Multidim Syst Sign Process https://doi.org/10.1007/s11045-018-0569-1

Asynchronous filtering for 2-D switched systems with missing measurements Shuang Shi1

· Yujunrong Ma1 · Shunqing Ren2

Received: 22 August 2017 / Revised: 21 March 2018 / Accepted: 26 March 2018 © Springer Science+Business Media, LLC, part of Springer Nature 2018

Abstract This work is concerned with H∞ filter design with missing measurements for a class of two-dimensional (2-D) switched systems represented by Fornasini–Marchesini local state-space model. The switching signal of the switched filters involve time delays, which result in the asynchronism between the filter and the system switching. The issues of asymptotic mean-square stability and 2 -gain analysis for the 2-D switched systems are addressed firstly, based on which mode-dependent filters are designed with mode-dependent average dwell time scheme. Finally, two examples are given to demonstrate the validity of the proposed technique. Keywords Asynchronous filtering · H∞ filtering · Mode-dependent average dwell time · Missing measurements · 2-D switched system

1 Introduction In the past decade, two-dimensional (2-D) systems have received considerable attention due to both their theoretical significance and wide applications (Ahn and Kar 2015; Kaczorek 1985; Shyu et al. 2014; Xu et al. 2010). In general, 2-D systems can be modeled by the Rosser model, Fornasini–Marchesini (FM) model, and Attasi model (Fornasini and Marchesini 2017; Roesser 2012). Especially, the Fornasini–Marchesini local state-space (FMLSS) model includes the Roesser model and the Attasi model as a special case (Kaczorek 1985; Li et al. 2012). Meanwhile, quantities of practical systems are subject to abrupt changes, and the switched systems provide a unified framework for characterizing these changes (Fei et al. 2017;

B

Shuang Shi [email protected]

1

Research Institute of Intelligent Control and Systems, Harbin Institute of Technology, Harbin 150000, China

2

Space Control and Inertial Technology Research Center, Harbin Institute of Technology, Harbin 150000, China

123

Multidim Syst Sign Process

Gao et al. 2011; Li et al. 2016; Zhao et al. 2012). The switching features may also include in 2-D systems. For example, a 2-D switched representation is needed when modeling the thermal processes in chemical reactors, heat exchangers, and pipe furnaces with multiple modes (Lo et al. 2008; Wu et al. 2015). In general, 2-D switched systems can also be formulated by Roesser model and FM model (Duan and Xiang 2013; Guan et al. 2017). So far, a number of meaningful results for 2-D switched systems have been reported in the literature (Benzaouia et al. 2011; Duan and Xiang 2013; Ghous et al. 2015; Shi et al. 2018; Wu et al. 2015; Xiang and Huang 2013). To mention a few, stability analysis and stabilization problems are discussed on 2-D discrete switched systems represented by FMLSS model in Fei et al. (2017). The stabilization problem formulated by Roesser type is investigated in Xiang and Huang (2013). In Shi et al. (2018), by designing a set of switching signals, 2-D switched systems are stabilized without any controller. It is well known that filtering problems play significant roles in signal processing. Filtering for 2-D systems have been widely reported (Ahn 2014; Ahn et al. 2017, 2015; Boukili et al. 2016; Du et al. 2000). In Boukili et al. (2016), Du et al. (2000), H∞ filter design is addressed for 2-D systems represented by Roesser and FM types, respectively. However, there are few results reported about filtering for 2-D switched systems. When considering filtering problems, one popular assumption is that the measurement always contain consecutive usable signals (Chen et al. 2015; Guan et al. 2016). However, the signals are vulnerable to be corrupted by the noise in practical applications, which results in the inconsecutive observations, namely, the system may have missing measurements (Yang et al. 2014; Zhang et al. 2009). Various factors may lead to inconsecutive measurements such as the high maneuverability of tracked target, intermittent sensor failures, accidental loss of some collected data and so on Sinopoli et al. (2004). One of the common approach to describe the missing measurement phenomenon is using the binary switching sequence. This sequence is specified by a conditional probability distribution which can be described by a Bernoulli distributed white sequence taking on values of 0 and 1. In Zhang et al. (2009), such a model is applied to design robust H∞ filters for stochastic time-delay systems. On the other hand, it takes time to identify the current mode of the system and apply the matched filter for switched systems, which will result in the asynchronous phenomenon between the system mode and the filter (Lian et al. 2013; Mahmoud and Shi 2012; Wang et al. 2013). In Zhang et al. (2011), asynchronous filtering for discrete-time switched systems is investigated under average dwell time (ADT) scheme. However, to the best of the authors’ knowledge, the asynchronous filtering is still unsolved for 2-D switched systems. In this paper, we concentrate on the asynchronous H∞ filtering with missing measurements for 2-D switched systems represented by FMLSS model. Mode-dependent average dwell time (MDADT) switching is adopted, which is more general compared with dwell time (DT) and ADT switching (Zhao et al. 2012). The remainder of this paper is organized as follows. In Sect. 2, the model of 2-D switched systems and missing measurements is established, and some definitions are provided. In Sect. 3, sufficient conditions to guarantee the stability and 2 -gain analysis are derived for 2-D switched systems. Then, the filter design is discussed in Sect. 4. In Sect. 5, two examples are presented to illustrate the effectiveness of the developed method. Finally, we conclude the paper in Sect. 6. Notation The notations used in this paper are fairly standard. The superscript “T ” stands for matrix transposition. Rn denotes the n-dimensional Euclidean space, and Z+ represents the set of nonnegative integers. Prob{·} indicates the occurrence probability of the event “·”.

123

Multidim Syst Sign Process

E{x} stands for the expectation of x. In addition, in symmetric block matrices or long matrix expressions, we use a “∗” as an ellipsis for the terms that are introduced by symmetry and diag{· · · } stands for a block-diagonal matrix. · refers to the Euclidean vector norm. I and 0 represent identify matrix and zero matrix with appropriate dimensions, respectively. The notation P > 0 means that P is real symmetric and positive definite. λmin {P} and λmax {P} denote the minimum and maximum eigenvalues of matrix P, respectively. The 2 norm of a  ∞ ∞ T 2-D signal ω(i, j) is defined by ω(·, ·)2 = i=0 j=0 ω (i, j)ω(i, j). The set of all vector functions such that ω(·, ·)22 < ∞ is denoted by 2 {[0, ∞), [0, ∞)}.

2 Problem formulation and preliminaries Consider a class of 2-D switched discrete-time systems given by x(i + 1, j + 1) = A1σ (i, j+1) x(i, j + 1) + A2σ (i+1, j) x(i + 1, j) + B1σ (i, j+1) ω(i, j + 1) + B2σ (i+1, j) ω(i + 1, j), z(i, j) = Cσ (i, j) x(i, j) + Dσ (i, j) ω(i, j),

(1) (2)

where (i, j) ∈ Z+ × Z+ , x(i, j) ∈ Rn is the state vector of the system, z(i, j) ∈ R p is the objective signal to be estimated, and ω(i, j) ∈ Rr is the exogenous disturbance input which belongs to 2 {[0, ∞), [0, ∞)}. σ (i, j) : (Z+ , Z+ ) → M is a switching signal, which takes its values in the finite set M  {1, 2, . . . , M} with M being the number of subsystems. A1 p , A2 p , B1 p , B2 p , C p and D p are constant real matrices with appropriate dimensions for σ (i, j) = p ∈ M . The measurements, which may contain missing data, can be described by y(i, j) = ρ(i, j)C yσ (i, j) x(i, j) + D yσ (i, j) ω(i, j),

(3)

where y(i, j) ∈ Rq is the measured output vector, C yp and D yp are constant real matrices with appropriate dimensions for p ∈ M , the stochastic variable ρ(i, j) ∈ R is a Bernoulli distributed white sequence taking the value 0 and 1 with Prob {ρ(i, j) = 1} = E {ρ(i, j)} := η, Prob {ρ(i, j) = 0} = 1 − E {ρ(i, j)} := 1 − η, and 0 < η < 1 is a known scalar. Assumption 1 For the system (1), the initial condition is satisfied x(0, j) = v j , ∀0 ≤ j ≤ z 2 , x(i, 0) = wi , ∀0 ≤ i ≤ z 1 , v0 = w0 , i = j = 0, x(0, j) = 0, ∀ j > z 2 , x(i, 0) = 0, ∀i > z 1 , where z 1 and z 2 are positive integers, v j and wi are given vectors. Assumption 2 The switching signal is assumed to be only dependent upon i + j (Duan and Xiang 2014; Wu et al. 2015).

123

Multidim Syst Sign Process

According to Assumption 2, the switching signal can be rewritten as σ (κ) : Z+ → M . The switching sequence can be described as (κ0 , κ1 , . . . , κl , κl+1 , . . .) with l = 0, 1, 2, . . ., κl denotes the l-th switching instant. Meanwhile, when σ (κl ) = p ∈ M , the p-th subsystem is active during [κl , κl+1 ). We are interested in designing a set of filter for the 2-D switched system (1)–(2) described by x(i ˆ + 1, j + 1) = G 1σ (κ+1−τ (κ+1)) x(i, ˆ j + 1) + G 2σ (κ+1−τ (κ+1)) x(i ˆ + 1, j) + K 1σ (κ+1−τ (κ+1)) [y(i, j + 1) − ηC yσ (κ+1−τ (κ+1)) x(i, ˆ j + 1)] + K 2σ (κ+1−τ (κ+1)) [y(i + 1, j) − ηC yσ (κ+1−τ (κ+1)) x(i ˆ + 1, j)], zˆ (i, j) = L σ (κ−τ (κ)) x(i, ˆ j),

(4) (5)

where x(i, ˆ j) and zˆ (i, j) are the estimate for x(i, j) and z(i, j), respectively, for σ (κ − τ (κ)) = p ∈ M , G 1 p , G 2 p , K 1 p , K 2 p and L p are filter parameters to be determined. τ (κ) is the uncertain switching delay satisfying 0 < τ (κ) ≤ τmax . Without loss of generality, we assume that the maximum switching delay τmax is a known priori, and τmax < κl+1 − κl , l = 1, 2, . . .. Assumption 3 Consider the switching occurs at κl , l = 1, 2, . . .. Since the delay considered in this paper is only related to the switching, we can assume that τ (κ) = τ (κl ) for ∀κ ∈ [κl , κl+1 ). Here, we consider σ (κl ) = p, σ (κl − 1) = q, ( p, q) ∈ M × M , p = q. When κ ∈ [κl , κl + τ (κl )), the q-th subsystem has switched to the p-th subsystem, but the q-th filter is still active because it takes time to identify the system modes and apply the matched filter. Combing (1)–(3) and (4)–(5), ∀( p, q) ∈ M × M , p = q, we obtain the augmented filtering error system ξ(i + 1, j + 1) = A1 p ξ(i, j + 1) + A2 p ξ(i + 1, j) + (ρ(i, j + 1) − η)Aη1 p ξ(i, j + 1) + (ρ(i + 1, j) − η)Aη2 p ξ(i + 1, j) + B1 p ω(i, j + 1) + B2 p ω(i + 1, j),

(6)

e(i, j) = C p ξ(i, j) + D p ω(i, j), where ξ(i, j) = [κl , κl + τ (κl )),



x T (i, j) xˆ T (i, j)

T

, e(i, j) = z(i, j) − zˆ (i, j), and when κ ∈

  0 A1 p 0 ˜ , Aη1 p = Aη1 p = A1 p ηK 1q C yp G 1q − ηK 1q C yp K 1q C yp    0 A2 p 0 A2 p = A˜ 2 p = , Aη2 p = A˜ η2 p = ηK 2q C yp G 2q − ηK 2q C yp K 2q C yp     B1 p B2 p B1 p = B˜ 1 p = , B2 p = B˜ 2 p = , K 1q D yp K 2q D yp   ˜ p = Dp, C p = C˜ p = C p −L q , D p = D = A˜ 1 p =

123



(7)

 0 , 0  0 , 0

Multidim Syst Sign Process

 when κ ∈ κl + τ (κl ), κl+1 ),    A1 p 0 0 A1 p = A¯ 1 p = , Aη1 p = A¯ η1 p = K 1 p C yp ηK 1 p C yp G 1 p − ηK 1 p C yp    A2 p 0 0 ¯ ¯ A2 p = A2 p = , Aη2 p = Aη2 p = K 2 p C yp ηK 2 p C yp G 2 p − ηK 2 p C yp     B1 p B2 p , B2 p = B¯ 2 p = , B1 p = B¯ 1 p = K 1 p D yp K 2 p D yp   ¯ p = Dp. C p = C¯ p = C p −L p , D p = D

 0 , 0  0 , 0

Here, we present the following definitions for further development. Definition 1 Fei et al. (2017) For any D ≥ r and switching signal σ , let Nσ p (r, D) denote the switching numbers of the p-th subsystem activated during the interval [r, D] and H p (r, D) denote the total running time of the p-th subsystem in [r, D], p ∈ M . We say that σ has a mode-dependent average dwell time τap if there exist positive numbers N0 p (we call N0 p the mode-dependent chatter bounds here) and τap such that Nσ p (r, D) ≤ N0 p +

H p (r, D) . τap

Definition 2 Duan and Xiang (2014) The 2-D switched system (1) with ω(i, j) ≡ 0 is said to be asymptotically mean-square stable under the switching signal σ (i, j), if for a given r ≥ 0, there exist ς > 0 and 0 < ε < 1, such that the solution x(i, j) satisfies ⎧ ⎧ ⎫ ⎫ ⎨  ⎨  ⎬ ⎬ x(i, j)2 ≤ ςε D−r E x(i, j)r2 , E ⎩ ⎩ ⎭ ⎭ i+ j=D

i+ j=r

for all D ≥ r , where x(i, j)r = sup {x(i, j) : i + j = r, i ≤ z 1 , j ≤ z 2 }. Assumption 4 The 2-D switched system (1) is asymptotically mean-square stable. Remark 1 To guarantee the asymptotic mean-square stability for the filtering error system (6)–(7), the prerequisite is that the original system (1) to be estimated, which exists no control, has to satisfy Assumption 4. Definition 3 Duan et al. (2013) For a given scalar 0 < α < 1, the 2-D switched system (1)–(2) is said to be with a weighted H∞ disturbance attenuation γ under switching signal σ if it satisfies the following conditions: 1. System (1) with ω(i, j) ≡ 0 is asymptotically mean-square stable; 2. Under zero boundary condition, it holds that ⎧ ⎫ ⎧ ⎫ ∞ ∞ ∞  ∞  ⎨ ⎬ ⎨ ⎬ ω E ¯ 22 , (α i+ j ¯z 22 ) < γ 2 E ⎩ ⎭ ⎩ ⎭ i=0 j=0

i=0 j=0

for all 0 = ω ∈ 2 {[0, ∞), [0, ∞)} , where the 2 -norm of 2-D discrete signal z(i, j) and ω(i, j) are defined as ¯z 22 = z(i + 1, j)22 + z(i, j + 1)22 , ω ¯ 22 = ω(i + 1, j)22 + ω(i, j + 1)22 .

123

Multidim Syst Sign Process

3 Stability and 2 -gain analysis In this section, an improved approach will be developed to solve the stability and 2 -gain analysis for (6)–(7). The switching signal is with the form of MDADT, and the asynchronous switching is taken into consideration. Lemma 1 Consider the 2-D switched system (6)–(7). For any ( p, q) ∈ M × M , p = q, let 0 < α p < 1, β p > 0 and μ p > 1 be given constants, if there exist matrices Pp > 0, Pq > 0, Q p > 0, Q q > 0, and a scalar γ > 0, such that ⎤ ⎡ Φ¯ 11 Φ¯ 12 Φ¯ 13 Φ¯ 14 ⎢ ∗ Φ¯ 22 Φ¯ 23 Φ¯ 24 ⎥ ⎥ (8) Φ¯ = ⎢ ⎣ ∗ ∗ Φ¯ 33 Φ¯ 34 ⎦ < 0, ∗ ∗ ∗ Φ¯ 44 ⎤ ⎡ Φ˜ 11 Φ˜ 12 Φ˜ 13 Φ˜ 14 ⎢ ∗ Φ˜ 22 Φ˜ 23 Φ˜ 24 ⎥ ⎥ (9) Φ˜ = ⎢ ⎣ ∗ ∗ Φ˜ 33 Φ˜ 34 ⎦ < 0, ∗ ∗ ∗ Φ˜ 44 Pq ≤ μ p Pp ,

(10)

Qq ≤ μ p Q p ,

(11)

where α¯ p = 1 − α p , β¯ p = 1 + β p , θ p = β¯ p /α¯ p , and T (P + Q ) A¯ Φ¯ 11 = A¯ 1T p (Pp + Q p ) A¯ 1 p + η(1 − η) A¯ η1 ¯ p Pp + C¯ Tp C¯ p , p η1 p − α p p Φ¯ 12 = A¯ 1T p (Pp + Q p ) A¯ 2 p , Φ¯ 13 = A¯ 1T p (Pp + Q p ) B¯ 1 p + C¯ Tp D¯ p , Φ¯ 14 = A¯ 1T p (Pp + Q p ) B¯ 2 p , T (P + Q ) A¯ Φ¯ 22 = A¯ 2T p (Pp + Q p ) A¯ 2 p + η(1 − η) A¯ η2 ¯ p Q p + C¯ Tp C¯ p , p η2 p − α p p Φ¯ 23 = A¯ 2T p (Pp + Q p ) B¯ 1 p , Φ¯ 24 = A¯ 2T p (Pp + Q p ) B¯ 2 p + C¯ Tp D¯ p ,

Φ¯ 33 = B¯ 1Tp (Pp + Q p ) B¯ 1 p + D¯ Tp D¯ p − γ 2 I, Φ¯ 34 = B¯ 1Tp (Pp + Q p ) B¯ 2 p , Φ¯ 44 = B¯ T (Pp + Q p ) B¯ 2 p + D¯ Tp D¯ p − γ 2 I, 2p

T (P + Q ) A˜ T Φ˜ 11 = A˜ 1T p (Pp + Q p ) A˜ 1 p + η(1 − η) A˜ η1 p η1 p − β¯ p P p + C˜ p C˜ p , p p Φ˜ 12 = A˜ 1T p (Pp + Q p ) A˜ 2 p , Φ˜ 13 = A˜ 1T p (Pp + Q p ) B˜ 1 p + C˜ Tp D˜ p , Φ˜ 14 = A˜ 1T p (Pp + Q p ) B˜ 2 p , T (P + Q ) A˜ T Φ˜ 22 = A˜ 2T p (Pp + Q p ) A˜ 2 p + η(1 − η) A˜ η2 p η2 p − β¯ p Q p + C˜ p C˜ p , p p Φ˜ 23 = A˜ 2T p (Pp + Q p ) B˜ 1 p , Φ˜ 24 = A˜ 2T p (Pp + Q p ) B˜ 2 p + C˜ Tp D˜ p ,

Φ˜ 33 = B˜ 1Tp (Pp + Q p ) B˜ 1 p + D˜ Tp D˜ p − γ 2 I, Φ˜ 34 = B˜ 1Tp (Pp + Q p ) B˜ 2 p , Φ˜ 44 = B˜ T (Pp + Q p ) B˜ 2 p + D˜ Tp D˜ p − γ 2 I. 2p

Then, for any switching signal with MDADT satisfying ∗ τap > τap =−

ln μ p + τmax ln θ p , ln α¯ p

(12)

the 2-D switched system (6)–(7) is asymptotically mean-square stable with a prescribed  τp τmax −1 M N0 p (α γ weighted H∞ disturbance attenuation level γs = max /αmin )θmax p=1 (θ p μ p ) and αmax = max p∈M {α p }, αmin = min p∈M {α p }, θmax = max p∈M {θ p }.

123

Multidim Syst Sign Process

Proof To establish the stability and H∞ performance of the 2-D switched system (6)–(7), we construct the following Lyapunov function: Vσ (κ) (x(i, j)) = Vσ1(κ) (x(i, j)) + Vσ2(κ) (x(i, j)), Vσ1(κ) (x(i, j)) = x T (i, j)Pσ (κ) x(i, j), Vσ2(κ) (x(i, j)) = x T (i, j)Q σ (κ) x(i, j),

(13)

where Pσ (κ) and Q σ (κ) are positive definite matrices for ∀σ (κ) ∈ M . Here, we define ΔVσ (κ) (x(i, j)) =Vσ (κ) (x(i + 1, j + 1)) − Vσ1(κ) (x(i, j + 1)) − Vσ2(κ) (x(i + 1, j)). (14)    Noting that E [ρ(i, j) − η]2 = (1 − η)η. When κ ∈ κl + τ (κl ), κl+1 ), the filter is matched with the mode. From (8), we obtain that   E{ΔVσ (κ) (x(i, j)) + ασ (κ) Vσ1(κ) (x(i, j + 1)) + Vσ2(κ) (x(i + 1, j)) + Γ (i, j)}   ¯ (i, j) < 0, = E ζ T (i, j)Φζ T ξ(i, j + 1) ξ(i + 1, j) ω T (i, j + 1) ω T (i, j + 1) , Γ (i, j) = e¯ T e¯ − T T   γ 2 ω¯ T ω¯ with e¯ = e T (i, j + 1) e T (i + 1, j) , ω¯ = ω T (i, j + 1) ω T (i + 1, j) . Thus, we can get where ζ (i, j) =



E{Vσ (κ) (x(i + 1, j + 1))} < E{α¯ σ (κ) [Vσ1(κ) (x(i, j + 1)) + Vσ2(κ) (x(i + 1, j))] − Γ (i, j)},

(15)

Consider κl + τ (κl ) < D < κl+1 . From (15), we obtain for κ ∈ [κl + τ (κl ), D), ⎧ E{Vσ (κ) (x(1, D − 1))} < E{α¯ σ (κ) [Vσ1(κ) (x(0, D − 1)) + Vσ2(κ) (x(1, D − 2))] ⎪ ⎪ ⎪ ⎪ ⎪ −Γ (0, D − 2)}, ⎪ ⎪ ⎪ E{V ⎪ (x(2, D − 2))} < E{ α¯ σ (κ) [Vσ1(κ) (x(1, D − 2)) + Vσ2(κ) (x(2, D − 3))] σ (κ) ⎪ ⎨ −Γ (1, D − 3)}, ⎪ .. ⎪ ⎪ ⎪ . ⎪ ⎪ ⎪ ⎪ (x(D − 1, 1))} < E{ α¯ σ (κ) [Vσ1(κ) (x(D − 2, 1)) + Vσ2(κ) (x(D − 1, 0))] E{V σ (κ) ⎪ ⎪ ⎩ −Γ (D − 2, 0)}. According to Assumption 1, we can get for κ ∈ [κl + τ (κl ), D), ⎫ ⎧ ⎬ ⎨  Vσ (κ) (x(i, j)) E ⎭ ⎩ i+ j=D ⎫ ⎧ ⎬ ⎨   Vσ (κ) (x(i, j)) − Γ (i, j) < E α¯ σ (κ) ⎭ ⎩ i+ j=D−1

< ... <

⎧ ⎨

D−(κ +τ (κl )) E α¯ σ (κl ) l ⎩

i+ j=D−2

 i+ j=κl +τ (κl )

Vσ (κl ) (x(i, j)) −

D−2 



s=κl +τ (κl )−1 i+ j=s

⎫ ⎬ α¯ σD−2−s (κl ) Γ (i, j)⎭ . (16)

123

Multidim Syst Sign Process

Similarly, consider κ ∈ [κl , κl + τ (κl )), in this situation the filter is mismatched with the subsystem. From (9), it holds that ⎧ ⎫ ⎨ ⎬  E Vσ (κ) (x(i, j)) ⎩ ⎭ i+ j=κl +τ (κl ) ⎧ ⎫ κl +τ (κl )−2  ⎨ ⎬  τ (κ ) κ +τ (κ )−2−s < E β¯σ (κll ) β¯σl(κl ) l Vσ (κl ) (x(i, j)) − Γ (i, j) . (17) ⎩ ⎭ s=κl −1

i+ j=κl

i+ j=s

 Combining (16) with (17), we obtain that for κ ∈ κl , κl+1 ), ⎧ ⎫ ⎨  ⎬ E Vσ (κ) (x(i, j)) ⎩ ⎭ i+ j=D ⎧ D−2 ⎨    l τ (κl ) ≤ E α¯ σD−κ θ V (x(i, j)) − α¯ σD−2−s σ (κ ) l (κl ) Γ (i, j) σ (κ (κ ) ) l l ⎩ i+ j=κl i+ j=s s=κl +τ (κl )−1 ⎫ κl +τ (κl )−2  ⎬ κl +τ (κl )−2−s α¯ σD−2−s θ Γ (i, j) . (18) − (κl ) σ (κl ) ⎭ s=κl −1

i+ j=s

On the other hand, according to (10)–(11), we conclude for all (σ (κl ) = p, σ (kl−1 ) = q) ∈ M × M , p = q, ⎧ ⎧ ⎫ ⎫ ⎨ ⎨  ⎬ ⎬  (19) E Vσ (κl ) (x(i, j)) ≤ E μσ (κl ) Vσ (κl−1 ) (x(i, j)) . ⎩ ⎩ ⎭ ⎭ i+ j=κl

i+ j=κl

We prove the stability firstly. Consider ω(i, j) ≡ 0, from (18), we have, ⎧ ⎫ ⎧ ⎫ ⎨  ⎬ ⎨ ⎬  l τ (κl ) E Vσ (κ) (x(i, j)) ≤ E α¯ σD−κ θσ (κl ) Vσ (κl ) (x(i, j)) . (κ ) 1 ⎩ ⎭ ⎩ ⎭ i+ j=D

i+ j=κl

Denote τ (κl ) as τl , and τmax = max p∈M {τ p }. From (19)–(20), we obtain ⎧ ⎫ ⎨  ⎬ E Vσ (κ) (x(i, j)) ⎩ ⎭ i+ j=D ⎫ ⎧ ⎬ ⎨  l τ (κl ) θ V (x(i, j)) < E μσ (κl ) α¯ σD−κ σ (κ ) l−1 (κl ) σ (κl ) ⎭ ⎩ i+ j=κl

< ... ⎧ ⎫ M M ⎨ ⎬   τ H (r,D) ≤E (μ p θ p p ) N0 p +H p (r,D)/τap α¯ p p Vσ (r ) (x(i, j)) ⎩ ⎭ p=1 p=1 i+ j=r ⎫ ⎧  M ⎬ ⎨   D−r  τ p N0 p (μ p θ p ) Vσ (r ) (x(i, j)) . ≤E max (μ p θ pτmax )1/τap α¯ p ⎭ ⎩ p∈M p=1

123

i+ j=r

(20)

Multidim Syst Sign Process

From (13), we know that there exist two positive scalars λ1 and λ2 such that ∀σ (κ) = p ∈ M,       λ1 E x(i, j)2 ≤ E V p (x(i, j)) ≤ λ2 E x(i, j)2 , where

  λ1 = min λmin (Pp ) + λmin (Q p ) ,   λ2 = max λmax (Pp ) + λmax (Q p ) .

Thus, we have E

⎧ ⎨  ⎩

x(i, j)2

i+ j=D

⎫ ⎬ ⎭

≤ ςε D−r E

⎧ ⎨ 

x(i, j)r2

⎩

i+ j=r

⎫ ⎬ ⎭

,

where ς=

M   λ2  τ τ (μ p θ p p ) N0 p > 0, ε = max (μ p θ p p )1/τap α¯ p , p∈M λ1 p=1

then if there exist constants τap , p ∈ M satisfying (12), we can get 0 < ε < 1. From Definition 2, it can be concluded that the system (6) under asynchronous switching is asymptotically mean-square stable. Then, we address the 2 -gain analysis. From (18)–(19), we obtain ⎫ ⎧ ⎬ ⎨  Vσ (κl ) (x(i, j)) E ⎭ ⎩ i+ j=D ⎧ D−2 ⎨    l τ (κl ) θ μ V (x(i, j)) − α¯ σD−2−s < E α¯ σD−κ σ (κ ) σ (κ ) l l−1 (κl ) Γ (i, j) (κ ) ) σ (κ l l ⎩ i+ j=κl s=κl +τ (κl )−1 i+ j=s ⎫ κl +τ (κl )−2  ⎬ κl +τ (κl )−2−s α¯ σD−2−s θ Γ (i, j) − (κl ) σ (κl ) ⎭ s=κl −1

< ...

⎧ ⎨

i+ j=s

κ −κ

τ (κ )

τ (r )

l < E α¯ σD−κ α¯ l l−1 · · · α¯ σκ1(r−r) θσ (κll ) . . . θσ (r ) μσ (κl ) . . . μσ (r ) ⎩ (κl ) σ (κl−1 )

D−κ − α¯ σ (κl )l

τ (κ ) τ (r ) . . . α¯ σκ1(r−r) θσ (κll ) . . . θσ (r ) μσ (κl ) . . . μσ (r )

D−κ − α¯ σ (κl )l

τ (κ ) τ (r ) . . . α¯ σκ1(r−r) θσ (κll ) . . . θσ (r ) μσ (κl ) . . . μσ (r )

×

r +τ (r )−2



r +τ (r )−2−s

s=r −1 i+ j=s

−

D−2 

α¯ σκ1(r−2−s θσ (r ) )



s=κl +τ (κl )−1 i+ j=s

κ 1 −2



Vσ (r ) (x(i, j))

i+ j=r



s=r +τ (r )−1 i+ j=s

α¯ σκ1(r−2−s Γ (i, j) )

Γ (i, j) − . . .

α¯ σD−2−s (κl ) Γ (i, j) −

κl +τ (κl )−2



s=κl −1

i+ j=s

κl +τ (κl )−2−s α¯ σD−2−s Γ (i, (κl ) θσ (κl )

⎫ ⎬ j) . ⎭ (21)

123

Multidim Syst Sign Process

Denote αmax = max{α p }, αmin αmax , α¯ min = 1 − αmin . Con = min{α p } and α¯ max = 1 −  sider zero initial condition, i.e., i+ j=r Vσ (r ) (x(i, j)) = 0, and i+ j=κl Vσ (κl ) (x(i, j)) ≥ 0. κ +τ p −2−s

Owing to 1 < θ p p

τ −1

< θ pp

, s ∈ [κ p − 1, κ p + τ p − 2), then we have

⎧ ⎡ ⎤⎫ M ⎨ D−2 ⎬    τp D−2−s N (s,D) T ⎣α¯ max E (θ p μ p ) σ p e¯ e¯⎦ ⎩ ⎭ s=r −1 i+ j=s p=1 ⎧ ⎤⎫ ⎡ M ⎬ ⎨ D−2    τ −1 τ p Nσ p (s,D) σ (s) 2 T ⎦ ⎣α¯ D−2−s (θ μ ) θ γ ω ¯ ω ¯ ≤E . p p min σ (s) ⎭ ⎩ s=r −1 i+ j=s

p=1

τ

Multiplying both sides of the above inequality by (θ p p μ p )−Nσ p (r,D) , it follows that ⎧ ⎤⎫ ⎡ M ⎬ ⎨ D−2    τ D−2−s ⎣α¯ max (θ p p μ p )−Nσ p (r,s) e¯ T e¯⎦ E ⎭ ⎩ s=r −1 i+ j=s p=1 ⎧ ⎡ ⎤⎫ M ⎨ D−2 ⎬    τp −Nσ p (r,s) τσ (s) −1 2 T ⎦ ⎣α¯ D−2−s ≤E (θ μ ) θ γ ω ¯ ω ¯ . p p min σ (s) ⎩ ⎭ s=r −1 i+ j=s

p=1

From Definition 1, we know −N0 p − H p (r, s)/τap ≤ −Nσ p (r, s) ≤ 0. Meanwhile, by noticing (12) we have ⎧ ⎤⎫ M ⎨ D−2 ⎬     τp τ D−2−s E α¯ max (θ p p μ p )−N0 p +H p (r,s) ln α¯ p / ln(θ p μ p ) e¯ T e¯⎦ ⎩ ⎭ s=r −1 i+ j=s p=1 ⎫ ⎧ ⎨ D−2 !⎬   D−2−s τmax −1 2 T α¯ min θmax γ ω¯ ω¯ . ≤E ⎭ ⎩ s=r −1 i+ j=s

Furthermore, ⎧ ⎡ ⎤⎫ ∞ D−2 M ⎨ ⎬    H (r,s) T D−2−s ⎣α¯ max E α¯ p p e¯ e¯⎦ ⎩ ⎭ p=1 D=2 s=r −1 i+ j=s ⎧ ⎫ ∞ D−2 ⎨ ⎬   D−2−s τmax −2 2 T α¯ min θmax γ ω¯ ω¯ , ≤ Θ p γ 2E ⎩ ⎭ D=2 s=r −1 i+ j=s

and E

⎧ ∞ ⎨  ⎩



D−2−s s α¯ max α¯ max e¯ T e¯

s=r −1 D=s+2 i+ j=s

≤ Θ p γ 2E

123

∞ 

⎧ ∞ ⎨  ⎩

∞ 



s=r −1 D=s+2 i+ j=s

⎫ ⎬ ⎭ ⎫ ⎬

D−2−s τmax −2 2 T α¯ min θmax γ ω¯ ω¯ , ⎭

Multidim Syst Sign Process

where Θ p = ∞ 

M

τp N0 p . p=1 (θ p μ p )



D−2−s α¯ max =

D=s+2 i+ j=s

Since 

1/αmax ,

i+ j=s

∞ 



D=s+2 i+ j=s

D−2−s α¯ min =



1/αmin ,

i+ j=s

thus, we conclude ⎫ ⎧ ⎫ ⎧ ∞ M ∞ ⎬  ⎨  ⎬ ⎨    α τ max τmax −1 2 α s e¯ T e¯ ≤ (θ p p μ p ) N0 p θmax γ E ω¯ T ω¯ , E ⎭ ⎩ ⎭ ⎩ αmin s=r −1 i+ j=s p=1 s=r −1 i+ j=s ⎫ ⎫ ⎧ ⎧ ∞ ∞ ∞  ∞  ⎬ ⎬ ⎨ ⎨ ω ¯ 22 ≤ γs2 E ¯ 22 , α i+ j e E ⎭ ⎭ ⎩ ⎩ i=0 j=0

i=0 j=0



τ

p τmax −1 M N0 p (α γ . where α = 1 − αmax and γs = max /αmin )θmax p=1 (θ p μ p ) According to Definition 3, we conclude that the 2-D switched system (6)–(7) is asymptotically stable with a prescribed weighted H∞ disturbance attenuation level γs , which ends the proof.



4 H∞ Filter Design According to 2 -gain analysis in the preceding section, we address asynchronous H∞ filter design with missing measurements for 2-D switched systems. Theorem 1 Consider the 2-D switched system (1)–(3). For any ( p, q) ∈ M × M , p = q, let 0 < α p < 1, β p > 0 and μ p > 1 be given constants, if there exist matrices Pp > 0, Q p > 0, X p , Y p , Z p , G F1 p , G F2 p , K F1 p , K F2 p , L F p , p ∈ M , and a scalar γ > 0, such that (10)–(11) hold, and ⎤ ⎡ Ξ¯ 11 0 Ξ¯ 13 Ξ¯ 14 Ξ¯ 15 ⎢ ∗ −γ 2 I Ξ¯ 23 0 Ξ¯ 25 ⎥ ⎥ ⎢ ⎢ ∗ (22) ∗ Ξ¯ 33 0 0 ⎥ ⎥ < 0, ⎢ ⎣ ∗ ∗ ∗ Ξ¯ 44 0 ⎦ ∗ ∗ ∗ ∗ −I ⎡ ˜ ⎤ Ξ11 0 Ξ˜ 13 Ξ˜ 14 Ξ˜ 15 ⎢ ∗ −γ 2 I Ξ˜ 23 0 Ξ˜ 25 ⎥ ⎢ ⎥ ⎢ ∗ (23) ∗ Ξ˜ 33 0 0 ⎥ ⎢ ⎥ < 0, ⎣ ∗ ⎦ ˜ ∗ ∗ Ξ44 0 ∗ ∗ ∗ ∗ −I where Ξ¯ 11 = diag{−α¯ p Pp , −α¯ p Q p }, Ξ¯ 33 = diag{Pp − R p − R Tp , Q p − R p − R Tp }, 1 2 Ξ¯ 44 = ηdiag{ ¯ Ξ¯ 33 , Ξ¯ 33 }, Ξ¯ 14 = diag{Ξ¯ 14 , Ξ¯ 14 },     1     1 1 1 C Tp 0 0 C Tp Ξ¯ 13 Ξ¯ 13 Ξ¯ 23 Ξ¯ 23 ¯ ¯ ¯ ¯ Ξ13 = ¯ 2 ¯ 2 , Ξ23 = ¯ 2 ¯ 2 , Ξ15 = , , Ξ25 = 0 −L TF p −L TF p 0 Ξ13 Ξ13 Ξ23 Ξ23

Ξ˜ 11 = diag{−β¯ p Pp , −β¯ p Q p }, Ξ˜ 33 = diag{Pp − Rq − RqT , Q p − Rq − RqT },

123

Multidim Syst Sign Process 1 2 ¯ Ξ˜ 33 , Ξ˜ 33 }, Ξ˜ 14 = diag{Ξ˜ 14 , Ξ˜ 14 }, Ξ˜ 44 = ηdiag{  1        1 1 1 ˜ ˜ ˜ ˜ C Tp 0 0 C Tp Ξ Ξ13 ˜ 23 = Ξ23 Ξ23 , Ξ˜ 15 = ˜ Ξ˜ 13 = ˜ 13 = , Ξ , , Ξ 25 2 Ξ 2 Ξ ˜2 ˜2 −L TFq 0 0 −L TFq Ξ13 Ξ˜ 23 13 23

with α¯ p = 1 − α p , β¯ p = 1 + β p , θ p = β¯ p /α¯ p , η¯ = (1 − η)η, and for r = 1, 2,   X p Yp Rp = , Z p Yp   T T T K T T T Ar p X p + ηC yp Fr p Ar p Z p + ηC yp K Fr p r Ξ¯ 13 , = T T K T G TFr p − ηC yp Fr p G Fr p − ηC yp K Fr p  T  T K T T C yp K Fr p C yp Fr p C yp K Fr p C yp K Fr p r = Ξ¯ 14 , 0 0 0 0   r T K T T T = BrTp X Tp + D yp Ξ¯ 23 Fr p Br p Z p + D yp K Fr p ,  T T  T K T T T Ar p X q + ηC yp Frq Ar p Z q + ηC yp K Frq r = Ξ˜ 13 , T K T K G TFrq − ηC yp G TFrq − ηC yp Frq Frq  T  T K T T C yp K Frq C yp Frq C yp K Frq C yp K Frq r ˜ Ξ14 = , 0 0 0 0   r T K T T T = BrTp X qT + D yp Ξ˜ 23 Frq Br p Z q + D yp K Frq , Then, for any switching signal with MDADT satisfying (12), the 2-D switched system (6)–(7) is asymptotically mean-square stable with a prescribed weighted H∞ disturbance  τp τmax −1 M N0 p (α attenuation level γs = γ and θmax = max p∈M {θ p }. max /αmin )θmax p=1 (θ p μ p ) Moreover, the filter parameters are given by G 1 p = Y p−1 G F1 p , G 2 p = Y p−1 G F2 p , K 1 p = Y p−1 K F1 p , K 2 p = Y p−1 K F2 p , and L p = L F p . Proof Choose (13) as the Lyapunov function. T T Noticing that −R p Pp−1 R Tp < Pp − R p − R Tp and −R −1 p Q p R p < Q p − R p − R p , denote G F1 p = Y p G 1 p , G F2 p = Y p G 2 p , K F1 p = Y p K 1 p , K F2 p = Y p K 2 p , and L F p = L p . Then taking congruent transformation and using Schur complement, we know that (22) ensures (8). By a similar procedure, we find that (9) can be guaranteed by (23). According to Lemma 1, we conclude that system (6)–(7) is asymptotically mean-square stable with a prescribed weighted H∞ disturbance attenuation level γs , which ends the proof.

When taking no account of the asynchronous switching, i.e., τ (κl ) = 0, we have the following corollary. Corollary 1 Consider the 2-D switched system (1)–(3). For any ( p, q) ∈ M × M , p = q, let 0 < α p < 1 and μ p > 1 be given constants, if there exist a set of matrices Pp > 0, Q p > 0, X p , Y p , Z p , G F1 p , G F2 p , K F1 p , K F2 p , L F p , and a scalar γ > 0, such that (10)–(11) and (22) hold, Then, for any switching signal with MDADT satisfying ∗ τap > τap =−

ln μ p , ln α¯ p

where α¯ p = 1 − α p , the 2-D switched system (6)–(7) is asymptotically mean-square stable with a prescribed weighted H∞ disturbance attenuation level γs

123

Multidim Syst Sign Process

 N0 p M = p=1 μ p (αmax /αmin )γ , αmax = max p∈M {α p }, αmin = min p∈M {α p }. Moreover, the filter parameters are given by G 1 p = Y p−1 G F1 p , G 2 p = Y p−1 G F2 p , K 1 p = Y p−1 K F1 p , K 2 p = Y p−1 K F2 p , and L p = L F p . Remark 2 Only switching delay is considered in this paper. As a matter of fact, state delay may also be involved in practical systems. There exist some results for 2-D switched delay systems (Duan et al. 2013; Ghous et al. 2015), one can develop the filter design for 2-D switched delay systems by using the similar methods. Remark 3 All above results are based on MDADT switching, which is more general than ADT switching (Fei et al. 2017). Recently, a more flexible switching logic, persistent dwell time (PDT) switching is applied in 1-D switching systems (Zhang et al. 2015). In future works, it is meaningful to extended PDT switching to 2-D switched systems.

5 Illustrative examples In this section, we use two examples to illustrate the effectiveness of the results developed in the above section. Example 1 Consider the 2-D switched systems (1)–(3) consisting of two subsystems as follow: Subsystem 1:         0.2352 0.0019 0.1 −0.0965 0.1002 0.01 A11 = , B11 = , A21 = , B21 = , −0.9648 0.2019 0.1 −0.0965 0.0002 0.01     C1 = 0.2 0.2 , D1 = −0.1, C y1 = 0.1 0.2 , D y1 = 0.1; Subsystem 2:         0.0313 0.1757 0.05 0 0 0 A12 = , B12 = , A22 = , B22 = , −0.5773 0.0515 0.1 0.1254 −0.0970 0.2     C2 = 0.2 0.2 , D2 = −0.1, C y2 = 0.1 0.2 , D y2 = 0.1. Our purpose here is to design a set of mode-dependent filters with missing measurements such that the filtering error system is asymptotically mean-square stable and has an prescribed H∞ performance. Set μ1 = μ2 = 1.1 and α1 = 0.4, α2 = 0.3, β1 = 1.2, β2 = 1.1, and assume τmax = 2, η = 0.9, γ = 1.2247. Then we find an admissible solution by using standard softwares. The ∗ = 5.2736, τ ∗ = 6.4275, and a set of filters are designed: corresponding τa1 a2     0.0379 −0.0593 −0.0140 0.0373 G 11 = , G 21 = , −0.1537 0.2619 0.0256 −0.0586       0.0005 −0.0388 , K 21 = , L 1 = 0.0050 −0.0986 , K 11 = −0.0411 0.1252     −0.0068 0.0447 −0.0030 0.0103 G 12 = , G 22 = , −0.0078 0.0018 0.0156 −0.0542       −0.0139 −0.0007 K 12 = , K 22 = , L 2 = 0.0160 −0.1535 . −0.0035 0.0014

123

Multidim Syst Sign Process

Here, we choose the switching signal as Fig. 1. The missing measurements satisfying η = 0.9 is shown in Fig. 2. Under the zero initial condition, the disturbance input is assumed to be ω(i, j) = sin(0.1π(i + j))e−0.2(i+ j) . Using the filters obtained by Theorem 1, we can get the filter state responses shown in Figs. 3, 4, and filter error responses shown in Fig. 5. It can be observed the augmented system is asymptotically stable. Example 2 Consider the thermal processes in heat exchangers, which can be described by a partial differential equation (Ghous et al. 2015): ∂ T (x, t) ∂ T (x, t) =− − aσ (x,t) T (x, t) + bσ (x,t) f (x, t), ∂x ∂t

Switching signal for subsystem σ(κ) Switching signal for filters

2

1

0

5

10

15

20

25

30

35

40

κ Fig. 1 The switching signal σ (κ)

1

0 40 30 20

j

20

10 0

0

Fig. 2 The missing measurement distribution ρ(i, j)

123

30

10

i

40

(24)

Multidim Syst Sign Process

10-3 1

xf

1

0 -1 -2 -3 40 30 20 10

j

0

0

10

30

20

40

i

Fig. 3 The filter state response xˆ1 (i, j)

10-3 6

xf

2

4 2 0 -2 40 30

j

20 10 0

10

0

30

20

40

i

Fig. 4 The filter state response xˆ2 (i, j)

0.02

e

0 -0.02 -0.04 -0.06 40 30 20

j

10 0

0

10

20

30

40

i

Fig. 5 The filter error response e(i, j)

123

Multidim Syst Sign Process

Switching signal for subsystem σ(κ) Switching signal for filters

2

1

0

5

10

15

20

25

30

35

40

κ Fig. 6 The switching signal σ (κ)

where T (x, t) is the temperature at x(space) ∈ [0, x f ] and t (time) ∈ [0, ∞), f (x, t) is the input function, and aσ (x,t) , bσ (x,t) are real coefficients, which are functions of σ (x, t). Similar to the technique used in Kaczorek (1985), here, define ∂ T (x, t) T (i, j + 1) − T (i, j) ≈ , ∂t Δt ∂ T (x, t) T (i, j) − T (i − 1, j) ≈ , ∂x Δx u(x, t) ≈u(i, j), where T (i, j) = T (iΔx, jΔt), u(i, j) = u(iΔx, jΔt), Δx and Δt are space and time discretization periods, respectively. Then, when viewing the disturbance input as the input function, we obtain that (24) can be rewritten in the form of (1) with     01 0 , B1σ (i, j) = , A1σ (i, j) = 00 0     0 0 0 A2σ (i, j) = Δt , B2σ (i, j) = . Δt bσ (i, j) Δt Δx 1 − Δx − aσ (i, j) Δt Consider this 2-D switched system with two modes and Δt = 0.2, Δx = 0.5, a1 = 2, a2 = 2.5, b1 = 0.25, b2 = 0.5. Here, assume that   C1 = C2 = C y1 = C y2 = 0.1 0.1 , D1 = D2 = −0.1, D y1 = D y2 = 0.1. Set μ1 = μ2 = 1.1, α1 = α2 = 0.35, β1 = 1.2, β2 = 1.1, and τmax = 1, η = 0.9, ∗ = 2.9435, τ ∗ = 3.0515. Choose the γ = 3.1623. According to Theorem 1, we have τa1 a2 switching signal as Fig. 6. Under the zero initial condition, the disturbance input is assumed j)) to be ω(i, j) = sin(0.1π(i+ . Using the filters obtained by Theorem 1, we can get the filter 0.1(i+ j)2 error responses as shown in Fig. 7, which satisfies above constraints. This illustrate the effectiveness of the proposed method.

123

Multidim Syst Sign Process

0.05

e

0 −0.05 −0.1 −0.15 40 30

j

20 10 0

0

10

30

20

40

i

Fig. 7 The filter error response e(i, j)

6 Conclusions In this paper, the problem of asynchronous H∞ filtering has been solved for 2-D switched systems with missing measurements. By constructing a class of mode-dependent Lyapunov function, sufficient conditions are proposed to guarantee the asymptotic mean-square stability and 2 -gain of the 2-D switched systems. Based on the results, a set of filters are designed. For future study, since the state delay is prevalent in practical systems, it would be interesting to extend the above results to 2-D switched delay systems. Meanwhile, more general switching logics could be applied to study 2-D switched delay systems.

References Ahn, C. K. (2014). 2 − ∞ suppression of limit cycles in interfered two-dimensional digital filters: A Fornasini–Marchesini model case. IEEE Transactions on Circuits and Systems II, 61(8), 614–618. Ahn, C. K., & Kar, H. (2015). Expected power bound for two-dimensional digital filters in the Fornasini– Marchesini local state-space model. IEEE Signal Processing Letters, 22(8), 1065–1069. Ahn, C. K., Shi, P., & Basin, M. V. (2017). Two-dimensional peak-to-peak filtering for stochastic FornasiniMarchesini systems. IEEE Transactions on Automatic Control,. https://doi.org/10.1109/TAC.2017. 2747761. Ahn, C. K., Shi, P., & Basin, M. V. (2015). Two-dimensional dissipative control and filtering for Roesser model. IEEE Transactions on Automatic Control, 60(7), 1745–1759. Benzaouia, A., Hmamed, A., Tadeo, F., & Hajjaji, A. (2011). Stabilisation of discrete 2D time switching systems by state feedback control. International Journal of Systems Science, 42(3), 479–487. Boukili, B., Hmamed, A., & Tadeo, F. (2016). Robust H∞ filtering for 2-D discrete Roesser systems. Journal of Control, Automation and Electrical Systems, 27(5), 497–505. Chen, X., Lam, J., & Lam, H. (2015). Positive filtering for positive Takagi–Sugeno fuzzy systems under 1 performance. Information Sciences, 299, 32–41. Du, C., Xie, L., & Soh, Y. (2000). H∞ filtering of 2-D discrete systems. IEEE Transactions on Signal Processing, 48(6), 1760–1768. Duan, Z., & Xiang, Z. (2013). State feedback H∞ control for discrete 2D switched systems. Journal of the Franklin Institute, 350(6), 1513–1530. Duan, Z., & Xiang, Z. (2014). Output feedback H∞ stabilization of 2D discrete switched systems in FM LSS model. Circuits Systems & Signal Processing, 33(4), 1095–1117. Duan, Z., Xiang, Z., & Karimi, H. (2013). Delay-dependent H∞ control for 2-D switched delay systems in the second FM model. Journal of the Franklin Institute, 350(7), 1697–1718.

123

Multidim Syst Sign Process Fei, Z., Guan, C., & Shi, P. (2017). Further results on control for discrete-time Markovian jump time delay systems. International Journal of Control, 90(7), 1505–1517. Fei, Z., Shi, S., Zhao, C., & Wu, L. (2017). Asynchronous control for 2-D switched systems with modedependent average dwell time. Automatica, 79, 198–206. Fornasini, E., & Marchesini, F. (2017). State-space realization theory of two-dimensional filters. IEEE Transactions on Automatic Control, 21(4), 484–492. Gao, H., Fei, Z., Lam, J., & Du, B. (2011). Further results on exponential estimates of Markovian jump systems with mode-dependent time-varying delays. IEEE Transactions on Automatic Control, 56(1), 223–229. Ghous, I., Xiang, Z., & Karimi, H. (2015). State feedback H∞ control for 2-D switched delay systems with actuator saturation in the second FM mode. Circuits, Systems, and Signal Processing, 34(7), 2167–2192. Guan, C., Fei, Z., Li, Z., & Xu, Y. (2016). Improved H∞ filter design for discrete-time Markovian jump systems with time-varying delay. Journal of the Franklin Institute, 353(16), 4156–4175. Guan, C., Fei, Z., Wang, Z., & Wu, L. (2017). Stabilization of continuous 2-D switched systems with all modes unstable. IET Control Theory and Applications,. https://doi.org/10.1049/iet-cta.2017.0736. Kaczorek, T. (1985). Two-dimensional linear systems. Berlin: Springer. Li, X., Gao, H., & Wang, C. (2012). Generalized Kalman–Yakubovich-Popov lemma for 2-D FM LSS model. IEEE Transactions on Automatic Control, 57(12), 3090–3103. Li, Z., Xu, Y., Huang, H., & Misra, S. (2016). Sparse control and compressed sensing in networked switched systems. IET Control Theory & Applications, 10(9), 1078–1087. Lian, J., Mu, C., & Shi, P. (2013). Asynchronous H∞ filtering for switched stochastic systems with timevarying delay. Information Sciences, 224, 200–212. Lo, W., Wang, L., & Li, B. (2008). Thermal transistor: Heat flux switching and modulating. Journal of The Physical Society of Japan, 77(5), 1–4. Mahmoud, M., & Shi, P. (2012). Asynchronous H∞ filtering of discrete-time switched systems. Signal Processing, 92(10), 2356–2364. Roesser, R. (2012). A discrete state-space model for linear image processing. IEEE Transactions on Automatic Control, 57(12), 3090–3103. Shi, S., Fei, Z., Sun, W., & Yang, X. (2018). Stabilization of 2-D switched systems with all modes unstable via switching signal regulation. IEEE Transactions on Automatic Control, 63(3), 857–863. Shyu, J., Pei, S., Huang, Y., & Chen, Y. (2014). A new structure and design method for variable fractional-delay 2-D FIR digital filters. Multidimensional Systems & Signal Processing, 25(3), 511–529. Sinopoli, B., Schenato, L., Franceschetti, M., & Poolla, K. (2004). Kalman filtering with intermittent observations. IEEE Transactions on Automatic Control, 49(9), 1453–1464. Wang, Y., Zhao, J., & Jiang, B. (2013). Stabilization of a class of switched linear neutral systems under asynchronous switching. IEEE Transactions on Automatic Control, 58(8), 2114–2119. Wu, L., Yang, R., Shi, P., & Su, X. (2015). Stability analysis and stabilization of 2-D switched systems under arbitrary and restricted switchings. Automatica, 59, 206–215. Xiang, Z., & Huang, S. (2013). Stability analysis and stabilization of discrete-time 2D switched systems. Circuits Systems & Signal Processing, 32(1), 401–414. Xu, H., Lin, Z., & Makur, A. (2010). Non-fragile H2 and H∞ filter designs for polytopic two-dimensional systems in Roesser model. Multidimensional Systems & Signal Processing, 21(3), 255–275. Yang, H., Chen, Y., & Ren, T. (2014). H∞ filtering for networked systems with bounded measurement missing. Mathematical Problems in Engineering, 2014(17), 1–10. Zhang, H., Chen, Q., Yan, H., & Liu, J. (2009). Robust H∞ filtering for switched stochastic system with missing measurements. IEEE Transactions on Signal Processing, 57(9), 3466–3474. Zhang, L., Cui, N., Liu, M., & Zhao, Y. (2011). Asynchronous filtering of discrete-time switched linear systems with average dwell time. IEEE Transactions on Circuits & Systems I, 58(5), 1109–1118. Zhang, L., Zhuang, S., & Shi, P. (2015). Non-weighted quasi-time-dependent H∞ filtering for switched linear systems with persistent dwell-time. Automatica, 54, 201–209. Zhao, X., Zhang, L., Shi, P., & Liu, M. (2012). Stability and stabilization of switched linear systems with mode-dependent average dwell time. IEEE Transactions on Automatic Control, 57(7), 1809–1815.

123