Systems & Control Letters 61 (2012) 292297

Contents lists available at SciVerse ScienceDirect

Systems & Control Letters

journal homepage: www.elsevier.com/locate/sysconle

Pole placement by parametric output feedback

Ulrich Konigorski
Institute of Automatic Control, Control Engineering and Mechatronics Lab., Technische Universitt Darmstadt, Landgraf-Georg-Strasse 4, 64283 Darmstadt, Germany

article

info

Article history:
Received 29 September 2011
Received in revised form
18 November 2011
Accepted 21 November 2011
Available online 3 January 2012
Keywords:
Pole placement
Output feedback
Eigenstructure assignment

abstract
This note presents a new analytical solution to the problem of pole placement via constant output
feedback under the condition m + p n, where n, m, and p are the number of states, inputs and
outputs, respectively. The approach is based upon parametric eigenstructure assignment of linear timeinvariant multivariable systems in combination with a special explicit formulation of the pole assignment
equations. Thus, the resulting analytical solution explicitly offers all remaining mp n degrees of freedom
beyond eigenvalue assignment which can be used for additional design goals such as response shaping,
minimizing the norm of the feedback matrix, and robust control, respectively.
2011 Elsevier B.V. All rights reserved.

1. Introduction
Besides optimal control the pole placement approach is one of
the most popular design methods in linear control theory. While
in the case of complete state feedback the problem of finding a
constant feedback matrix which assigns an arbitrary selected set
of self-conjugate complex numbers as spectrum of the closedloop system is completely solved (see e.g. [1]) to deal with static
output feedback is much harder. This is mainly due to the fact
that in multi-inputmulti-output (MIMO) systems pole placement
is a nonlinear problem and demands solving a set of nonlinear
algebraic equations in the unknown gain parameters whose
solution may not exist in the case of output feedback. However,
for controllable systems and state feedback these equations always
have a solution [1] and in MIMO systems this solution is even not
unique. In this case, there are additional degrees-of-freedom (dof)
beyond pole placement which can be used for further design goals
such as eigenvector or eigenstructure assignment.
Based on the results in [2] on eigenstructure assignment in
the case of state feedback several solutions to the problem of
pole placement by static output feedback have been reported
during the last three decades [310] and just recently another new
approach has been presented in [11,12]. Most of them rely on the
fundamental result of [13,14], which is also known as Kimuras
condition, that for the generic system all closed-loop poles can be
assigned almost arbitrary if m + p n + 1 where n, m, p denote the
system order and the number of inputs and outputs, respectively.

Tel.: +49 6151163014; fax: +49 6151166114.

E-mail address: ukonigorski@iat.tu-darmstadt.de.

0167-6911/$ see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.sysconle.2011.11.015

On the other hand in [15] it was shown that a necessary and

sufficient condition for arbitrary pole assignment for the generic
system is mp n if complex feedback gains are allowed and
finally in [16,17] Wang presented the important result that for
real static output feedback mp > n is sufficient for generic pole
assignability. However, up to now there is no closed-form solution
to the problem of finding a real feedback under Wangs condition
which is certainly due to the fact that in general pole placement
via static output feedback is N P -hard [18]. Moreover, if mp > n
one is interested in a solution which encompasses all remaining
mp n dof beyond pole placement. Meanwhile, under Kimuras
condition there exist several design techniques (see e.g. [6,7,10])
which offer such a parametric solution and in [11] the authors
presented a new noniterative approach to pole placement based
on eigenstructure assignment which improves Kimuras sufficient
condition to m + p n while in [12] this result was extended to
even encompass some cases for which m + p < n < mp. A general
overview on static output feedback which covers several different
design techniques can be found in [19].
In this note, we seize the suggestions from [10,11] and by combining them with the results from [20] we are able to develop
a straightforward noniterative procedure for pole placement by
parametric output feedback. In Section 2, after statement of the
problem the fundamental properties of eigenstructure assignment
are shortly reviewed and for m = p = 2 an analytical expression
for the direct solution of the pole assignment equation is developed. Based on these preliminary results a closed-form parametric
solution to the problem of pole placement by constant output feedback under the condition m + p n is presented in Section 3 while
Section 4 gives a numerical example before the main results of this
note are summarized in Section 5.

U. Konigorski / Systems & Control Letters 61 (2012) 292297

293

2. Problem statement and preliminaries

Now we come back to (8), (9) and substitute them into (7) to get
the homogeneous equation

In this section, the problem of pole placement by constant

output feedback is stated and some fundamental results on
eigenstructure assignment as well as a formulation of the problem
based on an explicit expression of the closed loop characteristic
polynomial are shortly reviewed.
Consider the completely controllable and observable linear
time-invariant multivariable system

(Mi KCNi )qi = 0

x = Ax + Bu,

y = Cx

(1)

where x Rn , u Rm , y Rp . The real constant matrices

A, B, and C are of appropriate dimensions and it is assumed that
rank(B) = m 2 and rank(C ) = p 2. If the real constant output
feedback
u = Ky

(2)

is applied to (1) the closed-loop state equation becomes

x = (A + BKC )x = Ac x.

(3)

The problem of pole placement or eigenvalue assignment by

constant output feedback then means to find the real matrix K in
(2) such that the spectrum {Ac } of Ac coincides with the given set
= {1 , 2 , . . . , n } of self-conjugate complex values.
2.1. Eigenstructure assignment
In the context of eigenstructure assignment by constant output
feedback the set = {1 , 2 } of closed-loop eigenvalues is
usually divided into two self-conjugate sets of arbitrarily selected
distinct complex numbers 1 = {1 , 2 , . . . , r } and 2 =
{r +1 , r +2 , . . . , n }. Then in a first step the set 1 is associated
with the spectrum of Ac and the closed-loop eigenvectors vi via
Ac vi = (A + BKC )vi = i vi ,

i = 1, . . . , r

(4)

which can also be written as

vi
[A i I , B]
= 0,
KC vi

i = 1, . . . , r .

(5)

If (A, B) is completely controllable rank[A i I , B] = n, i

C [21] and thus the right nullspace
ker{[A i I , B]} = ker{Si } =

which has a nonzero solution qi = 0 iff

det(Mi KCNi ) = 0.

(6)

(11)

Obviously, (11) can be used to assign i as closed-loop pole [22]

while the parameter vector qi associated with i via (10) then
explicitly offers (m 1) additional dof beyond eigenvalue
assignment. Thus, to assign the r numbers in 1 as closed-loop
eigenvalues K must solve the linear equation
K (CVr ) = Q r

(12)

where the qi = 0, i = 1, . . . , r are considered as free parameters

and
Vr = [N1 q1 , . . . , Nr qr ]

(13)

Q r = [M1 q1 , . . . , Mr qr ].

(14)

Obviously, a solution of (12) exists for almost any choice of the set
1 and qi = 0, i = 1, . . . , r if the condition rank(CVr ) = r holds
which in turn implies r p. For r = p, the usual choice in the literature on eigenstructure assignment, the solution K = Q r (CVr )1
of (12) explicitly exhibits all mp dof provided by K Rmp in the
shape of the p eigenvalues from 1 and the p corresponding parameter vectors qi = 0, i = 1, . . . , p. Thus, to assign the remaining
n p eigenvalues in 2 the parameter vectors qi = 0, i = 1, . . . , p
are not arbitrary but must undergo some restrictions. In the following this can be seen if all investigations carried out so far with right
eigenvectors vi , input directions hi and (right) parameter vectors qi
are accomplished with left eigenvectors wj , output directions lj and
corresponding (left) parameter vectors zj for the eigenvalues in 2
where the prime denotes transpose. To this end instead of (4) we
start with the relation

wj Ac = wj (A + BKC ) = j wj ,

j = r + 1, . . . , n

(15)

or

[wj , wj BK ]

Ni
Mi

(10)

A j I
C

= 0 ,

j = r + 1, . . . , n

(16)

where

is of dimension m. Then the closed-loop eigenvectors vi and the

input directions

A j I
rank
C

hi = KC vi

if (A, C ) is an observable pair [21]. Therefore, the p-dimensional

left nullspace of Tj can be calculated from

(7)
m

can be parameterized by nonzero parameter vectors qi C (see

e.g. [2,10,11])

vi = Ni qi
hi = Mi qi .

(8)
(9)

Remark 1. Since the columns of ker{Si } can arbitrarily be scaled

by any nonzero scalar the eigenvectors vi and input directions hi in
(8), (9) are only determined except for their length and so are the
parameter vectors qi . Thus, each parameter vector only provides
(m 1) dof to assign the corresponding eigenvector vi within the
m-dimensional subspace of Cn spanned by the columns of Ni [2].
Moreover, it directly follows from elementary matrix theory that
for a self-conjugate complex pair i1 = i2 in the set 1 we also

have vi1 = vi2

, Ni1 = Ni2 and this implies qi1 = qi2 . Therefore, the
parameter vectors qi associated with the set 1 are not completely
free but must also constitute a self-conjugate set. However, this
does not reduce the available dof provided by the set qi , i =
1, . . . , r since a complex qi offers m 1 complex and 2(m 1)
real dof, respectively.

= rank(Tj ) = n,

j C

[Dj , Ej ] Tj = 0

(17)

and the closed-loop left eigenvectors wj and output directions

lj = wj BK

(18)

are parameterized by nonzero left parameter vectors zj C

wj = zj Dj

(19)

lj = zj Ej .

(20)

Finally, from (18)(20) we get the dual version of (10)

zj (Ej Dj BK ) = 0

(21)

with the corresponding necessary condition

det(Ej Dj BK ) = 0

(22)

and

(Wn r B)K = Zn r

(23)

294

U. Konigorski / Systems & Control Letters 61 (2012) 292297

with
Wnr = [Dr +1 zr +1 , . . . , Dn zn ]

(24)

Znr = [Er +1 zr +1 , . . . , En zn ]
(25)
as dual version of (12). Combining (10) and (21) results in a bilinear
equation with respect to qi , i = 1, . . . , r and zj , j = r + 1, . . . , n
(see [10,11]). Other approaches to eigenstructure assignment
by output feedback solve (10), (21) via two coupled Sylvester
equations [4,5,8] or a bilinear generalized Sylvester equation [9].
In the sequel it is shown that under the condition m + p n a
closed form parametric solution to the pole assignment problem
can be obtained without referring to the solution of a bilinear
matrix equation. To that purpose (10), (21) are combined with
the direct evaluation of the pole-assignment Eq. (11). By means
of exterior algebra the following section summarizes the results
in [20] on the general formulation of the closed-loop characteristic
equation for the special case K R22 .

2 . To that purpose we choose r < p and nonzero column vectors

[q1 , . . . , qr ] such that (CVr ) in (12) has full rank r. Then K can be
solved from (12)
K = Q r (CVr )+ + K1 U1 = K0 + K1 U1

(28)

where K0 = Q r (CVr ) while (CVr )

= [(CVr ) (CVr )] (CVr )
denotes the MoorePenrose pseudo-inverse of CVr and U1 is a
(p r ) p orthonormal matrix whose rows form a basis for the
left kernel of (CVr ), i.e.,
+

U1 (CVr ) = C1 Vr = 0.

(29)

By assumption, rank(C ) = p and thus rank(U C ) = rank(C ) = p

where U = [U1 , U1 ] is a full rank p p orthonormal matrix with
U1 U1 = 0. Therefore, all rows of C are linear independent and the
downsized output matrix C1 = U1 C always has full rank p r. As
shown in the Appendix the output feedback K given by (28) assigns
the r numbers in 1 as closed-loop eigenvalues of

2.2. Direct solution of the pole assignment equation

Ac = A + BK0 C + BK1 U1 C = A1 + BK1 C1

As already discussed in the previous section for i to be a root

of the closed-loop characteristic polynomial Pc () = det(I
A BKC ) the constant output feedback matrix K must solve one
of the Eqs. (11) and (22), respectively. In [20] the expansion of
Pc () has been discussed in great detail and it turns out that for
arbitrary m and p the pole-assignment equations are multilinear
in the coefficients of K and therefore an analytical solution is not
obvious. However, in what follows we concentrate on the special
case K R22 , Mi R22 , and CNi R22 where an analytical
solution of (11) can still be obtained. To that purpose the evaluation
of the determinant in (11) by means of the exterior product of its
corresponding column vectors is used and we write

for arbitrary qi = 0, i = 1, . . . , r and K1 Rm(pr ) and

renders the pair (A1 , C1 ) unobservable for 1 . However,
since all unobservable eigenvalues of (A1 , C1 ) cannot be changed
by the constant output feedback K1 , if some eigenvalues of A1 not
belonging to 1 are unobservable, they also remain fixed in the
subsequent design steps. Applying the coordinate transformation
x = [Vr , R]x with R Vr = 0, R R = Inr to (A1 , C1 ) results in

+

Vr A1 Vr Vr+ A1 R
diag{1 }

A1 =
=
R A1 Vr
R A1 R
0

C1 = C1 Vr C1 R = 0 C1 R

(30)

Vr+ A1 R
R A1 R

Mi = [i1 , i2 ] = {ijl }

since according to (A.2) the columns of Vr are eigenvectors of A1

corresponding to the r unobservable eigenvalues in 1 . i.e. A1 Vr =
Vr diag{1 } and C1 Vr = 0. Then a straightforward investigation of
the PBH eigenvector test (A.2) for (
A1 ,
C1 ) reveals that 1 is
unobservable, if there exist a vector v 2 = 0 of dimension n r
which simultaneously solves the two equations

CNi = [1i , 2i ] = {jli }

C1 Rv 2 = U1 CRv 2 = 0

respectively. Then taking into account the special properties of

the exterior product x y of two vectors x, y (e.g. distributivity,
associativity, etc.) and after some elementary calculations the
evaluation of

det(Mi KCNi ) = ( K ) ( K ) = 0

v 2 = R C

k
K = [k1 , k2 ] = 11
k21

k12
k22

and

i
1

i
1

i
2

i
2

finally results in
det(Mi KCNi ) = det(Mi ) i k = 0

(26)

with

k = [k11 , k21 , k12 , k22 , det(K )]

i =

i22

i12

i21

i11

R A1 R Inr v 2 = 0.

Rv 2 = C + Vr
with , = 0 arbitrary vectors of appropriate dimensions.
Substituting these two latter expressions for Rv 2 and v 2 in (32)
finally results in the condition

= 0

(33)

n(np)

(nr )n

with Ho = (A I )C . Since C R
,R R
and
r < p (33) can only be fulfilled if rank(R Ho ) < n p or in other
words any eigenvalue 1 of A1 is observable if

(27)

(32)

Obviously, from (31) we get

R Ho = 0,

i
i
i
i = [11
, 12
]i , [21
, 22
]i , det(CNi )

(31)

rank(R Ho ) = n p

(34a)

With (10), (21) and (26) we are now able do derive the main result
of this note.

which can equivalently be written

3. Parametric pole assignment

(see also Theorem 2.3 and its proof in [23]). But since Vr (and
thus R) depend on the choice of the eigenvalues in 1 and the
parameter vectors q1 , . . . , qr either (34) is never true, which can
be seen as singular case, or only violated on a hypersurface of the
parameter space Po = {1 , q1 , . . . , qr }. Hence, we may assume
that generically rank(R Ho ) = n p and thus 1 is observable
for almost all parameter values in the set Po .

To avoid a bilinear relation between the right and left parameter

vectors qi and zj in what follows we consider a consecutive solution
of (10), (21) or (12), (23), respectively. In a first step we assign the r
eigenvalues in 1 via solving (12) while at the same time retaining
enough dof to subsequently solve the remaining n r Eqs. (23) for

Im(Ho ) Im(Vr ) = {0}

(34b)

U. Konigorski / Systems & Control Letters 61 (2012) 292297

Therefore K1 can subsequently be used to solve (23) for

(A1 , B, C1 ) whereas only the r eigenvalues 1 of A1 are unobservable and thus remain unchanged. So we choose self-conjugate
numbers = {r +1 , . . . , r + } from the set 2 and nonzero row
vectors [z1 , . . . , z ] such that (W B) in (23) has rank which implies m. Then K1 can be solved from (23)
K1 = (W B)+ Z + U2 K3 = K2 + U2 K3

(35)

where K2 = (W B)+ Z while (W B)+ = (W B) [(W B)(W B) ]1

denotes the MoorePenrose pseudo-inverse of W B and U2 is an
m (m ) orthonormal matrix whose columns form a basis for
the right kernel of W B, i.e.,

(W B)U2 = W B2 = 0.

(36)

Obviously, U2 and thus K3 and the downsized input matrix B2 =

BU2 only exist if < m and by a similar argumentation as before
for C1 we always have rank(B2 ) = m . Then it can be shown (see
Appendix) that K1 assigns the numbers from as closed-loop
eigenvalues of
Ac = A1 + BK2 C1 + BU2 K3 C1 = A2 + B2 K3 C1

(37)

for arbitrary self-conjugate zj =

0 , j = 1, . . . , and K3

R(m)(pr ) and renders the pair (A2 , B2 ) uncontrollable for

. Therefore, if r + < n in the last step K3 can be used to
solve (11) for (A2 , B2 , C1 ) and the = n r remaining
self-conjugate values from the set 2 without changing the r +
preassigned eigenvalues of A2 . But to assure the controllability of
the remaining eigenvalues of A2 , with W S = 0, S S = In and
Hc = B (A I ) a similar reasoning as before for the observability
of 1 results in the condition
rank(Hc S ) = n m

(38a)

or equivalently
Im(Hc ) Im(W ) = {0}

(38b)

for {1 , } which should generically be fulfilled. Then to

solve (11) we must bear in mind that in general for arbitrary m
2, p 2, 0 r < p, 0 m the resulting equations
det(Mi K3 C1 Ni ) = 0,

i = 1, . . . ,

m + p n. In this case K3 R22 , = n m p + 4 and

especially for m + p = n we thus have = 4 so that the solution of
(39) turns to the problem discussed in Section 2.2. In what follows
we therefore address the solution of this problem in more detail.
To this end for (A2 , B2 , C1 ), K3 R22 and the 4 remaining selfconjugate values {n3 , . . . , n } from the set 2 we refer to (26)
and set = [1 , . . . , 4 ], = [det(M1 ), . . . , det(M4 )], which
results in
k 3 = + + f = + f

(41)

where + = [ ]1 denotes the right inverse of the 4 5

matrix if rank() = 4 and f is a one-dimensional basis for the
right nullspace of , i.e., 1f = 0. At first glance there seems to
be an infinite set of solutions parameterized by the scalar but
to comply with (27) must solve a quadratic equation. To see
this according to the notation introduced in Section 2.2 we set
= [11 , 21 , 12 , 22 , 5 ] and f = [f11 , f21 , f12 , f22 , f5 ] . Then
with respect to (27) and 1 = [11 , 21 ] , 2 = [12 , 22 ] , f1 =
[f11 , f21 ] , f2 = [f12 , f22 ] (41) can be split into the matrix equation
K3 = [1 , 2 ] + [f1 , f2 ] = + F

(42)

and the scalar equation

det(K3 ) = 5 + f5

(43)

where in turn the left hand side of (43) is replaced by the exterior
product
det(K3 ) = (1 + f1 ) (2 + f2 )

= (1 2 ) + (1 f2 ) + ( f1 2 ) + ( f1 f2 )
= det( ) + {det([1 , f2 ]) + det([f1 , 2 ])} + 2 det(F )
to finally result in the quadratic equation
0 = {det( ) 5 } + {det([1 , f2 ]) + det([f1 , 2 ]) f5 }

+ 2 det(F )
0 = p0 + p1 + 2 p2 .

(44)

Obviously, only for a real solution of (44) we get a real k 3 in

(41) and a real K3 from (42), respectively. With this K3 the final
solution to the pole assignment problem is then given by (28) and
(35).
3.3. Main results

(39)

are multilinear in the elements of K3 and will only have a solution

if the number of equations does not exceed the number (m
)(p r ) of unknowns, i.e.,

(m )(p r ) n r .

295

(40)

Hence, in the following we concentrate on two special cases where

a solution to (39) is not required or an analytical solution to (39)
can be obtained by means of (26).
3.1. r = p 1(m + p n + 1)
For r = p 1 (40) yields Kimuras condition m + p n + 1
with arbitrary. Thus, to avoid solving (39) we set = n r which
results in = 0, m and the parametric solution to the pole
assignment problem is given by (28) and (35).
3.2. r = p 2 (m + p n)
For r = p 2 the evaluation of (40) gives m + p n +( m + 2)
and since m the choice = m and = m1 results in m+p
n + 2 and m + p n + 1, respectively, so that both conditions
are already covered by Kimuras condition (see Section 3.1). On the
other hand, for 0 m 2 only the choice = m 2 results in
an easily accessible analytical solution to (39) under the condition

Before proceeding with a numerical example the main results

of this note are summarized in the following propositions.
Proposition 1. For m 2, p 2 let m + p = n and 1 =
{1 , . . . , p2 }, 2 = {p1 , . . . , n4 }, 3 = {n3 , . . . , n } be
self-conjugate sets of distinct complex numbers and qi = 0, i =
1, . . . , p 2, zj = 0 , j = p 1, . . . , n 4 corresponding selfconjugate sets of right and left parameter vectors, respectively. If

(I) rank(CVp2 ) = p 2,

Vp2 Rn(p2)

(II) rank(Wm 2 B) = m 2,

Wm 2 R(m2)n

(III) rank() = 4, R45

(IV) (44) has a real solution
then there exists a real output feedback matrix K = K0 + K2 U1 +
U2 K3 U1 according to (28), (35), (42) that assigns to (A + BKC ) the
closed-loop spectrum = {1 , 2 , 3 }.
In Proposition 1 conditions (III), (IV) are obsolete if instead of
m + p = n the stronger condition m + p n + 1 (Kimuras
condition) holds.
Proposition 2. For m 2, p 2 let m + p n + 1 and
1 = {1 , . . . , p1 }, 2 = {p , . . . , n } be two self-conjugate sets
of distinct complex numbers and qi = 0, i = 1, . . . , p 1, zj =
0 , j = p, . . . , n corresponding self-conjugate sets of right and left

296

U. Konigorski / Systems & Control Letters 61 (2012) 292297

parameter vectors, respectively. If for = n + 1 p

(I) rank(CVp1 ) = p 1,
(II) rank(W B) = ,

Vp1 R

n(p1)

W Rn

then there exists a real output feedback matrix K = K0 + K2 U1 +

U2 K3 U1 according to (28), (35) with K3 arbitrary that assigns to
(A + BKC ) the closed-loop spectrum = {1 , 2 }.
Remark 2. As aforementioned the case m + p n + 1 is not
only covered by Kimuras condition (Section 3.1) but also by the
approach presented in Section 3.2 for m 1 m. Therefore,
to gain greater flexibility in the separation of the spectrum into
self-conjugate subsets one can freely choose between the two
approaches. For instance, if n = 6, p = 4, m = 4 the separation of
the spectrum according to Section 3.1 yields three eigenvalues in
1 and 2 , respectively. Thus, two closed-loop eigenvalues must
be real. However, the approach according to Section 3.2 results in
two subsets with r = p 2 = 2 and = m = 4 so that all
n = 6 closed-loop eigenvalues can be complex and K is given by
(28), (35).
Remark 3. Obviously, according to the derivations in Section 3 it
is generically always possible to place at least m + p 1 or m + p
closed loop poles, respectively. Therefore, as in [11] the approach
presented in this note can also be used for partial pole placement
although this might be of less practical importance.
Remark 4. Of course, instead of (A, B, C ) the dual system (A , C ,
B ) can be used in all preceding calculations to interchange the role
of p and m as appropriate.
A simple calculation shows that for m + p n (28), (35) provide
all remaining mp n dof beyond eigenvalue assignment. The r selfconjugate right parameter vectors qi = 0, i = 1, . . . , r in (28)
offer r (m 1) dof while (p r 1) dof are provided by the selfconjugate set zj = 0 , j = 1, . . . , of left parameter vectors. In
Proposition 1 r = p2, = m2 and therefore the left parameter
vectors zj = 0 are of dimension 2 and each of them offers 1
dof whereas in Proposition 2 r = p 1 and the left parameter
vectors zj = 0 are of dimension 1, i.e., nonzero scalars and offer
no additional dof. But if in Proposition 2 m + p = n + 1 + d with
d > 0 then there are d additional real dof provided by K3 Rd1
in (35).
For m + p = n we also have mp > n if m 3 and/or
p 3. Therefore, in this case Wangs condition for the generic
solvability of the pole assignment problem over the reals is always
fulfilled and the remaining mp n dof covered by the left and
right parameter vectors qi = 0, i = 1, . . . , p 2, zj = 0 , j =
p 1, . . . , n 4 can be used to generically assure one or two
real solutions of (44). This assertion has been verified by hundreds
of numerical test runs with randomly assigned matrices A, B and
C of order up to 20 and with different closed-loop eigenvalues
and combinations of m and p, respectively. All these runs gave a
numerically efficient and reliable solution. However, in some rare
cases at first a self-conjugate complex solution for the feedback
matrix K was obtained but finally after some additional runs with
the same A, B, C and closed-loop eigenvalues but randomly chosen
parameter vectors in all cases one or two real solutions have been
obtained.
However, if m = p = 2, m + p = mp = n = 4 the sets 1 and
2 in Proposition 1 are both empty and conditions (I) and (II) can
be omitted. Thus there are no dof beyond eigenvalue assignment
and the solution is solely given by K3 in (42) which depends on
the solutions of (44) and thus might be complex. Indeed, the
condition mp n is in general only necessary and sufficient for
the generic solvability of the pole assignment problem if complex
solutions are allowed [24,25]. Especially in the case m = p =
2, n = 4 it can be shown [26] that the problem is not generically
solvable over the reals.

As shown in [12] even for m + p < n < mp there are some cases
which allow for a direct solution of the pole assignment problem.
Of course, this interesting result is not covered by the approach
presented in this note. However, for r = p 3, = m 2
the evaluation of (40) gives m + p n 1 and especially for
m + p = n 1(A2 , B2 , C1 ) in (37) reduces to a system of McMillan
= 2 inputs and p = 3 outputs, respectively.
degree n = 6 with m
For this system we must assign the remaining = 6 self-conjugate
eigenvalues {n5 , . . . , n } from the set 2 by solving (39) for
K3 R23 .
Due to the results in [24,25] for this kind of system there
exists (generically) at least one real solution to the pole placement
problem. However, instead of (44) in this case the expansion of
(39) results in a system of 3 quadratic equations in 3 unknowns
(see e.g. [20]) which in general cannot be solved analytically. But
obviously the results in [12] suggest that the solution of these
special quadratic equations deserves further investigations.
4. Numerical example
In this section a numerical example from the literature is used to
illustrate the application of Proposition 1. The system data (A, B, C )
originate from [11] and describe a system with n = 5, m = 3, p =
2. Since r = p 2 = 0 the set 1 is empty and condition (I) in
Proposition 1 can be omitted. So we can directly evaluate (35) for
A1 = A, C1 = C and the = m 2 = 1 real eigenvalue from
the desired closed-loop spectrum = {2 , 3 } = {0.5, 3
2i, 2 2i}. For that purpose we must choose one nonzero real
left parameter vector associated with the eigenvalue 1 = 0.5
from the set 2 . With z1 = [1, 1] the evaluation of (44) gives two
real solutions for and thus we readily get from (35), (42) two
corresponding real feedback matrices
20.2391
7.5363
45.0720

K 1 =

788.530
= 83.736
4003.256

K 2

18.1572
6.9665 ,
39.1186

775.014
82.257
3933.974

which assign both the spectrum to the closed-loop system A +

BK1 C and A + BK2 C , respectively. Obviously, K2 K1
and even the norm of K1 is quite large and should be reduced. To
this end the single dof offered by z1 can be used, e.g., the choice
z1 = [1, 0.36] leads to the two solutions
3.0150
1.8156
2.8693

K 1 =

560.317
= 65.655
2839.943

K 2

2.3515
3.6562 ,
3.0012

547.490
64.091
2774.271

with much smaller entries of both feedback matrices.

This example shows the advantage over the approach in [11]
where for m + p = n not all mp dof are accessible by the designer. In
fact, there is always one dof missing and therefore in this example
the technique in [11] yields exactly one real solution which only
depends on the separation of the desired closed-loop spectrum.
However, as shown above there are in general two real solutions
which in addition depend on the remaining dof offered by z1 .
5. Conclusion
In this paper a closed form solution to the problem of pole
placement by constant output feedback has been presented.
Under the condition m + p n beside eigenvalue assignment
the approach explicitly offers all remaining mp n degrees of
freedom provided by two self-conjugate sets of right and left
parameter vectors, qi and zj and possibly a real matrix K3 . For

U. Konigorski / Systems & Control Letters 61 (2012) 292297

instance these dof can be used to minimize the norm of the

resulting feedback matrix K as shown by a numerical example.
Furthermore the maximization of the complex stability radius
or the minimization of the closed-loop eigenvalue sensitivity as
proposed in [20,27], respectively can be considered. Finally, the
less restrictive condition m + p + 1 = n mp deserves some
further investigation.
Appendix
From (8), (13), (14) we have vi = Ni qi = Vr i and Q r i = Mi qi
where i denotes the ith unit vector. Then with (28), (29) and (6)
for Ac in (30) we get

(i I Ac )vi = (i I A BQ r (CVr )+ C BK1 U1 C )vi

= (i I A)vi BQ r (CVr )+ (CVr )i BK1 U1 (CVr )i
= (i I A)Ni qi BMi qi = 0,

qi , i = 1, . . . , r .

This shows that for any choice of self-conjugate right parameter

vectors qi = 0 and arbitrary real K1 in (28) i 1 and vi = Ni qi
are eigenvalues and corresponding right eigenvectors of A1 = A +
BQ r (CVr )+ C = A + BK0 C , i.e.,

(i I A1 )vi = 0,

i = 1, . . . , r .

(A.1)

Moreover, the two Eqs. (A.1) and (29) directly correspond to the
PopovBelevitchHautus (PBH) eigenvector test for observability [21], i.e.,
A1 vi = i vi ,

C1 vi = 0,

i = 1, . . . , r

(A.2)

thus (A1 , C1 ) is an unobservable pair for the r eigenvalues in 1 .

Now following an analogue chain of proof as before by
application of (35), (36), (17) and with wj = zj Dj = j W and

j Z = zj Ej according to (19), (24), (25) we get for Ac in (37)

wj (j I Ac )
= j W (j I A1 B(W B)+ Z C1 BU2 K3 C1 )
= zj Dj (j I A1 ) zj Ej C1
= zj [Dj , Ej ]Tj = 0 ,

zj , j = 1, . . . , .

Therefore j 2 and wj = zj Dj are eigenvalues and accompanying left eigenvectors of A2 = A1 + B(W B)+ Z C1 = A1 + BK2 C1 for
any self-conjugate set of left parameter vectors zj = 0 and arbitrary real K3 in (35), i.e.,

wj (j I A2 ) = 0 ,

j = 1, . . . , .

(A.3)

Finally, the combination of (A.3) and (36) gives the PHB eigenvector
test for controllability

wj A2 = j wj ,

wj B2 = 0 ,

j = 1, . . . ,

(A.4)

hence (A2 , B2 ) is an uncontrollable pair for the selected eigenvalues from 2 .

297

References
[1] W.M. Wonham, On pole assignment in multi-input controllable linear
systems, IEEE Transactions on Automatic Control AC-12 (1967) 660665.
[2] B.C. Moore, On the flexibility offered by state feedback in multivariable
systems beyond closed loop eigenvalue assignment, IEEE Transactions on
Automatic Control AC-21 (1976) 689692.
[3] S. Srinathkumar, Eigenvalue/eigenvector assignment using output feedback,
IEEE Transactions on Automatic Control AC-23 (1978) 7981.
[4] L.R. Fletcher, J. Kautsky, G.K.G. Kolka, N.K. Nichols, Some necessary and
sufficient conditions for eigenstructure assignment, International Journal of
Control 42 (1985) 14571468.
[5] B.-H. Kwon, M.-J. Youn, Eigenvalue-generalized eigenvector assignment by
output feedback, IEEE Transactions on Automatic Control AC-32 (1987)
417421.
[6] G. Roppenecker, J.O. Reilly, Parametric output feedback controller design,
Automatica 25 (1989) 259265.
[7] C. Champetier, J.-F. Magni, On eigenstructure assignment by gain output
feedback, SIAM Journal on Control and Optimization 29 (1991) 865884.
[8] V.L. Syrmos, F.L. Lewis, Output feedback eigenstructure assignment using
two Sylvester equations, IEEE Transactions on Automatic Control 38 (1993)
495499.
[9] V.L. Syrmos, F.L. Lewis, A bilinear formulation for the output feedback
problem in linear systems, IEEE Transactions on Automatic Control 39 (1994)
410414.
[10] A.T. Alexandridis, P.N. Paraskevopoulos, A new approach to eigenstructure
assignment by output feedback, IEEE Transactions on Automatic Control 41
(1996) 10461050.
[11] O. Bachelier, J. Bosche, D. Mehdi, On pole placement via eigenstructure
assignment approach, IEEE Transactions on Automatic Control 51 (2006)
15541558.
[12] O. Bachelier, D. Mehdi, Non-iterative pole placement technique: a step further,
Journal of the Franklin Institute 345 (2008) 267281.
[13] H. Kimura, Pole assignment by gain output feedback, IEEE Transactions on
Automatic Control AC-20 (1975) 509516.
[14] E.J. Davison, S.H. Wang, On pole assignment in linear multivariable systems
using output feedback, IEEE Transactions on Automatic Control AC-20 (1975)
516518.
[15] R. Hermann, C.F. Martin, Applications of algebraic geometry to systems theory:
part I, IEEE Transactions on Automatic Control AC-22 (1977) 1925.
[16] X.A. Wang, Pole placement by static output feedback, IEEE Transactions on
Automatic Control AC-22 (1977) 1925.
[17] X.A. Wang, Grassmannian, central projection, and output feedback pole
assignment of linear systems, IEEE Transactions on Automatic Control AC-41
(1996) 786794.
[18] M. Fu, Pole placement via static output feedback is NP-hard, IEEE Transactions
on Automatic Control AC-49 (2004) 855857.
[19] V.L. Syrmos, C.T. Abdallah, P. Dorato, K. Grigoriadis, Static output feedback-a
survey, Automatica 33 (1997) 125137.
[20] L. Carotenuto, G. Franz, A general formula for eigenvalue assignment by static
output feedback with application to robust design, Systems & Control Letters
49 (2003) 175190.
[21] L. Kailath, Linear Systems, Prentice-Hall, New Jersey, 1980.
[22] U. Konigorski, A new direct approach to the design of structurally constrained
controllers, in: Preprints of the 10th IFAC World Congress, vol. 8, 1987,
pp. 293297.
[23] L.R. Fletcher, J.F. Magni, Exact pole assignment by output feedback, part 1,
International Journal of Control 45 (1987) 19952007.
[24] R.W. Brockett, C.I. Byrnes, Multivariable nyquist criteria, root loci, and pole
placement: a geometric viewpoint, IEEE Transactions on Automatic Control
AC-26 (1981) 271284.
[25] C. Giannakopoulos, N. Karcanias, Pole assignment of strictly proper and proper
linear systems by constant output feedback, International Journal of Control 42
(1985) 543565.
[26] J.C. Willems, W.H. Hesselink, Generic properties of the pole placement
problem, in: Poceedings of the 7th IFAC World Congress, 1978, pp. 17251729.
[27] J. Kautsky, N.K. Nichols, P. van Dooren, Robust pole assignment in linear state
feedback, International Journal of Control 41 (1985) 11291155.