Problem Set 6 Solutions: 6.245: Multivariable Control Systems by A. Megretski

Massachusetts Institute of Technology
Department of Electrical Engineering and Computer Science

6.245: MULTIVARIABLE CONTROL SYSTEMS
by A. Megretski
Problem Set 6 Solutions
1
Problem 6.1
A standard feedback control design setup is defined by the differential
equations
x
1
=ax
1
+u, x
2
=x
2
+w,
and by

cu
y=w +bx
2
, z= ,
x
1
where a, b, c are real parameters.
(a) Find matrices A, B
1
, B
2
, C
1
, C
2
, D
11
, D
12
, D
21
, D
22
for this setup.
By inspection,

0 0 1 0 0

A=
a
, B
1
= , B
2
= , C
1
= , C
2
= 0 b ,
0 1 0 1 0 1

0 c
D
11
= , D
12
= , D
21
=1, D
22
=0.
0 0
(b) Find all values of a, b, c for which the setup is singular, inficating
frequency, multiplicity, and type (control/sensor) of the singular-
ity.
1
Version of April 26, 2004

2
Matrix

as 0 1
M
u
(s)=
0 1s 0
0 c
1 0 0
is right invertible for all s = j, R. The largest degree of determinant of a
3-by-3 minor of M
u
(s) equals d
= 1 when c = 0, and 2 otherwise. Hence, for

0
u
c=0,thereisacontrolsingularityat=,multiplicity1(thedierencebetween
system order and d
). For c=0, there are no control singularities.

u
The determinant of

as 0 0
M
y
(s)=
0 1s 1

0 b 1
equals
y
(s)=det(M
y
(s))=(s+a)(s+b+1).
Hence,thesystemhassensorsingularitiesonlywhena=0orb=1,inwhichcase
the singularity is located at =0, and its multiplicity 1 equals the multiplicity of
the zero s=0 of
y
(s) (one unless a=b+1=0, which implies multiplicity 2).
Problem 6.2
Find H-Infinity and H2 norms of G(s)=G
a
(s) as a function of real param-
eter a:
(a) G(s)=1/(s+a)2/(s+2a);
The impulse response g=g(t) of G is given by
2at
g(t)=e
at
2e .
Since

1
2at 4at
g(t)|
2
dt= e 4e
3at
+4e dt= ,
6a
0
|
0
G
H2
=1/
6a.
Since
s 1
G(s)= = ,
s
2
+3as+2a
2
3a+s+2a
2
/s
wehave G(j) 1/3a,wheretheequalitytakesplaceat=
2a. HenceG
= | |
1/3a.
3
(b) G(s)=(1exp(a
2
s))/s.
The impulse response g=g(t) of G is given by
g(t)=
1, t[0,a
2
],
0, otherwise.
Hence G
H2
=a, and G
=a
2
.
Problem 6.3
Find L2 gains of systems described below (input f, output g, defined for
t0). You are not required to prove correctness of your answer.
(a) y(t)=e
at
f(t).
L2 gain equals 1 for a0, and innity for a<0.
(b) y(t)=f(t)/(1+a
2
f(t)
2
).
L2 gain equals 1 for all a.
(c) y(t)=f(t+a).
L2 gain equals 1 for a0, and innity for a>0.
Problem 6.4
A standard H2 optimization setup is defined by the following transfer
functions:

0 1/(s+a)
P
wz
(s)= , P
wy
(s)=1, P
uz
(s)= , P
uy
(s)=1/(s+a),
0 1
where a R is a parameter. For all values of a for which the setup
is non-singular, find the H2 optimal controller, together with the as-
sociated Hamiltonian matrices, solutions of Riccati equations, and con-
troller/observer gains.
A state space model of the setup is given by

x
x =ax+u, x(0)=x
0
, z= , y=wx.
u
The setup is non-singular for a=0 (when a=0, there is a sensor singularity at =0).
4
The full information abstract H2 optimization has the form
2
X
=aX+U, (|X
2
+ U )dtmin.
0
| | |
The corresponding Hamiltonian matrix is

a 1
fi
= . H

1 a
The stabilizing solution of the associated Riccati equation is
P
fi
=
a
2
+1a,
and the optimal full state feedback gain is given by
K
fi
=a
a
2
+1.
The state estimation abstract H2 optimization has the form
=aq, (0)=
0
, q
2
dtmin.
0
| |
The corresponding Hamiltonian matrix is

a 1
se
= . H

0 a
The stabilizing solution of the associated Riccati equation is
P
se
=
0,
2a,
for
for
a>0,
a<0,
and the optimal state estimator gain is given by
0, for a>0,
L
se
=
2a, for a<0.
For a > 0, the H2 optimal controller has zero transfer function. For a < 0, the H2
optimal controller can be written in the observer-based form
d
u=(a
a
2
+1) x+u2a( x, x=a xy),
dt
and its transfer function is
2a(a
a
2
+1)
K(s)= .
s+
a
2
+12a
5
Problem 6.5
Open-loop plant P(s) is strictly proper, has a double pole at s=0, and a
zero at s = 0.1. Controller C = C(s) stabilizes the system on Figure 6.1,
e f
r
C(s)
v
P(s)
Figure 6.1: A SISO Feedback Setup

and ensures good tracking (transfer function from r to e has gain less
than 0.1) for frequencies up to 1 rad/sec. Find a good lower bound on
the maximal gain from r to e.
Let S denote the closed loop transfer function from r to e. Then, by assumption,
S(0.1)=1. Hence
log S(j) d | |
0.
2
+0.1
2
0
Using the fact that S(j) 0.1 for 1, and S(j)

for 1, we get | | | | S
10
d
logS
2
+1
log10
2
+1
,
10 0
i.e.
arctan(10)
10
/2arctan(10)
10
14.76
. S
Problem 6.6
Find the square of the H2 norm of
s
G(s)=
s
2
+a
1
s+a
0
as a rational function of real parameters a
1
,a
0
.
A state space model of G is given by

0

0 1
A= , B = , C = 0 1 , D=0.
a
1
1 a
0
The solution P of the Lyapunov equation

AP +PA
=BB
6
has the form

P =
1/2a
0
a
1
0
0
1/2a
1
.
Hence
G
H2
=
CPC
= .
1
2a
1

Problem Set 6 Solutions: 6.245: Multivariable Control Systems by A. Megretski

Uploaded by

Copyright:

Available Formats

Problem Set 6 Solutions: 6.245: Multivariable Control Systems by A. Megretski

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Problem Set 6 Solutions: 6.245: Multivariable Control Systems by A. Megretski

Uploaded by

Copyright:

Available Formats

Massachusetts Institute of Technology

Department of Electrical Engineering and Computer Science

= 1 when c = 0, and 2 otherwise. Hence, for

). For c=0, there are no control singularities.

Figure 6.1: A SISO Feedback Setup

The solution P of the Lyapunov equation

You might also like