systems with observers. Section 10–8 discusses quadratic optimal regulator systems.

Note
that the state feedback gain matrix K can be obtained by both the pole-placement method
and the quadratic optimal control method. Finally, Section 10–9 presents robust control
systems. The discussions here are limited to introductory subjects only.

10–2 POLE PLACEMENT

In this section we shall present a design method commonly called the pole-placement or
pole-assignment technique. We assume that all state variables are measurable and are
available for feedback. It will be shown that if the system considered is completely state
controllable, then poles of the closed-loop system may be placed at any desired locations
by means of state feedback through an appropriate state feedback gain matrix.
The present design technique begins with a determination of the desired closed-loop
poles based on the transient-response and/or frequency-response requirements, such as
speed, damping ratio, or bandwidth, as well as steady-state requirements.
Let us assume that we decide that the desired closed-loop poles are to be at s=m1 ,
s=m2 , p , s=mn . By choosing an appropriate gain matrix for state feedback, it is pos-
sible to force the system to have closed-loop poles at the desired locations, provided
that the original system is completely state controllable.
In this chapter we limit our discussions to single-input, single-output systems. That
is, we assume the control signal u(t) and output signal y(t) to be scalars. In the deriva-
tion in this section we assume that the reference input r(t) is zero. [In Section 10–7 we
discuss the case where the reference input r(t) is nonzero.]
In what follows we shall prove that a necessary and sufficient condition that the
closed-loop poles can be placed at any arbitrary locations in the s plane is that the sys-
tem be completely state controllable. Then we shall discuss methods for determining
the required state feedback gain matrix.
It is noted that when the control signal is a vector quantity, the mathematical aspects
of the pole-placement scheme become complicated. We shall not discuss such a case in
this book. (When the control signal is a vector quantity, the state feedback gain matrix
is not unique. It is possible to choose freely more than n parameters; that is, in addition
to being able to place n closed-loop poles properly, we have the freedom to satisfy some
or all of the other requirements, if any, of the closed-loop system.)

Design by Pole Placement. In the conventional approach to the design of a single-

input, single-output control system, we design a controller (compensator) such that the
dominant closed-loop poles have a desired damping ratio z and a desired undamped
natural frequency vn . In this approach, the order of the system may be raised by 1 or 2
unless pole–zero cancellation takes place. Note that in this approach we assume the ef-
fects on the responses of nondominant closed-loop poles to be negligible.
Different from specifying only dominant closed-loop poles (the conventional design
approach), the present pole-placement approach specifies all closed-loop poles. (There is
a cost associated with placing all closed-loop poles, however, because placing all closed-
loop poles requires successful measurements of all state variables or else requires the in-
clusion of a state observer in the system.) There is also a requirement on the part of the
system for the closed-loop poles to be placed at arbitrarily chosen locations.The requirement
is that the system be completely state controllable. We shall prove this fact in this section.

Section 10–2 / Pole Placement 723

 Consider a control system
#
x = Ax + Bu (10–1)
y = Cx + Du

where x = state vector (n-vector)

y = output signal (scalar)
u = control signal (scalar)
A = n * n constant matrix
B = n * 1 constant matrix
C = 1 * n constant matrix
D = constant (scalar)
We shall choose the control signal to be
u = -Kx (10–2)
This means that the control signal u is determined by an instantaneous state. Such a
scheme is called state feedback. The 1*n matrix K is called the state feedback gain
matrix. We assume that all state variables are available for feedback. In the following
analysis we assume that u is unconstrained. A block diagram for this system is shown in
Figure 10–1.
This closed-loop system has no input. Its objective is to maintain the zero output.
Because of the disturbances that may be present, the output will deviate from zero. The
nonzero output will be returned to the zero reference input because of the state feed-
back scheme of the system. Such a system where the reference input is always zero is
called a regulator system. (Note that if the reference input to the system is always a
nonzero constant, the system is also called a regulator system.)
Substituting Equation (10–2) into Equation (10–1) gives
#
x(t) = (A - BK) x(t)
The solution of this equation is given by
x(t) = e(A - BK)tx(0) (10–3)
where x(0) is the initial state caused by external disturbances.The stability and transient-
response characteristics are determined by the eigenvalues of matrix A-BK. If matrix

.
u x x +
B + 兰 C +
+

A
Figure 10–1
Closed-loop control
system with –K
u=–Kx.

724 Chapter 10 / Control Systems Design in State Space

K is chosen properly, the matrix A-BK can be made an asymptotically stable matrix,
and for all x(0) Z 0, it is possible to make x(t) approach 0 as t approaches infinity. The
eigenvalues of matrix A-BK are called the regulator poles. If these regulator poles are
placed in the left-half s plane, then x(t) approaches 0 as t approaches infinity. The prob-
lem of placing the regulator poles (closed-loop poles) at the desired location is called a
pole-placement problem.
In what follows, we shall prove that arbitrary pole placement for a given system is
possible if and only if the system is completely state controllable.

Necessary and Sufficient Condition for Arbitrary Pole Placement We shall now
prove that a necessary and sufficient condition for arbitrary pole placement is that the
system be completely state controllable.We shall first derive the necessary condition.We
begin by proving that if the system is not completely state controllable, then there are
eigenvalues of matrix A-BK that cannot be controlled by state feedback.
Suppose that the system of Equation (10–1) is not completely state controllable.
Then the rank of the controllability matrix is less than n, or

rankCB 哸 AB 哸 p 哸 An - 1 BD = q 6 n

This means that there are q linearly independent column vectors in the controllability
matrix. Let us define such q linearly independent column vectors as f1 , f2 , p , fq . Also,
let us choose n-q additional n-vectors vq+1 , vq+2 , p , vn such that

P = Cf1 哸 f2 哸 p 哸 fq 哸 vq + 1 哸 vq + 2 哸 p 哸 vn D

is of rank n. Then it can be shown that

 = P -1 AP = c d, B̂ = P -1B = c d
A 11 A 12 B11
0 A 22 0

(See Problem A–10–1 for the derivation of these equations.) Now define

K̂ = KP = Ck1 哸 k2 D

Then we have

∑s I - A + BK∑ = @P -1(s I - A + BK)P@

= @s I - P -1 AP + P -1 BKP @
= @s I - Â + B̂K̂ @

= 2 sI - c d + c 11 d Ck1 哸 k2 D 2
A 11 A 12 B
0 A 22 0
s I q - A 11 + B11 k1 -A 12 + B11 k2
= 2 2
0 s I n - q - A 22

= @s I q - A 11 + B11 k1 @ ⴢ @s I n - q - A 22 @ = 0
where Iq is a q-dimensional identity matrix and In-q is an (n-q)-dimensional identity
matrix.

Section 10–2 / Pole Placement 725

 Notice that the eigenvalues of A22 do not depend on K. Thus, if the system is not
completely state controllable, then there are eigenvalues of matrix A that cannot be
arbitrarily placed. Therefore, to place the eigenvalues of matrix A-BK arbitrarily, the
system must be completely state controllable (necessary condition).
Next we shall prove a sufficient condition: that is, if the system is completely state
controllable, then all eigenvalues of matrix A can be arbitrarily placed.
In proving a sufficient condition, it is convenient to transform the state equation
given by Equation (10–1) into the controllable canonical form.
Define a transformation matrix T by

T = MW (10–4)

where M is the controllability matrix

M = CB 哸 AB 哸 p 哸 An - 1 BD (10–5)

and
an - 1 an - 2 p a1 1
an - 2 an - 3 p 1 0
   
W = G    W (10–6)
   
a1 1 p 0 0
1 0 p 0 0
where the ai’s are coefficients of the characteristic polynomial

∑s I - A∑ = sn + a1 sn - 1 + p + an - 1 s + an

Define a new state vector x̂ by

x = Tx̂
If the rank of the controllability matrix M is n (meaning that the system is completely
state controllable), then the inverse of matrix T exists, and Equation (10–1) can be
modified to
#
x̂ = T -1 ATx̂ + T -1 Bu (10–7)

where

0 1 0 p 0
0 0 1 p 0
   
T AT = G 
-1
   W (10–8)
   
0 0 0 p 1
-an -an - 1 -an - 2 p -a1

726 Chapter 10 / Control Systems Design in State Space

 0
0

T B = GW
-1
(10–9)

0
1

[See Problems A–10–2 and A–10–3 for the derivation of Equations (10–8) and (10–9).]
Equation (10–7) is in the controllable canonical form.Thus, given a state equation, Equa-
tion (10–1), it can be transformed into the controllable canonical form if the system is
completely state controllable and if we transform the state vector x into state vector x̂
by use of the transformation matrix T given by Equation (10–4).
Let us choose a set of the desired eigenvalues as m1 , m2 , p , mn . Then the desired
characteristic equation becomes

As - m1 BAs - m2 B p As - mn B = sn + a1 sn - 1 + p + an - 1 s + an = 0 (10–10)

Let us write

KT = Cdn dn - 1 p d1 D (10–11)

When u = -KTx̂ is used to control the system given by Equation (10–7), the system
equation becomes
#
x̂ = T -1 ATx̂ - T -1 BKTx̂

The characteristic equation is

@s I - T -1 AT + T -1 BKT@ = 0

This characteristic equation is the same as the characteristic equation for the system,
defined by Equation (10–1), when u = -Kx is used as the control signal. This can be
seen as follows: Since

#
x = Ax + Bu = (A - BK) x

the characteristic equation for this system is

∑s I - A + BK∑ = @T -1(s I - A + BK) T@ = @s I - T -1 AT + T -1 BKT @ = 0

Section 10–2 / Pole Placement 727

 Now let us simplify the characteristic equation of the system in the controllable canonical
form. Referring to Equations (10–8), (10–9), and (10–11), we have

@s I - T -1 AT + T -1 BKT@

0 1 p 0 0
   
   
= 6 sI - F V + F V Cdn dn - 1 p d1 D 6
   
0 0 p 1 0
-an -an - 1 p -a 1 1
s -1 p 0
0 s p 0
  
= 6 6
  
  
an + dn an - 1 + dn - 1 p s + a1 + d1

= sn + Aa1 + d1 Bsn - 1 + p + Aan - 1 + dn - 1 Bs + Aan + dn B = 0 (10–12)

This is the characteristic equation for the system with state feedback. Therefore, it must
be equal to Equation (10–10), the desired characteristic equation. By equating the
coefficients of like powers of s, we get
a1 + d1 = a1
a2 + d2 = a2



an + dn = an
Solving the preceding equations for the di ’s and substituting them into Equation (10–11),
we obtain
K = Cdn dn - 1 p d1 D T -1
= Can - an 哸 an - 1 - an - 1 哸 p 哸 a2 - a2 哸 a1 - a1 D T -1 (10–13)
Thus, if the system is completely state controllable, all eigenvalues can be arbitrarily
placed by choosing matrix K according to Equation (10–13) (sufficient condition).
We have thus proved that a necessary and sufficient condition for arbitrary pole
placement is that the system be completely state controllable.
It is noted that if the system is not completely state controllable, but is stabilizable,
then it is possible to make the entire system stable by placing the closed-loop poles at
desired locations for q controllable modes. The remaining n-q uncontrollable modes
are stable. So the entire system can be made stable.

728 Chapter 10 / Control Systems Design in State Space

 Determination of Matrix K Using Transformation Matrix T. Suppose that the
system is defined by
#
x = Ax + Bu
and the control signal is given by
u = -Kx
The feedback gain matrix K that forces the eigenvalues of A-BK to be m1 , m2 , p , mn
(desired values) can be determined by the following steps (if mi is a complex eigenvalue,
then its conjugate must also be an eigenvalue of A-BK):
Step 1: Check the controllability condition for the system. If the system is completely
state controllable, then use the following steps:
Step 2: From the characteristic polynomial for matrix A, that is,

∑s I - A∑ = sn + a1 sn - 1 + p + an - 1 s + an
determine the values of a1 , a2 , p , an .
Step 3: Determine the transformation matrix T that transforms the system state equa-
tion into the controllable canonical form. (If the given system equation is already in the
controllable canonical form, then T=I.) It is not necessary to write the state equation
in the controllable canonical form. All we need here is to find the matrix T. The
transformation matrix T is given by Equation (10–4), or
T = MW
where M is given by Equation (10–5) and W is given by Equation (10–6).
Step 4: Using the desired eigenvalues (desired closed-loop poles), write the desired
characteristic polynomial:

As - m1 BAs - m2 B p As - mn B = sn + a1 sn - 1 + p + an - 1 s + an
and determine the values of a1 , a2 , p , an .
Step 5: The required state feedback gain matrix K can be determined from Equation
(10–13), rewritten thus:

K = Can - an 哸 an - 1 - an - 1 哸 p 哸 a2 - a2 哸 a1 - a1 D T -1

Determination of Matrix K Using Direct Substitution Method. If the system

is of low order (n  3), direct substitution of matrix K into the desired characteristic
polynomial may be simpler. For example, if n=3, then write the state feedback gain
matrix K as

K = Ck1 k2 k3 D

Substitute this K matrix into the desired characteristic polynomial ∑s I - A + BK∑ and
equate it to As-m1 B As-m2 B As-m3 B, or

∑s I - A + BK∑ = As - m1 BAs - m2 BAs - m3 B

Section 10–2 / Pole Placement 729

 Since both sides of this characteristic equation are polynomials in s, by equating the
coefficients of the like powers of s on both sides, it is possible to determine the values
of k1 , k2 , and k3 . This approach is convenient if n=2 or 3. (For n=4, 5, 6, p , this
approach may become very tedious.)
Note that if the system is not completely controllable, matrix K cannot be determined.
(No solution exists.)

Determination of Matrix K Using Ackermann’s Formula. There is a well-known

formula, known as Ackermann’s formula, for the determination of the state feedback
gain matrix K. We shall present this formula in what follows.
Consider the system
#
x = Ax + Bu

where we use the state feedback control u=–Kx. We assume that the system is
completely state controllable. We also assume that the desired closed-loop poles are at
s=m1 , s=m2 , p , s=mn .
Use of the state feedback control
u = -Kx
modifies the system equation to
#
x = (A - BK) x (10–14)
Let us define
苲
A = A - BK
The desired characteristic equation is

∑s I - A + BK∑ = @s I - A @ = As - m1 BAs - m2 B p As - mn B
苲

= sn + a1 sn - 1 + p + an - 1 s + an = 0
苲
Since the Cayley–Hamilton theorem states that A satisfies its own characteristic
equation, we have

fAA B = A n + a1 A n - 1 + p + an - 1 A + an I = 0
苲 苲 苲 苲
(10–15)

We shall utilize Equation (10–15) to derive Ackermann’s formula. To simplify the

derivation, we consider the case where n=3. (For any other positive integer n, the
following derivation can be easily extended.)
Consider the following identities:

I = I
苲
A = A - BK
苲2 苲
A = (A - BK)2 = A2 - ABK - BKA
苲 苲 苲
A3 = (A - BK)3 = A3 - A2 BK - ABKA - BKA 2

730 Chapter 10 / Control Systems Design in State Space

Multiplying the preceding equations in order by a3 , a2 , a1 , and a0 (where a0=1),
respectively, and adding the results, we obtain
苲 苲 苲
a3 I + a2 A + a1 A 2 + A 3
= a3 I + a2(A - BK) + a1 AA2 - ABK - BKA B + A3 - A2BK
苲
苲 苲
- ABKA - BKA 2
苲
= a3 I + a2 A + a1 A2 + A3 - a2 BK - a1 ABK - a1 BKA - A2 BK
苲 苲
- ABKA - BKA 2 (10–16)
Referring to Equation (10–15), we have
a3 I + a2 A + a1 A 2 + A 3 = fAA B = 0
苲 苲 苲 苲

Also, we have
a3 I + a2 A + a1 A2 + A3 = f(A) Z 0
Substituting the last two equations into Equation (10–16), we have
fAA B = f(A) - a2 BK - a1 BKA - BKA 2 - a1 ABK - ABKA - A2 BK
苲 苲 苲 苲

Since fAA B = 0, we obtain

苲

f(A) = BAa2 K + a1 KA + KA 2 B + ABAa1 K + KA B + A2 BK

苲 苲 苲
苲 苲
a2 K + a1 KA + KA 2
= CB 哸 AB 哸 A2 BD C
苲
a1 K + KA S (10–17)
K
Since the system is completely state controllable, the inverse of the controllability matrix
CB 哸 AB 哸 A2 BD
exists. Premultiplying both sides of Equation (10–17) by the inverse of the controllability
matrix, we obtain
苲 苲
a2 K + a1 KA + KA 2
CB 哸 AB 哸 A2 BD f(A) = C
苲
S
-1
a1 K + KA
K
Premultiplying both sides of this last equation by [0 0 1], we obtain
苲 苲
a2 K + a1 KA + KA 2
苲
1]CB 哸 AB 哸 A2 BD f(A) = [0 1] C S = K
-1
[0 0 0 a1 K + KA
K
which can be rewritten as
1]CB 哸 AB 哸 A2 BD f(A)
-1
K = [0 0
This last equation gives the required state feedback gain matrix K.
For an arbitrary positive integer n, we have
K = [0 0 p 0 1]CB 哸 AB 哸 p 哸 An - 1 BD f(A)
-1
(10–18)

Section 10–2 / Pole Placement 731

 Equation (10–18) is known as Ackermann’s formula for the determination of the state
feedback gain matrix K.

Regulator Systems and Control Systems. Systems that include controllers can
be divided into two categories: regulator systems (where the reference input is constant,
including zero) and control systems (where the reference input is time varying). In what
follows we shall consider regulator systems. Control systems will be treated in Section
10–7.

Choosing the Locations of Desired Closed-Loop Poles. The first step in the
pole-placement design approach is to choose the locations of the desired closed-loop
poles. The most frequently used approach is to choose such poles based on experience
in the root-locus design, placing a dominant pair of closed-loop poles and choosing other
poles so that they are far to the left of the dominant closed-loop poles.
Note that if we place the dominant closed-loop poles far from the jv axis, so that the
system response becomes very fast, the signals in the system become very large, with
the result that the system may become nonlinear. This should be avoided.
Another approach is based on the quadratic optimal control approach.This approach
will determine the desired closed-loop poles such that it balances between the acceptable
response and the amount of control energy required. (See Section 10–8.) Note that
requiring a high-speed response implies requiring large amounts of control energy.Also,
in general, increasing the speed of response requires a larger, heavier actuator, which will
cost more.

EXAMPLE 10–1 Consider the regulator system shown in Figure 10–2. The plant is given by
#
x = Ax + Bu
where
0 1 0 0
A = C 0 0 1S , B = C0S
-1 -5 -6 1
The system uses the state feedback control u=–Kx. Let us choose the desired closed-loop poles
at
s = -2 + j4, s = -2 - j4, s = -10
(We make such a choice because we know from experience that such a set of closed-loop poles
will result in a reasonable or acceptable transient response.) Determine the state feedback gain
matrix K.

u x
B + 兰
+

Figure 10–2 –K
Regulator system.

732 Chapter 10 / Control Systems Design in State Space