Cambridge University Engineering Dept.

Third year

Module 3F2: Systems and Control

EXAMPLES PAPER 3
OBSERVERS AND STATE-FEEDBACK

Observability and Controllability

1. Consider the satellite example of Section 4.7 of Lecture Notes 3.

(a) Suppose the bias is on the star tracker (measurement of θ) rather than on the
rate gyro (measurement of θ̇). Show that the technique shown in the Notes,
based on appending the bias to the state vector, won’t work in this case,
because the system is not observable.
(b) Suppose that there is a constant but unknown disturbance torque d acting on
the satellite: J θ̈ = u + d, d˙ = 0, and the star tracker measures θ correctly
(without bias). Show that, if d is appended to the state vector then the
system is observable (using the star tracker output only), and hence that d
can be estimated.

2. (a) Derive state-space equations in standard form for the systems shown in
Figures 1 and 2.
(b) Investigate the controllability and observability of the systems obtained in
(a), as the parameter α varies.
(c) For values of α when either controllability or observability is lost determine
the set of states that can be reached from the origin and the set of initial
conditions that do not affect the output.
(d) In each case that observability or controllability is lost, show that the
corresponding eigenvalue does not appear as a pole of the transfer function,
because of a pole-zero cancellation.

- +
+ 
? x1 ?
- -
R
-
R -
- -
 x2  y
u
6+

α 

Figure 1:
Module 3F2: Examples Paper 3 2


u
+? +
+ x2  x1 
R R ?
- - - - - −3 - -
 + + y
+6 -
6

α 

Figure 2:

3. The inverted pendulum laboratory experiment has the linearized equations:

   
0 1 0 0 0
 0 0 ω02 − ω12 0   1 
ẋ = 
 0 0
x +  u
0 1   0 
2
0 0 ω0 0 1

(a) Show that the system is controllable.

(b) Will this system be observable if just one of the states is measured, i.e. if
y = xi for each of the four cases i = 1, 2, 3, 4?

4. Consider the linearized two link manipulator of Question 6 of Examples Paper 1.

(a) Is it controllable from

(i) T1 and T2 ,
(ii) T1 alone (T2 = T2e ),
(iii) T2 alone (T1 = T1e )?
(b) What implication can you draw on the achievable closed-loop behaviour for
this system for these three cases? Include a discussion of the achievable
steady-state conditions.
Module 3F2: Examples Paper 3 3

5. (a) In lectures we defined the controllability Gramian Wc (t) as

Z t
T
Wc (t) = eAτ BB T eA τ dτ.
0

Assume that the system is asymptotically stable; then

Wc (∞) = limt→∞ Wc (t) exists. By considering
d n Aτ T
o
e BB T eA τ
dτ
show that

AWc (∞) + Wc (∞)AT = −BB T

(b) For the system shown in Figure 1, with α = −1, find Wc (∞).
(Note that Wc (∞) is symmetric.)

Observers, State Feedback

6. Design an observer for Question 1(b), such that the state estimation error decays
with a time constant of 1 sec.

7. Design a state-feedback controller for the system of Figure 1, which places both
closed-loop poles at -10 sec−1 . (Assume that both state variables are available for
feedback.)

Minimum energy control

8. A “pirate ship”1 at a fairground is controlled by a suitable electric motor. An

idealised and simplified model of the system is given by:

I θ̈ = −M gLθ + τ

where θ is the angle (of the ship’s mast to the vertical), I = L2 M , g ' 10ms−2 ,
L = 10m, and τ (Nm) is the geared motor torque. It is desired to minimise the
losses in the motor which are assumed to be proportional to
Z t1
J= τ 2 (t)dt.
0

Show that the minimum possible value of J when moving the ship from rest to
θ(t1 ) = π/4, θ̇(t1 ) = 0, with t1 = nπ sec (n = 1, 2, · · · ) is given by,

πI 2
Jmin =
8n
1
For those who do not frequent fairgrounds this ride looks like a ship swinging from a pivot above it.
Module 3F2: Examples Paper 3 4

when the optimal input is

(−1)n+1 I sin(t)
τopt (t) = .
2n
Hint: Define the input to be the scaled torque: u = τ /I.

State feedback, Observers, Everything

9. A system satisfies the state equation

ẋ = Ax + Bu
y = Cx + Du

where
   
0 0 1  
A= , B= , C= 1 1 , D=0
0 −1 −2

and it is desired to design a feedback controller so that y(t) tracks an external

reference signal r(t) as closely as possible with satisfactory stability margins.

(a) Verify that the transfer function is

(1 − s)
G(s) =
s(s + 1)
with impulse response 1 − 2e−t . (Note the zero at s = +1 makes the response
initially go in the ‘wrong’ direction).
(b) Design a state feedback controller

u = −Kx + M r

so that the closed-loop poles are both at −β and y(t) → r when r is a step.
Calculate the resulting response of y(t) to a step change in r.
(c) Design a state observer with gain matrix L so that the poles oftheobserver
1
are both at −α. Calculate the state estimation error if x(0) = and the
  1
0
observer state, x̂(0) = .
0
(d) Show that if estimated state feedback is now used then the controller
equation will be
x̂˙ = (A − BK − LC) x̂ + Ly + BM r
u = −K x̂ + M r
and observe that the step response from r to y will be the same as in (b) if
x(0) = x̂(0) = 0.
Module 3F2: Examples Paper 3 5

(e) Now suppose that the system actually satisfies the state equations
ẋ = Aa x + Ba u
y = Ca x
and the measurement is (y + v) where v is observation noise.
Show that the resulting closed loop equation will be
       
d x Aa −Ba K x Ba M 0 r
= +
dt x̂ LCa A − BK − LC x̂ BM L v
       
y Ca 0 x 0 0 r
= +
u 0 −K x̂ M 0 v

(f) Show that the return-ratio (open-loop transfer function) of the system with
the loop ‘broken’ at the plant output will be,

H(s) = Ca (sI − Aa )−1 Ba K(sI − A + BK + LC)−1 L

and note that this transfer function will determine the stability margins with
respect to gain and phase uncertainty in the plant model.
(g) Investigate the appropriate choices for α and β by considering the speed of
the step responses from r to y, the amplitude of the required input, u, and
the gain and phase margins. A MATLAB .m file has been written to perform
these calculations. It can be invoked from MATLAB as follows (for example):
>> alpha = 1
>> beta = 1
>> fact = 1
>> Q83F2
The gain of G(s) is multiplied by the term fact.
The resulting plots and printout then give the Bode diagram for G(jω); the
eigenvalues of the closed loop A-matrix as in part (e); the step response from
r to y in the presence of approximately white noise on v (Normally
distributed sequence of standard deviation 0.01, and sampled at 0.01 s); and
the Nyquist diagram for the loop gain H(jω) as in part (f).
Module 3F2: Examples Paper 3 6

Answers

1. —
   
0 1 0
2. (a) Figure 1: ẋ = x+ u, y = [−1, 1]x + 0u.
α −1 1
   
−1 1 1
Figure 2: ẋ = x+ u, y = [−3, 1]x + 0u.
α 1 0
(b) and (c): Figure 1 — Always
 controllable. Not observable if α = 2, in which
1
case states of the form x = ρ are unobservable, for any scalar ρ.
1
Figure 2 —
 Not controllable if α = 0, in which case the reachable states are of the
ρ
form x = . Not observable if α = 3, in which case states of the form
  0
1
x= ρ are unobservable.
3
(d) Figure 1: When α = 2 the eigenvalues are 1, −2, but the transfer function is
1/(s + 2).
Figure 2: When α = 0 the eigenvalues are 1, −1 but the transfer function is
−3/(s + 1). When α = 3 the eigenvalues are 2, −2 but the transfer function is
−3/(s + 2).
3. (b) Observable from x1 only.
4. (a) Controllable from (u1 and u2 ) or u2 alone, but not from u1 alone.
(b) Can place poles using u2 alone, but cannot maintain a steady state value with
x1 6= 0.
 
1/2 0
5. (b) Wc (∞) =
0 1/2

6. Observer gain: L = [`1 , `2 , `3 ]T , where `1 = 3, `2 = 3, `3 = J. (This places all three

poles at −1 sec−1 , which is not the only possible solution.)
7. u = −Kx = −[k1 , k2 ]x, where k1 = 100 + α and k2 = 19.
8. —
9. (a) g(t) = 1 − 2e−t
(b) K = β 2 , 21 (1 − β)2 , M = β 2 , y(t) = 1 − e−βt (1 + β(1 + β)t).
 
   
α2 1 + α(1 − 2α)t
(c) L = , e(t) = exp (−αt) .
−(1 − α)2 1 + (1 − 2α)(1 − α)t
(g) Note the large noise amplification and poor stability margins when α or β are
chosen too large.

J.M.Maciejowski February 2011