1.

Impulse Response Models for LTI Systems

Denition of impulse response model g(n):

System

For an arbitrary input {u(n)}, output {y(n)} of LTI system with zero initial state is given by:

y(n) =
k=

u(k)g(n k)

All possible combinations such that the sum of arguments is equal to n

Kannan M. Moudgalya, Autumn 2007

2.

Importance of Impulse Response Models

Impulse response has all information about LTI system Given impulse response, can determine output due to any arbitrary input

Digital Control

Kannan M. Moudgalya, Autumn 2007

3.

Step Response, Relation with Impulse Response

The unit step response of an LTI system at zero initial state {s(n)} is the output when {u(n)} = {1(n)}:

s(n) =
k=

1(k)g(n k)

Apply the meaning of 1(k):

=
k=0

g(n k)

This shows that the step response is the sum of impulse response.
Digital Control

Kannan M. Moudgalya, Autumn 2007

4.

Relation between Step and Impulse Responses

We can also get impulse response from step response. (n) = 1(n) 1(n 1) Using linearity and time invariance properties, g(n) = s(n) s(n 1)

Digital Control

Kannan M. Moudgalya, Autumn 2007

5.

Causality of LTI Systems

If output depends only on past inputs, called causal If output depends on future inputs, not causal For LTI causal systems, g(n) = 0 for n < 0 Initial state is zero No input until n = 0 - impulse input So, impulse response can begin only from n = 0 Signal {u(n)} is causal if u(k) = 0 for k < 0

Digital Control

Kannan M. Moudgalya, Autumn 2007

6.

Output of Causal Systems to Causal Signals

Impulse response g(n), input u(n) are causal:

y(n) =
k=

u(k)g(n k) u(k)g(n k) (u(k) = 0 k < 0)

= =
k=0

u(k)g(n k) (g(k) = 0 k < 0)

Digital Control

Kannan M. Moudgalya, Autumn 2007

7.

Convolution Theorems

Convolution is Commutative: u(n) g(n) = g(n) u(n) Convolution is Associative:

Convolution Distributes over Addition: u(n) (g1(n) + g2(n)) = u(n) g1(n) + u(n)
u(n) g1 (n)
+

y2 (n)

u(n)

g1 (n) + g2 (n)

y1(n)
7

g2 (n)
Kannan M. Moudgalya, Autumn 2007

Digital Control

8.

External (BIBO) Stability of LTI Systems

If every Bounded Input produces Bounded Output, system is externally stable equivalently, system is BIBO stable

|g(n)| < BIBO Stability

Dont care about what unbounded input does...

Digital Control

Kannan M. Moudgalya, Autumn 2007

9.

Recall convolution example

{y(n)} =
k=

u(k){g(n k)} {u(n)} = {4, 5, 6}

{g(n)} = {1, 2, 3},

g, u start at n = 0. They are zero for n < 0. y(0) = u(0)g(0) = 4 y(1) = u(0)g(1) + u(1)g(0) = 13 y(2) = u(0)g(2) + u(1)g(1) + u(2)g(0) = 28 y(3) = u(1)g(2) + u(2)g(1) = 27 y(4) = u(2)g(2) = 18 All other terms that dont appear above are zero.
Digital Control

Kannan M. Moudgalya, Autumn 2007

10.

Polynomial Calculation Convolution

Also carryout multiplication:

u(0)g(0)+ (u(0)g(1) + u(1)g(0))z 1+ (u(0)g(2) + u(1)g(1) + u(2)g(0))z 2+ (u(1)g(2) + u(2)g(1))z 3+

z a position marker - coe. of z i at ith instant u(0)+u(1)z 1 +u(2)z 2 - a way of representing a sequence with three terms: {u(0), u(1), u(2)} Even if large no. of terms, can get compact form
Digital Control

10

Kannan M. Moudgalya, Autumn 2007

11.

Denition of Z-transform

Z-transform of a sequence {u(n)}, denoted by U (z), is dened as:

U (z) =
n=

u(n)z n

where z is such that there is absolute convergence. That is, z should be chosen so as to satisfy
n=

|u(n)z n| <

Digital Control

11

Kannan M. Moudgalya, Autumn 2007

12.

Important properties of transfer functions

Given a causal, BIBO stable system with impulse response g(n) Z-transform of g(n), namely G(z), will have poles inside unit circle g(n) is a causal sequence G(z) = D(z) with N (z) is a polynomial of degree n D(Z) is a polynomial of degree m n m
Digital Control

N (z)

12

Kannan M. Moudgalya, Autumn 2007

13.

Z-transform Theorems - Linearity

Given the following Z-transform pairs, u1(n) U1(z), u2(n) U2(z), the following relation, with arbitrary , , holds:
Z [{u1(n)} + {u2(n)}] = U1(z) + U2(z)

Digital Control

13

Kannan M. Moudgalya, Autumn 2007

14.

Example 1 - Linearity

Find the Z-transform of u1(n) = (n) 3(n 2):

U1(z) =
n=

(n)z n 3

n=

(n 2)z n

= 1 3z 2 z 1 nite

Digital Control

14

Kannan M. Moudgalya, Autumn 2007

15.

Z-transform - Shifting

Z [u(n + d)] =

u(n + d)z n

n= u(n + d)z (n+d) = zd n= = z dU (z)

Digital Control

15

Kannan M. Moudgalya, Autumn 2007

16.

Z-transform - Shifting

Example: If {u(n)} U (z), then {u(n + 3)} z 3U (z) {u(n 2)} z 2U (z)

Digital Control

16

Kannan M. Moudgalya, Autumn 2007

17.

Initial value, theorem for causal signals z

lim U (z) = u(0)

Digital Control

17

Kannan M. Moudgalya, Autumn 2007

18.

Under the conditions

Final value theorem for causal signals

U (z) converges for all |z| > 1, if all poles of U (z)(1 z 1) are inside unit circle,
k

lim u(k) = lim (1 z 1)U (z)

Digital Control

18

Kannan M. Moudgalya, Autumn 2007

19.

Examples for Final Value Theorem

Using the nal value theorem, nd the steady state value of (0.5n 0.5)1(n) and verify. z 0.5z n 0.5)1(n) (0.5 |z| > 1 z 0.5 z1 lim LHS = 0.5
n z1

lim (z 1)RHS = lim = 0.5

0.5z

z1 z 1

(z 1)

Digital Control

19

Kannan M. Moudgalya, Autumn 2007

20.

Examples for Final Value Theorem

Is it possible to use the nal value theorem on 2n1(n)? z n1(n) 2 |z| > 2 z2 Since RHS is valid only for |z| > 2, the theorem cannot even be applied. In the LHS also, there is a pole outside the unit circle thereby violating the conditions of the theorem.

Digital Control

20

Kannan M. Moudgalya, Autumn 2007

21.

Z-transform of Convolution

If u(n) U (z) g(n) G(z) then, g(n) u(n) G(z)U (z) Recall the motivation slide for Z-transform.

Digital Control

21

Kannan M. Moudgalya, Autumn 2007

22.

Z-transform of Discrete State Space Systems

x(n + 1) = Ax(n) + Bu(n) x(0) = x0 Invalid for n = 1: x(0) = Ax(1) + Bu(1) = 0 = x0 The fact that it is not valid for n < 0 is not explicitly stated. But if we write it as x(n + 1) = Ax(n) + Bu(n) + (n + 1)x0 and assume initial rest, all variables are zero until n = 0, problem is solved.
Digital Control

22

Kannan M. Moudgalya, Autumn 2007

23.

Z-transform of Discrete State Space Systems

Z-transform of x(n + 1) = Ax(n) + Bu(n) + (n + 1)x0 gives zX(z) = AX(z) + BU (z) + x0z (zI A)X(z) = BU (z) + x0z X(z) = (zI A)1BU (z) + z(zI A)1x0 Z-transform of y(n) = Cx(n) + Du(n) is Y (z) = CX(z) + DU (z) = C(zI A)1BU (z) + DU (z) + C(zI A)1zx(0) = [C(zI A)1B + D]U (z) + C(zI A)1zx(0)

Digital Control

23

Kannan M. Moudgalya, Autumn 2007

24.

Finding Transfer Function - an Example B= 0.02 0.001987

Find the transfer function of 1 0 , A= 0.19801 0.9802 C = 0 1 , D = 0, G = c(zI A)1B

z1 0 0.02 = 0 1 0.19801 z 0.9802 0.001987 0 1 z 0.9802 0 0.02 = (z 1)(z 0.9802) 0.19801 z 1 0.001987

Digital Control

24

Kannan M. Moudgalya, Autumn 2007

25.

Finding Transfer Function - an Example 0 1 0.02 0.001987

G=

z 0.9802 0 (z 1)(z 0.9802) 0.19801 z 1 0.19801 z 1 0.02 = (z 1)(z 0.9802) 0.001987 0.001987z + 0.0019732 = (z 1)(z 0.9802) z + 0.9931 = 0.001987 (z 1)(z 0.9802)

Digital Control

25

Kannan M. Moudgalya, Autumn 2007

26.
1

Finding Transfer Function - an Example

/ / / /

F = [ 0 0 ; 1 0.1]; G = [ 0 . 1 ; 0 ] ; C = [0 1 ] ; dt = 0 . 2 ; sys = syslin ( c ,F ,G,C ) ; s y s d = dscr ( s y s , d t ) ; H = ss2tf ( sysd ) ;

Digital Control

26

Kannan M. Moudgalya, Autumn 2007