You are on page 1of 10

2708

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: REGULAR PAPERS, VOL. 58, NO. 11, NOVEMBER 2011

Realization Using the Fornasini-Marchesini Model for Implementations in Distributed Grid Sensor Networks
M. G. Buddika Sumanasena, Member, IEEE, and Peter H. Bauer, Fellow, IEEE
AbstractA method for distributed information processing in grid sensor networks using the Fornasini-Marchesini (FM) local state space model has been proposed in the literature recently. The method can be used to implement linear systems in grid sensor networks. It can be shown that, the system matrices of the FM state space model have to satisfy particular conditions for the system to be implementable in real time in a sensor network. This constraint limits the type of systems implementable in real time on sensor networks. A necessary and sufcient condition for a causal transfer matrix to be realizable in the constrained FM model is established. A realization algorithm to derive an FM model that satises the desired condition, given an admissible causal transfer matrix, is also derived. The corresponding problem for the realization of noncausal transfer matrices is also addressed. Index TermsFM model, multidimensional signal processing, multidimensional systems, wireless sensor networks.

Using the Fornasini-Marchesini [11] local state space model, a method for distributed information processing in grid sensor networks has been presented in [12]. The method can be used to implement linear systems in grid sensor networks. The system to be implemented is realized in an FM state space model and is implemented on the sensor network. Stability of distributed 3-D systems implemented on grid sensor networks using the said method is studied in [13].Implementation of an FM model requires communication only between the adjacent sensor nodes. Therefore, the scheme is scalable and lends itself readily for distributed signal processing. A similar method for distributed information processing in sensor networks, based on the GivoneRoesser model [14], was presented in [15]. A. Motivation and Goals It can be shown that, unless information originating from a node can be conveyed over the entire sensor network in a single time slot, system matrices of the FM model have to satisfy a particular condition for the system to be implementable on a sensor network in real time. This constraint limits the type of systems implementable on a sensor network in real time using the method proposed in [12]. Implications of this constraint on the class of systems that are implementable in real time on a sensor network, is studied in this work. A necessary and sufcient condition for a transfer matrix to be realizable in the constrained FM model is established. The result can be used to deduce the type of impulse responses realizable in the constrained FM model. In order to make this paper self-contained the approach proposed in [12] for distributed information processing in grid sensor networks is briey described. The analogues problem for Roesser model based implementations discussed in [15] is addressed in [16]. B. Outline The FM model for 3-D systems is presented in Section II. The constraint that has to be satised by the system matrices of the FM model for implementation in a real-time sensor network is also derived in Section II. The main contribution of this paper is in Section III. A necessary and sufcient condition for a 3-D transfer matrix to be realizable in the constrained FM model is derived. In Section IV, examples are provided to illustrate the theoretical results. Concluding remarks are given in Section V. II. BACKGROUND The FM model [11] is a widely used local state space model for multidimensional systems. It was originally proposed for 2-D systems but can readily be extended to higher dimensions.

I. INTRODUCTION

IRELESS sensor networks are promising candidates for a rich variety of applications. Some applications require sensor nodes to be deployed in a regular grid. Agricultural and environmental monitoring are application areas that often prefer grid or mesh topology. Application of grid sensor networks for contaminant propagation detection and structural health monitoring is discussed in [1] and [2] respectively. Grid sensor networks have drawn considerable attention in the literature. Coverage and connectivity of grid sensor networks in the presence of node failure, sensor deployment strategies, robustness against deployment errors, reliability, routing schemes, and network capacity limits of grid sensor networks have been studied in [3][9]. Employing distributed schemes for signal processing in sensor networks is highly benecial in terms of scalability, bandwidth, and energy consumption. Furthermore such distributed algorithms readily support applications requiring local actuation in response to a local detection [10], yielding minimum response delays as compared to centralized schemes.

Manuscript received September 09, 2010; revised December 24, 2010; accepted April 01, 2011. Date of publication June 20, 2011; date of current version October 28, 2011. This work was supported in part by the National Science Foundation (NSF) under Grant IIS-0325252 (ITR Medium) at the University of Notre Dame, in part by the Defense Threat Reduction Agency (DTRA), and in part by the Naval Surface Warfare Center (NSWC)Crane division under Grant N00164-07-C-8510. This paper was recommended by Associate Editor C.-C. Tseng. The authors are with the Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN 46556 USA (e-mail: msumanas@nd.edu). Digital Object Identier 10.1109/TCSI.2011.2151110

1549-8328/$26.00 2011 IEEE

SUMANASENA AND BAUER: REALIZATION USING THE FORNASINI-MARCHESINI MODEL

2709

A linear system operating on measurements of multiple sensor nodes, that are sampled at periodic time instants, in a 2-D grid sensor network, can be interpreted as a 3-D system. Let there be a spatially and temporally causal 3-D linear process to be implemented on the sensor network. Such a process is rst octant causal and can be represented by the FM model. The FM model for 3-D systems is given by

(1) is the state vector, Z Z, Z Z, and Z Z. where and the output vector . Let the input vector Then, , , , , , , , and . Algorithms for realization of rational transfer functions in the FM model are given in [11], [17][21]. A sufcient condition for a rational transfer matrix to be realizable in the FM model is given in [22]. This condition has been shown to be also necessary in [20]. An algorithm for n-D rational transfer function realization in the Roesser model is given in [23]. A. Implementation in a Sensor Network Let the sensor network be of size . In the model (1), and be the horizontal and vertical let the spatial variables indexes of a node, respectively. Therefore, and . The tuple refers to a unique , , node in the sensor network. Vectors are the state vector, the output vector, and the and at time slot , respectively. The input vector of node following operations are performed by each node at the time slot . and Receive state vectors and input vectors and . Use (1) to compute . and Transmit Use (1) to compute the output. Fig. 1 illustrates the operation of nodes and communication of state vectors between nodes. A system described by model (1) is necessarily rst octant causal. A spatially noncausal impulse response can be decomposed into four quarter plane causal components. Each component can then be realized using the above model. Realization of noncausal 2-D systems is discussed in [24] and [25]. B. Real-Time Implementation Issues is evaluated at the node The state vector at time . It is required by nodes and at time to perform their computations. A real-time implementation thus requires data transmission with zero time delay, which is impossible. There are two options to work around this problem.

Fig. 1. Communication of state vectors between nodes in the network.

One is to allow a time lag between nodes which means the system is not real time. This could be a good solution for small sized sensor networks. The other option is to impose constraints on system matrices such that zero time delay data transmission is not required to perform computations. This would limit the type of systems that can be implemented. These two options are discussed next. 1) Delayed Response Implementation: In an implementaand tion, the problem is that nodes do not receive state vector at time to perform their computations at time . A simple solution is to allow those nodes to do the computations they should do at time slot at time slot instead. This means for each distance unit in either spatial direction there is a time lag of one time slot (one distance unit is equivalent to the distance between two nodes). This time lag could be signicant in a moderate sized sensor network. The time lag can be reduced signicantly if the computation front in the network propagates more than one distance unit . Let the computation front in a sampling interval propagates distance units along either spatial axis in one time slot. The maximum time delay in computing the output is for a sensor network of size [15]. If , the output at every node at time slot can be . computed within the interval 2) Real-Time Implementation: Let the system matrices in the FM model (1) satisfy

(2) (3) From (1) we have

2710

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: REGULAR PAPERS, VOL. 58, NO. 11, NOVEMBER 2011

By substituting for and from (1) in the above equation and using equations (2) and (3)

where , and are the degrees of the polynomial with respect to variables , and matrix respectively. Furthermore , and are the degrees of the polynomial matrix with respect to variables , and respectively. The transfer matrix (4) can be realized in an FM model (1) satisfying conditions (2) and (3), if and only if the conditions (5) and (6) are satised (5) (6) Proof: Necessity is proved by showing that, if a transfer matrix is realizable in an FM model (1) satisfying conditions (2) and (3), it can be represented in the form (4) with conditions (5) and (6) satised. In order to prove sufciency, a realization algorithm which realizes a transfer matrix (4) that satises conditions (5) and (6), in an FM model satisfying conditions (2) and (3), is presented. The proposed algorithm is a modication of the algorithm given in [18][20]. To prove the necessity, let the input-output transfer matrix (4) be realized by the FM model (1) with conditions (2) and (3) satised. From (1) we have

Hence it is sufcient, if node transmits to node , for the latter node to perform its computations at time . Similarly it can be shown only needs to transmit that, node to node , for the latter node to perform its computations at time . Therefore, node can transmit at time instead of which cannot be computed with the information it has at time . This does not affect computations and due to the conditions at nodes (2) and (3). Node can compute state vector and output vector in the interval , when the necessary information is available. The constraints (2) and (3) on system matrices restricts the impulse responses the system could have. III. REALIZABILITY IN THE FM MODEL In this section, a necessary and sufcient condition for a 3-D rational transfer matrix to be realizable in an FM model, that satises conditions (2) and (3), is established. A realization algorithm for the derivation of an FM model that satises desired conditions, given an admissible transfer matrix,1 is presented as part of the proof. A. Causal Transfer Matrices Theorem 1: Let an input-output transfer matrix be given by the right matrix fraction description [26] (4) and where matrices and respectively. Let th elements of matrices respectively2. Let are of size denote

Due to conditions (2) and (3)

(7) Similarly

(8)

and and

(9) denote the identity matrix of order . By substituting Let from (7) and (8) in (9), we have

1A transfer matrix realizable in an FM model that satises conditions (2) and (3). 2In order to make the notation concise the (l; m)th element of a polynomial matrix P (z ; z ; z ) is denoted by P (l; m) in this paper.

SUMANASENA AND BAUER: REALIZATION USING THE FORNASINI-MARCHESINI MODEL

2711

Elements of the transfer matrix th element of functions. The

are rational is denoted by

(12) (13) Then

Let

(10) Monomials in the denominators of the elements of result from products of monomials , and . Therein the denominators of the elements fore, monomials of satisfy . Monomials in the numerresult from products of ators of the elements of , and . Therefore, monomials monomials , , , in the numerators of the elements of satisfy . Due to (10), is equal to the rst . Hence column of where . In order to construct matrices , , , , , , , and that satisfy the above properties, should have the following properties. where should contain all the 1) Entries in monomials in the th column of . This condiand always exist and sattion guarantees that isfy conditions (11) and (12) respectively. say which is not , or 2) For every entry in there should be another element in say such , where . As will be that explained later this condition enables the construction of , , , , , and such that (13) is matrices satised. where should contain at least 3) Vectors one of the terms , and . This is necessitated by condition 2 above. , matrices , , For , , , and are constructed as follows. , if I) For such that . Otherwise . , if for II) For and such that . Otherwise . III) For , if and for and such that . Other. wise IV) if and otherwise. if and otherwise. V) if and otherwise. VI) . VII) VIII) . It can be easily seen that (14) Let where each for is a column vector of size whose elements are monomials of the form where , and are nonnegative integers. Matrices , , , , , , and are constructed such that (11)

such that the th element in the vector is unity and all other elements are zero. Then the same argument can be used to show that

Select

Therefore, can be represented in the form (4) while satisfying conditions (5) and (6). This completes the proof of necessity. In order to prove sufciency, let the transfer matrix (4) satisfy conditions (5) and (6). The realization algorithm given below realizes the transfer matrix (4) in the FM model (1) while satisfying conditions (2) and (3). Without loss of generality it can and be assumed that when . Let and

The basic idea behind the algorithm is to construct a matrix of size containing all the monomials required for the realization. Matrix is of the following form:

2712

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: REGULAR PAPERS, VOL. 58, NO. 11, NOVEMBER 2011

(15) Matrices , , , , , and satisfy condition (13). , , , and realize the Therefore, matrices , , , transfer matrix (4). In addition to realizing the transfer matrix (4), all system matrices are required to satisfy conditions (2) and (3). . There is no monomial in any of the Let due to (6). Therefore, polynomial elements of , . Hence (16) Similarly (17) Let be constructed such that, it does not contain monomials , and . Due to (5) and (6) monomials , and are not required by to satisfy (11) and (12). There is no element in such that . Therefore, for . Hence (18) Since there are no argument or terms in , using a similar (19) Due to (16), (17), (18), and (19) we can deduce that condition (3) is satised by , , , , , and . such that when is a Let and . monomial in the th column of Let for some . By construction, there can be one and only one nonzero element in any row of . be the nonzero element in the th row. Due to (11) Let is in the th column of . If , since is constructed according to steps I and VII, for some . Therefore, by construction of and (given by steps II and III and VII), and for , if for some . Hence (20) Let, if or tion of , Hence, either , if and is not equal to , or and satisfy , then such that . Let . Then by construcfor some . for some or . Therefore, by construction of and , then and for . Similarly if , then for . Therefore, (21) ,

From (3), (20), and (21), we can deduce that condition (2) is satised. Therefore, in addition to conditions 1, 2, and 3, if matrix satises the following conditions, the transfer matrix (4) can be realized in an FM model satisfying conditions (2) and (3). and are not contained in . 4) Monomials , is a monomial in and 5) When , such that . that is not equal to , or and satisfy 6) For , there exist a such that . 7) For not equal to , or satisfying there exist a such that . The algorithm given in Table I can be used to construct the matrix with the above properties. Lines 2 and 3 of the algorithm ensure that all the monomials are contained in . Lines 63 ensure that satin ises condition 5. The monomials missing in that are needed to satisfy conditions 2, 6 and 7 are inserted in the loop starting from line 14. Due to conditions (5) and (6), monomials in vectors satisfy for . Moresatisfy , the algoover, if monomial rithm starts inserting monomials to by inserting or . Therefore, for every monomial inserted into , is inserted into unless . inserted to , is inserted For every monomial . Therefore, conditions 6 and 7 to unless are satised by . It is easily seen that, the matrix constructed by the algorithm satises condition 4 if conditions (5) and (6) hold. This completes the proof of the theorem. For the sake of simplicity, implications of the theorem are discussed for the case of single-input single-output systems. In this case, the right matrix fraction description of the system given by (4), reduces to a rational transfer function. Let the impulse response of the system described by the transfer function (4) be . If conditions (5) and (6) are satised by the transfer function, then we have (22) Therefore, an impulse response, realizable in an FM model (1) which satises conditions (2) and (3), satises the condition (22). The FM model (1) [satisfying conditions (2) and (3)] is implementable in a grid sensor network in which information is conveyed over a single hop in one time slot. The effect of an input at a node propagates over a single hop in a single time slot. Let a rst octant causal impulse response that satises the . It can condition (22) have a rational Z-transform satises conditions (5) and easily be shown that (6). Hence any rst octant causal impulse response satisfying (22) with a rational Z-transform is realizable in an FM model (1) that satises conditions (2) and (3). B. Summary of the Realization Algorithm The algorithm to derive an FM model that satises conditions (2) and (3), given the right matrix fraction description (4) of an admissible transfer function, can be summarized as follows.

SUMANASENA AND BAUER: REALIZATION USING THE FORNASINI-MARCHESINI MODEL

2713

TABLE I ALGORITHM FOR CONSTRUCTING 8

C. Noncausal Transfer Matrices Systems described by the FM model (1) are necessarily rst octant causal. Systems causal in any of the four quadrants of the spatial plane can be realized as

(23) by appropriately choosing and . Here . For example when the region of support of the impulse response lies in the 3rd quadrant of the spatial plane. The following lemma which is a generalization of the Theorem 1 can be used to derive the nal result of this paper. Lemma 1: Let an input-output transfer matrix be given by the right matrix fraction description (24) and where matrices and are of size and respectively. Let th elements of matrices and be denoted by and respectively. Let where

where , and are the degrees of the polynomial with respect to variables , and rematrix , and are the degrees spectively. Furthermore of the polynomial matrix with respect to variables , and respectively. The transfer matrix (24) can be realized in an FM model of the form (23) while satisfying conditions (2) and (3), if and only if the conditions (25) and (26) are satised (25) (26) Compute . Compute using using Proof: For the case of the lemma is a restatement of Theorem 1. For other cases it can be proven using an identical approach to the proof of Theorem 1. A parallel combination of systems causal in each quadrant can be used to realize noncausal systems [24]. can be realized as Theorem 2: A transfer matrix a parallel combination of FM models of the form (23) while satisfying conditions (2) and (3), if and only if it can be expressed as

Construct the matrix using the algorithm given in Table I. and that satises (11) and Derive the matrices (12). Using the steps IVIII derive matrices , , , , , and . and using (15). Compute , .

(27)

2714

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: REGULAR PAPERS, VOL. 58, NO. 11, NOVEMBER 2011

where elements of matrices and are polynomials of , and satisfying the following th elements of and conditions: If the are given by

TABLE II THE SEQUENCE OF OPERATIONS IN THE FIRST ITERATION OF THE ALGORITHM

then (28) (29) Proof: The result trivially follows from Lemma 1. D. Comparison With Roesser Model Based Implementations Using the Roesser model, a method for distributed information processing in grid sensor networks is presented in [15]. It has been shown that unless information originating from a sensor node can be conveyed over the entire sensor network in a single time slot, for a system to be implementable in real time using the Roesser model, system matrices of the Roesser model have to assume a particular from. A causal transfer matrix of the form (4) is realizable in the constrained Roesser model if and only if conditions (5) and (6) are satised [16]. Therefore, a causal transfer matrix realizable in the constrained Roesser model is also realizable in the constrained FM model and vice versa. Similarly a transfer matrix realizable as a parallel combination of constrained Roesser models that are causal in each quadrant is also realizable as a parallel combination of constrained FM models causal in each quadrant.The converse is also true. IV. EXAMPLES The realization algorithm discussed in the above section is illustrated by examples in this section. A. Example 1 Let the transfer matrix to be realized be as follows: Since the system has two inputs, the loop starting at line 1 of the algorithm given in Table I runs twice. The sequence of operations in the rst iteration is given in Table II. At the end of the rst iteration (of the loop starting at line 1 of the algorithm) can be derived in the second iteration we obtain . Vector which is omitted for the sake of brevity. Using and , can be constructed as

(30) The transfer function (30) satises conditions (5) and (6). Therefore, according to Theorem 1 the above transfer matrix is realizable in the FM model (1) with conditions (2) and (3) satised. Matrix is given by:

Matrices as

and

can be derived using (11) and (12)

SUMANASENA AND BAUER: REALIZATION USING THE FORNASINI-MARCHESINI MODEL

2715

Matrix . Matrices , , , , and be derived using the procedure given in Section III-A

can

is used to detect and locate the abrupt variation of concentration Z Z across the propagating front of the contaminant. Here, is the unit impulse function. Coefcients , , , and are derived by an optimization procedure [1]. A delayed response implementation was proposed for the above lter in [1]. The above lter can be modied such that a real-time implementation is possible, as follows: (31) Above modication does not affect the performance if the contaminant propagation is slow. The transfer function of the system having the impulse response (31) is given by (32). Relationship between constant coefcients of the impulse response (31) and the transfer function (32) is not established since it is not required for the purpose of this example. The transfer function (32) is in the form (27) and satises conditions (28) and (29). Therefore, it can be realized as a parallel combination of four FM models of the form (23) while satisfying conditions (2) and (3). See (32)(33) at the top of the next page. , , , , , and , where Let , be the system matrices of the FM model that realizes the th quadrant component of the impulse response (31). Impulse response (31) is even symmetric across the axis axis. Therefore, and odd symmetric across the

where . The realization algorithm given in , , Section III-A can be used to derive the matrices , , , and which realizes the rst quadrant component of the transfer function (32). is given in (33). Vector , which is The matrix constructed using the algorithm given in Table I, is given below:

The sequence of operations of the algorithm is omitted for the sake of brevity. Matrices and can be derived using (11) and (12) as The state vector B. Example 2 A linear ltering based method, for contaminant propagation detection in grid sensor networks, is presented in [1]. Concentration of a contaminant in air is measured by nodes of a rectangular grid sensor network. The abrupt variation of concentration, across the propagating front of the contaminant, is used to detect the propagation of the contaminant. A 2-D edge detection lter having an impulse response of the form Matrix and Section III-A . Matrices , , , , can be derived using the procedure given in in this realization.

Matrix , All other elements of the matrix , trix

and . are equal to zero. Ma, and

2716

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: REGULAR PAPERS, VOL. 58, NO. 11, NOVEMBER 2011

(32) (33)

. All other elements of the matrix equal to zero

are

We have by

. Matrices for

, .

are given

V. CONCLUSION It has been shown that linear systems can be implemented in a distributed fashion in real time on a grid sensor network if the system equations satisfy a particular constraint in the FM model. A necessary and sufcient condition for a causal transfer matrix to be realizable in the constrained FM model was established. A realization algorithm to derive the FM model in the desired form, given an admissible transfer matrix, was also derived. The result can be used to draw conclusions on the type of impulse responses that are realizable in the constrained FM model, and hence are implementable in real time in a distributed grid sensor network. The results derived for the case of causal systems were further extended to noncausal systems. REFERENCES
[1] M. G. B. Sumanasena and P. H. Bauer, Distributed m-D ltering for wave front detection in grid sensor networks, in Proc. 20th IASTED Int. Conf. Parallel Distrib. Comput. Syst., Orlando, FL, Nov. 2008, pp. 423429.

[2] Y. Huang, K. Loewke, K. Schaaf, and S. Nemat-Nasser, Localized SHM with embedded sensor network, in Proc. 5th Int. Workshop Struct. Health Monitor., Stanford, CA, Sep. 2005, pp. 15541561. [3] S. Shakkottai, R. Srikant, and N. Shroff, Unreliable sensor grids: Coverage, connectivity and diameter, in Proc. IEEE INFOCOM, San Francisco, CA, Apr. 2003, pp. 10731083. [4] M. Leoncini, G. Resta, and P. Santi, Analysis of a wireless sensor dropping problem in wide-area environmental monitoring, in Proc. 4th Int. Symp. Inf. Process. Sensor Netw., Los Angeles, CA, 2005, pp. 239245. [5] K. Xu, G. Takahara, and H. Hassanein, On the robustness of grid-based deployment in wireless sensor networks, in Proc. 2006 Int. Conf. Wireless Commun. Mobile Comput., Vancouver, BC, Canada, Jul. 2006, pp. 11831188. [6] H. AboElFotoh, S. Iyengar, and K. Chakrabarty, Computing reliability and message delay for cooperative wireless distributed sensor networks subject to random failures, IEEE Trans. Rel., vol. 54, pp. 145155, Mar. 2005. [7] H. M. F. AboElFotoh, E. S. Elmallah, and H. S. Hassanein, A owbased reliability measure for wireless sensor networks, Int. J. Sensor Netw., vol. 2, pp. 311320, Jul. 2007. [8] A. Akbar, W. Mansoor, S. Chaudhry, A. Kashif, and K. Kim, Node-link-failure resilient routing architecture for sensor grids, in Proc. 8th Int. Conf. Adv. Commun. Technol., Phoenix, AZ, Feb. 2006, pp. 131135. [9] G. Barrenechea, B. Beferull-Lozano, and M. Vetterli, Lattice sensor networks: Capacity limits, optimal routing and robustness to failures, in Proc. 3rd Int. Symp. Inf. Process. Sensor Netw., Berkeley, CA, Apr. 2004, pp. 186195. [10] W. Hu, N. Bulusu, and S. Jhan, A communication paradigm for hybrid sensor/actuator networks, Int. J. Wireless Inf. Netw., vol. 12, pp. 4759, Oct. 2005. [11] E. Fornasini and G. Marchesini, Doubly-indexed dynamical systems: State-space models and structural properties, Math. Syst. Theory, vol. 12, pp. 5972, Dec. 1978. [12] D. A. Dewasurendra and P. H. Bauer, A novel approach to grid sensor networks, in 15th IEEE Int. Conf. Electron., Circuits Syst., Malta, Aug. 2008, pp. 11911194. [13] M. G. B. Sumanasena and P. H. Bauer, Stability of distributed 3-D systems implemented on grid sensor networks, IEEE Trans. Signal Process., vol. 58, pp. 44474453, Aug. 2010. [14] D. Givone and R. Roesser, Multidimensional linear iterative circuitsgeneral properties, IEEE Trans. Comput., vol. C-21, pp. 10671073, Oct. 1972. [15] M. G. B. Sumanasena and P. H. Bauer, A Roesser model based multidimensional systems approach for grid sensor networks, in Conf. Rec. 44th Asilomar Conf. Signals, Syst., Comput. (ASILOMAR), Pacic Grove, CA, Nov. 2010, pp. 21512155. [16] M. G. B. Sumanasena and P. H. Bauer, Realization using the Roesser model for implementations in distributed grid sensor networks, Multidimensional Syst. Signal Process., vol. 22, no. 13, pp. 131146, Mar. 2011. [17] M. Bisiacco, E. Fornasini, and G. Marchesini, Dynamic regulation of 2-D systems: A state space approach, Linear Algebra Its Appl., vol. 122, pp. 195218, 1989.

SUMANASENA AND BAUER: REALIZATION USING THE FORNASINI-MARCHESINI MODEL

2717

[18] L. Xu, L. Wu, Q. Wu, Z. Lin, and Y. Xiao, On realization of 2-D discrete systems by Fornasini-Marchecini model, Int. J. Control, Autom., Syst., vol. 3, pp. 631639, Dec. 2005. [19] L. Xu, Q. Wu, Z. Lin, and Y. Xiao, A new constructive procedure for 2-D coprime realization in Fornasini Marchesini model, IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 54, pp. 20612069, Sep. 2007. [20] H. Cheng, T. Saito, S. Matsushita, and L. Xu, Realization of multidimensional systems in Fornasini-Marchesini state-space model, Multidimensional Syst. Signal Process., pp. 115, Sep. 2010 [Online]. Available: http://www.springerlink.com/content/t134v786967m5u50/ [21] C. Dubi, Triangular realization of rational functions of n complex variables, Multidimensional Syst. Signal Process., vol. 19, pp. 123129, Mar. 2008. [22] D. Alpay and C. Dubi, A realization theorem for rational functions of several complex variables, Syst. Control Lett., vol. 49, pp. 225229, Jul. 2003. [23] L. Xu, H. Fan, Z. Lin, and N. K. Bose, A direct-construction approach to multidimensional realization and LFR uncertainty modeling, Multidimensional Syst. Signal Process., vol. 19, pp. 323359, Dec. 2008. [24] S. G. Lele and J. M. Mendel, Modeling and recursive state estimation for two-dimensional noncausal lters with applications in image restoration, IEEE Trans. Circuits Syst., vol. CAS-34, pp. 15071517, Dec. 1987. [25] L. Ntogramatzidis, M. Cantoni, and R. Yang, On the partial realization of noncausal 2-D linear systems, IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 54, pp. 18001808, Aug. 2007. [26] N. K. Bose, Multidimensional Systems Theory and Applications. Norwell, MA: Kluwer, 2003.

M. G. Buddika Sumanasena (M11) received the B.Sc. degree in engineering and the M.Sc. degree in telecommunication engineering from the Department of Electronics and Telecommunication Engineering, University of Moratuwa, Sri Lanka, in 2004 and 2006, respectively. He is currently working toward the Ph.D. degree in the Department of Electrical Engineering, University of Notre Dame, Notre Dame, IN. His research interests include digital signal processing, multidimensional systems theory, distributed signal processing, and computational photography.

Peter H. Bauer (F05) received the Diplom degree from the Technical University of Munich, Munich, Germany, in 1984 and the Ph.D. degree from the University of Miami, Coral Gables, FL, in 1987, both in electrical engineering. In 1999, he was the Director of the University of Notre Dame, Notre Dame, IN, EE/CSE London Program, London, U.K. He is currently a Professor in the Department of Electrical Engineering, University of Notre Dame, where he is the Head of the Mobile Sensor Systems (MOSES) Laboratory. He is the author or coauthor of over 150 technical papers and six book chapters. His research interests include sensor and actuator networks, mobile wireless sensing, congestion control, evidential ltering and fusion, stability theory of discrete-time systems, and control of power generation systems. Dr. Bauer served as an Associate Editor for the IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSPART II (19891991, 20032005) and served as an Associate Editor for the journal Multidimensional Systems and Signal Processing (20002007). He received the E. I. Jury Award from the University of Miami in 1987, the NASA Technical Innovator Award in 1992, the Alexander von Humboldt Fellowship in 1997, and the University of Notre Dame Kaneb Center Best Teacher Award in 2005.

You might also like