Answers to Selected Problems

CHAPTER 2

2.1 (a) Linear. (b) and (c) Nonlinear. In (b), shift the operating point to (0, yo ); then (u, ȳ) is linear,
where ȳ = y − yo .

2.3 Linear, time varying, causal.

2.5 No, yes, no for x(0) = 0. Yes, yes, yes for x(0) = 0. The reason is that the superposition property
must also apply to the initial states.

2.9 y(t) = 0 for t < 0 and t > 4.


 0.5t
2
for 0 ≤ t < 1
y(t) = −1.5t + 4t − 2 for 1 ≤ t ≤ 2
2

−y(4 − t) for 2 < t ≤ 4

2.10 ĝ(s) = 1/(s + 3), g(t) = e−3t for t ≥ 0.

2.12
 −1  
D11 (s) D12 (s) N11 (s) N12 (s)
Ĝ(s) =
D21 (s) D22 (s) N21 (s) N22 (s)

2.15 (a) Define x1 = θ and x2 = θ̇ . Then ẋ1 = x2 , ẋ2 = −(g/ l) sin x1 + (u/ml) cos x1 . If θ is
small, then
   
0 1 0
ẋ = x+ u
−g/ l 0 −1/mg
It is a linear state equation.
(b) Define x1 = θ1 , x2 = θ̇1 , x3 = θ2 , and x4 = θ̇2 . Then

ẋ1 = x2
ẋ2 = − (g/ l1 ) sin x1 + (m2 g/m1 l) cos x3 sin(x3 − x1 )
+ (1/m1 l) sin x3 sin(x3 − x1 ) · u
ẋ3 = x4
ẋ4 = − (g/ l2 ) sin x3 + (1/m2 l2 )(cos x3 )u
This is a set of nonlinear equations. If θi ≈ 0 and θi θj ≈ 0, then it can be linearized as

321
322 ANSWERS TO SELECTED PROBLEMS

   
0 1 0 0 0
 −g(m1 + m2 )/m1 l1 0 m2 g/m1 l1 0  
ẋ =  x +  0 u
 0 0 0 1   0 
0 0 −g/ l2 0 1/m2 l2

2.16 From

mḧ = f1 − f2 = k1 θ − k2 u
I θ̈ + bθ̇ = (l1 + l2 )f2 − l1 f1
we can readily obtain a state equation. Assuming I = 0 and taking their Laplace transforms can
yield the transfer function.

2.18 ĝ1 (s) = ŷ1 (s)/û(s) = 1/(A1 R1 s + 1), ĝ2 (s) = ŷ(s)/ŷ1 (s) = 1/(A2 R2 s + 1). Yes,
ŷ(s)/û(s) = ĝ1 (s)ĝ2 (s).
2.20
 
0 0 1/C1
ẋ =  0 0 1/C2 x
−1/L1 −1/L1 −(R1 + R2 )/L1
 
0 −1/C1  
u1
+ 0 0 
u2
1/L1 R1 /L1
y = [−1 − 1 − R1 ]x + [1 R1 ]u
ŷ(s) s 2 + (R2 /L1 )s
ĝ1 (s) = =    
û1 (s) R1 + R2 1 1 1
s2 + s+ +
L1 C1 C2 L1
ŷ(s) (R1 s + (1/C1 ))(s + (R2 /L1 ))
ĝ2 (s) = =    
û2 (s) R1 + R2 1 1 1
s +
2 s+ +
L1 C1 C2 L1
ŷ(s) = ĝ1 (s)û1 (s) + ĝ2 (s)û2 (s)

CHAPTER 3

1 
8 
3.1 3 3
, [−2 1.5] .
3.3
  
2 1
1  1
q1 = −3  q2 = 1
3.74 1.732
1 1

3.5 ρ(A1 ) = 2, nullity(A1 )=1; 3, 0; 3, 1.

 Answers to Selected Problems 323

3.7 x = [1 1] is a solution. Unique. No solution if y = [1 1 1] .

3.9 α 1 = 4/11, α 2 = 16/11. The solution
 
4 −8 −4 −16
x=
11 11 11 11
has the smallest 2-norm.

3.12
 
0 0 0 −8
1 0 0 20 
Ā = 
0

1 0 −18 
0 0 1 7

3.13
   
1 0 −1 1 0 0
Q3 =  0 1 0  Â3 =  0 1 0
0 0 1 0 0 2
   
5 4 0 0 1 0
Q4 =  0 20 1  Â4 =  0 0 1
0 −25 0 0 0 0

3.18

1 (λ) = (λ − λ1 )3 (λ − λ2 ) ψ1 (λ) = 1 (λ)

2 (λ) = (λ − λ1 ) 4
ψ2 (λ) = (λ − λ1 )3
3 (λ) = (λ − λ1 )4 ψ3 (λ) = (λ − λ1 )2
4 (λ) = (λ − λ1 )4 ψ4 (λ) = (λ − λ1 )

3.21
   
1 1 9 1 1 102
A10 = 0 0 1 A103 =  0 0 1 
0 0 1 0 0 1
 t t 
e e − 1 te − e + 1
t t

eAt =0 1 et − 1 
0 0 et

3.22
 
1 4t + 2.5t 2 3t + 2t 2
e A4 t =  0 1 + 20t 16t 
0 −25t 1 − 20t
324 ANSWERS TO SELECTED PROBLEMS

3.24
 
ln λ1 0 0
B= 0 ln λ2 0 
0 0 ln λ3
 
ln λ 1/λ 0
B =  0 ln λ 0 
0 0 ln λ

3.32 Eigenvalues: 0, 0. No solution for C1 . For any m1 , [m1 3 − m1 ] is a solution for C2 .

√
3.34 6, 1. 4.7, 1.7.

CHAPTER 4

4.2 y(t) = 5e−t sin t for t ≥ 0.

4.3 For T = π ,
   
−0.0432 0 1.5648
x[k + 1] = x[k] + u[k]
0 −0.0432 −1.0432
y[k] = [2 3]x[k]

4.5 MATLAB yields |y|max = 0.55, |x1 |max = 0.5, |x2 |max = 1.05, and |x3 |max = 0.52 for unit
step input. Define x̄1 = x1 , x̄2 = 0.5x2 , and x̄3 = x3 . Then
   
−2 0 0 1
x̄˙ =  0.5 0 0.5  x̄ +  0  u y = [1 − 2 0]x̄
0 −4 −2 1
The largest permissible a is 10/0.55 = 18.2.
4.8 They are not equivalent but are zero-state equivalent.
4.11 Using (4.34), we have
   
−3 0 −2 0 1 0
 0 −3 0 −2  0 1
ẋ = 
 1
x +  u
0 0 0  0 0
0 1 0 0 0 0
   
2 2 4 −3 0 0
y= x+ u
−3 −2 −6 −2 1 1

4.13
   
−3 1 0 0 2 2
 −2 0 0 0  
ẋ =   x +  4 −3  u
 0 0 −3 1   −3 −2 
0 0 −2 0 −6 −2
 Answers to Selected Problems 325
   
1 0 0 0 0 0
y= x+ u
0 0 1 0 1 1
Both have dimension 4.

4.16
 $t 2 
1 0 e0.5τ dτ
X(t) = 2
0 e0.5t
 2 $t 2 
1 −e0.5t t0 e0.5τ dτ

(t, t0 ) = 2 −t 2 )
0 e0.5(t 0

 t 
e−t e
X(t) =
0 2e−t
 
e−(t−t0 ) 0.5(et et0 − e−t et0 )

(t, t0 ) =
0 e−(t−t0 )

4.20
 
ecos t−cos t0 0

(t, t0 ) =
0 e− sin t+sin t0

4.23 Let x̄(t) = P(t)x(t) with

 
e− cos t 0
P(t) =
0 esin t
˙ = 0 · x̄ = 0.
Then x̄(t)

4.25
 
t 2 e−λt
ẋ = 0 · x +  −2te−λt  u y = [eλt teλt t 2 eλt ]x
e−λt
   
3λ −3λ2 λ3 1
ẋ =  1 0 0  x +  0  u y = [0 0 2]x
0 1 0 0

CHAPTER 5

5.1 The bounded input u(t) = sin t excites y(t) = 0.5t sin t , which is not bounded. Thus the system
is not BIBO stable.

5.3 No. Yes.

5.6 If u(t) = 3, then y(t) → −6. If u(t) = sin 2t , then y(t) → 1.26 sin(2t + 1.25).

5.8 Yes.
326 ANSWERS TO SELECTED PROBLEMS

5.10 Not asymptotically stable. Its minimal polynomial can be found from its Jordan form as ψ(λ) =
λ(λ + 1). λ = 0 is a simple eigenvalue, thus the equation is marginally stable.
5.13 Not asymptotically stable. Its minimal polynomial can be found from its Jordan form as ψ(λ) =
(λ − 1)2 (λ + 1). λ = 1 is a repeated eigenvalue, thus the equation is not marginally stable.
5.15 If N = I, then
 
2.2 1.6
M=
1.6 4.8
It is positive definite; thus all eigenvalues have magnitudes less than 1.

5.17 Because x Mx = x [0.5(M+M )]x, the only way to check positive definiteness of nonsymmetric
M is to check positive definiteness of symmetric 0.5(M + M ).
5.20 Both are BIBO stable.

5.22 BIBO stable, marginally stable, not asymptotically stable. P (t) is not a Lyapunov transformation.

CHAPTER 6

6.2 Controllable, observable.

6.5
   
−1 0 1
ẋ = x+ u y = [0 − 1]x + 2u
0 −1 0
Not controllable, not observable.

6.7 µi = 1 for all i and µ = 1.

6.9 y = 2u.

6.14 Controllable, not observable.

6.17 Using x1 and x2 yields

   
−2/11 0 −2/11
ẋ = x+ u y = [−1 − 1]x
3/22 0 3/22
This two-dimensional equation is not controllable but is observable.
Using x1 , x2 , and x3 yields
   
−2/11 0 0 −2/11
ẋ =  3/22 0 0  x +  3/22  u y = [0 0 1]x
1/22 0 0 1/22
This three-dimensional equation is not controllable and not observable.

6.19 For T = π , not controllable and not observable.

6.21 Not controllable at any t . Observable at every t .

 Answers to Selected Problems 327

CHAPTER 7

7.1
   
−2 1 2 1
ẋ =  1 0 0  x +  0  u y = [0 1 − 1]x
0 1 0 0
Not observable.

7.3
   
−2 1 2 a1 1
 1 0 0 a2  0
ẋ = 
 0
x +  u y = [0 1 − 1 c4 ]x
1 0 a3  0
0 0 0 a4 0
For any ai and c4 , it is an uncontrollable and unobservable realization.
   
−3 −2 1
ẋ = x+ u y = [0 1]x
1 0 0
A controllable and observable realization.

7.5

Solve the monic null vector of

  
−1 −1 0 0 −N0
 0 2 −1 −1   
   D0  = 0
 4 0 0 2   −N1 
0 0 4 0 D1
as [−0.5 0.5 0 1] . Thus 0.5/(0.5 + s) = 1/(2s + 1).

7.10
   
0 1 0
ẋ = x+ u y = [1 0]x
−2 −1 1

7.13

1 (s) = s(s + 1)(s + 3) deg = 3

2 (s) = (s + 1) (s + 2)
3 2
deg = 5

7.17 Its right coprime fraction can be computed from any left fraction as
  −1
2.5 s + 0.5 0.5s s 2 + 0.5s
Ĝ(s) =
2.5 s + 2.5 s − 0.5 −0.5
or, by interchanging their two columns,
328 ANSWERS TO SELECTED PROBLEMS

  −1
s + 0.5 2.5 s 2 + 0.5s 0.5s
Ĝ(s) =
s + 2.5 2.5 −0.5 s − 0.5
Using the latter, we can obtain
   
−0.5 −0.25 −0.25 1 −0.5
ẋ =  1 0 0 x + 0 0 u
0 0.5 0.5 0 1
 
1 0.5 2.5
y= x
1 2.5 2.5

CHAPTER 8

8.1 k = [4 1].

8.3 For
 
−1 0
F= k̄ = [1 1]
0 −2
we have
 
0 1/13
T= k = k̄T−1 = [4 1]
1 9/13

8.5 Yes, yes, yes.

8.7 u = pr − kx with k = [15 47 − 8] and p = 0.5.

8.9 u[k] = pr[k] − kx[k] with k = [1 5 2] and p = 0.5. The overall transfer function is
0.5(2z2 − 8z + 8)
ĝf (z) =
z3
and the output excited by r[k] = a is y[0] = 0, y[1] = a , y[2] = −3a , and y[k] = a for
k ≥ 3.
8.11 Two-dimensional estimator:
     
−2 2 0.6282 1
ż = z+ u+ y
−2 −2 −0.3105 0
 
−12 −27.5
x̂ = z
19 32
One-dimesnional estimator:

ż = −3z + (13/21)u + y
  
−4 21 y
x̂ =
5 −21 z
 Answers to Selected Problems 329

8.13 Select
 
−4 3 0 0
 −3 4 0 0 
F= 0 0 −5 4 


0 0 −4 −5
   
1 0 1 0 62.5 147 20 515.5
If K̄ = then K =
0 0 0 0 0 0 0 0
   
1 0 0 0 −606.2 −168 −14.2 −2
If K̄ = then K =
0 0 1 0 371.1 119.2 14.9 2.2

CHAPTER 9

9.2 C(s) = (10s + 20)/(s − 6), p = −0.2.

9.4 C(s) = (22s − 4)/(s − 16), p = 1. There is no need for a feed forward gain because ĝ(s)
contains 1/s .

9.6 C(s) = (7s 3 + 14s 2 + 34s + 25)/(s(s 2 + 4)).

9.8 Yes, no. The transfer function from r to u is ĝur (s) = s/(s + 1)(s − 2), which is not BIBO
stable.
9.10 Yes, no, no, no, yes, no (row-wise).

9.12 C1 (s) = −2(s + 3)/(s − 21), C2 (s) = (28s − 6)/(s − 21). A(s) = s − 21 is not Hurwitz. Its
implementation in Fig. 9.4(a) will not be totally stable. A minimal realization of [C1 (s) − C2 (s)]
is
   
r r
ẋ = 21x + [−48 − 582] y = x + [−2 − 28]
y y
from which an op-amp circuit can be drawn.

9.15 (1) a > 0 and b0 = −2b1 . (2) Yes, b0 = −2, b1 = −4.

9.16 The 1 × 2 compensator

1
C(s) = [3.5s + 12 − 2]
s + 3.5
will place the closed-loop poles at −2, −1 ± j, −3. No.
9.18 If we select
F(s) = diag((s + 2)(s 2 + 2s + 2)(s + 3), (s 2 + 2s + 2))
then we cannot find a feedforward gain to achieve tracking. If
 
(s + 2)(s 2 + 2s + 2)(s + 3) 0
F(s) =
1 s 2 + 2s + 2
then the compensator
330 ANSWERS TO SELECTED PROBLEMS

 −1  
s − 4.7 −53.7 −30.3s − 29.7 4.2s − 12
C(s) = A−1 (s)B(s) =
−3.3 s − 4.3 −0.7s − 0.3 4s − 1
and the feedforward gain matrix
 
0.92 0
P=
−4.28 1
will achieve the design.

9.20 The diagonal transfer matrix

 −2(s − 1) 
0
 2 
Ĝo (s) =  s + 2s + 2
−2(s − 1) 
0
s 2 + 2s + 2
is implementable. The proper compensators A−1 (s)L(s) and A−1 (s)M(s) with
   
s + 5 −14 −2(s + 3) 2(s + 3)
A(s) = L(s) =
0 s+4 2 −2s
 
−6 13s + 1
M(s) =
2 −2s
in the two-parameter configuration will achieve the design.