Lecture 4

Feedback linearisation
A single input-single output control ane sys-
tem obeys the following state and output equa-
tions:
x = f(x) +g(x)u
y = h(x)
(1)
where x
n
is a vector, u is a scalar input, y is
a scalar output, f :
n

n
and g :
n

n
are vector elds and h :
n
is a scalar
function. Suppose that f(), g() and h() are
dierentiable functions.
1
Obviously y does not depend explicitly on u
since u is not an argument of function h.
If u is changed instantaneously, there will be
no immediate change in y. The change comes
gradually via x.
To check the corresponding behaviour of y we
can dierentiate the output equation:
y =
h
x
x =
h
x
(f(x) +g(x)u)
=
h
x
f(x) +
h
x
g(x)u
(2)
This shows that y depends directly on u if and
only if [h/x]g(x) = 0. We will say that the
system has relative degree 1 if [h/x]g(x) = 0
2
Now, let us assume that [h/x]g(x) = 0. Dif-
ferentiating the output y once more, we obtain:
y =

x
_
h
x
_
f(x) +

x
_
h
x
f(x)
_
g(x)u (3)
The system is said to have relative degree 2 if:
h
x
g(x) = 0 and

x
_
h
x
f(x)
_
g(x) = 0 (4)
3
The idea of relative degree can be easily gener-
alised with the help of the following notation.
Denition 1 (Lie derivative). Let :
n

be a dierentiable function and f a vector eld,

x
, f(x)
_
=

x
f(x) (5)
The Lie derivative of along f is usually de-
noted as L
f
, so that:
L
f
(x) =
n

i=1

x
i
f
i
(x) (6)
for each given x U.
4
If L
f
is again dierentiated along another vec-
tor eld g, the following is obtained:
L
g
L
f
(x) =
(L
f
)
x
g(x) (7)
This operation could be used recursively along
the same vector eld. L
k
f
indicates that is
being dierentiated k times along f such that
L
k
f
(x) =

_
L
k1
f

_
x
f(x)
L
0
f
(x) = (x)
(8)
5
Example 1. Suppose that x = [x
1
, x
2
]
T
, (x) =

x
1
and f(x) = [x
2
, sinx
1
]
T
. Therefore the
Lie derivative is:
L
f
(x) =

x
f(x)
=
_
1
2

x
1
, 0
_ _
x
2
sinx
1
_
=
x
2
2

x
1
(9)
6
Denition 2 (Relative degree). The SISO
system given by Equation (1) is said to have a
relative degree r at a point x
o
if
1. L
g
L
k1
f
h(x) = 0 for all x in a neighbour-
hood of x
o
k = 1, . . . , r 1.
2. L
g
L
r1
f
h(x
o
) = 0
The following formula describing the time deriva-
tives of the output is an immediate conse-
quences of Denition 2:
d
k
y
dt
k
= y
(k)
=
_
_
_
L
k
f
h(x) , k = 1, . . . , r 1
L
k
f
h(x) +L
g
L
k1
f
h(x)u, k = r
(10)
7
Example 2. Consider the controlled van der
Pol equation:
x
1
= x
2
x
2
= x
1
+(1 x
2
1
)x
2
+u, > 0
(11)
If the output is y = x
1
, the derivatives of the
output are given by:
y = x
1
= x
2
y = x
2
= x
1
+(1 x
2
1
)x
2
+u, > 0
(12)
Hence the system has relative degree r = 2 in

2
.
8
In terms of Lie derivatives we have:
L
g
L
0
f
h =
h
x
g = [1, 0]
_
0
1
_
= 0 (13)
L
f
h =
h
x
f = [1, 0]
_
x
2
x
1
+(1 x
2
1
)x
2
_
= x
2
(14)
L
g
L
f
h =
(x
2
)
x
g = [0, 1]
_
0
1
_
= 1 = 0 (15)
It is not dicult to show that if the output was
y = x
2
then the relative degree of the system
would be r = 1 in
2
.
9
Remark 1. For SISO linear systems:
x = Ax +Bu
y = Cx
(16)
The relative degree is the dierence in degree
between the nominator and denominator of its
equivalent transfer function:
G(s) = C(sI A)
1
B (17)
Equivalently, the relative degree of a SISO lin-
ear system is the least positive integer r such
that:
L
g
L
r1
f
h = CA
r1
B = 0 (18)
10
Denition 3 (Strong relative degree). A sys-
tem is said to have a strong relative degree if
the relative degree is r for all x
o

n
.
Input-output linearisation
Consider a system with a strong relative degree
r. Then we have that the r-th derivative of the
output is given by:
y
(r)
= L
r
f
h +L
g
L
r1
f
hu, L
g
L
r1
f
h = 0 (19)
There is an interesting possibility. Introduce
the feedback law:
u =
1
L
g
L
r1
f
h
(v L
r
f
h) (20)
where v is a reference signal.
11
The resulting relationship is between v and y
(r)
is:
y
(r)
= v (21)
which is a linear system. Taking the Laplace
transform:
y(s) =
1
s
n
v(s) (22)
By using the feedback law (20), we have ob-
tained a system that is linear from the refer-
ence signal v to the output y.
12
Proposition 1. Suppose a system of the form
of Equation 1 has a relative degree r at a point
x
o
. Dene,

1
(x) = h(x)
.
.
.

r
(x) = L
r1
f
h(x)
(23)
If r is strictly less than n, it is always possible to
nd nr arbitrary functions
r+1
(x), . . . ,
n
(x)
such that the mapping
(x) = [
1
(x), . . . ,
r
(x), . . . ,
n
(x)]
T
(24)
has a jacobian matrix which is nonsingular at
x
o
and therefore represents a local coordinates
transformation in a neighbourhood of x
o
. More-
over, it is always possible to choose
r+1
(x),
. . . ,
n
(x) in such a way that
L
g

i
(x) = 0 , r +1 i n , (25)
and all x around x
o
.
Proof. See the book by Isidori (1995).
13
Dierentiating Equation 23 with respect to time
and using Equation 10, we obtain:
d
1
(x)
dt
=
dh(x)
dt
= L
f
h(x) =
2
(x)
.
.
.
d
r1
(x)
dt
= L
r1
f
h(x) =
r
(x)
d
r
(x)
dt
= L
r
f
h(x) +L
g
L
r1
f
h(x)u
(26)
By introducing the variables:
=
_

1
.
.
.

r
_

_ =
_

1
(x)
.
.
.

r
(x)
_

_ (27)
=
_

1
.
.
.

nr
_

_ =
_

r+1
(x)
.
.
.

n
(x)
_

_ (28)
14
Then Equation 26 can be written as follows:
d
1
dt
=
2
(t)
.
.
.
d
r1
dt
=
r
(t)
d
r
dt
= b(, ) +a(, )u
(29)
where y =
1
a(, ) = L
g
L
r1
f
h(
1
(, ))
b(, ) = L
r
f
h(
1
(, ))
(30)
Notice that
x =
1
(, ) (31)
15
As far as the other new co-ordinates are con-
cerned, corresponding to = [
1
, . . . ,
nr
]
T
,
we cannot expect any special structure in the
corresponding equations. However, if
r+1
, . . . ,
n
have been chosen such that L
g

r+i
= 0, i =
1, . . . , (n r), then:
d
i
dt
= L
f

r+i
(x(t)), i = 1, . . . , (n r) (32)
By dening q
i
(, ) = L
f

r+i
(
1
(, )), then
we have that the derivatives of
i
can be writ-
ten as follows:
d
i
dt
= q
i
(, ), i = 1, . . . , (n r) (33)
or, in vector form:
d
dt
= q(, ) (34)
16
In summary, the normal form of a control
ane SISO nonlinear system of Equation 1,
with relative degree r around a point x
o
, is
given by:

1
=
2

2
=
3
.
.
.

r1
=
r

r
= b(, ) +a(, )u
= q(, )
y =
1
(35)
17
Notice that if we choose the feedback law:
u =
1
a(, )
(v b(, ))
=
1
L
g
L
r1
f
h(
1
(, ))
(v L
r
f
h(
1
(, )))
=
1
L
g
L
r1
f
h(x)
(v L
r
f
h(x))
(36)
Then the resulting system from the reference
input v to the output y =
1
is linear. The
normal form is given by:

1
=
2

2
=
3
.
.
.

r1
=
r

r
= v
= q(, )
y =
1
(37)
Notice, however, that the internal dynamics
= q(, ) are possibly nonlinear, so that the
system has not been fully linearised by the
feedback law.
18
If the feedback law is changed to: :
u =
1
a(, )
(v b(, )
0

1

r1

r
)
(38)
Then the normal form is given by:

1
=
2

2
=
3
.
.
.

r1
=
r

r
=
0

1

r1

r
+v
= q(, )
y =
1
(39)
and the transfer function from input to output
is:
y(s)
v(s)
=
1

r1
s
r
+ +
0
(40)
19
Example 3. Consider the system:
x
1
= x
3
2
x
2
= u
y = x
1
+x
2
(41)
Since y = x
3
2
+u the system has a strong rela-
tive degree r = 1. Introduce the following new
states to dene the normal form:
=
1
= y = x
1
+x
2
=
2
= x
1
(42)
Notice that
2
satises L
g

2
= 0 and that the
transformation:
_

_
= (x) =
_
1 1
1 0
_ _
x
1
x
2
_
(43)
is non-singular and has the following inverse:
x =
1
(, ) =
_
0 1
1 1
_ _

_
(44)
20
The normal form is:

=
3
+u
= ( +)
3
y =
(45)
The feedback law:
u = a +v
3
, a > 0 (46)
results in the following normal form:

= a +v
= ( +)
3
y =
(47)
which is linear from the reference v to the out-
put y, with transfer function:
y(s)
v(s)
=
1
s +a
(48)
21
It is easy to see that the unforced internal dy-
namics (or zero dynamics) which are dened
for = 0 ( =
3
), are asymptotically stable
around = 0, so that the control signal should
be bounded for bounded v and . Notice that
is bounded if v is bounded since with a > 0
the transfer function y(s)/v(s) is stable.
What would happen with the control signal if
the internal dynamics were unstable?
22