay function considered in Fridman et al. (2004) for continuous-time sampled-data systems. {x(t+1)=Ax(t)+Adx(t−h−τ(t)) ∀t∈[tk,tk+1)Zx(θ)=x0(θ), ∀θ∈[−h, 0]Z,  (2.4) where the delay $$h$$ is constant and where the sampling delay $$\tau(t)$$ is given in (2.3). In this article, we aim at providing stability conditions for this peculiar class of systems using a method based on discontinuous Lyapunov functionals. This article can be seen as the discrete-time counterpart of the recent article (Liu & Fridman, 2012). To this end, we will provide a stability analysis of such class of systems, where novel Wirtinger-based terms are added to ‘nominal’ Lyapunov functionals for the stability analysis of the discrete-time systems with the constant delay $$h$$. 3 Preliminaries on summation inequalities 3.1 Discrete-time Wirtinger inequality Wirtinger inequalities are integral inequalities issued from the Fourier analysis. The continuous-time versions of this inequality have already shown their potential for the stability analysis of partial differential equation (Fridman & Orlov, 2009), sampled-data systems (Liu & Fridman, 2012) or time-delay systems (Seuret & Gouaisbaut, 2013). In this article, we aim at showing that this class of inequalities also serves for the stability analysis of discrete-time systems. Indeed a discrete-time version of these inequalities have been extended to the discrete-time framework. It is stated in the following lemma taken from Ky Fan et al. (1955). Lemma 3.1 For a given $$N\in\mathbb N_{\geq0}$$, consider a sequence of $$N$$ real scalars $$x_0,x_1,\dots,x_N$$ such that $$x_0=0$$. Then, the following inequality holds ∑i=0N−1(xi−xi+1)2≥λN2∑i=0N−1xi2, (3.1) where $$\lambda_N =2sin\left(\pi/(2(2N+1))\right)$$. A straightforward corollary of this lemma is provided for $$n$$-dimensional sequences $$z$$ and is stated below. Corollary 3.1 For a given $$N\in\mathbb N_{\geq0}$$, consider a sequence of $$N$$ real $$n$$-dimensional vectors $$z_0,z_2,\dots,z_N$$ such that $$z_0=0$$. Then, the following inequality holds, for any symmetric positive definite matrix $$W\in \mathbb S^n_+$$. ∑i=0N−1(zi−zi+1)TW(zi−zi+1)≥λN2∑i=0N−1ziTWzi, (3.2) where $$\lambda_N =2sin\left(\pi/(2(2N+1))\right)$$. Proof. Since $$W>0$$, there exists an orthogonal matrix $U=\left[\begin{array}{ccccc} U_1^T&U_2^T&\dots&U_n^T\end{array}\right]^T$ and a positive definite diagonal matrix $$\it {\Delta}=diag(\it {\Delta}_1,\it {\Delta}_2,...\it {\Delta}_n)$$ such that $$W=U^T\it {\Delta} U$$. It holds λN2∑i=0N−1ziTWzi =λN2∑i=0N−1∑j=1n(Ujzi)TΔj(Ujzi) =∑j=1nΔjλN2∑i=0N−1(Ujzi)2. (3.3) Following the same procedure, we also have ∑i=0N−1(zi−zi+1)TW(zi−zi+1)=∑j=1nΔj∑i=0N−1(Uj(zi−zi+1))2. (3.4) Since the vector $$z_0=0$$, the scalar $$U_jz_0$$ is zero for all $$j=1,\dots, n$$. Hence the Wirtinger inequality in Lemma 3.1 ensures that ∑i=0N−1(Uj(zi−zi+1))2≥λN2∑i=0N−1(Ujzi)2,j=1,…,n. Finally computing the sum over $$j=1,\dots,n$$ of the previous inequality weighted by $$\it {\Delta}_j$$ leads to ∑j=1nΔj∑i=0N−1(Uj(zi−zi+1))2≥∑j=1nΔjλN2∑i=0N−1(Ujzi)2 which is equivalent to (3.2) thanks to (3.3) and (3.4). □ 3.2 Summation inequalities In this section, two summation inequalities are recalled. The first one is the Jensen inequality and is stated here. Lemma 3.2 For a given symmetric positive definite matrix $$Z\in \mathbb S^n_+$$, any sequence of discrete-time variable $$x$$ in $$[-h,0]_\mathbb Z \rightarrow \mathbb R^n$$, where $$h\geq 1$$, the following inequality holds: ∑i=−h+10yT(i)Zy(i)≥1hΘ0TZΘ0, (3.5) where $$y(i)=x(i)-x(i-1)$$ and $$\Theta_0=x(0)-x(-h).$$ The second lemma is a recent inequality proposed in Seuret et al. (2015) that extends the Wirtinger-based integral inequality (see Seuret & Gouaisbaut, 2013) to the discrete-time case. Lemma 3.3 For a given symmetric positive definite matrix $$Z\in \mathbb S^n_+$$, any sequence of discrete-time variable $$x$$ in $$[-h,0]_ \mathbb Z \rightarrow \mathbb R^n$$, where $$h\geq 1$$, the inequality ∑i=−h+10yT(i)Zy(i)≥1h[Θ0Θ1]T[Z003(h+1h−1)Z][Θ0Θ1] (3.6) holds where y(i) =x(i)−x(i−1),Θ0 =x(0)−x(−h),Θ1 =x(0)+x(−h)−2h+1∑i=−h0x(i). Proof. The proof is provided in Seuret et al. (2015) and is therefore omitted. □ Remark 3.1 The inequality provided in Lemma 3.3 implies ∑i=−h+10yT(i)Zy(i)≥1hΘ0TZΘ0, which is exactly the Jensen summation inequality. Therefore, Lemma 3.3 is less conservative than the celebrated Jensen inequality since a positive quantity is added in the right-hand side of the inequalities. 4 Stability analysis 4.1 Wirtinger-based functional In this section, we aim at proposing a new functional to deal with the discrete-time sampled-data system (2.4) by an appropriate use of the discrete time Wirtinger inequality resumed in Lemma 3.1. This contribution is proposed in the following lemma. Lemma 4.2 Consider the following Lyapunov functional, for a given matrix $$W\in \mathbb S^n_+$$, a given $$k\in\mathbb N$$ and for all $$t\in [t_k,\ t_{k+1})_ \mathbb Z$$ VW(xt)=∑i=tk−ht−1yT(i)Wy(i)−λT2σ(t,xt), (4.1) where y(i) =x(i+1)−x(i)∀i∈NλT =2sin(π2(2Tmax+1))σ(t,xt) ={∑i=tkt−1ν(i)Wν(i)t∈[tk+1, tk+1−1]0t=tk, ν(i) =x(i−h)−x(tk−h),i∈[tk, tk+1−1]. (4.2) Then, the forward difference of the functional $$V_W$$ satisfies the inequality ΔVW(xt)≤(x(t+1)−x(t))TW(x(t+1)−x(t))−λT2νT(t)Wν(t) (4.3) holds, for all $$t\in[t_k,t_{k+1})_ \mathbb Z$$, and for any sampling satisfying (2.2). Proof. For a given $$k\in \mathbb N$$, consider first $$t\in[t_k, t_{k+1}-2]_ \mathbb Z$$. Then the computation of $$\it {\Delta} V_W$$ straightforwardly leads, for all $$t\in[t_k,t_{k+1}-2]_ \mathbb Z$$, to ΔVW(xt) =∑i=tk−htyT(i)Wy(i)−∑i=tk−ht−1yT(i)Wy(i)−λT2(σ(t+1,xt+1)−σ(t,xt)) =yT(t)Wy(t)−λT2(σ(t+1,xt+1)−σ(t,xt)). (4.4) From the definition of $$\sigma(t,x_t)$$, it is easy to see that, if $$t\neq t_k$$, we have σ(t+1,xt+1)−σ(t,xt)=∑i=tktkν(i)Wν(i)−0=νT(t)Wν(t) and if $$t=t_k$$ σ(tk+1,xtk+1)−σ(tk,xtk)=∑i=tktν(i)Wν(i)−∑i=tkt−1ν(i)Wν(i)=νT(tk)Wν(tk). This ensures that, for all $$t\in[t_k, t_{k+1}-2]_\mathbb Z$$, the following equality holds ΔVW(xt)=(x(t+1)−x(t))TW(x(t+1)−x(t))−λT2νT(t)Wν(t). (4.5) Consider now the remaining case $$t=t_{k+1}-1$$. The computation of $$\it {\Delta} V_W$$ leads to ΔVW(xt) =∑i=tk+1−htk+1−1yT(i)Wy(i)−∑i=tk−htk+1−2yT(i)Wy(i)−λT2(0−∑i=tk−htk+1−h−2νT(i)Wν(i)) =yT(t)Wy(t)−λT2νT(t)Wν(t)−ψ, where ψ=∑i=tk−htk+1−h−1yT(i)Wy(i)−λT2∑i=tk−htk+1−h−1νT(i)Wν(i). By noting that ν(tk)=0,∀k∈Nν(i+1)−ν(i)=y(i)∀i∈[tk, tk+1)Z,tk+1−tk≤Tmax∀k∈N, the assumptions of the Wirtinger inequality in Corollary 3.1 are satisfied, which guarantees that $$\psi\geq0$$. It thus holds that, for $$t=t_{k+1}-1$$ ΔVW(xt)≤(x(t+1)−x(t))TW(x(t+1)−x(t))−λT2νT(t)Wν(t). (4.6) Then, combining (4.4) and (4.6) proves the result. □ Note that inequality (4.6) is actually an equality when $$t\neq t_{k+1}-1$$ and is an inequality only when $$t=t_{k+1}-1$$. The computation of this inequality only relies on the computation of the forward increment of functional $$V_W$$ and the use of the Wirtinger inequality. The objective in the remainder of this article is to include this functional in the stability analysis of discrete sampled-data systems. Next, we will propose two stability theorems which rely on the use of the Jensen inequality and on Lemma 3.3. 4.2 Jensen-based theorem The following theorem holds Theorem 4.1 For given $$h$$ and $$T_{max}$$ in $$\mathbb N$$, assume that there exist $$n\times n$$ matrices $$P$$, $$Q$$, $$Z$$ and $$W\in\mathbb S^n_+$$ such that the LMI condition Π(Tmax,h)<0 (4.7) holds where Π(Tmax,h) =[Q00∗−Q0∗∗−λT2W]+M1TPM1−M2TPM2+M0T(W+h2Z)M0−M3TZM3,M0 =[A−IAd−Ad],M1 = [AAd−Ad],M2 =[I00],M3 = [I−I0],λT =2sin(π2(2Tmax+1)). (4.8) Then system (2.4) is asymptotically stable for the constant delay $$h$$ and any aperiodic sampling satisfying (2.2). Proof. Consider the functional V(xt)=V1(xt)+VW(xt) where the functional $$V_W$$ have been defined previously. The fucntional $$V_1$$ is built to assess stability of the delayed term $$A_1x(t-h)$$ which appears in equation (2.4). Indeed a classical functional for discrete-time delay system is given by V1(xt)=xT(t)Px(t)+∑i=t−ht−1xT(i)Qx(i)+h∑i=−h+10∑j=t+i−1t−1yT(j)Zy(j), (4.9) where $$y(i)=x(i+1)-x(i)$$. Define the increment of the Lyapunov–Krasovskii functional as follows $$\it {\Delta} V(x_t)=V(x_{t+1})-V(x_t)$$. From Lemma 4.1, we show that ΔV(xt) =xT(t+1)Px(t+1)−xT(t)Px(t)+xT(t)Qx(t)−xT(t−h)Qx(t−h) +h2(x(t+1)−x(t))TZ(x(t+1)−x(t))+(x(t+1)−x(t))TW(x(t+1)−x(t)) −λT2νT(t)Wν(t)−h∑j=t−ht−1yT(j)Zy(j). Applying Jensen’s inequality to the summation term ensures that ΔV(xt) ≤xT(t+1)Px(t+1)−xT(t)Px(t)+xT(t)Qx(t)−xT(t−h)Qx(t−h) +h2(x(t+1)−x(t))TZ(x(t+1)−x(t))−(x(t)−x(t−h))TZ(x(t)−x(t−h)) +(x(t+1)−x(t))TW(x(t+1)−x(t))−λT2νT(t)Wν(t). It follows from the previous calculations that ΔV(xt)≤[x(t)x(t−h)ν(t)]TΠ(Tmax,h)[x(t)x(t−h)ν(t)]. Then asymptotic stability results from the condition $$\it {\Pi}(T,h)<0$$, which concludes the proof. □ 4.3 Improved stability Theorem As it was noticed in Seuret et al. (2015), the conservatism induces by the Jensen inequality can be notably reduced by considering the refined summation inequality provided in Lemma 3.3. The resulting analysis leads to the following theorem. Theorem 4.2 For given $$h$$ and $$T_{max}$$ in $$\mathbb N$$, assume that there exist a $$2n\times 2n$$ matrix $$P>0$$ and $$n\times n$$ matrices $$Q>0$$, $$Z>0$$ and $$W>0$$ such that the LMI condition Φ(Tmax,h)<0 (4.10) holds where Φ(Tmax,h) =[Q0000−Q000000000−λT2W]+N1TPN1−N2TPN2+N0T(W+h2Z)N0−N3TZ~N3N0 =[A−IAd0−Ad]N1 =[AAd0−Ad0−I(h+1)I0]N2 =[I000−I0(h+1)I0]N3 =[I−I00II−2I0]Z~ =[Z0 03h+1h−1Z]λT =2sin(π2(2Tmax+1)). (4.11) Then system (2.4) is asymptotically stable for the constant delay $$h$$ and any aperiodic sampling satisfying (2.2). Proof. Consider the functional V(xt)=V2(xt)+VW(xt), where we use the same definition for the functional $$V_W$$ as in Theorem 4.1. In order to fully take advantages of the summation inequality provided in Lemma 3.3, we select the following functional $$V_2$$ given by V2(xt) =[x(t)∑i=t−ht−1x(i)]TP[x(t)∑i=t−ht−1x(i)]+∑i=t−ht−1xT(i)Qx(i)+h∑i=−h+10∑j=t+i−1t−1yT(j )Zy(j), (4.12) where $$y(i)=x(i+1)-x(i)$$. This functional has been build according to the method provided in Seuret et al. (2015). The forward difference of the Lyapunov–Krasovskii functional yields ΔV(xt) =[x(t+1)∑i=t−h+1tx(i)]TP[x(t+1)∑i=t−h+1tx(i)]−[x(t)∑i=t−ht−1x(i)]TP[x(t)∑i=t−ht−1x(i)] +xT(t)Qx(t)−xT(t−h)Qx(t−h)+h2(x(t+1)−x(t))TZ(x(t+1)−x(t))−h∑j=t−ht−1yT(j)Zy(j). Define the $$\xi(t)=\frac{1}{h+1}\sum\limits_{i=t- h}^{t}x(i)$$ and applying the summation provided in Lemma 3.3 to the last term ensures that ΔV(xt) ≤[x(t+1)(h+1)ξ(t)−x(t−h)]TP[x(t+1)(h+1)ξ(t)−x(t)] −[x(t)(h+1)ξ(t)−x(t)]TP[x(t)(h+1)ξ(t)−x(t)] +h2(x(t+1)−x(t))TZ(x(t+1)−x(t))−(x(t)−x(t−h))TZ(x(t)−x(t−h))−3(h+1h−1)(x(t)+x(t−h)−2ξ(t))TZ(x(t)+x(t−h)−2ξ(t)). It follows from the previous calculations that ΔV(xt)≤[x(t)x(t−h)ξ(t)ν(t)]TΦ(Tmax,h)[x(t)x(t−h)ξ(t)ν(t)]. Then asymptotic stability results from the condition $$\Phi(T_{max},h)<0$$, which concludes the proof. □ Remark 4.1 In the previous developments, we only focussed on the case of discrete-time delay systems with a single delay and a single sampling. However, the methodology can be extended to the case of multiple delays and multiple sampling by introducing additional functional terms. For the sake of consistency, this problem is not addressed in this article. 4.4 Comparison with approaches from the literature In this article, we consider functionals of the form V(xt)=V1(xt)+VW(xt), where the functional $$V_1$$ (or $$V_2)$$ aims at assessing the stability of system (2.1) without sampling and where the functional $$V_W(x_t)$$ aims at ensuring the robustness with respect to the sampling. In Seuret et al. (2015), the functional can also be split into two parts where the first one is again the same $$V_1$$ (or $$V_2$$) but the second part is related to the time-varying delay case. Therefore the conditions provided in Seuret et al. (2015) only address the stability of the system driven by x(t+1)=Ax(t)+Adx(t−h(k)), where the delay $$h(k)$$ can take any values between $$h$$ and $$h+T$$, without respecting the constraint imposed in this article on the sawtooth form of the delay. Therefore, the conditions provided in this article and the one provided in Seuret et al. (2015) does not treat the same problem. However, it is correct to say that the conditions of Seuret et al. (2015) guarantee stability of the sampled-data system (2.1) but also to a larger class of delay systems. The idea of this article is to propose a dedicated construction of the functional to cope with the stability analysis of sampled and delayed closed-loop system driven by (2.1). 4.5 Example 1 Consider the continuous time sampled-data system linear driven for all $$\mathbf{t}\in[kT_0, (k+1)T_0)_ \mathbb Z$$ by x˙(t)=Acx(t)+BcdKx((k−h)T0), (4.13) where $$\mathbf{t}$$ represents the continuous time and where Ac=[010−0.1],Bcd=[00−0.375−1.15],K=[−0.375−1.15] and $$T_0$$ is the discretization period, $$h$$ is the delay and $$k$$ is a positive integer. The associated discretized system is given by discrete-time systems with delay given in (2.4) with the matrices A=eAcT0,Ad=∫0T0eAc(T0−s)dsBcdK. The stability conditions provided in this article and from the literature are tested on this system for several values of the discretization period $$T_0=0.1$$ and $$T_0=0.01$$. The results and a comparison with existing results from the literature are presented in Tables 2 and 3. Table 2 Results derived from various theorems showing the maximal admissible constant delay $$h$$ for several values of the sampling period for system with $$T_0=0.1$$ $$T_{max}$$ 1 2 3 4 5 6 7 8 9 10 11 Liu Zhang (2012) 9 8 7 6 5 4 3 2 1 - - Seuret et al. (2015) 9 8 7 6 5 4 3 2 1 - - Theorem 4.1 9 8 7 6 5 4 4 3 2 1 1 Theorem 4.2 9 8 7 6 6 5 4 3 2 1 1 $$T_{max}$$ 1 2 3 4 5 6 7 8 9 10 11 Liu Zhang (2012) 9 8 7 6 5 4 3 2 1 - - Seuret et al. (2015) 9 8 7 6 5 4 3 2 1 - - Theorem 4.1 9 8 7 6 5 4 4 3 2 1 1 Theorem 4.2 9 8 7 6 6 5 4 3 2 1 1 Table 2 Results derived from various theorems showing the maximal admissible constant delay $$h$$ for several values of the sampling period for system with $$T_0=0.1$$ $$T_{max}$$ 1 2 3 4 5 6 7 8 9 10 11 Liu Zhang (2012) 9 8 7 6 5 4 3 2 1 - - Seuret et al. (2015) 9 8 7 6 5 4 3 2 1 - - Theorem 4.1 9 8 7 6 5 4 4 3 2 1 1 Theorem 4.2 9 8 7 6 6 5 4 3 2 1 1 $$T_{max}$$ 1 2 3 4 5 6 7 8 9 10 11 Liu Zhang (2012) 9 8 7 6 5 4 3 2 1 - - Seuret et al. (2015) 9 8 7 6 5 4 3 2 1 - - Theorem 4.1 9 8 7 6 5 4 4 3 2 1 1 Theorem 4.2 9 8 7 6 6 5 4 3 2 1 1 Table 3 Results derived from various theorems showing the maximal admissible constant delay $$h$$ for several values of the sampling period for system with $$T_0=0.01$$ $$T_{max}$$ 5 10 25 50 75 100 125 133 Liu Zhang (2012) 101 96 81 56 31 4 — — Seuret et al. (2015) 109 103 86 58 33 7 — — Theorem 4.1 102 98 85 65 45 25 7 1 Theorem 4.2 110 106 91 69 47 26 7 1 $$T_{max}$$ 5 10 25 50 75 100 125 133 Liu Zhang (2012) 101 96 81 56 31 4 — — Seuret et al. (2015) 109 103 86 58 33 7 — — Theorem 4.1 102 98 85 65 45 25 7 1 Theorem 4.2 110 106 91 69 47 26 7 1 Table 3 Results derived from various theorems showing the maximal admissible constant delay $$h$$ for several values of the sampling period for system with $$T_0=0.01$$ $$T_{max}$$ 5 10 25 50 75 100 125 133 Liu Zhang (2012) 101 96 81 56 31 4 — — Seuret et al. (2015) 109 103 86 58 33 7 — — Theorem 4.1 102 98 85 65 45 25 7 1 Theorem 4.2 110 106 91 69 47 26 7 1 $$T_{max}$$ 5 10 25 50 75 100 125 133 Liu Zhang (2012) 101 96 81 56 31 4 — — Seuret et al. (2015) 109 103 86 58 33 7 — — Theorem 4.1 102 98 85 65 45 25 7 1 Theorem 4.2 110 106 91 69 47 26 7 1 The stability conditions from Liu & Zhang (2012) and Seuret et al. (2015) address the problem of stability analysis of discrete-time systems subject to an unknown time-varying delay but which belongs to the interval $$[h,\ h+T]$$. To the best of our knowledge, these results are the most efficient conditions for the stability analysis of discrete systems with interval time-varying delays The sawtooth delay addressed in this article is only a particular case of this more general class of time-varying delays. Table 1 compares the complexity of Theorem 4.1 and 4.2 to theorems taken from Liu & Zhang (2012) and Seuret et al. (2015). Tables 2 and 3 show that our theorems essentially reduce the complexity of the conditions provided in Liu & Zhang (2012) and Seuret et al. (2015) leading to less conservative results. Table 1 Complexity of the LMI conditions tested in the example Theorems Complexity Liu Zhang (2012) $$(3.5n^2+2.5n)\times 4n$$ Seuret et al. (2015) $$(10n^2+3.5n)\times 12n$$ Theorem 4.1 $$(2n^2+2n)\times 3n$$ Theorem 4.2 $$(3.5n^2+2.5n)\times 4n$$ Theorems Complexity Liu Zhang (2012) $$(3.5n^2+2.5n)\times 4n$$ Seuret et al. (2015) $$(10n^2+3.5n)\times 12n$$ Theorem 4.1 $$(2n^2+2n)\times 3n$$ Theorem 4.2 $$(3.5n^2+2.5n)\times 4n$$ Table 1 Complexity of the LMI conditions tested in the example Theorems Complexity Liu Zhang (2012) $$(3.5n^2+2.5n)\times 4n$$ Seuret et al. (2015) $$(10n^2+3.5n)\times 12n$$ Theorem 4.1 $$(2n^2+2n)\times 3n$$ Theorem 4.2 $$(3.5n^2+2.5n)\times 4n$$ Theorems Complexity Liu Zhang (2012) $$(3.5n^2+2.5n)\times 4n$$ Seuret et al. (2015) $$(10n^2+3.5n)\times 12n$$ Theorem 4.1 $$(2n^2+2n)\times 3n$$ Theorem 4.2 $$(3.5n^2+2.5n)\times 4n$$ Finally, Fig. 1 depicts the solutions of system (4.13) taken with a sampling period of the continuous time systems $$T_0=0.1$$ and the input delay $$h=6$$. From Table 2, the maximal length between two successive control update$$t_{k+1}-t_k$$ is upper bounded by $$6$$. Figure 1 shows two simulations of a periodic and an aperiodic implementation of the control input $$u=Kx(t_k-h)$$ where it can be seen that the solutions of the systems remain stable in both cases. It is also worth noting that the system remains stable with the periodic implementation up to $$T=12$$, which means that the stability conditions resulting from the functional term related to the Wirtinger inequality are still conservative even if they already improve the condition issued from the time-varying delay case, which, again, allows assessing stability of the system with a larger class of delay functions than sawtooth delays. Fig. 1. View largeDownload slide Example 1. Simulation results representing the state $$x$$ (top), the timer $$\tau=t-t_k$$ (middle), and the control inputs $$u=Kx(t_k-h)$$ (bottom), of System (4.13) with $$T_0=0.1$$ with $$h=5$$, $$T_{max}=6$$ and aperiodic (a) and periodic (b) samplings. Fig. 1. View largeDownload slide Example 1. Simulation results representing the state $$x$$ (top), the timer $$\tau=t-t_k$$ (middle), and the control inputs $$u=Kx(t_k-h)$$ (bottom), of System (4.13) with $$T_0=0.1$$ with $$h=5$$, $$T_{max}=6$$ and aperiodic (a) and periodic (b) samplings. 5 Model reduction and predictor control 5.1 Definitions Consider the linear discrete-time system driven by {x(t+1)=Ax(t)+Bu(tk−h)∀t∈[tk,tk+1)∩Nx(θ)=x0(θ)∀θ∈[−h, 0],  (5.1) where $$x(t)\in\mathbb{R}^{n}$$ is the state vector, $$x_0$$ is the initial condition and $$A$$, $$A_d\in\mathbb{R}^{n\times n}$$ are constant matrices. with the initial condition $$x_0$$. The prediction-based approach consists in considering the following control law u(t) =Kx¯(t+h)x¯(t+h) =Ahx(t)+∑i=0h−1Ah−i−1Bu(t−h+i). (5.2) In this formulation, the vector $$\bar x$$ is the $$h-$$step ahead state prediction. Lemma 5.1 The closed-looped system (5.1) with the control scheme (5.2) can be expressed as z(t+1)=(A+BK)z(t)−AhBK[z(t−h)−z(tk−h)]. (5.3) Proof. Define the new state $$z(t)=\bar x(t+h),$$ where $$\bar x$$ is given in (5.2). It holds z(t+1) =Ah(Ax(t)+Bu(tk−h))+∑i=0h−1Ah−i−1Bu(t+1−h+i) =Ah(Ax(t)+Bu(tk−h))+A∑i=0h−1Ah−i−1Bu(t−h+i)+Bu(t)−AhBu(t−h) =A[Ahx(t)+∑i=0h−1Ah−i−1Bu(t−h+i)]+Bu(t)+AhBu(tk−h))−AhBu(t−h) =Az(t)+Bu(t)+AhB[u(tk−h))−u(t−h)]. Finally, reinjecting the definition of $$u=Kz$$ in the previous equation leads to the result. □ 5.2 Stability conditions The following theorem holds Theorem 5.1 For a given controller gain $$K$$ and a given delay $$h$$, assume that there exists two $$n\times n$$ matrices $$P>0$$ and $$W>0$$ such that the LMI condition [−P0(A+BK−I)TW(A+BK)TP∗−λT2W−(AhBK)TW−(AhBK)TP∗∗−W0∗∗∗−P]<0 (5.4) holds where $$\lambda_T=2sin\left(\frac{\pi}{2(2T+1)}\right)$$. Proof. Consider the functional Vt(xt)=xT(t)Px(t)+VW(xt), where the functionals $$V$$ and $$V_W$$ have been defined previously. It follows from the previous calculations that ΔVt(xt)≤[x(t)ν(t)]TΨ2[x(t)ν(t)] where Ψ2=[−P0∗−λT2W]+[(A+BK)T−I−(AhBK)T]W[(A+BK)T−I−(AhBK)T]T+[(A+BK)T−(AhBK)T]P[(A+BK)T−(AhBK)T]T. (5.5) Then asymptotic stability results from the condition $${\it \Psi}_2<0$$, which is equivalent to (5.4) by application of the Schur complement. □ 5.3 Example 2 Consider the linear discrete-time systems with delay given in (2.4) with the matrices taken from Gao et al. (2004) A=[0.800.050.9],Ad=[−0.10−0.2−0.1]. The results are presented in Table 4. One can see from this example that the robustness of the predictor control is reinforced for large delays. This means that the more the delay, the more the maximal allowable sampling period is obtained. A possible interpretation for such behavior is that the matrix $$A_0$$ is Schur stable. Therefore the matrix $$A_0^h$$ in the LMI conditions becomes smaller when the delay increases so that the contribution of $$\lambda_T$$ becomes sufficiently great to ensure robustness with respect to the sampling period. Table 4 Evolution of the maximal admissible sampling period $$T$$ for several values of the input delay $$h$$ T 1 2 3 4 5 6 7 8 9 10 Number of decision variables Theorem 5.1 3 7 10 13 15 16 17 19 20 21 $$3.5n^2+2.5n$$ T 1 2 3 4 5 6 7 8 9 10 Number of decision variables Theorem 5.1 3 7 10 13 15 16 17 19 20 21 $$3.5n^2+2.5n$$ Table 4 Evolution of the maximal admissible sampling period $$T$$ for several values of the input delay $$h$$ T 1 2 3 4 5 6 7 8 9 10 Number of decision variables Theorem 5.1 3 7 10 13 15 16 17 19 20 21 $$3.5n^2+2.5n$$ T 1 2 3 4 5 6 7 8 9 10 Number of decision variables Theorem 5.1 3 7 10 13 15 16 17 19 20 21 $$3.5n^2+2.5n$$ A counter part of this numerical results is that the performances of the closed loop systems may be affected. This means that increasing the delay $$T$$ for large values of $$h$$ would lead to power performances. In order to measure the performance degradation, one may look at $$\mathcal L_2$$ performance criteria or exponential stability criteria with guaranteed decay rate. For the latter solution, one would need to lightly modify the Wirtinger-based functional to account for exponential stability. This can be achieved following the idea developed for the continuous-time case in Lemma 1 of Selivanov & Fridman (2016). 6 Conclusions This article addresses the stability analysis of discrete time sampled-data systems. The approach developed in this article can be interpreted as the counterpart of the recent result on continuous-time systems from Liu & Fridman (2012). Two stability theorems have been provided and are tested on a simple example showing the efficiency of the method. Funding This work was supported by the ANR project SCIDiS contract number 15-CE23-0014. References Chen T. & Francis B. A. ( 1995 ) Optimal Sampled-Data Control Systems . London : Springer . Fan K. , Taussky O. & Todd J. ( 1955 ) Discrete analogs of inequalities of wirtinger. Monatshefte für Mathematik , 59 , 73 – 90 . Google Scholar CrossRef Search ADS Fridman E. ( 2010 ) A refined input delay approach to sampled-data control. Automatica , 46 , 421 – 427 . Google Scholar CrossRef Search ADS Fridman E. ( 2014 ) Introduction to Time-Delay Systems: Analysis and Control . Systems and Control: Foundations and Applications. Switzerland : Birkhäuser Basel . Fridman E. & Orlov Y. ( 2009 ) Exponential stability of linear distributed parameter systems with time-varying delays. Automatica , 45 , 194 – 201 . Google Scholar CrossRef Search ADS Fridman E. , Seuret A. & Richard J. P. ( 2004 ) Robust sampled-data stabilization of linear systems: an input delay approach. Automatica , 40 , 1441 – 1446 . Google Scholar CrossRef Search ADS Fridman E. & Shaked U. ( 2003 ) Delay dependent stability and $$H_{\infty}$$ control: constant and time-varying delays. Int J Contr , 76 , 48 – 60 . Google Scholar CrossRef Search ADS Fujioka H. ( 2009 ) Stability analysis of systems with aperiodic sample-and-hold devices. Automatica , 45 , 771 – 775 . Google Scholar CrossRef Search ADS Gao H. , Lam J. , Wang C. & Wang Y. ( 2004 ) Delay-dependent output-feedback stabilisation of discrete-time systems with time-varying state delay. IEE Contr Theor & Appl, IET , 151 , 691 – 698 . Google Scholar CrossRef Search ADS Gao H. & Chen T. ( 2007 ) New results on stability of discrete-time systems with time-varying state delay. IEEE Trans. on Autom. Cont. , 52 , 328 – 334 . Google Scholar CrossRef Search ADS Hespanha J. , Naghshtabrizi P. & Xu Y. ( 2007 ) A survey of recent results in networked control systems. P IEEE , 95 , 138 – 162 . Google Scholar CrossRef Search ADS Kammler W. D. ( 2007 ) A First Course in Fourier Analysis . Cambridge : Cambridge University Press . Liu K. & Fridman E. ( 2012 ) Wirtinger’s inequality and Lyapunov-based sampled-data stabilization. Automatica , 48 , 102 – 108 . Google Scholar CrossRef Search ADS Liu K. , Suplin V. & Fridman E. ( 2010 ) Stability of linear systems with general sawtooth delay. Special issue on Time-delay Systems in IMA J Math Control I , 27 , 419 – 436 . Liu J. & Zhang J. ( 2012 ) Note on stability of discrete-time time-varying delay systems. Contr Theor & Appl, IET , 6 , 335 – 339 . Google Scholar CrossRef Search ADS Mikheev Yu. , Sobolev V. & Fridman E. ( 1988 ) Asymptotic analysis of digital control systems. Automat Rem Contr , 49 , 1175 – 1180 . Mirkin L. ( 2007 ) Some remarks on the use of time-varying delay to model sample-and-hold circuits. IEEE Trans. on Autom. Cont. , 52 , 1109 – 1112 . Google Scholar CrossRef Search ADS Naghshtabrizi P. , Hespanha, J. & Teel A. ( 2008 ) Exponential stability of impulsive systems with application to uncertain sampled-data systems. Syst Contr Lett , 57 , 378 – 385 . Google Scholar CrossRef Search ADS Selivanov A. & Fridman E. ( 2016 ) Observer-based input-to-state stabilization of networked control systems with large uncertain delays. Automatica , 74 , 63 – 70 . Google Scholar CrossRef Search ADS Seuret A. ( 2012 ) A novel stability analysis of linear systems under asynchronous samplings. Automatica , 48 , 177 – 182 . Google Scholar CrossRef Search ADS Seuret A. & Gouaisbaut F. ( 2013 ) Wirtinger-based integral inequality: Application to time-delay systems. Automatica , 49 , 2860 – 2866 . Google Scholar CrossRef Search ADS Seuret A. , Gouaisbaut F. & Fridman E. ( 2015 ) Stability of discrete-time systems with time-varying delays via a novel summation inequality. IEEE T Automat Contr , 60 , 2740 – 2745 . Google Scholar CrossRef Search ADS Shao H. & Han Q.-L. ( 2011 ) New stability criteria for linear discrete-time systems with interval-like time-varying delays. IEEE Trans Automat Contr , 56 , 619 – 625 . Google Scholar CrossRef Search ADS Zampieri S. ( 2008 ) Trends in networked control systems. Proceedings of the 17th IFAC World Congress , Vol. 41, Seoul, Korea, pp. 2886 – 2894 . © The authors 2017. Published by Oxford University Press on behalf of the Institute of Mathematics and its Applications. All rights reserved.. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png IMA Journal of Mathematical Control and Information Oxford University Press

Wirtinger-like Lyapunov–Krasovskii functionals for discrete-time delay systems

Loading next page...
 
/lp/ou_press/wirtinger-like-lyapunov-krasovskii-functionals-for-discrete-time-delay-iafYesAHZS
Publisher
Oxford University Press
Copyright
© The authors 2017. Published by Oxford University Press on behalf of the Institute of Mathematics and its Applications. All rights reserved..
ISSN
0265-0754
eISSN
1471-6887
D.O.I.
10.1093/imamci/dnx002
Publisher site
See Article on Publisher Site

Abstract

Abstract Time-dependent Lyapunov functionals appeared to be very efficient for sampled-data systems. Recently, new Lyapunov functionals were constructed for sampled-data control in the presence of a constant input delay. The construction of these functionals was based on Wirtinger’s inequality leading to simplified and efficient stability conditions in terms of linear matrix inequalities. In this article, we extend the latter results to the discrete-time sampled-data systems. We show that the proposed approach is less conservative on some examples with a lower number of decision variables. 1. Introduction Sampled-data systems have been studied extensively over the past decades (see e.g. Chen & Francis, 1995; Fridman et al., 2004; Mirkin, 2007; Naghshtabrizi et al., 2008; Fujioka, 2009 and the references therein). Modeling of continuous-time systems with digital control in the form of continuous-time systems with time-varying delay (Mikheev et al., 1988) and the extension of Krasovskii method to systems with fast varying delays (without any constraints on the delay derivative as in Shaked (2003) and to discontinuous delays (Fridman et al., 2004) have allowed the development of the time-delay approach to sampled-data and to network-based control (see Section 7 of Fridman, 2014 for details). Till Fridman (2010), the conventional time-independent Lyapunov functionals $$V(x_t,\dot x_t)$$ for systems with fast-varying delays were applied to sampled-data systems (Fridman et al., 2004). These functionals did not take advantage of the sawtooth evolution of the delays induced by sampled-and-hold. The latter drawback was removed in Fridman (2010) and Seuret (2012), where time-dependent Lyapunov functionals (inspired by Naghshtabrizi et al., 2008) were constructed for sampled-data systems. A different time-dependent Lyapunov functional was suggested in Liu & Fridman (2012) which is based on Wirtinger’s inequality (see for instance Kammler, 2007, Liu et al., 2010): Let $$z(t):(a,b)\to \mathbb R^n$$ be absolutely continuous with $$\dot z\in L_2[a,b]$$ and $$z(a)=0$$. Then for any $$n\times n$$ matrix $$W>0$$ Wirtinger’s inequality holds: ∫abz˙T(ξ)Wz˙(ξ)dξ≥π24(b−a)2∫abzT(ξ)Wz(ξ)dξ. The Wirtinger-based linear matrix inequality (LMI) is a single LMI with fewer decision variables than the LMIs of Fridman (2010) and Seuret (2012). More important, differently from the Lyapunov functionals of Fridman (2010) and Seuret (2012), the extension of the Wirtinger-based Lyapunov functionals to a more general sampled-data system in the presence of a constant input/output delay leads to efficient stability conditions (see e.g. Liu & Fridman, 2012). In this article, we aim at extending the results of Liu & Fridman (2012) to discrete-time sampled-data systems. Unlike the continuous-time case, the discrete-time formulation has surprisingly attracted only few attention in the literature even if the formulation represents an efficient way to model the dynamics of discrete-time systems subject to control packet losses. The problem of packet losses indeed appears in many applications of networked control systems (see for instance Hespanha et al., 2007, Zampieri, 2008. As in the continuous-time case, the Wirtinger-based Lyapunov functionals essentially reduce the numerical complexity of the resulting LMIs leading in some examples to less restrictive conditions. Similarly to the continuous-time case, discrete-time sampled-data can be seen as a discrete-time system subject to a particular time-varying delay, for which there exist many stability conditions (see e.g. Gao & Chen, 2007, Shao & Han, 2011, Liu & Fridman, 2012). However, such approaches do not account accurately the particularities of the sawtooth delay. In the continuous-time case, the analysis of this class of functionals is made possible by considering that the functionals do not grow at the sampling instants. A translation of such analysis in the discrete-time framework is not easy and requires a dedicated analysis. The article is organized as follows. Section 2 describes the problem formulation. Section 3 shows some preliminary summation inequalities including a Wirtinger’s and Jensen’s inequality as well as a recent summation inequality that includes the Jensen’s inequality as a consequence. This last inequality is the counter part of the Wirtinger-based inequality provided in Seuret & Gouaisbaut (2013). Section 4 presents the main results on the stability analysis of discrete-time sampled-data systems. Section 6 shows the efficiency of the proposed method on some examples. Finally Section 7 draws some conclusions. Notations: Throughout the article, $$\mathbb Z$$ ($$\mathbb N$$) denotes the set of (positive) integers, $$\mathbb R^n$$ the $$n$$-dimensional Euclidean space with vector norm $$|\cdot|$$, $$\mathbb R^{n\times m}$$ the set of all $$n\times m$$ real matrices. For any symmetric matrix $$P \in \mathbb R^{n\times n}$$, the notation $$P> 0$$ (or $$P<0$$) means that $$P$$ is positive (or negative) definite. The set $$\mathbb S_{n}^+$$ refers to the set of symmetric positive definite matrices. For any matrices $$A$$, $$B$$ in $$\mathbb R^{n\times n}$$, the notation $$diag(A,B)$$ denotes the block diagonal matrix $\left[\begin{smallmatrix} A&0\\ {\ast}& B \end{smallmatrix}\right]$ . For any square matrix, $$\mathrm{He}(A)$$ stands for $$A+A^T$$. Along the paper, for any real number $$a<b$$, the notation $$[a,\ b]_ \mathbb Z$$ denotes $$[a,\ b]\cap \mathbb Z$$. The same notations will also hold for open intervals. 2 Problem formulation Consider a linear discrete-time time-delay system of the form: {x(t+1)=Ax(t)+Adx(tk−h)∀k∈[ki,ki+1)Zx(θ)=x0(θ)∀θ∈[−h, 0],  (2.1) where $$x(t)\in\mathbb{R}^{n}$$ is the state vector, $$x_0$$ is the initial condition and $$A$$, $$A_d\in\mathbb{R}^{n\times n}$$ are constant matrices. The delay $$h\in \mathbb N$$ is assumed to be constant and known and the sequence of integers $$\{t_k\}_{k\in \mathbb N}\subset\mathbb N$$ satisfies ∀k∈Ntk+1−tk∈[1, Tmax]Z,limk→+∞tk=+∞. (2.2) Following the principles of the input delay approach for discrete-time systems, we can define an input delay function $$\tau$$ given by τ(t)=t−tk,∀t∈[tk,tk+1)Z,τ(t+1)={τ(t)+1ift<tk+10ift=tk+1.  (2.3) The system (2.1) can then be rewritten as a discrete-time system subject to a time-varying delay. The delay function can be seen as the discrete-time version of the sawtooth delay function considered in Fridman et al. (2004) for continuous-time sampled-data systems. {x(t+1)=Ax(t)+Adx(t−h−τ(t)) ∀t∈[tk,tk+1)Zx(θ)=x0(θ), ∀θ∈[−h, 0]Z,  (2.4) where the delay $$h$$ is constant and where the sampling delay $$\tau(t)$$ is given in (2.3). In this article, we aim at providing stability conditions for this peculiar class of systems using a method based on discontinuous Lyapunov functionals. This article can be seen as the discrete-time counterpart of the recent article (Liu & Fridman, 2012). To this end, we will provide a stability analysis of such class of systems, where novel Wirtinger-based terms are added to ‘nominal’ Lyapunov functionals for the stability analysis of the discrete-time systems with the constant delay $$h$$. 3 Preliminaries on summation inequalities 3.1 Discrete-time Wirtinger inequality Wirtinger inequalities are integral inequalities issued from the Fourier analysis. The continuous-time versions of this inequality have already shown their potential for the stability analysis of partial differential equation (Fridman & Orlov, 2009), sampled-data systems (Liu & Fridman, 2012) or time-delay systems (Seuret & Gouaisbaut, 2013). In this article, we aim at showing that this class of inequalities also serves for the stability analysis of discrete-time systems. Indeed a discrete-time version of these inequalities have been extended to the discrete-time framework. It is stated in the following lemma taken from Ky Fan et al. (1955). Lemma 3.1 For a given $$N\in\mathbb N_{\geq0}$$, consider a sequence of $$N$$ real scalars $$x_0,x_1,\dots,x_N$$ such that $$x_0=0$$. Then, the following inequality holds ∑i=0N−1(xi−xi+1)2≥λN2∑i=0N−1xi2, (3.1) where $$\lambda_N =2sin\left(\pi/(2(2N+1))\right)$$. A straightforward corollary of this lemma is provided for $$n$$-dimensional sequences $$z$$ and is stated below. Corollary 3.1 For a given $$N\in\mathbb N_{\geq0}$$, consider a sequence of $$N$$ real $$n$$-dimensional vectors $$z_0,z_2,\dots,z_N$$ such that $$z_0=0$$. Then, the following inequality holds, for any symmetric positive definite matrix $$W\in \mathbb S^n_+$$. ∑i=0N−1(zi−zi+1)TW(zi−zi+1)≥λN2∑i=0N−1ziTWzi, (3.2) where $$\lambda_N =2sin\left(\pi/(2(2N+1))\right)$$. Proof. Since $$W>0$$, there exists an orthogonal matrix $U=\left[\begin{array}{ccccc} U_1^T&U_2^T&\dots&U_n^T\end{array}\right]^T$ and a positive definite diagonal matrix $$\it {\Delta}=diag(\it {\Delta}_1,\it {\Delta}_2,...\it {\Delta}_n)$$ such that $$W=U^T\it {\Delta} U$$. It holds λN2∑i=0N−1ziTWzi =λN2∑i=0N−1∑j=1n(Ujzi)TΔj(Ujzi) =∑j=1nΔjλN2∑i=0N−1(Ujzi)2. (3.3) Following the same procedure, we also have ∑i=0N−1(zi−zi+1)TW(zi−zi+1)=∑j=1nΔj∑i=0N−1(Uj(zi−zi+1))2. (3.4) Since the vector $$z_0=0$$, the scalar $$U_jz_0$$ is zero for all $$j=1,\dots, n$$. Hence the Wirtinger inequality in Lemma 3.1 ensures that ∑i=0N−1(Uj(zi−zi+1))2≥λN2∑i=0N−1(Ujzi)2,j=1,…,n. Finally computing the sum over $$j=1,\dots,n$$ of the previous inequality weighted by $$\it {\Delta}_j$$ leads to ∑j=1nΔj∑i=0N−1(Uj(zi−zi+1))2≥∑j=1nΔjλN2∑i=0N−1(Ujzi)2 which is equivalent to (3.2) thanks to (3.3) and (3.4). □ 3.2 Summation inequalities In this section, two summation inequalities are recalled. The first one is the Jensen inequality and is stated here. Lemma 3.2 For a given symmetric positive definite matrix $$Z\in \mathbb S^n_+$$, any sequence of discrete-time variable $$x$$ in $$[-h,0]_\mathbb Z \rightarrow \mathbb R^n$$, where $$h\geq 1$$, the following inequality holds: ∑i=−h+10yT(i)Zy(i)≥1hΘ0TZΘ0, (3.5) where $$y(i)=x(i)-x(i-1)$$ and $$\Theta_0=x(0)-x(-h).$$ The second lemma is a recent inequality proposed in Seuret et al. (2015) that extends the Wirtinger-based integral inequality (see Seuret & Gouaisbaut, 2013) to the discrete-time case. Lemma 3.3 For a given symmetric positive definite matrix $$Z\in \mathbb S^n_+$$, any sequence of discrete-time variable $$x$$ in $$[-h,0]_ \mathbb Z \rightarrow \mathbb R^n$$, where $$h\geq 1$$, the inequality ∑i=−h+10yT(i)Zy(i)≥1h[Θ0Θ1]T[Z003(h+1h−1)Z][Θ0Θ1] (3.6) holds where y(i) =x(i)−x(i−1),Θ0 =x(0)−x(−h),Θ1 =x(0)+x(−h)−2h+1∑i=−h0x(i). Proof. The proof is provided in Seuret et al. (2015) and is therefore omitted. □ Remark 3.1 The inequality provided in Lemma 3.3 implies ∑i=−h+10yT(i)Zy(i)≥1hΘ0TZΘ0, which is exactly the Jensen summation inequality. Therefore, Lemma 3.3 is less conservative than the celebrated Jensen inequality since a positive quantity is added in the right-hand side of the inequalities. 4 Stability analysis 4.1 Wirtinger-based functional In this section, we aim at proposing a new functional to deal with the discrete-time sampled-data system (2.4) by an appropriate use of the discrete time Wirtinger inequality resumed in Lemma 3.1. This contribution is proposed in the following lemma. Lemma 4.2 Consider the following Lyapunov functional, for a given matrix $$W\in \mathbb S^n_+$$, a given $$k\in\mathbb N$$ and for all $$t\in [t_k,\ t_{k+1})_ \mathbb Z$$ VW(xt)=∑i=tk−ht−1yT(i)Wy(i)−λT2σ(t,xt), (4.1) where y(i) =x(i+1)−x(i)∀i∈NλT =2sin(π2(2Tmax+1))σ(t,xt) ={∑i=tkt−1ν(i)Wν(i)t∈[tk+1, tk+1−1]0t=tk, ν(i) =x(i−h)−x(tk−h),i∈[tk, tk+1−1]. (4.2) Then, the forward difference of the functional $$V_W$$ satisfies the inequality ΔVW(xt)≤(x(t+1)−x(t))TW(x(t+1)−x(t))−λT2νT(t)Wν(t) (4.3) holds, for all $$t\in[t_k,t_{k+1})_ \mathbb Z$$, and for any sampling satisfying (2.2). Proof. For a given $$k\in \mathbb N$$, consider first $$t\in[t_k, t_{k+1}-2]_ \mathbb Z$$. Then the computation of $$\it {\Delta} V_W$$ straightforwardly leads, for all $$t\in[t_k,t_{k+1}-2]_ \mathbb Z$$, to ΔVW(xt) =∑i=tk−htyT(i)Wy(i)−∑i=tk−ht−1yT(i)Wy(i)−λT2(σ(t+1,xt+1)−σ(t,xt)) =yT(t)Wy(t)−λT2(σ(t+1,xt+1)−σ(t,xt)). (4.4) From the definition of $$\sigma(t,x_t)$$, it is easy to see that, if $$t\neq t_k$$, we have σ(t+1,xt+1)−σ(t,xt)=∑i=tktkν(i)Wν(i)−0=νT(t)Wν(t) and if $$t=t_k$$ σ(tk+1,xtk+1)−σ(tk,xtk)=∑i=tktν(i)Wν(i)−∑i=tkt−1ν(i)Wν(i)=νT(tk)Wν(tk). This ensures that, for all $$t\in[t_k, t_{k+1}-2]_\mathbb Z$$, the following equality holds ΔVW(xt)=(x(t+1)−x(t))TW(x(t+1)−x(t))−λT2νT(t)Wν(t). (4.5) Consider now the remaining case $$t=t_{k+1}-1$$. The computation of $$\it {\Delta} V_W$$ leads to ΔVW(xt) =∑i=tk+1−htk+1−1yT(i)Wy(i)−∑i=tk−htk+1−2yT(i)Wy(i)−λT2(0−∑i=tk−htk+1−h−2νT(i)Wν(i)) =yT(t)Wy(t)−λT2νT(t)Wν(t)−ψ, where ψ=∑i=tk−htk+1−h−1yT(i)Wy(i)−λT2∑i=tk−htk+1−h−1νT(i)Wν(i). By noting that ν(tk)=0,∀k∈Nν(i+1)−ν(i)=y(i)∀i∈[tk, tk+1)Z,tk+1−tk≤Tmax∀k∈N, the assumptions of the Wirtinger inequality in Corollary 3.1 are satisfied, which guarantees that $$\psi\geq0$$. It thus holds that, for $$t=t_{k+1}-1$$ ΔVW(xt)≤(x(t+1)−x(t))TW(x(t+1)−x(t))−λT2νT(t)Wν(t). (4.6) Then, combining (4.4) and (4.6) proves the result. □ Note that inequality (4.6) is actually an equality when $$t\neq t_{k+1}-1$$ and is an inequality only when $$t=t_{k+1}-1$$. The computation of this inequality only relies on the computation of the forward increment of functional $$V_W$$ and the use of the Wirtinger inequality. The objective in the remainder of this article is to include this functional in the stability analysis of discrete sampled-data systems. Next, we will propose two stability theorems which rely on the use of the Jensen inequality and on Lemma 3.3. 4.2 Jensen-based theorem The following theorem holds Theorem 4.1 For given $$h$$ and $$T_{max}$$ in $$\mathbb N$$, assume that there exist $$n\times n$$ matrices $$P$$, $$Q$$, $$Z$$ and $$W\in\mathbb S^n_+$$ such that the LMI condition Π(Tmax,h)<0 (4.7) holds where Π(Tmax,h) =[Q00∗−Q0∗∗−λT2W]+M1TPM1−M2TPM2+M0T(W+h2Z)M0−M3TZM3,M0 =[A−IAd−Ad],M1 = [AAd−Ad],M2 =[I00],M3 = [I−I0],λT =2sin(π2(2Tmax+1)). (4.8) Then system (2.4) is asymptotically stable for the constant delay $$h$$ and any aperiodic sampling satisfying (2.2). Proof. Consider the functional V(xt)=V1(xt)+VW(xt) where the functional $$V_W$$ have been defined previously. The fucntional $$V_1$$ is built to assess stability of the delayed term $$A_1x(t-h)$$ which appears in equation (2.4). Indeed a classical functional for discrete-time delay system is given by V1(xt)=xT(t)Px(t)+∑i=t−ht−1xT(i)Qx(i)+h∑i=−h+10∑j=t+i−1t−1yT(j)Zy(j), (4.9) where $$y(i)=x(i+1)-x(i)$$. Define the increment of the Lyapunov–Krasovskii functional as follows $$\it {\Delta} V(x_t)=V(x_{t+1})-V(x_t)$$. From Lemma 4.1, we show that ΔV(xt) =xT(t+1)Px(t+1)−xT(t)Px(t)+xT(t)Qx(t)−xT(t−h)Qx(t−h) +h2(x(t+1)−x(t))TZ(x(t+1)−x(t))+(x(t+1)−x(t))TW(x(t+1)−x(t)) −λT2νT(t)Wν(t)−h∑j=t−ht−1yT(j)Zy(j). Applying Jensen’s inequality to the summation term ensures that ΔV(xt) ≤xT(t+1)Px(t+1)−xT(t)Px(t)+xT(t)Qx(t)−xT(t−h)Qx(t−h) +h2(x(t+1)−x(t))TZ(x(t+1)−x(t))−(x(t)−x(t−h))TZ(x(t)−x(t−h)) +(x(t+1)−x(t))TW(x(t+1)−x(t))−λT2νT(t)Wν(t). It follows from the previous calculations that ΔV(xt)≤[x(t)x(t−h)ν(t)]TΠ(Tmax,h)[x(t)x(t−h)ν(t)]. Then asymptotic stability results from the condition $$\it {\Pi}(T,h)<0$$, which concludes the proof. □ 4.3 Improved stability Theorem As it was noticed in Seuret et al. (2015), the conservatism induces by the Jensen inequality can be notably reduced by considering the refined summation inequality provided in Lemma 3.3. The resulting analysis leads to the following theorem. Theorem 4.2 For given $$h$$ and $$T_{max}$$ in $$\mathbb N$$, assume that there exist a $$2n\times 2n$$ matrix $$P>0$$ and $$n\times n$$ matrices $$Q>0$$, $$Z>0$$ and $$W>0$$ such that the LMI condition Φ(Tmax,h)<0 (4.10) holds where Φ(Tmax,h) =[Q0000−Q000000000−λT2W]+N1TPN1−N2TPN2+N0T(W+h2Z)N0−N3TZ~N3N0 =[A−IAd0−Ad]N1 =[AAd0−Ad0−I(h+1)I0]N2 =[I000−I0(h+1)I0]N3 =[I−I00II−2I0]Z~ =[Z0 03h+1h−1Z]λT =2sin(π2(2Tmax+1)). (4.11) Then system (2.4) is asymptotically stable for the constant delay $$h$$ and any aperiodic sampling satisfying (2.2). Proof. Consider the functional V(xt)=V2(xt)+VW(xt), where we use the same definition for the functional $$V_W$$ as in Theorem 4.1. In order to fully take advantages of the summation inequality provided in Lemma 3.3, we select the following functional $$V_2$$ given by V2(xt) =[x(t)∑i=t−ht−1x(i)]TP[x(t)∑i=t−ht−1x(i)]+∑i=t−ht−1xT(i)Qx(i)+h∑i=−h+10∑j=t+i−1t−1yT(j )Zy(j), (4.12) where $$y(i)=x(i+1)-x(i)$$. This functional has been build according to the method provided in Seuret et al. (2015). The forward difference of the Lyapunov–Krasovskii functional yields ΔV(xt) =[x(t+1)∑i=t−h+1tx(i)]TP[x(t+1)∑i=t−h+1tx(i)]−[x(t)∑i=t−ht−1x(i)]TP[x(t)∑i=t−ht−1x(i)] +xT(t)Qx(t)−xT(t−h)Qx(t−h)+h2(x(t+1)−x(t))TZ(x(t+1)−x(t))−h∑j=t−ht−1yT(j)Zy(j). Define the $$\xi(t)=\frac{1}{h+1}\sum\limits_{i=t- h}^{t}x(i)$$ and applying the summation provided in Lemma 3.3 to the last term ensures that ΔV(xt) ≤[x(t+1)(h+1)ξ(t)−x(t−h)]TP[x(t+1)(h+1)ξ(t)−x(t)] −[x(t)(h+1)ξ(t)−x(t)]TP[x(t)(h+1)ξ(t)−x(t)] +h2(x(t+1)−x(t))TZ(x(t+1)−x(t))−(x(t)−x(t−h))TZ(x(t)−x(t−h))−3(h+1h−1)(x(t)+x(t−h)−2ξ(t))TZ(x(t)+x(t−h)−2ξ(t)). It follows from the previous calculations that ΔV(xt)≤[x(t)x(t−h)ξ(t)ν(t)]TΦ(Tmax,h)[x(t)x(t−h)ξ(t)ν(t)]. Then asymptotic stability results from the condition $$\Phi(T_{max},h)<0$$, which concludes the proof. □ Remark 4.1 In the previous developments, we only focussed on the case of discrete-time delay systems with a single delay and a single sampling. However, the methodology can be extended to the case of multiple delays and multiple sampling by introducing additional functional terms. For the sake of consistency, this problem is not addressed in this article. 4.4 Comparison with approaches from the literature In this article, we consider functionals of the form V(xt)=V1(xt)+VW(xt), where the functional $$V_1$$ (or $$V_2)$$ aims at assessing the stability of system (2.1) without sampling and where the functional $$V_W(x_t)$$ aims at ensuring the robustness with respect to the sampling. In Seuret et al. (2015), the functional can also be split into two parts where the first one is again the same $$V_1$$ (or $$V_2$$) but the second part is related to the time-varying delay case. Therefore the conditions provided in Seuret et al. (2015) only address the stability of the system driven by x(t+1)=Ax(t)+Adx(t−h(k)), where the delay $$h(k)$$ can take any values between $$h$$ and $$h+T$$, without respecting the constraint imposed in this article on the sawtooth form of the delay. Therefore, the conditions provided in this article and the one provided in Seuret et al. (2015) does not treat the same problem. However, it is correct to say that the conditions of Seuret et al. (2015) guarantee stability of the sampled-data system (2.1) but also to a larger class of delay systems. The idea of this article is to propose a dedicated construction of the functional to cope with the stability analysis of sampled and delayed closed-loop system driven by (2.1). 4.5 Example 1 Consider the continuous time sampled-data system linear driven for all $$\mathbf{t}\in[kT_0, (k+1)T_0)_ \mathbb Z$$ by x˙(t)=Acx(t)+BcdKx((k−h)T0), (4.13) where $$\mathbf{t}$$ represents the continuous time and where Ac=[010−0.1],Bcd=[00−0.375−1.15],K=[−0.375−1.15] and $$T_0$$ is the discretization period, $$h$$ is the delay and $$k$$ is a positive integer. The associated discretized system is given by discrete-time systems with delay given in (2.4) with the matrices A=eAcT0,Ad=∫0T0eAc(T0−s)dsBcdK. The stability conditions provided in this article and from the literature are tested on this system for several values of the discretization period $$T_0=0.1$$ and $$T_0=0.01$$. The results and a comparison with existing results from the literature are presented in Tables 2 and 3. Table 2 Results derived from various theorems showing the maximal admissible constant delay $$h$$ for several values of the sampling period for system with $$T_0=0.1$$ $$T_{max}$$ 1 2 3 4 5 6 7 8 9 10 11 Liu Zhang (2012) 9 8 7 6 5 4 3 2 1 - - Seuret et al. (2015) 9 8 7 6 5 4 3 2 1 - - Theorem 4.1 9 8 7 6 5 4 4 3 2 1 1 Theorem 4.2 9 8 7 6 6 5 4 3 2 1 1 $$T_{max}$$ 1 2 3 4 5 6 7 8 9 10 11 Liu Zhang (2012) 9 8 7 6 5 4 3 2 1 - - Seuret et al. (2015) 9 8 7 6 5 4 3 2 1 - - Theorem 4.1 9 8 7 6 5 4 4 3 2 1 1 Theorem 4.2 9 8 7 6 6 5 4 3 2 1 1 Table 2 Results derived from various theorems showing the maximal admissible constant delay $$h$$ for several values of the sampling period for system with $$T_0=0.1$$ $$T_{max}$$ 1 2 3 4 5 6 7 8 9 10 11 Liu Zhang (2012) 9 8 7 6 5 4 3 2 1 - - Seuret et al. (2015) 9 8 7 6 5 4 3 2 1 - - Theorem 4.1 9 8 7 6 5 4 4 3 2 1 1 Theorem 4.2 9 8 7 6 6 5 4 3 2 1 1 $$T_{max}$$ 1 2 3 4 5 6 7 8 9 10 11 Liu Zhang (2012) 9 8 7 6 5 4 3 2 1 - - Seuret et al. (2015) 9 8 7 6 5 4 3 2 1 - - Theorem 4.1 9 8 7 6 5 4 4 3 2 1 1 Theorem 4.2 9 8 7 6 6 5 4 3 2 1 1 Table 3 Results derived from various theorems showing the maximal admissible constant delay $$h$$ for several values of the sampling period for system with $$T_0=0.01$$ $$T_{max}$$ 5 10 25 50 75 100 125 133 Liu Zhang (2012) 101 96 81 56 31 4 — — Seuret et al. (2015) 109 103 86 58 33 7 — — Theorem 4.1 102 98 85 65 45 25 7 1 Theorem 4.2 110 106 91 69 47 26 7 1 $$T_{max}$$ 5 10 25 50 75 100 125 133 Liu Zhang (2012) 101 96 81 56 31 4 — — Seuret et al. (2015) 109 103 86 58 33 7 — — Theorem 4.1 102 98 85 65 45 25 7 1 Theorem 4.2 110 106 91 69 47 26 7 1 Table 3 Results derived from various theorems showing the maximal admissible constant delay $$h$$ for several values of the sampling period for system with $$T_0=0.01$$ $$T_{max}$$ 5 10 25 50 75 100 125 133 Liu Zhang (2012) 101 96 81 56 31 4 — — Seuret et al. (2015) 109 103 86 58 33 7 — — Theorem 4.1 102 98 85 65 45 25 7 1 Theorem 4.2 110 106 91 69 47 26 7 1 $$T_{max}$$ 5 10 25 50 75 100 125 133 Liu Zhang (2012) 101 96 81 56 31 4 — — Seuret et al. (2015) 109 103 86 58 33 7 — — Theorem 4.1 102 98 85 65 45 25 7 1 Theorem 4.2 110 106 91 69 47 26 7 1 The stability conditions from Liu & Zhang (2012) and Seuret et al. (2015) address the problem of stability analysis of discrete-time systems subject to an unknown time-varying delay but which belongs to the interval $$[h,\ h+T]$$. To the best of our knowledge, these results are the most efficient conditions for the stability analysis of discrete systems with interval time-varying delays The sawtooth delay addressed in this article is only a particular case of this more general class of time-varying delays. Table 1 compares the complexity of Theorem 4.1 and 4.2 to theorems taken from Liu & Zhang (2012) and Seuret et al. (2015). Tables 2 and 3 show that our theorems essentially reduce the complexity of the conditions provided in Liu & Zhang (2012) and Seuret et al. (2015) leading to less conservative results. Table 1 Complexity of the LMI conditions tested in the example Theorems Complexity Liu Zhang (2012) $$(3.5n^2+2.5n)\times 4n$$ Seuret et al. (2015) $$(10n^2+3.5n)\times 12n$$ Theorem 4.1 $$(2n^2+2n)\times 3n$$ Theorem 4.2 $$(3.5n^2+2.5n)\times 4n$$ Theorems Complexity Liu Zhang (2012) $$(3.5n^2+2.5n)\times 4n$$ Seuret et al. (2015) $$(10n^2+3.5n)\times 12n$$ Theorem 4.1 $$(2n^2+2n)\times 3n$$ Theorem 4.2 $$(3.5n^2+2.5n)\times 4n$$ Table 1 Complexity of the LMI conditions tested in the example Theorems Complexity Liu Zhang (2012) $$(3.5n^2+2.5n)\times 4n$$ Seuret et al. (2015) $$(10n^2+3.5n)\times 12n$$ Theorem 4.1 $$(2n^2+2n)\times 3n$$ Theorem 4.2 $$(3.5n^2+2.5n)\times 4n$$ Theorems Complexity Liu Zhang (2012) $$(3.5n^2+2.5n)\times 4n$$ Seuret et al. (2015) $$(10n^2+3.5n)\times 12n$$ Theorem 4.1 $$(2n^2+2n)\times 3n$$ Theorem 4.2 $$(3.5n^2+2.5n)\times 4n$$ Finally, Fig. 1 depicts the solutions of system (4.13) taken with a sampling period of the continuous time systems $$T_0=0.1$$ and the input delay $$h=6$$. From Table 2, the maximal length between two successive control update$$t_{k+1}-t_k$$ is upper bounded by $$6$$. Figure 1 shows two simulations of a periodic and an aperiodic implementation of the control input $$u=Kx(t_k-h)$$ where it can be seen that the solutions of the systems remain stable in both cases. It is also worth noting that the system remains stable with the periodic implementation up to $$T=12$$, which means that the stability conditions resulting from the functional term related to the Wirtinger inequality are still conservative even if they already improve the condition issued from the time-varying delay case, which, again, allows assessing stability of the system with a larger class of delay functions than sawtooth delays. Fig. 1. View largeDownload slide Example 1. Simulation results representing the state $$x$$ (top), the timer $$\tau=t-t_k$$ (middle), and the control inputs $$u=Kx(t_k-h)$$ (bottom), of System (4.13) with $$T_0=0.1$$ with $$h=5$$, $$T_{max}=6$$ and aperiodic (a) and periodic (b) samplings. Fig. 1. View largeDownload slide Example 1. Simulation results representing the state $$x$$ (top), the timer $$\tau=t-t_k$$ (middle), and the control inputs $$u=Kx(t_k-h)$$ (bottom), of System (4.13) with $$T_0=0.1$$ with $$h=5$$, $$T_{max}=6$$ and aperiodic (a) and periodic (b) samplings. 5 Model reduction and predictor control 5.1 Definitions Consider the linear discrete-time system driven by {x(t+1)=Ax(t)+Bu(tk−h)∀t∈[tk,tk+1)∩Nx(θ)=x0(θ)∀θ∈[−h, 0],  (5.1) where $$x(t)\in\mathbb{R}^{n}$$ is the state vector, $$x_0$$ is the initial condition and $$A$$, $$A_d\in\mathbb{R}^{n\times n}$$ are constant matrices. with the initial condition $$x_0$$. The prediction-based approach consists in considering the following control law u(t) =Kx¯(t+h)x¯(t+h) =Ahx(t)+∑i=0h−1Ah−i−1Bu(t−h+i). (5.2) In this formulation, the vector $$\bar x$$ is the $$h-$$step ahead state prediction. Lemma 5.1 The closed-looped system (5.1) with the control scheme (5.2) can be expressed as z(t+1)=(A+BK)z(t)−AhBK[z(t−h)−z(tk−h)]. (5.3) Proof. Define the new state $$z(t)=\bar x(t+h),$$ where $$\bar x$$ is given in (5.2). It holds z(t+1) =Ah(Ax(t)+Bu(tk−h))+∑i=0h−1Ah−i−1Bu(t+1−h+i) =Ah(Ax(t)+Bu(tk−h))+A∑i=0h−1Ah−i−1Bu(t−h+i)+Bu(t)−AhBu(t−h) =A[Ahx(t)+∑i=0h−1Ah−i−1Bu(t−h+i)]+Bu(t)+AhBu(tk−h))−AhBu(t−h) =Az(t)+Bu(t)+AhB[u(tk−h))−u(t−h)]. Finally, reinjecting the definition of $$u=Kz$$ in the previous equation leads to the result. □ 5.2 Stability conditions The following theorem holds Theorem 5.1 For a given controller gain $$K$$ and a given delay $$h$$, assume that there exists two $$n\times n$$ matrices $$P>0$$ and $$W>0$$ such that the LMI condition [−P0(A+BK−I)TW(A+BK)TP∗−λT2W−(AhBK)TW−(AhBK)TP∗∗−W0∗∗∗−P]<0 (5.4) holds where $$\lambda_T=2sin\left(\frac{\pi}{2(2T+1)}\right)$$. Proof. Consider the functional Vt(xt)=xT(t)Px(t)+VW(xt), where the functionals $$V$$ and $$V_W$$ have been defined previously. It follows from the previous calculations that ΔVt(xt)≤[x(t)ν(t)]TΨ2[x(t)ν(t)] where Ψ2=[−P0∗−λT2W]+[(A+BK)T−I−(AhBK)T]W[(A+BK)T−I−(AhBK)T]T+[(A+BK)T−(AhBK)T]P[(A+BK)T−(AhBK)T]T. (5.5) Then asymptotic stability results from the condition $${\it \Psi}_2<0$$, which is equivalent to (5.4) by application of the Schur complement. □ 5.3 Example 2 Consider the linear discrete-time systems with delay given in (2.4) with the matrices taken from Gao et al. (2004) A=[0.800.050.9],Ad=[−0.10−0.2−0.1]. The results are presented in Table 4. One can see from this example that the robustness of the predictor control is reinforced for large delays. This means that the more the delay, the more the maximal allowable sampling period is obtained. A possible interpretation for such behavior is that the matrix $$A_0$$ is Schur stable. Therefore the matrix $$A_0^h$$ in the LMI conditions becomes smaller when the delay increases so that the contribution of $$\lambda_T$$ becomes sufficiently great to ensure robustness with respect to the sampling period. Table 4 Evolution of the maximal admissible sampling period $$T$$ for several values of the input delay $$h$$ T 1 2 3 4 5 6 7 8 9 10 Number of decision variables Theorem 5.1 3 7 10 13 15 16 17 19 20 21 $$3.5n^2+2.5n$$ T 1 2 3 4 5 6 7 8 9 10 Number of decision variables Theorem 5.1 3 7 10 13 15 16 17 19 20 21 $$3.5n^2+2.5n$$ Table 4 Evolution of the maximal admissible sampling period $$T$$ for several values of the input delay $$h$$ T 1 2 3 4 5 6 7 8 9 10 Number of decision variables Theorem 5.1 3 7 10 13 15 16 17 19 20 21 $$3.5n^2+2.5n$$ T 1 2 3 4 5 6 7 8 9 10 Number of decision variables Theorem 5.1 3 7 10 13 15 16 17 19 20 21 $$3.5n^2+2.5n$$ A counter part of this numerical results is that the performances of the closed loop systems may be affected. This means that increasing the delay $$T$$ for large values of $$h$$ would lead to power performances. In order to measure the performance degradation, one may look at $$\mathcal L_2$$ performance criteria or exponential stability criteria with guaranteed decay rate. For the latter solution, one would need to lightly modify the Wirtinger-based functional to account for exponential stability. This can be achieved following the idea developed for the continuous-time case in Lemma 1 of Selivanov & Fridman (2016). 6 Conclusions This article addresses the stability analysis of discrete time sampled-data systems. The approach developed in this article can be interpreted as the counterpart of the recent result on continuous-time systems from Liu & Fridman (2012). Two stability theorems have been provided and are tested on a simple example showing the efficiency of the method. Funding This work was supported by the ANR project SCIDiS contract number 15-CE23-0014. References Chen T. & Francis B. A. ( 1995 ) Optimal Sampled-Data Control Systems . London : Springer . Fan K. , Taussky O. & Todd J. ( 1955 ) Discrete analogs of inequalities of wirtinger. Monatshefte für Mathematik , 59 , 73 – 90 . Google Scholar CrossRef Search ADS Fridman E. ( 2010 ) A refined input delay approach to sampled-data control. Automatica , 46 , 421 – 427 . Google Scholar CrossRef Search ADS Fridman E. ( 2014 ) Introduction to Time-Delay Systems: Analysis and Control . Systems and Control: Foundations and Applications. Switzerland : Birkhäuser Basel . Fridman E. & Orlov Y. ( 2009 ) Exponential stability of linear distributed parameter systems with time-varying delays. Automatica , 45 , 194 – 201 . Google Scholar CrossRef Search ADS Fridman E. , Seuret A. & Richard J. P. ( 2004 ) Robust sampled-data stabilization of linear systems: an input delay approach. Automatica , 40 , 1441 – 1446 . Google Scholar CrossRef Search ADS Fridman E. & Shaked U. ( 2003 ) Delay dependent stability and $$H_{\infty}$$ control: constant and time-varying delays. Int J Contr , 76 , 48 – 60 . Google Scholar CrossRef Search ADS Fujioka H. ( 2009 ) Stability analysis of systems with aperiodic sample-and-hold devices. Automatica , 45 , 771 – 775 . Google Scholar CrossRef Search ADS Gao H. , Lam J. , Wang C. & Wang Y. ( 2004 ) Delay-dependent output-feedback stabilisation of discrete-time systems with time-varying state delay. IEE Contr Theor & Appl, IET , 151 , 691 – 698 . Google Scholar CrossRef Search ADS Gao H. & Chen T. ( 2007 ) New results on stability of discrete-time systems with time-varying state delay. IEEE Trans. on Autom. Cont. , 52 , 328 – 334 . Google Scholar CrossRef Search ADS Hespanha J. , Naghshtabrizi P. & Xu Y. ( 2007 ) A survey of recent results in networked control systems. P IEEE , 95 , 138 – 162 . Google Scholar CrossRef Search ADS Kammler W. D. ( 2007 ) A First Course in Fourier Analysis . Cambridge : Cambridge University Press . Liu K. & Fridman E. ( 2012 ) Wirtinger’s inequality and Lyapunov-based sampled-data stabilization. Automatica , 48 , 102 – 108 . Google Scholar CrossRef Search ADS Liu K. , Suplin V. & Fridman E. ( 2010 ) Stability of linear systems with general sawtooth delay. Special issue on Time-delay Systems in IMA J Math Control I , 27 , 419 – 436 . Liu J. & Zhang J. ( 2012 ) Note on stability of discrete-time time-varying delay systems. Contr Theor & Appl, IET , 6 , 335 – 339 . Google Scholar CrossRef Search ADS Mikheev Yu. , Sobolev V. & Fridman E. ( 1988 ) Asymptotic analysis of digital control systems. Automat Rem Contr , 49 , 1175 – 1180 . Mirkin L. ( 2007 ) Some remarks on the use of time-varying delay to model sample-and-hold circuits. IEEE Trans. on Autom. Cont. , 52 , 1109 – 1112 . Google Scholar CrossRef Search ADS Naghshtabrizi P. , Hespanha, J. & Teel A. ( 2008 ) Exponential stability of impulsive systems with application to uncertain sampled-data systems. Syst Contr Lett , 57 , 378 – 385 . Google Scholar CrossRef Search ADS Selivanov A. & Fridman E. ( 2016 ) Observer-based input-to-state stabilization of networked control systems with large uncertain delays. Automatica , 74 , 63 – 70 . Google Scholar CrossRef Search ADS Seuret A. ( 2012 ) A novel stability analysis of linear systems under asynchronous samplings. Automatica , 48 , 177 – 182 . Google Scholar CrossRef Search ADS Seuret A. & Gouaisbaut F. ( 2013 ) Wirtinger-based integral inequality: Application to time-delay systems. Automatica , 49 , 2860 – 2866 . Google Scholar CrossRef Search ADS Seuret A. , Gouaisbaut F. & Fridman E. ( 2015 ) Stability of discrete-time systems with time-varying delays via a novel summation inequality. IEEE T Automat Contr , 60 , 2740 – 2745 . Google Scholar CrossRef Search ADS Shao H. & Han Q.-L. ( 2011 ) New stability criteria for linear discrete-time systems with interval-like time-varying delays. IEEE Trans Automat Contr , 56 , 619 – 625 . Google Scholar CrossRef Search ADS Zampieri S. ( 2008 ) Trends in networked control systems. Proceedings of the 17th IFAC World Congress , Vol. 41, Seoul, Korea, pp. 2886 – 2894 . © The authors 2017. Published by Oxford University Press on behalf of the Institute of Mathematics and its Applications. All rights reserved..

Journal

IMA Journal of Mathematical Control and InformationOxford University Press

Published: Feb 27, 2017

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off