Você está na página 1de 34

ELEC2400 Signals & Systems

Chapter 3. Modelling, Differential Equations and System Properties


Brett Ninnes brett@newcastle.edu.au. School of Electrical Engineering and Computer Science The University of Newcastle

Slides by Juan I. Yuz (jiyuze@ee.newcastle.edu.au) - July 24, 2003 p.1/100

3. Modelling, Differential Equations and System Properties

In most engineering scenarios, signals are related by the laws of physics. The equations involved can be very complicated, although the essential characteristics can be captured by a simplied analysis. Physical models can be very often expressed as differential equations. This provide insight into the general principles to derive, describe and analyse these models.

Chapter 3. Modelling, Differential Equations and System Properties p.2/100

Electrical Circuits Modelling


Kirchoff Current and Voltage conservation laws: (KCL)
k

ik (t) = 0 vn (t) = 0
n

(KVL)

Fundamental passive components: Resistor (Ohms Law) Inductor Capacitor


vn (t) = R ik (t) vn (t) = L d ik (t) dt d ik (t) = C vn (t) dt

Chapter 3. Modelling, Differential Equations and System Properties p.3/100

We have that
i(t)

+
v (t)

vR (t)

by Kirchoffs Voltage Law: v (t) = vR (t) + vC (t)


+
vC (t)

by Ohms Law: vR (t) = R i(t) by the capacitive action: d i(t) = C vC (t). dt

Replacing in KVL, this yields:


v (t) = RC

d 1 1 d vC (t) + vC (t) vC (t) + vC (t) = v (t) dt dt RC RC

Chapter 3. Modelling, Differential Equations and System Properties p.4/100

Mechanical System Modelling


For a given a body, we have: position
u(t)
d dt u(t) d dt v (t) d2 dt2 u(t)

velocity v (t) = acceleration a(t) =

The sum of forces on a body of mass m, moving only in one axis, makes it accelerate: (Newtons 2nd Law)
k

Fk (t) = m a(t)

Chapter 3. Modelling, Differential Equations and System Properties p.5/100

Mechanical System Modelling


As well, many physical mechanical components obey laws expressible via differential equation. For a spring:
Fk (t) = ks xk (t)

where ks is the spring constant, and positive xk (t) corresponds to spring stretching. For a damper (or dash-pot):
Fk (t) = kd d xk (t) dt

where kd is the damper viscosity, and positive xk (t) corresponds to extension (of the damper).
Chapter 3. Modelling, Differential Equations and System Properties p.6/100

CAR CHASSIS

F1 (t) = mg

Spring

Damper d dt

F2 (t) = ks [y (t) u(t)]

F3 (t) = kd

[y (t) u(t)]

y (t)

Road

u(t)

Three forces:

(gravity) F1 (t) = mg (the spring) F2 (t) = ks [y (t) u(t)]

d (the damper) F3 (t) = kd d t [y (t) u(t)]


Chapter 3. Modelling, Differential Equations and System Properties p.7/100

Example 1: Car Shock Absorber By the Newton 2nd Law: F1 (t) + F2 (t) + F3 (t) = m a(t) mg ks [y (t) u(t)] kd Which (rearranging) yields: d2 kd d ks kd d ks y (t) + y (t) + y (t) = g + u(t) + u(t) 2 dt m dt m m dt m And, making y (t) = y (t) mg/ks : d2 kd d ks kd d ks y (t) + y (t) + y (t) = u(t) + u(t) 2 dt m dt m m dt m
Chapter 3. Modelling, Differential Equations and System Properties p.8/100

d d2 [y (t) u(t)] = m 2 y (t) dt dt

Example 2: Car Velocity


F1 (t) F2 (t)

x(t)

Consider a car subject to the engine force F1 (t), and the wind resistance F2 (t) proportional to velocity: d x(t) dt And the acceleration a(t) is the time rate of change of velocity v (t): d a(t) = v (t) dt F2 (t) = kv (t) , v (t)
Chapter 3. Modelling, Differential Equations and System Properties p.9/100

F1 (t)

F2 (t)

x(t)

Using Newtons 2nd law, we get a relationship between the motor induced force F1 (t) = f (t) and the car velocity v (t): d F1 (t) F2 (t) = f (t) kv (t) = m v (t) dt Which may be re-written as: k 1 d v (t) + v (t) = f (t) dt m m
Chapter 3. Modelling, Differential Equations and System Properties p.10/100

Example 3: Water Tank


u(t) Suface Area = A m2

h(t)

y (t)

Consider a circular water tank of surface area A [m2 ], subject to an inow u(t) [m3 /s] and a resulting outow y (t) [m3 /s]. The total volume V (t) of water in the tank at any time is: V (t) = Ah(t) Thus: d d V (t) = A h(t) = u(t) y (t) dt dt
Chapter 3. Modelling, Differential Equations and System Properties p.11/100

Example 3: Water Tank


u(t) Suface Area = A m2

h(t)

y (t)

The output ow rate y (t) is proportional to the pressure at the bottom of the tank, and in turn this pressure is proportional to the height h(t): y (t) = Kh(t) for some K R+ Thus, the nal differential equation model is: A d y (t) + y (t) = u(t) K dt
Chapter 3. Modelling, Differential Equations and System Properties p.12/100

u(t) Suface Area = A1 m2 Suface Area = A2 m2

h2 (t) x(t) h1 (t) y (t)

Consider now that the outow x(t) of one tank of surface area A1 becomes the inow of a second tank of surface area A2 . As per the previous example, we have that: d h1 (t) = u(t) x(t) dt d A2 h2 (t) = x(t) y (t) dt A1

Chapter 3. Modelling, Differential Equations and System Properties p.13/100

Example 4: Coupled Water Tanks


u(t) Suface Area = A1 m2 Suface Area = A2 m2

h2 (t) x(t) h1 (t) y (t)

The ow rate x(t) is proportional to the pressure difference, which is proportional to h1 (t) h2 (t). That is: x(t) = K1 [h1 (t) h2 (t)] for some K1 R+ (Note that x(t) can be negative) Finally, as in the previous example: y (t) = K2 h2 (t) for some K2 R+

Chapter 3. Modelling, Differential Equations and System Properties p.14/100

Example 4: Coupled Water Tanks


Substituting the expressions for x(t) and h2 (t), we have: A1 d 1 h1 (t) = u(t) K1 h1 (t) y (t) dt K2 A2 d 1 y (t) = K1 h1 (t) y (t) y (t) K2 dt K2

()

Which can be rewritten as: d K1 1 1 h1 (t) = h1 (t) + u(t) + y (t) dt A1 A1 K2 K1 + K2 A2 d h1 (t) = y (t) + y (t) K1 K2 dt K1 K2 Differentiating () and replacing h1 (t), we obtain the nal differential equation model relating input ow rate u(t) to output ow rate y (t): A1 A2 K1 K2 d2 y (t) + dt2 A1 + A 2 A1 + K2 K1 d y (t) + y (t) = u(t) dt

Chapter 3. Modelling, Differential Equations and System Properties p.15/100

For a wide class of physical phenomena, two signals y (t) and u(t) are related by a differential equation:
dn1 d dn y ( t ) + a y (t) + + a1 y (t) + a0 y (t) = n 1 n n 1 dt dt dt dm d bm m u(t) + + b1 u(t) + b0 u(t) dt dt

which is linear, time-invariant and nite dimensional. In the previous examples:


1 d dt vC (t) + RC vC (t) d2 s d d (t) + k (t) (t) + k m dt y my dt2 y

= =

1 RC v (t) kd d ks m dt u(t) + m u(t)

Chapter 3. Modelling, Differential Equations and System Properties p.16/100

First Order Differential Equations


Given a differential equation which describes the relationship between y (t) and u(t), can an explicit formula for y (t) be given? We consider a simple case, where n = 1 and m = 0:
d y (t) + a0 y (t) = b0 u(t). dt

It is called a rst order differential equation (only rst derivatives present). We are interested in obtaining a solution y (t) in terms of u(t) and the initial condition y (t0 ) = y0 .

Chapter 3. Modelling, Differential Equations and System Properties p.17/100

Solution of First Order Differential Equations

To solve the equation we use an integrating factor, ea0 t :


e a0 t d y (t) + ea0 t a0 y (t) = ea0 t b0 u(t) dt d a0 t e y (t) = ea0 t b0 u(t) dt

And, using the fact that:


t t0

d f ( ) d = f (t) f (t0 ) d
t

we can integrate both sides to give:


y (t) ea0 t = y (t0 ) ea0 t0 +

ea0 b0 u( ) d
t0
Chapter 3. Modelling, Differential Equations and System Properties p.18/100

In summary, the rst order differential equation:


d y (t) + a0 y (t) = b0 u(t) dt

has solution y (t) which is given by:


t

y (t) = y0 ea0 (t0 t) +


t0

ea0 (t) b0 u( ) d

; t t0

This solution, at time t = t0 passes through the point:


y (t0 ) = y0

Chapter 3. Modelling, Differential Equations and System Properties p.19/100

Example: First Order System with Step Input

Consider u(t) = 1(t) and y (0) = y0 , for the equation: d y (t) + a0 y (t) = b0 u(t). dt The solution is given by y (t) = y0 ea0 t +
b0 a0

b0 1 ea0 t . a0

y (t) = y0 ea0 t +

b0 [1 ea0 t ] a0

y0 0

Time (seconds)

5/a0

Chapter 3. Modelling, Differential Equations and System Properties p.20/100

State Space Descriptions


More generally, n, m > 1 in:
dn dn1 d y ( t ) + a y (t) + + a1 y (t) + a0 y (t) = n 1 dtn dtn1 dt dm d bm m u(t) + + b1 u(t) + b0 u(t) dt dt

State-space descriptions reduce high order scalar differential equations to vector rst order ones. The associated vector quantity is known as the state-vector, and its dimension is equal to n. The advantage of reduction to rst-order vector differential equation is ability to provide solutions.

Chapter 3. Modelling, Differential Equations and System Properties p.21/100

Consider a second order differential equation (n = 2):


d d d2 y (t) + a1 y (t) + a0 y (t) = b1 u(t) + b0 u(t) 2 dt dt dt

It may be re-written as:


d d y (t) + a1 y (t) b1 u(t) + a0 y (t) = b0 u(t) dt dt

Introduce the new signals:


x1 (t) x2 (t) y (t) d y (t) + a1 y (t) b1 u(t) dt
Chapter 3. Modelling, Differential Equations and System Properties p.22/100

State Space Descriptions: The Second Order Case

Replacing, we have:
d x1 (t) + a1 x1 (t) b1 = x2 (t) (denition of x2 (t)) dt d x2 (t) + a0 x1 (t) = b0 u(t) (original 2nd order eq.) dt

These two coupled rst-order differential equations may be compactly expressed in matrix-vector form as:
d x1 (t) dt x2 (t) = a1 1 a0 0 1 0 x1 (t) b1 + u(t) x2 (t) b0

y (t) =

x1 (t) x2 (t)
Chapter 3. Modelling, Differential Equations and System Properties p.23/100

State Space Descriptions: The Second Order Case

We then have a state-space description for the 2nd order differential equation:
d x(t) = Ax(t) + Bu(t) dt y (t) = Cx(t).

where:
x(t) x1 (t) , x2 (t) A a1 1 , a0 0 B b1 , b0 C 1 0

Chapter 3. Modelling, Differential Equations and System Properties p.24/100

For a car shock absorber, with y (t) = y (t) mg/ks , we obtained a 2nd order model of the form:
d2 ks k d ks k d (t) + y (t) = d u(t) + u(t) y (t) + d y dt2 m dt m m dt m

The corresponding state-space description is given by: kd kd 1 d x(t) = m x(t) + m u(t) ks dt ks 0


m m

y (t) = 1 0 x(t)

Chapter 3. Modelling, Differential Equations and System Properties p.25/100

State Space Descriptions: The General Case

Consider the general differential equation, with m n 1:


n

k =1

dk ak k y (t) + a0 y (t) = dt

n1

b
=1

d u(t) + b0 u(t) dt

It may be re-written as:


d dt
n

ak
k =1

dk1 y (t) dtk1

n1

b
=1

d 1 u(t) = a0 y (t) +b0 u(t). dt 1


x1 (t)

xn (t)

Which implies a rst-order relationship:


d xn (t) = a0 x1 (t) + b0 u(t). dt
Chapter 3. Modelling, Differential Equations and System Properties p.26/100

State Space Descriptions: The General Case

This re-parameterisation can be repeated:


n

xn (t)
k =1

dk1 ak k1 y (t) dt
n

n1

b
=1

d 1 u(t) dt 1 b
=2

d dt

ak
k =2

dk2 y (t) dtk2

n1

d 2 u(t) +a1 y (t) b1 u(t) dt 2


x1 (t)

xn1 (t)

so that:
d xn1 (t) = xn (t) a1 x1 (t) + b1 u(t) dt
Chapter 3. Modelling, Differential Equations and System Properties p.27/100

Repeating this process n 1 times yields n coupled rst order differential equations:
d xn (t) = a0 x1 (t) + b0 u(t) dt d xn1 (t) = xn (t) a1 x1 (t) + b1 u(t) dt d xn2 (t) = xn1 (t) a2 x1 (t) + b2 u(t) dt . . . . . . d x1 (t) = x2 (t) an1 x1 (t) + bn1 u(t) dt and where y (t) = x1 (t)
Chapter 3. Modelling, Differential Equations and System Properties p.28/100

State Space Descriptions: The General Case

Putting in matrix-vector form: bn1 x1 (t) an1 1 0 0 x1 (t) x2 (t) an2 0 1 0 x2 (t) bn2 d . . ... ... . + . . u(t) . . . . = . . . . . . dt 0 1 xn1 (t) b1 xn1 (t) a1 0 xn (t) a0 0 0 0 b0 xn (t) x1 (t) x2 (t) . . y (t) = 1 0 0 0 . xn1 (t) xn (t)

Chapter 3. Modelling, Differential Equations and System Properties p.29/100

State Space Descriptions: The General Case

In summary, the nth order differential equation:


dn1 d dn y ( t ) + a y (t) + + a1 y (t) + a0 y (t) = n 1 dtn dtn1 dt dn1 bn1 n1 u(t) + + b0 u(t). dt

may be re-expressed as a state-space representation, in terms of the n dimensional state vector x(t):
d x(t) = Ax(t) + Bu(t) dt y (t) = Cx(t)

Chapter 3. Modelling, Differential Equations and System Properties p.30/100

Where:
an1 an2 . . . a1 a0 0 0 1 0 ... ... . . . , 0 0 1 0 0 0 1 0 0 0 bn1 bn2 . . . , b1 b0

1 0

This formulation is known as the observer canonical form.

Chapter 3. Modelling, Differential Equations and System Properties p.31/100

State Transformations and Canonical Forms

The state space representation of a differential equation is not unique. Specically, for any invertible matrix M Rnn , the state vector x(t) Rn1 can be transformed to a new state vector z(t), such that:
z(t) = Mx(t) x(t) = M1 z(t)

In this case:

d d z(t) = Mx(t) = MAx(t) + MBu(t) dt dt = MAM1 z(t) + MBu(t) y (t) = Cx(t) = CM1 z(t)
Chapter 3. Modelling, Differential Equations and System Properties p.32/100

and

State Transformations and Canonical Forms

That is, the state space representation:


d z(t) = A z(t) + B u(t) dt y (t) = C z(t)

where:
A MAM1 , B MB, C CM1

describes exactly the same relationship between the signals y (t) and u(t)

Chapter 3. Modelling, Differential Equations and System Properties p.33/100

For example, controller canonical form:


A an1 an2 0 1 0 1 . . . . . . 0 0 a1 a0 0 0 0 0 . . ... . . . . 1 0 b 1 b0 1 0 . . . 0 0

bn1 bn2

There are also observability and controllability canonical forms, similar to the preceding.
Chapter 3. Modelling, Differential Equations and System Properties p.34/100

State Space Descriptions: The Feed-through Term

We have considered the general differential equation:


dn d2 d y ( t ) + + a y (t) + a1 y (t) + a0 y (t) = 2 n 2 dt dt dt dm bm m u(t) + + b0 u(t) dt

where we restricted m < n. We now consider the case m = n, and hence there is a dn term d tn u(t) on the right hand side:
n

ak
k =0

dk dn u(t) + y ( t ) = b n dtn dtk

n1

bk
k =0

dk u(t). dtk

Chapter 3. Modelling, Differential Equations and System Properties p.35/100

State Space Descriptions: The Feed-through Term

We can rewrite the equation, remembering an = 1:


n

ak
k=0 n k

dk dn y ( t ) b u(t) = n dtk dtn d ak bn k u(t) = dt k=0


n k

n1

bk
k=0 n1

dk u(t) dtk
n1

d ak k y (t) dt k=0
n k

dk bk k u(t) dt k=0 (bk ak bn )

ak b n
k=0

dk u(t) dtk

ak
k=0

d (y (t) bn u(t)) = dtk


z (t)

n1

k=0

dk u(t) dtk

bn

Considering z (t) as output variable, this can now be written in state-space form.
Chapter 3. Modelling, Differential Equations and System Properties p.36/100

Then we have:
d x(t) = Ax(t) + Bu(t) dt z (t) = Cx(t)

and using z (t) = y (t) bn u(t), it is equivalent to:


d x(t) = Ax(t) + B u(t) dt y (t) = Cx(t) + D u(t)

where D = bn , and it is called the feed-through term: it is the proportion of u(t) directly fed-through to y (t).
Chapter 3. Modelling, Differential Equations and System Properties p.37/100

State Space Descriptions: The Feed-through Term

In summary, the differential equation relationship:


dn1 dn dn y (t) + an1 n1 y (t) + + a0 y (t) = bn n u(t) + + b0 u(t) dtn dt dt

may be expressed as a state-space representation:


d x(t) = Ax(t) + Bu(t) dt y (t) = Cx(t) + Du(t)

In observer canonical form A, B, C, D are given as:


an1

1 0 . . . 0 0 0

0 1

.. .

bn1 bn an1

an2 . . .

0 . . .

bn2 bn an2 . . .

a1 a0

1 0

b 1 b n a1

b 0 b n a0

bn

Chapter 3. Modelling, Differential Equations and System Properties p.38/100

State Space Descriptions: The Feed-through Term

While, in controller canonical form the matrices A, B, C, D are given as:


an1

an2 0 1 . . . 0
T

.. .

a1 0 0 . . . 1

a0

0 0 . . .

0 . . . 0

0 . . .

bn1 bn an1

bn2 bn an2 . . .

bn

b 1 b n a1 b 0 b n a0

Chapter 3. Modelling, Differential Equations and System Properties p.39/100

Sometimes, in physical modelling, it may be natural to progress directly to a state space form: Kirchoffs laws:
iC (t) = C
vL (t)

+
i(t) C

+
vR (t)

d vC (t) = i(t) iL (t) dt d vL (t) = L iL (t) = vC (t) RiL (t) dt

+
vC (t) R

Which are expressible as:


d 1 1 vC (t) = iL (t) + i(t) dt C C 1 R d iL (t) = vC (t) iL (t) dt L L
Chapter 3. Modelling, Differential Equations and System Properties p.40/100

Direct State-Space Modelling

Writing these two differential equations in matrix-vector form gives: State-space model:
vL (t)

+
i(t) C

+
vR (t)

+
vC (t) R

vC (t) 1/C d vC (t) 0 = dt iL (t) iL (t) 1/L R/L


x(t) A x(t)

1/C 0
B

i(t)

Chapter 3. Modelling, Differential Equations and System Properties p.41/100

Direct State-Space Modelling

u(t) Suface Area = A1 m2 Suface Area = A2 m2

h2 (t) x(t) h1 (t) y (t)

Recall the coupled tank system, where:


A1 d h1 (t) = u(t) x(t) dt d A2 h2 (t) = x(t) y (t) dt

The ow between the tanks is x(t) = K1 [h1 (t) h2 (t)]. The output ow is y(t) = K2 h2 (t).
Chapter 3. Modelling, Differential Equations and System Properties p.42/100

u(t) Suface Area = A1 m2 Suface Area = A2 m2

h2 (t) x(t) h1 (t) y (t)

This yields the state-space model:


K1 /A1 h1 (t) 1/A1 d h1 (t) K1 /A1 + u(t) = dt h2 (t) h2 (t) K1 /A2 (K1 + K2 )/A2 0 h1 (t) y (t) = 0 K2 h2 (t)
Chapter 3. Modelling, Differential Equations and System Properties p.43/100

Differential equation solution via state-space

Recall that the 1st order scalar differential equation:


d y (t) = a0 y (t) + b0 u(t), dt y (t0 ) = y0

is solved, for a given signal u(t), by:


t

y (t) = ea0 (tt0 ) y0 +


t0

ea0 (t) b0 u( ) d.

What is solution the 1st order vector state-space differential equation:


d x(t) = Ax(t) + Bu(t), dt x(t0 ) = x0

Chapter 3. Modelling, Differential Equations and System Properties p.44/100

Differential equation solution via state-space

Using the previous methods, the solution of:


d x(t) = Ax(t) + Bu(t), dt
t

x(t0 ) = x0 eA(t) Bu( ) d

is:

x(t) = eA(tt0 ) x0 +
t0

where the Matrix-Exponential eAt is dened as:


exp(At) = eAt = I + At + A 2 t2 + ... = 2!

k =0

A k tk k!

which satises

d At e = AeAt = eAt A dt
Chapter 3. Modelling, Differential Equations and System Properties p.45/100

Matrix Exponential: Example 1 2 . Consider the matrix A = 3 4

Then according to the denition: 2 3 1 0 1 2 1 2 1 2 1 1 + + + + eA = 2 3 4 6 3 4 0 1 3 4 51.9690 74.7366 = 112.1048 164.0738

Note that this is quite different to the matrix: e1 e2 2.7183 7.3891 = e3 e4 20.0855 54.5982
Chapter 3. Modelling, Differential Equations and System Properties p.46/100

Differential equation solution via state-space

In summary, the solution y (t) of the equation:


dn dn1 dn y (t) + an1 n1 y (t) + . . . + a0 y (t) = bn n u(t) + . . . + b0 u(t) n dt dt dt

is given by:
y (t) = CeA(tt0 ) x0 +
t t0

CeA(t) Bu( ) d + Du(t)

where:

d x(t) = Ax(t) + Bu(t) dt y (t) = Cx(t) + Du(t)

The solution y (t) satises y (t0 ) = Cx0 + Du(t0 ).


Chapter 3. Modelling, Differential Equations and System Properties p.47/100

Differential Equation Solution via State-Space: Example

Consider the 2nd order equation (, R):


d d d2 y (t) + ( + ) y (t) + y (t) = b1 u(t) + b0 u(t). 2 dt dt dt

In state-space form:
d x(t) = Ax(t) + Bu(t) dt y (t) = Cx(t)

where:
A ( + ) 1 , 0 B b1 , b0 C 1 0

Chapter 3. Modelling, Differential Equations and System Properties p.48/100

The solution is:


t

y (t) = Cx(t) = CeA(tt0 ) x0 +


t0

g (t )u( ) d

where g (t)

CeAt B. To obtain eAt , we diagonalize A:

A = S1 S Ak = S1 SS1 S S1 S = S1 k S

At

=
k =0

A k tk = S1 k!

k =0

k tk k!

Chapter 3. Modelling, Differential Equations and System Properties p.49/100

Differential Equation Solution via State-Space: Example

Therefore:
g (t) = CeAt B et et 1 [1, 0] = ( ) et et = 1 et + 2 et b 1 t t b0 e e et et

And, if we consider u(t) = 1(t), x0 = 0, and t0 = 0:


t t

y (t) =
t0

g (t )1( ) d =
0

1 e(t) + 2 e (t) d

1 2 = (1 et ) + (1 et )
Chapter 3. Modelling, Differential Equations and System Properties p.50/100

Differential equation solution via state-space

In the example, the assumptions


a1 = + a0 =

for some , R, are not completely general. We didnt consider the cases = or , C. The difculty with which the matrix exponential eAt was obtained, is an indication that state-space formulations may not be the best method. This will motivate our study of Laplace-Transform methods (Chapter 4).

Chapter 3. Modelling, Differential Equations and System Properties p.51/100

An important point is that the solution:


t

y (t) = CeA(tt0 ) x0 +
t0

CeA(t) Bu( ) d + Du(t)

is unique only up to the choice of x(t0 ) = x0 . The uniqueness of the solution depends upon the initial conditions at time t = t0 for y (t), dy (t)/dt, . . . , dn1 y (t)/dtn1 and u(t), du(t)/dt, . . . , dn1 u(t)/dtn1 .

Chapter 3. Modelling, Differential Equations and System Properties p.52/100

Initial Conditions
Rewriting the solution y (t) as:
y (t) = CeA(tt0 ) x0 + CeAt
t t0

eA Bu( ) d + Du(t)

then, differentiating with respect to t, yields:


d y (t) = CAeA(tt0 ) x0 + dt CAeAt
t t0

eA Bu( ) d + CBu(t) + D

d u(t) dt

Therefore, at time t = t0 :
d d y (t0 ) = CAx0 + CBu(t0 ) + D u(t0 ) dt dt
Chapter 3. Modelling, Differential Equations and System Properties p.53/100

Initial Conditions
Repeating this process, differentiating up to the n 1st derivative and setting t = t0 , a pattern clearly emerges:
y (t0 ) C D 0 0 u(t0 )
d dt u(t0 ) d2 dt2 u(t0 )

CB CA D 0 2 = CA x0 + CAB CB D . . . .. . . . . . . . dn1 n1 n2 y ( t ) CA 0 CA B dtn1


d dt y (t0 ) d2 dt2 y (t0 ) y0 O

..

CB D

0 . . . . . .

. . .

dn1 dtn1 u(t0 ) u0

which is expressible as y0 = Ox0 + u0 .

Chapter 3. Modelling, Differential Equations and System Properties p.54/100

A natural question is: given the initial conditions for u(t) and its derivatives (in u0 ), how should x0 be chosen to satisfy the initial condition on the solution y (t) and its derivatives (in y0 ) ? The answer is simple:
, O C

x0 = O1 [y0 u0 ]

The matrix O is called the observability matrix.


Chapter 3. Modelling, Differential Equations and System Properties p.55/100

CA CA2 . . . CAn1

Initial Conditions: Example Consider 2nd order system:


d2 d d y (t) + 3 y (t) + 2y (t) = 3 u(t) + 2u(t). dt2 dt dt

The state space description (observer canonical form) is:


3 1 3 d x(t) + u(t) x(t) = dt 2 0 2 y (t) = 1 0 x(t)

Then where

y (t) = CeAt x0 +
0

g (t )u( ) d

g (t) = CeAt B = 4e2t et


Chapter 3. Modelling, Differential Equations and System Properties p.56/100

Initial Conditions: Example The observability matrix is given by:


O= 1 0 C CA = 1 0 3 1

Now, suppose y0 = [0.5, 1]T and u0 = [0, 0]T , then:


= y0 = 3 1 1 2.5 2e2t et et e2t 0.5 At = 1.5e2t 2et Ce x 0 = 1 0 2.5 2 e2t et 2et e2t x0 = O
1

0.5

0.5

so the complete solution is:


t

y (t) = 1.5e2t 2et +


0

4e2(t) e(t) u( ) d.
Chapter 3. Modelling, Differential Equations and System Properties p.57/100

For some state space realizations, the observability matrix O, could be non invertible. For example:
d2 d d y (t) + 3 y (t) + 2y (t) = u(t) + u(t) dt2 dt dt

has observer canonical form matrices:


A= 3 1 2 0 ,C = 1 0 O= C CA = 1 0

3 1

But, the controller canonical form matrices are:


A= 3 2 1 0 ,C = 1 1 O= C CA = 1 1

2 2

Chapter 3. Modelling, Differential Equations and System Properties p.58/100

Initial Conditions: Observability

This means that the state space model:


d x(t) = Ax(t), dt y (t) = Cx(t) x(t0 ) = x0

describes a richer class of systems than:


dn dn1 d y ( t ) + a y (t) + + a1 y (t) + a0 y (t) = 0 n 1 dtn dtn1 dt

which only describes one input-output relationship. If for a pair {A, C}, the observability matrix O is non-singular, then the realisation is called observable.

Chapter 3. Modelling, Differential Equations and System Properties p.59/100

Initial Conditions: Controllability

The concept of controllability involves the question of whether the state can be driven to an arbitrary value x(t) from an arbitrary initial point x0 . To address this issue, we can assume D = 0 and x0 = 0. The state trajectory then is:
t

x(t) =
t0

eA(t) Bu( ) d

where, the matrix exponential term is:


exp(At) = eAt = I + At + A 3 t3 A 2 t2 + + = 2! 3!
k=0

A k tk k!

Chapter 3. Modelling, Differential Equations and System Properties p.60/100

The CayleyHamilton theorem establishes that if A is n n, then for some constant 0 , , n1 :


An = 0 I + 1 A + 2 A2 + + n1 An1

As a consequence, every power of A can be expressed in terms of {I, A, . . . , An1 }. Applying the theorem above, we have that:
eAt = I + At + A 3 t3 A 2 t2 + + = 2! 3!
k=0

A k tk k!
n1

= 0 (t)I + 1 (t)A + 2 (t)A2 + + n1 (t)An1 =


k=0

k (t)Ak

Chapter 3. Modelling, Differential Equations and System Properties p.61/100

Initial Conditions: Controllability

Substituting into the state trajectory:


t n1 n1 t

x(t) =
t0 k=0

k (t )Ak Bu( ) d =
k=0

Ak B
t0

k (t )u( ) d

where zk (t)

z (t) 0 z1 (t) = C z(t) = B, AB, , An1 B . . . C zn1 (t)


t

k (t )u( ) d
t0

The matrix C is known as the controllability matrix: Any x(t) that are achievable by manipulation of u(t) must be a linear combination of the columns of C .
Chapter 3. Modelling, Differential Equations and System Properties p.62/100

Initial Conditions: Controllability

In summary, given an n-dimensional linear time invariant system with state space description:
d x(t) = Ax(t) + Bu(t) dt y (t) = Cx(t) + Du(t)

the n n controllability matrix C is dened as:


C B, AB, A2 B, , An1 B

The system is termed controllable if and only if C is an invertible matrix (full rank or det C = 0). Otherwise the system is termed uncontrollable.
Chapter 3. Modelling, Differential Equations and System Properties p.63/100

The solution of any differential equation:


dn1 dn dn y (t) + an1 n1 y (t) + + a0 y (t) = bn n u(t) + + b0 u(t). n dt dt dt

is expressible as:
y (t) = CeA(tt0 ) x0 Natural Response
t

+
t0

CeA(t) Bu( ) d + Du(t) . Forced Response

The Natural Response dependes on the coefcients {an1 , . . . , a0 , bn , . . . , b0 } and the initial conditions x0 . The Forced Response dependes on the coefcients {an1 , . . . , a0 , bn , . . . , b0 } and the input signal u(t).
Chapter 3. Modelling, Differential Equations and System Properties p.64/100

Forced and Natural Response: Example

Recall the shock absorber model,


d2 kd d ks kd d ks y (t) + y (t) + y (t) = u(t) + u(t) 2 dt m dt m m dt m

with m = 1000 kg, ks = 2000 N/m, kd = 3000 Ns/m, and initial conditions y (0) = 0.5 and dy (0)/dt = 1. The response of the shock absorber is:
t

y (t) =

1.5e2t 2et

+
0

4e2(t) e(t) u( ) d .

Natural Response

Forced Response
Chapter 3. Modelling, Differential Equations and System Properties p.65/100

Forced and Natural Response: Example

Suppose that u(t) = 1(t 1), then for t > 1:


t

4e2(t) e(t) 1( 1) d = 12e2(t1) +e(t1)


0 Forced Response 1 Total Response

0 2 0.5 4 Natural Response 6 8 t

Chapter 3. Modelling, Differential Equations and System Properties p.66/100

Suppose that: Initial time t0 , i.e., a long time ago, The initial conditions, given by the vector quantities y0 , u0 and, hence x0 , are all zero at t0 = . This yields only the forced response part:
t

y (t) =

g (t )u( ) d + Du(t)

where g (t)

CeAt B.

Clearly, given u(t), the response y (t) depends on the function g (t) and the feed-through term D.

Chapter 3. Modelling, Differential Equations and System Properties p.67/100

Impulse Response
In particular, if u(t) = (t) (Dirac delta), for t > 0:
t

y (t) =

g (t ) ( ) d + D (t) = g (t) + D (t)

Thus, the system impulse response is dened as:


h(t) g (t) + D (t) 0 ;t 0 ;t < 0

For t < 0, h(t) is zero because of the zero initial conditions. From physical modelling, commonly D = bn = 0, so h(t) = g (t).
Chapter 3. Modelling, Differential Equations and System Properties p.68/100

Impulse Response: Example 1

For an RC-circuit, we have seen that:


d 1 1 vC (t) = vC (t) + v (t) dt RC RC
x(t) A x(t) B

Therefore:
1 t/RC g (t) = CeAt B = e RC
1 RC

1 et/RC h(t) = RC 0

;t 0 ;t < 0

1 RC

et/RC

t Chapter 3. Modelling, Differential Equations and System Properties p.69/100

For a car shock absorver, we obtained:


kd d ks kd d ks d2 y (t) + y (t) + y (t) = u(t) + u(t) dt2 m dt m m dt m

If m = 1000 kg, ks = 1000 N/m, kd = 500 Ns/m, then:


h(t) = CeAt B = 1.0328e0.25t cos(0.9682t 1.0654), t>0

1.0328e0.25t cos(0.9682t 1.0654)

Chapter 3. Modelling, Differential Equations and System Properties p.70/100

Convolution
A key result we have is that for any input u(t), and zero initial conditions:
t

y (t) =
t

g (t )u( ) d + Du(t),

g (t)

CeAt B

(g (t ) + D (t )) u( ) d h(t )u( ) d

Noting: = [t, ) (t ) (, 0] h(t ) = 0. We say that y (t) is the convolution of h(t) with u(t). The new limits give a more general expression, often necessary (for example, for lter design).
Chapter 3. Modelling, Differential Equations and System Properties p.71/100

Convolution

y (t) =

h(t )u( ) d = [h

u] (t)

Intuitively, it means that y (t) is the sum of the effects of u(t), weighted by h(t). Graphically, h(t) characterizes the system:
u(t) y (t)

u(t)

Linear System h(t) y (t) =

h(t )u( ) d

Chapter 3. Modelling, Differential Equations and System Properties p.72/100

Commutative Property:

[h

u](t) =

h(t )u( ) d =

h(x)u(t x) dx h](t).

u(t x)h(x) dx = [u

Linearity:

[h

(u + w)] (t) = =

h(t ) (u( ) + w( )) d

h(t )u( ) d +

h(t )w( ) d

= [h

u](t) + [h

w](t).

Chapter 3. Modelling, Differential Equations and System Properties p.73/100

Properties of Convolution
Graphically, linearity means that linear combination of signals can be done before or after a convolution operation, leading to the same nal result.

Gain u(t)
+

Gain y (t) u(t) Linear System h(t)


+

y (t)

w(t)
+

Linear System h(t)

y (t)

w(t)
+

w(t)

Gain

w(t)

Linear System h(t)

Gain

[h

(u + w)](t) = [h

u](t) + [h

w](t)

Chapter 3. Modelling, Differential Equations and System Properties p.74/100

Properties of Convolution
Convolution and Dirac functions:

[h

](t) =

h(t ) ( ) d = h(t)

Convolution and Time Shifted Signals:

[h

g ](t) =

h(t )g ( ) d h(t )u( T ) d

and changing variable x = T :

[h

g ](t) =

h((t T ) x)u(x) dx = [h

u](t T )

Chapter 3. Modelling, Differential Equations and System Properties p.75/100

Associative property:

[h

(u

f )](t) =

h(t )[u h(t )


f ]( ) d

= =

u( x)f (x) dx d

h(t )u( x) d f (x) dx

changing variable = x:

[h

(u

f )](t) =

h(t x )u( ) d f (x) dx [h

= = [(h

u](t x)f (x) dx f ](t).


Chapter 3. Modelling, Differential Equations and System Properties p.76/100

u)

Properties of Convolution
Combining commutativity and associativity, we have, for example:
[h (u f )](t) = [h (f u)](t) = [(h f ) u](t) = [(f h) u](t)
u(t) y (t)

Same Input u(t)

u(t)

Linear System h(t)

Linear System f (t)

Same Output y (t)

Order Reversed

u(t)

Linear System f (t)

Linear System h(t)

Chapter 3. Modelling, Differential Equations and System Properties p.77/100

Convolution: Example

To apply the convolution:

y (t) =

h(t )u( ) d

in practice, we have to consider: the effect of changing h( ) to h(t ) involves the ipping and sliding of that signal, the integral is then the area obtained for different intervals, and conmutativity ensures that we can also ip and slide u(t), obtaining the same nal result.

Chapter 3. Modelling, Differential Equations and System Properties p.78/100

For an RC Circuit we already obtained:


1 et/RC h(t) = RC 0 ;t 0 ;t < 0
0
1 RC

h( ) =

1 /RC e RC

Assume that the voltage is:


v (t) = 1(t) 1(t 1)
1 v ( )

Note that,ipping and shifting, we obtain h(t ):

h(t ) 0

Chapter 3. Modelling, Differential Equations and System Properties p.79/100

Convolution Example: RC Circuit Note that:


y (t) = 0, t0
0 1 h(t ), t = 0 v ( )

y (t) =
0

1 (t)/RC e d RC
t/RC

h(t ), t = 0.5

v ( )

=1e

h(t )u( ) d

0<t1

0.5

y (t) =
0

1 (t)/RC e d RC

h(t ), t > 1

v ( )

= et/RC e1/RC 1

h(t )u( ) d

t>1

Chapter 3. Modelling, Differential Equations and System Properties p.80/100

Convolution Example: RC Circuit

Combining the 3 previous results: 0 vC (t) = 1 et/RC 1/RC e 1 et/RC


1 et/RC 1 e1/RC vC (t) =

;t 0 ; t (0, 1] ;t > 1

h( )v (t ) d

t [e1/RC 1]et/RC
Chapter 3. Modelling, Differential Equations and System Properties p.81/100

The previous example can be repeated, but using the commutative property:

h(t )v ( ) d =

v (t )h( ) d

In this case, the impulse response remains static for all t as:
1 RC

h( ) =

1 /RC e RC

However, now v is ipped and shifted.

Chapter 3. Modelling, Differential Equations and System Properties p.82/100

Convolution Example: RC Circuit (again) Then we have:


y (t) = 0, t0
1
t

v (t ), t = 0 h( )

y (t) =
0

1 /RC e d RC

v (t ), t = 0.5 h( )

= 1 et/RC 0<t1

t 1 = 0.5 0

y (t) =
t1

1 /RC e d RC

h( )

v (t ), t > 1

= et/RC e1/RC 1

t>1

0 t1

Chapter 3. Modelling, Differential Equations and System Properties p.83/100

Convolution Example: RC Circuit (again)

Combining the 3 results, we obtain the same previous solution:


0 vC (t) = 1 et/RC e1/RC 1 et/RC 1 et/RC

;t 0

; t (0, 1]

;t > 1

1e

1/RC

vC (t) =

h( )v (t ) d

1 [e
1/RC

t 1]e
t/RC

This veries that it does not matter which of the two function in a convolution is ipped and shifted.
Chapter 3. Modelling, Differential Equations and System Properties p.84/100

As a nal more abstract example of convolution, suppose that u( ) and h( ) are rectangular signals:
1 u( )

0 1

1 h( )

0.5

In this case, if we elect to ip and slide h as h(t ):


h(t ) 0

t 0.5 t

Chapter 3. Modelling, Differential Equations and System Properties p.85/100

Convolution Example: Rectangular Signals Then we have:


y (t) = 0, t 1.5
0
t0.5

u( )

h(t ), t = 1.5

1.5 h(t ), t = 1.25

y ( t) = =

d
1 =t0.5 =1

u( )

h(t )u( ) d

= 1.5 t

1 t < 1.5
t

0.75

1.25

y (t) = =

t0.5 =t0.5 =t

h(t ), t = 0.75

u( )

h(t )u( ) d

= 0.5

0.5 t < 1

0 0.25 0.75 1

Chapter 3. Modelling, Differential Equations and System Properties p.86/100

Convolution Example: Rectangular Signals

Using similar arguments, if 0 t < 0.5:


t

y (t) =
0

d =

=t =0

=t

And nally, if t 0, the product h(t )u( ) is zero for all and hence:
y (t) = 0 ; t < 0.

Combining all the results, we have the full solution:


0 1.5 t 0.5

; t 1.5 ; t [1, 1.5) ; t [0.5, 1) ; t [0, 0.5) ;t < 0 0 0.5 0.5

y (t) =

h( )u(t ) d

y (t) =

1.5

Chapter 3. Modelling, Differential Equations and System Properties p.87/100

Via electrical, mechanical and differential equation modelling examples we have gradually introduced the idea of a system. In a more abstract and general sense we have: A system S is a governing inuence which regulates the relative behaviour of signals. The notation used to denote this regulation is that the nature of a signal y (t) relative to another signal u(t) is given as
y (t) = S [u](t).

Chapter 3. Modelling, Differential Equations and System Properties p.88/100

System Properties
Linearity/Non-linearity: A system S is linear if, for any signals u(t), x(t) and any constant numbers , :
S [u + x](t) = S [u](t) + S [x](t)

The systems expressible as convolution with an impulse response, S [u](t) = [h u](t), are linear:
S [u + x](t) = [h (u + x)](t) = [h u](t) + [h x](t) = S [u](t) + S [x](t)

The system S [u](t) = u2 (t) is non-linear since:


S [u+x](t) = (u(t) + x(t)) = 2 u2 (t)+ 2 x2 (t)+2u(t)x(t) = u2 (t) + x2 (t) = S [u](t) + S [x](t)
Chapter 3. Modelling, Differential Equations and System Properties p.89/100

System Properties
Causality: A system is causal or non-anticipative if y(t ) = S [u](t ) does not depend on future values of u(t) (i.e. t > t ). All physical systems are causal (except in certain quantum phenomena). The linear system:
y (t) = [h u](t) is causal h(t) = 0 for t < 0.

To see this, in

y (t ) = [h

u](t ) =

h( )u(t ) d.

note that if h( ) = 0 for < 0, then y (t) depends on future values of u(t), because t > t .
Chapter 3. Modelling, Differential Equations and System Properties p.90/100

Memory: A causal system S is memory-less if y(t ) = S [u](t ) depends only on u(t) at t = t (otherwise it has memory, since y(t ) is affected by prior values of u). The simple amplier system y(t) = S [u](t) = Ku(t) for some K R is a memory-less system. The linear system
y (t) = [h u](t) is memory-less h(t) = K (t), K R.

To see this, in

y (t ) = [h

u](t ) =

h( )u(t ) d

note that, if h( ) = 0 h( )u(t ).

for > 0 then y(t ) depends on


Chapter 3. Modelling, Differential Equations and System Properties p.91/100

System Properties
Time Invariance: A system S is time-invariant if, a time time shifting in the input generates a time shifting in the output:
x(t) = u(t T ) z (t) = S [x](t) = S [u](t T ) = y (t T )

The simple amplier system:


y (t) = S [u](t) = Ku(t)

for some K R is time invariant. But, if the gain is time-varying:


y (t) = S [u](t) = | sin t|u(t)

would be time-varying.
Chapter 3. Modelling, Differential Equations and System Properties p.92/100

System Properties
System Dimension: A system described by an n-th order differential equations can be represented as:
d x(t) = Ax(t) + Bu(t) dt y (t) = Cx(t) + Du(t)

where x(t) is a state vector of dimension n. In this case, the system is said to have dimension n. When n < , the system is nite dimensional. For the non-linear case:
d x(t) = f (x(t), u(t)) dt y (t) = g (x(t), u(t))

Again, the dimension is that of the vector x(t).


Chapter 3. Modelling, Differential Equations and System Properties p.93/100

Stability: Intuitively, a system is stable if small perturbations to it are only capable of producing small responses.
t

y (t) =

CeA(tt0 ) x0 Natural Response

+
t0

g (t )u( ) d + Du(t) ; g (t) Forced Response

CeAt B.

Asymptotic Stability: It concerns on the natural response only. If we diagonalize the matrix A = S1 S, then:
y (t) = CeA(tt0 ) x0 = CS1 e(tt0 ) Sx0

where contains the eigenvalues of A.


Chapter 3. Modelling, Differential Equations and System Properties p.94/100

System Properties
Asymptotic Stability. If all the eigenvalues 1 , . . . , n have negative real part:
= e1 (tt0 ) 0 . . . 0 0 e2 (tt0 ) .. . 0 0 0 . . . en (tt0 ) 0 t

(tt0 )

Hence y (t) 0, for any x0 . Then we say that a linear, nite dimensional and time invariant system:
d x(t) = Ax(t) + Bu(t) dt y (t) = Cx(t) + Du(t)

is asymptotically stable Re{k } < 0 ; k = 1, . . . , n.


Chapter 3. Modelling, Differential Equations and System Properties p.95/100

System Properties
Bounded-Input Bounded-Output (BIBO) Stability: This involves the idea that the system response y (t) is bounded, for any bounded input u(t), regardless of the initial conditions.

|u(t)| M <

|y (t)| = M

h(t )u( ) d

|h(t )|d

And

|h( )|d =
0

CeAt B + D (t) dt CS1 et SB dt + D


0 n

=
k=1

|k |
0

ek t dt + D

Chapter 3. Modelling, Differential Equations and System Properties p.96/100

Bounded-Input Bounded-Output (BIBO) Stability: Then, we have that a linear, nite dimensional and time invariant system:
d x(t) = Ax(t) + Bu(t) dt y (t) = Cx(t) + Du(t)

is bounded-input bounded-output stable if all the eigenvalues 1 , , n of A, satisfy: Re{k } < 0;


k = 1, 2, , n.

However, the converse is not true.

Chapter 3. Modelling, Differential Equations and System Properties p.97/100

System Properties
Stability: Example
Again, we consider the general 2nd order linear time invariant system: d d d2 y (t) + a1 y (t) + a0 y (t) = b1 u(t) + b0 u(t) dt2 dt dt In state-space representation: 1 + 2 1 , B A 1 2 0 The eigenvalues of A are 1,2 = 1 2

b1 b0

1 0

a1

a2 1 4a0

Thus, the nature of the response y (t) is completely captured by the terms e1 t and e2 t .
Chapter 3. Modelling, Differential Equations and System Properties p.98/100

System Properties
Stability: Example
With x0 = 0, the response to u(t) = 1(t) for t > 0 is: 1 b1 + b0 b0 + 2 b1 y (t) = (e1 t 1) + (1 e2 t ) 1 1 2 2 1 2
2 T If u(t) = 0 and the initial condition is x0 = [x1 0 , x0 ] , then:

y (t) =

1 1 x2 0 + x0 2 1

e 1 t +

2 x1 0 + 2 x0 1 2

e 2 t ,

t>0

If 1,2 = j , then e1,2 t = et [cos t j sin t]. If = Re{1,2 } < 0, then e1,2 t will tend to zero as t grows. If > 0, then e1,2 t will explode towards as t grows. Therefore, the system is bounded-input bounded-output and asymptotically stable if Re{1,2 } < 0.

Chapter 3. Modelling, Differential Equations and System Properties p.99/100

Note that a system can be BIBO stable even if an eigenvalue violates the condition Re{k } < 0.
Example. Consider the case of: d2 d d y (t) + y (t) 2 y (t) = b1 u(t) + b0 u(t) dt2 dt dt
1 1 The eigenvalues of the matrix A = are 1 , 2 = 2, 1, and 2 0 hence Re{2 } = 1 > 0. However, the response to u(t) = 1(t) (with x0 = 0) is: 1 b1 + b0 b0 + 2 b1 (e1 t 1)+ (1 e2 t ) ; t > 0 y (t) = 1 1 2 2 1 2

But if we choose b1 = 1 and b0 = 1, then 1 + 1 2 + (1) (e2t 1)+ y (t) = 2 2 1 1 2 1 and thus the system is BIBO stable.

(1et ) =

(1e2t )

Chapter 3. Modelling, Differential Equations and System Properties p.100/100

Você também pode gostar