Robust Adaptive Control by Petros A. Ioannou, Jing Sun - HTML preview

PLEASE NOTE: This is an HTML preview only and some elements such as links or page numbers may be incorrect.
Download the book in PDF, ePub, Kindle for a complete version.

for all i.

Because f is uniformly continuous, there exists a number δ( 0) > 0 such that

|f ( t) − f ( ti) | < 0 for every t ∈ [ t

2

i, ti + δ( 0)]

Hence, for every t ∈ [ ti, ti + δ( 0)], we have

|f ( t) | = |f ( t) − f ( ti) + f( ti) | ≥ |f( ti) | − |f( t) − f( ti) |

0

0 = 0

2

2

which implies that

ti+ δ( 0)

ti+ δ( 0)

f ( τ ) =

|f ( τ ) | dτ > 0 δ( 0)

(3.2.4)

t

2

i

ti

where the first equality holds because f ( t) retains the same sign for t ∈ [ ti, ti+ δ( 0)].

On the other hand, g( t) = t f ( τ ) has a limit as t → ∞ implies that there exists 0

a T ( 0) > 0 such that for any t 2 > t 1 > T ( 0) we have

|g( t 1) − g( t 2) | < 0 δ( 0)

2

i.e.,

t 2

f ( τ ) dτ < 0 δ( 0)

t

2

1

which for t 2 = ti + δ( 0) , t 1 = ti contradicts (3.2.4), and, therefore, lim t→∞ f( t) = 0.

3.2. PRELIMINARIES

77

The proof of Lemma 3.2.5 follows directly from that of Lemma 3.2.6 by

noting that the function f p( t) is uniformly continuous for any p ∈ [1 , ∞)

because f, ˙

f ∈ L∞.

The condition that f ( t) is uniformly continuous is crucial for the results

of Lemma 3.2.6 to hold as demonstrated by the following example.

Example 3.2.4 Consider the following function described by a sequence of isosceles

triangles of base length 1 and height equal to 1 centered at n where n = 1 , 2 , . . . ∞

n 2

as shown in the figure below:

f ( t)

1

❈✄

❈✄

✄ ❈

· · · · · ·

✄ ❈

· · · · · ·

✄ ❈

✄ ❈

1

✛2

n

2

1

1

4

n 2

This function is continuous but not uniformly continuous. It satisfies

t

1 1

π 2

lim

f ( τ ) =

=

t→∞ 0

2

n 2

12

n=1

but lim t→∞ f( t) does not exist.

The above example also serves as a counter example to the following situ-

ation that arises in the analysis of adaptive systems: We have a function V ( t)

with the following properties: V ( t) 0, ˙

V ≤ 0. As shown by Lemma 3.2.3

these properties imply that lim t→∞ V ( t) = V∞ exists. However, there is no

guarantee that ˙

V ( t) 0 as t → ∞. For example consider the function

t

V ( t) = π −

f ( τ )

0

where f ( t) is as defined in Example 3.2.4. Clearly,

V ( t) 0 ,

˙

V = −f ( t) 0 ,

∀t ≥ 0

and

π 2

lim V ( t) = V∞ = π −

t→∞

12

78

CHAPTER 3. STABILITY

but lim

˙

t→∞ V ( t) = lim t→∞ f ( t) does not exist. According to Barb˘

alat’s

lemma, a sufficient condition for ˙

V ( t) 0 as t → ∞ is that ˙

V is uniformly

continuous.

3.2.3

Positive Definite Matrices

A square matrix A ∈ Rn×n is called symmetric if A = A . A symmet-

ric matrix A is called positive semidefinite if for every x ∈ Rn, x Ax ≥ 0

and positive definite if x Ax > 0 ∀x ∈ Rn with |x| = 0. It is called neg-

ative semidefinite ( negative definite) if −A is positive semidefinite (positive

definite).

The definition of a positive definite matrix can be generalized to non-

symmetric matrices. In this book we will always assume that the matrix

is symmetric when we consider positive or negative definite or semidefinite

properties.

We write A ≥ 0 if A is positive semidefinite, and A > 0 if A is positive

definite. We write A ≥ B and A > B if A − B ≥ 0 and A − B > 0,

respectively.

A symmetric matrix A ∈ Rn×n is positive definite if and only if any one

of the following conditions holds:

(i) λi( A) > 0 , i = 1 , 2 , . . . , n where λi( A) denotes the i th eigenvalue of A, which is real because A = A .

(ii) There exists a nonsingular matrix A 1 such that A = A 1 A 1 .

(iii) Every principal minor of A is positive.

(iv) x Ax ≥ α|x| 2 for some α > 0 and ∀x ∈ Rn.

The decomposition A = A 1 A 1 in (ii) is unique when A 1 is also symmetric.

In this case, A 1 is positive definite, it has the same eigenvectors as A, and

its eigenvalues are equal to the square roots of the corresponding eigenvalues

1

of A. We specify this unique decomposition of A by denoting A 1 as A 2 , i.e.,

1

1

A = A 2 A 2 where A 2 is a positive definite matrix and A / 2 denotes the

transpose of A 1 / 2.

A symmetric matrix A ∈ Rn×n has n orthogonal eigenvectors and can

be decomposed as

A = U Λ U

(3.2.5)

where U is a unitary (orthogonal) matrix (i.e., U U = I) with the eigen-

3.3. INPUT/OUTPUT STABILITY

79

vectors of A, and Λ is a diagonal matrix composed of the eigenvalues of A.

Using (3.2.5), it follows that if A ≥ 0, then for any vector x ∈ Rn

λmin( A) |x| 2 ≤ x Ax ≤ λmax( A) |x| 2

Furthermore, if A ≥ 0 then

A 2 = λmax( A)

and if A > 0 we also have

1

A− 1 2 = λmin( A)

where λmax( A) , λmin( A) is the maximum and minimum eigenvalue of A,

respectively.

We should note that if A > 0 and B ≥ 0, then A + B > 0, but it is not

true in general that AB ≥ 0.

3.3

Input/Output Stability

The systems encountered in this book can be described by an I/O mapping

that assigns to each input a corresponding output, or by a state variable

representation. In this section we shall present some basic results concerning

I/O stability. These results are based on techniques from functional analysis

[42], and most of them can be applied to both continuous- and discrete-

time systems. Similar results are developed in Section 3.4 by using the state

variable approach and Lyapunov theory.

3.3.1

Lp Stability

We consider an LTI system described by the convolution of two functions

u, h : R+ → R defined as

t

t

y( t) = u ∗ h =

h( t − τ ) u( τ ) =

u( t − τ ) h( τ )

(3.3.1)

0

0

where u, y is the input and output of the system, respectively. Let H( s) be

the Laplace transform of the I/O operator h( ·). H( s) is called the transfer

80

CHAPTER 3. STABILITY

function and h( t) the impulse response of the system (3.3.1). The system

(3.3.1) may also be represented in the form

Y ( s) = H( s) U ( s)

(3.3.2)

where Y ( s) , U ( s) is the Laplace transform of y, u respectively.

We say that the system represented by (3.3.1) or (3.3.2) is Lp stable if

u ∈ Lp ⇒ y ∈ Lp and y p ≤ c u p for some constant c ≥ 0 and any

u ∈ Lp. When p = , Lp stability, i.e., L∞ stability, is also referred to as

bounded-input bounded-output (BIBO) stability.

The following results hold for the system (3.3.1).

Theorem 3.3.1 If u ∈ Lp and h ∈ L 1 then

y p ≤ h 1 u p

(3.3.3)

where p ∈ [1 , ∞] .

When p = 2 we have a sharper bound for y p than that of (3.3.3) given by

the following Lemma.

Lemma 3.3.1 If u ∈ L 2 and h ∈ L 1 , then

y 2 sup |H( ) | u 2

(3.3.4)

ω

For the proofs of Theorem 3.3.1, Lemma 3.3.1 see [42].

Remark 3.3.1 It can be shown that (3.3.4) also holds [232] when h( ·) is of

the form

0

t < 0

h( t) =

i=0 fiδ( t − ti) + fa( t)

t ≥ 0

where fa ∈ L 1 , ∞

i=0 |fi| < ∞ and ti are nonnegative finite constants.

The Laplace transform of h( t) is now given by

H( s) =

fie−sti + Ha( s)

i=0

which is not a rational function of s. The biproper transfer functions

that are of interest in this book belong to the above class.

3.3. INPUT/OUTPUT STABILITY

81

Remark 3.3.2 We should also note that (3.3.3) and (3.3.4) hold for the

truncated functions of u, y, i.e.,

yt p ≤ h 1 ut p

for any t ∈ [0 , ∞) provided u ∈ Lpe. Similarly,

yt 2 sup |H( ) | ut 2

ω

for any t ∈ [0 , ∞) provided u ∈ L 2 e. This is clearly seen by noticing

that u ∈ Lpe ⇒ ut ∈ Lp for any finite t ≥ 0.

It can be shown [42] that inequality (3.3.3) is sharp for p = because

h 1 is the induced norm of the map T : u → T u = y from L∞ into L∞, i.e.,

T ∞ = h 1. Similarly for (3.3.4) it can be shown that the induced norm

of the linear map T : L 2 → L 2 is given by

T 2 = sup |H( ) |

(3.3.5)

ω∈R

i.e., the bound (3.3.4) is also sharp.

The induced L 2 norm in (3.3.5) is referred to as the H∞ norm for the

transfer function H( s) and is denoted by

H( s) = sup |H( ) |

ω∈R

Let us consider the simple case where h( t) in (3.3.1) is the impulse response

of an LTI system whose transfer function H( s) is a rational function of s.

The following theorem and corollaries hold.

Theorem 3.3.2 Let H( s) be a strictly proper rational function of s. Then

H( s) is analytic in Re[ s] 0 if and only if h ∈ L 1 .

Corollary 3.3.1 If h ∈ L 1 , then

(i) h decays exponentially , i.e., |h( t) | ≤ α 1 e−α 0 t for some α 1 , α 0 > 0

(ii) u ∈ L 1 ⇒ y ∈ L 1 L∞, ˙ y ∈ L 1 , y is continuous and lim t→∞ |y( t) | = 0

(iii) u ∈ L 2 ⇒ y ∈ L 2 L∞, ˙ y ∈ L 2 , y is continuous and lim t→∞ |y( t) | = 0

(iv) For p ∈ [1 , ∞] , u ∈ Lp ⇒ y, ˙ y ∈ Lp and y is continuous

82

CHAPTER 3. STABILITY

For proofs of Theorem 3.3.2 and Corollary 3.3.1, see [42].

Corollary 3.3.2 Let H( s) be biproper and analytic in Re[ s] 0 . Then u ∈

L 2 L∞ and lim t→∞ |u( t) | = 0 imply that y ∈ L 2 L∞ and lim t→∞ |y( t) | =

0 .

Proof H( s) may be expressed as

H( s) = d + Ha( s)

where d is a constant and Ha( s) is strictly proper and analytic in Re[ s] 0. We have

y = du + ya,

ya = Ha( s) u

where, by Corollary 3.3.1, ya ∈ L 2

L∞ and |ya( t) | → 0 as t → ∞. Because

u ∈ L 2

L∞ and u( t) 0 as t → ∞, it follows that y ∈ L 2

L∞ and |y( t) | → 0

as t → ∞.

Example 3.3.1 Consider the system described by

e−αs

y = H( s) u,

H( s) = s + β

for some constant α > 0. For β > 0, H( s) is analytic in Re[ s] 0. The impulse response of the system is given by

e−β( t−α)

t ≥ α

h( t) =

0

t < α

and h ∈ L 1 if and only if β > 0. We have

1

h 1 =

|h( t) |dt =

e−β( t−α) dt =

0

α

β

and

e−αjω

1

H( s) = sup

=

ω

+ β

β

Example 3.3.2 Consider the system described by

2 s + 1

y = H( s) u,

H( s) = s + 5

3.3. INPUT/OUTPUT STABILITY

83

The impulse response of the system is given by

2 δ

h( t) =

∆( t) 9 e− 5 t

t ≥ 0

0

t < 0

where ha = 9 e− 5 t ∈ L 1. This system belongs to the class described in Re-

mark 3.3.1. We have

1

1 + 2

1 + 4 ω 2 2

H( s) = sup

= sup

= 2

ω

5 +

ω

25 + ω 2

Hence, according to (3.3.4) and Remarks 3.3.1 and 3.3.2, for any u ∈ L 2 e, we have