IEEE TRANSACTIONS ON EDUCATION, VOL. 39, NO.

1, FEBRUARY 1996

97

REFERENCES
R. Inigo and J. Morton, Simulation of the dynamics of an industrial robot, IEEE Trans. Educ., vol. 34, no. 1, pp. 89-99, Feb. 1991. H. Pota, A prototype flexible robot arm-An interdisciplinary undergraduate project, IEEE Trans. Educ., vol. 35, no. 1, pp. 83-90, Feb. 1992. W. Richard, An educational image processing/machine vision system, IEEE Trans. Educ., vol. 34, no. 1, pp. 129-133, Feb. 1991. C. Vibet, Elements to teach robot control design, IEEE Trans. Educ., vol. 36, no. 1, pp. 45-51, Feb. 1993. N. Vira and E. Tunstel, Use of symbolic computation in robotics education, ZEEE Trans. Educ., vol. 35, no. 1, pp. 18-31, Feb. 1992.

MIMO Systems--Transfer Function to State-Space

Hemanshu Roy Pota
Abstruct-In this short note we present a very simple method, useful for classroom teaching, to get a minimal state-space representation (with the exception of systems where there is a pole-zero cancellation) from a transfer function matrix. The method is direct and does not involve the intermediate step of obtaining nonminimal realization-which further requires system reduction routines-and hence is suitable as a compact self-contained topic which has been hitherto neglected in undergraduate linear control theory curriculum.

is attacking problems where transfer function matrices arise naturally from the physics of the problem [6]. Most of the practical control systems are multivariable systems, and the students should be taught the connection between a transfer function matrix and its state-space realization in a simple way. Students, when taught this connection using methods given in [4] and [5],lose the physical feel behind the definition of system state and presume that controllability and observability are more important issues than the system representation itself. In this short note, a simple procedure is given that can be used to get a minimal state-space realization (with the exception of systems where there is a pole-zero cancellation) directly from a transfer function matrix by hand calculation, without going through the intermediate nonminimal realization stage. The theory behind this realization is not new [3] and the main motivation for writing this short note arises from the fact that the author could not find a single book where such a simple method is given. This work aims at providing material in a suitable form to enable the inclusion of the realization of state-space from a transfer function matrix in an undergraduate curriculum, giving students a tool to have a first go at multivariable control design for problems such as those that arise in [6]. 11. TRANSFER FUNCTION TO STATE-SPACE Let a general multivariable system (1 inputs and p outputs) transfer function matrix be given as follows:

I. INTRODUCTION
The undergraduate linear control theory curriculum can be divided into two parts. The introductory part of the curriculum deals with classical frequency domain material and the advanced part is devoted to state-space theory. Single-input single-output (SISO) transfer functions are central to the teaching of the classical frequency domain theory. Considerable part of the curriculum is devoted to deriving state-space canonical forms from the SISO transfer functions [ 2 ] , [ 4 ] ,[ 5 ] . The idea of minimal state order and the associated notions of controllability and observability are discussed as a prelude to the state-space theory. With the state-space theory, the fact that the system under study is SISO or multi-input multi-output (MIMO) is less relevant. Hence, the knowledge of the state-space theory enables the students to attack MIMO control problems, provided a state-space representation is available; this is the case in some practical situations, while in other situations transfer function matrices arise naturally. Many books on linear systems theory [4], [5] discuss the derivation of a state-space representation from a transfer function matrix. The discussion normally starts with obtaining either block observer or block controller state-space (nonminimal) realization, which is followed by algorithms to obtain minimal realization (both controllable and observable) from these nonminimal realizations. These algorithms [5] are suitable only for a digital computer implementation and are more of an exercise in linear algebra than in controller design. The consequence is that this material has to be left out of classroom teaching (even in standard text books such as [l] and [2], this material is omitted) and the students get little confidence
Manuscript received December 31, 1992; revised June 9, 1993

Let us consider two examples before giving a general procedure to derive a state-space representation of the above MIMO system (1).

A. Examples The first example is of a multi-input single-output system and the second example is of a two-input two-output system. In the general method, to be presented later in this paper, every multivariable system ( I inputs and p outputs) is first reduced to p multi-input singleoutput systems. Each of the inputs to these p multi-input single-output systems is an output of a single-input single-output system. The procedure to get a state-space representation for a multi-input singleoutput system used in the first example can be used to get the final state-space representation. The second example is chosen so that this two-step procedure is made clear to the reader. Example 1: The transfer function matrix for this example system is

Y ( s )=

(s

+ a)(. + b)
+

(s

The author is with the Department of Electrical Engineering, University College, University of New South Wales, Australian Defence Force Academy, Canberra ACT 2600, Australia. Publisher Item Identifier S 0018-9359(96)02136-X.

The first step in writing state-space equations for a MIS0 system is to pull out all the common factors in the denominator polynomials. Notice that the factor ( s a ) is common to both the columns. Pulling this term out, the above transfer function matrix can be written as

0018-9359/96$05.00 0 1996 IEEE

98

IEEE TRANSACTIONS ON EDUCATION, VOL. 39, NO 1, FEBRUARY 1996

As a first step in getting the state-space equations, the system left after taking the common factor ( s a) is considered. Since there is no common term left between the two transfer functions, each of them can be considered as a separate SISO system and the state variables defined as follows: Zll(S) A ELM Z,,(S) A sz+ds+e) (s+b) 211 bz11 = u l ( t ) 321 i 2 1 d ezzl = u ~ ( t ) A . A Lc1 = 211 2 2 = ai, 2 3 = 221 21 = -bzl ul(t) kz = -dxZ - ex3 u ~ ( t ) 5 3 = x2.

The above system (8) can be seen as two separate two-input oneoutput systems with the following state-space representation:

+ + +
+

+ by1 = + = 23 = -b23 + +
$1

211

Z21

$2
24

+ cy2 = +
Zll

Z21
2 2

23

Yl

= yz

21

2 2

54

=-CZ~

+ +
~1

Again collecting all the differential equations involving the state variables, a compact matrix representation can be written as follows:

Note that sZ,,(s) = iz,(t) (lower case is used for the variable representation in the time-domm and the upper case for the Laplace domain). With the above definition of state variables and the inter, system equations mediate output variables Z l l ( s ) and Z ~ l ( s )the can be rewritten as follows:

The state-space representation of the above system can be written by using the standard observer canonical form of the SISO systems [5] (note from above that zl1 = x1 and 221 = x3)
y

+ a y = x i + xz + cz3 . + + ex3 - ax4 .

22

Let

24

6 y ( t ) , then we can write

54 = 21

The state order four, which is as minimal as can be verified from the result in [3]. B. General Procedure In this section a procedure to derive a state-space representation of a general MIMO system (1) is given. Each term of the matrix, is a ratio of rational polynomials such that the order of d,,(s) is greater than or equal to the order of n that a transfer function matrix is available with all its constituent polynomials given in pole-zero form. The procedure to get a minimal state-space realization of system (l), directly fro function matrix, can be stated as: 1) Set i = 1. 2) Select the ish column of the transfer function matrix. 3 ) Find the largest common factor in the denominator polynomials of the ith column vector. Let this common factor be D z l ( s ) , then define Z,, (s) A 4) Pull out the common factor D z l ( s )from the ith column and then find a factor that is common to most of the remainand then define ing denominator polynomials, call it D Z 2 ( s ) , z22(s) 4z , l o

Putting all the above equations together in a compact matrix notation, we have -b 0 0 0 -d -e = 0 1 0 1 l c

#,

m.

Example 2: The transfer function matrix for the second example system is given below.

To get a state-space representation of (7),we first consider one input at a time, i.e., one column of the transfer function matrix Here, let us consider both the columns side by side and define
Zll(S) 5 %
,211
21

= Zll

a511

(s+a)

= ui(t)

a UZ(S) 221(s) = (s+d) dZZld = uz(t)

22

5 ) Pull out D12( s ) from the column vector in Step 4) and then repeat Step 4) until there is no common factor between any of the two denominator polynomials of the column vector. 6) If i < I, set i = i 1 and go to Step 2). 7) After all the common terms of each column have been duly considered, we can rewrite system (1) as

D,z(a).

51 = --a21

=z z 1

u1(t)

i z = -ax2

+ W ( t ).
where a) Wz3(s) (input variables for the modified system) IS equal to the last Z,, (s) that was pulled o element of the transfer function matrix = Zll(s). In Example 2, W ~ l ( s ) h Z 3 ( s are ) the terms of the denominator of the (i,g)th element of the transfer function matrix that remain after

There is an order in how the various intermediate variables, Z,, (s), are defined. First, all of the common factors in the denominators of 1 1 ( s ) is defined as the intermediate column 1 are pulled out and 2 output due to that common denominator; 221 ( s ) is similarly defined. ) 2 2 1 ( s ) , we can write the With the above definitions of Z ~ i ( s and system transfer function matrix (7) as

b)

IEEE TRANSACTIONS ON EDUCATION, VOL. 39, NO. 1, FEBRUARY 1996

99

having pulled out all the common terms in Steps 3)-5). In Example 2, h l l ( s ) = ( 3 b ) .

representation can be written by hand calculation-an for classroom teaching.

ideal situation

8) Now the above system (11) can be seen as p independent single-input multi-output systems (i = 1,.. . , p )

1 1 1 . CONCLUSION
A direct method to write down state-space equations (of minimal order) for a given transfer function matrix is presented in this paper. The method is simple enough for hand implementation and does not divert the attention from the simple physical meaning of the system state to issues relating to controllability and observability. This method is ideal for classroom teaching and is just as useful in analyzing many practical systems [6].
R E F E R E N C E S [I] G. F. Franklin, J. D. Powell, and A. Emami-Naeini, Feedback Control of Dynamic Systems. Sydney, Australia: Addison-Wesley, 1986. [2] J. J. DAzzo and C. H. Houpis, Linear Control Systems Analysis and Design. Sydney, Australia: McGraw-Hill, 1981. [3] H. H. Rosenbrock, Efficient computation of the least order for a given transfer function matrix, Electron. Lett., vol. 3, pp. 413-414, 1967. [4] M. R. Chidambara and S . Ganapathy, An Znfroduction to Control of Dynamic Systems. Faridabad, India: Sehgal Educational Consultants, 1979. [5] T. Kailath, Linear Systems. Englewood Cliffs, NJ: Prentice-Hall, 1980. [6] H. R. Pota and T. E. Alberts, Multivariable transfer functions for a slewing piezoelectric laminate beam, ASME J. Dynamic Systems, Measurement, and Control, vol. 117, no. 3, pp. 352-359, Sept. 1995. [7] P. K. Sinha, Multivariable Control. New York Marcel Dekker, 1984.

Note that the order of this multi-input single-output system [3] is the least common denominator of all the terms in (12) above. The obvious, mechanical procedure to wnte the statespace representation of the system (12) is to write the right-hand side of (12) over its lowest common denominator and then write an observer canonical [5] form of state-space representation. Although considerable savings in efforts to write the equations can be achieved if common factors in h t 3 ( s )are first spotted and pulled out as was done in Example 1. The procedure given above will result in a minimal realization provided there is no system pole-zero [7] cancellation; it should be noted that it is important to consider the cases where there is a pole-zero cancellation, but for a majority of cases [6] one can proceed without considering it and adequate check can be performed in due course. This method can be easily implemented on a computer, and in most cases where the denominator polynomials of the transfer function matrix has large common factors, a state-space