Sei sulla pagina 1di 6

Linear systems

Lecture notes

Pavankumar Tallapragada

Electrical Engineering Department


Indian Institute of Science

August 10, 2017


Contents

1 Linear systems - preliminaries 2


1.1 Examples of difference equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 State-space representation of linear systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Properties of linear systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4.1 0 is always an equilibrium point for a homogeneous system . . . . . . . . . . . . . . . 4
1.4.2 Causality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4.3 Time invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4.4 Linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1
Chapter 1

Linear systems - preliminaries

In this chapter, we will cover some preliminary concepts about linear systems. In this course, although we
will mostly work with continuous time systems (differential equations), we will also look at discrete time
systems (difference equations) from time to time. In fact, we will start with some simple difference equations
that all of us have already encountered.

1.1 Examples of difference equations


Consider the sequence
1, 2, 4, 8, 16, . . .
This is a geometric sequence. What are some compact representations of this sequence? Lets denote the k th
number in this sequence by x(k). Then, a representation is x(k) = 2k for k N0 . We could also represent
this sequence recursively
x(k + 1) = 2x(k), x(0) = 1.
This representation is useful because it clearly represents the change from one time step to the next. Further,
it is also a more general description of the sequence. In fact, the difference equation represents a whole family
of sequences - for c R, if we set x(0) = c then we get the sequence in which x(k) = 2k c. So, there is
something special about the factor 2k here. We can also say that each sequence in this family, except
for the one generated by x(0) = 0, is unbounded. Observe that if x(0) = 0 then x(1) = 0 and in fact x(k) = 0
for any k. So, 0 is a fixed point or an equilibrium point for this system. Finally, it is an unstable equilibrium
point because no matter how close to 0 we start, the sequence diverges away from 0.
Now, consider the sequence
1 1 1 1
1, , , , , . . .
2 4 8 16
This geometric sequence belongs to the family generated by
1
x(k + 1) = x(k), x(0) = c.
2
And x(k) = ( 21 )k x(0). So, all sequences in this family converge to 0, which by the way is again an equilibrium
point. But now, it is a stable equilibrium.
Still continuing with difference equations, consider the following.
0, 1, 1, 2, 3, 5, 8, 13, . . .
This is the Fibonacci sequence. It is generated by
F (k + 1) = F (k) + F (k 1), F (0) = 0, F (1) = 1.
It is a two dimensional system. Let  
F (k)
x(k) = .
F (k + 1)

2
Then     
F (k + 1) 0 1 F (k)
x(k + 1) = = := Ax(k),
F (k + 2) 1 1 F (k + 1)
and the k th element in the sequence is the output of the system
 
F (k) = 1 0 x(k),
 
0
if the initial condition is appropriately chosen. The point (vector) is again an equilibrium point.
0
In future, we will refer the zero vector simply by 0 irrespective of the dimension of the state
space.
Closed-form expression for the state
x(k) = Ak x(0).
We will postpone the justification for this but notice the generalization compared to the scalar case. And
the k th element of the Fibonacci sequence is given by
 
0
F (k) = 1 0 x(k) = 1 0 Ak
   
.
1

Question: What kind of an equilibrium is 0 for this system? Does there exist an x(0) R2 ,
with x(0) 6= 0, such that limk x(k) = 0?

1.2 Differential equations


Consider the first order constant-coefficient, linear, homogeneous ODE
dx
x := = ax, x(0) = c.
dt
We know its solution is x(t) = eat c = eat x(0). Now, consider a second order linear, homogeneous ODE
d2 z
z := = a0 (t)z + a1 z.
dt2
Note that we need both z(0) and z(0) to uniquely specify a solution. Let us write the ODE as a system of
two first order ODEs. Let    
x z
x := 1 := .
x2 z
Then,  
0 1
x = x.
a0 (t) a1
Question: Does x(t) again depend on x(0) linearly, meaning, does there exist some matrix
Q(t) such that x(t) = Q(t)x(0) for all x(0)? What about higher dimensional systems?

1.3 State-space representation of linear systems


The general state-space representation of a continuous-time linear system is
x(t) = A(t)x(t) + B(t)u(t) (1.1a)
y(t) = C(t)x(t) + D(t)u(t). (1.1b)
The signals
x : [0, ) Rn , u : [0, ) Rm , y : [0, ) Rp
are the state, the input and the output of the system. The matrices A(t), B(t) C(t) and D(t) are time-
varying and of appropriate dimensions. The differential equation (1.1a) is the state equation and the algebraic
equation (1.1b) is the output equation.

3
For a given input signal u, the system (1.1) generates a whole collection of solutions or state and output
signals or trajectories. Given an input signal u and the initial condition x(0), the trajectories x and y
are uniquely determined.
The state x(t) at time t is a completely captures the effect of the initial conditions and the input signal
u up to time t.
y(t) depends on x(t) and u(t) only algebraically.
If m = 1 then the system is single input (SI) otherwise it is multi input (MI).
If p = 1 then the system is single output (SO) otherwise it is multi output (MO).
SISO, MIMO
If the matrices A(t), B(t), C(t) and D(t) are constant in time then it is a linear time-invariant system
(LTI) otherwise it is a linear time-varying (LTV) system.
For conciseness, we will drop time arguments of the signals when there is no ambiguity. For
system parameters and matrices we will always use the time argument if they are time-varying.

x = A(t)x + B(t)u, y = C(t)x + D(t)u, x Rn , u Rm , y Rp .


State space representation of discrete-time linear systems:
x(t + 1) = A(t)x(t) + B(t)u(t) (1.2a)
y(t) = C(t)x(t) + D(t)u(t). (1.2b)
The time t N0 := {0, 1, 2, 3, . . .}. More concisely,
x+ = A(t)x + B(t)u, y = C(t)x + D(t)u, x Rn , u R m , y R p .

1.4 Properties of linear systems


The following properties hold for both continuous and discrete time systems. The proofs in the discrete-time
case are similar to those below.

1.4.1 0 is always an equilibrium point for a homogeneous system


Equilibrium point is a point where there is no change. So, in the continuous-time case, x = 0 at an equilibrium
point. And in the discrete-time case, x+ = x at an equilibrium point. So, we see that
x = A(t)0 = 0.

1.4.2 Causality
The state and the output of the system at any time t depend only on the past inputs.
Consider the initialized system
x = A(t)x + B(t)u, x(0) = x0 ,
and lets make a copy of it
z = A(t)z + B(t)r, z(0) = x0 .
Let us suppose the signals u and r are identical up to time t, that is, u(s) = r(s) for all s [0, t]. Then, for
s [0, t],
e := x z = A(t)x + B(t)u A(t)z B(t)r, e(0) = 0
= A(t)e, e(0) = 0.
So, e(s) = 0 or x(t) = z(t) for all s [0, t]. Thus, the outputs are also the same on the interval [0, t] in the
two cases.

4
1.4.3 Time invariance
We say a system is time-invariant if time-shifted inputs cause a time-shift in the evolution of the state and
the output, when the initial conditions of the state are appropriately changed.
Earlier, we said

x = Ax + Bu
y = Cx + Du

is a time-invariant system. We need to verify this claim given the above definition. Let T be time-shift under
consideration and let
r(t) = u(t + T ), t [0, )
be the time-shifted version of the signal u. As in the proof of causality, lets make a copy of the system

z = Az + Br, z(0) = x(T ),

Lets define e(t) := z(t) x(t + T ). Then,

e(t) = Az(t) + Br(t) Ax(t + T ) Bu(t + T )


= Ae(t),

where we have used the definitions of r(t) and e(t). By definition, e(0) = 0 and since 0 is an equilibrium
point for a homogeneous system, we see that e(t) = 0 for all t [0, ). In other words, z(t) = x(t + T ) for
all t [0, ). Similarly,
Cz + Dr = Cx(t + T ) + Du(t + T ) = y(t + T ).
So, the claim follows.

1.4.4 Linearity
A linear combination of the inputs generates the exact same linear combination of the corresponding state
trajectories and outputs under appropriate initial conditions.
Consider

x = A(t)x + B(t)u1 , x(0) = x0 ,


z = A(t)z + B(t)u2 , z(0) = z0 ,
v = A(t)v + B(t)(u1 + u2 ), v(0) = x0 + z0 .

Claim: v = x + z for all t [0, ).


Proof: Exercise.

Construction of all outputs given an input signal


Linearity allows us to construct all the outputs corresponding to a given input u from a single output yf .
Let y be another output that u generates. Then,
0 = u u = that the zero input (input equal to 0 for all t 0) generates y yf .

Suppose yh is an output corresponding to the zero input (unforced or homogeneous system). Since
u = u + 0, y = yf + yh is another output corresponding to u.
These two observations mean that the set of all outputs corresponding to the input u can be found by finding
one solution yf corresponding to u and the set of all solutions yh to the unforced system.

Potrebbero piacerti anche