Today’s goals

• State space so far

– Definition of state variables
– Writing the state equations
– Solution of the state equations in the Laplace domain
– Phase space and phase diagrams
• Today
– Stability in state space
– State feedback control

2.004 Fall ’07 Lecture 28 – Wednesday, Nov. 14

State space overview
x1
w tower
sway
k1
m1

b1

From the Equation of Motion to the State—Space representation:

µ ¶ µ ¶
x q1
mẍ(t)+bẋ(t)+kx(t) = w(t) → ≡ q(t) = state, y(t) ≡ ẋ(t) output
ẋ q2
µ ¶ µ ¶µ ¶ µ ¶ µ ¶
q̇1 0 1 q1 0 q1
⇒ q̇(t) = = + w(t); y(t) = (0 1) ≡ cq.
q̇2 −k/m −b/m q2 1 q2

Solution to the state equations:

µ ¶
0 1
A= sq̂(s) = Aq̂(s) + bW (s) ⇒
−k/m −b/m
µ ¶ −1
0 q̂(s) = (sI − A) bW (s).
b=
1 −1
Y (s) = cq̂(s) = c (sI − A) bW (s).

2.004 Fall ’07 Lecture 28 – Wednesday, Nov. 14

State space solution to the uncompensated 2.004 Tower model
µ ¶
−1 1 1/m1
q̂(s) = (sI − A) bW (s) = 2 W (s).
s + (b1 /m1 ) s + (k1 /m1 ) s/m1

From this result we can obtain transfer functions for position, velocity:
−1
for position choose c = (1 0) , X(s) ≡ Y (s) = c (sI − A) bW (s) ⇒

X(s) 1/m1
= 2 .
W (s) s + (b1 /m1 ) s + (k1 /m1 )
−1
for velocity choose c = (0 1) , V (s) ≡ Y (s) = c (sI − A) bW (s) ⇒
V (s) s/m1
= 2 .
W (s) s + (b1 /m1 ) s + (k1 /m1 )

position velocity phase diagram

velocity position
2.004 Fall ’07 Lecture 28 – Wednesday, Nov. 14
 Poles are the eigenvalues of A | Stability
Consider the eigevalue problem for the matrix A: The system represented by A is
stable if A’s eigenvalues
Aξ = μξ,
have negative real part
where the solutions for μ are the eigenvalues and ξ are the eigenvectors. (i.e., are on the left—hand half—plane.)
To solve the eigenvalue problem, we set det (μI − A) = 0. The phase diagram is then oriented
That is, the eigenvalues are the roots of the determinant towards the origin (“sink.”)
of the matrix (μI − A).

Recall that the state—space solution was

−1 adj (sI − A)

velocity
q̂(s) = (sI − A) bW (s) = bW (s) =
det (sI − A)
µ ¶
1 1/m1
= 2 W (s),
s + (b1 /m1 ) s + (k1 /m1 ) s/m1
where adj(.) denotes the adjoint. The same denominator det (sI − A)
appears in the transfer functions for both velocity and position.
This denominator is also referred to as characteristic equation. position
Therefore, the poles of the system are the roots of the determinant
of the matrix (sI − A), i.e., the eigenvalues. The system represented by A is
The uncompensated 2.004 Tower is a 2nd order system with unstable if A’s eigenvalues
have positive real part (phase diagram
k1 b1 explodes outwards — “source”)
ωn2 = ; ζ= √
m1 2 k1 m1 and marginally stable if A’s eigenvalues
have zero real part (phase diagram
Therefore the eigenvalues/poles are
rotates around the origin
³ p ´ without either approaching or moving away.)
s± = −ωn ζ ± ζ 2 − 1 .

2.004 Fall ’07 Lecture 28 – Wednesday, Nov. 14

 Eigenvectors and modes
Let’s do a specific example: m1 = 1, b1 = 1,pk1 = 1, The two eigenvectors ξ (+) , ξ (−)
that is ωn = 1, ζ = 1/2, s± = −(1/2) ± j 3/2. are referred to as the modes of the system.
This system is stable and, in fact, underdamped, The imaginary parts of the corresponding poles
consistent with ζ < 1 and poles off the real axis. are the eigenfrequencies of the modes.
In the uncompensated 2.004 Tower,
Now let’s compute the eigenvectors, starting with ξ (+) the two modes are degenerate
corresponding to the eigenvalue s+ : because the two poles are conjugate
µ ¶ ( ¡ √ ¢ (+) (i.e., they have the same imaginary
(+)
0 1 (+) (+) ξ2 = −1 + j 3 ξ1 /2 parts with ± signs.)
ξ = s+ ξ ⇒ (+) (+) ¡ √ ¢ (+)
−1 −1 ξ 1 + ξ2 = −1 + j 3 ξ2 /2. This is true for any 2nd order system.
The compensated 2.004 Tower
It can be verified that the two equations are equivalent. is a 4th order system, and so it has
Therefore, the eigenvector corresponding to s+ is two non—degenerate modes.
⎛ 1 ⎞
√ The significance of a mode
⎜ 2 √ ⎟ is that if the system is excited
ξ (+)
=⎜⎝
⎟ α(+) ,
1 3 ⎠ with a sinusoid of frequency equal
−√ + j √
2 2 to the mode’s eigenfrequency,
then the response of the system will
where α(+) is any arbitrary real number. By convention,
¯ ¯ be the mode itself (i.e., the eigenvector.)
the eigenvector is written so that if α(+) = 1 ⇒ ¯ξ (+) ¯ = 1. At other frequencies, the response
Similarly we can find the eigenvector ξ (−) is a mixture of modes.
corresponding to the eigenvalue s− :
⎛ 1 ⎞
√
⎜ 2 √ ⎟
ξ (−) ⎜
=⎝ ⎟ α(−) .
1 3 ⎠
−√ − j √
2 2

2.004 Fall ’07 Lecture 28 – Wednesday, Nov. 14

State space representation as block diagram

q̇(t) = Aq(t) + Bw(t),

y(t) = Cq(t).

w +
.
q q y
B C
+

Figure by MIT OpenCourseWare.

Equivalent block diagram representation as transfer function:

W (s) Y (s)
C (sI − A)−1 B

2.004 Fall ’07 Lecture 28 – Wednesday, Nov. 14

 State feedback

r + w +
.
q q y
B C
- +

Figure by MIT OpenCourseWare.

w(t) = r(t) − Kq(t) ⇒

³ ´ ³ ´
q̇(t) = Aq(t) + B r(t) − Kq , q̇(t) = A − BK q(t) + Br(t),
⇒
y(t) = Cq(t). y(t) = Cq(t).

Y (s) −1
Closed—Loop TF: = C (sI − A + BK) B
R(s)

2.004 Fall ’07 Lecture 28 – Wednesday, Nov. 14

 Example W(s) 20(s + 5) Y(s)
G(s) =
(Nise 12.1) s(s + 1)(s + 4)
Figure by MIT OpenCourseWare.

Design problem: We are given transient response requirements of 9.5% over- Let the gain matrix be
shoot and 0.74sec settling time. Moreover, we would like to approximately ⎛⎞ ⎛ ⎞
cancel the zero in the opn—loop transfer function. 0 0 0 0
To meet these goals, we select two closed—loop poles at −5.4 ± j7.2. These K = (k1 k2 k3 ) ⇒ BK = ⎝ 0 ⎠ (k1 k2 k3 ) = ⎝ 0 0 0 ⎠
meet the transient reponse requirements. Moreover, we select an additional 1 k1 k2 k3
closed—loop pole at −5.1 to approximately cancel the open—loop zero. Therefore, ⎛ ⎞
the desired closed—loop transfer function should be proportional to 0 1 0
⇒ A − BK = ⎝ 0 0 1 ⎠
(s + 5) (s + 5) −k1 −(4 + k2 ) −(5 + k3 )
= 3 .
(s + 5.1)(s + 5.4 − j7.2)(s + 5.4 + j7.2) s + 15.9s2 + 136s + 413
The denominator of the transfer function is the determinant of the matrix
Next we convert the given transfer function to a state—space representation. ⎛ ⎞
s −1 0
This is done by first considering a system without the open—loop zeros, i.e. of sI − A + BK = ⎝ 0 s −1 ⎠
the form s + k1 s + (4 + k2 ) s + (5 + k3 )
X(s) 1 1 ³ ´
= = 3 ⇔ x(3) + 5ẍ + 4ẋ = w.
W (s) s(s + 1)(s + 4) s + 5s2 + 4s det sI − A + BK = s3 + (5 + k3 )s2 + (4 + k2 )s + k1 .

The state variable are selected as Equating coefficients with the desired denominator (characteristic equation) s3 +
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ 15.9s2 + 136s + 413, we obtain the gains
q1 x 0 1 0 0
q = ⎝ q2 ⎠ = ⎝ ẋ ⎠ ⇒ A = ⎝ 0 0 1 ⎠, b = ⎝ 0 ⎠. k1 = 413; k2 = 132; k3 = 10.9.
q3 ẍ 0 −4 −5 1
The state space representation of the closed—loop system is
This choise of state variables is also known as phase—variable form, because ⎛ ⎞ ⎛ ⎞
it agrees with the phase diagram representation that we saw earlier (except in 0 1 0 0
this case we have a 3rd order system, and hence three state/phase variables: q̇ = ⎝ 0 0 1 ⎠ q + ⎝ 0 ⎠ r.
position, velocity, acceleration.) −413 −136 −15.9 1
We also need to determine the observation matrix C. Since the open—loop
transfer function has a zero, the response includes a derivative term; that is, y = (100 20 0) q.
h i The closed—loop transfer function is
Y (s) = 20(s+5)X(s) ⇒ y(t) = 20 ẋ(t) + 5x(t) = 20(q2 +5q1 ) ⇒ c = (100 20 0) .
Y (s) 20(s + 5)
= 3 .
R(s) s + 15.9s2 + 136s + 413

2.004 Fall ’07 Lecture 28 – Wednesday, Nov. 14