ECE504: Lecture 4

ECE504: Lecture 4

D. Richard Brown III

Worcester Polytechnic Institute

23-Sep-2008

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 1 / 28

 ECE504: Lecture 4

Lecture 4 Major Topics

We are now starting Part II of ECE504: Quantitative and

qualitative analysis of systems
mathematical description → results about behavior of system
Today:
1. Solution of state equations for discrete-time systems
2. Solution of state equations for continuous-time systems
3. Some necessary linear algebra (and calculus review)
4. Examples

You should be reading Chen Chapter 4 now. You should also read
Chen 3.2-3.3 to learn about “basis”, “linear independence”, and
solutions to linear algebraic equations like Ax = y.

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 2 / 28

 ECE504: Lecture 4

Linear State-Space Description of Discrete-Time Systems

x[k + 1] = A[k]x[k] + B[k]u[k]

y[k] = C[k]x[k] + D[k]u[k]

We assume a general model with p inputs, q outputs, and n states.

Given an initial time k0 ∈ Z, an initial state x[k0 ] ∈ Rn , how does the

state evolve for k = k0 + 1, k0 + 2, . . . ?

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 3 / 28

 ECE504: Lecture 4

Solution to State Equation

Following our induction, for all k ≥ k0 , we can write
k−1
X
x[k] = Φ[k, k0 ]x[k0 ] + Φ[k, ℓ + 1]B[ℓ]u[ℓ]
ℓ=k0

where Φ is an n × n matrix valued function with two time arguments:


undefined
 k<j
Φ[k, j] = In k=j


A[k − 1]A[k − 2] · · · A[j] k > j
Remarks:
◮ I n is the n × n identity matrix.
◮ The order of the product A[k − 1]A[k − 2] · · · A[j] is important
because matrices don’t usually commute.
◮ The matrix function Φ : Z2 7→ Rn×n is called the state transition
matrix (STM) corresponding to A[k].
Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 4 / 28
 ECE504: Lecture 4

Zero-Input Response
Recall that linear systems have the nice property that we can
separately analyze the zero-input response and the zero-state
response.

Zero-input response: Given u[k] = 0 for all k ≥ k0 , we can write

x[k] = Φ[k, k0 ]x[k0 ]
The state transition matrix Φ[k, k0 ] describes how the state at time
k0 evolves to the state at time k ≥ k0 (in the absence of an input).

x[k0 ]
Φ[k, k0 ]
x[k]

If the STM Φ[k, k0 ] is invertible, then Φ−1 [k, k0 ] = Φ[k0 , k]. But
there is no guarantee that it is invertible. This operation is one-way.
Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 5 / 28
 ECE504: Lecture 4

Zero-State Response

Zero-state response: Given x[k0 ] = 0, we can write

k−1
X
x[k] = Φ[k, ℓ + 1]B[ℓ]u[ℓ]
ℓ=k0

In this case, we have to compute several state transition matrices:

Φ[k, k0 + 1], Φ[k, k0 + 2], . . . , Φ[k, k].

This looks like it might require a lot of computation as k gets larger.

Fortunately, there are some nice properties of the state transition
matrix that can ease the computational burden...

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 6 / 28

 ECE504: Lecture 4

Some Basic Properties of the State Transition Matrix

1. Φ[j, j] = I n for all j ∈ Z.

2. Φ[k + 1, j] = A[k]Φ[k, j] for all k ≥ j.
3. If ℓ ≤ j ≤ k, then Φ[k, ℓ] = Φ[k, j]Φ[j, ℓ].
This last property is called the “semigroup” property. It intuitively
says that the transition from x[ℓ] to x[k] is the same as the transition
from x[ℓ] to x[j] followed by the transition from x[j] to x[k].

x[k]
x[ℓ] Φ[j, ℓ]
x[j]
Φ[k, j]

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 7 / 28

 ECE504: Lecture 4

Special Case: A[k] ≡ A for all k ≥ k0

When A[k] ≡ A for all k ≥ k0 , the product
A[k − 1]A[k − 2] · · · A[j] = AA · · · A
How many A’s are involved in this product?

Hence, when A[k] ≡ A for all k ≥ k0 , the state transition matrix can be
written as

undefined k < j

Φ[k, j] = In k=j

 k−j
A k > j.
In this case, the solution to the DT state-update difference equation is
k−1
X
x[k] = Ak−k0 x[k0 ] + Ak−ℓ−1 B[ℓ]u[ℓ]
ℓ=k0

for all k ≥ k0 .
Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 8 / 28
 ECE504: Lecture 4

Discrete-Time Output Solution

For all k ≥ k0 , we can just plug our solution to the state equation into our
state-space output equation to get
k−1
X
y[k] = C[k]Φ[k, k0 ]x[k0 ] + C[k] Φ[k, ℓ + 1]B[ℓ]u[ℓ] + D[k]u[k]
| {z }
zero-input response ℓ=k 0
| {z }
zero-state response

If the system is time-invariant, then we can write

k−1
X
y[k] = CAk−k0 x[k0 ] + C Ak−ℓ−1 Bu[ℓ] + Du[k]
| {z }
zero-input response ℓ=k0
| {z }
zero-state response

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 9 / 28

 ECE504: Lecture 4

Remarks on Discrete-Time State-Space Solutions

For causal, linear, lumped discrete-time systems with p input terminals, q

output terminals, and n states, we have shown that, given x[k0 ] and u[k]
for all k ≥ k0 , there exists a unique solution to the discrete-time
state-update difference equation:
k−1
X
x[k] = Φ[k, k0 ]x[k0 ] + Φ[k, ℓ + 1]B[ℓ]u[ℓ]
ℓ=k0

for all k ≥ k0 with Φ[k, j] as defined earlier.

This also implies that, given x[k0 ] and u[k] for all k ≥ k0 , there exists a
unique solution to the discrete-time output equation.

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 10 / 28

 ECE504: Lecture 4

Discrete-Time State-Space Example

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 11 / 28

 ECE504: Lecture 4

Continuous-Time Linear Systems

ẋ(t) = A(t)x(t) + B(t)u(t) (1)

y(t) = C(t)x(t) + D(t)u(t) (2)

Theorem
For any t0 ∈ R, any x(t0 ) ∈ Rn , and any u(t) ∈ Rp for all t ≥ t0 , there
exists a unique solution x(t) for all t ∈ R to the state-update differential
equation (1). It is given as
Z t
x(t) = Φ(t, t0 )x(t0 ) + Φ(t, τ )B(τ )u(τ ) dτ t ∈ R
t0

where Φ(t, s) : R2 7→ Rn×n is the unique function satisfying

d
Φ(t, s) = A(t)Φ(t, s) with Φ(s, s) = I n .
dt

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 12 / 28

 ECE504: Lecture 4

Theorem Remarks

◮ Note that this theorem claims two things:

1. A solution to the state-update equation always exists.
2. The solution is unique.
◮ Our strategy to prove the theorem:
1. We will first show that, given two solutions to the
state-update equation, they must be identical. This
establishes uniqueness.
2. We will then establish existence constructively by giving a
solution and showing that it satisfies the state-update
equation.

Before doing any of this, however, we are going to need to learn

some more linear algebra (and a calculus refresher)...

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 13 / 28

 ECE504: Lecture 4

Euclidean Norm of a Vector

Definition
For x ∈ Rn , the Euclidean norm of x is given as

1/2
kxk := x21 + · · · + x2n .

The Euclidean norm of vectors in R1 , R2 , or R3 is just your normal notion

of distance/length.

Some useful facts (easy to show from the definition):

◮ kxk2 = x⊤ x.
◮ kαxk = |α|kxk for any α in R.
◮ kx + yk ≤ kxk + kyk for any x ∈ Rn and any y ∈ Rn . This is often
called the triangle inequality.

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 14 / 28

 ECE504: Lecture 4

Induced Euclidean Norm of a Matrix

Definition
For A ∈ Rn×n , the induced Euclidean norm of the matrix A is given as

kAk := max kAxk.

x∈Rn and kxk=1

◮ The set of vectors x where kxk = 1 is a unit-sphere in Rn .

◮ The induced Euclidean norm of A is the maximum value of kAxk as
x ranges over all points on this unit-sphere.
◮ Intuitively, kAk gives a measure of how much A can magnify the
length (Euclidean norm) of a vector in Rn .
Some useful facts (not too hard to show from the definition):
◮ kA + Bk ≤ kAk + kBk for any A ∈ Rn×n and any B ∈ Rn×n .
◮ kABk ≤ kAkkBk for any A ∈ Rn×n and any B ∈ Rn×n .
◮ kAxk ≤ kAkkxk for any A ∈ Rn×n and any x ∈ Rn .
Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 15 / 28
 ECE504: Lecture 4

Schwarz Inequality

Theorem
Given x ∈ Rn and y ∈ Rn , then

|x⊤ y| = |x1 y1 + · · · + xn yn | ≤ kxkkyk

Proof:

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 16 / 28

 ECE504: Lecture 4

Leibniz’ rule
Theorem
If f (t, τ ) is continuous and all of the necessary derivatives exist, then
Z w(t) Z w(t)
d d
f (t, τ ) dτ = ẇ(t)f (t, w(t)) − v̇(t)f (t, v(t)) + f (t, τ ) dτ
dt v(t) v(t) dt

The proof can be found in most calculus textbooks.

Two particularly useful special cases are

Z
d t
f (τ ) dτ = f (t)
dt a
Z
d a
f (τ ) dτ = −f (t)
dt t

where a is not a function of t.

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 17 / 28
 ECE504: Lecture 4

Back to the Theorem: Uniqueness Proof

ẋ(t) = A(t)x(t) + B(t)u(t)

We first want to show that any solution x(t) to this state-update

differential equation must be unique.

To show this, suppose we had two solutions to the state-update differential

equation, x1 (t) and x2 (t) for t ∈ [s1 , t1 ], both of which satisfy the initial
condition x1 (t0 ) = x2 (t0 ) = x(t0 ). Let’s prove that x1 (t) must be
identical to x2 (t)...

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 18 / 28

 ECE504: Lecture 4

Theorem: Existence Proof Warmup

We now know that, if a solution to the state-update DE exists, it must be

unique. We now need to show that a solution always exists.

To develop some intuition, let’s first assume that everything is scalar,

i.e. p = q = n = 1. Our state update equation becomes

ẋ(t) = a(t)x(t) + b(t)u(t)

Let
Z t 
φ(t, s) := exp a(τ ) dτ
s

What is φ(s, s)?

d
What is dt φ(t, s)?

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 19 / 28

 ECE504: Lecture 4

Theorem: Existence Proof Warmup

nR o
t
Note that φ(t, s) = exp s a(τ ) dτ always exists and satisfies its own
differential equation:
d
φ(t, s) = a(t)φ(t, s) with φ(s, s) = 1.
dt
Now lets try the following solution to the scalar state-update differential
equation with initial state condition x(t0 ):
Z t
x(t) = φ(t, t0 )x(t0 ) + φ(t, τ )b(τ )u(τ ) dτ ∀t ∈ R
t0

To see that this solution is valid, we should confirm two things:

1. Does our solution satisfy the initial condition requirement of the
scalar state-update DE?
2. Does our solution really solve the scalar state-update DE?
Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 20 / 28
 ECE504: Lecture 4

Theorem: Existence Proof

For the general (non-scalar) case, we propose the solution

Z t
x(t) = Φ(t, t0 )x(t0 ) + Φ(t, τ )B(τ )u(τ ) dτ (3)
t0

where the state transition matrix satisfies the matrix differential equation
d
Φ(t, s) = A(t)Φ(t, s) with Φ(s, s) = I n . (4)
dt
To complete the existence proof, we need to:
1. Show that (3) with Φ defined according to (4) satisfies the initial
condition requirement of the state-update DE.
2. Show that (3) with Φ defined according to (4) is indeed a solution to
the state-update DE.
3. Show that there always exists a solution to the matrix DE (4).
Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 21 / 28
 ECE504: Lecture 4

Remarks on the CT State-Transition Matrix Φ(t, s)

1. Computation of Φ(t, s) is almost always difficult.
2. The Peano-Baker series is only one way to compute Φ(t, s). Other
(perhaps better?) ways:
◮ Directly solve the matrix state-update differential equations (not always
possible)
◮ Fundamental matrix method (see Chen 4.5)
◮ Other methods...
3. Question: Is it possible that different methods for computing the
STM will lead to different Φ(t, s)?
4. Unlike the DT-STM Φ[k, j], the CT-STM Φ(t, s) is defined for any
(t, s) ∈ R2 . This means that we can specify an initial state x(t0 ) and
compute the system response at times prior to t0 .
5. Φ(t, s) possesses the semi-group property, i.e.
Φ(t, τ ) = Φ(t, s)Φ(s, τ )
for any (t, τ, s) ∈ R3 .
Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 22 / 28
 ECE504: Lecture 4

Important Special Case: A(t) ≡ A

When A(t) ≡ A, the state-transition matrix Peano-Baker series becomes

∞
X
Φ(t, s) = M k (t, s)
k=0
∞ Z tZ
X τ1 Z τk−1
= ··· AA · · · A} dτk · · · dτ1
| {z
k=0 s s s
k−fold product
∞
X Z tZ τ1 Z τk−1
= Ak ··· dτk · · · dτ1
k=0 s s s

To compute M k (t, s), let’s look at k = 0, 1, 2, . . . to see the pattern:

◮ What is M 0 (t, s)?
◮ What is M 1 (t, s)?
◮ What is M 2 (t, s)?
◮ What is M 3 (t, s)?
Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 23 / 28
 ECE504: Lecture 4

Important Special Case: A(t) ≡ A

By induction, we can show that

1
M k (t, s) = Ak (t − s)k
k!
hence
∞
X 1
Φ(t, s) = Ak (t − s)k
k!
k=0

Suppose, for x ∈ C, we have

∞
X xk
f (x) =
k!
k=0

Math trivia question: What is f (x)?

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 24 / 28

 ECE504: Lecture 4

Matrix Exponential

Definition (Matrix Exponential)

Given W ∈ Cn×n , the matrix exponential is defined as
∞
X Wk
exp(W ) =
k!
k=0

Note that the matrix exponential is not performed element-by-element, i.e.

   w 
w11 w12 e 11 ew12
exp 6= w21 w22
w21 w22 e e

Matlab has a special function (expm) that computes matrix exponentials.

Calling exp(W) will not give the same results as expm(W).

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 25 / 28

 ECE504: Lecture 4

Important Special Case: A(t) ≡ A

Putting it all together, when A(t) ≡ A, we can say that

Φ(t, s) = exp {(t − s)A}

Then the solution to the state-update DE is

Z t
x(t) = exp {(t − t0 )A} x(t0 ) + exp {(t − τ )A} B(τ )u(τ ) dτ
t0

and the output equation is

Z t
y(t) = C(t) exp {(t − t0 )A} x(t0 ) + C(t) exp {(t − τ )A} B(τ )u(τ ) dτ + D(t)u(t)
t0

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 26 / 28

 ECE504: Lecture 4

Contrast/Comparison Between CT and DT Solutions

Similarities
◮ Results have same “look”.
◮ Both have state transition matrices with same intuitive properties,
e.g. semigroup.

Differences
◮ In DT systems, x[k] is only defined for k ≥ k0 because the DT-STM
Φ[k, k0 ] is only defined for k ≥ k0 .
◮ In CT systems, x(t) is only defined for all t ∈ R because the CT-STM
Φ(t, t0 ) is defined for all (t, t0 ) ∈ R2 .
◮ We didn’t prove this, but the CT-STM Φ(t, t0 ) is always invertible.
This is not true of the DT-STM Φ[k, k0 ].

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 27 / 28

 ECE504: Lecture 4

Conclusions

◮ Solution to LTI or LTV discrete-time state-space

difference equations (existence and uniqueness)
◮ Solution to LTI or LTV continuous-time state-space
differential equations (existence and uniqueness)
◮ Special case: time-invariant A matrix
◮ LTI discrete-time systems: Ak−j
◮ LTI continuous-time systems: exp{(t − τ )A}

Worcester Polytechnic Institute D. Richard Brown III 23-Sep-2008 28 / 28