State-Space Representation

The State-Space: Why Do I Need It?

The state of a system as any set of quantities which must be specified at a given time in order to completely determine the
behavior of the system.

The quantities constituting the state are called the state variables, and the hypothetical space spanned by the state variables
is called the state-space.

Example 1.1

The governing differential equation for the motion of a simple pendulum on which no external input is applied (Figure 1.1)
is given by Eq. (1.1).
If we apply a torque, M(t), about the hinge, O, as an input to the pendulum, the
governing differential equation can be written as

where θ(2) (t) represents the second order time derivative of θ(t), as per our notation
(i.e. d2θ(t)/dt2 = θ(2) (t)). Let the output of the system be the angle, θ(t), of the
pendulum. We would like to determine the state of this system.
To begin, we must know how many quantities (i.e. state variables) need to be specified to completely determine the motion
of the pendulum.
we know that for a system of order n, we have to specify precisely n initial conditions to solve the governing differential
equation.
Hence, it must follow that the state of an nth order system should consist of precisely n state variables, which must be
specified at some time (e.g. t = 0) as initial conditions in order to completely determine the solution to the governing
differential equation.
Here we are dealing with a second order system - which implies that the state must consist of two state
variables
Let us call these state variables x1(t) and x2(t), and arbitrarily choose them to be the following:

The state-space is thus two-dimensional for

the simple pendulum whose axes are x1(t) and
x2(t) (Figure 1.2).
It is now required that we express the governing differential equation (Eq. (1.2)) in terms of the state variables defined
by Eqs. (1.3) and (1.4). Substituting Eqs. (1.3) and (1.4) into Eq. (1.2), we get the following first order differential
equation:

Have we transformed a second order differential equation (Eq. (1.2)) into a first order differential equation (Eq. (1.5))
by using the state-variables?
Not really, because there is a another first order differential equation that we have forgotten about - the one obtained by
substituting Eq. (1.3) into Eq. (1.4), and written as

Equations (1.5) and (1.6) are two first order differential equations, called the state equations, into which the governing
equation (Eq. (1.2)) has been transformed.
The order of the system, which is its important characteristic, remains unchanged when we express it in terms of the
state variables. In addition to the state equations (Eqs. (1.5) and (1.6)), we need an output equation which defines the
relationship between the output, θ (t), and the state variables x1(t) and x2(t). Equation (1.3) simply gives the output
equation as

The state equations, Eqs. (1.5) and (1.6), along with the output equation, Eq. (1.7), are called the state-space
representation of the system.
Instead of choosing the state variables as θ (t) and θ(1)(t), we could have selected a different set of state variables, such as

Although the state-space representation given by Eqs. (1.8)-(1.12) is different from that given by Eqs. (1.5)-(1.7), both
descriptions are for the same system.
Hence, the state-space representation of a system is not unique, and all legitimate state-space representations should give
the same system characteristics.
While we have freedom to choose our state variables, we have to ensure that we have chosen the minimum number of
state variables that are required to describe the system.
However, if we are deriving the state-space representation from a transfer function (or transfer matrix), some poles may be
canceled by the zeros, thereby yielding an erroneous order of the system which is less than the correct order.

Example 1.2
Consider a system with input, u(t), and output, y ( t ) , described by the following differential
equation:

where a and b are positive constants. The transfer function, Y(s)/U(s), of this system can be obtained by taking the Laplace
transform of Eq. (1.13) with zero initial conditions, and written as follows:

In Eq. (1.14), if we cannot resist the temptation to cancel the pole at s = a with the zero at s = a, we will be left with the
following transfer function:

which yields the following incorrect differential equation for the system:

This example illustrates one of the hazards associated with the transfer function description of a system, which can be
avoided if we directly obtain state-space representation from the governing differential equation.
Another cause of illegitimacy in a state-space representation is when two (or more) state variables are linearly dependent.

For example, if x1(t) = θ(t) is a state variable, then X2(t) = Lθ(t) cannot be another state variable in the same state-space
representation, because that would make x1(t) and x2(t) linearly dependent.
In general, for a system of order n, if x1(t), x2(t),..., xn-1(t) are state variables, then xn(t) is not a legitimate state variable if it
can be expressed as a linear combination of the other state variables given by:

Example 1.3
Consider an inverted pendulum on a moving cart, for which the governing differential equations are the following:

where m and L are the mass and length, respectively, of the

inverted pendulum, M is the mass of the cart, θ(t) is the angular
position of the pendulum from the vertical, x(t) is the horizontal
displacement of the cart, f(t) is the applied force on the cart in
the same direction as x(t) (see Figure 1.3), and g is the
acceleration due to gravity. Assuming f ( t ) to be the input to
the system, and x(t) and θ(t) to be the two outputs, let us derive
a state-space representation of the system.
The system is described by two second order differential equations; hence, the order of the system is four.

Thus, we need precisely four linearly independent state-variables to describe the system.

When dealing with a physical system, it is often desirable to select physical quantities as state variables.

Let us take the state variables to be the angular position of the pendulum, θ(t), the cart displacement, x(t) , the angular
velocity of the pendulum, θ(1)(t)and the cart's velocity, x(1)( t ). We can arbitrarily number the state variables as follows:
Note that due to the nonlinear nature of the system, we cannot express the last two state-equations (Eqs. (1.26), (1.27))
in a form such that each equation contains the time derivative of only one state variable.
Such a form is called an explicit form of the state-
equations.
If the motion of the pendulum is small about the equilibrium point,θ = 0, we can linearize Eqs. (1.26) and (1.27) by
assuming cos(θ(t))= cos(x1(t) =1, sin(θ(t) = sin(x1(t)) = x1(t), and [θ(1) (t)]2 sin(θ(t)) = [ x3(1) (t)]2 sin ( x1(t)) =0.
The corresponding linearized state equations can then be written in explicit form as follows:

The linearized state-equations of the system, can be expressed in the following matrix form, where all coefficients are
collected together by suitable coefficient matrices :
Taking a cue from Example 1.3, we can write the state-equations of a general linear system of order n, with m inputs and
p outputs, in the following matrix form:
 Example 1.4
Consider the electrical network presented in Fig. 1.4 whose governing differential equations are as follows:

Looking at Eq. (1.37), we find that the time derivative of the input appears on the right-hand side.
For a linear, time-invariant state-space form of Eqs. (1.34) and (1.35), the state variables must be selected in such a way that
the time derivative of the input, e(1)(t), vanishes from the state and output equations.
One possible choice of state variables which accomplishes this is the following:

‫اختيار الستيس‬
‫فرضية حسب‬
‫سلوك الدائرة‬

Then the first state-equation is obtained by substituting Eq. (1.38) into Eq. (1.39), and expressed
as

Substitution of Eqs. (1.38) and (1.39) into Eq. (1.37) yields the second state-equation, given by

The output equation is given by using Eq. (1.36) as follows:

In the matrix notation, Eqs. (1.40)-(1.42) are expressed as

A general linear (or nonlinear) system may have several higher-order time derivatives of the input in its governing
differential equations such as Eq.(1.45)
To simplify the selection of state variables in such cases, it is often useful to first draw a schematic diagram of the
governing differential equations.
Example 1.5
The system of Example 3.4 has two governing equations, Eqs. (1.36) and (1.37). While Eq. (1.36) is an algebraic
equation (i.e. a zero order differential equation), Eq. (1.37) is a second order differential equation. Let us express Eq.
(1.37) in terms of a dummy variable (so called because it is neither a state variable, an input, nor output) z ( t ) , such that

The schematic diagram of Eqs. (1.46) and (1.47) is drawn in Figure 1.6
The state-equations is given by

The output equation is obtained as follows: 𝒚 𝟏 ( 𝒕 )=𝒊𝟏 ( 𝒕 ) − 𝒊𝟐 (𝒕 )

From Figure 1.6 :

Substitute in Eq. 1.47:

𝑹𝟑 𝒙 𝟐 (𝒕 )
𝒊𝟐 ( 𝒕 ) = (𝟏 . 𝟓𝟏 )
𝑳 ( 𝑹 𝟏 + 𝑹𝟑 )

( )
𝟐
From Eqs. (1.36) and (1.51) 𝟏 𝑹𝟑
𝒊𝟏 ( 𝒕 ) = 𝒆 (𝒕)+ 𝒙 𝟐 ( 𝒕 ) ( 𝟏. 𝟓𝟐)
𝑹𝟏 + 𝑹𝟑 𝑳 ( 𝑹 𝟏 + 𝑹𝟑 )

Substitute Eqs. (1.51) and (1.52) in the output equation we have

Example 1.6
Let us find a state-space representation using the schematic diagram for the system with input, u(t), and output, y(t),
described by the following differential equation:

Since the right-hand side of Eq. (1.56) contains time derivatives of the input, we should introduce a dummy variable, z(t),
as follows:
Figure 1.7 shows the schematic diagram of Eqs. (1.57) and (1.58). Note that Figure 1.7 has n integrators arranged in a series.

let us choose the state variables to be the integrator outputs, and number them beginning from the right of Figure 1.7. Then
the state-equations are as follows:
The output equation is obtained by substituting the definitions of the state variables, namely, x1(t)= z(t), x2(t) = z(1)(t), ...,
xn(t) = z(n-1)(t) into Eq. (1.58), thereby yielding:

and substituting Eq. (1.59d) into Eq. (1.60), the output equation is expressed as follows:
Such a state-space representation has a name: The controller companion form or Controllable Canonical form
Another companion form, called the observer companion form, is obtained as follows for the system obeying Eq. (1.56). In
Eq. (1.56) the terms involving derivatives of y(t) and u(t) of the same order are collected, and the equation is written as
follows:

On taking the Laplace transform of Eq. (1.64) subject to zero initial conditions, we get the following:

Dividing Eq. (1.65) by Sn leads to

Figure 1.8 shows the schematic diagram of Eq. (1.66).
If we select the outputs of the integrators as state variables beginning from the left of Figure 1.8, we get the following state
equations:
Example 1.7
Consider the electrical network for an amplifier-motor shown in Figure 1.9.

It is desired to change the angle, θ(t), and angular velocity, θ(1)(t), of a load attached to the motor by changing the input
voltage, e(t), in the presence of a torque, TL(t)applied by the load on the motor.
The governing differential equations for the amplifier-motor are the following:

where J, R, L, and b are the moment of inertia, resistance, self-inductance, and viscous damping-coefficient of the motor,
respectively, and a is a machine constant. R0 and KA are the resistance and voltage amplification ratio of the amplifier.

Since the loading torque, TL(t)acts as a disturbance to the system, we can consider it as an additional input variable. The
input vector is thus given by u(t) = [e(t); TL(t)]T ,The output vector is given by y(t)=[θ(t); θ(1)(t)]T.
Let us select the state variables as x1(t) = θ(t), x2(t) = θ(1)(t) and x3(t) = i(t). Then the state-equations can be written as
follows:
In matrix form, the state – equation and output equation are written as follows:

(1.77)

[ ]
𝑥 1 (𝑡 )
[ 𝑦 1 (𝑡 )
𝑦2( 𝑡 )
=
] [ 1
0
0
1
0
0 ] 𝑥 2 (𝑡 )
𝑥 3 (𝑡 )
( 1 . 78 )
Linear Transformation of State-Space Representations
Since the state-space representation of a system is not unique, we can always find another state-space representation for the
same system by the use of a state transformation.
State transformation refers to the act of producing another state-space representation, starting from a given state-space
representation.
If a system is linear, the state-space representations also are linear, and the state transformation is a linear transformation in
which the original state-vector is pre-multiplied by a constant transformation matrix yielding a new state-vector.
Suppose T is such a transformation matrix for a linear system described by Eqs. (1.34) and (1.35). Let us find the new state-
space representation in terms of T and the coefficient matrices, A, B, C, D. The transformed state-vector, x'(f), is expressed
as follows:

Equation (1.79) is called a linear state-transformation with transformation matrix, T. Note that for a system of order n, T
must be a square matrix of size (n × n), because order of the system remains unchanged in the transformation from x(t) to
x’(t).
It is possible to transform the new state-vector, x’(t), back to the original state-vector, x(t), with the use of the following
inverse transformation:
To find the transformed state-equation, yielding

Pre-multiplying both sides of Eq. (1.81) by T, we get the transformed state-equation as follows:

We can write the transformed state-equation Eq. (1.81) in terms of the new coefficient matrices A', B', C', D', as follows:

where A' = TAT-1, and B' = TB, C' = CT-1, and D' = D.
System Characteristics from State-Space Representation
The characteristics of a system can be obtained by its characteristic equation, whose roots are the poles of the system. The
locations of the poles indicate a system's performance - such as natural frequency, damping factor, system type - as well
as whether the system is stable. Let us see how a system's characteristic equation can be derived from its state-space
representation.
The characteristic equation is defined in to be the denominator polynomial of the system's transfer function (or transfer
matrix) equated to zero. Hence, we should first obtain an expression for the transfer matrix in terms of the state-space
coefficient matrices, A, B, C, D.
Taking the Laplace transform of both sides of the matrix state-equation assuming zero initial conditions (i.e. x(0) = 0)
yields the following result:
Example 1.8
For a linear system described by the following state coefficient matrices, let us determine the transfer function and the
characteristic equation:
Special State-Space Representations: The Canonical Forms

The companion forms are members of a special set of state-space representations, called canonical forms. In addition to the
companion forms, another canonical form is the Jordan canonical form, which is derived from the partial fraction
expansion of the system's transfer matrix as described below.

For simplicity, consider a single-input, single-output system with the transfer function given by the following partial
fraction expansion:

Let us select the state-variables of the system as follows:

Taking the inverse Laplace transform of Eq. (1.92), we get the following state-equations:
Example 1.9
Let us find the Jordan canonical form of the following system: