AC Lec6

You might also like

You are on page 1of 17

Introduction to Linear Systems

Insoon Yang

Department of Electrical and Computer Engineering


Seoul National University

CORE
Control + Optimization Research Lab

1 / 17
Dynamical systems
Time index (discrete): k = 0, 1, 2, . . .
State variables: x1 (k), x2 (k), . . . , xn (k)
System of equations:
x1 (k + 1) = f1 (x1 (k), x2 (k), . . . , xn (k), k)
x2 (k + 1) = f2 (x1 (k), x2 (k), . . . , xn (k), k)
..
.
xn (k + 1) = fn (x1 (k), x2 (k), . . . , xn (k), k)
with initial conditions x1 (0), x2 (0), . . . , xn (0)
Example: position, velocity, voltage, current, temperature, chemical
concentration, etc.
fi ’s define how the state evolves.
Given initial conditions, the system has a unique solution.
2 / 17
Linear systems

A simple but popular class of dynamical systems is linear systems:

x1 (k + 1) = a11 (k)x1 (k) + a12 (k)x2 (k) + · · · + a1n (k)xn (k) + w1 (k)
x2 (k + 1) = a21 (k)x1 (k) + a22 (k)x2 (k) + · · · + a2n (k)xn (k) + w2 (k)
..
.
xn (k + 1) = an1 (k)x1 (k) + an2 (k)x2 (k) + · · · + ann (k)xn (k) + wn (k)

aij (k)’s: coefficients or parameters


wi (k)’s: forcing terms
Vector form:
x(k + 1) = A(k)x(k) + w(k),
where x(k) = (x1 (k), . . . , xn (k)), w(k) = (w1 (k), . . . , wn (k)), and
A(k) is the n × n matrix consisting of the coefficients aij (k)’s.
3 / 17
Inputs
In most applications, the forcing terms are derived from specific
inputs to the system.
These inputs are often designed to guide the behavior of the system.
Control variables or input variables: u1 (k), . . . , um (k)
Linear system with inputs:
x1 (k + 1) = a11 (k)x1 (k) + a12 (k)x2 (k) + · · · + a1n (k)xn (k)
+ b11 u1 (k) + b12 u2 (k) + · · · + b1m (k)um (k)
..
.
xn (k + 1) = an1 (k)x1 (k) + an2 (k)x2 (k) + · · · + ann (k)xn (k)
+ bn1 u1 (k) + bn2 u2 (k) + · · · + bnm (k)um (k)
Vector form:
x(k + 1) = A(k)x(k) + B(k)u(k),
where B(k) is the n × m matrix consisting of the coefficients bij (k)’s.
4 / 17
State vector and state space
x(k) ∈ Rn , called the state vector, is a complete description of the
system at time k, at least in so far as determining future behavior.
The future behavior can be calculated recursively from the system of
equations once the current state is known.
State space: the n-dimensional space in which the state vector is
defined.
One can visualize the evolution of a dynamical system in terms of the
state vector moving within the state space.
0.2

-0.2

-0.4
y

-0.6

0
-0.8 /2

3 /2
-1
-1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4
x

5 / 17
Homogeneous systems

A linear system is said to be homogeneous if there’s no forcing term:

x(k + 1) = A(k)x(k)

Solution?
x(1) = A(0)x(0)
x(2) = A(1)x(1) = A(1)A(0)x(0)
..
.
x(k) = A(k − 1)A(k − 2) · · · A(0) x(0)
| {z }
=:Φ(k,0) state-transition matrix

Multiplication of any initial state vector by the state-transition matrix


yields the state at time k.
6 / 17
State-transition matrix
Definition
The state-transition matrix of the homogeneous system is

Φ(k, l) = A(k − 1)A(k − 2) · · · A(l), k>l


Φ(k, k) = I

Equivalently, it is the matrix satisfying

Φ(k + 1, l) = A(k)Φ(k, l), k>l


Φ(l, l) = I

Note that x(k) = Φ(k, l)x(l) (state-transition property).


Unfortunately, there is no shortcut procedure for calculating the
state-transition matrix.
However, the special structure of certain systems often enables one to
calculate it easily.
7 / 17
Example: Time-invariant system

Consider a time-invariant linear system

x(k + 1) = Ax(k),

where A is fixed, independent of k.

It is easy to see that

Φ(k, 0) = Ak , k≥0

or, more generally,

Φ(k, l) = Ak−l , k ≥ l.

8 / 17
Fundamental sets of solutions
Assume A(k) is nonsingular for all k.

Fundamental set of solutions x1 (k), . . . , xn (k) ∈ Rn : linearly


independent collection of n solutions to x(k + 1) = A(k)x(k)

Fundamental matrix of solutions

X(k) := x1 (k) x2 (k) · · · xn (k)


 

Since each column of X(k) is a solution,

X(k + 1) = A(k)X(k)

Note that X(k) is nonsingular for all k due to the nonsingularity of


A(k)’s.
9 / 17
Fundamental matrix of solutions vs. state-transition matrix

Let x(k) be the solution with initial condition x(0).

Define
x̄(k) := X(k) X(0)−1 x(0) = X(k)α
| {z }
=:α

Thus, x̄(k) is a linear combination of solutions. By the linearity of the


system, this linear combination is itself a solution.

Since x̄(0) = x(0), the two solutions x̄(k) and x(k) must be identical:

x(k) = X(k)X(0)−1 x(0)

Thus,
Φ(k, 0) = X(k)X(0)−1

10 / 17
Fundamental matrix of solutions vs. state-transition matrix

Proposition
Let X(k) be a fundamental matrix of solutions to x(k + 1) = A(k)x(k).
Then, the state-transition matrix is given by

Φ(k, l) = X(k)X(l)−1 , k ≥ l.

11 / 17
Example
Consider a linear homogeneous systems
    
x1 (k + 1) 1 k + 1 x1 (k)
=
x2 (k + 1) 0 1 x2 (k)

Apply two linearly independent initial conditions (1, 0), (0, 1).
One can find the following two solutions
x1 (k) = (1, 0), x2 (k) = (k(k + 1)/2, 1).

Fundamental matrix of solutions:


 
1 k(k + 1)/2
X(k) =
0 1

Given the initial conditions x(0) = (α1 , α2 ), the general solution is


x(k) = X(k)X(0)−1 x(0) = (α1 + α2 k(k + 1)/2, α2 )
12 / 17
Linear control systems

Discrete-time linear system with a control input

x(k + 1) = A(k)x(k) + B(k)u(k)

Proposition
Given x(0) and u(k), the solution of the system can be expressed as
k−1
X
x(k) = Φ(k, 0)x(0) + Φ(k, l + 1)B(l)u(l)
l=0

13 / 17
Proof
Pk−1
To show: x(k) = Φ(k, 0)x(0) + l=0 Φ(k, l + 1)B(l)u(l)

14 / 17
Superposition

The linearity of the system implies that the solution can be computed
by the principle of superposition:
The total response due to several inputs is the sum of their individual
responses, plus an initial condition term.

This can be seen in


k−1
X
x(k) = Φ(k, 0)x(0) + Φ(k, l + 1)B(l)u(l)
l=0

15 / 17
Time-invariant systems

x(k + 1) = Ax(k) + Bu(k)

State-transition matrix:

Φ(k, l + 1) = Ak−l−1

General solution:
k−1
X
x(k) = Ak x(0) + Ak−l−1 Bu(l)
l=0

16 / 17
Impulse response
Impulse response: The response due to u(0) = 1 at time 0 is

x(k) = Ak−1 b

General version: The response due to u(l) = 1 at time l is

x(k) = Ak−l−1 b

(shifted by l time units)

Recall that the general


Pk−1solution is
x(k) = Ak x(0) + l=0 Ak−l−1 bu(l).

Thus, the response of a linear time-invariant system to an arbitrary


sequence of inputs is made up from the basic response pattern of the
impulse response.
17 / 17

You might also like