MASSACHUSETTS INSTITUTE OF TECHNOLOGY

Department of Electrical Engineering and Computer Science

6.241: Dynamic Systems—Fall 2003

Homework 8 Solutions

Exercise 17.1 The system is not well posed if negative feedback is assumed. Consider the Figure
17.1 of the notes and replace the (+) with a (-) sign. After some algebraic manipulation one obtains

u(t) = −r(t + 2) t ≥ 0

which corresponds to a non causal relationship.

Exercise 17.2 To check external stability of the interconnected (closed-loop) system we need
to verify BIBO stability of the transfer functions from ( w1 w2 )T to ( y u )T . The system can
be represented by a transfer function matrix:
� � � �� �
y (I + P K)−1 P (I + P K)−1 P K w1
= .
u (I + KP )−1 (I + KP )−1 K w2

Given � � � �
s α s+1
s+1 − s+1 s(s+5) 0
P = 1 1 , K= s+1 s+1
s+1 s+1 − s(s+5) s+5

we can see that � �

s(s+5) αs(s+5)
(I + P K)−1 = s2 +6s+α (s+6)(s2 +6s+α)
s+5
0 s+6

which is stable for α > 0. Notice that P is stable as well, therefore (I + P K)−1 P is stable for
α > 0. Other entries are given by:
� � � α(s+5)
�
s+α αs(s+5) s+5
−1 2
s +6s+α
− 2
(s+6)(s +6s+α) −1 s+6 (s+6)(s 2 +6s+α)
(I + P K) P K = 1
, (I + KP ) = s(s+5)
0 s+6 0 s2 +6s+α
� α(s+1)
�
s+1
s2 +6s+α (s+6)(s2 +6s+α)
(I + KP )−1 K = K(I + P K)−1 = s+1 s(s+1)
− s2 +6s+α s2 +6s+α
We can see that for α > 0 all of these transfer functions are stable, therefore the closed loop system
is stable.

b) If α = 0 the transfer function (I + KP )−1 K will have poles at 0 (integrators), therefore the
closed loop system is not stable.

Exercise 17.3 Given:

1
P (s) = , K(s) = k,
10s + 1
with no measurement noise.

1
(i) In order for the feedback system to be stable, we would like to have the closed loop from r to y
is stable. The closed loop transfer function from r to y is expressed as follows:

Y (s) = (I + P K)−1 P KR(s)

PK
= ∵ SISO .
1 + PK
K
10s+1
= K
R(s)
1 + 10s+1
K
10
∴ Y (s) = R(s).
s + 1+K
10

Thus the feedback system is stable iff

1+K
> 0 → K > −1. (1)
10
Since we are to find the least positive gain k, the first condition does not give us any information.

(ii) In order to evaluate e(t), we need to have the closed loop transfer function from R(s) to E(s),
which has the following form:
1
E(s) = R(s).
1 + PK
Yet, since the input of our interest here is a unit step, so that
1
R(s) = ,
s
thus E(s) has the following expression:

1
E(s) = R(s)
PK
1 1
= 1
1 + 10s+1 k s
10s + 1 1
= ,
10s + 1 + k s
with which we can use the final value theorem to evaluate e(∞) as follows:

e(∞) = lim sE(s)

s→0
10s + 1 1
= lim s
s→0
10s + 1 + k s
10s + 1
= lim
s→0 10s + 1 + k
1
∴ e(∞) = ,
1+k
whose absolute value has to be less than 0.1. Thus

2
 � 1 �
� � ≤ 0.1 → K ≥ 9, (2)
1+k
since k has to be positive.

(iii) As indicated in the problem, first we would like to show that the L2 to L∞ induced norm of a
SISO system is given by H2 norm of the system. It can be proven as follows:

Here we have the following convolution relationship:

y(t) = h(t) ∗ d(t),

then

� ∞
y(t) = h(t − τ )d(τ )dτ
�−∞∞
→ |y(t)| = | h(t − τ )d(τ )dτ |
−∞
� ∞
≤ |h(t − τ )||d(τ )|dτ
−∞
� ∞ � ∞
2 1/2
≤ ( h(t − τ ) dτ ) ( d(τ )2 dτ )1/2 , ∵ Cauchy-Schwartz inequality
−∞ −∞
∴ |y(t)| ≤ h 2 d 2.

Thus by taking sup of both sides yields

sup |y(t)| = y ∞ ≤ h 2 d 2.
t

Now we have to show that the equality can be achieved. Consider

d(t) = h(−t),
then clearly

d 2 = h 2 .,
Thus, at t = 0,

� ∞
|y(0)| = | h(0 − τ )h(−τ )dτ |
∞
= h 22 = h 2 d 2.

Thus the equality can be achieved.

Hence the requirement can be interpreted as to have h 2 , since

y ∞ ≤ h 2 d 2 ≤ h 2 ≤ 0.1

3
with d 2 ≤ 1.
Here, the closed loop transfer function from d(t) to y(t) is given by

1
H(s) =
1 + PK
1
= 1
1 + 10s+1 k
k
= 1− .
10s + 1 + k
Since the transfer function is proper its H2 norm is infinite (this corresponds to the fact that
one can always find a finite-energy input signal that will result in an output with infinite amplitude,
take for example t−1/4 for 0 ≤ t ≤ 1, and 0 elsewhere) and, there exists no K such that the third
requirement is met. Therefore, there exists no K to meet all the requirements.

Exercise 17.4 1) First, in order for the closed loop system to be stable, the transfer func-
tion from ( w1 w2 )T to ( y u )T has to be stable. The transfer function from w1 to y is given
by (I − P K)−1 P and is called system response function. The transfer function from w1 to u is
given by (I − KP )−1 and is called input sensitivity function. The transfer function from w2 to y is
(I − P K)−1 P K and is called the complementary sensitivity function. The transfer function from
w2 to u is given by (I − KP )−1 K. Therefore, we have the following :
� � � �� �
y (I + P K)−1 P (I + P K)−1 P K w1
= .
u (I + KP )−1 (I + KP )−1 K w2
So, if K is given as

K = Q(I − P Q)−1 = (I − QP )−1 Q,

then

(I + P K)−1 P = (I + P Q(I − P Q)−1 )−1 P

= (((I − P Q) + P Q)(I − P Q)−1 )−1 P
= (I − P Q)P
−1
(I + P K) P K = (I − P Q)P Q(I − P Q)−1
= P (I − QP )(I − QP )−1 Q
= PQ
−1
(I + KP ) = (I + (I − QP )−1 QP )−1
= ((I − QP + QP )(I − QP )−1 )−1
= I − QP
−1
(I + KP ) K = (I − QP )(I − QP )−1 Q
= Q.

Thus, the closed loop transfer function can be now written as follows:
� � � �� �
y (I − P Q)P P Q w1
= .
u (I − QP ) Q w2

4
In order for the closed loop system to be stable, then all the transfer functions in the large matrix
above must be stable as well.

(I − P Q)P ≤ I − PQ P ≤ ( I + PQ ) P
2
≤ ( I + P Q ) P ≤ P + P Q
PQ ≤ P Q
I − QP ≤ I + QP ≤ I + Q P .

Since P and Q are stable from the assumptions, we know that all the transfer functions are stable.
Therefore the closed loop system is stable if K = Q(I − P Q)−1 = (I − QP )−1 Q.

2) From 1), we can express Q in terms of P and K in the following manner.

K = Q(I − P Q)−1
K(I − P Q) = Q
K − KP Q = Q
K = (I + KP )Q
→ Q = (I + KP )−1 K = K(I + P K)−1 ,

by push through rule.

For some stable Q, the closed loop is stable for a stable P . by the stabilizing controller K =
Q(I − P Q)−1 . Yet, not all stable Q can be used for this formulation because of the well-posedness
of the closed loop. In the state space descriptions of P and Q, in order for the interconnected
system, in this case K(s) to be well-posed, we have to have the condition ( 17.4 ) in the lecture
note, i.e., (I − DP Q(∞)) is invertible.

3) Suppose P is SISO, w1 is a step, and w2 = 0. Then, we have the following closed loop transfer
function:
� � � �
Y (s) (I − P Q)P 1
= ,
U (s) I − QP s
1
since the Laplace transform of the unit step is s we have
1
U (s) = (1 − Q(s)P (s)) .
s
Then using the final value theorem, in order to have the steady state value of u(∞) to be zero, we
need:

1
u(∞) = lim s(1 − Q(s)P (s)) = 0
s→0 s
→ 1 − Q(0)P (0) = 0
→ Q(0) = 1/P (0).

5
Therefore, Q(0) must be nonzero and is equal to 1/P (0). Note that this condition implies that P
cannot have a zero at s = 0 because then Q would have a pole at s = 0, which contradicts that Q
is stable.

Exercise 18.2 Given :

� �
s−1
s+1 −5
G(s) = s+2 s−1 .
(s+1)2 s+1

Then we would like to verify whether or not there exists a controller K such that S = (I + P K)−1
satisfies S ∞ < 1.

1) No. Although G11 and G22 have nonminimum-phase zero at s = 1, it does not imply that the
plant G(s) has nonminimum-phase zeros. In order to find the zeros of G(s), we have to find a
variable coefficient c(s) such that G(s) loses its rank. It can be proceeded as follows:
� � � �
s−1
s+1 −5
s+2 = c(s) s−1 ,
(s+1)2 s+1

from the first equation, we have c(s) = −(1/5) s−1

s+1 . Substituting c(s) to the second equation yields

� �� �
s+2 1 s−1 s−1
= −
(s + 1)2 5 s+1 s+1
→ −5(s + 2) = (s − 1)2
→ s2 + 3s + 11 = 0
√
−3 ± i 35
∴s = .
2
Thus the zeros of the system G(s) are minimum-phase zeros.

2) For this G(s) we can indeed invert G(s) and scale it in order to tune the sensitivity is uniformly
less than one. G(s)−1 is computed as follows:

� �
s−1
1 5
G(s)−1 = s+1
s+2 s−1
s−1 s−1
+ 5 s+2
2
− (s+1) 2 s+1
s+1 s+1 (s+1)
� �
1 (s + 1)(s − 1) 5
= .
s2 + 3s + 11 −(s + 2) (s + 1)(s − 1)

Its poles are

√
−3 ± 35
s= .
2
Thus using the controller K(s) = αG(s)−1 , S(s), where α ∈ R, is expressed as follows:
1
S(s) = (I + GK)−1 = (I + αI)−1 = I
1+α

6
Thus in order to have S ∞ < 1, we need the following condition on α :

α > 0 , α < −2.

Finally, check that this is indeed a stabilizing controller, and the right half plane zeros that we
cancelled do not appear as right half plane poles in any of the transfer functions from all block
inputs to all block outputs.