4 Kundur DT Analysis

Discrete-Time LTI Systems
Discrete-time Systems
Input-Output Description of Dst-Time Systems

input/
excitation
Discrete-Time LTI Systems and Analysis
x(n)
Discrete-time
System
y(n)
output/
response
Discrete-time
signal
Discrete-time
signal
Dr. Deepa Kundur

University of Toronto
Input-output description (exact structure of system is unknown

or ignored):
y (n) = T [x(n)]
black box representation:

T
x(n) y (n)
Dr. Deepa Kundur (University of Toronto)
1 / 61
System Properties
2 / 61
Terminology: Implication
If A then B
Shorthand: A = B
Why is this so important?

I
mathematical techniques developed to analyze systems are often

contingent upon the general characteristics of the systems being
considered
for a system to possess a given property, the property must hold
for every possible input to the system
I
I
Example 1:
it is snowing = it is at or below freezing temperature
Example 2:
5.2 = is positive
Note: For both examples above, B 6= A
to disprove a property, need a single counter-example

to prove a property, need to prove for the general case
3 / 61
4 / 61
Terminology: Equivalence
Common Properties
If A then B
Shorthand: A = B
and
If B then A
Time-invariant system: input-output characteristics do not

change with time
I
a system is time-invariant iff
Shorthand: B = A
x(n) y (n) = x(n n0 ) y (n n0 )
can be rewritten as
for every input x(n) and every time shift n0 .
Shorthand: A B
A if and only if B
We can also say:
I
A is EQUIVALENT to B
A=B
5 / 61
6 / 61
Additivity:
Common Properties
I
Linear system: obeys superposition principle

I
a system is linear iff

T [a1 x1 (n) + a2 x2 (n)] = a1 T [x1 (n)] + a2 T [x2 (n)]
for any arbitrary input sequences x1 (n) and x2 (n), and any
arbitrary constants a1 and a2 .
Homogeneity:
7 / 61
8 / 61
Common Properties
I
The Convolution Sum
The Convolution Sum
Causal system: output of system at any time n depends only on

present and past inputs
I
a system is causal iff
Recall:
y (n) = F [x(n), x(n 1), x(n 2), . . .]
x(n) =
for all n.
x(k)(n k)
k=
I
Bounded Input-Bounded output (BIBO) Stable: every bounded

input produces a bounded output
I
a system is BIBO stable iff

|x(n)| Mx < = |y (n)| My <
for all n and for all possible bounded inputs.
9 / 61
The Convolution Sum
The Convolution Sum
10 / 61
The Convolution Sum
The Convolution Sum
Let the response of a linear time-invariant (LTI) system denoted T to

the unit sample input (n) be h(n).
Therefore,
(n) h(n)
T
(n k) h(n k)
y (n) =
(n k) h(n k)
x(k)h(n k) = x(n) h(n)
k=
x(k) (n k) x(k) h(n k)
X
X
T
x(k)(n k)
x(k)h(n k)
k=
for any LTI system.
k=
T
x(n) y (n)
11 / 61
12 / 61
The Convolution Sum
Causality and Convolution
Stability and Convolution
For a causal system, y (n) only depends on present and past inputs
values. Therefore, for a causal system, we have:
y (n) =
It can also be shown that
h(k)x(n k)
k=
1
X
h(k)x(n k) +
k=
|h(n)| < LTI system is BIBO stable
n=
h(k)x(n k)
Note:
I means that the two statements are equivalent
I BIBO = bounded-input bounded-output
k=0
h(k)x(n k)
The Convolution Sum
k=0
where h(n) = 0 for n < 0 to ensure causality.
13 / 61
The Convolution Sum
PROOF
For a stable system, y (n) is bounded if x(n) is bounded. What are the implications on h(n)?
We have:
X
|y (n)| = |
h(k)x(n k)|
|h(k)x(n k)| =
k=
P
To prove the reverse implication (i.e., necessity), assuming
n= |h(n)| = we must find a
bounded input x(n) that will always result in an unbounded y (n). Recall,
|h(k)|Mx = Mx
Therefore,
k=
y (0)
Consider x(n) = sgn(h(n)); note: |x(n)| 1.

y (0)
h(k)x(k)
k=
|h(k)| <
h(k)sgn(h((k))) =
k=
and we can write:
=
|h(n)| <
h(k)x(k)
k=
|h(k)|
k=
h(k)x(0 k) =
k=
k=
h(k)x(n k)
k=
|h(k)| < is a sufficient condition to guarantee:
y (n) Mx
|h(k)| |x(n k)|

| {z }
k=
|x(n)|Mx <
k=
y (n)
The Convolution Sum
PROOF
k=
14 / 61
LTI system is stable
h(k)sgn(h(k))
k=
|h(n)| =
n=
n=
15 / 61
16 / 61
The Convolution Sum
The z-Transform and System Function
The Direct z-Transform
PROOF
Therefore,
|h(n)| =
Direct z-Transform:
n=
guarantees that there exists a bounded input that will result in an unbounded output, so it is
also a necessary condition and we can write:
X (z) =
x(n)z n
n=
|h(n)| <
n=
Notation:
Putting sufficiency and necessity together we obtain:
|h(n)| <
X (z)
Z{x(n)}
x(n) X (z)
n=
Note: means that the two statements are equivalent.
17 / 61
Region of Convergence
18 / 61
z-Transform Properties
Property
Notation:
the region of convergence (ROC) of X (z) is the set of all values

of z for which X (z) attains a finite value
The z-Transform is, therefore, uniquely characterized by:
1. expression for X (z)
2. ROC of X (z)
Linearity:
Time shifting:
Time Domain
x(n)
x1 (n)
x2 (n)
a1 x1 (n) + a2 x2 (n)
x(n k)
z-Domain
X (z)
X1 (z)
X1 (z)
a1 X1 (z) + a2 X2 (z)
z k X (z)
z-Scaling:
Time reversal
Conjugation:
z-Differentiation:
Convolution:
an x(n)
x(n)
x (n)
n x(n)
x1 (n) x2 (n)
X (a1 z)
X (z 1 )
X (z )
dX (z)
z dz
X1 (z)X2 (z)
ROC
ROC: r2 < |z| < r1
ROC1
ROC2
At least ROC1 ROC2
ROC, except
z = 0 (if k > 0)
and z = (if k < 0)
|a|r2 < |z| < |a|r1
1
< |z| < r1
r1
2
ROC
r2 < |z| < r1
At least ROC1 ROC2
among others . . .
19 / 61
20 / 61
Common Transform Pairs

1
2
3
4
5
6
Signal, x(n)
(n)
u(n)
an u(n)
nan u(n)
an u(n 1)
nan u(n 1)
(cos(0 n))u(n)
(sin(0 n))u(n)
(an
10
(an sin(0 n)u(n)
cos(0 n)u(n)
The System Function
z-Transform, X (z)
1
1
1z 1
1
1az 1
az 1
(1az 1 )2
1
1az 1
az 1
(1az 1 )2
1z 1 cos 0
12z 1 cos 0 +z 2
z 1 sin 0
12z 1 cos 0 +z 2
1az 1 cos 0
12az1 cos 0 +a2 z 2
1az 1 sin 0
12az1 cos 0 +a2 z 2
ROC
All z
|z| > 1
|z| > |a|
|z| > |a|
|z| < |a|
|z| < |a|
h(n) H(z)
Z
time-domain z-domain
Z
impulse response system function
Z
y (n) = x(n) h(n) Y (z) = X (z) H(z)
|z| > 1
|z| > 1
Therefore,
|z| > |a|
H(z) =
|z| > |a|
Discrete-Time Fourier Analysis
21 / 61
DTFT
Discrete-Time Fourier Transform (DTFT)
Y (z)
X (z)
22 / 61
DTFT
Periodicity of the DTFT

Consider
DTFT pair:
X ( + 2) =
1
x(n) =
X ()e jn d
2 2
X
X () =
x(n)e jn
n=
I
X () is the decomposition of x(n) into its frequency

components.
X
n=
X
n=
X
n=
x(n)e j(+2)n
x(n)e jn e j2n
x(n)e
jn
1=
x(n)e jn = X ()
n=
Therefore, X () is periodic with a period of 2.
23 / 61
24 / 61
DTFT

I
I
I

I
Since X () = X ( + 2), when dealing with discrete

frequencies, only a continuous frequency range of length 2
(representing one period) needs to be considered.
I
DTFT
Continuous-Time Sinusoids: Frequency and Rate of Oscillation:

x(t) = A cos(t + )
Minimum frequency for = 2k, k Z

Maximum frequency for = (2k + 1), k Z
Convention is to use [0, 2) or (, ]
T =
2
1
=
Frequency range of a discrete-time signal is considered to be

(, ]
Rate of oscillation increases as increases (or T decreases).
25 / 61
DTFT
smaller
26 / 61
DTFT
larger, rate of oscillation higher
27 / 61
28 / 61
DTFT
DTFT

MINIMUM OSCILLATION
Discrete-Time Sinusoids: Frequency and Rate of Oscillation:

x[n] = A cos(n + )
MAXIMUM OSCILLATION
Rate of oscillation increases as increases UP TO A POINT then

decreases again and then increases again and then decreases again
....
MINIMUM OSCILLATION
-3
-3
-2
-2
29 / 61
DTFT
-1
-1
-3
-3
31 / 61
-2
-2
30 / 61
DTFT
-1
-1
32 / 61
DTFT
Discrete-Time LTI Filtering
DTFT Theorems and Properties
LTI Filtering
y (n) =
Property
Notation:
Linearity:
Time shifting:
Time reversal
Convolution:
Correlation:
Time Domain
x(n)
x1 (n)
x2 (n)
a1 x1 (n) + a2 x2 (n)
x(n k)
x(n)
x1 (n) x2 (n)
rx1 x2 (l) = x1 (l) x2 (l)
Wiener-Khintchine:
rxx (l) = x(l) x(l)
Frequency Domain
X ()
X1 ()
X1 ()
a1 X1 () + a2 X2 ()
e jk X ()
X ()
X1 ()X2 ()
Sx1 x2 () = X1 ()X2 ()
= X1 ()X2 () [if x2 (n) real]
Sxx () = |X ()|2
Y ()
= H()X ()
where
x(n)
F
h(n)
F
y (n)
H()
among others . . .
|H()|
H() = ()
x(k)h(n k)
k=
33 / 61
X ()
H()
Y ()
|H()|e j()
system gain for freq
phase shift for freq
34 / 61
Complex Nature of X (j)
Complex Nature of X (j)
Recall, Fourier Transform:

Z
X (j)
x(t)e jt dt
Rectangular coordinates: rarely used in signal processing
X (j) = XR (j) + j XI (j)
and Inverse Fourier Transform:

Z
1
x(t) =
X (j)e jt d
2
Z 0
Z
1
1
jt
=
X (j)e d +
X (j)e jt d
2
2 0
where XR (j), XI (j) R.
X (j) = |X (j)| e jX (j)
Note: If x(t) is real, then the imaginary part of the negative frequency sinusoids
(i.e., e jt for <0) cancel out the imaginary part of the positive frequency
sinusoids (i.e., e jt for >0)
Polar coordinates: more intuitive way to represent frequency content
35 / 61
where |X (j)|, X (j) R.
36 / 61
Magnitude and Phase of X (j)
Magnitude and Phase of X (j)

x(t)
|X (j)|: determines the relative presence of a sinusoid e jt in

x(t)
=
=
=
X (j): determines how the sinusoids line up relative to one

another to form x(t)
37 / 61
Z
1
X (j)e jt d
2
Z
1
|X (j)| e jX (j) e jt d
2
Z
1
|X (j)| e j(t+X (j)) d
2
Recall, e j(t+X (j)) = cos(t + X (j)) + j sin(t + X (j)).
The larger |X (j)| is, the more prominent e jt is in forming x(t).
X (j) determines the relative phases of the sinusoids (i.e. how they line
up with respect to one another).
38 / 61
LTI Filtering
LTI Systems as Frequency-Selective Filters

y (n)
x(k)h(n k)
k=
Y ()
= H()X ()
where
x(n)
F
h(n)
F
y (n)
X ()
H()
Filter: device that discriminates, according to some attribute of

the input, what passes through it
For LTI systems, given Y () = H()X ()

I
Y ()
I
H()
|Y ()| =
Y ()
H() acts as a weighting or spectral shaping function of the

different frequency components of the signal
LTI system is known as a frequency shaping filter
LTI system filter
|H()|e j()
|H()||X ()|
= () + X ()
39 / 61
40 / 61
Causal FIR Filters
Causal IIR Filters
Definition: a discrete-time finite impulse response (FIR) filter is one

in which the associated impulse response has finite duration.
y (n) =
Definition: a discrete-time infinite impulse response (IIR) filter is one

in which the associated impulse response has infinite duration.
h(k)x(n k)
k=
M1
X
y (n) =
k=
h(k)x(n k)
k=0
I
I
h(k)x(n k)
k=0
lower limit of k = 0 is from causality requirement

upper limit of 0 M 1 < is from the finite duration
requirement; in this case the support is M consecutive points
starting at time 0 and ending at M 1
h(k)x(n k)
I
I
41 / 61
lower limit of k = 0 is from causality requirement

necessary upper limit of is from the infinite duration
requirement
42 / 61
LCCDEs
LCCDEs
Linear constant coefficient difference equations (LCCDEs) are an

important class of filters that we consider in this course:
Q: Why does an LCCDE have a rational system function?

y (n)
N
X
ak y (n k) +
k=1
y (n) =
N
X
k=1
ak y (n k) +
M
X
bk x(n k)
a0 y (n) =
k=0
They have a rational system function:

PM
k
polynomial in z
k=0 bk z
H(z) =
=
P
N
another polynomial in z
1 + k=1 ak z k
ak y (n k) =
k=0
Z{
ak y (n k) +
N
X
ak y (n k)}
N
X
43 / 61
ak Z{y (n k)}
k=0
M
X
M
X
bk x(n k)
a0 1
k=0
bk x(n k)
k=0
k=0
Depending on the values of N, M, ak and bk they can correspond to

either FIR or IIR filters.
N
X
bk x(n k)
k=0
k=1
N
X
M
X
Z{
M
X
bk x(n k)}
k=0
M
X
bk Z{x(n k)}
k=0
44 / 61
z-Transform Properties
Property
Notation:
LCCDEs
Linearity:
Time shifting:
Time Domain
x(n)
x1 (n)
x2 (n)
a1 x1 (n) + a2 x2 (n)
x(n k)
z-Domain
X (z)
X1 (z)
X1 (z)
a1 X1 (z) + a2 X2 (z)
z k X (z)
z-Scaling:
Time reversal
Conjugation:
z-Differentiation:
Convolution:
an x(n)
x(n)
x (n)
n x(n)
x1 (n) x2 (n)
X (a1 z)
X (z 1 )
X (z )
dX (z)
z dz
X1 (z)X2 (z)
N
X
ROC
ROC: r2 < |z| < r1
ROC1
ROC2
At least ROC1 ROC2
ROC, except
z = 0 (if k > 0)
and z = (if k < 0)
|a|r2 < |z| < |a|r1
1
< |z| < r1
r1
2
ROC
r2 < |z| < r1
At least ROC1 ROC2
k=0
ak Z{y (n k)}
|
{z
}
k=0
z k Y (z)
N
X
ak z k Y (z)
k=0
Y (z)
M
X
M
X
bk z k X (z)
ak z k
= X (z)
Y (z)
X (z)
PM
k=0
H(z)
45 / 61
M
X
bk z k
k=0
H(z) =
z k X (z)
k=0
N
X
k=0
bk z k
a0 1
k
k=0 ak z
PM
k
k=0 bk z
P
N
z 0 + k=1 ak z k
PN
among others . . .
bk Z{x(n k)}
|
{z
}
PM
=
k=0
1+
PN
bk z k
k=1 ak z
46 / 61
FIR LCCDEs
Block Diagram Represenation
y (n) =
M1
X
bk x(n k) =
H(z) =
Adder:
h(k)x(n k)
k=
k=0
M1
X
Unit delay:
bk z
k=0
Constant multiplier:
Please note: upper limit is M 1 opposed to M (which is used for the general
LCCDE case) to meet common FIR convention of an M-length filter.
Unit advance:

h(n) =
bn 0 n M 1
0 otherwise
Signal multiplier:
47 / 61
By inspection:
48 / 61
FIR Filter Implementation

y (n) =
M1
X
IIR LCCDEs
bk x(n k)
k=0
y (n) =
...
+
...
ak y (n k) +
k=1
PM
H(z) =
N
X
1+
k=0
P
N
bk z k
k=1
ak z k
M
X
k=0
M
X
bk x(n k)
1
PN
1 + k=1 ak z k
{z
}
} |
bk z k
|k=0 {z
H1 (z)
H2 (z)
= H1 (z) H2 (z)
Requires:
I M multiplications
I M 1 additions
I M 1 memory elements
49 / 61
Adder:
50 / 61
Direct Form I IIR Filter Implementation
Direct Form II IIR Filter Implementation
Unit delay:
+Unit
LTI All-zero system
...
...
...
advance:
LTI All-pole system
LTI All-pole system
Requires: M + N + 1 multiplications, M + N additions, M + N memory locations

...
...
...
...
...
Signal multiplier:
51 / 61
LTI All-zero system
Requires: M + N + 1 multiplications, M + N additions, M + N memory locations

52 / 61
Direct Form II IIR Filter Implementation

+
Stability of Rational System Function Filters
y (n) =
N
X
ak y (n k) +
k=1
...
...
H(z) =
1+
k=0
P
N
bk x(n k)
k=0
PM
...
M
X
bk z
k=1
ak z k
...
Recall, for BIBO stability of a causal system the system poles must
be strictly inside the unit circle.
For N>M
Why?
Requires: M + N + 1 multiplications, M + N additions, max(M, N) memory

locations
53 / 61
Adder:
54 / 61
Unit delay:
H(z) =
Unit advance:
Recall,
|H(z)|
n=
h(n)z n
|h(n)z
|=
|h(n)||z n |
X
Signal multiplier:
|h(n)| < LTI system is stable
n=
n=
n=
When evaluated for |z| = 1 (i.e., on the unit circle),

|H(z)|
|h(n)|<
n=
Therefore, BIBO stability = ROC includes unit circle

ROC includes unit circle = BIBO stability is also true.
55 / 61
56 / 61

. . . because
Therefore,
|h(n)| <
n=
H(z) ROC
LTI system
includes
is stable
unit circle
For a causal rational system function, the ROC includes the unit
circle if all the poles are inside the unit circle.
57 / 61
ARMA, MA and AR Filters
Other commonly used terminology for the filters described include:

I Autoregressive moving average (ARMA) filter:

I Moving average (MA) filter:
y (n) =
H(z) =
I
I
N
X
ak y (n k) +
k=1
PM
k
k=0 bk z
P
+ Nk=1 ak z k
M
X
bk x(n k)
y (n) =
k=0
H(z) =
M
X
k=0
M
X
bk x(n k)
bk z k
k=0
has both poles and zeros

IIR
58 / 61
I
I
59 / 61
has zeros only; no poles; is BIBO stable

FIR
60 / 61

I Autoregressive (AR) filter:
y (n) =
N
X
ak y (n k)
k=1
H(z) =
I
I
1+
1
PN
k=1
ak z k
has poles only; no zeros

IIR
61 / 61

4 Kundur DT Analysis

Загружено:

Сведения о документе

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

4 Kundur DT Analysis

Загружено:

Авторское право:

Доступные форматы

Discrete-Time LTI Systems

Input-Output Description of Dst-Time Systems

Discrete-Time LTI Systems and Analysis

Dr. Deepa Kundur

Input-output description (exact structure of system is unknown

black box representation:

Discrete-Time LTI Systems and Analysis

Discrete-Time LTI Systems

Dr. Deepa Kundur (University of Toronto)

Discrete-Time LTI Systems and Analysis

Discrete-Time LTI Systems

Why is this so important?

mathematical techniques developed to analyze systems are often

to disprove a property, need a single counter-example

Dr. Deepa Kundur (University of Toronto)

Discrete-Time LTI Systems and Analysis

Dr. Deepa Kundur (University of Toronto)

Discrete-Time LTI Systems and Analysis

Discrete-Time LTI Systems

Discrete-Time LTI Systems

Time-invariant system: input-output characteristics do not

a system is time-invariant iff

x(n) y (n) = x(n n0 ) y (n n0 )

for every input x(n) and every time shift n0 .

Dr. Deepa Kundur (University of Toronto)

Discrete-Time LTI Systems and Analysis

Discrete-Time LTI Systems

Discrete-Time LTI Systems and Analysis

Discrete-Time LTI Systems

Dr. Deepa Kundur (University of Toronto)

Linear system: obeys superposition principle

a system is linear iff

Dr. Deepa Kundur (University of Toronto)

Discrete-Time LTI Systems and Analysis

Dr. Deepa Kundur (University of Toronto)

Discrete-Time LTI Systems and Analysis

Discrete-Time LTI Systems

Discrete-Time LTI Systems

The Convolution Sum

The Convolution Sum

Causal system: output of system at any time n depends only on

a system is causal iff

y (n) = F [x(n), x(n 1), x(n 2), . . .]

Bounded Input-Bounded output (BIBO) Stable: every bounded

a system is BIBO stable iff

Dr. Deepa Kundur (University of Toronto)

Discrete-Time LTI Systems and Analysis

Discrete-Time LTI Systems

Dr. Deepa Kundur (University of Toronto)

The Convolution Sum

Discrete-Time LTI Systems and Analysis

Discrete-Time LTI Systems

The Convolution Sum

The Convolution Sum

The Convolution Sum

Let the response of a linear time-invariant (LTI) system denoted T to

x(k)h(n k) = x(n) h(n)

x(k) (n k) x(k) h(n k)

for any LTI system.

Dr. Deepa Kundur (University of Toronto)

Discrete-Time LTI Systems and Analysis

Dr. Deepa Kundur (University of Toronto)

Discrete-Time LTI Systems and Analysis

Discrete-Time LTI Systems