Lecture 3 - Laplace Transforms Linear Time Invariant Systems (LTI)

K. J. Åström The linear time invariant system

1. Review of LTI standardform 1 dn y d n− 1 y d n− 1 u

n
+ a1 n−1 + . . . + an y = b1 n−1 + . . . + bn u
dt dt dt
2. Introduction
is characterized by two polynomials
3. Laplace Transforms
4. The Transfer Function A( s) = sn + a1 sn−1 + a2 sn−2 + . . . + an−1 s + an

5. Qualitative Understanding of Signals and Systems B ( s) = b1 sn−1 + b2 sn−2 + . . . + bn−1 s + bn

6. Summary • The roots of A(s) are called poles of the system.

Theme: Simplify manipulation of equations and block dia- • The roots of B (s) are called zeros of the system.
grams. B ( s)
• The transfer function of the system is G (s) =
A(s)

Linear Time Invariant Systems (LTI) Interpretation of the Impulse Response


dn y d n− 1 y d n− 1 u n t
+ a1 + . . . + a n y = b1 + . . . + bn u y( t) = Ck−1( t) eα kt + g( t − τ )u(τ ) dτ
dtn dtn−1 dtn−1 0
k=1

has the solution

Let the system be initially at rest, i.e. Ck = 0 and let the input


n  t be an impulse at time 0. The output is then
y( t) = Ck−1 ( t) eα kt + g( t − τ )u(τ ) dτ
k=1 0
y( t) = g( t)
where {α k} are roots of the characteristic equation A, C k (t)
If the input is a unit step the output (the step response) is
polynomials and g is the impulse response, which has the form
 t  t


n
y ( t) = g( t − τ ) dτ = g(τ ) dτ
α kt
g( t) = C̄k−1( t) e 0 0
k=1
Experimental determination of step and impulse responses.
Notice appearance of α k again!

c K. J. Åström August, 2000 1

 1. Introduction Recall Cruise Control
To analyze the feedback system Process model
y sp e u y dv
+ 0.02v = u − 10θ
Σ Controller Process dt

PI controller
 t
u = k(v r − v) + ki (vr − v(τ )) dτ
−1 0

The closed loop system is described by (differentiate both

where the process and the controller are described as linear
equations and add them), introduce e = v r − v
time invariant systems we must manipulate the equations. We
will now introduce methods that simplify the manipulatins. d2 e de dθ
+ (0.02 + k) + ki e = 10
We will also develop methods that give an intuitive insight into dt2 dt dt
the behavior of systems. The mathematical tool of Laplace transforms is ideally suited
for these type of calculations.

 ∞
2. Laplace Transforms L f = F ( s) = e− st f ( t) dt
0
Consider a function f defined on 0 ≤ t < ∞ and a real number
σ > 0. Assume that f grows slower than eσ t for large t. The Transform of the derivative
Laplace transform F = L f of f is defined as  ∞ ∞  ∞
df f ( t) 
L = e− st dt = e− st f ( t) + s e− st f ( t) dt = − f (0) + sL f
 ∞ dt 0 dt 0 0
L f = F ( s) = e− st f ( t) dt
0 Initial value theorem
 ∞  ∞
Example 1:
lim sF ( s) = lim se− st f ( t) dt = lim e−v f ( t/ s) dv = f (0)
 ∞ ∞ 1 s→∞ s→∞ 0 s→∞ 0
1 
f ( t) = 1 , F ( s) = e− stdt = − e− st =
0 s 0 s Final value theorem
 ∞  ∞
Example 2: lim sF ( s) = lim se − st
f ( t) dt = lim e−v f ( t/ s) dv = f (∞)
s→0 s→0 s→0

1 − st∞
0 0
∞
1
f ( t) = e− at, F ( s) = e−( s+a) tdt = − e  =
0 s+a 0 s+a Behavior of f (t) for small t is similar to behavior of sF (s) for
large s and vice versa!
c K. J. Åström August, 2000 2
 Properties Inverse Transforms
Linearity: L (a f + bg) = aL f + bL g A simple way to find time functions corresponding to a rational
Differentiation: L = sL f − f (0)
df Laplace transform. Write F (s) in a partial fraction expansion
dt
t
Integration: L 0 f (τ )dτ = 1s L f B ( s) B ( s) C1 C2 Cn
F ( s) = = = + + ... +
A( s) ( s − a1 )( s − a2 ) . . . ( s − an ) s − a1 s − a2 s − an
Time shift: L f (t − T ) = e−sT L f B ( a k)
Ck = lim ( s − ak ) F ( s) =
Time stretch: L f (at) = 1a F ( 1a ) s→ a k ( ak − a1 ) . . . ( ak − ak−1 )( s − ak+1 ) . . . ( ak − an )
t
Convolution: L 0 f (t − τ )g(τ )dτ = F (s) G (s) The time function corresponding to the transform is
Final value Theorem: lim s→0 sF (s) = lim t→∞ f (t)
f ( t) = C1 ea1 t + C2 ea2 t + . . . + Cn ean t
Initial value Theorem: lim s→∞ sF (s) = limt→0 f (t)
• Note: Valid only if limits exist! Parameters a k give shape and numbers Ck give magnitudes.
Notice that a k may be complex numbers. With multiple roots
the constants Ck are instead polynomials.

Manipulating LTI Systems Cruise Control

The differentiation property L df = sL f − f (0) makes dv
dt Process model: + 0.02v = u − 10θ
the Laplace transform very convenient for dealing with LTI dt
t
systems, particularly if all initial values are zero. Differentiation PI controller: u = k(v r − v) + ki 0 (vr − v(τ ))dτ
of the time functions then simply corresponds to multiplication
of the transform with s. We then obtain the following recipe for Taking Laplace transforms
dealing with linear systems: ( s + 0.02) V ( s) = U ( s) − 10Θ( s)
• Take Laplace transforms of the equations E( s) = Vr ( s) − V ( s)
ki
• Take Laplace transforms of the signals acting on the U ( s) = u0 + kE( s) + E( s)
s
system
• Solve linear algebraic equations to obtain the transforms of Pure algebra gives relation between slope Θ and E by elimi-
the interesting signals nating V , U and V r
 
• Convert the Laplace transform to a time function s( s + 0.02) + ks + k i E( s) = 10Θ( s)

c K. J. Åström August, 2000 3

 More Cruise Control More Cruise Control ...
 
s( s + 0.02) + ks + ki E( s) = 10sΘ( s) 10θ 0
e ( t) = te−ω 0 t
ω0
With the chosen controller k = 2ζ ω 0 − 0.02 (ζ = 1) and k i = ω 02 The largest error e max = 10θ 0 e−1 occurs for t = 1/ω 0 . Compare
and a step change of magnitude θ 0 in slope we have graph below θ 0 = 0.04 ζ = 1, ω 0 = 0.05 (dotted), ω 0 = 0.1
Θ( s) = θ 0 /s (solid) and ω 0 = 0.2 (dashed)
3

and we find that the Laplace transform of the error is 2.5

e [m/s]
2
10s 10θ 0
E( s) = 2 Θ( s) = 1.5

s + 2ω 0 s + ω 0
2 ( s + ω 0 )2 1

0.5
Converting this to a time function we get 0
0 10 20 30 40 50 60 70 80 90 100
t [s]
10θ 0
e( t) = te−ω 0 t
ω0 What can we conclude?

Discussion 3. The Transfer Function

• What do we mean by a solution to a problem? The transfer function of an LTI system was introduced in
• A historical perspective Lecture 2. Using Laplace transforms it can also be defined as
follows. Consider an LTI system with input u and output y. The
Closed form expressions, tables, curves transfer function is
• The role of computers
Y ( s) Ly
G ( s) = =
• The necessity of insight and understanding U ( s) Lu
• The need to check results where the Laplace transforms are calculated under the
• What properties can we find easily using “back of an assumption that all initial values are zero.
envelope” calculation.
Transfer functions and Laplace transforms are ideal to deal
• A perspective on use of Laplace transforms in control with block diagram. A block is simply characterized by
engineering
Y ( s) = G ( s) U ( s)
• A more general (biased personal) perspective. Technology
changes fast engineering education changes slowly.
Signals and and systems have the same representations.
c K. J. Åström August, 2000 4
 Working with Block Diagrams Solution
d n
Consider the system r e u x y
Σ C Σ P Σ
d n
r e u x y −1
Σ C Σ P Σ
Introduce Laplace transforms and transfer functions. We have
−1  
E = R − N + P( D + CE)
How is the error e related to the signals r  and n?
Solving for E gives
1 1 P
E= R− N− D
The Audience is Thinking ... 1 + PC 1 + PC 1 + PC

• Superposition
• The form of the equations

Car Model in Cruise Control PID Controller

Process model The error e is the input and the control signal u is the output
dv  t
+ 0.02v = u − 10θ de
dt u = ke + ki e(τ )) dτ + kd
0 dt
Transfer function from control u to velocity v Transfer function
V ( s) 1 U ( s) ki
Gvu ( s) = = G ( s) = = k+ + kd s
U ( s) s + 0.02 E( s) s

Transfer function from slope θ to velocity v

Θ( s) 10
Gvθ ( s) = =−
U ( s) s + 0.02

c K. J. Åström August, 2000 5

 Car, Double Integrator Time Delay
A simple model of a car on a horizontal road tells how its Consider a system where the output y is the input u delayed T
position y depends on the throttle. Let the mass be m and time units. The input output relation is
assume that the propelling force is proportional to the throttle
we find y( t) = u( t − T )

d2 y and the transfer function is

m = F = ku
dt2
Y ( s)
G ( s) = = e− sT
The transfer function is U ( s)

Y ( s) k
G ( s) = =
U ( s) ms2

With suitable units we get G (s) = s −2 a double integrator.

Linear Time Invariant Systems 4. Qualitative Insight

Consider the system Time responses can in principle be computed. Tables of
Laplace transforms is a help but the work is quite tedious.
dn y d n− 1 y d n− 1 u Time responses are easy to compute using different types of
+ a1 + . . . + a n y = b1 + . . . + bn u
dtn dtn−1 dtn−1
software.
Assuming that y and u and all their derivatives are zero • It is a good rule to always make order of magnitude
initially. Taking Laplace transforms we get calculations to make sure that results are reasonable
( sn + a1 sn−1 + a2 sn−2 + . . . + an−1 s + an ) Y ( s) whenever you use software.
= ( b1 sn−1 + b2 sn−2 + . . . + bn−1 s + bn ) U ( s) • Much insight can be obtained form very simple calcula-
tions (series expansions and factorization).
The transfer function is • Some results will be presented.
n− 1 n− 2
b1 s + b2 s + . . . + bn−1 s + bn • It will be discussed more in future lectures
G ( s) =
sn + a1 sn−1 + a2 sn−2 + . . . + an−1 s + an

c K. J. Åström August, 2000 6

 Insight into behavior of Signals Example
Consider a signal specified by a rational Laplace transform Consider the signal
Y (s). Make a partial fraction expansion
B ( s) s+5
Y ( s) = =
B ( s) B ( s) C1 C2 Cn A( s) s( s + 1)( s + 2)
Y ( s) = = = + + ... +
A( s) ( s − a1 )( s − a2 ) . . . ( s − an ) s − a1 s − a2 s − an
B ( a k) The polynomial A(s) is already in factored form and its zeros
Ck = lim ( s − ak ) F ( s) = are 0, -1 and -2. This means that the time function has the
s→ a k ( ak − a1 ) . . . ( ak − ak−1)( s − ak+1 ) . . . ( ak − an )
components
Parameters a k (roots of A(s)) are easy to compute. The signal
y(t) has the form constant, e − t, and e−2t

y( t) = C1 ea1 t + C2 ea2 t + . . . + Cn ean t the second and third component decay exponentially and
after a short time only the constant components remains. The
Parameters a k give shape and Ck give magnitudes. amplitude of the constant component is 2.5. (lim s→0 sY (s)).

Example ... Example

For small s the Laplace transform is Y (s)  2.5/s, which Consider the signal
implies that for large t the time function is y(t)  2.5. For large
B ( s) s+5
s we have Y (s)  1/s 2 . This means that for small t the time Y ( s) = = 4
C( s) s + 5s3 + 7s2 + 5s + 6
function is approximately y(t)  t.
3 The matlab command
2.5

1.5
A=[1 5 7 5 6]
1
roots(A)
0.5

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
gives the roots of A(s) to be −2, −3, and ±i. This means that
1 the time function has the components
0.8

0.6
e−2t, e−3t, sin t, and cos t
0.4

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
the first components decay exponentially and after a short time
only the periodic components remain.
c K. J. Åström August, 2000 7
 Insight from Transfer Functions Insight from Transfer functions
B ( s)
• Derive transfer function G (s) = A(s)
• Make a series expansion of G (s) for small s
• Compute poles a k (roots of A(s) = 0) G ( s) = c0 + c1 s + . . . + ck sk + . . .
Free motion of system has component Ce ak t
If c0 = 0 like a static system gain c0 .
• Compute zeros b k (roots of B (s) = 0)
If c0 = c2 = . . . = ck−1 = 0 and ck = 0 like k
The system blocks transmission of the signal e bk t differentiators.
• Look at behavior for small s (large t, low frequencies) and • Make a series expansion of G (s) for large s (high fre-
large s (small t, high frequencies) quency behavior)
c− 1 c− 2 ck
G ( s) = + 2 + ... + k + ...
s s s

If c−1 = 0 like an integrator.

If c−1 = c−2 = . . . = c−k+1 = 0 and ck = 0 like k
integrators.

Examples Summary
E XAMPLE 1—PID CONTROLLER • Finding relations between signals in linear time invariant
The PID controller has the transfer function systems is very simple by using Laplace transforms
ki ki • Fits very well to the block diagram description
G ( s) = k + + kd s = + k + kd s
s s • It is natural to think in terms of transfer functions
This is already in series expansion form. For slow signals • The similarity of signals and systems
(small s) it behaves like an integrator and for fast signals (large • Qualitative reasoning (small s corresponds to large t and
s) it behaves like a differentiator. large s correspond to small t) is very useful to get a quick
insight into the behavior of signals and systems
E XAMPLE 2—PID CONTROLLER WITH D ERIVATIVE F ILTER
 1 sTd 
G ( s) = k 1 + +
sTi 1 + sTd/ N

Behaves like a static gain k(1 + N ) for fast signals (large s).

c K. J. Åström August, 2000 8