Call 3220

c Kevin Buckley - 2013
-10
ECE3220 Signal Processing

Fall 2013
Villanova University
ECE Department
Prof. Kevin M. Buckley
x[n]
b0
y[n]
b0
x[n]
b1
b1
a1
a2
b2
+
+
b M1
+
+
adder
a N1
bM
delay
b M1
b2
bM
aN
adder
delay
multiplier
multiplier
(a)
(b)
1.5
amplitude
y[n]
1
Thomas Friedman
Roger Waters
h[k]
1
0.5
0.5
0
0.5
0
t
Alan Oppenheim
0.5
0.5
0
amplitude
0.5
1
1.5
0.5
0
t
0
40
1
0.5
0
40
10
y[n]
0.5
0
20
memory time k
40
0.5
Ayn Rand
20
Stephen Hawking
x[nk]; n=0
amplitude
0.5
1
1.5
20
0
20
memory time k
Lauren Visualization
40
yc (t), y[n] = yc (nT)
0.5
1
5
0.5
0 t
0.5
John Butler
0
30
0
50
0
5
Magnitude Response (dB)
4
6
8
time t (milliseconds)
10
3
x 10
100
4
20
10
0
10
sample time n
20
30
Chase Utley
2
0
2
Freq. (radians/sample)

.
-9
-8
Contents
1 Introduction
1.1 Motivation & Background . . . . . . . . . . . . . . . . . . . .
1.2 Signal Processing Examples . . . . . . . . . . . . . . . . . . .
1.2.1 Discussion on Sampling & Reconstruction . . . . . . .
1.2.2 An RC Circuit . . . . . . . . . . . . . . . . . . . . . .
1.2.3 Channel Equalization . . . . . . . . . . . . . . . . . . .
1.2.4 An FIR Filter Example . . . . . . . . . . . . . . . . . .
1.2.5 A Simple Discrete-Time (DT) System . . . . . . . . . .
1.3 Course Objective, Linear Combinations & Two Basic Concepts
1.4 Discrete & Continuous Time Signals & Operators . . . . . . .
1.4.1 Basic Discrete-Time Signals & Operators . . . . . . . .
1.4.2 Basic Continuous-Time Signals & Operators . . . . . .
1.4.3 Signal Classes . . . . . . . . . . . . . . . . . . . . . . .
1.4.4 Periodic Signals and Sinusoids . . . . . . . . . . . . . .
1.4.5 Basic Signal Sets and Transforms . . . . . . . . . . . .
1.5 Linear Time-Invariant (LTI) Systems . . . . . . . . . . . . . .
1.5.1 System Examples . . . . . . . . . . . . . . . . . . . . .
1.5.2 System Properties . . . . . . . . . . . . . . . . . . . . .
1.5.3 Linear Time-Invariant (LTI) Systems . . . . . . . . . .
1.5.4 Linear Constant Coecient System I/O Equations . .
1.6 Practicum 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.7 Appendix 1A: Complex Numbers and Signals . . . . . . . . .
1.7.1 Complex Numbers . . . . . . . . . . . . . . . . . . . .
1.7.2 Algebra with Complex Numbers . . . . . . . . . . . . .
1.7.3 Complex-Valued Signals . . . . . . . . . . . . . . . . .
1.7.4 Why Consider Complex-Valued Signals? . . . . . . . .
1.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2 Time Domain Analysis of LTI Systems

2.1 DT Signal Impulse Expansion . . . . . . . . . . . . . . . . . . . .
2.2 DT LTI System I/O Calculation: The Convolution Sum . . . . .
2.2.1 The Impulse Response of a DT LTI System & Convolution
2.2.2 Approaches for the Calculation of the Convolution Sum . .
2.2.3 Convolution Sum Properties . . . . . . . . . . . . . . . . .
2.3 DT LTI System Issues . . . . . . . . . . . . . . . . . . . . . . . .
2.3.1 System Properties . . . . . . . . . . . . . . . . . . . . . . .
2.3.2 DT LTI Subsystem Interconnections . . . . . . . . . . . .
2.3.3 DT LTI System Response to Sinusoidal Inputs . . . . . . .
2.4 CT Signal Impulse Expansion . . . . . . . . . . . . . . . . . . . .
2.5 CT LTI System I/O Calculation: Convolution Integral . . . . . .
2.5.1 The Impulse Response of a CT LTI System . . . . . . . .
2.5.2 Convolution Integral . . . . . . . . . . . . . . . . . . . . .
2.5.3 Convolution Integral Properties and Examples . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . .
. . .
Sum
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
. . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
3
5
5
6
7
8
9
11
15
15
22
25
28
32
34
34
41
48
51
53
59
59
60
62
65
67
.
.
.
.
.
.
.
.
.
.
.
.
.
.
79
80
81
81
94
95
100
100
102
105
108
109
109
109
110

2.6
2.7
2.8
2.9
CT LTI System Issues . . . . . . . . .

Generalizing the Utility of Convolution
Practicum 2 . . . . . . . . . . . . . . .
Problems . . . . . . . . . . . . . . . . .
-7
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3 Continuous Time Transforms

3.1 Some Motivation for the Study of Transforms . . . . . . . . .
3.1.1 Preliminary View of Spectral Representation . . . . . .
3.1.2 System Response to Complex Exponential Signals . . .
3.2 CT Fourier Series (CTFS) Representation of Periodic Signals .
3.2.1 A Few CTFS Properties . . . . . . . . . . . . . . . . .
3.3 The Continuous Time Fourier Transform (CTFT) . . . . . . .
3.3.1 The Generalized CTFT . . . . . . . . . . . . . . . . . .
3.4 The Laplace Transform (LT) . . . . . . . . . . . . . . . . . . .
3.4.1 The Bilateral Laplace Transform (BLT) . . . . . . . . .
3.4.2 Relationship Between the BLT and the CTFT . . . . .
3.4.3 The Inverse Bilateral Laplace Transform . . . . . . . .
3.4.4 The Unilateral Laplace Transform (ULT) . . . . . . . .
3.4.5 Some Useful Rules-of-Thumb on When to Use Dierent
3.5 Continuous Time Transform Properties . . . . . . . . . . . . .
3.6 Practicum 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4 Applications of CT Transforms
4.1 Continuous Time LTI Systems . . . . . . . . . . . . .
4.1.1 CT LTI System I/O Descriptions . . . . . . .
4.1.2 Partial Fraction Expansion (PFE) . . . . . . .
4.1.3 Stability, Causality, the j Axis and ROC . .
4.1.4 Frequency Response and Pole/Zero Locations
4.1.5 The Unilateral Laplace Transform and System
4.1.6 CT LTI System Block Diagrams . . . . . . . .
4.2 Signal Processing Functions . . . . . . . . . . . . . .
4.2.1 Filtering . . . . . . . . . . . . . . . . . . . . .
4.2.2 Sampling . . . . . . . . . . . . . . . . . . . .
4.2.3 Modulation . . . . . . . . . . . . . . . . . . .
4.2.4 RLC Circuits . . . . . . . . . . . . . . . . . .
4.2.5 Spectrum Estimation . . . . . . . . . . . . . .
4.3 Practicum 4a . . . . . . . . . . . . . . . . . . . . . .
4.4 Practicum 4b . . . . . . . . . . . . . . . . . . . . . .
4.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
Transforms
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
Initial Conditions
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
117
120
121
127
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
145
146
146
152
155
163
166
171
173
173
177
178
180
180
181
206
215
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
227
227
229
232
237
240
241
243
245
246
253
262
269
271
272
277
283
5 Discrete Time Transforms

299
5.1 The Frequency Response & Transfer Function of a DT LTI System . . . . . 301
5.2 The DTFS Representation of DT Periodic Signals . . . . . . . . . . . . . . . 306
5.3 The DTFT Representation of DT Energy Signals . . . . . . . . . . . . . . . 310
-6
5.4
5.5
5.6
5.7
The DTFT Representation of Periodic Signals .

Properties of the DTFT . . . . . . . . . . . . .
5.5.1 Periodicity of X(ej ) . . . . . . . . . . .
5.5.2 Symmetry of X(ej ) . . . . . . . . . . .
5.5.3 Time Delay . . . . . . . . . . . . . . . .
5.5.4 Linearity . . . . . . . . . . . . . . . . . .
5.5.5 Convolution . . . . . . . . . . . . . . . .
5.5.6 Parsevals Theorem . . . . . . . . . . . .
5.5.7 Multiplication . . . . . . . . . . . . . . .
5.5.8 Modulation . . . . . . . . . . . . . . . .
The z-Transform . . . . . . . . . . . . . . . . .
5.6.1 The Bilateral z-Transform and its Region
5.6.2 Selected Properties of the z-Transform .
Problems . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
of Convergence
. . . . . . . . .
. . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
315
318
318
320
322
323
324
327
328
331
334
334
339
342
6 Applications of DT Transforms
6.1 The DTFT and DT LTI Systems . . . . . . . . . . . . . .
6.1.1 The Frequency Response . . . . . . . . . . . . . . .
6.1.2 Bringing in the I/O Dierence Equation . . . . . .
6.2 The z-Transform and the LTI System Transfer Function . .
6.2.1 Frequency Response . . . . . . . . . . . . . . . . .
6.2.2 Partial Fraction Expansion (PFE) of Rational H(z)
6.2.3 Stability, Causality, the Unit Circle and ROC . . .
6.2.4 DT LTI System Block Diagrams . . . . . . . . . . .
6.3 Signal Processing Functions & Implementation . . . . . . .
6.3.1 Channel Equalization . . . . . . . . . . . . . . . . .
6.3.2 Sampling: A DT Perspective . . . . . . . . . . . . .
6.3.3 Filter Banks . . . . . . . . . . . . . . . . . . . . . .
6.3.4 Spectrum Estimation . . . . . . . . . . . . . . . . .
6.3.5 Real-Time DT Systems . . . . . . . . . . . . . . . .
6.4 Practicum 4c . . . . . . . . . . . . . . . . . . . . . . . . .
6.5 Practicum 5a . . . . . . . . . . . . . . . . . . . . . . . . .
6.6 Practicum 5b . . . . . . . . . . . . . . . . . . . . . . . . .
6.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
349
350
350
351
354
357
362
369
372
375
375
379
379
379
379
380
385
391
401
7 Introduction to Random Processes

7.1 Review of Random Variables . . . . . . . .
7.2 DT Random Processes . . . . . . . . . . .
7.3 Power Spectral Density . . . . . . . . . . .
7.4 DT Random Processes & DT LTI Systems
7.5 Problems . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
419
420
426
431
434
437
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
-5
List of Figures
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
An illustration of sampling. . . . . . . . . . . . . . . . . . . . . . . . . . . .
Ideal sampling and reconstruction block diagrams. . . . . . . . . . . . . . . .
A simple parallel RC circuit. . . . . . . . . . . . . . . . . . . . . . . . . . . .
A multipath communication channel. . . . . . . . . . . . . . . . . . . . . . .
The FIR lter structure (the D block represents a delay; combined, the set
of delays form a delay line, a.k.a. a shift register). . . . . . . . . . . . .
A simple DT system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Representation of a CT square wave as a linear combination of sinusoids. The
idea of harmonics will be established in Chapter 3 of this Course. . . . . .
A delayed (a.k.a. shifted) step. . . . . . . . . . . . . . . . . . . . . . . . . .
A folded and shifted step. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A right-sided decaying exponential. . . . . . . . . . . . . . . . . . . . . . . .
A pulse formed from two steps. . . . . . . . . . . . . . . . . . . . . . . . . .
An exponential pulse. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
y[n] = x[n] h[4 n]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Block diagram of an accumulator. . . . . . . . . . . . . . . . . . . . . . . . .
Basic CT signals: (a) step; (b) ramp; (c) pulse; (d) impulse. . . . . . . . . .
Dierentiation and integration involving impulses. . . . . . . . . . . . . . . .
Compression (scaling with a > 1). . . . . . . . . . . . . . . . . . . . . . . . .
Expansion (scaling with 0 < a < 1). . . . . . . . . . . . . . . . . . . . . . . .
A periodic pulse train. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A delayed (therefore phase shifted) cosine signal. . . . . . . . . . . . . . . . .
Two periodic signals (to be summed). . . . . . . . . . . . . . . . . . . . . . .
A simple DT system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The FIR lter structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A parallel interconnection of subsystems. . . . . . . . . . . . . . . . . . . . .
A cascade interconnection of subsystems. . . . . . . . . . . . . . . . . . . . .
A feedback interconnection of subsystems. . . . . . . . . . . . . . . . . . . .
Inverting the eect of a degrading physical system. . . . . . . . . . . . . . .
Assuming linearity, given outputs for x1 (t) and x2 (t), what are the outputs
for x3 (t) and x4 (t)? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Illustration of the test for time-invariance. . . . . . . . . . . . . . . . . . . .
Assuming time-invariance, given the output for x1 (t), what are the outputs
for x2 (t) and x3 (t)? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
An LTI system and the impulse I/O pair. . . . . . . . . . . . . . . . . . . . .
An LTI system and the impulse I/O pair. . . . . . . . . . . . . . . . . . . . .
A Discrete-Time Linear Time-Invariant (DT LTI) system. . . . . . . . . . .
Assuming linearity and time-invariance, given the output for x1 (t), what are
the outputs for x2 (t) and x3 (t)? . . . . . . . . . . . . . . . . . . . . . . . . .
DT LTI system structures: a) FIR lter; and b) general IIR lter. . . . . . .
A complex number x in the complex plane. . . . . . . . . . . . . . . . . . . .
An in-phase/quadrature (I/Q) receiver. . . . . . . . . . . . . . . . . . . . . .
Representation of a DT signal as linear combination of delayed impulses. . .
5
6
6
7
9
9
14
16
18
18
19
19
19
20
22
23
24
24
27
28
31
35
36
39
40
40
41
44
45
46
46
47
48
49
52
59
66
80
-4
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
A DT LTI system and the convolution sum. . . . . . . . . . . . . . . . . . . 81

Example 2.1 convolution sum illustration. . . . . . . . . . . . . . . . . . . . 82
Plots generated by the Convolution Sum Demo. . . . . . . . . . . . . . . . . 88
The FIR lter structure (implementing a convolution sum directly). . . . . . 89
Given convolution result for Example 2.15. . . . . . . . . . . . . . . . . . . . 97
Illustration of the overlap-and-add convolution sum calculation. . . . . . . . 99
Parallel interconnection of DT LTI subsystems. . . . . . . . . . . . . . . . . 102
Cascade interconnection of DTLTI subsystems. . . . . . . . . . . . . . . . . 103
Example 2.23 interconnection of DTLTI subsystems. . . . . . . . . . . . . . 104
Representation of a CT LTI system using its impulse response h(t). . . . . . 109
A CT LTI system and the convolution integral. . . . . . . . . . . . . . . . . 109
Example 2.31 CT convolution integral illustration. . . . . . . . . . . . . . . . 112
Convolution integral result for Example 2.38. . . . . . . . . . . . . . . . . . . 115
CTLTI subsystem interconnection for Example 2.41. . . . . . . . . . . . . . . 118
Serial RLC circuit for Example 2.43. . . . . . . . . . . . . . . . . . . . . . . 119
Serial RC circuit for Example 2.44. . . . . . . . . . . . . . . . . . . . . . . . 119
An example of the spectrum of a sum of sinusoids. . . . . . . . . . . . . . . . 148
The plot of an AM signal and its spectrum. . . . . . . . . . . . . . . . . . . 149
Frequency spectra: (a) as a representation of frequency content for all time;
(b) the time-frequency spectrum of a real-valued sinusoid; (c) the time-frequency
spectrum of a musical scale; and (d) an estimate of the time-frequency spectrum.150
The CT LTI system response for an exponential input. . . . . . . . . . . . . 153
CTFS approximations of a square wave using dierent numbers of harmonics. 159
A sawtooth waveform and its DC, 1-st harmonic approximation. . . . . . . . 160
A fullwave rectied sinusoid. . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
Example 3.11 signals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
CTFT of a periodic signal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
CTFT of a impulse train. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
CTFT of an AM signal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
ROC illustrations for dierent types of signals. . . . . . . . . . . . . . . . . . 175
CTFT of an AM modulated sinc2 signal. . . . . . . . . . . . . . . . . . . . . 190
Illustration of the multiplication property of the CTFT. . . . . . . . . . . . . 193
Transform based I/O representations of a CT LTI system. . . . . . . . . . . 228
Converting between dierent CT LTI system I/O descriptions. . . . . . . . . 230
Illustration of the relationship between pole positions, ROC and causality of
impulse response components. . . . . . . . . . . . . . . . . . . . . . . . . . . 237
s-plane illustration for Example 4.9. . . . . . . . . . . . . . . . . . . . . . . . 239
(a) Pole/zero locations; and (b) frequency response for a CT notch lter. . . 240
Block diagram for Example 4.11. . . . . . . . . . . . . . . . . . . . . . . . . 243
Block diagram for Example 4.12. . . . . . . . . . . . . . . . . . . . . . . . . 244
Typical CT lowpass lter design specications. . . . . . . . . . . . . . . . . . 250
Pole plot (there are no zeros) for a Butterworth CT lowpass lter. . . . . . . 252

80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
-3
Frequency response (magnitude-squared in db) and phase response for a Butterworth CT lowpass lter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
An idealistic C-to-D converter. . . . . . . . . . . . . . . . . . . . . . . . . . . 253
An idealistic D-to-C converter. . . . . . . . . . . . . . . . . . . . . . . . . . . 255
An antialiasing lter as a preprocessor before a C-to-D converter. . . . . . . 255
A spectral illustration of aliasing. . . . . . . . . . . . . . . . . . . . . . . . . 256
A visualization of sampling, as modeled using an impulse train multiplication
and studied using the CTFT. . . . . . . . . . . . . . . . . . . . . . . . . . . 258
Reconstruction of x(t) from xT (t) (i.e. from its samples) . . . . . . . . . . 259
AM modulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
A practical asynchronous AM transmitter and receiver. . . . . . . . . . . . . 264
System for Example 4.22. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Laplace transform representation of: (a) a resistor; (b) an inductor; (c) a
capacitor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
Serial RLC circuit for Example 4.26. . . . . . . . . . . . . . . . . . . . . . . 270
RLC circuit for Example 4.27. . . . . . . . . . . . . . . . . . . . . . . . . . . 270
A DT LTI system and the convolution sum. . . . . . . . . . . . . . . . . . . 301
Frequency response for Example 5.1. . . . . . . . . . . . . . . . . . . . . . . 303
Transform based I/O representations of a DT LTI system. . . . . . . . . . . 305
Illustration of the ambiguity of DTFS frequencies. . . . . . . . . . . . . . . . 306
DTFS coecients for Example 5.4. . . . . . . . . . . . . . . . . . . . . . . . 307
DTFT for Example 5.7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
DTFT for Example 5.10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
DTFT for Example 5.14. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
DTFT for Example 5.15. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
DTFT for Example 5.16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
DTFT for Example 5.17. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
DTFT for Example 5.18. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
Phase response for Example 5.20. . . . . . . . . . . . . . . . . . . . . . . . . 321
DTFT for Example 5.21. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
Frequency domain circular convolution for Example 5.29. . . . . . . . . . . . 329
The DTFT convolution property and DT LTI systems. . . . . . . . . . . . . 350
A DT LTI system and its transfer function. . . . . . . . . . . . . . . . . . . 354
The eect of poles and zeros on the transfer function H(z). . . . . . . . . . . 356
(a) evaluation of the z-plane on the unit circle (illustrating the DTFT/ztransform relationship); (b) relationships between DT LTI system descriptions.357
-2
121
122
123
124
125
126
127
128
129
130
Pole/zero plot, and corresponding magnitude response, for Example 6.7. . .

Pole/zero locations and frequency responses for two DT lowpass lters. . . .
Pole/zero plot, and corresponding magnitude response, for Example 6.9. . .
Illustration of the relationship between pole positions, ROC and causality of
impulse response components. . . . . . . . . . . . . . . . . . . . . . . . . . .
Pole/zero plot, and possible ROCs, for Example 6.16. . . . . . . . . . . . . .
DT LTI model of the digital communication channel equalization problem. .
The uniform pdf. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The Gaussian pdf. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A 2 dimensional pdf. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Illustration of a realization of DT white noise. . . . . . . . . . . . . . . . . .
360
361
361
369
371
375
421
421
422
426
-1
List of Tables
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
A Summation Table. . . . . . . . . . . . . . . . . . . . .
Basic Signal Sets and corresponding transforms. . . . . .
Convolution Sum Results. . . . . . . . . . . . . . . . . .
Convolution Integral Results. . . . . . . . . . . . . . . .
Transforms. . . . . . . . . . . . . . . . . . . . . . . . . .
Continuous Time Fourier Series (CTFS) Pairs. . . . . . .
Continuous Time Fourier Transform (CTFT) Pairs. . . .
Bilateral Laplace Transform (BLT) Pairs. . . . . . . . . .
Continuous Time Transform Properties. . . . . . . . . .
Continuous Time Fourier Transform (CTFT) Properties.
Bilateral Laplace Transform (BLT) Properties. . . . . . .
Discrete Time Fourier Series (DTFS) Pairs. . . . . . . .
Discrete Time Fourier Transform (DTFT) Pairs. . . . . .
DTFT Properties. . . . . . . . . . . . . . . . . . . . . . .
z-Transform Pairs. . . . . . . . . . . . . . . . . . . . . .
z-Transform Properties. . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
21
32
94
111
151
161
168
176
182
204
205
309
314
333
338
339
Introduction
This Course is an introduction to the theory and methods of:

1. signal analysis and synthesis; and
2. system analysis, design and implementation.
We consider both continuous-time (CT) and discrete-time (DT) signals and systems. For
electrical engineering students our emphasis is slightly more on CT than DT. For computer
engineers our focus will be on DT, though it will be necessary to consider CT signals as they
relate to the DT signals derived from them via sampling. As we will see, the methods for
DT are very similar to those for CT, so both the EE student and CPE student approaches
provide useful experience for both CT and DT signals and systems.
There are two basic topics that encompass the methods developed throughout this Course.
These are:
1. convolution; and
2. transforms.
These basic topics are relevant to both CT and DT systems. The objective of this Course is
to become familiar with these two basic topics, and to learn how they are applied to signals
and systems.
This Chapter of the Course consists of introductory material. Its overall objective is
threefold:
1. to motivate the topics of this Course;
2. to establish required background; and
3. to introduce, informally, several basic concepts which are central to this Course.
After completing this Chapter, you should be prepared to begin learning the theory and
methods covered in this Course. Refer to the Chapter 1 Objective Checklist on the next
page for a list of specic Chapter 1 objectives.
A word of caution although no principal Course topics are covered in this Chapter, we
do establish required prerequisites, important notation and basic concepts. Over the rst
two weeks, as we progress through this introduction, make sure you are comfortable with
the material we cover (e.g. complex sinusoids, signal classes, the basic concept of linear
combinations as it relates to transforms, basic system properties). Students who struggle
with this Course are usually students who do not apply themselves during this introduction.
Chapter 1 Objective Checklist

Be comfortable with complex numbers, including simple algebraic operations (e.g. addition, multiplication, powers) and the use of Eulers Identities.
Be familiar with basic DT and CT signals (e.g. impulses, steps, ramps, exponentials)
and basic operations on them (e.g. fold, shift, fold and shift).
Be comfortable with real and complex-valued DT and CT sinusoids.
Be comfortable with the application of integration as a tool for the solution of basic
signal processing problems.
Be comfortable with the application of summation, including the use of Table 1, as a
tool for the solution of basic signal processing problems.
Be able to compute the energy and power of DT and CT signals.
Be comfortable with periodic signals.
Understand the idea of a set of basic signals.
Understand the basic concept of linear combinations of basic signals.
Understand the the rst basic concept of this Course to represent and analyze signals
as linear combinations of basic signals.
Be familiar with some simple DT systems. (You are already familiar with some CT
systems from your circuits course.)
Be able to compute the output of some simple DT systems due to simple inputs. This
includes the ability to identify the impulse response of a simple DT system.
Be able to compute the output of some simple CT & DT systems due to complex
sinusoidal inputs.
Be familiar with basis (CT and DT) system properties (i.e. linearity, time-invariance,
causality and stability).
Understand the the second basic concept of this Course to describe and analyze
systems in terms of their response to basic signals.
1.1
Motivation & Background
As with many engineering areas, signals and systems is about the eective use of mathematical and/or scientic methods to application problems. For this area, the methods are
primarily mathematical, and the applications are very diverse. In this Subsection we suggest motivation for studying signals and systems, discuss some applications, and provide an
overview and some background for the topics we will consider.
Motivation
Professionally, engineers are valued in part because they can solve complex technical design
and analysis problems. The topics covered in this Course (i.e. signals and systems topics)
provide the most important basic tools for a systematic approach to solving complex technical
problems. That is, they are the tools for converting challenging problems into easy ones.
Engineering is interesting because of the challenge of solving complex technical problems,
and because the problems engineers work on are practical and real-world in nature. Although the topics covered in this Course are mathematical, they are all about applications.
Signal processing engineers work on some very interesting and important applications. Here
we list just a few of these. In the rst class we will briey discuss a few of these, and
throughout the semester (for example, in the Practicums) we will consider examples of how
signals and systems topics relate to applications.
Applications
Here is a list of just few of the many applications of signals and systems.
Multimedia signal processing: audio & video processing, data compression, A/D and
D/A converters, HDTV, audio special eects.
Communications: cell phones, satellite communications, deep space communication.
Physical sciences: geophysical exploration, astrophysical exploration.
Biomedical engineering: biomedical imaging, cardiac pacemakers and debrillators,
health monitoring, cognitive studies, hearing aids, Brain/Computer Interface (BCI).
Biomedical Signal Processing (BSP) is the topic of ece5251.
Traditional electronic systems: SONAR, RADAR.
Control systems: manufacturing process control, trajectory control.
Machine diagnostics: engine monitoring and control, fault detection.
Circuit and electronic modeling.
Forecasting: weather, economic trends.
From this list, it is clear that signal processing occurs all around us and we use signal
processing systems on a regular basis.
In the rst lecture we will discuss signal processing applications. Prepare for this by
thinking a little about: 1) how and why you use signal processing in your daily life; 2) what
problems you are aware of that signal processing might solve; and 3) what signal processing
problems you may be interested in professionally.
Implementation
Engineers implement useful things. Signal processing is implemented in various ways that
electrical and computer engineers have expertise in. This is one reason signals and systems
is an important core course in any ECE curriculum. Implementation platforms include the
following:
Continuous-time hardware: circuits, electronics.
Digital hardware: microprocessors, ASICs, FPGAs.
Digital software implemented on general purpose digital systems: C++, Java, Matlab.
Digital software implemented on microprocessors (e.g. on Digital Signal Processing
(DSP) chips): assembly language, cross-compiled C programs.
In this Course we will implement signal processing, digitally, using Matlab. This is what the
Practicums are about. In ece5790 we cover more advanced implementation issues. To learn
how to implement signal processing in software on DSP chips, consider ece7710. To learn
how to implement signal processing in hardware on FPGAs, consider ece7711.
Prerequisites
The formal prerequisites for this Course are listed on the Course Information page. The
topics you need to be familiar coming into this Course are:
Calculus (i.e. integration, dierentiation): assumes background algebra and trigonometry.
Dierential equations: in particular, linear, constant coecient (LCC) dierential
equations.
Complex numbers and basic complex variables (see Appendix 1A, Section 1.7 of these
Notes, for a review).
Basic circuits - linear circuits (e.g. RLC circuits, natural response, step response,
frequency response) and superposition.
Matlab.
If you are concerned about your background in any of these prerequisite topic, please feel
free to discuss your concerns with me.
My experience with this Course is that, with these prerequisites, you should be ne as
long as you keep working. This is not a guarantee, but this Course is set up so that if you
keep trying theres a very high probability that you will be OK. That said, remember that
it is very important to take this rst Chapter seriously. It may not look like much is going
on, but familiarity with the basic concepts and the notation developed over this Chapter, as
well as the math reviewed, is critical as we move on.
1.2
1.2.1
Signal Processing Examples

Discussion on Sampling & Reconstruction
First, to develop a little notation that will facilitate the discussion, consider the ContinuousTime (CT, i.e. analog) signal xc (t) shown in Figure 1(a). The subscript c xc (t) indicates
that the signal is CT. The plotted signal is an exponential, of the general form
c eat
0
xc (t) =
t0
t<0
(1)
where in this case c = 5 is the peak amplitude and a = 2 is the decay rate. This xc (t) is
real-valued (as opposed to complex-valued).
6
(a)
xc (t)
4
2
0
yc (t), y[n] = yc (nT)
2
time t (seconds)
(b)
5
y[n]
4
5
6
time t (milliseconds)
10
3
x 10
(c)
5
6
time n (samples)
10
Figure 1: An illustration of sampling.

Now consider the plot of some signal yc (t) shown in Figure 1(b). This could, for example,
be an audio or communications signal. Also shown in Figure 1(b) is the Discrete-Time (DT)
signal y[n] obtained by sampling yc (t) with sampling interval T = 0.001. Specically, ideal
sampling is illustrated, which means that
y[n] = yc (nT ) ;
n = , 1, 0, 1, 2, 3,
(2)
where n is the integer sample index. Figure 2 show typical ideal sampling and reconstruction
block diagrams.
For the xc (t) in Eq (1),
x[n] = xc (nT ) =
c eanT
0
nT 0
nT < 0
c n
0
n0
n<0
where = eaT . Figure 1(c) shows the DT signal y[n] by itself, plotted vs. n.
(3)
y (t)
c
y[n] = y (nT)
Ideal Sampler
Ideal
Reconstruction
(T)
y (t)
c
(T)
Figure 2: Ideal sampling and reconstruction block diagrams.

Since signals of interest are usually CT, and they are often processed with DT processors,
sampling is an important signal processing function. Two fundamental questions of interest
when sampling and reconstructing a signal are:
Can yc (t) be exactly reconstructed from its samples in y[n]? Generally the answer is
no, as you may intuitively expect since when sampling yc (t) we seemingly loose any
indication of its value between the samples in y[n]. However, as shown in this Course,
it is possible to exactly reconstruct yc (t) from y[n] if the frequency content of yc (t) is
properly limited (i.e. if yc (t) is not changing too quickly) and if T is small enough.
This is a surprising and extremely important result.
How can sampling and reconstruction be implemented? In this Course we describe sampling and reconstruction mathematically. However, these descriptions directly suggest
a variety of part digital, part analog implementation approaches.
1.2.2
An RC Circuit
ECE juniors are familiar with RLC circuits such as the simple circuit shown below.
+
x(t)
y(t)
Figure 3: A simple parallel RC circuit.

For this, we know that a simple node equation can be used to generate the following dierential equation relating input and output:
C
d
1
y(t) +
y(t) = x(t) ,
dt
R
(4)
where x(t) in the input current and y(t) is the output voltage. In your circuits course you
learned how to determine the output of this and other fairly simple circuits for particular
types of inputs (e.g. a step x(t) = u(t), a complex sinusoid x(t) = ejt ). In your dierential equations course you learned to solve some specic dierential equations, and perhaps
realized that no single approach exists for solving all types of dierential equations. In this
course you will learn how to use both the convolution operation and transforms to solve
(i.e. compute outputs for given inputs) this and much more complex circuits for any given input. For example, the following integral equation, which is a convolution, solves this system
for any input x(t) (assuming initial condition y(0 ) = 0):
y(t) =
t
0
1 (t )/RC
e
x( ) d
C
t0 .
(5)
So, regardless of what the input x(t) is, you plug it into Eq (5) and solve for output y(t).
1.2.3
Channel Equalization
Consider the problem of transmitting a continuous-time signal xc (t) over a communication

channel. A communication channel is illustrated in Figure 4(a). For example, the channel
for your cell phone communications system is the physical space between your cell phone
and a base station.
(a)
(b)
xc (t)
xc (t)
channel
vc (t)
vc (t)
channel
equalizer
?
yc (t) = xc (t)
(c)
x[n]
modulator
xc (t)
vc (t)
demodulator
(d)
x[n]
discretetime
channel
model
v[n]
?
y[n] = x[n]
v[n]
equalizer
?
discretetime y[n] = x[n]
equalizer
Figure 4: A multipath communication channel.

Figure 4(a) depicts what is referred to as a multipath channel for which the transmitted
signal arrives at the receiver antenna as a superposition of arrivals from multiple paths. That
is, the received signal vc (t) is a superposition of multiple scaled and delayed inputs xc (t).
At the receiver, the challenge is to process vc (t) to recover xc (t). As illustrated in Figure
4(b), the signal processor that is used to recover xc (t) is called a channel equalizer. For
digital communications a discrete-time sequence of symbols representing a number of bits is
transmitted. At the transmitter, a modulator is used to convert the sequence, x[n], to the
continuous-time signal xc (t) to be transmitted over the physical channel. At the receiver,
a demodulator is used to convert the received vc (t) back to discrete-time form. Figure 4(c)
illustrates this, where v[n] represents the sequence of received symbols which is distorted by
the channel. Figure 4(d) shows a discrete-time model of the discrete-time channel from input
sequence x[n] to output v[n]. It also shows a discrete-time equalizer used to invert the eect
of the channel. This discrete-time channel model, which is commonly used for digital communication system design, encompasses the physical channel, the modulator/demodulator
and the transmitter/receiver antennae and front end electronics.
The objective at the receiver is to acquire the transmitted signal xc (t) from the receiver
signal vc (t), or in the digital communications system to recover x[n] from v[n]. The discretetime equalizer operates to invert the eect of the channel, ideally providing the output
y[n] = x[n]. In Practicum 1 we will explore the digital communication channel equalization
problem. We will consider a relatively simple discrete-time channel model of the form
v[n] = x[n] + a x[n 1] .
(6)
At symbol time n, the channel output is the desired symbol x[n] superimposed with the
scaled/delayed previous symbol (e.g. via the reection o a building). We will observe the
eectiveness of two proposed equalizers.
Eq (6) represents a very simple channel. The two equalizers considered in Practicum 1
were designed specically for this simple channel. In a realistic cell phone application, the
channel can be much more complex, it is not known (depending of the cell phone location
relative to the base station), and it varies over time (as the cell phone moves). As we develop
design and analysis tools in this Course, we will see how the two Practicum 1 equalizers were
designed, and we will begin to be able to handle more realistic channel equalization problems.
1.2.4
An FIR Filter Example
A lter is a system that separates things. It takes a mixture of several things as an input,
and works to provide an output which contains only some of these things. In DT systems,
a lter processes a DT input signal, say
x[n] = s[n] + n[n]
(7)
where s[n] is a desired signal and n[n] is additive interfering noise, creating an output, say
y[n], which is an approximate of s[n]. The most common type of discrete-time lter in termed
an FIR lter1 . An FIR lter forms the output y[n] at time n by processing the present and
previous N 1 input values, i.e. {x[n], x[n 1], , x[n (N 1)]}, as follows
y[n] = b0 x[n] + b1 x[n 1] + + bN 1 x[n N + 1] =
N 1
k=0
bk x[n k]
(8)
where the bk are the multipliers applied to the input values, i.e. x[n] is multiplied by b0 and
then summed with all the other multiplier outputs. Figure 5 is a block diagram illustration
of this type of system.
Suppose, for example that N = 10, bk = 1; k = 0, 1, , 9, s[n] = 1 for all time n, and
n[n] = (1)n for all n. Let the output y[n] be expressed as
y[n] = ys [n] + yn [n] ,
1
We will explain this terminology later.
(9)
x[n]
D
b0
D
b1
D
b N2
b2
b N1
y[n]
Figure 5: The FIR lter structure (the D block represents a delay; combined, the set of
delays form a delay line, a.k.a. a shift register).
i.e. the signal and noise components of the output, respectively. It is pretty easy to show2
that ys [n] = 10 and yn [n] = 0 for all n. So this system amplies s[n] by a factor of 10 and
completely attenuates n[n]. That is, this system does a great job of separating the given s[n]
from n[n].
1.2.5
A Simple Discrete-Time (DT) System
Since you have already studied CT systems in your circuits courses, you probably have more
experience with CT systems. So, lets start by looking at a simple DT system.
Let x[n] and y[n] denote, respectively, the input and the output to a DT system. Let the
input/output (I/O) relationship for this DT system be described by the I/O equation
y[n] = 0.9y[n 1] + x[n] .
(10)
That is, the output at time n is the input at time n plus the previous output multiplied by
0.9. Figure 6 illustrates this DT system in block diagram form.
adder
x[n]
y[n]
+
+
D
delay
0.9
.9 y[n1]
y[n1]
multiplier
Figure 6: A simple DT system.

We now calculate the output for several cases, and make an important observation about
systems.
2
Try showing this yourself. We will formalize approaches to computing outputs to FIR lters later in this
Course.
10
a) Find the impulse response, which is denoted below as ya [n]. That is, nd the output
to the impulse input
1
n=0
xa [n] =
(11)
0
n=0
when the initial condition is ya [1] = 0.
Solution: use either Eq (10) or Figure (6).
b) Find the step response, which is denoted below as yb [n]. That is, nd the response to
the step
1
n0
xb [n] =
(12)
0
n<0
when the initial condition is yb [1] = 0.
Solution: This isnt too bad. Well learn formal approaches for doing this later in the
Course. For now, heres the answer. Verify and sketch it yourself.
yb [n] =
Sketch:
n
k
k=0 0.9
10.9n+1
10.9
n0
n<0
(13)
11
c) Find the response yc [n] to initial condition yc [1] = 1 when there is no input.
Solution: Try this one yourself. Its pretty straightforward. The answer is
yc [n] =
0.9n+1
0
n 1
n < 1
(14)
Note that this is just ya [n] shifted to the right (i.e. advanced in time) by one sample.
Sketch:
d) Find the response yd [n] to input

xd [n] = 3xa [n] 4xb [n]
(15)
when the initial condition is yd [1] = 5.

Solution: Even though the system is very simple, and the input is not too complex, this
is already starting to get challenging. Fortunately, for reasons identied later in the
course, this system has a very useful property called linearity (a.k.a. superposition).
Thus,
yd [n] = 3ya [n] 4yb [n] + 5yc [n] .
(16)
An observation: Linearity is a useful property! Are there other useful system properties?
1.3
Course Objective, Linear Combinations & Two Basic Concepts
Course Objective: to introduce the theory, methods and applications of signal processing
and system analysis. This theory and these methods eectively decompose complex problems
into simpler ones. Case d) of the Subsection 1.2.5 example illustrates this. The principal
approach is to decompose signals into simpler ones using transforms.
Linear Combinations (a.k.a weighted sums) and signal expansions: as an example,
we will describe this with CT signals. Its the same idea for DT signals. Let
xi (t) ;
i = 1, 2, , N
(17)
12
be some set of N basic (i.e. simple) signals. We say that the xi (t) is a discrete or countable
set since there are integer N of them. Let
i = 1, 2, , N
ci ;
(18)
be any N constants (a.k.a. weights; multipliers). A linear combination of the xi (t) signals is
N
x(t) =
ci xi (t) .
(19)
i=1
The Matlab demo on the following two pages illustrates this idea of representing a general
signal as a linear combination of simple signals.
We consider Eq (19) to be a decomposition or expansion of the signal x(t) as a linear
combination of basic signals3 . With transforms, we use a set of basic signals, called basis
signals, that allow us to exactly expand any signal of interest as shown in Eq (19). The
transform tells us how to identify the weights of this expansion for a given signal x(t).
The same idea applies for the continuum of signals and constants. Let x (t) be a set
of signals parameterized by the continuous-valued variable . A good example of this is
x (t) = ejt , were is the angular frequency of complex sinusoid x (t). Consider the set of
signals
x (t) ;
W1 W2
(20)
and corresponding constants
W1 W2 .
c() ;
(21)
Since c() has a value for each over the continuum W1 W2 , it should be considered
a function of . A linear combination of the x (t) signals is
x(t) =
W2
W1
c() x (t) d .
(22)
In this equation, we should interpret the integral as a weighted sum of signals dened over
a continuum. Note that x(t) is not a function of since we are integrating over .
So a linear combination of signals can be a sum over a discrete set of signals or an integral
(continuous sum) over a continuum of signals. Make sure you understand this notation, and
what it suggests. This underlying idea (linear combinations of signals) will be central to
what we consider throughout much of this Course. We will continue our discussion of this
general idea a little later, in Subsection 1.4.5.
Two Basic Concepts:
i) Signal Representation & Analysis: General (classes of) signals can be represented as
or decompose into linear combinations of a set of basic signals.
ii) System Design & Analysis: Many systems can be described & analyzed in terms of
their responses to a set of basic signals.
3
If you have taken a course on linear algebra, then note that this is the same idea as representing a
vector as a linear combination of basis vectors. In fact, expansion of a signal as a linear combination of basis
signals is the fundamental concept that is common to both the transforms considered in the Course and the
representation of vectors in terms of basis vectors.
echo on
%
%
Matlab Demo
%
Representing a CT Square Wave as a Linear Combination of Sinusoids
%
%
x(t) == unit magnitude square wave of period 1, w_0 = 2 \pi
%
%
x(t) =? 0.5 + 1/pi cos(2*pi*t) - 1/(3pi) cos(6*pi*t) +
%
1/(5pi) cos(10*pi*t) - 1/(7pi) cos(14*pi*t) + ...
%
%
Enter N (odd) == highest harmonic before running
%
pause
%
Construct and plot samples of x(t) for -1 <= t <= 1.
%
t = -1:.005:1;
x=ones(1,401);
x(51:151) = zeros(1,101), x(251:351) = zeros(1,101);
pause
subplot(221)
plot(t,x),xlabel(t),ylabel(amplitude),title(square wave)
text(0.5,1.5,Thomas Friedman),axis([-1 1 -0.5 1.5])
pause
%
Generate array of required CTFS coefficients
%
Ak = zeros(1,N+1);
Ak(1) = 0.5;
% DC
kk = 1:N;
Ak(2:N+1) = sin(kk*pi/2)./(kk*pi);
%
%
Approximate x(t) with harmonics 1,3
%
omega0 = 2*pi;
x1 = Ak(1)*ones(1,length(t));
for k=1:3
x1 = x1 + 2*abs(Ak(k+1)*cos(k*omega0*t+angle(Ak(k+1)));
end
pause
subplot(222)
plot(t,x1,b),xlabel(t),ylabel(amplitude),title(1,3 harmonics)
text(0.5,1.5,Alan Oppenheim),axis([-1 1 -0.5 1.5])
pause
%
Approximate x(t) with harmonics 0,1,3,5,7,9,11
%
13
14
x2 = x1;
for k=4:11;
end
pause
subplot(223)
plot(t,x2,r),xlabel(t),ylabel(amplitude),title(0,1,3,5,7,9,11 harmonics)
text(0.5,1.5,Khaled Hosseini),axis([-1 1 -0.5 1.5])
pause
%
Approximate x(t) with harmonics 0,1,3, ... , N (odd)
%
x3 = x2;
for k=12:N;
end
pause
subplot(224)
plot(t,x3,m),xlabel(t),ylabel(amplitude),title(0,1,3,..,N harmonics)
text(0.5,1.5,Robert Pirsig, Ayn Rand),axis([-1 1 -0.5 1.5])
square wave
1,3 harmonics
1.5
1.5
Thomas Friedman
Alan Oppenheim
1
amplitude
amplitude
0.5
0.5
1
0.5
0.5
0
t
0.5
0.5
1
1,3,5,7,9,11 harmonics
0.5
1.5
Khaled Hosseini
Robert Pirsig, Ayn Rand

1
amplitude
1
amplitude
0
t
1,3,..,N harmonics
1.5
0.5
0.5
1
0.5
0.5
0.5
0
t
0.5
0.5
1
0.5
0
t
0.5
Figure 7: Representation of a CT square wave as a linear combination of sinusoids. The idea

of harmonics will be established in Chapter 3 of this Course.
1.4
15
Discrete & Continuous Time Signals & Operators
The purpose of this Section is to establish some basic notation and concepts that we will
rely heavily on throughout the Course.
1.4.1
Basic Discrete-Time Signals & Operators
Examples of Signals:
Impulse (or unit impulse)
[n] =
1
0
n=0
n=0
(23)
u[n] =
1
0
n0
n<0
(24)
r[n] =
n
0
n0
n<0
(25)
Step
Ramp
Pulse (of length N)

1
0
pN [n] =
n = 0, 1, , N 1
otherwise
(26)
Plot: impulse
step
ramp
pulse
Exponential: For complex constants c and , consider

x[n] = c n
(27)
Let c = A ej and = r ejo . Then

x[n] = A ej
r ejo
= A ej r n ejo n = A r n ej(o n+)
= A r n cos(o n + ) + jA r n sin(o n + ) .
Note that r and determines the decay and oscillation rates, respectively. One form
of Eulers identity is used in going from the rst to second line of the equation above.
We will see much more on complex exponentials and Eulers identities later, so review
exponential algebra and Eulers identities now as needed.
16
Eulers Identities: relating real-valued and complex-valued sinusoids

ejx = cos x + j sin x
1 jx
1 jx
e +
e
cos x =
2
2
1 jx
1 jx
sin x =
e
e
.
2j
2j
(28)
(29)
(30)
Real & complex valued DT sinusoids: much more on these later.

Basic Operators: The following operations on signals could be considered as very simple
systems, but they are so elementary we introduce them here, before any formal discussion
of systems.
Shift (i.e. delay): A delay shifts a signal in time. Mathematically, we replace the
discrete-time variable n with n m, where m is the (constant) amount of shift.
Example 1.1: Consider the step u[n], plot x[n] = u[n m] for m = 4.
We have established that
n0
n<0
1
0
u[n] =
So,
nm 0
nm< 0
1
0
u[n m] =
or
nm
n<m
1
0
u[n m] =
u[n4]
1
.....
1
Figure 8: A delayed (a.k.a. shifted) step.

Note, as shown in Figure 8, that m = 4 shifts the signal to the right (by
4). This is a positive delay. m < 0 shifts the signal to the left. This is a
negative delay.
17
Fold: A fold is a reection of a signal about n = 0. Mathematically, we replace the

discrete-time variable n with n.
Example 1.2: Plot x[n] = u[n].
Solution:
Example 1.3: Plot y[n] = x[n], where

x[n] =
0.5n
0
n0
n<0
Solution:
y[n] =
0.5n
0
n 0
n < 0
2n
0
n0
n>0
Plot:
This x[n] is called a right-sided signal since it is zero as n and not

zero as n . y[n] is a left-sided signal.
18
Fold & Shift: First we fold the signal, then shift. Mathematically, we replace the
discrete-time variable n with m n where m is the amount of shift.
Example 1.4: Plot x[n] = u[m n] for m = 5.
mn0
mn<0
1
0
y[n] =
n0
n<0
1
0
x[n] =
.
1
0
nm
n>m
u[5n]
1
.....
1
Figure 9: A folded and shifted step.

Note, as shown in gure above, that after the fold we shift to the right (by
5). This is because m is positive. For negative m, after the fold we would
shift to the left.
Later, when studying linear time-invariant (LTI) systems, we will learn about
convolution. Convolution involves the fold & shift operation, so it is important to get used to it now!
Addition and Multiplication: These operators are shown below:
v[n] = w[n] s[n]
(31)
z[n] = x[n] + y[n]
(32)
Each operates on two signals sample-by-sample.

Example 1.5 - Multiplication: Plot x[n] = .5n u[n].
n
.5 u[n]
1
.....
2 1
Figure 10: A right-sided decaying exponential.

Note that the step eectively turns the exponential on and o. Notationally,
this is a common use of a step function.
19
Example 1.6 - Addition: Plot pN [n] = u[n] u[n N].

p [n]
u[nN]
u[n]
.....
2 1
N2
.....
N+2
2 1
.....
N1 N
Figure 11: A pulse formed from two steps.

The step is added to the negative of a delayed step, forming a pulse.
Example 1.7: Plot x[n] = an pN [n]. (Assume 0 < a < 1).
n
a p N [n]
1
.....
2 1
N1 N
Figure 12: An exponential pulse.

The pulse eectively turns the exponential on and o.
Example 1.8: Given two signals x[n] = .5n u[n] and h[n] = u[n], plot
y[n] = x[n] h[4 n].
h[n]
h[4n]
....
....
n
y[n]
x[n]
....
0
Figure 13: y[n] = x[n] h[4 n].

Example 1.8 takes us one step closer convolution the LTI system operation
covered in Chapter 2 of this Course. That is, there is a fold, a shift and a
multiplication. As we will see, all that is left for convolution will be an add.
Note that y[n] = .5n p5 [n].
20
Accumulator: Let x[n] be a DT signal. At any time n, an accumulator sums all inputs
up to time n to form the output y[n] at time n. Thus an input/output (I/O) equation
for an accumulator is
n
y[n] =
x[k] =
k=
x[n k] .
k=0
(33)
Think of n as representing current time and k as memory time (i.e. all time up to
n). At any time n we sum all x[n] up to time n. The variable k is sometimes called a
dummy variable. An alternative I/O expression for an accumulator is
y[n] = y[n 1] + x[n] .
(34)
To convince yourself that Eq (34) does the job, plug y[n 1] = y[n 2] + x[n 1] into
Eq (33), then y[n 2] = y[n 3] + x[n 2], etc.. By induction, Eq (34) describes the
I/O of an accumulator. Figure 14 illustrates an accumulator operating on and input
x[n].
x[n]
y[n]
Accumulator
x[n]
y[n]
D
y[n1]
Figure 14: Block diagram of an accumulator.

Although both Eq (33) and Eq (34) describe the I/O of an accumulator, they suggest
dierent implementations. Would you prefer to compute y[n] using Eq (33) or Eq (34)?
In general, there can be many (literally innite) ways to implement a desired I/O
function. For a given processing platform, this gives us alternative implementations
to choose from. So we can select that which in some sense best uses our resources. In
this Course we will develop tools that can be used to derive alternative I/O equations
for a given I/O function.
Example 1.9: For the accumulator, let the input be x[n] = an u[n]. Determine
y[n] for all n. Make use of the Geometric series equations the rst three
entries in Table 1 on the following page.
Solution:
21
We will be performing summations throughout the Course, since this is what DT systems
and transforms typically do. So become familiar with Table 1.
Note that summations are like integrals. How do you do an integral? Well, although you
may use a few properties and rules to convert a required integral into one that you know,
in the end you use an integral you know. If you have a large table of integrals, then you
know a lot of integrals. So, how do you do a summation? If possible, you use a table (and
perhaps a few properties and rules). A more interesting question from and engineering point
of view is: why and how do we use summations to solve signal processing problems?
Table 1: A Summation Table.
N2
aN1 aN2 +1
(Geometric Series)4 ;
1a
k=N1
n1
1 an
ak =
(special case of GS)
1a
k=0
1
(special case of GS; |a| < 1)
ak =
1a
k=0
n
n(n + 1)
k =
2
k=1
n
n(n + 1)(2n + 1)
k2 =
6
k=1
n
2
n (n + 1)2
k3 =
4
k=1
ak =
k=1
n1
(2k 1) = n2
a(1 nan1 + (n 1)an )

(1 a)2
k=1
a
k ak =
(1 a)2
k=1
n1
2nan+1
n2 an
(1 an )(a2 + a)
2 k
+
+
k a =
(1 a)3
(1 a)2 (1 a)
k=1
2
(a + a)
k 2 ak =
(1 a)3
k=1
1
2
=
2
6
k=1 k
k ak =
(|a| < 1)
(|a| < 1)
[4] For N2 = |a| < 1 must hold. For N1 = , |a| > 1 must hold.
22
1.4.2
Basic Continuous-Time Signals & Operators
Examples of Signals:
Step: See Figure 15(a).
u(t) =
1
0
t0
t<0
(35)
r(t) =
t
0
t0
t<0
(36)
Ramp: See Figure 15(b).
Pulse: See Figure 15(c).
|t| T
2
otherwise
1
0
pT (t) =
(37)
Note that, as opposed to a DT pulse that starts at time n = 0, as dened here a CT

pulse is centered at time t = 0.
Impulse: an impulse is denoted (t) and dened by the following properties:
* (t) = 0 for t = 0.
* For any small positive constant ,
area).
/2
/2
(t) dt = 1 (i.e. an impulse has unit
Figure 15(d) shows the standard graphical representation of an impulse. Note that
(t) can be dened as (t) = lim0 1 p (t).
u(t)
r(t)
area
....
....
(a)
(1)
(t)
p (t)
(b)
T/2
(c)
t
0
T/2
(d)
Figure 15: Basic CT signals: (a) step; (b) ramp; (c) pulse; (d) impulse.
Exponentials: Given complex constants c = A ej and s = + j0 , a CT complex
exponential is of the form
x(t) = c est = A ej e(+j)t
= A et ej(0 t+) .
(38)
23
Basic Operators:
Shift (delay): x(t) x(t )
Fold: x(t) x(t)
Fold and Shift: x(t) x( t)
Multiplication & Addition: (point by point in time)
Example 1.10: Let x(t) = e.5t u(t). Plot y(t) = x(4 t).
Solution:
Dierentiation and Integration: Dierentiation (i.e. slope calculation) is as dened in

your calculus course. Integration, dened as
y(t) =
x( ) d =
x(t ) d ,
(39)
is analogous to the accumulator for DT signals for any time t you accumulate the
input up to that time.
Example 1.11: Dierentiate the scaled step. Integrate the scaled impulse.
Solution:
d
c u(t) = c (t)
dt
The slope is zero everywhere. The derivative at the discontinuity results in
an impulse (innite slope) with area equal to the height of the discontinuity.
t
c ( ) d = c u(t)
For t < 0 we are not integrating across the impulse. For t > 0 we are
integrating across the scaled impulse, which has area c.
t
c
t
(c)
(c)
( ) d
d
dt
Figure 16: Dierentiation and integration involving impulses.
c
t
24
Time scale: For positive constant a, x(t) x(at).
for 0 < a < 1, its a signal expansion
for a > 1, its a signal Contraction
Example 1.12: Let x(t) = et u(t). Then x(5t) = e5t u(5t) = e5t u(t).
x(t)
x(5t)
Figure 17: Compression (scaling with a > 1).

Example 1.13: Let x(t) = p1 (t). Then x(0.5t) = p1 (0.5t) = p2 (t).
x(0.5t)
x(t)
1
0.5
0.5
Figure 18: Expansion (scaling with 0 < a < 1).

Symmetry: This discussion applies to both CT and DT signals. It is presented in terms of
CT signals. The following are denitions of even & odd (symmetric) signals:
even
odd
x(t) = x(t)
x(t) = x(t) .
Any signal can be decomposed into its even & odd components as follows:
1
1
x(t) = xe (t) + xo (t);
xe (t) =
[x(t) + x(t)]
xo (t) =
[x(t) x(t)]
2
2
A complex-valued signal is complex symmetric if its real part is even symmetric and its
imaginary part is odd symmetric. That is, given a signal x(t) = xr (t) + j xi (t) where realvalued xr (t) and xi (t) are the real and imaginary parts respectively, it is complex symmetric
if xr (t) = xr (t) and xi (t) = xi (t). If this is the case, then x(t) = x (t), |x(t)| = |x(t)|
and x(t) = x(t).
Example 1.14: Is the signal x(t) = ej0 t complex symmetric?
Solution: Using Eulers identity, x(t) = xr (t) + j xi (t) where xr (t) = cos(o t) and
xi (t) = sin(o t). cos(o t) and sin(o t) are, respectively, even and odd functions.
Thus x(t) is complex symmetric. Note that |x(t)| = x2 (t) + x2 (t) is even since
i
r
cos2 (o t) and sin2 (o t) are. Also, since tan1 is an odd function,
x(t) = tan1
sin(0 (t))
cos(0 (t))
= tan1
That is, x(t) is an odd function.
sin(0 t)
cos(0 t)
= tan1
sin(0 t)
cos(0 t)
= x(t) .

1.4.3
25
Signal Classes
A signal class is a set of signals sharing a common characteristic of interest. As a topic,

signal classes is important because it dierentiates the dierent transforms we will learn in
this course. That is, all transforms do basically the same thing break complex signals into
simpler ones. We use dierent transforms on dierent classes of signals. Below we dene the
following signal classes: DT, CT, energy, power and periodic.
DT vs. CT Signals: As already established, DT signals are a function of a discretevalued (i.e. integer-valued) parameter, say n. The set of all possible DT signals forms
the class of DT signals. It should be noted that although in this Course we refer to
these signals at discrete time, the important characteristic is that the independent
parameter is discrete-valued. So, for example, in Practicums 1 & 2 we process rows
of an image the same way we process a DT signal, because the positions in a row are
indexed as a discrete-valued variable.
Similarly, we refer to signals that are a function of a continuous-valued variable, say t,
as CT signals. The set of all possible CT signals forms the class of CT signals.
Energy Signals: For DT and CT signals, respectively, energy is dened as
E =
|x[n]|2
(40)
|x(t)|2 dt
(41)
n=
E =
An energy signal is a signal which has nite energy, i.e. E < . The class of DT
energy signals, for example, is the set of all possible DT signals with E < .
Example 1.15: Is x[n] = .5n u[n + 3] an energy signal? Using the geometric
series,
E =
=
(0.5n )2 =
n=3
3
n=3
3
0.52n =
n=3
(0.52 )n =
0.25n
n=3
0.25 0
4
1
=
= 85 .
1 0.25
0.75
3
Note that x[n] has innite duration, yet it has nite energy. So some innite
duration signals have nite energy, but of course some dont.
Question: Again, how in general does someone do a summation like the one
required in Example 1.15? How do you do integrals?
26
Example 1.16: Is x(t) = cos(10t) an energy signals? What is its energy in

one period?
Since x2 (t) has innite area, i.e. since clearly
E =
cos2 (10t) dt = ,
x(t) is not an energy signal. Note that its energy over one period is
E1
period
=
=
1/5
0
1/5
0
1 1
+ cos(20t)
2 2
0
1
1
cos(20t) dt =
.
2
10
cos2 (10t) dt =
1
dt +
2
1/5
0
1/5
dt
Power Signals: For DT and CT signals, respectively, power is dened as

P = lim
N
1
|x[n]|2
2N + 1 n=N
(42)
1 T
|x(t)|2 dt
(43)
T
2T T
That is, the power is the average energy over all time. A power signal is a signal which
has nonzero but nite power, i.e. 0 < P < . The class of CT power signals, for
example, is the set of all possible CT signals with 0 < P < .
P = lim
The limits in Eqs (42,43) can be dicult to compute. Note, however, that in this
Course we will be interested in only a subclass of power signals periodic signals.
We will talk more about periodic signals a little later. For now note that for periodic
signals with period N (for DT) or T (for CT), power can be computed, respectively,
as follows:
1 N 1
P =
|x[n]|2
(44)
N n=0
1 T
|x(t)|2 dt .
(45)
T 0
That is, the power is the average energy over a period. Note that average energy can
be computed over any period.
P =
Example 1.17: Is x(t) = et u(t) a power signal?

Solution:
An energy signal has zero power, and a power signal has innite energy.
27
k=
Example 1.18: Is x(t) =
p2 (t 4k) a power signal?
Solution:
x(t)
....
....
4
Figure 19: A periodic pulse train.
P =
1
4
2
2
x2 (t) dt =
1
4
1
1
dt =
1
.
2
This signal has nite power (and thus innite energy). It is a power signal.
Example 1.19: What is the power of the signal x[n] = 5?
Solution:
P = 52 = 25 .
The average energy of a constant signal is just the square of its amplitude.
Example 1.20: Determine the power of the signal x[n] =
where x1 [n] = (0.9)n p10 [n].
Solution:
k=
x1 [n 10k]
28
1.4.4
Periodic Signals and Sinusoids
General Denitions: First consider CT signals. Let T be a positive, real number.

The CT signal x(t) is periodic if for all t and some T
x(t + T ) = x(t) .
(46)
The fundamental period T0 is the smallest T such that this equation holds. The fundamental frequency, in radians/second, is 0 = 2 .
T0
For DT signals, Let N be a positive integer. The DT signal x[n] is periodic if for all n
and some N
x[n + N] = x[n] .
(47)
The fundamental period N0 is the smallest N such that this equation holds. The
2
fundamental frequency, in radians/sample, is 0 = N0 .
CT Sinusoids: A real-valued CT sinusoid has the form:
x(t) = A cos(0 t + )
(48)
where A is the amplitude, 0 is the frequency (in radians/sec.), and is the phase (in
radians). It is easy to show (using trigonometric identities or by plotting) that this
2
signal is periodic with fundamental period T0 = 0 . So the fundamental frequency, in
1
Hz, is f0 = 0 = T0 .
2
Time shift: Consider the sinusoid x(t) = A cos(0 t) and its delay y(t) = A cos(0 (t
0 )) illustrated in Figure 20.
y(t)
A
....
....
0
T0
Figure 20: A delayed (therefore phase shifted) cosine signal.

We have that
y(t) = A cos(0 t 0 0 ) = A cos(0 t + 0 )
(49)
where 0 = 0 0 . The point is that a delay of 0 corresponds to a phase shift of

0 0 . The amount of phase shift is proportional to both the frequency and the delay.
29
A complex-valued sinusoid is a special case of the CT exponential signal x(t) = c et ej(0 t+)
introduced is Subsection 1.4.2 of these Notes. For the general CT exponential of Subsection 1.4.2, let c = A and = 0. We then have
x(t) = A ej(0 t+)
(50)
As in Subsection 1.4.2, A is the amplitude, 0 the frequency (in radians/second), and

is the phase (in radians).
Is this signal periodic? Sure, let T =
2
.
0
Then
x(t+ T ) = A ej(0 (t+T )+) = A ej(0 t++2) = A ej(0 t+) ej2 = A ej(0 t+) . (51)
This proves it is periodic (i.e. that x(t + T ) = x(t) for all t). The fundamental period
2
is T0 = 0 .
DT Sinusoids: Real-valued DT sinusoids are of the form

x[n] = A cos(0 n + ) .
(52)
Complex-valued DT sinusoids are of the form

x[n] = A ej(0 n+) .
(53)
A is the amplitude, 0 the frequency (now in radians/sample), and is the phase (in
radians).
Example 1.21: Let x[n] = ej0.5n . Find the fundamental period N0 .
Solution:
x[n + N0 ] = ej0.5(n+N0 ) = ej0.5n ej0.5N0 .
For N0 = 4, ej0.5N0 = ej2 = 1. Thus, for N0 = 4,
x[n + N0 ] = ej0.5n = x[n]
n .
N0 = 4 is the smallest positive integer such that this is true. So, x[n] is
periodic (of course?) with fundamental period N0 = 4.
30
Example 1.22: Let x[n] = ej0.5n . Find the fundamental period N0 .

Solution:
x[n + N0 ] = ej0.5n ej0.5N0 .
Note that ej0.5N0 = 1 for any integer N0 . Thus
x[n + N0 ] = ej0.5n = x[n]
n .
This DT sinusoid is not periodic!!

Unlike CT sinusoids, not all DT sinusoids are periodic!! To be periodic, the frequency
0
must be rational. (This artifact of DT sinusoids shows up from time to time, but it
2
is not a problem in signal processing.)
Ambiguity of DT Sinusoids: We now explore another way in which DT sinusoids
are dierent than CT sinusoids. This dierence is critically important. It holds the
key to understanding sampling CT signals and its limitations.
Consider two DT complex sinusoids, x[n] = ej0 n and y[n] = ej(0 +2)n .
y[n] = ej0 n ej2n = x[n]
(54)
Point by point, they are the same signal!! Sinusoids of frequencies 0 and 0 + 2 are
ambiguous since we can not tell one from the other. Extending this, for any 0 , the set
of frequencies {0 + k2; k = 0, 1, 2, } are ambiguous. Thus, when considering
DT signals, we need only consider frequencies over a range of 2, say in the range
<
(55)
It can not be over emphasized how important this frequency ambiguity phenomenon is
to the understanding of processing DT signals.
Sums of Periodic Signals: This discussion applies to both CT and DT signals. We
illustrate the point using CT. Consider two periodic signals, and their sum
x1 (t) = x1 (t + T1 )
x2 (t) = x2 (t + T2 )
x(t) = x1 (t) + x2 (t) .
The sum x(t) is periodic if and only if, for some integers N and M,
N T1 = M T2
(56)
For the smallest integers N and M such that this is true, the fundamental period of
x(t) is
T0 = N T1 = M T2 .
(57)
31
An Illustration:
x1 (t)
....
....
T1
x2 (t)
T 0 = 3 T1 = 2 T2
....
....
t
T2
Figure 21: Two periodic signals (to be summed).

Example 1.23: Given
x(t) = 6 cos(10t) + 3 sin(14t) ,
determine the fundamental frequency of x(t).
1
1
Solution: Note that with 1 = 10, T1 = 5 . Similarly, 2 = 14, T2 = 7 . So
that N T1 = M T2 = T0 ,
N
T2
1/7
5
=
=
=
.
M
T1
1/5
7
So, T0 = 5 T1 = 7 T2 = 1 .
Sums of Same-Frequency Sinusoids: This discussion applies to both CT and DT

signals. We illustrate the point using DT signals.
The objective is to simplify into a single sinusoid the sum of sinusoids of the same
frequency but dierent amplitudes and phases. Specically, starting with
N
Ak cos(o n + k ) ,
(58)
k=1
the objective is to write it in the form B cos(o n + ).

N
Ak cos(on + k ) =
k=1
k=1
Re{Ak ej(o n+k ) } =
k=1
Re{Ak ejk ejo n }
= Re{
{Ak ejk
k=1
Let B ej =
N
k=1
ejo n } .
(59)
Ak ejk . Then
N
k=1
Ak cos(o n + k ) = Re{B ej ejo n } = Re{B ej(o n+) } = B cos(0 n + ) . (60)
32
1.4.5
Basic Signal Sets and Transforms
In Section 1.3 we introduced the general concept of representing a signal as a linear combination of basic signals. Now that we have introduced a number of basic signals, we can
describe the main topics of this Course in terms of this general concept.
The main topics of this Course are the entries of the third column of Table 2. The purpose
of this table is to show that each of these topics can be interpreted in terms of a process of
representing a general class of signals (listed in the second column) as a linear combination
(i.e. a weighted sum) of a set of basic signals (listed in the rst column). The rst four rows
represent CT signals and systems topics. The last four rows represent DT topics. For both
the CT and the DT entries, the rst row leads to a very useful time domain approach to
systems analysis called convolution. This is the topic of Chapter 2 of this Course. The last
three rows of both the CT and the DT entries of Table 2 represent transforms. Note that the
majority of these transforms concern the representation of signals as linear combinations of
complex-valued sinusoids. These transforms, which facilitate transform domain approaches
to signal & system design & analysis, will be covered in Chapters 3 & 5 of this Course.
Table 2: Basic signal sets and corresponding transforms.
Signal Set
Class Represents
Transform
{(t ) ; }
all CT signals
none, but leads to convolution
{ejk0 t ; k = 0, 1, 2, }
0 = 2
T
all periodic CT signals

with period T
CT Fourier Series (CTFS)
{ejt ; }
all CT energy signals
CT Fourier Transform (CTFT)
{est ; all complex s }
all CT signals
Laplace Transform (LT)
{[n k] ; k = 0, 1, 2, }
all DT signals
none, but leads to convolution
{ejk0n ; k = 0, 1, , N 1}
0 = 2
N
all periodic DT signals

with period N
DT Fourier Series (DTFS)
{ejn ; }
all DT energy signals
DT Fourier Transform (DTFT)
{z n ; all complex z }
all DT signals
z-Transform (ZT)
33
Consider, for example, entry #3 of Table 2 the Continuous Time Fourier Transform
(CTFT). This entry suggests that the CTFT can be employed to represent any CT energy
signal as a linear combination of the CT complex-valued sinusoids ejt were we use all
frequency over the continuum of range . When we study the CTFT we
will see specically how to do this. Generally, we need to identify the weighting on each
CT complex-valued sinusoid that is required to represent a CT energy signal x(t) under
consideration. For this x(t), denote this weighting function as X(). X() is a function
of frequency since we need to weight each sinusoid dierently in order to represent the
given x(t). To represent an x(t), Entry #3 of Table 2 indicates that in general we will need
to identify X() over the entire range . Then, following the idea of linear
combinations proposed in Subsection 1.3, Entry #3 of Table 2 suggests that we can construct
this x(t) from CT complex-valued sinusoids as follows:
x(t) =
X() ejt d .
(61)
In this equation, the ejt are the basic signals being used to represent or construct the general
signal x(t). X() is the function that describes the weighting applied to each basic signal.
The integral does the combining (summing) of the weighted basic signals, and implies that
we need basic signals dened over a continuum (of frequency in this case). The integral
bounds indicate that the continuum goes from to (in this case, in frequency).
In this Course we will establish how to determine the weighting functions for these transforms. This is relatively easy once you are comfortable with this general idea of a transform.
We will also consider how this basic idea, and these specic transforms, are employed for a
wide range of signal and system design and analysis problems. At this point consider, as an
example, that for those transforms that use complex-valued sinusoids as the basic signals,
the weighting function tells us the frequency content of the signal. That is, computing the
weighting function is equivalent to performing spectrum estimation (i.e. frequency analysis).
Example 1.24: You are interested in the amount of base and treble in a score of
music you have recorded on an analog tape. Specically, of the transforms listed
in Table 2, which would you use to determine the frequency content of a one
second segment of your music signal.
Solution: On the one hand, your music is stored as a CT signal, and one second
of music will be an energy signal. This suggests that the CTFT (Entry #3 of
Table 2) could be used.
On the other hand, if you have an A/D converter that you can apply to your
stored music (e.g. a sound card on your laptop), and you wish to process your
signal digitally (e.g. with Matlab), then you will be working with a DT energy
signal. So perhaps it is the DTFT (Entry # 7 of Table 2) that you are interested
in.
34
1.5
Linear Time-Invariant (LTI) Systems
The term system can mean a lot of dierent things. As considered in this Course, a system
operates on an input signal to generate an output signal. The system input/output (I/O)
characterization is a description of how the system maps inputs to outputs. Although in
general there are other important system considerations for example how they are implemented/applied and what happens internal to a system in this Course we will focus on I/O
characterizations. However, we will be implementing a number of DT systems in Matlab
and discuss applications from time to time in class.
We begin this Subsection by briey considering a number of common DT and CT systems.
We then identify several important system characteristics. Linearity, which we have already
considered informally, is one of these. Time-invariance is another. We will conclude by
introducing one very common type of system - the linear, time-invariant (LTI) system.
1.5.1
System Examples
Since I/O characterization of systems is a central theme of this Course, with each system
example below some aspect of I/O is discussed.
An RC Circuit: Consider the RC circuit discussed in Subsection 1.2.2 above. An I/O
t
1
equation for it, which is equivalent to Eq (5), is y(t) = C e(t )/RC x( ) d .
Example 1.25: For this RC circuit, determine an expression for its output for
complex sinusoidal input x(t) = ej0 t .
Solution:
35
Simple DT Feedback System

In Subsection 1.1.2 and in Practicum 1 we considered a DT system which fed back the
present output through a multiplier to compute the next output. In Figure 22 we generalize
this system by using a arbitrary multiplier with value a. In equation form, the I/O is
y[n] = a y[n 1] + x[n] .
(62)
adder
x[n]
y[n]
+
+
D
delay
a
a y[n1]
y[n1]
multiplier
Figure 22: A simple DT system.

One thing we did in Subsection 1.1.2 was to compute the impulse response. (i.e. the
response due to x[n] = [n]). We will see that the impulse response of a system is a very
useful characterization of the system so much so that we give it its own notation. In this
Course we will follow convention by denoting the impulse response of a DT system h[n] (we
use h(t) for a CT system). The impulse response of the system in Figure 22 is
h[n] = an u[n] .
(63)
We will learn to do a lot with the impulse response, both to compute system outputs to
general inputs, and to study system properties. For now note that if |a| > 1, the impulse
response magnitude gets bigger as n increases i.e. it blows up. A system whose output
blows up for any expected or reasonable input should be avoided.
N Point Averagers and Finite Impulse Response (FIR) Filters
An N point averager is a DT system that averages the N most recent inputs to produce
the current output. That is,
y[n] =
N 1
k0
1
1
x[n k] =
N
N
N 1
k0
x[n k] .
(64)
Averagers are used quite a bit in signal processing, for example to smooth signals. One
attractive computational aspect of this system is that it requires only one multiply per
1
output. Although we apply the N multiplier once per output to the sum of the N most
recent inputs, eectively we are applying the multiplier to each input used to compute the
output (i.e. to the current and past N 1 inputs). Also note that, in Eq (64), both n and
k represent time n output time, and k memory time.
36
More generally, a dierent multiplies can be applied to the dierent inputs. That is, we
can consider the system
y[n] =
N 1
bk x[n k] ,
k0
(65)
where the bk ; k = 0, 1, 2, , N 1 are the multiplier (a.k.a. coecient) values. If bk =

1
; k = 0, 1, 2, , N 1, then this is an N point averager. Generally, we can choose the
N
bk s the control I/O characteristics. In this Course we will learn how to evaluate these I/O
characteristics for a given set of bk coecients. In the senior level DSP Course, ece5790,
we learn how to design this type of system (i.e. how to choose the coecients to realize
specied I/O characteristics). This type of DT system used in a Subsection 1.1.3 example
and in Practicum 1 to model a multipath communications channel and to equalize a channel.
Figure 23 illustrates this general DT system. Letting x[n] = [n], the output is y[n] =
h[n], the impulse response. Just plugging x[n] = [n] into Figure 23, it is easy to see that
h[n] =
bn
0
n = 0, 1, 2, , N 1
otherwise
(66)
The impulse response is nite length. Thus, this very common type of DT system is referred
to as an FIR (nite impulse response) lter. The terms lter and system are often used
interchangeably since systems are commonly used to separate a signal of interest from noise
superimposed on it.
x[n]
D
b0
D
b1
D
b2
D
b N2
b N1
y[n]
Figure 23: The FIR lter structure.

In Practicum 1 you develop Matlab code which will implement a general FIR lter. That
is, you will build an FIR lter in Matlab software. Incomplete code for this is provided in
the Practicum 1 write-up (in Section 1.6 of these Notes).
Example 1.26: Consider the 10-point summer with I/O equation
9
y[n] =
k=0
x[n k] .
Recall that here n represents output time and k represents memory time.

a) What is the impulse response, h[n], for this system? Rewrite the I/O equation given above using the general h[n] notation.
b) Sketch this FIR lter twice, once using the general h[n] notation and once
for the specic h[n]. For general time n, label each delay line node with the
correct delayed x[n].
c) Visualize the input shifted through this system. To do this plot the impulse
response as a function of k (i.e. plot h[k]). Now, sketch some input x[n].
For this input, directly below your h[k] plot, sketch x[n k] as a function
of k for n = 0. Now visualize how x[n k] shifts by h[k] as n varies.
d) Using your part c) visualization, write down the expression for y[n] as a sum
of the products, over k, of the samples in your h[k] and x[n k] plots.
e) Use the visualization you developed in part c) to determine the output of

this 10-point summer for input x[n] = p20 [n].
Solution:
37
38
Concerning part b), note that h[n] depicts the memory structure of this system
(i.e. it shows how the system operates on each input in the system memory to
form the output).
Concerning part c), the data in your x[n k]; n = 0 plot over the range k =
0, 1, , 9 is the data in the FIR lter delay line at time n = 0. This data lines
up with the FIR lter multipliers plotted in h[k]. In terms of the basic signal
operators discussed in Subsection 1.4.1, as a function of memory time k, x[n k]
is folded and shifted by h[k]. That is, the input is folded and then shifted through
the memory structure of the system.
Concerning part d), the result is just your FIR lter I/O equation from part a).
What this example develops is a general process for determining the output to an
FIR lter. It involves: 0) plotting h[k], the impulse response as a function of k;
1) plotting x[n k]; n = 0 as a function of k (i.e. folding the input); 2) shifting
the input by the impulse response (i.e. shifting the input); 3) multiplying h[k]
and x[n k] for each output time n of interest; and 4) adding this multiplier
results to form the output.
You are already familiar with the steps of this process (they are basic operators
introduced and exemplied in Subsection 1.4.1). This process, which we will
formalize and generalize in Chapter 2, is called convolution. Convolution is the
principle topic of Chapter 2. So if you take the time to understand this example
now, you will be well on your way to mastering Chapter 2 of this Course.
Example 1.27: Consider the N point averager described in Subsection 1.5.1 as
an example of a system. This is an FIR lter. Determine an expression for its
output for complex sinusoidal input x[n] = ej0 n .
Solution:
39
Innite Impulse Response (IIR) Filters

Since we have a name (FIR) for DT systems with nite impulse response, we might
as well have one (IIR) for DT systems with innite impulse response. The simple DT
feedback system, described earlier in this Subsection as a generalization of the Subsection
1.2.5 example, is an IIR lter (with innitely long impulse response h[n] = an u[n]).
Subsystem Interconnections
In engineering we often encounter systems that can be considered as congurations of
components. The components are referred to a subsystems of the larger system. For example, subsystems are often used together as components of some larger, more complex system.
Each component provides a required function. Together they implement an overall system
specication. This modular approach can simplify design and implementation while providing a exible capability. Similarly, we often work with systems composed of one subsystem
in a physical system interconnected with others are the devices which are used to it to eect
the physical systems performance. Fortunately, the techniques developed in this Course
provide a systematic framework for design and analysis of complex systems congured as an
interconnection of simpler subsystems.
Here we introduce the three most common types of subsystem interconnections. They are
described and illustrated for DT systems, though they are applicable for CT systems as well.
Parallel Interconnections
Figure 24 shows M subsystems connected in parallel. The subsystems share a common
input x[n], and there individual outputs are summed together to form the overall output
y[n].. That is,
M
y[n] =
yi [n] ,
(67)
i=1
where yi [n] is the output of the ith subsystem due to input x[n]. One example of a parallel
interconnection is an audio graphic equalizer, where each subsystem amplies or attenuates
a dierent range of audio frequencies.
y [n]
subsystem #1
y [n]
subsystem #M
y[n]
....
subsystem #2
....
x[n]
y [n]
M
Figure 24: A parallel interconnection of subsystems.
40
Cascade (Serial) Interconnections
Figure 25 shows M subsystems connected in cascade. The output of one subsystem is the
input to the next, with x[n] (the overall system input) being the input of the rst and y[n]
(the overall system output) being the output of the last. In Practicum 1 we consider two
applications of cascaded interconnections: 1) a channel equalizer processing the output of a
physical communications channel; and 2) an image deblurrer processing an image captured
by a camera in motion.
v1 [n]
subsystem #1
subsystem #2
v2 [n]
....
x[n]
v [n]
subsystem #M
y[n]
Figure 25: A cascade interconnection of subsystems.

Feedback Interconnections
Figure 26 shows 2 subsystems congured in a feedback interconnection. Subsystem #
2 processes the output of subsystem # 1 and feeds it back into subsystem #1 (i.e. the
subsystem #2 output is superimposed onto the overall input x[n] and fed into subsystem #1).
The negative sign on the input summer is pretty standard, suggesting negative feedback.
The overall output is the subsystem # 1 output. Feedback is commonly used in controls
systems, where subsystem # 1 represents a physical system, and subsystem # 2 is a controller
which monitors the physical system output so as to eect its performance by altering its
input.
x[n]
v[n]
+
subsystem #1
y[n]
subsystem #2
Figure 26: A feedback interconnection of subsystems.
Inverse Systems
In may situations we are interested in inverting the degrading eect of some physical system we are stuck with. Call this the original system. The channel equalization problem
(Subsection 1.2.3 and Practicum 1) and image deblurring problem (Practicum 1) are examples of this. Typically, we process the degraded signal at the output of the original system
so a to invert its eect. Thus we are dealing with a cascade of a physical system and an
inverse system we design. Figure 27 illustrates this for the case where the original system is
a simple FIR lter and the inverse system is a simple IIR system. Employing the induction
41
approach we used previously in Subsection 1.2.5, we have that

y[n] =
=
=
=
=
=
=
v[n] + 0.5y[n 1]
(68)
v[n] + 0.5{v[n 1] + 0.5y[n 2]}
v[n] + 0.5v[n 1] + 0.25 {v[n 2] + 0.5y[n 3]}
v[n] + 0.5v[n 1] + 0.25v[n 2] + 0.25{v[n 3] + 0.5y[n 4]}
v[n] + 0.5v[n 1] + 0.25v[n 2] + 0.25v[n 3] + 0.125v[n 4] +
{x[n] 0.5x[n 1]} + 0.5{(x[n 1] 0.5x[n 2]} + 0.25{(x[n 2] 0.5x[n 3]} +
x[n] .
Thus, the inverse system perfectly inverts the eect of the original system. In this Course
we will learn the tools we need to analyze this problem and design eective inverse systems.
As pointed out in Subsection 1.2.3, inverting an original system is not usually as easy as this
simple example might suggest. In ece5790 we learn how to design inverse systems.
x[n]
v[n] = x[n] 0.5 x[n1]
physical system
y[n] = v[n] + 0.5 y[n1]
inverse system
Figure 27: Inverting the eect of a degrading physical system.
1.5.2
System Properties
In this Subsection we introduce the following system properties:

[1.5.2.1] causality (and the associated property of memory);
[1.5.2.2] linearity;
[1.5.2.3] time-invariance; and
[1.5.2.4] stability.
These properties are basically the same for CT and DT system, so we will cover both
simultaneously. For each property we start with its denition, give a few examples, and
discuss how to generally go about testing if a system exhibits the property. At this time we
will not spend too much time with these tests, and you will not be responsible for being able
to apply them. Later in the Course we will be developing new analysis tools and introducing
system descriptions that will make testing systems for these properties both systematic and
simple.
42
1.5.2.1 Causality (and the associated property of memory)
Causality: A system is causal it, for any time t1 (or n1 ), the output y(t1 ) (or y[n1 ]) in
not a function of inputs x(t) for times t > t1 (or x[n] for times n > n1 ). That is: the
present output is not a function of future inputs.
Test: Basically by inspection. (The right system I/O description makes it easy.) A
real-time system, that operates on inputs as they occur to generate outputs, must
be causal. However it is possible and not uncommon to implement concausal systems
which process input data that is stored, say on a CD.
Example 1.28: The following is an I/O expression for a parallel RC circuit
(see Subsection 1.5.1 above). Is this circuit casual?
y(t) =
t
0
1 1 (t )
x( ) d
e RC
C
t>0
Solution: Yes. From the I/O equation, note that for time t > 0 only the
present and past input is used to compute the present output, i.e. for y(t),
only x( ); 0 t is used.
Example 1.29: Is each of the following accumulators causal?
a) y1 [n] =
k=0 x1 [n k]. Solution: Yes, only x1 [k]; k n is
used to compute y1 [n].
b) y2 [n] = n+1 x2 [k]. Solution: No, x2 [n + 1] is used to compute y2 [n].
k=
Memory (stated for CT, but applicable for DT too): A system is memoryless if the
output y(t) at any time t1 is a function of the input x(t) for only that time t1 . A
system has memory if: any present output is a function of either past or future inputs.
Example 1.30: Do the following systems have memory?
a) y(t) = cos(x(t)). Solution: No, the output at time t is a function of
the input for only time t.
b) y[n] = x[n + 1]. Solution: Yes, this system remembers 1 sample into
the future. It has negative memory, so it is noncausal.
d
d
c) y(t) = dt x(t). Solution: Yes, from circuits we know that dt means
input energy storage. The slope of a CT signal at a given time depends
on values of that signal incrementally close to that time.
For some system I/O descriptions, i.e. for the RC circuit in Example 1.28 and the
accumulators in Example 1.29, the memory structure is obvious. For example, with
the accumulator I/O expressions, you can see by inspection of the range of memory
time k what inputs eect the output at output (or current) time n. Similarly, for the
RC circuit, represents memory time and t output (or current) time. With other I/O
descriptions, the memory structure might be less obvious.
43
1.5.2.2 Linearity (also called superposition)

Here we describe linearity for CT systems, but as with the other properties considered
here, the description is similar for DT systems. Consider a system. Given any inputs
x1 (t) and x2 (t) with corresponding outputs y1 (t) and y2 (t), and given any constants a1
and a2 , the system is linear if and only if input
x(t) = a1 x1 (t) + a2 x2 (t)
(69)
y(t) = a1 y1 (t) + a2 y2 (t) .
(70)
results in the output

Note that if this is true, then given any N inputs xi (t); t = 1, 2, , N and corresponding outputs yi (t); t = 1, 2, , N, any input of the form
N
x(t) =
ai xi (t)
(71)
ai yi (t) .
(72)
i=1
gives output
N
y(t) =
i=1
Test: To prove a system is nonlinear, you need only determine a counter example. In
general, proving a system is linear requires a test that shows the system is linear for
any pairs of inputs xi (t) and weights ai . This can be challenging. However, fortunately
there are several common system I/0 forms for which linearity/nonlinearity can be
deduced by inspection.
Example 1.31: Is the following system linear?
y[n] = x2 [n]
By counter example:
x1 [n]
1 x1 [n]
y1 [n] = x2 [n]
1
2
x1 [n] = 1 y1 [n] .
So we see that the system is nonlinear.

Basically, if you see a signal raised to a power in the I/O expression, or more generally
if you see signals multiplied together, then the system is nonlinear.
44
Example 1.32: Given that a CT system is linear, and given the following two
I/O pairs, if possible determine the output due to the each of the two inputs
shown.
x 1 (t)
y 1 (t)
x 2 (t)
y 2 (t)
x 3 (t)
Solution:
1/2
1
1/2
x 4 (t)
1
Figure 28: Assuming linearity, given outputs for x1 (t) and x2 (t), what are the outputs for
x3 (t) and x4 (t)?
Example 1.32 illustrates the power of linearity. Given the response of a linear system
to two inputs, we know its response to any linear combination of those two inputs.
This is useful and important enough that, if possible, we would like to work with (i.e.
develop or interact with) linear systems. This example also shows that linearity, by
itself, has its limitations.
45
1.5.2.3 Time-Invariance (TI)

Basically, a system is time-invariant if it does not change with time that is, if the
way it operates on inputs does not change with time.
Test: A general test is illustrated in Figure 29 for a CT system, where x(t) is any input
and T is any delay. The same test applies for DT systems.
system
y(t)
y(tT)
any delay T
any x(t)
same delay T
x1 (t) = x(tT)
same system
y1 (t) = y(tT)
Figure 29: Illustration of the test for time-invariance.

What this gure means is that if we apply the same arbitrary input to a system at any
two dierent times, the outputs are the same, except separated in time by the same
amount as the inputs. So, for example, if you build a circuit in lab on Tuesday and
successfully test it for some input x(t) (i.e. it gives the required output y(t)), then if
you demonstrate it to the lab instructor on Thursday for the same input (i.e. delayed
by two days x(t 2 days)), if the system is time-invariant (e.g. it still works) you
will observe the same output (i.e. delayed by two days y(t 2 days)).
Example 1.33: Are the following systems time-invariant?
a) y[n] = x[n] cos(0 n). Solution: This system is time varying, i.e.
cos(0 n) is a time varying multiplier. Test:
y[n N1 ] = x[n N1 ] cos(0 n)
(73)
x1 [n] = x[n N1 ]
(74)
y1 [n] = x1 [n] cos(0 n) = x[n N1 ] cos(0 n) = y[n N1 ]
in general.
t
b) y(t) = x( ) d . Solution: This system in time invariant, i.e. it
is composed of constant multipliers (of values 1) and constant memory
structure (from present time back to ). The formal test would
involve a change of variable on the integration.
2t
c) y(t) = x( ) d . Solution: This system is time varying, i.e. the
memory structure changes with time (e.g. for t < 0 this system is causal,
but for t > 0 it is not).
Example 1.33 illustrates the fact that if a systems memory structure or multipliers are
time varying, then the system is time varying.
46
Example 1.34: Given a CT TI system, and given the following I/O pair, if
possible determine the output due to the each of the two inputs shown.
x 1 (t)
y 1 (t)
Solution:
x 2 (t)
1
x 3 (t)
1/2
1
1.2
Figure 30: Assuming time-invariance, given the output for x1 (t), what are the outputs for
x2 (t) and x3 (t)?
Example 1.35 shows the power of time-invariance. Given the response of a TI system
to any input, we know its response to any delayed version of that input. This is useful
and important enough that, if possible, we would prefer to work with TI systems. This
example also shows that time-invariance, by itself, has its limitations.
Example 1.35: Given a CT LTI system, and given the impulse response h(t)
shown below, if possible determine the output due to the inputs shown.
(t )
h(t)
(1)
Solution:
y(t)
1
x(t)
....
.... ....
(1)
T
2T
....
0 T
2T
Figure 31: An LTI system and the impulse I/O pair.

Examples 1.32 & 1.34 illustrate that linearity and time-invariance are useful
and desirable system properties, though separately they are limited. Example 1.35 shows that together they can be quite powerful if you know
the response to one input, you know the response to any linear combination
of all the delays of that one input. Example 1.36 (below) also shows this.
As a result, in engineering, Linear Time-Invariance (LTI) systems are very
common. In Subsection 1.5.3 we will expand on this idea of a LTI system.
47
Example 1.36: Given an DT LTI system, and given that input p2 [n] produces
output p4 [n], if possible determine the output due to the given x[n].
Solution:
y[n]
x[n]
=
=
p [n]
2
+
p2 [n1]
Figure 32: An LTI system and the impulse I/O pair.

1.5.2.4 Stability
Bounded-Input Bounded-Output (BIBO) Stability is the only denition of stability of
concern in this course. A system is BIBO stable if and only if for any bounded input,
i.e. for any x[n] such that
|x[n]| < B1 ,
n
(75)
where B1 is nite, the output is bounded, i.e.
|y[n]| < B2 ,
(76)
where B2 is nite. (Note that the notation n reads for all n.) This denition applies
to CT systems too.
Test: For now, we can prove a system is not BIBO stable via counter example (i.e.
nd one bounded input that results in an unbounded output). We can show that a
system is BIBO stable by showing it is BIBO stable for the worst possible input case.
Later, when we have easy I/O descriptions to work with, the test (for some types of
systems) will be much easier.
Example 1.37: Is the following system stable?
y(t) =
x( ) d
Solution: No. Proof is by counter example. Let x(t) = u(t), which is

bounded.
0
t<0
y(t) =
.
t
t0
0 dr = t
So, limt y(t) = , which is unbounded.
48
Example 1.38: Is the following system stable? Assume 0 < a < 1.

y[n] =
k=0
ak x[n k] ;
0<a<1
Solution: Yes. Proof is by worst case. Let x([n] = B1 < for all n. This
x[n] is bounded. It is the worst case input because, with 0 < a < 1, if all
x[n] are positive, all x[n] have a positive contribution to the output. So the
output is greatest if all inputs are a large as possible (i.e. equal to the signal
bound). Then,
y[n] =
ak B1 = B1
k=0
k=0
ak =
B1
,
1a
which is bounded.
1.5.3
Linear Time-Invariant (LTI) Systems
Figure 33 illustrates a DT LTI system. A CT LTI system illustration would be similar.

Linearity, represented by the 3-rd input line, is shown for a linear combination of two inputs.
It extends to any number N of inputs.
y [n]
x 1 [n]
x 2 [n]
x[n] = a 1 x 1 [n] + a 2 x 2[n]
x 1 [nN]
DTLTI
System
y2 [n]
y[n] = a 1 y1 [n] + a 2 y2 [n]
y1 [nN]
Figure 33: A Discrete-Time Linear Time-Invariant (DT LTI) system.
From a system analysis and synthesis point of view, systems which are both linear and timeinvariant have denite advantages. We will see some of these advantages throughout the
Course. In this Subsection we emphasize two advantages of LTI systems.
Advantage 1: For a LTI system, if you know the response y[n] for any one signal x[n], then
you know the response to any linear combination of all delayed versions of x[n]. This is a
direct consequence of the two system properties it does not need further proof. So, if you
know the output to just one input signal, you know a lot! This advantage of LTI systems in
the topic of Chapter 2 of this Course. It is illustrated in the next two examples.
49
Example 1.39: Given a CT LTI system, and given the following I/O pair, if
possible determine the output due to the each of the two inputs shown.
x 1 (t)
y 1 (t)
Solution:
x 2 (t)
....
....
x 3 (t)
....
....
1/2
1
1/2
Figure 34: Assuming linearity and time-invariance, given the output for x1 (t), what are the
outputs for x2 (t) and x3 (t)?
Example 1.40: Consider a DT LTI system which for input x[n] = [n] gives output
y[n] = .5n u[n]. Find the output y1 [n] due to input x1 [n] = 2[n] 0.5[n 2].
Solution:
50
Advantage 2: For a LTI system, the output due to a complex sinusoid input will be a complex
sinusoid of the same frequency the eect of the system is simply to scale this input. We
have not proved this property of LTI systems. We will do so in Chapter 2. However, this
property of LTI systems has already been illustrated in Examples 1.25 & 1.27 (i.e. the RC
circuit and the FIR lter considered in these examples are LTI systems).
Example 41: Consider the simple IIR lter introduced in Subsection 1.2.5. It
can be shown that this is a LTI system. Generally, this system has I/O equation
y[n] = a y[n 1] + x[n]. In Subsection 1.2.5 we used a = 0.9. For the
accumulator considered in Subsection 1.4.1, a = 1. It is straightforward to show
that the I/O of this simple IIR lter can be written as
y[n] =
k=0
ak x[n k] .
Determine the output for complex sinusoidal input x[n] = A ej0 n+ .

Solution:
Example 42: Consider FIR lter with I/O equation y[n] =

Determine the output for input x[n] = ej0 n .
Solution:
5
k=0
(1)k x[n k].

1.5.4
51
Linear Constant Coecient System I/O Equations
The I/O response of a system is often represented as an I/O dierential (CT) or dierence
(DT) equation. For example, performing loop and/or node analysis of an RLC circuit yields
an I/O dierential equation which is linear (as long as the circuit components are linear) with
constant coecient (as long as the circuit components are time invariant). Furthermore,
the simple IIR and FIR DT systems we have dealt with thus far, as Examples and in
Practicum 1, have I/O equations that are dierence equations which are linear with constant
coecients. On the other hand, we have just established that LTI systems have very desirable
characteristics. Below we establish a very important relationship between LTI systems and
dierential/dierence equation I/O representation.
LCC Dierential Equation Representation of CT LTI Systems:
A CT system is LTI if and only if it can be represented as follows:
N
al y (l) (t) =
l=0
where x(i) (t) =
di
x(t).
dti
bk x(k) (t)
(77)
k=0
This is the form of a LCC dierential equation.
Comments:
The al s and bk s are time-invariant constants (so the system is TI).
y(t) and x(t), and their derivatives, appear as linear combinations only (i.e. no products
of signals, powers of signals, etc.) (so the system is L).
If a systems I/O can be represented as a LCC dierential equation, the system is LTI.
The solution, y(t); t 0 depends on the input and the initial conditions
y (i) (0 ); l = 0, 1, , N 1 and x(k) (0 ); k = 0, 1, , M 1.
Given an input and initial conditions, such a system can be solved (i.e. the I/O LCC
dierential equations can be solved) in a systematic way using Laplace transforms or
Continuous-Time Fourier Transforms (CTFTs). We will learn how to do this later.
So any CT LTI systems I/O can be expressed in the Eq (77) form. Also, any CT system
what has an I/O relationship in the form of Eq (77) is a CT LTI system. Reecting back to
your circuits courses, recall that any RLC circuit (with input and output which are voltages
and/or currents) has an I/O expression in the form of Eq (77). In this Course we will learn
how to handle I/O issues for CT LTI systems in systematic ways, regardless on how large
M on N is (i.e. regardless of how complicated the system is), and for any type of input.
52
LCC Dierence Equation Representation of DT LTI Systems:

A DT system is LTI if and only if it can be represented as follows:
N
l=0
al y[n l] =
bk x[n k]
k=0
(78)
This is the form of a LCC dierence equation. This is also often written as:
M
y[n] =
k=0
bk x[n k]
l=1
al y[n l]
(79)
where we have assumed that a0 = 1.

Comments:
See the rst 4 Comments for dierential equations above, replacing the derivatives with
the corresponding dierences, e.g.
y (l) (t)
y[n l] .
(80)
Given an input and initial conditions, such a system can be solved (i.e. the I/O LCC
dierence equations can be solved) in a systematic way using z-transforms or DiscreteTime Fourier Transforms (DTFTs). We will learn how to do this later.
Figure 35 shows structures5 for implementing a general DT LTI system that follows directly
from Eq (79). Note that Figure 35(a) implements the general FIR lter introduced as an
example of a system at the beginning of this Section. Concerning Eq (79), an FIR lter
corresponds to an LCC dierence equation with no output feedback (i.e. with N = 0).
Figure 35(b) implements the more general LCC dierence equation, allowing for output
feedback. Note that the simple IIR lter introduced as an example of a system at the
beginning of this Section corresponds to this structure, with M = 0, b0 = 1 and N = 1.
x[n]
b0
y[n]
b0
x[n]
b1
b1
b2
a2
b M1
adder
a N1
bM
delay
b M1
b2
bM
adder
aN
delay
multiplier
(a)
a1
+
1
+
1
y[n]
+
1
multiplier
(b)
Figure 35: DT LTI system structures: a) FIR lter; and b) general IIR lter.
5
There is a reason that each delay in Figure 35 is represented as z 1 . This reason will become apparent
later in the Course when we use the z-transform to represent DT-LTI systems.
1.6
53
Practicum 1
Introduction to Discrete Time Signals & Systems

Reporting Requirements: At the beginning of the semester it will be announced whether
the practicum reporting requirement will be: 1) a full practicum report; or 2) submission of
a practicum Sign O Sheet & Report Form only. Either way, the Sign O Sheet & Report
Form, which appears at the back of the Practicum Section, should be submitted.
The full report, if required, should be headed with your name, the course name, and the
practicum number and title. The rst section should be titled Objectives and Conclusions and should describe your understanding of objectives and conclusions. Nominally,
this section should contain two or three paragraphs. The second section should be titled
Results and Comments and should be organized in the same way that the procedures in
the practicum are organized. Provide all derivations, comments, answers, plots and comparisons requested. (The Matlab Diary command might be used to generate the body of this
section.) Each plot should have a title, x,y coordinate labels, and your name printed on the
plot using Matlab (e.g. using the text command). Include printouts of all Matlab .m les
used (i.e. to compute values and generate plots). As a heading for each program printout,
include a comment line with your name on it. The Sign O Sheet & Report Form should be
attached as the last page of the full report.
Background: Read Subsection 1.2.3 of these Course Notes.
Procedures: Procedures 1(a), 2(a) and 3(a) should be completed before the rst lab session.
1. Finite Impulse Response (FIR) Filter: Consider the following DT system:
y[n] =
M 1
k=0
bk x[n k] .
(1)
(a) On paper, for input x[n] = [n], determine the output y[n] = h[n], i.e. nd
the impulse response. (h[n] is standard notation for the impulse response of a
system.) Note that we can also express the FIR lter I/O equation in terms of
the impulse response as
M 1
h[k] x[n k] .
(2)
y[n] =
k=0
Why is this general type of system called an FIR lter?

(b) In le FIR mine.m, write a Matlab function that implements a general FIR lter
which is called by the command
y = FIR_mine (h,x)
where
x is the input array containing the N input values x[n]; n = 0, 1, , N 1,
h is the M element array of impulse response values
h[k] = bk ; k = 0, 1, , M 1, and
y is the output array containing y[n]; n = 0, 1, , N 1.
Your code should look something like the following (completed):
54
function
y = FIR_mine(h,x)
%
% h = impulse response
% x = input
%
M = length(h);
N = length(x);
% N>M assumed
%
% initial outputs
%
for n=0:M-2
y(n+1) = 0.;
for k=0:n
y(n+1) = y(n+1) + h(k+1)*x(n+1-k);
end
end
%
% steady state
%
for n=M-1:N-1
y(n+1) = 0.;
for k=0:M-1
???????????? (fill in)
end
end
(c) Test your FIR mine function for a 20-point averager (i.e. h[n] =
input x[n] = p20 [n]. Plot the output of n = 0, 1, , 50.
1
20
p20 [n]) with
2. A Simple Innite Impulse Response (IIR) Filter: Consider the following simple
DT system:
y[n] = x[n] + a y[n 1] ,
(3)
where a called the feedback coecient.
(a) On paper, for input x[n] = [n], determine the output y[n] = h[n]. Why is this
system called an IIR lter?
(b) In le IIR simple.m, write a Matlab function that implements this simple IIR lter
which is called by the command
y = IIR_simple (a,x)
where
x is the input array containing x[n]; n = 0, 1, , N 1,
a is the feedback coecient; and
y is the input array containing y[n]; n = 0, 1, , N 1.
55
Your code should look something like the following (completed):

function
y = IIR_simple(a,x)
%
% a = feedback coefficient
% x = input
%
N = length(x);
%
% initial output
y(1) = x(1);
%
% steady state
for n=2:N
???????????? (fill in)
end
(c) Test your IIR simple function for an accumulator (i.e. a
x[n] = p20 [n]. Plot the output of n = 0, 1, , 50.
1) with input
3. Discrete Time Channel Equalization: Consider the following two DT systems.

The rst has input x[n] and output v[n]. The second has input v[n] and output y[n].
v[n] = x[n] .9 x[n 1]
(4)
y[n] = v[n] + .9 v[n 1] + .81 v[n 2] + .729 v[n 3] + .6561 v[n 4] . (5)
For example, the rst system might be a discrete model of a multipath communication
channel, and the 2-nd system might be designed as part of a receiver to compensate
for (or equalize or invert the eect of) the channel.
(a) On paper, determine and plot the impulse responses of the channel and the equalizer. Do the same for the overall impulse response of the cascade of the two systems
(i.e. the response from the input to the rst system to the output of the second,
if the output of the 1-st system is the input to the second).
(b) Use FIR mine to compute the response of the rst system to input x[n] = p2 [n].
Plot this for n = 0, 1, , 10. Do the same for the output of the cascade of the
two systems. Comment on the eectiveness of the second system at equalizing
the rst. Reecting on characteristics of y[n], suggest how the equalizer could be
improved, describing this in terms of the impulse response of the equalizer.
(c) Instead of using the second system to equalize the rst, consider the third system:
y[n] = v[n] + .9y[n 1] .
(6)
Repeat Part (a).

(d) For the cascade of the rst and third systems, use your FIR mine and IIR simple
functions to compute the impulse response and the response to x[n] = p2 [n]. Plot
the equalizer output for n = 0, 1, , 10.
56
(e) In terms of the impulse responses of the two equalizers, explain why the 2-nd one
works better than the 1-st.
4. Image Deblurring: For this procedure you will process a black and white image.
There are three les you will need, lighthouse.mat, lighthouse1.mat and show img.m,
which can be found in directory
v:/Electrical Computer/ece3240
(a) Copy the three les into your working directory or otherwise assure that you have
access to them within Matlab.
(b) Load lighthouse.mat into Matlab. In Matlab, what is the image name? What
size is it? Display the image using the command show img(xx). Comment on the
quality of the image. Do the same for lighthouse1.mat.
(c) The image yy[i, n] has been processed to simulate the eects of horizontal camera
motion. That is, the original image xx[i, n] has been blurred along the horizontal
to generate yy. The blurring was simulated using the following equation for each
horizontal image line (i.e. each row)
10
yy[i, n] =
j=0
0.8j xx[i, n j] ;
i = 1, 2, , 326 ;
n = 1, 2, , 426 . (7)
To deblur the given image yy, use your FIR mine function to process each row as
follows to generate the deblurred image zz.
zz[i, n] = yy[i, n] 0.8 yy[i, n1] ;
i = 1, 2, , 326 ; n = 1, 2, , 426 . (8)
Use yy[i, 0] = 0.0, i = 1, 2, , 326.
(d) Display the deblurred image zz. Compare its quality to that of the yy image, and
the original image xx stored in le lighthouse.mat.
(e) How do the systems and results of this (the Image Deblurring) Section relate to
those of the previous (Channel Equalization) Section?
57
Practicum 1: Instructor/TA Sign O Sheet, & Report Form

Students Name:
1. Procedures 1(c): 20-pt. averager output
2. Procedure 2(c): accumulator output
3. Procedure 3(a): below, sketch the three impulse responses.
4. Procedure 3(b): output plot of 1-st equalizer

Record requested comments.
5. Procedure 3(c): sketch the impulse response from the channel input to the 2-nd equalizer output.
6. Procedure 3(d,e): output plot of 2-nd equalizer

7. Procedure 4: displays of original, blurred and deblurred images

58
.
1.7
59
Appendix 1A: Complex Numbers and Signals
In this Appendix we review complex numbers and signals, and common operations applied
to them. We assume that students in this course are familiar with these topics, but that
some will benet from a review which is focused on the specics required for this Course.
The objective of this Appendix is to provide students on this Course a convenient reference
on complex-valued signals that we can build on in support of the topics of this Course.
1.7.1
Complex Numbers
Let j = 1. Consider a complex-valued number x such that

x = xr + j xi = |x| ej
(1)
where xr = Re{x} and xi = Im{x} are, respectively, the real and imaginary parts of x,
and where |x| and x are, respectively, the magnitude and phase of x. Figure 36 illustrates
a complex number x visualized in the complex plane. Note that xr , xi , |x| and x are all
real-valued. Although x can be any real number, since it represents an angle relative to
the positive real axis, ej x = ej( x +2) for any x (i.e. ej x = ej[( x)modulo 2] ). Thus, in
any discussion, we typically limit this angle as either 0 x < 2 or < x .
Im{x}
complex plane
xi
|x|
x
Re{x}
xr
Figure 36: A complex number x in the complex plane.

We sometimes consider the {xr , xi } representation of x to be the rectangular coordinate
representation of x, and {|x|, x} to be the polar coordinate representation. The conversions
between these two representations are
xr = |x| cos( x) ;
|x| =
x2 + x2 ;
i
r
xi = |x| sin( x)
x =
xi
tan1 xr
+ tan1
1
+ tan
(2)
xr 0
xr < 0 and xi > 0
xi
xr
xi
xr
(3)
xr < 0 and xi < 0
There are various equivalent ways to express the conversion from rectangular coordinates to
x. Above, the conversion is given which results in the range < x . Also note that
tan1 (inverse tangent) and arctan (arc tangent) mean the same thing.
The conjugate of a complex number, denoted x for complex-valued x, is dened as
x = xr j xi = |x| ej
So, Re{x} = Re{x }, Im{x} = Im{x }, |x| = |x |) and x = x .
(4)
60
Example a1: Given w = 1 + j, x = 1 + j2, y = 2 ej/2 and z = 3 ej/5 ,

determine: a) w in polar representation; b) x in polar coordinates;
c) y in rectangular representation; and d) z in rectangular coordinates.
Solution: Sketch each of these numbers in the complex plane so as to verify the
plausibility of the results given below.
1
a) w = |w| ej w = 12 + 12 ej tan (1/1) = 2 ej/4 .
1
b) x = |x| ej x = (1)2 + (2)2 ej(+tan (2/(1)) = 5 ej(2.0344) .
c) y = yr + j yi = 2 cos(/2) + j 2 sin(/2) = j 2.
d) z = zr + j zi = 3 cos(/5) + j 3 sin(/5) = 2.4271 j 1.7634 .
Before proceeding, you should make sure you are comfortable with the fact that to following
results can be obtained by inspection: j = ej/2 ; ej = ej = 1; ej6 = ej2 = 1;
and ej7/2 = ej3/2 = ej /2 = j.
1.7.2
Algebra with Complex Numbers
Consider two complex-valued numbers x and y. Then

x + y = (xr + yr ) + j (xi + yi )
x y = (xr + yr ) + j (xi yi )
x y = |x| |y| ej(
x
|x| j( x
=
e
y
|y|
(5)
(6)
x+ y)
(7)
y)
(8)
Note that when adding or subtracting complex numbers, it is advantageous to represent

them in terms of rectangular coordinates. When multiplying or dividing them, it is better
to represent them in polar coordinates. So be willing to convert between rectangular and
polar coordinates as needed.
61
Example a2: Given x = 2 ej3/4 and y = 1 j 2, determine: a) za = x + y;

b) zb = 2 x + 3 y; c) zc = x y; d) zd = 2 x y; and e) ze = 2 x .
y
Solution:
a) za = (1+j)+(1+j(2)) = (11)+j(12) = 2j = 2.2361 ej(2.6779) .
b) zb = 2(1 + j) + 3(1 + j(2)) = (2 3) + j(2 6) = 5 j4 =

6.4031 ej(2.4669) .
c) zc = 2 ej3/4 5 ej(2.0344) = 10 ej(2.35622.0344) = 3.1623 ej(0.3218) .
d) zd = 2 zc = 6.3246 ej(0.3218) .
e) ze = 2 5
2 ej3/4
ej(2.0344)
=2
2
5
ej(2.3562+2.0344) = 1.2649 ej(4.3906) = 1.2649 ej(1.8926) .
Note that, for consistency, when presented in terms of polar coordinates, angles
of results have all been represented in the range {, }.
As an extension of the rule for multiplication of complex numbers, let n be a positive
integer and x a complex number. We have that
xn = (|x| ej
x n
= |x|n (ej
x n
= |x|n ej
(9)
As an extension of the multiplication and division rules, let xi ; i = 1, 2, , M and

yj ; j = 1, 2, , N be sets of complex numbers. Then
M
i=1
N
i=j
xi
=
yj
M
i=1
N
i=j
|xi | j
e
|yj |
M
i=1
xi
N
j=1
yj
Example a3: Consider the complex numbers x = 3 + j 4 and y = ej/12 .

Determine: a) va = x3 ; b) vb = x100 , c) vc = y 12 and vd = y 24 .
Solution:
a) va = (3 + j4)3 = (5 ej(0.2953) )3 = 53 (ej(0.2953) )3 = 125 ej3(0.2953) =
125 ej(0.8855) .
b) vb = 5100 ej100(0.2953) = 7.886 1069 ej(29.53) = 7.886 1069 ej(1.53) .
c) vc = (ej/12 )12 = ej12/12 = ej = 1.
d) vd = ej24/12 = ej2 = 1.
(10)
62
1.7.3
Complex-Valued Signals
A complex-valued DT signal is just a sequence of complex-valued numbers. So, for a complexvalued signal x[n], we have that, for each sample time n,
x[n] = xr [n] + j xi [n] = |x[n]| ej
x[n]
(11)
Conversion between real/imaginary parts (i.e. rectangular coordinates) and magnitude/phase

(i.e. polar coordinates) is as described above for each complex number (i.e. for each sample time). When adding, subtracting, multiplying and dividing signals, the arithmetic rules
for complex numbers described above apply for each sample time. For complex-valued CT
signals, its the same story. That is, at each time t, a complex-valued CT signal is complex
number.
Example a4: Consider the DT complex sinusoid x[n] = ej(/4)n . Consider an N
point averager, i.e.
1
y[n] =
N
N 1
k=0
x[n k] .
a) Determine a general expression for y[n]. b) Determine y[n] for N = 16.

c) Determine y[n] for N = 12.
Solution:
a) For general N,
y[n] =
1
N
N 1
k=0
ej(/4)(nk) =
1 j(/4)n N 1 j(/4)k
e
e
N
k=0
1 1 ej(/4)N
.
N 1 ej(/4)
The last equation was derived using the Geometric series summation.
b) For N = 16,
= ej(/4)n
1
11
1 1 ej(4)
= ej(/4)n
= 0.
y[n] = e
j(/4)
16 1 e
16 1 ej(/4)
A 16-point averager completely attenuates a complex sinusoid of frequency
= .
4
c) For N = 12,
j(/4)n
1 1 (1))
1 1 ej(3)
= ej(/4)n
j(/4)
12 1 e
12 1 ej(/4)
2
1
1
= ej(/4)n
= ej(/4)n 1.3066 ej(1.1781) = 0.2178 ej((/4)n1.1781) .
j(/4)
12 1 e
6
The output of a 12-point averager to a complex sinusoid input of frequency
= is a complex sinusoid of the same frequency. It is attenuated by a
4
factor of 0.2178 and phase shifted by 1.1871 radians/sample.
y[n] = ej(/4)n
63
Example a5: Consider the DT complex sinusoid x[n] = ej(0 n) . Consider a

10-point averager, i.e.
1
y[n] =
10
9
k=0
x[n k] .
a) Determine a general expression for y[n]. b) Determine y[n] for 0 = 0.

c) Determine y[n] for 0 = .
Solution:
a) In general we have
1
10
y[n] =
ej0 (nk) = ej0 n

k=0
= ej0 n
j0 10
1 1e
10 1 ej0
1
10
ej0 k
k=0
b) For 0 = we have
y[n] = ejn
1 1 ej10
1
11
= ejn
= 0 .
j
10 1 e
10 1 (1)
Note that the input is completely attenuated.

c) For 0 = 0 the input is just x[n] = 1, and the output is
y[n] =
1
10
9
k=0
x[n k] =
1
10
1 = 1 .
k=0
Note that we did not use the general output expression derived in a) to
solve this part of the Example. Lets resolve this problem using this general
expression.
y[n] = ej0n
1 11
1 1 ej010
=
.
j0
10 1 e
10 1 1
0
This 0 is a problem which you know how to deal with using tools from your
calculus courses. Try using LHospitals rule on your own. Here we will use
a Taylor series expansion approach. Starting with
y[n] = ej0 n
1 1 ej0 10
,
10 1 ej0
and using the Taylor series expansion ex = 1 + x + x2 + x3 + , we have

that
y[n] = ej0 n
1 1 (1 + 0 10 + (0 10)2 + )
1 (0 10 + (0 10)2 + )
= ej0 n
.
2
2
10
1 (1 + 0 + 0 + )
10
(0 + 0 + )
64
For 0 = 0, we consider the limit
lim
0 0
0 10
(0 10 + (0 10)2 + )
=
= 10 ,
2
(0 + 0 + )
0
(i.e. we keep only the lowest order term of each expansion), so

y[n] = ej(0)n
1
10 = 1 .
10
Either way, we see that the output is passed through the system without
change. We call a system that attenuated some inputs while passing others
without change a lter.
Example a6: Consider the DT signal x[n] = ej(0 n) Consider the simple DT
system
y[n] = 0.9 y[n 1] + x[n]
which can be shown to have output
y[n] =
k=0
0.9k x[n k] .
a) Determine a general expression for y[n]. b) Determine y[n] for 0 = 0.

c) Determine y[n] for 0 = .
Solution:
a) In general we have
y[n] =
0.9k ej0 (nk) = ej0 n
k=0
= ej0 n
(0.9 ej0 )k
k=0
1
.
1 0.9 ej0
b) For 0 = 0, that is for x[n] = 1, we have

y[n] = ej(0)n
1
= 10 .
1 0.9(1)
The input is amplied by a factor of 10.

c) For 0 = , that is for x[n] = (1)n , we have
y[n] = ejn
1
= 0.5263 ejn
1 0.9(1)
The input is slightly attenuated (by a factor of 0.5263).
65
Example a7: Consider the CT complex sinusoid x[n] = ej0 t and the system
y(t) = x(t) ej100t .
So this system simply multiplies the input x(t) by a complex sinusoid of frequency
100 radians/second. Determine a general expression for y(t).
Solution:
y(t) = ej0 t ej100t = ej(0 +100)t .
So the output is also a complex sinusoid, with frequency 0 + 100. That is,
the frequency of the input has been shifted by an amount equal to the frequency
of the sinusoid the system multiplies it by. We call this process of shifting the
frequency of an input modulation. We will consider modulation later in this
Course.
1.7.4
Why Consider Complex-Valued Signals?
At this point you may be wondering why we consider complex-valued numbers and signals.
The math and algebra seems to be harder (if you are not used to it), and complex numbers
dont appear to exist in the real engineering world. So why?
There are several ways to answer this question. One has to do with working with sinusoids.
First you have to acknowledge that we are centrally interested sinusoidal signals. That is,
we are interested in them almost all of the time. At the end of this Course you should
understand this. At this point you should at least understand the we naturally think of
signals in terms of their sinusoidal content (e.g. the base and treble of an audio signal, or
the transmission frequency band of a cell phone or a radio station, or the color of light).
So then the question is why consider complex-valued sinusoids as opposed to real-valued
sinusoids?
One answer to this is that the algebra of complex-valued sinusoids is easier. Easier algebra
can be advantageous in the analysis and the design of signals and systems. Consider, for
example, the simple task of multiplying two CT sinusoids and expressing the result as a
weighted-sum of sinusoids. First consider the product cos(1 t) cos(2 t). You probably
recall that there is a trigonometric identity for this, but lets say you dont remember it.
You could look it up, or to save time you could use Eulers identities, i.e.
1 j1 t
1 j2 t
(12)
e
+ ej1 t
e
+ ej2 t
2
2
1 j(1 +2 )t
(13)
e
+ ej(1 +2 )t + ej(1 2 )t + ej(1 2 )t
=
4
1
1
=
cos((1 + 2 )t) +
cos((1 2 )t) .
(14)
2
2
So complex sinusoids can be used to derive trigonometric identities. To a mathematician
this might be happy hour conversation, but an engineer this is not a big deal. The point is
that, in contrast, the multiplication of two complex-valued sinusoids is simply
cos(1 t) cos(2 t) =
ej1 t ej2 t = ej(1 +2 )t .
(15)
66
This illustrates that when you are analyzing a signal or designing a system, and you encounter
operations involving sinusoids, if you couch the derivations in terms of complex-valued sinusoids, the math will generally be easier. Would you rather spend your time looking through
trig identity tables, or proceeding with little resistance to a solution. What would your
employer rather pay you $400.00 a day to do?
Another argument for the use of complex-valued sinusoids involves what are called inphase/quadrature (I/Q) receivers. These are commonly found in systems such as mobile
phones, radio, RADAR and SONAR where modulation is required. An I/Q receiver is shown
in Figure 37. The input x(t) is the signal from the receiver antenna, after some preconditioning (e.g. amplication and ltering). The two real-valued outputs shown, termed the
in-phase output xi (t) and the quadrature output xq (t), are treated in subsequent processing
as a single complex-valued output
xiq (t) = xi (t) + j xq (t) = x(t) ej0 t .
(16)
cos ( o t)
x i (t)
x(t)
x q (t)
sin (o t)
Figure 37: An in-phase/quadrature (I/Q) receiver.

Since this is a very common receiver structure, that generates complex-valued signals
that must be processed, this is a strong motivation for getting comfortable with complex
numbers and signals. But why are I/Q receivers so common in the rst place? Basically,
because subsequent processing is simplied in the same way that algebra with complexvalued sinusoids is simplied. Processors which are analogous to I/Q receivers are found
commonly in Digital Signal Processing (DSP) systems for many applications.
1.8
67
Problems
Chapter Topics:
1.1-8 (basic math, signals and operators)
1.9-13 (energy and power of signals)
1.14-16 (linear combinations)
1.17-20 (simple DT systems)
1.21-28 (sinusoidal response of simple DT systems)
1.29-33 (system properties)
1.34-37 (LTI systems)
1. Given x = 1 + j, y = 4 ej/3 and z = 4 + j 3, determine the following:
(a) 1/x; x2 ; x5 ; x2 ; x5
(b) y 1 ; y 2; y + y 2 ;
1
1+y
(c) z 1 ; z 10 ; (1 + z)2
(d) x + y; x + z; y + z; x y z;
x+z
;
xy
(z x)4
2. Determine each of the following summations in closed form (use Table 1 from the
Course Notes):
(a)
k=0
(.5)k
(b)
3
k=0
(.5)k
(c)
k=4
(.5)k
(d)
0
k=
(e)
10
k=1
(f)
25
k=5
k2
(g)
k=1
k (.9)k
(h)
k=10
k (.9)k
(i)
k=2
k 2 (.4)k
3k
3. Plot each of the following signals over the range indicated:

(a) w[n] = p10 [n 7]; 0 n 20
(b) x(t) = p10 (t 7); 0 t 20
(c) y[n] = (0.9)n u[n + 1]; 2 n 6.
(d) z(t) =
k=
(t 3k); 5.5 t 5.5
(e) a(t) = r(t) 2r(t 2) + r(t 4); 1 t 5
(f) b[k] = r[k] 3r[k 10] + 2r[k 16] + 4u[k 20]; 2 k 24
(g) c[n] = ejn + ej2n ; 0 n 10.
68
jn/4
4. Using the geometric series equation, evaluate the sum 7
. The answer is a
n=0 e
specic real-valued number. As a means of verifying your answer, in the complex plane,
plot the 8 values that are summed, and comment on why their positions lead to your
answer.
5. Plot the following two signals over the range n = 0, 1, 2, , 9. (Use Matlab if you
wish.)
(a) v1 [n] = 2 cos( 2 n)
10
(b) v2 [n] = 2 cos( 22 n)
10
Compare the plots, commenting on and explaining similarities and dierences.
6. Basic signal operators:
(a) Plot the discrete-time signal x[n] = u[n 2] u[n 6]. Express x[n] as a
shifted pulse (i.e. determine N and m for x[n] = pN [nm]) & as a sum of shifted
impulses.
(b) Given that z1 [n] = (0.5)n p4 [n], plot z2 [n] = z1 [4 n].
(c) Consider the CT signal
3
1
1
3
x(t) = r((t + 1)) p1 (t + ) + p1 (t + ) + 2 p1 (t ) r((t 2)) p1 (t ) .
2
2
2
2
i. Sketch x(t).
ii. Sketch x(2 t).
3
iii. Sketch x(t) [(t + 2 ) (t 3 )].
2
(d) Consider the DT signal

x[n] = [n + 4]
1
1
1
[n + 3] +
[n + 2] + p4 [n + 1] +
[n 3] .
2
2
2
i. Sketch x[n].
ii. Sketch x[n] u[3 n].
(e) Consider the signal x(t) = 3u(t + 2) 5u(t 1) + 3u(t 4). Sketch it.
d
Determine and sketch y(t) = dt x(t).
7. For each of the following DT signals, determine if it is periodic. If periodic, determine
its period.
(a) sin( n )
8
(b) cos( 6 n + 1)
7
(c) 2 cos( n) + sin( n) 2 cos( n +
4
8
2
(d) 3 cos( n) sin( 34 n)

3
18
)
4
69
8. Periodic signals:
(a) First plot the signal
x1 [n] = [n] [n 1] + [n 2]
over the range 4 n 6. Now consider the signal
x[n] =
k=
x1 [n 6k] .
Plot this signal over the range 0 n 15.
(b) Consider the periodic continuous-time signal

x(t) =
k=
x1 (t 2k) ;
x1 (t) = e|t| p2 (t) .
Sketch both x1 (t) and x(t).

(c) Sketch
x(t) =
k=
x1 (t 2k)
where
x1 (t) = 3 cos(t) p2 (t 1) .
9. Determine the energy and power for the following DT signals:
(a) x[n] =
1
u[n
n
j2n
(b) x[n] = e
3].
. x1 [n] = 3 ej/2 ej(3/7)n . x2 [n] = x1 [n] p20 [n].
(c) y[n] = n .5n u[n].

(d) x1 [n] = 3 (0.2)n4 u[n 4]. x3 [n] = 0.91/2 0.5n u[n]. x4 [n] = 1.2n u[n 20].
x5 [n] = 4 (0.01)n u[n + 3].
(e) x2 [n] = 3 cos(n) 2.
(f) x3 [n] =
i=
[n 2i].
(g) x[n] = .9|n| . (Hints: Draw x[n] to see what it looks like, take advantage of
symmetry, and use the geometric series formula.)
(h) y[n] = 2 (.8)|n| .
(i) x[n] = r[n] p20 [n].
10. Determine the energy and power for the following CT signals:
(a) w(t) = 2 p8 (t 100.2).
70
(b) x(t) = 5 et u(t 1).
(c) y(t) =
9
k=0
(d) z(t) =
k=
r(t k) p1 (t 1 k).
2
(e) x2 (t) = ej100t .
z1 (t 10k);
z1 (t) = t2 p10 (t 5).
11. Determine the energy and power for the following DT signals:
(a) v[n] = 5 ej((2/18)n+(/4)) .
(b) x[n] =
(c) x[n] =
k=
k=
x1 [n 20k] where x1 [n] = (.5)n p10 [n]
x1 [n 8k] where x1 [n] = 5 p3 [n].
(d) x[n] =
k=
(e) x[n] =
k=
(f) x[n] =
k=
(g) x6 [n] =
3
l=1
x1 [n 8k] where x1 [n] = [n] 3[n 2] + 4[n 5]
x1 [n 20k] where x1 [n] = r[n] p10 [n]
x1 [n 12k] where x1 [n] = n p10 [n 1].
l[n l].
12. Determine the energy and power for the following CT signals:
(a) y(t) =
n=
3 p2 (t 4n)
(b) x(t) =
k= x1 (t
(c) y(t) =
k= (t
(d) z(t)
(t)
2k) where x1 (t) = e|t| p2 (t)
k) p1 (t k)
=
k= (t
= lim0 1 p (t)).
k) (Hint: consider the denition of a CT impulse
13. Let N be some positive integer. Consider the DT energy signal

x1 [n] = r[n] pN [n]
and DT periodic power signal
x[n] =
k=
x1 [n kN] .
(a) For N = 6, plot x1 [n]; 2 n 8.
(b) For general N, determine the energy in x1 [n]. Use Table 1, p. 14 to determine
this expression in closed form.
(c) For general N, what is the power of x[n].
(d) For N = 22, what is the energy in x1 [n].
14. Write x(t) = cos2 (2t + ) as a linear combination of complex sinusoids.
2
15. Represent the signal as a linear combination of basic signals from the set:
71
(a) signal 1 t < 0

0t<1
1t<2
otherwise
1
2
x3 (t) =
1
set - all delayed steps
(b) signal
2t
0t<1
4 2t 1 t < 2
y1 (t) =
0
otherwise
set - all delayed ramps

y (t)
x (t)
2
1
0
(c) signal - x1 (t) = | 2 cos(0.5t) |;

set of all delayed x(t) where x(t) = cos(.5t)[u(t + ) u(t )].
16. Use Matlab to plot the following linear combinations:
(a) Let xk (t) = cos(2kt) for integer k be some basic signals. Consider
10
x(t) =
k=1
1
xk (t) .
k
Plot x(t); 0 t 4, using the Matlab plot function and a dense sampling of
t. What type of signal does x(t) look like? What do you think would happen if
we increased the upper summation bound?
(b) Let s(t) be a symbol shape used in an Amplitude Modulated (AM) digital communications scheme. Specically, let s(t) = cos(0 t) pT (t) where T is the
symbol duration and 0 is called the carrier frequency. For a particular digital
AM scheme, let T = 1, 0 = 10 and let ak {3, 1, 1, 3} be the possible transmission amplitudes (with four amplitudes, we can represent 2 bits per transmitted
symbol). The transmitted signal would be of the form
x(t) =
k=
ak s(t kT ) ,
1
i.e. a linear combination of delayed symbol shapes. So the symbol rate is T
2
and the bit rate is T . Let {a0 , a1 , a2 , a3 } = {1, 3, 1, 3}. Use Matlab to plot
x(t); 0 t 4 using the Matlab plat function and a dense sampling of t.
72
17. Finding the impulse response of simple DT IIR systems:
(a) Consider the discrete-time signal x[n] = [n], and the simple feedback system
y[n] = x[n] .9 y[n 1] ,
where x[n] is the input, y[n] is the output, and y[1] = 0. Determine an expression
for y[n]; n = 0, 1, 2, 3 .
(b) Consider the discrete-time signal x[n] = [n], and the simple feedback system
y[n] = y[n 1] + 0.5x[n] ,
where x[n] is the input, y[n] is the output, and y[1] = 0. Determine an expression
for y[n]; n = 0, 1, 2, ....
(c) Find the impulse response (i.e. the output when the input is an impulse and the
initial conditions are zero) of the following system:
y[n] y[n 1] = x[n] x[n 1] .
(d) Consider the DT system
y[n] = 0.8 y[n 2] + x[n]
with input x[n] = [n] and initial conditions y[1] = y[2] = 0. Determine an
expression for y[n] for all time 0 n , and plot y[n] for 0 n 10.
18. Finding the output of simple DT IIR systems:
(a) Consider the DT system
y[n] = 0.8 y[n 1] + x[n]
with input x[n] = [n] 0.8 [n 1] and y[1] = 0. Plot the output y[n] for
all time 0 n 10.
(b) Consider the discrete-time signal x[n] = p2 [n] as the input to a simple feedback
system
y[n] = x[n] .5 y[n 2] ,
where y[n] is the output. Assume initial conditions y[1] = y[2] = 0. Determine
and plot y[n] for 0 n 7 .
(c) Consider the DT system
y[n] = 0.5 y[n 1] + x[n 1].

For input x[n] = [n] + 0.5 [n 1] and y[1] = 0, determine the output y[n]
for all time n 0.
(d) Consider the DT system
y[n] = 0.5 y[n 1] + x[n 1].

For input x[n] = [n] 0.5 [n 1] and y[1] = 0, determine the output y[n]
for all time n 0.
73
19. Finding the output of DT FIR systems:

(a) Consider the discrete-time signal x[n] = 2(u[n]u[n5]), and the simple summer
5
y[n] =
k=0
x[n k] ,
where again x[n] is the input and y[n] is the output. Determine and sketch
y[n]; n = 0, 1, 2, 3, .
(b) Consider the DT system
y[n] = x[n] + .5 x[n 1] + .25 x[n 2] + .125 x[n 3] + .0625 x[n 4]

with input x[n] = [n] .5 [n 1].
i. Plot x[n] over 2 n 10.

ii. Determine and plot y[n] over 2 n 10.
(c) Consider the following FIR lter:
y[n] =
k=0
0.9k x[n k] .
a) Sketch the delay-line block diagram representation of this system.

b) Sketch the impulse response h[n], and for N = 4 determine its energy Eh .
c) Now let the input be x[n] = u[n]. Determine an expression for the output
y[n], and for N = 4 determine its energy Ey .
(d) Consider the following DT system with input x[n] = [n] 4[n 2] and output
y[n]
y[n] = x[n] 2x[n 1] + x[n 2] .
i. Describe the output y[n] for all time n. Plot it for 2 n 8.
ii. Determine the output energy.
(e) Consider an input x[n] = r[n] p4 [n] to the DT system
3
y[n] =
k=0
(1)k x[n k] .
i. Plot x[n] for 2 n 10.

ii. Determine the energy of x[n].
iii. Determine and plot the output y[n] for 2 n 10.
(f) Consider the FIR lter which is a simple 10-point summer:

9
y[n] =
k=0
x[n k] .
74
i. Determine the output when the input is the impulse [n] (i.e. nd the impulse
response). What is the energy and power of this output?
ii. Determine the output when the input is the step u[n] (i.e. nd the step
response). What is the energy of this output?
(g) Consider the FIR lter which is a simple 10-point summer:
4
(2)k x[n k] .
y[n] =
k=0
i. Sketch the delay-line block diagram (i.e. with the input delay line).
ii. Determine the output for input [n] (i.e. nd the impulse response).
iii. Determine the output for input x[n] = [n] + 2[n 1]. What is the energy
of this output?
20. Finding the output of simple DT FIR systems: Consider the DT FIR system
5
y[n] =
k=0
x[n k] .
The impulse response for this system is h[n] = p6 [n].

a) Describe for all time n the output y1 [n] due to input x1 [n] = [n].
What is the Energy E1 of y1 [n]?
b) Describe for all time n the output y2 [n] due to input x2 [n] = u[n].
c) Describe for all time n the output y3 [n] due to input x3 [n] = (1)n u[n].
21. Consider the following simple summer
5
y[n] =
k=0
x[n k] .
Determine the output for the following inputs:

(a) x1 [n] = 1 for all n.
(b) x1a [n] = 5 for all n.
(c) x2 [n] = ej(/3)n for all n.
(d) x3 [n] = ej(/2)n .
22. Consider the simple system (a summer)
4
y[n] =
k=0
x[n k] ,
where x[n] is the input and y[n] is the output.
75
(a) For input x[n] = r[n] p5 [n], determine and sketch y[n]. Determine Ey , the energy
of the output y[n].
(b) Consider input
x[n] = 5 ej2n/5 .
Determine the output y[n] for all time n. As a means of verifying your answer for
time n = 0, in the complex plane, plot the 5 values (phasors) that are summed,
and comment on why their positions lead to your answer. Determine Py , the
power of the output y[n].
23. Consider the following DT system:
9
y[n] =
k=0
.5k x[n k] .
(a) First let the input be x1 [n] = 1 for all time n. The output will be a constant, i.e.
y1 [n] = c1 . Determine c1 . (To obtain a simplied approximate answer, assume
that .510 0.) Is this input amplied or attenuated?
(b) Now let the input be x2 [n] = ejn for all time n. The output will in the form
y2 [n] = c2 ejn . Determine c2 . (Again, to obtain a simplied approximate
answer, assume that .510 0.) Is this input amplied or attenuated?
24. Consider a DT system with output y[n] = x[n + 2] + x[n 2] and input x[n] = ej0 n .
(a) The output is of the form y[n] = K ej0 n where K is a constant depending on 0 .
Determine K explicitly as a real-valued constant.
(b) For 0 = 0, determine y[n].
(c) For 0 = , determine y[n].
25. Consider the FIR lter:
10
y[n] =
k=0
(1)k x[n k] .
a) For input x[n] = ej0 n , determine an expression for the output which is in closed
form (i.e. do the geometric series).
b) Using results from a), determine the simplest expression you can for y[n] when
0 = 0.
c) Using results from a), determine the simplest expression you can for y[n] when
0 = 11 .
d) Determine the simplest expression you can for y[n] when 0 = . (Hint: either
be very careful using part a) results; or go back and use Eq (1) directly, noting
that x[n] = (1)n .)
76
21
y[n] =
k=0
(0.9)k x[n k] .
Let x[n] = ej0 n for all time n.

(a) Determine a closed form expression for y[n] (closed form means do the summation;
assume that .922 0.1.)
(b) Determine y[n] specically for 0 = 0 (i.e. for x[n] = 1).

(c) Determine y[n] specically for 0 = .
(d) For which frequency, 0 = 0 or 0 = , does the system amplify the input?
Attenuate?
(e) Determine y[n] specically for 0 = /2.
y[n] =
k=0
.8k x[n k] .

(a) Determine a closed form expression for y[n] (Closed form means do the summation.)
y[n] =
k=0
(0.9)k x[n k] .

(a) Determine a closed form expression for y[n] (closed form means do the summation.)
(d) For which frequency, 0 = 0 or 0 = , does the system amplify the input?
Attenuate?
(e) Determine y[n] specically for 0 = /2.
29. For each system below, determine which of the following properties hold: (1) memoryless, (2) time invariance, (3) linear, (4) casual, (5) stable.
77
(a) y(t) = x(t 2) + x(2 t).
(b) y(t) = cos(3t) x(t).

2t
(c) y(t) =
x( ) d .
30. For each of the following systems, with input x and output y, determine whether the
corresponding system is: linear; time-invariant; and causal.
(a) y[n] = n2 x[n + 1]
(b) y[n] = |x[n]|
(c) y(t) =
(d) x[n] =
t
0
x(t ) d
n+1
k=
x[n k]
31. We know the following I/O pairs for a CT linear system:

x(t) = ej2t
x(t) = ej2t
y(t) = ej3t
y(t) = ej3t
(a) Let x1 (t) = cos(2t). Determine the corresponding output y1 (t).

1
(b) Let x2 (t) = cos(2(t 2 )). Determine the corresponding output y2 (t).
32. Determine whether or not the following systems are stable:

(a) y[n] =
k=
(b) y[n] =
10
k
k=0 2
(c) y[n] =
k
k=0 2
.5|n| x[n k]
x[n k]
x[n k]
33. Consider a linear system. Say that, for any integer k, we know that input
xk (t) = cos((t k)) p1 (t k) results in output yk (t) = 2 p1 (t k).
(a) Sketch x0 (t) and y0 (t), and x1 (t) and y1 (t).
(b) Let x(t) = 3 | cos(t)|. Determine the resulting output y(t). Sketch both x(t)
and y(t) for 2 t 2.
34. The impulse response of a DT LTI system is h[n] = u[n]. Determine and plot the
output y[n] due to input x[n] = [n + 3] 2[n 1] + 2[n 4].
35. Consider a DT LTI system which for input x[n] = u[n] results in output y[n] = [n].
Determine the output y1 [n] do to the input x1 [n] = 3 p4 [n 6k]. Sketch y1 [n]; 0
k=0
n 20.
36. Consider the following DT LTI system y[n] = x[n] + .5x[n 1].
(a) First determine the response to input p2 [n]. Call this y2 [n].
78
(b) Now, in terms of y2 [n], determine an expression for the output y[n] due to input
x[n] =
(1)k p2 [n 2k] .
k=0
Plot this y[n] for n = 0, 1, 2, 3, 4.

3
37. Consider the CT LTI system that has response y1 (t) = p1 (t 1 ) p1 (t 2 ) to input
2
1
x1 (t) = p1 (t 2 ). Determine the response y(t) to input
1
3
5
x(t) = p1 (t ) + p1 (t ) + p1 (t ) .
2
2
2
Sketch y(t); 0 t 5.
38. Consider the following DT LTI system and input:

4
y[n] =
k=0
0.5k x[n k] ,
1
0
1
0
00 0
11 1
1 11
0 00
1 11 1 1
0 00 0 0
1
0
1 1 11 1 1
0 0 00 0 0
00 00 00
11 11 11
1
0
x[n]
1
0.5
0.5
a) Express the input x[n] as a weighted-sum of delayed impulses.

b) Let h[n] denote the system impulse response. For the given x[n], express the
output y[n] as a weighted-sum of delayed h[n].
c) Sketch the impulse response h[n] of the given DT LTI system.
d) Combining your results from b) & c), determine and plot the output y[n].
39. Match exactly one impulse response with each of the DT LTI system descriptions.
Impulse response h[n]
DT LTI system description
ha [n] = 0.5n ej(/2)n u[n]
a delay
hb [n] = u[n]
causal, stable IIR
hc [n] = ej(/2)n
an averager
hd [n] = 0.5n ej(/2)n u[n + 3]
non causal, stable IIR
he [n] = 10 [n]
non causal, non stable IIR
hf [n] = p7 [n]
a summer
hg [n] =
hh [n] = [n 3]
1
5
p5 [n]
IIR accumulator
an amplier
79
Time Domain Analysis of LTI Systems
In this Chapter we establish that for an LTI system the impulse response is an important
I/O description, and we show how it can be employed to compute output of a LTI system
given any input. We start with DT LTI systems, and then consider CT. For each, we rst
show how to represent any signal as a linear combination of delayed impulses. Recall that
in Section 1.3 we emphasized that expressing general signals as linear combinations of basic
signals was an important concept. Also recall that the rst two rows of Table 2 in Subsection
1.4.5 suggest that using a set of delayed impulses as basic functions would be useful. This
representation of a signal as a linear combination of delayed impulses is easier to visualize
for DT signals, which is why we cover DT rst.
We know from Chapter 1 that, for a LTI system, once we have the response to an impulse,
we can compute the response to any linear combination of delayed impulses. That is, for a
LTI system, the output due to a given linear combination of delayed impulses is computed as
the same linear combination of delayed impulse responses. This computation is a convolution.
So convolution is the primary topic of this Chapter. We will rst study the convolution sum
for DT LTI systems, and then the convolution integral for CT.
In Chapter 1 we considered numerous examples of computing the output of a LTI system
given its impulse response and input. In these examples we were eectively computing
convolutions. In this Chapter we formalize this procedure (convolution), focus on techniques
and properties that assist in its computation, and generalize its applicability. We also analyze
the impulse response as a useful characterization of a LTI system.
Convolution has important applications beyond LTI system I/O calculation. We will
see some of these applications later in the Course. The objectives in this Chapter are: to
become procient at performing convolution; and to understand how it is used to compute
LTI system outputs.
Be comfortable with the idea of expanding any DT signal as a linear combination of
delayed impulses.
Understand how the expansion of a DT signal as a linear combination of delayed
impulses leads to the convolution sum as a DT LTI system I/O computation.
Understand that the impulse response is an I/O characterization of a DT LTI system,
and realize how it is used to compute I/O (as a convolution of the impulse response
and the input).
Be procient with the graphical approach to the convolution sum.
Be procient with the analytical approach to the convolution sum.
Be able to use convolution sum and DT LTI system properties to assist in performing
a convolution sum.
Have a mature ability to perform convolution sums, such that you can pick the best
approach or combination of approaches for a given problem.
80
Understand how the impulse response of a DT LTI system can be used to check for
causality and stability.
Be familiar with combining DT LTI systems impulse responses for parallel and cascaded
interconnections of subsystems.
Extend all of the objectives above to CT signals and systems.
2.1
DT Signal Impulse Expansion
Consider an arbitrary x[n] shown in Figure 38.

x[n]
x[2]
....
x[0]
x[1]
x[6]
x[3]
....
n
x[5]
x[1]
......
....
x[4]
x[1] [n+1]
x[1]
....
n
x[0] [n]
x[0]
....
....
n
x[1] [n1]
....
....
n
......
x[1]
Figure 38: Representation of a DT signal as linear combination of delayed impulses.

By direct observation of Figure 38, we see that any x[n] is easily represented as a linear
combination of delayed impulses {[n k]; k = , , 1, 0, 1, , }. Mathematically,
x[n] =
k=
Examples:
(a) u[n] =
k=0
(b) an pN [n] =
[n k].
N 1
k=0
ak [n k].
x[k] [n k] .
(1)
2.2
81
DT LTI System I/O Calculation: The Convolution Sum
With Chapter 1 of this Course behind us, we are familiar with the impulse response h[n] of
DT LTI systems. We were able to identify h[n] for some relatively simple systems. Later
in the Course we well see how to systematically determine h[n] for any DT LTI system,
regardless of how complex it is. For now, assume that for any DT LTI system we are
interested in, we have h[n].
2.2.1
The Impulse Response of a DT LTI System & Convolution Sum
Figure 99 develops the convolution sum of the input and the impulse response as the timedomain I/O calculation of a DT LTI system.
[n]
DT LTI system
impulse resp. h[n]
[nk]
x[k]
h[n]
(by the TI property)
[nk]
x[k]
(by the LTI properties)
[nk]
Figure 39: A DT LTI system and the convolution sum.

So, from the box in Figure 99, the I/O calculation for any DT LTI system is
y[n] =
k=
x[k] h[n k] .
(2)
This is the convolution sum of input x[n] with impulse response h[n]. Since h[n] is all we
need to know about the DT LTI system to compute the output, the impulse response h[n] is
a very useful I/O characterization. From Chapter 1, we already have some experience with
nding the impulse response of simple LTI systems. Later in the Course we will establish a
general approach for identifying the impulse response which can be systematically applied
to any LTI system, no matter how complicated it is. In this Chapter we will assume that
we know the impulse response.
Considering Eq (2), with the change of variables k n k, we equivalently have
y[n] =
k=
h[k] x[n k] .
(3)
Since the convolution of two signals, in this case an input x[n] and an impulse response h[n],
is such an important operation, it has its own notation. We represent Eqs (2,3) as
y[n] = h[n] x[n] = x[n] h[n]
(4)
Note that in Eqs (2,3) the index k can be thought of as memory time. That is, the output
at time n is in general eected by the input for all memory time k.
82
Graphical Approach to the Convolution Sum:
Recall that in Chapter 1 we spent time practicing a number of simple DT signal operators,
including: signal fold, signal shift, signal fold and shift, signal multiplication, and signal
accumulation. Now we see why. Referring to Eq (2) (or equivalently Eq (3)), the convolution
sum involves the following procedure (stated for Eq (2):
1) Plot h[k]
2) Fold x[k] to form x[k], e.g. x[n k]; n = 0
3) For each n:
a) Plot x[n k], formed by shift of x[k] by n (e.g. for positive n, shift right)
b) Multiply h[k] with x[n k] (point-by-point in memory time k)

c) Accumulate h[k] x[n k] over all k to form y[n]
4) Compile y[n] results for all n of interest.

This is termed the graphical approach because it employs plots to visualize the computation
and assist in the details involved in performing a convolution by hand.
To begin getting comfortable with computing the convolution sum, we now consider a
series of fairly straightforward graphical convolution examples (Examples 2.1-2.5).
Example 2.1: Given an FIR lter with impulse response h[n] =
x[n] = p2 [n], determine the output y[n] for all n.
1
5
p5 [n] and input
Solution: Consider the gure below. The rst column shows h[k], x[n k] n = 0,
and h[k] x[0 k]. That is, it shows the fold, the shift (by n = 0) and the
multiplication. To compute y[0], the last step is to accumulate across h[k]x[0k].
The result, y[0] = 1 , is shown in the plot on the right. The second column shows
5
the similar procedure for computing y[1]. Note that x[k] is now shifted one to
the right, forming x[n k]; n = 1. The result is y[1] = 2 .
5
For other n, try to visualize x[n k] and h[k] x[n k]. In particular, focus on
visualizing, for a given n, the range of k for which the multiplication h[k]x[nk]
is nonzero. This is the range of k where nonzero h[k] and x[n k] overlap. Then
see if you can mentally identify the accumulation, which is y[n]. The objective is
to get familiar enough with this graphical approach so that you can get by with
plotting only h[k] and x[n k]; n = 0, and then visualize the rest.
h[k]
h[k]
1/5
1/5
x[nk]; n=0
x[nk]; n=1
y[n]
2/5
1/5
1/5
sum
0
h[k] x[1k]
h[k] x[0k]
1/5
sum
0
Figure 40: Example 2.1 convolution sum illustration.
83
Example 2.2: Use the graphical convolution approach to convolve h[n] = 1 p8 [n]
8
and x[n] = [n] [n 1].
Solution:
Example 2.3: Convolve h[n] = p5 [n] with x[n] = p10 [n].

Solution: Lets fold and shift x[n]. In the gure below, x[k] and h[n k]; n = 0
are drawn.
For n < 0, h[n k] is shifted left. There is no overlap. x[k] h[n k] = 0 for all
k. So, y[n] = 0; n < 0.
For n = 0 we see from the plots that x[k] and h[n k] overlap at k = 0 only.
x[k] h[n k] = 0 except at k = 0 where the product is 1. Summing across this,
we get y[0] = 1.
For n > 0, we see that initially there is more and more overlap of nonzero x[k]
and h[n k], so y[n] increases linearly (i.e. since all nonzero values of x[k] and
x[n k] are equal to 1) until when there is complete overlap of nonzero values
y[n] levels o. Eventually, we start shifting x[n k] past x[k] so y[n] decreases.
Finally, for n > 13 (i.e. for (n 9) > 4), there is no overlap so y[n] = 0. The las
plot in the gure below shows y[n].
x[k]
1
h[nk]; n=0
1
n4
y[n]
5
1
0
13
84
Example 2.4: Using the graphical approach, nd h[n] = h1 [n] h2 [n] where
h1 [n] = 2n p10 [n] and h2 [n] = 2n p10 [n]. Specically,
1. Sketch h1 [k] and h2 [2 k].
2. Sketch h1 [k] h2 [2 k].
3. Determine h[2].
4. Determine, for all n, h[n].

Solution: h1 [k], h2 [2 k] and h1 [k] h2 [2 k] are shown in part a) of the gure
below. h1 [k] h2 [2 k] is shown in part b). As indicated in part c) of the gure,
h[2] is computed as accumulation of h1 [k] h2 [2 k]. The result is h[2] = 12.
a)
d)
h 1 [k]
h 1 [k]
512
....
2
1
1
h 2 [2k]
2
....
2
1
1
h 1 [k] h 2 [2k]
n=0
h 1 [k] h 2 [nk]; 0 < n < 9
....
2
1
1
h[2] =
n 9= 9
c)
512
2
1
b)
h 2 [nk]; n=0
nk
512
....
2
1
....
512
h 1 [k] h 2 [2k]= 4 + 4 + 4 = 12
k

Part d) of this gure shows the plots that, hopefully with practice, will be sucient to determine h[n] for all n. Note the this rst plot in part d) is just h1 [k]
(the signal not to be folded) plotted as a function of k. Note that the nonzero
values of h1 [k] are labeled with their values. Then h2 [n k] (the signal is which
folded and shifted) is plotted as a function of k for n = 0. Again note that
its nonzero values are labeled with their values. Also note that the leading an
lagging edges of the signal are labeled (as a function of n). All these labels assist
in determining the solution.
Now visualize shifting the signal in this plot left (for n < 0) and right (for n > 0),
and for dierent shifts identify the range of k where h1 [k] and h2 [n k] overlap.
For each n, this will be the range of k that must be accumulated. For example:
For n < 0 there will be no overlap, so h1 [k] h2 [n k] will be zero for all k,
so its accumulation (i.e. h[n]) will be zero.
85
or 0 n 9, the last plot shows h1 [k] h2 [n k]. We see that, for this range
of n, h1 [k] and h2 [n k] will overlap from k = 0 ((when h1 [k] turns on) to
k = n (after which h2 [n k] turns o). We see this by observing where the
signal edges are, as a function of k, which are identied in the plots. Noting
what the h1 [k] and h2 [n k] values are for this range of k (i.e. from the plot
labels h1 [k] = 2k and h2 [n k] = 2nk ), we have that h1 [k] h2 [n k] = 2n
over its nonzero range. Accumulating across these nonzero values, we have
that h[n] = (n + 1) 2n .
For 10 n 18, the overlap of nonzero h1 [k] and h2 [nk] is from k = n9

(when h2 [n k] turns on) till k = 9 (after which h1 [k] turns o). Again
h1 [k] h2 [n k] = 2n over its nonzero range. Thus, h[n] = (9 (n 9) +
1) 2n = (19 n) 2n over this range.
Finally, for n > 18 h[n] = 0 since there is no overlap of nonzero h1 [k] and
h2 [nk]. This is because the lagging edge of h2 [nk] (which is at k = n9)
we greater that the leading edge of h1 [k] (which is at k = 9).
In summary, we have that

0
(n + 1) 2n
h[n] =
n
(19 n) 2
n<0
0n9
10 n 18
n > 18
(5)
Note that in Example 2.4 we are dealing with signals h1 [n] and h2 [n], the implication being
that these are not necessarily the input and impulse response of a DT LTI system. This
choice of signal notation drives home the point that the convolution sum is a general operation
applied to two signal that has utility that goes beyond an I/O calculation for DT LTI systems.
If you can follow Example 2.4, you should be able to handle any graphical convolution,
regardless of its complexity. Again, note the utility of labeling the nonzero values of the
plotted signals (e.g. h1 [k] and h2 [n k] in Example 2.4) and labeling the leading and lagging
edges of these signals. Also note that the advantage of this graphical approach to convolution
stems from the fact that it assists in determining: 1) the dierent regions of n where there
are dierent types of overlap (over k); and 2) what range of overlap in k there is for each
region of n.
Example 2.5: Convolve h[n] = p10 [n] with x[n] = p15 [n 5], rst graphically then
using Matlab.
Solution: This Example is a lot like Example 2.3. The pulses are of dierent
width, and one is delayed, but you can still expect the result to look like a
triangle with its top chopped o (as in Example 2.3). The graphical approach
can be used to identify the details of the result (i.e. when the triangle turns on
an o, and where the top is chopped o). Try this yourself, and compare your
result to the Matlab generated plots below.
86
echo on
%
%
Matlab Convolution Sum Demo
%
h[n] = p_10 [n]; x[n] = p_15 [n-5]
%
clear all
clf
clear function
%
% First just do the convolution, using the "conv" command
%
nx=(-30):30;
h=zeros(1,61);
h(31:40)=ones(1,10); % h[n] is now generated for n=(-30):30
pause
%
subplot(311)
stem(nx,h)
xlabel(sample time - n)
ylabel(h[n])
text(-28,0.8,Robbie Robertson)
pause
%
x=zeros(1,61);
x(36:50)=ones(1,15); % x[n] is now generated for n=(-30):30
pause
%
subplot(312)
stem(nx,x)
ylabel(x[n])
text(-28,0.8,Rick Danko)
pause
%
y=conv(x,h);
pause
%
subplot(313)
stem(nx,y(31:91))
ylabel(y[n])
text(-28,9,Ronnie Hawkins)
pause
%
% Clear figure, and do it again as the input shifts through
87
%
clf
pause
subplot(311)
stem(nx,h)
xlabel(memory time - k)
ylabel(h[k])
text(-28,0.8,Roger Waters)
pause
%
for i=1:30;
n=i-1
xfold=zeros(1,61);
xfold(n+12:n+26)=ones(1,15);
subplot(312)
stem(nx,xfold)
xlabel(memory time - k)
ylabel(x[n-k]; n=0)
text(-28,0.8,Stephen Hawking)
%
ytmp=zeros(1,61);
ytmp(1:31+n)=y(31:61+n);
subplot(313)
stem(nx(1:31+n),ytmp(1:31+n))
axis([-30 30 0 10])
ylabel(y[n])
text(-28,9,Lauren Visualization)
pause
end
Figure 43 show the plots generated by this demo. Note that the impulse response represents the memory structure of the system Observe how the input x[n] is presented to the
system folded and shifted through its memory structure. The input in the system memory
is multiplied by the impulse response weights and summed to form the output. Thus, the
convolution sum is a fold, shift, multiply and accumulate operation. The Lauren Visualization plot shows the output computed up through sample time n = 20. This code
will be run in class. You should try it yourself.
88
1
h[n]
Robbie Robertson
0.5
0
30
20
10
0
sample time n
10
20
30
10
0
sample time n
10
20
30
10
0
sample time n
10
20
30
10
0
memory time k
10
20
30
10
0
memory time k
10
20
30
10
0
sample time n
10
20
30
x[n]
Rick Danko
0.5
0
30
20
10
y[n]
Ronnie Hawkins
0
30
20
h[k]
Roger Waters
0.5
0
30
20
x[nk]; n=0
1
Stephen Hawking
0.5
0
30
20
10
y[n]
Lauren Visualization
0
30
20
Figure 43: Plots generated by the Convolution Sum Demo.
89
As already noted, these rst convolution sum examples were solved using the graphical
approach. This is as opposed to several other approaches that we will learn about below. In
the end, you will be familiar with a number of approaches that can be used separately or in
combination to solve any convolution sum problem. Before we move on with this, lets step
back and make sure we have an appreciation of the physical relevance of convolution. We
do this be considering an FIR lter.
Figure 44 illustrates an FIR lter. It consists of a delay line which stores past input values,
multipliers which each weight a stored value of the input, and summers which accumulate
the multiplier outputs to produce the system output. Each sample time, the input is shifted
through the delay line, and an output is computed. The multiplier values are the nonzero
values of the FIR lter impulse response.
x[n1] x[n2]
x[n]
D
h[0]
D
h[1]
x[n(M1)]
D
h[2]
x[nM]
D
h[M1] h[M]
y[n]
Figure 44: The FIR lter structure (implementing a convolution sum directly).
Recall Example 1.26 which followed the introduction of the FIR lter. We saw that the
FIR lter I/O equation is
M
y[n] =
k=0
h[k] x[n k] .
(6)
In Example 1.26 we used plots to assist us in visualizing how input data ows through the FIR
lter and is operated on to compute the output. We observed that the input data appeared
to be folded and then shifted across the lter impulse response. We then multiply and
accumulate. So the FIR lter computation is fold, shift, multiply, accumulate. Note
that Eq (6) is a special case of the convolution sum (for which the range of memory k is
nite). So the FIR lter is a direct computation of the convolution sum for a DT system
with impulse response. The FIR lter provides us with a good visualization of convolution in
that it illustrates how a LTI system employs memory of the input signal in the computation
of the output signal. Run the Example 2.5 Matlab demo, step-by-step, imagining the folded
input shifting through the FIR lter register, and being tapped o of, and then multiplied
by summed to form outputs.
The convolution sum I/O expression derived at the beginning of this Section indicates
that all DT LTI systems, whether FIR or IIR, can be interpreted as input data being folded
and shifted through the a system memory structure, with the impulse response governing
how that input in memory is linearly combined to form the output. We will see that this
same interpretation applies to CT LTI systems.
90
The convolution sum examples considered to this point have involved two nite duration
signals. The next two examples consider some innitely long signals. If anything, innitely
long signals can be easier to graphically convolve because they result in fewer regions of n
for which dierent ranges of summation over k must identied and solved.
Example 2.6: For a DT LTI system with impulse response h[n] = an u[n], with
a = 0.9, determine the output due to input x[n] = [n] .9[n 1].
Solution: For the graphical approach, we choose to fold and shift x[n]. The gure
below shows h[k] and x[n k]; n = 0 as functions of k.
h[k]
0.9
.....
y[n]
k
.....
x[nk]; n=0
0
n1
n
0.9
Visualizing the shift of x[n k], we get
0
1
y[n] =
0.9 0.9 = 0
0.9n 0.9n = 0
n<0
n=0
n=1
n1
= [n] .
How does this Example relate to the result using the second equalizer considered
in Practicum 1?
Example 2.7: Convolve h[n] = an u[n] with x[n] = bn u[n]. Assume that a = b,
and for plotting purposes that 0 < a, b < 1.
Solution:
91
Example 2.8: Convolve h[n] = an u[n] with x[n] = p10 [n + 2].

Solution:
For the next four examples, it is easier to use the convolution sum expression directly. This
is called the analytical approach.
Example 2.9: Convolve h[n] = an u[n] with x[n] = ej0 n . Assume |a| < 1 (so the
the innitely long geometric series summation involved in the solutions can be
solved). Note that for this example you will probably nd it easier to fold and
shift x[n]. However, you might try it both ways.
Solution:
Example 2.10: Convolve h[n] = p10 [n] with x[n] = ej0 n . First determine the
answer for general 0 , and then specically for 0 = 2 .
10
Solution:
y[n] =
k=
h[k] x[n k] =
ej0 (nk) = ej0 n

k=0
ej0 k
k=0
j0 10
1 e
.
1 ej0
From this general solution, for 0 = 2 we have that
10
= ej0 n
1 1
1 ej(2/10)10
= ej(2/10)n
= 0.
j(2/10)
1 e
1 ej(2/10)
Note that this 10-point summer completely attenuates a complex-valued sinusoid
of frequency 0 = 2 . What other frequencies are completely attenuated?
10
y[n] = ej(2/10)n
92
Example 2.11: Lets go back an solve Example 2.7, this time using the analytical
approach. Convolve h[n] = an u[n] with x[n] = bn u[n]. Assume that a = b.
Solution:
Consider the convolution of a DT LTI system impulse response h[n] = [n2] with a general
input x[n]. The output is
y[n] = [n 2] x[n] =
k=
x[k] [n k 2] .
For each n, [n k 2] is zero for all k except for k = n 2. Thus, for each n, the innite
sum has only one nonzero entry, the k = n 2 term, which is x[n 2]. Thus,
y[n] = x[n 2] .
Note that we have identied the impulse response of a new type of simple DT LTI system.
The impulse response of an m sample delay is simply h[n] = [n m].
93
In the next example, we take advantage of a previous result (specically from Example 2.7).
Example 2.12: Find the step response of the simple IIR lter impulse response
h[n] = an u[n].
Solution:
In the next two examples we take advantage of both previous results and properties of the
LTI system that the convolution calculation represents.
Example 2.13: Determine the y[n] of the LTI system with h[n] = an u[n] due to
input x[n] = bn5 u[n 5]. Use the Example 2.7 result and exploit system time
invariance.
Solution: This example is like Example 2.7, except the input is delayed got
m = 5. Since the system is time invariant, the output is just the output from
Example 2.7 delayed by m = 5.
y[n] =
bn4 an4
u[n 5] .
b a
Example 2.14: Determine the output y[n] of the LTI system with h[n] = an u[n]
due to input x[n] = bn p20 [n]. Use the Examples 2.7,2.13 results and exploit
system linearity.
Solution: First note that
x[n] = bn u[n] bn u[n 20] = bn u[n] b20 bn20 u[n 20] .
Then, exploiting the linearity and time invariance properties of the system,
n19
an19
bn+1 an+1
20 b
u[n] b
u[n 20] .
y[n] =
b a
b a
94
2.2.2
Approaches for the Calculation of the Convolution Sum
With the examples presented above, we have already been exposed to several approaches to
computing convolution sums. We now organizes these and a few others into a list.
1) Graphical: See Examples 2.1-2.7
2) Analytical: See Examples 2.8-2.11.
3) Convolution table: i.e. use previously derived results.
4) Linearity & time invariance properties of DT LTI systems: See Examples 2.13-2.14.
5) Convolution properties: This is the topic of Subsection 2.2.3.
6) Any combination of techniques 1) through 5).
7) Numerical: When implementing a DT LTI system, the output is computed numerically,
either directly (as illustrated in the Convolution Demo above), of eectively (e.g. when
an IIR lter is implemented).
Table 3 provides a list of convolution sum results that can be used with approaches 3-5).
Table 3: Convolution Sum Results.
x
y = x h
h
bn u[n];
bn+1 an+1
ba
an u[n]
an u[n]
an u[n]
(n + 1) an u[n]
pN [n]
pN [n]
tri2N [n] = r[n + 1] 2r[n (N 1)] + r[n (2N 2)]
pN1 [n]
pN2 [n]; N1 = N2
triangle w/ top o (see Exs. 2.3,2.5)
[n m]
[n p]
[n (m + p)]
an u[n]
[n] a[n 1]
[n]
ej0 n
an u[n]
ej0 n
1
1 a ej0
ej0 n
pN [n]
ej0 n
1 ej0 N
1 ej0
b=a
u[n]

2.2.3
95
Convolution Sum Properties
We have already exploited the linearity and time invariance properties of DT LTI systems
in solving convolution sum problems. In this Subsection we introduce four convolution sum
properties that can be used to the same advantage. The proofs of these properties are given.
You are not responsible for these proofs, but following them will reinforce mathematical
techniques that are often used in practice.
Commutativity:
h[n] x[n] = x[n] h[n] .
(7)
Proof: Starting with the convolution sum expression rst derived at the beginning
of Section 2.2, and changing the variable as m n k, we have
h[n] x[n] =
=
k=
m=
h[k] x[n k]
h[n m] x[m] =
m=
x[m] h[n m] = x[n] h[n] .
We had already established this, without a formal proof, at the beginning of Section 2.2
when we developed the convolutions sum as a DT LTI system time domain I/O calculation.
In terms of the procedure for computation a convolution sum, it means that it does not
matter which signal is folded and shifted.
Distributivity:
{a1 x1 [n] + a2 x2 [n]} h[n] = a1 {x1 [n] h[n]} + a2 {x2 [n] h[n]} .
(8)
Proof: Using the linearity property of addition, we have

{a1 x1 [n] + a2 x2 [n]} h[n] =
=
k=
{a1 x1 [n] + a2 x2 [n]} h[n k]

a1 x1 [n] h[n k] +
k=
= a1
k=
k=
x1 [n] h[n k] + a2
a2 x2 [n] h[n k]
k=
x2 [n] h[n k]
= a1 {x1 [n] h[n]} + a2 {x2 [n] h[n]} .

This distributivity property of convolution is an extension of the linearity property of DT
LTI systems. Note that, because of commutativity, we also have that
x[n] {a1 h1 [n] + a2 h2 [n]} = a1 {x[n] h1 [n]} + a2 {x[n] h2 [n]} .
(9)
The distributivity property expands in an obvious manner to the convolution of any two
linear combinations of any number of signals. We will see in Section 2.3 that this property
is important when dealing with a parallel interconnection of DT LTI subsystems.
96
Delay: Given that y[n] = x[n] h[n], then
x[n m] h[n p] = y[n (m + p)] .
(10)
Proof: Let x1 [n] = x[n m] and h1 [n] = h[n p].
x[n m] h[n p] = x1 [n] h1 [n] =
=
=
k=
l=
k=
x1 [k] h1 [n k]
x[k m] h[n p k] =
x[l] h[n p l m] =
l=
l=
x[l] h[n p l m]
x[l] h[n (p + l) m]
= y[n (p + m)] .
To get the 2-nd sum in the 2-nd line in the above equation we use the substitution
l = k m.
This delay property of the convolution sum is an extension of the time invariance property
of DT LTI systems.
Associativity:
{x[n] h1 [n]} h2 [n] = x[n] {h1 [n] h2 [n]} .
(11)
Proof:
x[n] h1 [n] =
v1 [n] h2 [n] =
=
k=
m=
m=
x[k] h1 [n k] = v1 [n]
v1 [m] h2 [n m]
k=
k=
x[k] h1 [m k]
x[k]
m=
k=
x[k]
l=
h2 [n m]
h1 [m k] h2 [n m]
h1 [l] h2 [(n k) l]
= x[n] {h1 [n] h2 [n]} .
In going from the 3-rd to 4-th line above, we use the substitution l = m k. In
going from the 4-th to 5-th line, dene v2 [n k] =
l= h1 [l] h2 [(n k) l],
so v2 [n] = h1 [n] h2 [n].
Generally, when sequentially convolving any number of signals, the order in which they are
convolved does not eect the overall result. The notation chosen in Eq (11) implies a cascade
interconnections of systems h1 [n] and h2 [n], i.e. the output of subsystem h1 [n], due to input
x[n], is the input to subsystem h2 [n]. We will see in Section 2.3 that this commutativity
property is important when dealing with a cascaded interconnections of DT LTI subsystems.
97
Example 2.15 (Delay Property): Given y[n] = p10 [n] p20 [n] as shown below,
determine and sketch y1 [n] = p10 [n + 12] p20 [n 5].
y[n]
10
1
0
20
29
Figure 45: Given convolution result for Example 2.15.

Solution:
Example 2.16 (Distributivity/Delay): Using the 1-st entry of Table 3, determine

y[n] = x[n] h[n] where x[n] = p5 [n] and h[n] = bn u[n].
Solution:
Example 2.17 (Distributivity/Delay): Determine y[n] = x[n] h[n] where x[n] =

p5 [n] and h[n] = bn u[n 5].
Solution: This is just the Example 2.16 result, delayed by 5 and multiplied by
b5 . That is,
y[n] = b5
bn4 1
bn9 1
u[n 5] b5
u[n 10] .
b 1
b 1
98
Example 2.18 (Distributivity): Determine the output of an N-point averager

given input x[n] = cos(0 n).
1
Solution: The impulse response of an N-point averager is h[n] = N pN [n]. By
1 j0 n
1 j0 n
and Eulers identity, x[n] = 2 e
+ 2 e
. Using Table 3 Entry # 8, and
distributivity, we have,
y[n] =
1
2
ej0 n
1 1 ej0N
1 1 ej0 N
+ ej0 n
N 1 ej0
N 1 ej0
Note that the two additive terms in the solution form a complex conjugate pair.
Thus the form of this solution can be simplied into a real-valued expression,
which should be expected since both the input and impulse response are realvalued. Instead of pursuing this simplied form here, we will postpone this issue
until Subsection 2.3.3 below, where we specically study the sinusoidal response
of DT LTI systems. Specically, we will revisit this problem in Example 2.27.
Example 2.19: Consider using the Matlab conv function to convolve an FIR lter
impulse response h[n] which is nonzero only over the range 0 n N 1 with
an indenitely duration signal x[n] that turns on at time n = 0.
Solution: This can be be accomplished by breaking x[n] into contiguous blocks
of length P as follows:
x[n] = x1 [n] + x2 [n P ] + x3 [n 2P ] +
x[n (i 1)P ] n = 0, 1, , P 1
xi [n] =
0
otherwise
i = 1, 2,
We can now use conv to convolve each xi [n] with h[n] to form
yi [n] = xi [n] h[n]
i = 0, 1, 2,
By the linearity and delay properties, the overall FIR lter output is then
y[n] = y1 [n] + y2 [n P ] + y3 [n 2P ] +
This approach is called the overlap-and-add method of FIR lter implementation,

so called because the outputs of the dierent input blocks, which are overlapped
in time, are added together. Figure 46 illustrates this for N = 8 and P = 17.
99
OverlapandAdd Convolution Illustration

h[n]
1
0.5
y3[n]
y2[n]
y1[n]
x3[n]
x2[n]
x1[n]
x[n]
0
10
10
10
20
30
40
50
60
10
20
30
40
50
60
10
20
30
40
50
60
10
20
30
40
50
60
10
20
30
40
50
60
10
20
30
40
50
60
10
20
30
40
50
60
10
20
30
40
50
60
10
20
30
40
50
60
0
10
10
10
0
10
10
10
0
10
10
10
0
10
10
50
0
50
10
50
0
50
10
50
0
50
10
50
y[n]
0
50
10
time (n)
Figure 46: Illustration of the overlap-and-add convolution sum calculation.
100
2.3
DT LTI System Issues
In Section 1.5 of this Course we considered four system properties: linearity, time-invariance,
causality and stability. In this Section we rst relate these properties to the impulse response
h[n] of a DT LTI system. Concerning the system interconnections described earlier in Subsection 1.5.1, we then consider relationships between the overall impulse response of a DT
LTI system and those of the constituent DT LTI subsystems.
2.3.1
System Properties
By assumption, a DT LTI system is linear and time-invariant. Recall that one thing this
means is that if we are given a DT LTI system description in the form of a LCC dierence
equation, we know that the system is liner and we know it is time-invariant. So we need
only consider causality and stability.
Causality: Generally, a DT LTI system,
y[n] =
k=
h[k] x[n k]
(12)
uses past, present and future inputs to compute an output. This is clear from this general
convolution I/O expression. However, for
h[n] = 0
we have that
y[n] =
k=0
n<0
h[k] x[n k] ,
(13)
(14)
which uses only past and present inputs to compute the output. Thus a DT LTI system is
causal if and only if
h[n] = 0
n<0 .
(15)
Is an FIR lter Causal? The answer is yes, usually. That is, usually an FIR lter is
characterized as a DT LTI system that has an impulse response with h[n] = 0 only for
n = 0, 1, 2, , N 1, even though, generally speaking, a nite impulse response lter impulse
response my range over any nite interval of n. Under the usual characterization of an FIR
lter,
y[n] =
N 1
k=0
h[k] x[n k] ,
and the system is causal. Thus the answer is a qualied yes.
(16)
101
Stability: Recall that for BIBO stability, for any bounded x[n], i.e.
|x[n]| B1 <
all n
(17)
all n .
(18)
the output y[n] is bounded, i.e.

|y[n]| B2 <
Consider any DT LTI system. We have that
|y[n]| =
h[k] x[n k]
(19)
| h[k] x[n k] |
(20)
k=
k=
k=
| h[k]| |x[n k] |
(21)
For worst the case bounded input, i.e. for |x[n]| = B1 , we then have that
|y[n]|
k=
= B1
| h[k]| B1
k=
| h[k]| .
(22)
(23)
From this we conclude that |y[n]| is bounded if and only if
k=
| h[k]| < .
(24)
This is a very useful result, in that for DT LTI systems it provides an simple test for stability.
Later in the Course we will see that it also contributes to the importance of another LTI
description in terms of poles and zeros.
Example 2.20: Is the DT LTI system with h[n] = .5|n| causal? Stable?
Solution:
102
Example 2.21: We have already noted that an FIR lter (i.e. with impulse
response h[n] restricted to be nonzero only over n = 0, 1, 2, , N 1) is causal.
Will an FIR necessarily be stable?
Solution: The answer is yes, since by assumption none of the impulse response
values h[n] = 0 only for n = 0, 1, 2, , N 1 are innite in magnitude. That
is, we can always assume an FIR lter is stable. That is one of its key advantages
over an IIR lter.
2.3.2
DT LTI Subsystem Interconnections
Complex systems are often constructed as interconnections of simpler subsystems. Here,

for several common types of interconnections, we establish rules for identifying the impulse
response of a DT LTI system in terms of the the impulse responses of its component DT
LTI subsystems.
Parallel Interconnections:
y [n]
1
h 1 [n]
y [n]
2
....
y[n]
....
x[n]
h 2 [n]
hM[n]
y [n]
M
Figure 47: Parallel interconnection of DT LTI subsystems.

For M = 2,
y[n] =
=
=
=
y1 [n] + y2 [n]
x[n] h1 [n] + x[n] h2 [n]
x[n] {h1 [n] + h2 [n]}
x[n] h[n] ,
by distributivity
where h[n] = h1 [n] + h2 [n]. Generally, given M DT LTI subsystems in parallel, the overall
system is DT LTI with impulse response
M
hi [n] .
h[n] =
i=1
(25)
103
Cascade (Series) Interconnections:
h 1 [n]
v [n]
v [n]
h 2 [n]
hM[n]
....
v [n]
x[n]
y[n]
Figure 48: Cascade interconnection of DTLTI subsystems.

For M = 2,
y[n] =
=
=
=
v[n] h2 [n]
{x[n] h1 [n]} h2 [n]
x[n] {h1 [n] h2 [n]}
x[n] h[n] ,
by associativity
where h[n] = h1 [n] h2 [n]. Generally, given M DT LTI subsystems in cascade, the overall
system is DT LTI with impulse response
h[n] = h1 [n] h2 [n] hM [n] .
(26)
Feedback Interconnections: This representation of DT LTI systems, in terms of impulse

responses, does not easily lead to an overall system I/O expression in terms of the individual
subsystems.
Example 2.22: Determine the impulse responses of the combined channel/equalizer
for the channel and each of the equalizers considered in Practicum 1.
Solution:
104
Example 2.23: Consider the following block diagram. Determine the overall
system impulse response.
Figure 49: Example 2.23 interconnection of DTLTI subsystems.
h 2 [n] = 1 p5 [n]
5
x[n]
h 1 [n] = a u[n]
y[n]
+
Solution:
Example 2.23 illustrates that the parallel and cascade interconnection rules can be used in
sequence to systematically identify the impulse response of any complex parallel/cascade
interconnection of DT LTI subsystems.

2.3.3
105
DT LTI System Response to Sinusoidal Inputs
In Examples 2.8, 2.9 & 2.18, as illustrations of the analytical approach and convolution
properties approaches to performing a convolution sum, we considered applying sinusoidal
inputs to a DT LTI system. Here, through a sequence of examples, we more systematically
focus on how DT LTI systems process sinusoidal inputs.
1
Example 2.24: Consider an FIR lter with impulse response h[n] = 16 p16 [n] (i.e.
a 16-point averager). Determine its response to complex-valued sinusoidal input
x[n] = 1 + ej(25/16)n .
Solution: This problem is similar to Example 2.18 in that we have the superposition of two complex-valued sinusoids (in this case one is DC i.e. 0 = 0) into
a DT LTI system (in both cases a 16-point averager). Paralleling the Example
2.18 solution,
y[n] =
15
k=
1
16
h[k] x[n k] =
k=0
k=0
15
15
15
1 + ej2(5/16)(nk)
k=0
15
1
1
x[n k] = =
16
16
=
1
16
1 +
k=0
1
16
x[n k]
ej2(5/16)(nk)
k=0
15
1 1 ej2(5/16)16
1 j2(5/16)n
e
ej2(5/16)k = 1 + ej2(5/16)n
16
16 1 ej2(5/16)
k=0
1
1 1
1 +
= 1 .
N 1 ej2(5/16)
= 1 +
=
Note that the sinusoidal component of the input, with frequency 0 = 2(5/16),
is completely ltered out (i.e. attenuation). The DC component of the input
is not ltered out. In fact it appears at the output without distortion. Filtering
is one of the main things we do with LTI systems.
The lter analysis problem is to determine the ltering characteristics of a given LTI
system. We focus on this later in this Course.
The lter design problem is to select (in the FIR case) N and h[n]; n = 0, 1, , N 1
so as to realize some given ltering specication. Filter design is a topic of a senior level
Digital Signal Processing (DSP) course.
106
Example 2.25: Again consider the 16-point averager.
a) Determine the response for a general complex-valued sinusoid x[n] = A ej(0 n+) ,
where A and are constants and 0 is some value of frequency.
b) For what frequencies 0 over the range < o is the output y[n] = 0
for all n?
c) You will see from the 1-st part of this problem that for this system and
a complex-valued sinusoidal input, the output will be a scaled version of
the input. Note that in general this scaling is complex-valued, thus both
the magnitude and phase of the input signal are altered. As a function
of the input frequency 0 , determine this complex-valued scaling function.
Roughly sketch its magnitude for < 0 .
Solution:
a) This follows the Example 2.24 solution.
y[n] =
15
k=
1
=
16
h[k] x[n k] =
k=0
1
1
x[n k] = =
16
16
15
j(20 (nk)+)
Ae
j(0 n+)
= Ae
k=0
= A ej(0 n+)
15
k=0
x[n k]
1 15 j0 k
e
16 k=0
1 1 ej0 16
.
16 1 ej0
b) y[n] = 0 for 0 such that (1 ej0 16 ) = 0 as long as (1 ej0 ) = 0.

For < 0 , this occurs for 0 = 2 k; k = 7, 6, 5, , 8, except
16
for k = 0. For k = 0, i.e. for 0 = 0, the denominator (1 ej0 ) = 0. For
this case, to determine y[n] you can note from Example 2.24 that a DC is
unchanged by the system.
c) We can express the output as y[n] = A ej(0 n+) H(0 ), where
H() =
1 1 ej16
ej16/2 1 ej16/2 ej16/2
=
= ej15/2
16 1 ej
ej/2 N
ej/2 ej/2
i.e. the output is the input times H(0), where H(0) = H()|=0 . For a
given 0 , H(0 ) is a (frequency dependent) complex-valued constant. Let
H(0 ) = |H(0 )| ej H(0 ) . Then
y[n] = A |H(0)| ej(0 n++
Frequency response magnitude plot:
H(0 ))
1 sin(16/2)
16 sin(/2)
107
In Example 2.25, |H()| indicates the sinusoidal amplitude gain as a function of frequency.
H() is the sinusoidal phase shift as a function of frequency. |H()| is termed the frequency
response of the system.
Example 2.26: Again consider the 16-point averager. Determine the response for
a general real-valued sinusoid x[n] = A cos(0 n + ). Express the result as a
real-valued sinusoid.
Solution: This is basically Example 2.18, except we set N = 16 and we generalize
the input by including an arbitrary input magnitude A and phase . This time,
we will express the answer in terms of the frequency response H() identied
directly above in Example 2.25. Generalizing the result from Example 2.18 we
have that that
y[n] =
1
2
A ej(0 n+)
1 1 ej0 16
1 1 ej0 16
+ A ej(0 n+)
16 1 ej0
16 1 ej0
Following Example 2.25 above, let H() =

Note that H() =
1 1 ej16
16 1 ej
1
16
1 1 ej16
.
16 1 ej
j16
1 e
1 ej
, i.e. H() = H (). Thus,
1
A ej(0 n +) |H(0)| ej H(0 ) + A ej(0 n+) |H(0)| ej
2
= A |H(0)| cos(0 n + + H(0)) .
y[n] =
H(0 )
Example 2.27: Consider the DT LTI system y[n] .8y[n 1] = x[n]. determine
its response to input x[n] = A ej(0 n+) .
Solution:
Examples 2.8, 2.9, 2.18 & 2.24-27 all illustrate an extremely important characteristic of
any LTI system that its output to a sinusoidal input is a sinusoid of the same frequency.
The system simply alters the magnitude and phase of the input sinusoid. This, combined
with the assumption that the system is linear, suggests that the output of a LTI system will
contain only sinusoidal components that are present in the input. That is, LTI systems do
not generate new frequency components. This basic characteristic of LTI systems will be
formally established later in the Course using transforms.
108
2.4
CT Signal Impulse Expansion
In Sections 2.1 through 2.3 we developed the convolution sum representation of the I/O
calculation of a DT LTI system. To do this, in Section 2.1 we rst established how to
represent an arbitrary DT signal as a linear combination of delayed impulses. This lead,
in Section 2.2, to the derivation of the convolution sum as an I/O calculation of a DT LTI
system, and the identication of the impulse response as a useful DT LTI system description.
In this Section we also established a number of approaches to performing convolution sums.
In Section 2.3 we considered a number of DT LTI system issues related to the convolution
sum, including: system properties, system response to input sinusoids (i.e. the frequency
response), and subsystem interconnections.
We now proceed, In Section 2.4 through 2.6, with a parallel discussion for CT signals
and LTI systems. This discussion follows the convolution sum discussion very closely, since
the two topics are very similar. This CT discussion takes advantage of and reinforces the
previous DT discussion. Because of this, the following CT discussion is rather brief, though
it is as important as the DT discussion. The DT discussion was presented rst only because
the CT signal impulse expansion is a little less obvious, and there is no simple CT LTI
system example that visualizes convolution as clearly as the DT FIR lter.
We rst develop the representation of any CT signal x(t) as a linear combination of shifted
impulses. The result is similar to the impulse representation of a DT signal. However the
development is a little less obvious, since we are less familiar with the CT impulse (t).
Consider the following equation that we rst saw when we dened the impulse and then
looked at the dierentiation and integration operators:
u(t) =
(t ) d .
(27)
In this equation the step u(t) is represented as a linear combination of delayed impulses.
Delayed impulses (t ) over the delay range 0 are used, corresponding to the
range {0 t } where u(t) is nonzero. The weightings on these delayed impulses are
all one, corresponding to the values of u(t) over 0 t . How do you think this will
generalize to any CT signal x(t)?
Let
x(t) =
k=
x(k) (t k)
(28)
where
p (t ) .
(29)
2
In the limit as 0, we have that x(t) x(t), (t) (t), the summation becomes
an integral, k , and d (an incremental delay). So,

(t) =
lim
x(t) = x(t) =
x( ) (t ) d .
(30)
This is the desired result. Any signal x(t) can be represented as a linear combination of
delayed impulses, where the weighting function is just the function of signal values (i.e .the
signal itself).
2.5
109
CT LTI System I/O Calculation: Convolution Integral
Below, we will see that the topics of this Section closely parallel those in Section 2.2.
2.5.1
The Impulse Response of a CT LTI System
Figure 50 illustrates a CT LTI system. By convention, the impulse response of a CT LTI is

denoted h(t). This gure suggests that h(t) will be a system characterization.
x(t)
CT LTI system
h(t)
(t)
y(t)
h(t)
Figure 50: Representation of a CT LTI system using its impulse response h(t).
Paralleling our DT LTI systems discussion, this is the CT LTI system representation we
consider in this Chapter. For now, assume that we know h(t). In Section 2.6. we will
consider several simple circuits examples for which we will be able to identify h(t). Later in
the Course we will see how this representation can be systematically identied for any CT
LTI system.
2.5.2
Convolution Integral
Figure 51 develops the convolution integral of the input and the impulse response as the
time-domain I/O calculation of a CT LTI system.
(t)
(t )
h(t)
CT LTI system
impulse resp. h(t)
(by the TI property)
x( ) (t )
x( )
(t ) d
Figure 51: A CT LTI system and the convolution integral.

So, the I/O calculation for any CT LTI system is
y(t) =
=
x( ) h(t ) d = x(t) h(t)
(31)
h( ) x(t ) d = h(t) x(t) .
(32)
The second expression can be derived from the rst by a simple change of variables. Note the
similarities/dierences between the convolution sum and integral. The impulse response h(t)
is an I/O representation of a CT LTI system. The convolution integral is an I/O calculation.
110
2.5.3
Convolution Integral Properties and Examples
The properties of the convolution integral, and their utility, are the same as those we established earlier for the convolutions sum.
Commutativity: x(t) h(t) = h(t) x(t).
Distributivity: {a1 x1 (t) + a2 x2 (t)} h(t) = a1 {x1 (t) h(t)} + a2 {x2 (t) h(t)}
Associativity: {h1 (t) h2 (t)} h3 (t) = h1 (t) {h2 (t) h3 (t)}.
Delay: Given that y(t) = x(t) h(t), then x(t 1 ) h(t 2 ) = y(t (1 + 2 )).
Example 2.28: Graphically convolve x(t) = p1 (t) with h(t) = p3 (t).
Solution:
111
Example 2.29: First graphically, then analytically, nd y(t) = x(t) h(t) where
x(t) = eat u(t) and h(t) = ebt u(t) and a = b and Re{a, b} > 0 (i.e. the right
sided exponentials are decaying).
Solution:
The results from Examples 2.28 & 2.29, along with some other common CT convolution
integral results, are compiled in Table 4.
Table 4: Convolution Integral Results.
x
1 eat u(t)
y = x h
ebt u(t);
b=a
1
ba
eat ebt
u(t)
u(t)
u(t)
r(t)
pT (t)
pT (t)
tri2T (t) = r(t + T ) 2r(t) + r(t T )
u(t)
tm u(t); m = 0, 1, 2,
r(t)
tm u(t); m = 0, 1, 2,
tm+1
m+1
tm+2
1
m+1
u(t)
1
m+2
u(t)
112
Example 2.29 illustrates the general result that if both x(t) and h(t) are 0 for all t < 0, then
the convolution integral reduces to
t
y(t) = x(t) h(t) =
x( ) h(t ) d
u(t) .
What is the analogous rule for the DT convolution sum?

Example 2.30: Determine y(t) = x(t) h(t), where x(t) = r(t) and h(t) = u(t).
Solution: Since both x(t) and h(t) are 0 for all t < 0,
y(t) =
t
0
x( ) h(t ) d
u(t) =
t
0
u(t) =
1 2
t u(t) .
2
Note that we could have used Entry #4 or #5 of Table 4.

Next we consider a more challenging graphical convolution example.
Example 2.31: Using the graphical approach, determine y(t) = x(t) h(t) where
x(t) = e|t| and h(t) = p2 (t). You should determine closed form expressions for
y(t) for the dierent ranges of t.
Solution:
x( )
e
h(t ); t=0
1
t1
1
t+1
Figure 52: Example 2.31 CT convolution integral illustration.
y(t) =
t+1
t1 e d
0
t+1
t1 e d + 0
t+1
d
t1 e
(t+1)
(t1)
e
e
(t1)
=
2e
e(t+1)
(t1)
e
e(t+1)
t 1
1 t 1
t1
t 1
1 t 1
t1
(33)

The next seven examples illustrate the application of convolution integral properties.
Example 2.32: Consider the convolution y(t) = x(t) h(t), where h(t) = p2 (t)
and x(t) is the triangle x(t) = r(t) 2r(t 2) + r(t 4). First, use Table 4 (or
Example 2.31 results) and properties to solve this for all time t. Then use the
graphical approach, and specically:
1. Plot x( ) and h(1 ) as functions of .
2. Plot x( )h(1 ) as a function of .
3. Determine y(1). The answer is a number.
4. For what value of t does y(t) peak? Explain your answer

5. For what values of t is y(t) = 0? Explain your answer.
Solution:
113
114
Example 2.33: Convolve x(t) = eat u(t) and h(t) = u(t).
Solution: Using Entry # 1 of Table 4 directly, with b = 0, we have

y(t) = x(t) h(t) =
1
a
1 eat
u(t) .
Example 2.34: Convolve x(t) = Aea(tT ) u(t T ) and h(t) = u(t).

Solution: Using Entry # 1 of Table 4, along with the convolution integral delay
property and linearity, we have
y(t) = x(t) h(t) =
A
a
1 ea(tT )
u(t T ) .
Example 2.35: Convolve x(t) = A1 ea1 t u(t) + A2 ea2 t u(t) and h(t) = ebt u(t).
Solution: Again using Entry # 1 of Table 4, along with the convolution integral
distributivity property, we have
y(t) = x(t) h(t) =
A1
b a1
ea1 t ebt
u(t) +
A2
b a2
ea2 t ebt
u(t) .
Example 2.36: Convolve x(t) = ebt u(t 5) with h(t) = eat p10 (t 5).
Solution: Again using Entry # 1 of Table 4, along with the convolution integral
distributivity and delay properties, we have
y(t) = x(t) h(t) = eb5 eb(t5) u(t 5)
=
eb5
ba
ea(t5) eb(t5)
u(t 5)
eat u(t) ea10 ea(t10) u(t 10)
eb5 ea10
ba
ea(t15) eb(t15)
u(t 15) .
115
Example 2.37: Convolve x(t) = u(t+1)u(t1) with h(t) = 3(t2)+ebt u(t).

(Hint: note that for any signal v(t), (t T ) v(t) = v(t T ).)
Solution:
y(t) = u(t + 1) [3 (t 2)] u(t 1) [3 (t 2)]
+ u(t + 1) ebt u(t) u(t 1) ebt u(t)
1
= 3 u(t 1) + 3 u(t 3) +
1 eb(t+1))
b
u(t + 1)
1
b
1 eb(t1)
Example 2.38: The convolution y(t) = h(t) x(t) where h(t) = t2 [u(t) u(t 2)]
and x(t) =
k= (t 2k) is periodic with period T = 2. Compute the power
of y(t).
Solution:
y(t)
....
2
....
0
Figure 53: Convolution integral result for Example 2.38.
u(t 1) .
116
This next example illustrates the use of the analytical approach to convolution to determine
the response of a CT LTI system to a complex sinusoidal input.
Example 2.39: For a CT LTI system with impulse response h(t) = C eat u(t),
determine to response y(t) to input x(t) = A ej(t+) . Assume Re{a} > 0.
Solution:
In Example 2.39, note that the output is a sinusoid of the same frequency as the input. The
system alters the magnitude and phase of the input as dictated by the function
H() =
h(t) ejt dt .
(34)
This is a function of frequency, indicating that the eect of the system on the input sinusoid
depends on the frequency of that sinusoid. Also note that this eect can be derived from
the LTI system impulse response.
We have seen this kind of result before that for a sinusoidal input, the output of a LTI
system is consists solely of a sinusoid of the same frequency. For DT LTI systems, this was
the point of Subsection 2.3.3. We also saw this in Chapter 1 with Examples 1.39 & 1.40.
2.6
117
CT LTI System Issues
Since the tools for working with CT and DT LTI systems are so similar, for the most
part this Subsection looks like an abbreviated version of Subsection 2.2.4 (DT LTI system
issues). Paralleling Subsection 2.2.4, we briey cover system properties and interconnections.
However, in two respects the discussions diverge. First, for CT LTI systems, there are no
common nite impulse response systems, so we dont emphasize this. Second, with several
examples we do emphasize CT LTI circuits.
System Properties:
Since we are considering CT LTI systems, linearity and time-invariance are covered. A
CT LTI system is causal if and only if
h(t) = 0 ;
t<0 .
(35)
|h(t)| dt < .
(36)
A CT LTI system is stable, if and only if
System Interconnections:
If M CT LTI systems, with impulse responses hi (t); i = 1, 2, , M are connected in
parallel, than the overall system impulse response is
M
hi (t) .
h(t) =
(37)
i=1
If M CT LTI systems, with impulse responses hi (t); i = 1, 2, , M are connected in cascade,

than the overall system impulse response is
h(t) = hi (t) h2 (t) hM (t) .
(38)
For feedback systems, there is no direct expression for the overall impulse response in terms
of the impulse responses of the subsystems.
Example 2.40: A CT LTI system has impulse response h(t) = (t+1) e100t u(t+5).
Is this system: Linear? Time-invariant? Causal? Stable?
Solution:
118
Example 2.41: Determine h(t), the overall system impulse response, for the system shown below.
x(t)
h 2(t) = (t)
Delay T 1:
h 1 (t) = (t T 1 )
y(t)
Multiply a:
h 3 (t) = a (t)
Integrator:
h 4 (t) = u(t)
Delay T 2:
h 5 (t) = (t T 2 )
Figure 54: CTLTI subsystem interconnection for Example 2.41.

Solution:
Example 2.42: Determine the impulse response h(t) from the step response.
Solution:
The DT analogy is h[n] = s[n] s[n 1] (i.e. [n] = u[n] u[n 1]).
119
Example 2.43: Consider the following series RLC circuit, where R = 280,
L = 0.1H and C = 0.4F . The step response, which can be identied using
techniques from your circuits course, is
s(t) = u(t) e1400t cos(4800t)u(t)
7 1400t
e
sin(4800t)u(t) .
24
Find the impulse response h(t).

R
L
+
y(t)
x(t)
Figure 55: Serial RLC circuit for Example 2.43.

Solution:
h(t) =
d
s(t) = (t) (t) + 1400 e1400t cos(4800t) u(t) + 4800 e1400t sin(4800t) u(t)
dt
+ 408.4 e1400t sin(4800t) u(t) 1400 e1400t cos(4800t) u(t)
= 5204.4 e1400t sin(4800t) u(t) .
Example 2.44: For the following series RC circuit, which has step response
s(t) =
1 et/RC
u(t) .
R
+
+
x(t)
y(t)
Figure 56: Serial RC circuit for Example 2.44.

(a) Find the impulse response h(t).
(b) Determine the output for input xb (t) = A eat u(t).
(c) Determine the output for input xc (t) = 10 cos(200t).
(d) Determine the output for input
xd (t) =
k=
p (t kT ) ;
<T .
120
Solution:
2.7
Generalizing the Utility of Convolution
In this Chapter of the Course we introduced convolution as the I/O calculation for a LTI
system. Note that this is not the only application of convolution we have considered recall
that the impulse response of two LTI systems in cascade is the convolution of the subsystem
impulses responses. In engineering, the convolution operation crops up in a wide variety of
circumstances. To a large extent this is due to the nature of linear interactions between one
dimensional or multidimension signals and systems.
Here, within the context of examples of the convolution operator, we suggest several new
applications of convolution.
Example 2.45 - Correlation:
See Problem 2.101 in the back of this Chapter.
Example 2.46 - Spectral Smearing:
See Problem 2.102 in the back of this Chapter.
Example 2.47 - Image Smearing:
See Parcticums 1 & 2.
2.8
121
Practicum 2
Convolution
Reporting Requirements: See the Reporting Requirements at the beginning of Practicum
1.
Procedures
Before the lab session, Procedure 1(a) should be completed and the example in Appendix 1
should be studied. Also, identify the impulse response for Procedure 2(d).
1. Convolutions sum for nite duration DT signals:
(a) By hand, compute and plot the convolution of the signals x1 [n] = p20 [n] and
h1 [n] = p10 [n].
(b) Now use Matlab to compute and plot the convolution of these two signals. Specifically, do this twice, using the two methods illustrated in the attached Appendix
1: 1) by using the Matlab conv command; and 2) by programming the convolution
operation yourself. For both methods, plot the result for n = 0, 1, 2, , 27, 28.
(c) Repeat (b) using your FIR mine function. You can just directly input your arrays
from the 2-nd part of (b) (i.e. from your convolution program).
(d) Let q be the number of contiguous nonzero elements of h[n] (i.e. the length of
h[n]). Let p be the length of x[n]. Generalizing the results observed above, what
is the length of y[n] = x[n] h[n]?
2. Dierential Code Decoding:
As for Practicum 1, for this part of the practicum you will process a black and
white image. There are three les you will need, lighthouse.mat, lighthouse2.mat and
show img.m, which can be found in directory
v:/Electrical Computer/ece3240
(a) Copy the three les into your working directory or otherwise assure that you have
access to them within Matlab.
(b) Load lighthouse2.mat into Matlab. In Matlab, what is the image name? What
size is it? Display the image using the command show img( . ). Do the same for
lighthouse.mat.
(c) To reduce the numerical range of the image amplitudes prior to image compression, the image xx[i, n] has been dierentially encoded along the horizontal to
create zz[i, n] using the following equation for each row
zz[i, n] = xx[i, n] 0.9 xx[i, n1] ;
i = 1, 2, , 326 ; n = 1, 2, , 426 , (1)
with xx[i, 0] = 0.0, i = 1, 2, , 326. Note that each row is processed with a
system the looks like the channel for Procedure 2 of Practicum 1. How did you
invert the eect of that channel?
122
The following lter6 is to be used on each row of zz to decode this image:

yy[i, n] =
N 1
k=0
.9k zz[i, n k] ,
(2)
where N = 25. Describe the impulse response h[n] corresponding to this lter.
(d) Use the Matlab conv command to implement this decoding lter. Display the
decoded image yy. Compare its quality to that of the encoded zz image and the
original image xx.
(e) Try several other values of N and compare results with those obtained in 2(d).
(f) The decoder used in Procedures 2(c-e) is computationally expensive, and it does
not prefectly decode the dierential code. Can you think of a cheaper, perhaps
better decoder? Implement it if you have time, but dont worry about submitting
the results. (Hint: consider your experience with channel equalization and image
deblurring from Practicum 1.)
Appendix 1
echo on
%
Practicum 2, Appendix 1, The convolution sum
%
%
x(n) = u(n-5) - u(n-11)
%
h(n) = (.9)^(n-5) [u(n-5) - u(n-21)]
%
%
To illustrate one way of handling delayed signals below we load
%
zeros into the data arrays so that the first element of each
%
array corresponds to time n=0. However, using the convolution
%
delay property (to keep track of what time the array elements
%
correspond to), there is really no need to prepend or append
%
zeros to the input arrays to the conv function.
clear
clear functions
clf
pause
x=ones(1,6);
% Generate 6 samples of x[n] for n=5,..,10
k=1:16;
% Generate 16 samples of h[n] for n=5,..,20
h(k)=(.9).^(k-1);
pause
y=conv(x,h);
% Using the MATLAB conv command,
pause
% convolve h and x to form the output y
%
y1(1:40)=zeros(1,40);
% Plot 40 samples of y[n] for n=0,1,...,39
6
The term lter is often used to refer to a LTI system.
123
y1(11:31)=y;
n=0:1:39;
subplot(211)
plot(n,y1,*),xlabel(n),ylabel(y(n)),
title(1-st Convolution Result),text(25,4,John Butler)
pause
%
Now lets do it again without using the MATLAB conv command.
%
Well do it ourselves.
clear
x=zeros(1,11);
% Generate 11 samples of x[n] for n=0,..,10
x(6:11)=ones(1,6);
%
h=zeros(1,21);
% Generate 21 samples of h[n] for n=0,..,20
k=0:1:15;
%
h(6:21)=(.9).^k;
%
pause
%
x(12:40)=zeros(1,29);
% Directly convolve h and x --> output y
h(22:40)=zeros(1,19);
for n=1:40;
y(n)=0.;
for k=1:n;
y(n)=y(n)+x(k)*h(n+1-k);
end
end
pause
%
clear n
n=0:1:39;
subplot(212)
plot(n,y,*),xlabel(n),ylabel(y(n))
title(2-nd Convolution Result),text(25,4,Garret Dutton)
124
.
125
Practicum 2
Instructor/TA Sign O Sheet, & Report Form
Students Name:
1. Pract. 2, Procedure 1(b,c): output plots
2. Pract. 2, answer for Procedure 1(d):
(a function of p and q)
3. Pract. 2, Procedure 2(d,e): displays of original, coded and decoded images

4. Pract. 2, for Procedure 2(c), discuss the similarities and dierences in characteristics
and performance between: 1) the coder and decoder for this Procedure; and 2) the
channel and 1-st equalizer used in Practicum 1.
5. Pract. 2, for Procedure 2(f), describe the decoder which is analogous to the 2-nd
equalizer in Practicum 1.
126
.
2.9
127
Problems
Chapter Topics:
2.1-10 (graphical DT convolution sum with basic nite duration signals)
2.11-24 (graphical DT convolution sum with basic innite duration signals)
2.25-43 (graphical DT convolution sum with more challenging signals)
2.44-50 (analytical DT convolution sum, with sinusoids)
2.51-58 (DT convolution sum properties)
2.59-74 (DT convolution sum and DT LTI system issues)
2.75-87 (graphical CT convolution integral)
2.88-92 (CT convolution integral properties)
2.93-100 (CT convolution integral and CT LTI system issues)
2.101-102 (other convolution applications)
1. Using the graphical approach, nd y[n] = x[n] y[n] where h[n] = p4 [n] and x[n] =
(1)n u[n].
a) Sketch x[n] and h[n].
b) Sketch x[k] and h[n k]; n = 3.
c) Working o of part b) results, sketch x[k] h[3 k].
d) Working o of part c) results, determine y[3].

e) Plot y[n] for 2 n 10.
2. Consider a DT LTI system with impulse response
h[n] = [n] + a2 [n 2] + a4 [n 4] + a6 [n 6] .
Determine the output y[n] for input x[n] = [n] a2 [n 2]. First, use the linearity
and time-invariance properties of the system directly. Then convolve h[n] with x[n].
3. Consider a DT LTI system with inpulse response h[n] = [n] [n 1] + [n
2] [n 3] and input x[n] = r[n] p6 [n]. Using the graphical approach, nd
y[n] = x[n] h[n]. Specically
(a) sketch x[k] and h[n k]; n = 4;
(b) sketch x[k] h[4 k];

(c) determine y[4]; and
(d) plot y[n] for 2 n 10.

4. Consider two sequences h1 [n] = [n] 0.9 [n 1] and h2 [n] = 0.9n p5 [n].
(a) Plot these and use the graphical approach to determine the convolution
h[n] = h1 [n] h2 [n].
(b) With the h[n] you derived in (a), and x[n] = p2 [n], use the graphical approach to
determine y[n] = x[n] h[n].
128
5. Using the graphical approach, nd y[n] = x[n] h[n] where h[n] = .5n p4 [n] and
x[n] = [n] + .5 [n 1] .5 [n 2]. Specically,
(a) Sketch h[k] and x[n k]; n = 1.
(b) Sketch h[k] x[1 k].

(c) Determine y[1].
(d) Plot y[n] for 2 n 6.

6. A signal x[n] = [n] + 0.2 [n 1] 0.8 [n 2] is the input to a DT LTI system with
impulse response h[n] = 0.8n p25 [n]. Determine y[n] = x[n] h[n]. Specically:
(a) Sketch h[k] and x[0 k] as functions of k.
(b) Sketch h[k] x[0 k] as a function of k.
(c) Determine y[0]. The answer is a number.
(d) Sketch y[n] over the range of n that it is nonzero.

(e) List the times n and corresponding values y[n] of the nonzero outputs.
7. Using the graphical approach, nd y[n] = x[n] h[n] where h[n] = 2n5 p5 [n] and
x[n] = [n] + 2 [n 1]. Specically,
a) Sketch h[k].
b) Sketch x[n k] for n = 0.
c) Sketch h[k] x[n k] for n = 1.
d) Determine and plot y[n] for 2 n 8. What is y[n] outside this range of n?
8. Convolving pulses:
(a) Convolve x[n] = p3 [n 1] with h[n] = p5 [n + 3]. Plot y[n].
(b) Convolve x[n] = p5 [n] with h[n] = p10 [n 3]. Plot y[n].
9. Two DT LTI systems are connected in cascade. The 1-st has impulse response
h1 [n] = 2n p5 [n] and the 2-nd h2 [n] = [n] + 2[n 1] + [n 2]. Assume the
resulting system has impulse response h[n] = h1 [n] h2 [n]. Using the graphical
convolution approach, determine h[n].
Specically:
(a) Sketch h1 [k] and h2 [2 k] as functions of k.
(b) Sketch h1 [k] h2 [2 k] as a function of k.
(c) Determine h[2]. The answer is a number.
(d) Sketch h[n] over the range of n that it is nonzero. Label time and h[n] values.
10. Using the graphical approach, nd the output y[n] = x[n] h[n] of a DT LTI system
with impulse response h[n] = p10 [n] and x[n] = n p5 [n]. Minimally, you should:
129
(a) Sketch x[k] and h[n k]; n = 0 as a function of k.
(b) Sketch y[n] over its entire non-zero range, making sure to show details.
11. Using the graphical approach, nd y[n] = x[n] h[n] where h[n]
x[n] = (n 1 ) u[n 1]. Specically:
2
2 u[n] and
a) Sketch x[n] and h[n].

b) Sketch x[k] and h[n k]; n = 0.
c) Working o of part b) results, list each region of n that has a dierent y[n]
expression.
d) For each region of n from part c), determine an expression for y[n].
e) Describe y[n] as a single closed-form expression.
12. Consider y[n] = x[n] h[n], where h[n] = 0.5n p5 [n] and x[n] = u[n]. Using the
graphical convolution approach, determine y[n] (i.e. the step response). Specically:
(a) Plot h[k] and x[4 k] as functions of k.
(b) Plot h[k]x[4 k] as a function of k.
(d) Determine y[2010].

(e) Determine expressions for y[n] for all ranges of n.
13. Using the graphical approach, nd y[n] = x[n] h[n] where h[n] = 0.5n u[n] and
x[n] = 0.5n p20 [n + 10]. Specically,
(a) Sketch x[k] and h[n k]; n = 0.
(b) Determine the dierent ranges of n that have dierent expressions for y[n].
(c) For each of your ranges of n determine a closed-form expression for y[n].
14. Consider a simple IIR DT-LTI system with impulse response h[n] = 0.5n u[n] and
input x[n] = p3 [n].
(a) Plot, as a function of memory index k, h[k] and x[n k]; n = 0.
(b) Plot, as a function of k, h[k] x[n k]; n = 2.

(c) Using your result from (b), determine y[2].
(d) Determine y[0] and y[1]. Draw additional graphs only if you need to.
(e) For n 2, the output has the form: y[n] = K 0.5n . Determine the constant K.
Show all work.
15. Determine y2 [n] = x2 [n] h2 [n] where x2 [n] = (1)n u[n] and h2 [n] = p4 [n].
16. Consider y[n] = x[n] h[n], where h[n] = 0.5n u[n] and x[n] = p3 [n]. Using the
graphical convolution approach, determine y[n]. Specically:
130
(d) Determine an expression for y[n] for all n.

17. Consider y[n] = x[n] h[n], where h[n] = 0.99n u[n] and x[n] = u[n 5]. Specically:
(a) Plot h[k] and x[n k], as functions of k, for n = 0.
(b) Plot h[k]x[n k], as a function of k, for n = 0.

(c) Determine an expression for y[n] for all n.
(d) Determine y[100] and y[]. The answers are numbers.

18. Using what ever approach you wish, convolve x[n] = .5n u[n] with h[n] = .5n u[n].
Describe the result for all n. Show all work.
19. Consider a DT LTI system with impulse response h[n] = n (.5)n u[n] and input
x[n] = u[n 10]. The output is y[n] = h[n] x[n]. Using the graphical approach
to convolution, determine:
(a) y[0].
(b) y[10].
(c) y[12].
(d) A closed-form (i.e. do summation) expression for y[n] that is good for all n.
(e) y[].
20. Graphically solve y[n] = h[n] x[n] for x[n] = 6n p20 [n] and h[n] = 0.5 (0.8)n u[n].
Specically,
a) Plot x[k] (vs. k), labeling the axes and important values of k and x[k].
b) Plot h[n k] (vs. k), labeling the axes and important values of k and h[n k].
c) Identify the dierent regions of n that result in dierent convolution sum expressions.
d) For each region identied in c), determine the summation expression for y[n].
(You must clearly show the ranges of k summed over and the expressions inside
the summations. Leave the answers in summation form.)
e) For the highest positive region of n identied in procedure c), the output is of the
form y[n] = K (0.8)n . Determine K.
21. Graphically solve y[n] = h[n] x[n] for x[n] = 10n u[n] and h[n] = 0.4n u[n]. Specically,
a) Plot h[k] (vs. k), labeling the axes and important values of k and h[k].
b) Plot x[n k] (vs. k), labeling the axes and important values of k and x[n k].
131
c) Identify the dierent regions of n that result in dierent convolution sum expressions.
d) For each region identied in c), determine the summation expression for y[n].
22. Consider the impulse response h[n] = u[n] of a DT LTI system with input
x[n] = p3 [n] p3 [n 3]. Graphically compute y[n] = h[n] x[n]. Specically,
a) Plot x[k] and a function of k.
b) Plot h[n k]; n = 0 as a function of k.
c) Using a) and b), determine and plot y[n] for 4 n 10.
d) Now let x1 [n] = x[n 6m]. Graphically determine and plot y1 [n] =
m=0
h[n] x1 [n] for 4 n 14.
23. Using the graphical approach, determine y[n] = h[n] x[n], for all n, where h[n] =
0.9n u[n 6] and x[n] = u[n] + 2 u[n 10]. Specically,
a) Sketch h[k] and h[n k]; n = 0 as a function of k.
b) Using your sketches, determine all regions of n that have dierent expressions for
y[n].
c) For each of your regions of n, determine the expression for y[n]. Leave expressions
in summation form. That is, dont solve the summations. However, make sure
the sum limits and expressions in the summation are correct.
d) Determine y[40].
24. Graphically convolve h1 [n] = n (.8)n u[n] with h2 [n] = (.8)n p20 [n 3] to form h[n].
Specically,
(a) Plot h1 [k] vs. k and h2 [n k]; n = 0 vs. k.
(b) Identify the dierent regions of n that have dierent expressions for h[n].
(c) For each region of n determine the expression for h[n] (i.e. in summation form).
(d) For the region including n = 15, determine the closed form expression for h[n]
(i.e. do the summation). Use this to determine h[15].
25. Determine y5 [n] = tri10 [n] u[n].
26. Plot x[n] = p5 [n + 2] and h[n] = r[n] p5 [n]. Determine the convolution
y[n] = x[n] h[n] and plot the resulting y[n] over the range of n where y[n] = 0.
27. Graphically solve y[n] = h[n] x[n] for x[n] = n2 p10 [n] and h[n] = p20 [n]. Specically,
(a) Plot x[k] (vs. k), labeling the axes and important values of k and h[k].
(b) Plot x[n k] (vs. k), labeling the axes and important values of k and x[n k].
132
(c) Identify the dierent regions of n that result in dierent convolution sum expressions.
(d) For each region identied in c), determine the summation expression for y[n].
(You must clearly show the ranges of k summed over and the expression inside
the summation. Leave the answers in summation form.)
(e) What are y[7] and y[35]?
28. Your are given input x[n] = n p21 [n] to a DT LTI system with impulse response
h[n] = p10 [n 3]. Derive a closed form expression for the output y[n] = x[n] h[n] for
each range n. Using your expressions, specically determine y[6] and y[20].
29. Consider a DT-LTI system with impulse response h[n] = u[n 5] and input
x[n] = 3 (.5)n u[n] + 2 u[n 4] 7 r[n 10]. Determine output y[n] as follow.
(a) Determine ya [n] = u[n] (.5)n u[n].
(b) Determine yb [n] = u[n] u[n].

(c) Determine yc [n] = u[n] r[n].
(d) Use your results from parts (a-c) to determine an expression for y[n].
(e) Use your result from part (d) to determine y[6].
30. Combined graphical/property convolution problem:
(a) First determine h3 [n] = h1 [n] h2 [n] where h1 [n] = u[n] and h2 [n] = r[n].
(b) Using convolution properties and your result from part (a) above, determine an
expression for h[n] = h1 [n] h3 [n], where
h3 [n] = tri10 [n] = r[n + 1] 2r[n 4] + r[n 8] .
(3)
(c) Using your expression from (b), determine h3 [5].
31. Let x[n] = ( 0.5)n u[n]. Determine the energy of w[n] = x[n] x[n]. Feel free to
use the summation and convolution tables provided in the Course Notes.
32. Consider y1 [n] = x1 [n] h1 [n], where h1 [n] = 0.5n u[n 3] and x1 [n] = .5n p20 [n].
Specically:
(a) Plot x1 [k] and h1 [3 k] as functions of k.
(b) Plot h1 [k] x1 [3 k] as a function of k.
(c) Determine y1 [3]. The answer is a number.
(d) Determine an expression for y1 [n] for all n.

33. Convolve x3 [n] = n u[n] with h3 [n] = n2 u[n]. Use either the graphical or analytical
approach. Determine a closed form expression for the resulting y3 [n] for all n.
34. Using the graphical approach, convolve x[n] = 2n u[n] with h[n] = .9n u[n]. Describe
the result for all n. Specically, what is y[5]?
133
35. Using the graphical approach, determine y[n] = h[n] x[n], for all n, where
h[n] = 0.9n u[n] and x[n] = n 0.9n u[n]. Specically,
(a) plot x[k], h[k] and h[4 k],
(b) plot x[k] h[4 k],

(c) determine y[4],
(d) determine y[n] for all n.

36. Graphically solve y[n] = h[n] x[n] for x[n] = n p11 [n] and h[n] = r[n]. Specically,
a) Identify the regions of n that result in dierent convolution sum expressions.
b) For each region identied in a), determine the summation expression for y[n].
c) For n = 20, use the summation table to evaluate the required summation from b),
and determine y[20].
37. Determine y[n] = x[n] h[n] for n 0, where x[n] = r[n] (a ramp) and h[n] = p6 [n].
Plot y[n] for 0 n 10. For n 6 y[n] can be written in the form y[n] = c1 n + c2 .
Determine the constants c1 and c2 .
38. Using the graphical approach, determine y[n] = h[n] x[n], for all n, where h[n] =
0.5n u[n] and x[n] = n p10 [n]. Specically,
(a) Determine all the regions of n which have dierent expressions for y[n].
(b) For each region of n, determine the expression for y[n].
(c) Determine y[0] and y[].
39. Using the graphical approach, determine the convolution y[n] = x[n] h[n], where
h[n] = n p10 [n] and x[n] = n u[n]. Specically:
(a) Describe y[n] for n , evaluating all summations.
(b) Determine y[17].
40. Consider a DT LTI system with impulse response h[n] = (an + bn )u[n] and input
x[n] = (an + bn )u[n].
(a) Assume a = b. Using whichever approach to convolution you wish, determine the
expression for the output y[n] for all n.
(b) Repeat a) for a = b.
41. Consider y[n] = x[n] h[n], where h[n] = p10 [n 1] and x[n] = r[n]. Using the
graphical convolution approach, determine y[n]. Specically:
134
42. Using the graphical approach, determine y[n] = x[n] h[n] where
x[n] = 0.1|n|
and
h[n] = 0.4n u[n] .
43. Given that x[n] = h[n] =
0.9 u[n], determine :
(a) y[n] = x[n] h[n],
(b) the energy of y[n].
44. Consider input x12 [n] = A ej(0 n+) u[n] to a DT LTI system with impulse response
h12 [n] = p6 [n]. Determine closed form expressions for the output y12 [n] for all times
n. (Note that x12 [n] turns on at n = 0.)
45. Consider a general N point averager, i.e. with impulse response h[n] =
1
N
pN [n].
(a) Determine its output y[n] due to complex sinusoidal input x[n] = ejo n . Specically, what is the output when o = 0? o = , assuming N is even?
(b) Let N = 16 and
x[n] = 5 + 3 ej(/4)n
Determine the output y[n].

46. Using the analytical approach to convolution, determine y[n] = h[n]x[n] where h[n] =
bn u[n] and x[n] = (a ej )n . Perform any summations. Evaluate y[n] specically for
a = 1, b = .5 and = .
47. Consider a DT LTI system with impulse response h[n] = (0.8)n u[n].
(a) Determine its output y[n] due to complex sinusoidal input x[n] = ejo n . Specically, what is the output when o = 0? o = ?
(b) Determine its output y[n] due to input x[n] = ejo n u[n]. Specically, what is the
output when o = 0? o = ?
48. Let x11 [n] = ej0 n be the input to a DT LTI system with impulse response h11 [n] =
[n] [n 2]. Determine an expression for the output y11 [n]. Over the range
0 , for what values of 0 is y11 [n] = 0?
49. Given a DT LTI system with impulse response h[n] = n (.9)n u[n] and input
x[n] = ej0 n ,
135
(a) Analytically determine y[n] = h[n] x[n].
(b) For 0 = 0, determine y[n]. Is this input amplied or attenuated?

(c) For 0 = , determine y[n]. Is this input amplied or attenuated? What is the
phase shift from input to output?
50. Given h[n] = (1)n p12 [n] and x[n] = A ej(n+) ,
(a) Analytically, determine y[n] = h[n] x[n]. Since x[n] is a complex sinusoid, y[n]
is of the form y[n] = x[n] H(). What is H()?
(b) For = 0, determine y[n].
(c) For = , determine y[n]. You should be a little more careful with this one.
51. Using the convolution table in the Course Notes, determine x[n] h[n] for:
(a) h[n] = p10 [n] and x[n] =
k=
(b) h[n] = p20 [n] and x[n] = r[n 3].
p10 [n 20k].
(c) h[n] = r[n] p10 [n] and x[n] = r[n].
(d) h[n] = .9n u[n] and x[n] = .9n p10 [n 10].

52. Determine x[n] = [n 2] [n + 3] u[n + 1] u[n 4].
53. Consider the input x[n] = 3n u[n 1] +
response h[n] =
1
4
1
3
u[n] a DT LTI system with impulse
u[n + 3].
(a) Determine y[n] = x[n] h[n] without using the distributivity property of convolution.
(b) Determine y[n] = x[n] h[n] using the distributivity property of convolution.
54. Perform the convolution y[n] = x[n] h[n], where
x[n] = p10 [n] ;
h[n] = [n 3] + u[n] .5n u[n] .
Specically,
a) State the approaches you will use (e.g. graphical, analytical, tables, properties).
b) Derive an expression for y[n]. Do all convolutions, but dont worry too much
about making the answer compact.
c) Evaluate your expression to obtain y[3].
55. In this problem you are to perform the convolution y[n] = x[n] h[n], where
x[n] = p10 [n] and h[n] is plotted below. You are to use convolution properties as
detailed below.
Note that h[n] = r[n] r[n 10] r[n 15] + r[n 25].
a) First, determine the convolution y1 [n] = r[n] u[n].
136
h[n]
10
......
.
..
..
.
..
.......
10
..
..
...
. ...
24
15
b) Now, using the delay and distributivity properties, determine an expression for
y[n] in terms of y1 [n]. Using this expression, determine y[15].
56. In this problem you are to perform the convolution y[n] = x[n] h[n], where x[n] =
r[n 3] and h[n] is plotted below. You are to use convolution properties as detailed
below.
h[n]
10
.
......
0
......
.
..
..
..
..
..
...
10
19
Note that h[n] = r[n] 2 r[n 10] + r[n 20].

a) First, graphically perform the convolution y1 [n] = r[n] r[n]. Provide a closed
form expression good for all n. Using this expression, determine y1 [10].
b) Now, using the delay and distributivity properties, determine an expression for
y[n] in terms of y1 [n]. Using this expression, determine y[13].
57. Consider the convolution y[n] = x[n] h[n], where h[n] = 0.9n u[n] and
x[n] = [n 3i]. Specically:
i=0
(a) Plot x[k] and h[6 k] as functions of k.
(b) Plot x[k]h[6 k] as a function of k.
(d) For what values of n is y[n] = 0? Explain your answer.

58. You are given a DT LTI system with impulse response h[n] =
10
k=1
k p4 [n 8k].
(a) First determine s1 [n] = u[n] p4 [n] any way you wish.
(b) Now use convolution properties to determine the step response s[n] of this DT
LTI system in terms of s1 [n].
59. For each of the following LTI systems, nd the impulse response. Is the system casual?
Stable? Explain your answers in terms of your impulse response characteristics.
137
(a) LTI system (a).

y[n] =
k=0
ej(/3)k x[n k] .
(b) LTI system (b).

y[n] = x[n + 5] + 2 y[n 1] .
60. Consider a DT LTI system with impulse response h[n] = 0.7n p30 [n]. Use system
linearity to determine the output for input x[n] = [n 1] + u[n 7] + 0.7n u[n].
Determine the general expression, and y[5].
61. Consider a LTI DT system with impulse response h1 [n] = 5 p11 [n].
(a) Is this system linear? Why?
(b) Is this system time invariant? Why?
(c) Is this system causal? Why?
(d) Is this system stable? Why?
(e) Determine the output y1 [n] due to input x1 [n] = 1.
(f) Determine the output y2 [n] due to input x2 [n] = ejn .
(g) Now nd the output y3 [n] due to input x2 [n] and the LTI DT system with impulse
response h2 [n] = 5 [n] h1 [n].
1
62. Consider a DT LTI system with impulse response h[n] = 16 ej0 n p16 [n] and input.
x[n] = 3 ej1 n . The output is a constant times the input. What is the expression for
this constant?
63. Consider two parallel DT LTI subsystems, one with impulse response h1 [n] = 0.7n u[n]
and the other with h2 [n] = 0.49 [n 2] 0.7n u[n]. Determine the overall system
impulse response h[n].
64. Consider a DT LTI system that consists of two subsystems in cascade, h1 [n] and h2 [n],
which together are in parallel with subsystem h3 [n]. The hi [n]; i = 1, 2, 3 are subsystem
impulse responses. Given that
h1 [n] = [n] .9[n 1]
h2 [n] = [n] + .9[n 1] + .81[n 2] + .729[n 3]
h3 [n] = .6561[n 4],
determine the overall system impulse response h[n].
65. The DT LTI system consists of two subsystems connected in parallel. One has impulse
response h1 [n] = (.5)n u[n] and the other h2 [n] = .5n u[n].
(a) Determine h[n], the overall impulse response of the system.
138
(b) For input x[n] = p2 [n], plot output y[n] for 2 n 5.
66. Consider the cascade of two DT LTI subsystems, where the 1-st subsystem has impulse
response h1 [n] = sin(8n) and the 2-nd has impulse response h2 [n] = an u[n] where
|a| < 1. Consider input x[n] = [n] a[n 1] to the 1-st subsystem. Determine
y[n], the output of the 2-nd subsystem. Use of the associativity and commutativity
properties of the convolution sum can greatly simplify the solution to this problem.
67. Two LTI DT subsystems are connected in cascade. The rst system, with input x[n]
and output v[n], is described as
v[n] = x[n] 0.5 x[n 1] .
The second system, with input v[n] and output y[n], is described as
y[n] = v[n] + .5 v[n 1] + .25 v[n 2] + .125 v[n 3] .
(a) Determine & plot the impulse responses, h1 [n] and h2 [n], of the two subsystems.
(b) Determine & plot the overall system impulse response h[n].
(c) Determine & plot the system output y[n] for input x[n] = p3 [n].
68. DT LTI systems:
(a) Let h1 [n] = 100 an u[n m] be the impulse response of a DT LTI system. For
what ranges of real-valued a and integer m is the system both causal and stable?
(b) Determine and sketch the overall impulse response h[n] of the following interconnection of DT LTI subsystems? (The subsystem impulses responses are shown.)
(c) Given h[n] = 12n p15 [n], is the corresponding LTI system causal? Stable?
(d) Determine the output y[n] = 3 y[n 1] + x[n], where y[1] = 0 and x[n] = [n].
(e) Find h[n] = h1 [n] ([n] h2 [n]), with h1 [n] = 2n u[n] and h2 [n] = 16 h1 [n 4].
x[n]
p [n]
30
[n1]
p [n]
y[n]
10
p [n10]
10
69. This is the Practicum 1 image deblurring problem. Consider a DT LTI system with
impulse response hc [n] = 0.8n p11 [n].
(a) Consider cascading hc [n] with a DT LTI system with impulse response he1 [n] =
[n] 0.8 [n1]. Determine the overall system impulse response ht1 [n]. Simplify
your result as much as possible.
139
(b) Alternatively, consider cascading hc [n] with a DT LTI system with impulse response
he2 [n] = he1 [n] + 0.811 he1 [n 11]. Determine the overall system impulse
response ht2 [n]. Simplify your result as much as possible.
(c) Given your answers from part (a) and (b), deduce what the impulse response is
for the he [n] which perfectly inverts hc [n].
70. Consider the following DT LTI system. System 1 is a 10 pt. averager. For input x1 [n]
and output y1 [n], System 2 is described by I/O dierence equation
y1 [n] = x1 [n 5] 0.9 y1 [n 1] .
Determine the overall system impulse response. Is the overall system causal? Explain
why.
System #1
System #2
System #2
x[n]
System #1
y[n]
71. Two DT LTI subsystems are connected in cascade. The rst system, with input x[n]
and output v[n], is described as
v[n] = x[n] 0.5 x[n 1] .
The second system, with input v[n] and output y[n], is described as
y[n] = v[n] + .5 v[n 1] + .25 v[n 2] + .125 v[n 3] .
(a) Determine & plot the impulse responses, h1 [n] and h2 [n], of the two subsystems.
(b) Determine & plot the overall system impulse response h[n].
(c) Determine & plot the system output y[n] for input x[n] = p3 [n].
(d) Is the overall system stable? Why?
(e) Is the overall causal stable? Why?
72. You are given a DT LTI channel with impulse response hc [n] = 0.9[n] + [n 1]
and a DT LTI equalizer with impulse response
he [n] = (0.9)4 [n] (0.9)3 [n 1] + (0.9)2 [n 2] (0.9) [n 3] + [n 4] .
(a) Determine the impulse response h[n] from channel input to equalizer output.
(b) For your h[n] from (a), what is the corresponding I/O dierence equation? (Let
x[n] and y[n] denote, respectively, the channel input and the equalizer output).
140
(c) Say you do not mind a delay between the channel input and equalizer output.
Considering your experience with equalizers from Practicum 1 and your subsequent experience with designing similar systems in Practicum 2, describe a more
eective equalizer and give the resulting channel-input to equalizer-output I/O
dierence equation.
73. Consider two DT LTI systems connected in cascade. These have impulses responses
h1 [n] = [n] 2[n 1] and h2 [n] = 2n u[n].
(a) Is the h1 [n] system causal? Stable? Why?
(b) Is the h2 [n] system causal? Stable? Why?
(c) Is the overall system, with impulse response h[n], causal? Stable? Why?
(d) Is the DT LTI system with impulse response h3 [n] = ej0 n u[n] causal? Stable?
Why?
74. Any convolution required in this problem should be done graphically. Consider two DT
LTI systems connected in cascade. The rst has impulse response h1 [n] = n (.8)n u[n].
The second has impulse response h2 [n] = (.8)n u[n].
(a) Determine the overall system impulse response h[n].
(b) Specically, using your expression from (a), determine h[5].
(c) Is the overall system causal? Why?
(d) Is the overall system causal? Prove your answer?
75. Consider the convolution y(t) = x(t) h(t), where h(t) = p2 (t) and x(t) is the triangle
x(t) = r(t) 2r(t 2) + r(t 4). Specically:
(a) Plot x() and h(2 ) as functions of .
(b) Plot x()h(2 ) as a function of .
(c) Determine y(2). The answer is a number.
(d) For what value of t does y(t) peak? Explain your answer.
(e) For what values of t is y(t) = 0? Explain your answer.
76. Using the graphical approach, determine y(t) = x(t) h(t) for 0 t 4, where
x(t) = e(t1) u(t 1) 3e(t3) u(t 3) and h(t) = u(t 1) u(t 5). Specically,
i) plot x( ) and h(t ) as a function of for t = 0
ii) plot x( )h(3 ) as a function of
iii) determine y(3)
iv) determine and plot y(t) for 0 t 4.

77. Graphically convolve x(t) = p4 (t) with h(t) = et p2 (t 1). Specically:
(a) Identify each region of t that has a dierent output expression.
141
(b) For each region, completely and exactly identify the integral equation that describes the output, but dont bother doing all the integrals.
(c) Determine y(1) (the answer is a number).
78. Convolve x(t) = ebt u(t 5) with h(t) = eat p10 (t 5).
79. Given h(t) = e2t u(t) and x(t) = e2t p4 (t 2), using the graphical convolution
approach, determine y(t) = h(t) x(t) for all t. What is y(4)?
80. Using the graphical approach, determine y(t) = h(t) x(t), for all t, where
h(t) = p2 (t 1) and x(t) = p4 (t + 3). Specically,
(a) Determine all the regions of t which have dierent expressions for y(t).
(b) For each region of t, determine the integral expression for y(t).
(c) Plot y(t) for 10 t 10.
81. Graphically convolve x(t) = e2t p3 (t 1.5) and h(t) = e2t u(t 1).
82. Using the graphical approach, determine y(t) = x(t) h(t) for 0 t 3, where
x(t) = et u(t) 3e(t2) u(t 2) and h(t) = u(t 1) u(t 5). Specically,
(a) sketch x( ) and h(t ) as a function of for t = 0
(b) sketch x( ) h(2 ) as a function of

(c) determine y(2)
(d) determine and sketch y(t) for 0 t 3.

83. Consider the periodic signal
x(t) =
k=
x1 (t
k
);
60
x1 (t) = 4 sin(260t) p(1/120) (t
1
).
240
Also consider a CT LTI system with impulse response h(t) = p(1/120) (t
1
).
240
(a) Sketch x1 (t), x(t) and h(t).

(b) Using the graphical convolution approach, perform the convolution
y1 (t) = x1 (t) h(t). Specically, for each region of t, provide the integral that
when solved gives y1 (t) over that region. Dont bother solving the integrals.
(c) In terms of y1 (t), describe y(t) = x(t) h(t).
84. Consider the CT LTI system with impulse response h(t) = u(t).
1
a) First consider input x1 (t) = t2 p1 (t 2 ). Use the graphical approach to determine
the output y1 (t). Clearly identify the output for each range of time t. Sketch this
output for 2 t 10.
142
b) Now consider the input x(t) = t2 p1 (t 1 ) t2 p1 (t 3 ). Use your result from

2
2
part a), plus convolution integral properties, to determine an expression for the
output y(t). Explain how you used properties to determine this expression.
1
c) Extra credit (2 pts.): Continuing part b), determine y( 2 ) and y(100).
85. Determine z(t) = u(t) u(t) u(t).

86. Given input x(t) = e5t u(t) to a LTI system with impulse response h(t) = e5t u(t1),
use the graphical approach to convolution to determine the output y(t) = x(t) h(t).
87. Using the graphical approach, determine y(t) = h(t) x(t), for all t, where
h(t) = eat u(t T ) and x(t) = r(t) = t u(t). Specically,
(a) Determine all the regions of t which have dierent expressions for y(t).
(b) For each region of t, determine the expression for y(t). Leave expressions in
integral form. That is, dont solve the integrals. However, make sure the integral
limits and integrands are correct.
88. Using convolution properties, convolve h(t) = p2 (t) with
x(t) =
k=
(t k) .
Sketch the result for 2 t 2.

89. Given the following results on convolution
tm u(t) u(t) =
tm u(t) r(t) = tm+2
tm+1
u(t) ; m = 0, 1, 2,
m+1
1
1
u(t) ; m = 0, 1, 2,
m+1 m+2
use convolution properties to determine y(t) = h(t) x(t) where:

(a) x(t) = 3 t p2 (t 1) and h(t) = 4 u(t 3). Determine y(6).
(b) h(t) = r(t 2) and x(t) = (t2 1) u(t 1). What is y(6)? (Hint: t2 = (t 1)2 +
2(t 1) + 1.)
90. Use the convolution table & properties to determine an expression for y(t) = h(t) x(t)
for h(t) = r(t + 1) and x(t) = u(t) + u(t 1) 2u(t 2). Determine y(3).
91. Using entries in the provided convolution table, and convolution properties, determine
h(t) = h1 (t) [h2 (t) h3 (t) h4 (t)] ,
where h1 (t) = p10 (t 3), h2 (t) = p20 (t 3), h3 (t) = (t 10) and h4 (t) = p20 (t 3).
Sketch h(t) over the range that it is non-zero.
143
92. Use convolution properties and the table provided to determine y(t) = x(t) h(t)
where x(t) = (t 4) + r(t 3) and h(t) = (t 3) + u(t) u(t 3). What is y(5)?
93. Given the input x(t) = 5 ej2t to a CT LTI system with impulse response
h(t) = e2t u(t), use the analytical convolution approach to determine the output
y(t). The result is a complex sinusoid. What is its frequency, amplitude and phase?
94. Let h1 (t) = eat u(t), h2 (t) = ea(t1) u(t 1), h3 (t) = ea(t3) u(t 3) and
h4 (t) = ea(t4) u(t 4). Using convolution and LTI system interconnection properties,
determine the overall impulse response h(t), where h1 (t) in parallel with h2 (t) is cascaded to h3 (t) in parallel with h4 (t). Use the fact that eat u(t) eat u(t) = teat u(t).
Simplify your answer as much as possible.
95. Consider two LTI subsystems with impulse responses h1 (t) = et u(t) and h2 (t) = p2 (t).
(a) Determine the impulse response, h(t), of the cascade of these subsystems.
(b) What is the output y(t) of this cascaded system for input x(t) = (t 3).
96. Consider the following CT LTI system (all components are CT LTI).
(a) For just the part of the system between points a, b, nd the impulse response
hab (t).
(b) Is hab (t) a stable system?
(c) What is the impulse response h(t) of the overall system?
(d) Is h(t) a causal system?
x(t)
(a)
h1 (t) = e u(t)
(b)
h5 (t) = e
y(t)
u(t)
h2 (t) = (t1)
h3 (t) = e u(t)
h4 (t) = e (t)
97. For each LTI system, determine if it is stable and/or causal. For each, show why.
(a) h(t) = e(3+j10)t u(t + 2).
98. For the following CT LTI system, nd the overall impulse response. Is the system
casual? Stable? Explain your answers in terms of your impulse response characteristics.
Which subsystems are causal? Stable?
99. Consider an RC circuit with impulse response h(t) =
Consider input x(t) = ej0 t u(t).
1
RC
et/RC u(t). Assume RC = 2.
(a) Use analytical convolution to determine the output y(t) for all time t.
144
x(t)
h1 (t) = e
0.5t
y(t)
u(t)
h4 (t) = u(t)
h (t) = (t3)
2
h (t) = e
3
(t)
1.5
(b) The steady state response for this problem is the output expression for large
enough values of time t so that the eect of turning the input on at t = 0 is neg1
ligible). Determine the steady state response for 0 = 2 . What is the frequency,
amplitude and phase of this steady state response?
100. Channel equalization (i.e. the cascade of two DT LTI systems): Consider a DT LTI
channel with impulse response hc [n] = [n] + 2[n 1]. The ideal equalizer would
have impulse response he [n] such that hc [n] he [n] = [n].
(a) Consider the equalizer he1 [n] = [n] 2[n 1] + 4[n 2] 8[n 3].
Determine the channel/equalizer impulse response h1 [n] = hc [n] he1 [n]. What
is the problem with this equalizer?
(b) Consider the equalizer y[n] = x[n] 2y[n1], where x[n] is the input and y[n] is
the output. What is its impulse response he2 [n]. Determine the channel/equalizer
impulse response h2 [n] = hc [n] he2 [n]. What is the problem with this equalizer?
1
1
(c) Consider the equalizer he3 [n] = 16 [n + 4] + 1 [n + 3] 4 [n + 2] + 1 [n + 1].
8
2
Determine the channel/equalizer impulse response h3 [n] = hc [n] he3 [n]. What
is the problem with this equalizer?
(d) Given your experience with (a-c), suggest an eective, causal, stable equalizer for
this channel? A delay from the channel input to equalizer output is OK.
101. Correlation: For many signal processing applications it is necessary to compare two
signals. Often, the most eective operator for comparing signals x[n] and v[n] is the
correlation
c[n] = x[n] v[n] .
When this operator is used to compare a signal to itself, it is called the autocorrelation.
Determine, for all n, the autocorrelation of x[n] = 0.9n u[n].
102. Consider two functions X() and W () of the independent variable which is dened
for continuous values over . Using the graphical approach to convolution,
determine V () = X() W (), for all , where X() = p2 ( 1) and
W () = p4 ( + 3). Specically,
(a) Determine all the regions of which have dierent expressions for V ().
(b) For each region of , determine the integral expression for V ().
(c) Plot V () for 10 10.
145
Continuous Time Transforms
In Subsection 1.4.5 of the Course Notes we introduced the general concept of representing
classes of signals as linear combinations of signals from basic signal sets. Table 2 (in that
Subsection) lists the topics of this Course in terms of this concept. In Chapter 2 we employed
this concept when we used signal representations in terms of delayed impulses to establish
the convolution operation as the time domain calculation of the output of a LTI system
given the input and impulse response. This corresponds to rows #1 and #5 of Table 2. In
this and the next Chapter we introduce transform based analysis of signals and systems.
Transforms are mathematical procedures for representing signals from a general class as a
linear combinations of basic signals. The transforms we consider are representations in terms
of complex-valued sinusoids and exponentials. In this Chapter we consider the following CT
transforms: the CT Fourier Series (CTFS), the CT Fourier Transform (CTFT), and the
Laplace Transform (LT). These correspond, respectively, to rows #2, #3 and #4 of Table
2. In Chapter 5 we cover analogous DT transforms. Chapter 3 is organized as follows:
1. In Section 3.1 we begin with two motivating discussions: one on the spectrum of a
signal, which displays the signals frequency content; and the other on consideration
of how CT LTI systems respond to exponential inputs, which both illustrates an important characteristic of CT LTI system, and provides a motivation for considering
decompositions of signals in terms of sinusoids and exponentials.
2. In Section 3.2 we then introduce the CTFS as a transform for representing CT periodic
signals in terms of harmonically related sinusoids. This yields the frequency content
of periodic signals.
3. In Section 3.3 we next show how the CTFT is used to represent CT energy signals as
linear combinations of complex sinusoids of all frequencies. We generalize the CTFT
to include periodic signals (incorporating the CTFS into the CTFT).
4. In Section 3.4 we next develop the LT representation of any CT signal in terms of complex exponentials. Since complex sinusoids are complex exponentials, we can consider
the LT as a generalization of the CTFT, though the CTFT, with its frequency content
connotation, is worth separate treatment.
5. In Section 3.5 we look at transform properties. Since all transforms are based on the
common theme of representing signals in terms of basic signal sets, these dierent
transforms have properties which are similar. Thus we will focus primarily on one
transform, the CTFT, and bring in other transforms when there are specic results
worth emphasizing. A transform property usually relates directly to one or more
important engineering problems.
146
Understand, on a basic level, what a transform is.
Understand specically what class of signals are represented by and what basic signals
are employed for the CTFS, the CTFT and the Laplace transform.
Be able to use the CTFS, CTFT and Laplace transform synthesis/analysis equations
and understand what information these equations derive.
Be familiar with the transform properties covered in this Chapter, understanding which
are more important in this Course and why.
Be able to identify transforms and inverse transforms of signals, using analysis/synthesis
equations and tables/properties.
3.1
Some Motivation for the Study of Transforms
We begin with two preliminary discussions that suggest advantages of transforms: 1) on the
spectrum of a signal, which displays the signals frequency content; and 2) on consideration
of how CT LTI systems respond to exponential inputs.
3.1.1
Preliminary View of Spectral Representation
In this Subsection of the Course we begin our study of frequency spectrum representation
and analysis of signals. This topic is universally referred to as Fourier analysis, honoring
the fundamental contributions of Jean Baptiste Fourier starting in the later 1700s. We rst
consider CT signals which are sums of real-valued sinusoids, since these signals are easiest
to visualize in terms of frequency content. The resulting spectrum is a representation of a
signals frequency content over all time. It is this type of spectrum that we will focus on
in this Course. However, in some cases we are interested in characterizing the frequency
content of a signal whose frequency characteristics are dierent at dierent times. That is,
we are interested in a time-frequency spectrum. For example, the music signals you generated
for Practicum 3 have time-varying frequency content, and the spectrogram Matlab function
was used to analyze this. To close this Subsection, we briey discuss this issue.
147
Spectrum of a Linear Combination of Real-Valued Sinusoids

Consider the following real-valued CT signal x(t) which is comprised of a DC (constant)
term plus a linear combination of N real-valued sinusoids:
N
x(t) = A0 + 2
Ak cos(k t + k ) .
(1)
k=1
The k th sinusoid has amplitude Ak , frequency k , and phase k . We assume k > 0; k =

1, 2, , N. To formalize our representation of this signal in terms of its frequency content,
lets rewrite Eq (1), using an Eulers identity, as
N
Ak ej(k t+k ) + Ak ej(k t+k )
(2)
Ak ejk ejk t + Ak ejk ejk t
(3)
ak ejk t + a ejk t
k
x(t) = A0 +
(4)
k=1
N
= A0 +
k=1
N
= a0 +
k=1
Eq(4) shows x(t) as a sum of complex-valued sinusoids. a0 = A0 is the DC (zero frequency)

term. The ejk t are the positive frequency terms, and the ejk t are the negative frequency
terms. The ak = Ak ejk ; k = 1, 2, , N are the complex-valued weights (representing both
magnitude and phase) applied to the N positive frequency components that constitute x(t).
The a = Ak ejk ; k = 1, 2, , N (which are the complex conjugates of the ak s) are the
k
complex-valued weights applied to the N negative frequency components that constitute
x(t).
Qualitatively, the frequency spectrum for a weighted sum of sinusoids is the content of
the signal as a function of frequency. Quantitatively, it is the plot or function of the ak
vs. the k . It indicates the magnitude and the phase of the sinusoidal components that
constitute the signal7 . In this Course we will use the terms spectrum and frequency spectrum
interchangeably.
Since the ak are generally complex-valued, we usually consider magnitude and phase spectrum plots which are, respectively, the plots of the |ak | vs. and the ak = k vs. . For
the sum of real-valued sinusoids considered here, both positive and negative frequencies are
typically plotted, and the magnitude spectrum is even symmetric, and the phase spectrum
is odd symmetric. The angle a0 is either 0 or , depending on whether a0 is positive or
negative (note that a0 is assumed real-valued).
More generally, a spectrum of a signal is obtained from a representation of that signal as a weights sum
(i.e. linear combination) of basic signals. The spectrum is the plot or function of weights vs. an index
or parameter representing the basic signals. A frequency spectrum is obtained when the basic signals are
sinusoids. Although there are other types of spectra (e.g. wavelet spectra), when dealing with only frequency
spectra, it is common practice to just use the term spectrum to imply frequency spectrum.
148
Example 3.1: Plot the spectra for
x(t) = 5 + 3 cos(2t +
) 4 cos(9t ) .
2
2
Solution: Using an Eulers identity, we can rewrite x(t) as

3 j(2t+(/2))
3 j(2t+(/2))
e
+
e
2 ej(9t(/2)) 2 ej(9t(/2))
2
2
3
3
= 5 + ej/2 ej2t + ej/2 ej2t 2ej/2 ej9t 2ej/2 ej9t .
2
2
The following gures are, respectively, the spectrum, the magnitude spectrum
and the phase spectrum for this signal.
x(t) = 5 +
ak
2j
spectrum
3
2 j
3
2
2j
(not to scale,
just a visualization)
9
ak
5
magnitude
spectrum
3
2
(to scale)
9
3
2
2
2
ak
phase
spectrum
(to scale)
9
/2
/2
2
/2
/2
9
Figure 57: An example of the spectrum of a sum of sinusoids.

Example 3.2: Consider the carrier sinusoid x(t) = cos(c t) and an information
signal v(t) (usually all positive-valued). The signal
x(t) = v(t) cos(c t)
is called an amplitude modulated (AM) signal. Given v(t) = 5 + 2 cos(40t) and
c = 400, sketch x(t) and plot the spectrum of x(t).
Solution: The rst plot shown below illustrates v(t) and x(t). To derive the
spectrum, rewrite x(t) as:
x(t) =
(5 + 2 cos(40t)) cos(400t)
1 j400t 1 j400t
e
+ e
= 5 + ej40t + ej40t
2
2
1 j360t 1 j360t
1
1
5 j400t 5 j400t
e
+ e
+ e
+ e
+ ej440t + ej440t .
=
2
2
2
2
2
2
(5)
The 2-nd and 3-rd plots show the spectra of v(t) and x(t) respectively.
149
x(t)
v(t)
t
ak
5
1
40 0
40
ak
5/2
5/2
1/2
440
1/2
400
360
360
400
440
Figure 58: The plot of an AM signal and its spectrum.

So the idea of a frequency spectrum is to identify or describe the frequency components of
a signal i.e. the sinusoidal components that constitute it. In this Chapter we introduce two
transforms which are used to determine a signals frequency content. One, the ContinuousTime Fourier Series (CTFS), is applicable to continuous-time periodic signals. The other,
the Continuous-Time Fourier Transform (CTFT), is appropriate for continuous-time energy
signals.
Time-Frequency Spectrum
In this Course we simply introduce the idea of a time-frequency spectrum. You will not
be responsible for this on homeworks, quizzes or tests. In Practicum 3 you computed timefrequency spectrum estimates using the Matlab spectrogram function.
Consider the sinusoid x(t) = A cos(0 t + ). Its spectrum is illustrated in Figure 59(a).
This spectrum is applicable for all time. More generally, the CTFS describes the frequency
content of a periodic signal for all time. An alternative spectral representation is shown in
Figure 59(b). This spectrum describes the frequency content of the signal as a function of
time. (As is common with this type of plot, only the positive frequency content is shown,
and frequency in Hertz is used.) For the x(t) we are considering here, the frequency content
is constant for all time.
150
(a)
A
2
A
2
2 f0
2 0
f
(b)
f (frequency)
f0
t (time)
(c)
fk
262
294
330
349
392 440
494
523
middle
note
Each note has a

0.2 sec. duration.
They are concatenated.
f (frequency)
523
494
440
392
349
330
294
262
0.2
.4
.6
.8
1.0
1.2
1.4
t (time)
(d)
Figure 59: Frequency spectra: (a) as a representation of frequency content for all time; (b)
the time-frequency spectrum of a real-valued sinusoid; (c) the time-frequency spectrum of a
musical scale; and (d) an estimate of the time-frequency spectrum.
Some signals have frequency content that changes over time. If we are interested in how
it changes, then we want a frequency characterization of the signal that changes with time.
That is, we want a time-frequency spectrum which is often referred to as a spectrogram. As an
example, consider the stepped frequency signal described as a sequence of musical notes in
Practicum 3, as illustrated in Figure 59(c). Figure 59(d) shows the time-frequency spectrum
estimate you will generate in Practicum 3.
151
Generalizing the Idea of Spectral Representation

If you understand the above discussion on frequency spectrum, and if you buy into the
notion that knowing the frequency content of a signal may be useful, then the logical extension is to pursue a more general approach to determining the spectrum of any signal. In
Table 5 we present six transforms which are used to identify the frequency content of various
classes of signals. Note that, as presented in Table 5, these transforms are applicable to
dierent classes of signals, e.g. the column #2 transforms are for CT signals, and the row
#2 transforms are for energy signals. We will cover these six transforms in this Course.
Table 5: Transforms
P
e
r
i
o
d
i
c
E
n
e
r
g
y
Discrete Time
Discrete Time Fourier Series
(DTFS): Period N
x[n] =
X[k] =
k=<N >
1
N
n=<N >
x[n] ej(2/N )nk
Discrete Time Fourier Transform

(DTFT)
x[n] =
1
2
X(ej ) =
G
e
n
e
r
a
l
X[k] ej(2/N )kn
X(ej ) ejn d
n=
x[n] ejn
z-Transform
x[n] =
1
2j
X(z) =
X(z) z n1 dz
n=
x[n] z n
Continuous Time
Continuous Time Fourier Series
(CTFS): Period T0 ; 0 = 2
T0
k=
x(t) =
ak =
1
T0
ak ejk0t
x(t) ejkot dt
<T0 >
Continuous Time Fourier Transform

(CTFT)
x(t) =
1
2
X(j) =
X(j) ejt d
x(t) ejt dt
Laplace Transform
x(t) =
1
2j
X(s) =
c+j
cj
X(s) est ds
x(t) est dt
As an example of how these transforms address the concept of a frequency spectrum,

consider the Continuous Time Fourier Series (CTFS) in row #1, column #2 of Table 5.
Note from its position in the table that the CTFS is for CT periodic (power) signals. As
with the other transforms, two equations are given. The rst equation is called the synthesis
equation because it indicates how we can construct a given signal as a linear combination
of basic signals. In the CTFS case, we see from the synthesis equation that the basic
signals are the set of all complex-valued sinusoids that have frequencies that are integer
152
multiples of 0 = 2 , where T is the period of the signal. These frequencies are called
T
the harmonics of 0 . This set of basic signals has an innite number of members, but the
membership is countable. This synthesis equation suggests that in general a CT periodic
signal is composed of an innite number of harmonic frequency components. The spectrum
is the plot or function of the ak vs. the k = k0 . The second equation is called the analysis
equation because to indicates how to analyze the signal x(t) to determine the ak . That is, it
computes the spectrum.
In the context of spectral representation, the CTFS is the easiest of the six transforms
in Table 5 to understand. So we will cover the CTFS rst, in Section 3.2. Then, in the
remainder of this Chapter, we will introduce and consider the application of the other CT
transform in Table 5 (i.e. column #2). In Chapter 5 we introduce the DT transforms in
Table 5.
We will see that there are many similarities between these transforms. There should be,
since they basically do the same thing they represent, via a synthesis and an analysis
equation, general signals as linear combinations of basic signals. As the old saying goes,
if youve seen one transform, youve seen them all.
That is, if you understand one transform, then you understand the basic idea behind all
transforms. The key, then, to making a lot of money and to earning the clear adoration of
your peers and loved ones, is to know how to use transforms.
3.1.2
System Response to Complex Exponential Signals
Consider a CT LTI system with impulse response h(t). Let the input be x(t) = est , where
s = + j is a complex valued constant of arbitrary value8 . Then the output is
y(t) = h(t) x(t) =
=
= est
h( ) x(t ) d
h( ) es(t ) d
h( ) es d .
(6)
(7)
(8)
Note that on the right side of this equation,
h( ) es d
(9)
is not a function of t. However, it is a function of s. In general it is complex-valued.
x(t) = et ejt is a complex exponential. Its oscillation rate is dictated by . Its decays as dictated by
. If < 0, it decays as t . If decay rate > 0, it decays as t . For = 0, it does not decay
(i.e. its a complex sinusoid).
153
Let
H(s) =
h(t) est dt = |H(s)| ej
H(s)
(10)
denote the complex-valued term on the right of Eq(8). Then, for input x(t) = est , with
s = + j, we get output
y(t) = H(s) est = |H(s)|et ej(t+
H(s))
(11)
This simple result implies something very important. It shows that

if we input a complex exponential into a LTI system, we get that same complex
exponential out, altered in magnitude and phase by the system.
The input magnitude multiplied by |H(s)| and input phase added by
qualication is that, for a given s, the integral
H(s) =
H(s).The only
h(t) est dt
= |H(s)| ej
(12)
H(s)
(13)
must exist (i.e. be nite). Otherwise, for that s, input est results in an innite output.
Figure 60 illustrates this CT LTI system result for exponential inputs.
e
st
CT LTI system
h(t); H(s)
H(s) =
H(s) e st
h(t) est dt
Figure 60: The CT LTI system response for an exponential input.

So, why is this important, and what does it have to do with transforms? Well, considering
Eqs (10,11), H(s) appears to be another useful CT LTI system description that we can
employ to compute outputs for exponential inputs. Can we use H(s) for other types inputs?
Say we have input
N
ak esk t ,
x(t) =
(14)
k=1
where the ak s and sk s are complex-valued constants. Exploiting the linearity property of
LTI systems, we have output
N
H(sk ) ak esk t .
y(t) =
(15)
k=1
So, for any input x(t) that can be represented as a linear combination of complex-valued
exponentials, H(s) can be used directly to determine the output y(t). Eq (12) is the Laplace
transform of h(t) i.e. in Table 5 it is the 3-rd row, 2-nd column entry.
We will get back to this idea when we consider CT LTI systems in Chapter 4. For now,
this result motivates a closer look at representing signals as linear combinations of complexvalued exponentials in general, and complex-valued sinusoids in particular.
154
In the next two examples we use the H(s) function identied above to determine the
output of a CT LTI system due to exponential-like inputs. These examples should seem
familiar. They are very similar to numerous Chapter 1 & 2 examples involving complexvalued sinusoidal inputs to DT LTI systems.
Example 3.3: Consider a CT LTI system with impulse response h(t) = 4 e3t u(t).
First determine the output for general exponential input x(t) = 5 est . Then nd
the output specically for x(t) = e7t .
Solution: First lets determine H(s):
H(s) =
=
h(t) est dt =
4
e(s+3)t
(s + 3)
4 e3t est dt = 4
e(s+3)t dt
4
,
s+3
assuming that Re{s} > 3 so that the integral exists. Then, in general,
x(t) = est
y(t) =
4
est
s+3
y(t) =
4 7t
e .
10
as long as Re{s} > 3. So,

x(t) = e7t
Example 3.4: Repeat Example 3.3 for input x(t) = 5 cos(10t).

Use an Eulers identity it write x(t) as a sum of complex-valued exponentials.
Then use the general result from Example 3.3, along with linearity, to determine
the output y(t). Noting that for this system H(j10) = H (j10), i.e.
H(j10) = |H(j10)| ej
H(j10)
H(j10) = |H(j10)| ej
express y(t) explicitly as a real-valued sinusoid.

Solution: First,
H(j10) =
4
4
=
ej1.48 .
j10 + 3
31.56
Then, for
x(t) =
5 j10t
5 j10t
e
+
e
,
2
2
we have that
5 4
5 4
ej1.48 ej10t +
e+j1.48 ej10t
2 31.56
2 31.56
20
20
j(10t1.48)
e
+
ej(10t1.48)
=
2 31.56
2 31.56
= 0.62 cos(10t 1.48) .
y(t) =
H(j10)
3.2
155
CT Fourier Series (CTFS) Representation of Periodic Signals
In this Section we consider how to express any CT periodic signal as a linear combination
of complex sinusoids. In pursuit of this, rst consider the complex sinusoid ej0 t which has
fundamental frequency 0 and therefore fundamental period T where
2
(radians/sec.) .
(16)
0 =
T
For integers k = 1, 2, 3, , the frequencies k0 are called the harmonics of o , and
the set of signals
2
k (t) = ejk0 t = ejk T t ;
k = 0, 1, 2, 3,
(17)
are the set harmonically related complex sinusoids (including DC, the k = 0 term).
Every signal k (t) has a period of T (not necessarily its fundamental period). Thus, any
signal of the form
x(t) =
=
k=
ak k (t)
ak ejko t
(18)
k=
is periodic with period T . For all possible weights ak , Eq (18) forms the class of all signals
that are linear combinations of signals from the basic set {ejko t ; k = 0, 1, 2, }. All
signals from this class are periodic with period T .
Combined, the k = N components of the x(t) in Eq (18) are called the N th harmonic.
As mentioned earlier, the k = 0 component is the DC (a.k.a. constant, zero-th harmonic)
term. We can write x(t) from Eq (18) in the form
x(t) = a0 +
= a0 +
ak ejko t +
k=1
k=1
ak ejko t
(19)
k=
ak ejkot + ak ejko t
(20)
which shows the DC component a0 and the sum for all the other harmonics.
Let ak = Ak ejk , where the Ak and k , i.e. the magnitudes and phases respectively, are
real-valued. Then for the x(t) considered in Eq (20) we have
x(t) = a0 +
Ak ej(ko t+k ) + Ak ej(ko tk )
k=1
(21)
If ak = a (i.e. if the ak coecients are complex symmetric in k), then ak = Ak ejk , so

k
that
x(t) = a0 +
k=1
= a0 + 2
Ak ej(ko t+k ) + Ak ej(ko t+k )

Ak cos(ko t + k ) .
k=1
In this case, if we assume a0 is real-valued, x(t) is real-valued.
(22)
(23)
156
The CTFS: So far we just are just describing a sum of harmonically related sinusoids. Now
lets consider how to represent a periodic signal as a sum of harmonically related sinusoids.
Theorem: Under very general conditions (i.e. over a period x(t) must be absolutely integrable, and have a nite number of maxima/minima and discontinuities) any CT periodic signal with period T can be expressed as the CTFS
x(t) =
ak ejko t
(24)
x(t) ejk0 t dt ,
(25)
k=
where
ak =
1
T
<T >
where 0 = 2 , and the < T > argument on the integral means that we can
T
integrate over any period (i.e. over any duration of t of length T ).
Proof9 :
x(t) =?
k=
=
=
T /2
T /2
T /2
T /2
1
T
x( )
T /2
T /2
1
T
x( ) ejk0 d
ejk0 (t )
ejk0t
(26)
(27)
k=
x( ) (t ) d = x(t) .
(28)
In going from the 1-st to 2-nd lines above (i.e. switching the order of the summation and integral) the general conditions noted above are required. In going
from the 2-nd to 3-rd lines above, we make use of that fact that
1
T
k=
ejk0(t ) = (t )
T
T
t
2
2
(29)
which is a result from generalized functions (a topic beyond the scope of this
Course).
Eq (24) is called the exponential CTFS. Eq (23), if it applies (i.e. if ak = a ), is called
k
the trigonometric CTFS. Mostly we will use Eq (24).
Looking more closely at ak = Ak ejk , Ak is the magnitude and k the phase of the coecient
ak . Eq (24) in general, and Eq (23) if ak = a , show us the meaning of Ak and k . Ak is
k
the magnitude of the k th complex or real-valued sinusoidal component and k the phase.
9
You are not responsible for the proofs of any of the transforms covered in this Course. However, to have
condence in these transforms, it is useful to see the proof of at least one of them. Also, following the proof
is a good mathematical exercise, and it reinforces the concept of a transform.

Example 3.5: Let x(t) =
CTFS coecients.
157
l=
p2T1 (t lT ) where T > 2T1 . Determine the
Solution:
Example 3.6: Continuing Example 3.5, let T1 =

cients.
T
.
4
Determine the CTFS coe-
Solution:
The following Matlab Demo provides a visualization of the CTFS. The periodic signal
from Example 3.6 is considered. For several values of N, the periodic signal is approximated
by summing all harmonic components up to N. Figure 61 shows the result, for maximum
N = 151. As N increases, the quality of the approximation improves. Run this program
yourself, and on the 4-th plot zoom in on a transition region of the square wave.
158
echo on
%
%
Matlab CTFS Demo
%
Constructing a CT Square Wave from Sinusoids
%
%
x(t) == unit magnitude square wave of period 1, w_0 = 2 \pi
%
%
x(t) =? 0.5 + 1/pi cos(2*pi*t) - 1/(3pi) cos(6*pi*t) +
%
1/(5pi) cos(10*pi*t) - 1/(7pi) cos(14*pi*t) + ...
%
%
Enter N (odd) == highest harmonic before running
%
pause
%
Construct and plot samples of x(t) for -1 <= t <= 1.
%
t = -1:.005:1;
x=ones(1,401);
x(51:151) = zeros(1,101), x(251:351) = zeros(1,101);
pause
subplot(221)
plot(t,x),xlabel(t),ylabel(amplitude),title(square wave)
text(0.5,1.5,Thomas Friedman),axis([-1 1 -0.5 1.5])
pause
%
Generate array of required CTFS coefficients
%
Ak = zeros(1,N+1);
Ak(1) = 0.5;
% DC
kk = 1:N;
Ak(2:N+1) = sin(kk*pi/2)./(kk*pi);
%
%
Approximate x(t) with harmonics 1,3
%
omega0 = 2*pi;
x1 = Ak(1)*ones(1,length(t));
for k=1:3
end
pause
subplot(222)
plot(t,x1,b),xlabel(t),ylabel(amplitude),title(1,3 harmonics)
text(0.5,1.5,Alan Oppenheim),axis([-1 1 -0.5 1.5])
pause
%
Approximate x(t) with harmonics 0,1,3,5,7,9,11
%
159
x2 = x1;
for k=4:11;
end
pause
subplot(223)
plot(t,x2,r),xlabel(t),ylabel(amplitude),title(0,1,3,5,7,9,11 harmonics)
text(0.5,1.5,Khaled Hosseini),axis([-1 1 -0.5 1.5])
pause
%
Approximate x(t) with harmonics 0,1,3, ... , N (odd)
%
x3 = x2;
for k=12:N;
end
pause
subplot(224)
plot(t,x3,m),xlabel(t),ylabel(amplitude),title(0,1,3,..,N harmonics)
text(0.5,1.5,Robert Pirsig, Ayn Rand),axis([-1 1 -0.5 1.5])
square wave
1,3 harmonics
1.5
1.5
Thomas Friedman
Alan Oppenheim
1
amplitude
amplitude
0.5
0.5
1
0.5
0.5
0
t
0.5
0.5
1
1,3,5,7,9,11 harmonics
0.5
1.5
Khaled Hosseini
Robert Pirsig, Ayn Rand

1
amplitude
1
amplitude
0
t
1,3,..,N harmonics
1.5
0.5
0.5
1
0.5
0.5
0.5
0
t
0.5
0.5
1
0.5
0
t
0.5
Figure 61: CTFS approximations of a square wave using dierent numbers of harmonics.
160
Example 3.7: Given
x(t) =
l=
1
x1 (t) = r(t) p1 (t ) ,
2
x1 (t l) ;
determine the approximation x(t) based on only the DC and 1st harmonic terms
of the CTFS expansion.

Solution: First determine the DC term a0 . Pick the period 0 < t 1 to integrate
over, since for this range x(t) is easily described.
a0
1
=
1
1
0
t2
t dt =
2
=
0
1
2
This is just the average value of x(t). Note that 0 = 2. For a1 and a1 = a ,
1
a1 =
1
1
1
0
t ej2t dt =
1 j/2
j
=
e
= A1 ej1 .
2
2
Then,
x(t) = a0 + 2 A1 cos(0 t + 1 )
1
1
+
cos(2t + (/2))
=
2
1
1
=
sin(2t) .
2
The gure below shows a sketch of x(t) and its approximation based on only the
DC and 1-st harmonic components.
x(t)
DC
DC + 1st harmomic
1/2
Figure 62: A sawtooth waveform and its DC, 1-st harmonic approximation.
Table 6 provides a list of some useful CTFT pairs. In Example 3.5 we derived Entry #4,
and in Example 3.7 we derived some of Entry #10. As with all tables, the idea is to attempt
to use the table to identify a required result before resorting to deriving it. In Examples 2.8
& 2.9 we will expand on our table, we derive Entries #7 & #9.
Notationally, we refer to a CTFS pair as follows:
x(t)
ax .
k
(30)
The superscript x on ax is often dropped whenever it is clear which signal the coecients
k
correspond to.
161
Table 6: Continuous Time Fourier Series (CTFS) Pairs.
signal
type
DC
Periodic Signal
(x(t) = x(t + T0 ), all t)
x(t) = A (i.e. a DC signal)
cosine
x(t) = A cos(2f0 t + )
2f0
sine
x(t) = A sin(2f0 t + )
2f0
pulse
x(t) =
x1 (t nT );
n=
train
5
square
impulse
CTFS
(ak ; k = 0, 1, 2, , )
a0 = A
ak = 0, otherwise
a1 = A ej ; a1 = A ej
2
2
ak = 0, otherwise
a1 = A ej(/2) ; a1 = A ej(/2)
2
2
ak = 0, otherwise
A
k
2
T
ak =
2
T
a0 = a2 = a4 = = 0
sin(k /T ), all k
x1 (t) = A p (t); < T
x(t) =
x1 (t nT );
n=
wave
x1 (t) = 2A pT /2 (t T ) A
4
x(t) =
n=
A (t nT )
ak =
2A
k
2
T
ak =
A
;
T
2
T
a0 =
A
;
4
T
ak = A (1)2 1) , k = 2, 4
(k
ak = 0, otherwise
a0 = 2A
ejk/2 , otherwise
all k
train
x1 (t nT );
halfwave
x(t) =
x1 (t) = A cos
2
t
T
rectied
cosine
fullwave
rectied
cosine
x(t) = A cos
2
t
T
n=
every
a1 = a1 =
A
4
(k+2)/2
pT /2 (t)
ak =
x(t) =
x1 (t nT );
n=
2
T
2A (1)k
,
(14k 2 )
k = 1, 2
a2 =
A
,
4j
A
4j
a2 =
other
sine
10
saw
x1 (t) =
x(t) =
A sin((4/T )t) 0 t T /2
0
T /2 < t T
x1 (t nT );
n=
tooth
11
exp. ak s
ak = ej(k1)/2 2A sin(k/2) otherwise

(4k 2 )
2
T
x1 (t) = A t pT (t T )
2
1c2
1+c2 2c cos(o t)
a0 =
AT
2
AT
ak = j 2k , otherwise
ak = c|k|
(1 < c < 1)
162
Example 3.8: Consider the signal y(t) = |x(t)| where x(t) = cos(t). Determine
the CTFS coecients of y(t).
x(t) = cos(t)
....
y(t) = | x(t) | = | cos(t) |
....
....
....
Figure 63: A fullwave rectied sinusoid.

Solution: T = , 0 =
ay =
k
/2
/2
2
T
= 2. Note that y(t) = cos(t); (/2) < t (/2).
cos(t) ej2kt dt
ej2kt
(j2k cos(t) + sin(t))
1 4k 2
/2
/2
1
1
ej2k(/2) (j2k cos(/2) + sin(/2)) ej2k(/2) (j2k cos(/2) + sin(/2))
1 4k 2
2 (1)k
=
.
1 4k 2
2
Note that ay = . Thus the input has a zero DC component (ax = 0) while the
0
o
output does not. With this example, we have derived Entry #8 of Table 6.
Example 3.9: Let x(t) be periodic with period T0 = 4, and

x(t) =
0t2
2<t4
sin(t)
0
Determine the CTFS coecients.

Solution:
1
ak =
T
1
=
4
1
=
8j
.
.
.
T
0
2
0
x(t) ej0 kt dt =
1
4
2
0
sin(t) ej(/2)kt dt
1
ejt ejt j(/2)kt
e
dt =
2j
8j
ej((/2)k)t
ej(+(/2)k)t
j( (/2)k)
j( + (/2)k)
2j ejk/2 sin(k/2)
(4k 2 )
1
4j
1
4j
This is Entry #9 of Table 6.
k = 2
k=2
k = 2
ej((/2)k)t ej(+(/2)k)t
0
2
0
dt

3.2.1
163
A Few CTFS Properties
Properties are important features of transform. They make the transform easier to work
with, and thus extend the transforms utility. Furthermore, as we will see later in this
Course, a transform property is often directly related to one or more important engineering
problems. Since transform all do the same thing (i.e. decompose a general signal into a
linear combination of basic signals), properties for dierent transforms are often similar.
So learning a set of properties for one transform directly contributes to understanding and
employing other transforms.
In this Subsection we introduce just a few of the many properties of the CTFS. Later in
this Section, in a more general context of CT transforms, we will more systematically and
extensively cover transform properties. Becoming familiar with the properties presented here
will reduce the stress of learning properties later.
Linearity: Consider two signals, x1 (t) and x2 (t), each periodic with period T , with CTFS
coecients ax1 and ax2 respectively. Given any two constants c1 and c2 , the CTFS pair is
k
k
c1 x1 (t) + c2 x2 (t)
c1 ax1 + c2 ax2 .
k
k
(31)
Time Shift (i.e. Delay): Given a signal x(t), periodic with period T , with CTFS coecients ak , then for any delay , x(t ) is also periodic with period T with CTFS pair
x(t )
ejk0 ak = ejk0 Ak ejk = Ak ej(k k0 ) .
(32)
This property makes sense. If you delay a periodic signal x(t), then you delay each of its
sinusoidal components. This does not alter the sinusoidal components magnitudes. It only
changes their phases. The more delay, the more phase shift. The higher the frequency, the
greater the phase shift.
Symmetry: For real-valued periodic x(t), with CTFS coecients ak , we have that
ak = a .
k
(33)
That is, the CTFS coecients are complex symmetric.
These three properties, and Parsevals Theorem which we present after a few examples,
are enough to serve our objectives for the time being. We will develop more properties later,
after we introduce the CTFT (and the generalization of it that includes the CTFS) and the
Laplace transform. Below we use the properties just introduced, along with known CTFS
pairs from Table 6, to solve some new CTFS problems.
164
Example 3.10: Determine the CTFS of the following periodic signal
y(t) =
n=
(t T1 nT0 ) .
Hint: Use the Entry #6 of Table 6 along with the delay property.
Solution:
In this example, what happens if T1 = T0 ?

Example 3.11: Consider the square wave
x(t) =
l=
x1 (t lT ) ;
x1 (t) = pT /2 (t (T /4)) pT /2 (t (3T /4)) .
Determine the CTFS coecients. Hint: let x3 (t) denote the signal from Example
3.6, and note that x(t) = 2 x3 (t (T /4)) 1.
x 3 (t)
x(t)
1
....
....
....
....
t
T T
4 2
T
Figure 64: Example 3.11 signals.
Solution:
2 ax3 1
0
2 ax3 ej(2/T )k(T /4) = 2 ax3 ejk/2
k
k
ax =
k
k=0
k=0
Using Entry #4 of Table 6,

ax
k
2 ax3 1 = 0
0
= 2 ax3 ejk/2 = 0
k
2
2 ax3 ejk/2 = k ejk/2
k
k=0
k = 2, 4 ,
k = 1, 3 ,
This is Entry # 5 of Table 6. Observe that x(t) has no DC component (i.e. the
average value of x(t) is zero.)
165
Example 3.12: Determine the CTFS of the following periodic signal

z(t) =
n=
z1 (t 4n) ;
z1 (t) = (4 t) p4 (t 2) .
Solution: Let z3 (t) denote the sawtooth wave in Entry #10 of Table 6, with
A = 1 and T = 4. Then
z(t) = 4 z3 (t)
so
az
k
4 az3 = 2
k
2
az3 = j k
k
k=0
k=0
This is the signal you will analyze in Practicum 4a.

Parsevals Theorem: The power of a CT periodic signal with period T can be computed
in either the time domain or frequency domain as follows
P =
1
T
<T >
|x(t)|2 dt =
k=
|ak |2 .
(34)
By the symmetry property, if x(t) is real-valued,

P = |a0 |2 + 2
k=1
|ak |2 .
Example 3.13: For the signal in Example 3.7, determine the percentage of total
power that is in DC and the 1-st harmonic combined.
Solution:
(35)
166
3.3
The Continuous Time Fourier Transform (CTFT)
In this Section we introduce the CTFT. This transform provides the decomposition of CT
energy signals, so it is the column #2, row #2 entry in our transform table Table 5. The
decomposition is in terms of complex-valued sinusoids, so it generates a frequency spectrum of
an CT energy signal. We begin by introducing the transform equations, and then consider a
number of examples that generate entries in a CTFT table. We then develop a generalization
of the CTFT so as to include periodic signals, thus enabling us to represent both periodic
power and energy signals using a single frequency transform. After introducing the Laplace
Transform (LT) in Section 3.4, we will return the our study of the CTFT in Section 3.5
where we cover both CTFT and LT properties.
Let x(t) be an energy signal. The CTFT indicates that x(t) can be represented as
x(t) =
1
2
X(j) ejt d .
(36)
That is, x(t) can be represented as or decomposed into a linear combination of all the CT
complex-valued sinusoids ejt over the frequency range . Eq (36), called the
Inverse CTFT (ICTFT), is the synthesis equation since it generates x(t) from basic signals.
The CTFT is
X(j) =
x(t) ejt dt .
(37)
It is the analysis equation because it derives the weighting function X(j) for the synthesis
equation10 . The reason for the j in the argument of X(j) has to do with the relationship
between the CTFT and the LT (we will explain this later). For now remember that X(j)
is simply a function of .
10
Proof of the CTFT involves plugging Eq (37) into Eq (36) and simplifying to show that the right side
of Eq (36) does reduce to x(t). This derivation, specically a change of the order of two nested integrals,
requires certain assumptions. These assumptions are that x(t): be absolutely integrable, and have a nite
number of minima/maxima and discontinuities. The absolutely integrable requirement essentially (but not
exactly) means that x(t) be an energy signal.

Example 3.14: Determine the CTFT of the signal x(t) = p2T1 (t). Based on the
result, comment on the frequency content of the signal.
Solution:
Note that X(j) has innite extent, indicating that it contains innitely high
frequency components. This should not be surprising since x(t) has discontinuities, which require innitely high frequency components to synthesize. Also note
the X(j) is largest for lower frequencies, indicating that in some sense x(t) in
mostly a low frequency signal.
This Example derives Entry #8 of the CTFT table, Table 7, on the next page.
Example 3.15: Determine the ICTFT of X(j) = p2W (). Compare characteristics of x(t) and X(j).
Solution:
Note that with the X(j) given in this example, x(t) is a purely low frequency
signal. The manifestation of this in the time domain is that x(t) is smooth (e.g.
there are no discontinuities).
This Example derives Entry #9 of Table 7.
167
168
Table 7: Continuous Time Fourier Transform (CTFT) Pairs.

#
Signal
( t)
CTFT
( )
(t)
(t )
ej
u(t)
1
j
eat u(t);
+ ()
1
a+j
teat u(t); Re{a} > 0
Re{a} > 0
tn1 at
e u(t);
(n1)!
ea|t| ;
1
(a+j)2
1
(a+j)n
Re{a} > 0
2a
a2 + 2
Re{a} > 0
eat cos(0 t) u(t); Re{a} > 0
j+a
(j+a)2 + 2
eat sin(0 t) u(t); Re{a} > 0
0
(j+a)2 + 2
10
pT (t) = u(t + T ) u(t T )

2
2
11
1
t
sin(W t)
sin
T
2
p2W ()
12
sin2 (W t)
(t)2
13
c
c2 +t2
ec||
14
ej0 t
2( 0 )
15
cos(0 t)
( 0 ) + ( + 0 )
16
sin(0 t)
(
j
17
ak ejk0t
18
ak ejk0 t
k=
19
(t nT )
n=
1
2
p2W () p2W ()
0 ) ( + 0 )
j
2ak ( k0 )
2ak ( k0 )
k=
2
T
k=
2
k
T
169
Example 3.16: Determine the CTFT of the signal x(t) = (t). Based on the
result, comment on the frequency content of the signal.
Solution:
Note the consistence between the time and frequency domain representations
of this signal. x(t) changes innitely over zero time, which implies very high
frequency components. In fact, X(j) indicates that the impulse consists of
equal content over all frequency. Its the most wideband signal.
This Example derives Entries #1 & #2 of Table 7.
Example 3.17: Determine the CTFT of x(t) = eat u(t) where Re{a} > 0.
Solution:

Example 3.18: Determine the CTFT of x(t) = ea|t| where Re{a} > 0. Comparing the solution derivation and result to those of Example 3.17, can you anticipate
one or more properties of the CTFT?
Solution:
x(t) = eat u(t) + eat u(t)
X(j) =
=
eat ejt dt +
eat ejt dt =
1
2a
1
+
= 2
j + a
j + a
a + 2
1
+
j + a
e(ja)t dt
How about linearity? Perhaps theres some kind of symmetry property.

170
Example 3.19: Determine the ICTFT of X(j) = 2( 0 ). Note the x(t) is

a periodic (power) signal. Try deriving this X(j) from your x(t).
Solution:
This is not an energy signal! Whats up? See Subsection 3.3.1 below.
Example 3.20: Starting with the result of Example 3.19 (i.e. starting with the
CTFT pair established in Example 3.19), determine the CTFT of
x(t) = ak ejk0t , where ak is an arbitrary complex constant and k0 is the
frequency of sinusoidal x(t). Does this result suggest a useful CTFT property?
Solution: By the linearity property of the integral (in the CTFT equation),
ak ejk0 t
ak 2 ( k0 )

Example 3.21: Extending Example 3.20, determine the CTFT of
x(t) =
ak ejk0 t .
k=
Solution: To extend the Example 3.20 result, we anticipate the linearity property
of the CTFT (recall this property for the CTFS).
x(t) =
jk0 t
ak e
k=
X(j) =
k=
ak 2 ( k0 )
X(j )
(a 02 )
CTFT
Spectrum
(a 12 )
.... (a 2 )
1
(a 22 )
....
2 0
Figure 65: CTFT of a periodic signal.

This Example derives Entry #18 of Table 7, and leads the main point on the
next Subsection.

3.3.1
171
The Generalized CTFT
For CT energy signals, the generalized CTFT is just the CTFT described above. Concerning
CT periodic signals, rst recall form our CTFS discussion that for any periodic power signal
x(t) with fundamental frequency 0 has CTFS representation
x(t) =
ak ejk0 t
(38)
k=
From Example 3.21, we then have that, in terms of the CTFS coecients, the generalized
CTFT of a periodic signal is
X(j) =
k=
2 ak ( k0 ) .
(39)
Thats it. We now have a single CT transform, the generalize CTFT (or, for now on, the
CTFT for short) that can be used to represent and analyze the frequency content of both
energy and periodic CT signals. Note, however, that we cant just ignore the CTFS. For the
CTFT representation of periodic signals, we still need to know the CTFS coecients.
Example 3.22: Determine the CTFT of the impulse train
x(t) =
n=
(t nT ) .
Solution: Using Entry # 6 of the CTFS table (Table 6),

2
T
X(j) =
k=
( k0 )
X(j )
(2 /T)
x(t)
(1)
....
CTFT
....
2T
2T
3T
0 = 2 /T
t
....
....
20
2 0
3 0
Figure 66: CTFT of a impulse train.

This is Entry #19 of Table 7. In Chapter 4 we will see that this CTFT pair, while
not representing a realizable signal, is key to understanding why sampling works
(i.e. why and when we can exactly reconstruct a CT signal from its samples).
172
Example 3.23: Determine the CTFT of
x(t) = (1 + 0.25 cos(2t)) cos(40t) .
Solution:
x(t) =
1 j40t
1 j40t
1 j38t
1 j38t
1 j42t
1 j42t
e
+
e
+
e
+
e
+
e
+
e
2
2
16
16
16
16
1
1
1
1
( 40) + ( + 40) +
( 38) +
( + 38)
2
2
16
16
1
1
( 42) +
( + 42)
+
16
16
X(j) = 2
V(j )
(2)
(/4)
2
(/8 e
(/8 e
j /2
42
j /2
40
38
( e
(/8 e
(/4)
X(j )
( e
j /2
j /2
(/8 e
j /2
38
j /2
(/8 e
40
42
j /2
Figure 67: CTFT of an AM signal.

You will use this signal, and its CTFS and CTFT which you now know, in
Practicum 4b as a test case for a spectrum analyzer you develop.
3.4
173
The Laplace Transform (LT)
The LT introduced rst below, in Subsections 3.4.1-3, is termed the Bilateral LT (BLT)
because both sides of the time axis (i.e. t < 0 and t > 0) are considered. The BLT is also
called the two-sided LT. In most situations where a LT is useful, the BLT will be used. In
Subsection 3.4.4 we will introduce the Unilateral LT (ULT), also called the one-sided LT,
which deal with the t 0 part of a signal only. The reason for considering a ULT is that it
facilitates a consideration of any initial conditions of a system.
3.4.1
The Bilateral Laplace Transform (BLT)
Let x(t) be a signal with unrestricted extent in time. The bilateral Laplace Transform is,
for complex-valued variable s = a + j,
X(s) =
x(t) est dt .
(40)
If we evaluate at the BLT specically for s = j (i.e. for = 0, for s on the j axis of the
s-plane), we have
X(j) =
x(t) ejt dt .
(41)
Does this look familiar? Its the CTFT equation.

Example 3.24: Let x1 (t) = eat u(t). Find its BLT.
Solution:
This example derives Entries #2 & #4 of Table 8.

The BLT of x(t) consists of the function X(s) along with the Region Of Convergence (ROC)
Rx . The ROC is the region of the s plane (i.e. all values of complex s) such that the BLT
integral can be performed. In Example 3.24 Rx is the half-plane to the right of Re{a}.
174
Example 3.25: Let x2 (t) = eat u(t). Find the BLT X(s).
Solution:
This example derives Entries #3 & #5 of Table 8.

The BLT functions X1 (s) and X2 (s) from Examples 3.24 & 3.25 are equivalent. The ROCs
are dierent. To uniquely identify a signal, both the BLT function and ROC must be
specied.
Example 3.26: Find the inverse BLT x(t) of X(s) =
1
s5
with Rx = Re{s} < 5.
Solution: (Hint - use Table 8 entry #4 or #5)
Example 3.26 illustrates the most common way in which inverse LTs are found by nding a
similar LT pair in a table, and manipulating it to get the desired result. The formal approach
to manipulating the table entry is the use of LT properties. We will discuss inverse LTs
in more depth a little later. After we establish a relationship between the BLT and CTFT,
in Subsection 3.4.2, we will cover BLT and CTFT properties together in Section 3.5.
175
Example 3.27: Find the bilateral LT, including the ROC, of

x(t) = 3 e5t u(t) + e4t u(t) .
Solution:
1
3
+
Rx = 5 < Re{s} Re{s} < 4
s+5
s4
2s 17
= 2
Rx = 5 < Re{s} < 4
s + s 20
To be able to take the BLT of this x(t) you need to be able to take the BLT of
both of its additive components. This the ROC of x(t) is the intersection of the
ROCs of it individual additive components.
X(s) =
When working with the Laplace transform, we will most often be interested in signals which
are linear combinations of exponentials. As illustrated in Example 3.27, such a signal will
have a BLT function which is the sum of rational functions of s, and thus is itself a rational
function. This makes an ability to deal with rational function of s important, so we will
focus on this a little later. (Does the term partial fraction expansion ring a bell?)
Also illustrated in Example 3.27 is the fact that a CT signal which is the sum of exponentials will have an ROC which is the intersection of right and/or left sided half-planes in the
s plane. This suggests the following properties of ROCs of such signals:
a) Right-sided signal: The ROC is a right sided half plane in s.
b) Left-sided signal: The ROC is a left sided half plane in s.
c) Two-sided signal: The ROC, if it exists at all, is a vertical strip in s.
d) Finite duration signal: The ROC is the entire s plane.
splane
Im(s)
....
Im(s)
....
splane
....
....
Re(s)
Re(s)
....
....
Im(s)
splane
Im(s)
.... ....
splane
b)
.... ....
a)
....
....
....
....
Re(s)
d)
....
....
c)
Re(s)
Figure 68: ROC illustrations for dierent types of signals.

Figure 68 illustrates the ROC for the four cases listed above. It turns out that these ROC
properties are general, i.e. they apply to any signal. Thus, the ROC will always be a vertical
strip, i.e. for real valued boundaries rl and rr such that rl rr , the ROC will
be rl < Re{s} < rr .
176
Example 3.28: Let x(t) = e5|t| = e5t u(t) + e5t u(t). Find the BLT.
Solution:
The ROCs of the individual components of x(t) do not intersect. Thus there are
no values of s such that the BLT exists. That is, this x(t) does not have a BLT.
Table 8 provides a list some common BLT pairs.
Table 8: Bilateral Laplace Transform (BLT) Pairs.
#
Signal
( t)
(t)
u(t)
1
s
Re{s} > 0
u(t)
1
s
Re{s} < 0
eat u(t)
1
s+a
Re{s} > Re{a}
eat u(t)
1
s+a
Re{s} < Re{a}
tn1 at
e u(t)
(n1)!
1
(s+a)n
Re{s} > Re{a}
t
(n1)! eat u(t)
1
(s+a)n
Re{s} < Re{a}
eat cos(o t)u(t)
s+a
2
(s+a)2 +o
Re{s} > Re{a}
eat sin(o t)u(t)
0
2
(s+a)2 +o
Re{s} > Re{a}
n1
BLT
(with ROC)
1;

3.4.2
177
Relationship Between the BLT and the CTFT
Consider the CTFT and BLT equations for a signal x(t):

X(j) =
X(s) =
x(t) ejt dt
x(t) est dt
(42)
.
(43)
Comparing these, we see that

X(j) = X(s)|s=j .
(44)
This necessitates that the ROC of x(t) includes the j axis, and states that the CTFT is
(essentially) the BLT evaluated on the j axis. It should now be clear why we use the
j argument for the CTFT (i.e. why we use the X(j) notation). It is because of this
CTFT/BLT relationship.
Example 3.29: Given the following BLT pair (i.e. Entry # 9 in Table 8),
et sin(0 t) u(t)
0
;
2
(s + )2 + 0
Re{s} >
determine the CTFT of x(t) = e2t sin(10t) u(t) (which is not in Table 7).
Solution:
178
3.4.3
The Inverse Bilateral Laplace Transform
Let real-valued be in the ROC for a signal x(t), and let X(s) be the BLT function. Then
the inverse BLT is the line integral
x(t) =
1
2j
+j
j
X(s) est ds .
(45)
This equation shows x(t) expressed as a linear combination of the basic signals est for
s = + j for over the continuous range . The weighting function, applied
to these basic signals, is X(s) . So, the BLT computes the coecients of this expansion of
2
x(t). Note that dierent choices of , within the ROC, result in dierent expansions. Thus
the BLT provides a class of expansions of x(t) (one of which is the CTFT expansion if the
ROC includes the j axis). That is, if for a given x(t) the ROC includes = 0, and in the
inverse BLT equation we let s = j (i.e. = 0), we have
1
2j
1
=
2j
1
=
2
0+j
x(t) =
0j
X(s) est ds
s = j
(46)
X(j) ejt j d
(47)
X(j) ejt d .
(48)
So again we see that the CTFT can be viewed as the BLT evaluated on the j axis.
An important issue is this: given X(s) and the ROC, how can we derive x(t)? There
are several approaches. The direct one is to evaluate the inverse BLT line integral directly.
We show a general example of this directly above, for which we chose = 0, assuming the
ROC includes the j axis. You will note be responsible for this approach. A more common
approach, which you are responsible for, is to use BLT tables and properties. Example 3.26
above provides an illustration of the use of tables to determine inverse BLTs.
We will cover BLT properties, along with those of the CTFS and CTFT, in Section 3.5 of
the Course. We will also make extensive use of BLT properties in Chapter 4 of this Course
when discussing CT LTI system analysis. In the example directly below we make use of the
following BLT linearity property:
Given two signals x1 (t) & x2 (t) and corresponding BLTs X1 (s) & X2 (s) and
ROCs Rx1 & Rx2 , and given constants a1 and a2 ,
x(t) = a1 x1 (t) + a2 x2 (t)
X(s) = a1 X1 (s) + a2 X2 (s)
Rx = Rx1 Rx2 .
(49)
179
Example 3.30: Given the following BLT,

H(s) =
8s2 + 17s + 37
s3 + 4s2 + 9s + 10
Rh = Re{s} > 1
determine the inverse BLT h(t).

Solution: First note that since the ROC is a right half plane in the s-plane, h(t)
will be a right-sided signal.
Next note that H(s) is not in Table 8 of these Notes. Thus, we can not use an
entry in that table directly to determine the inverse BLT. However, it can be
shown that
H(s) =
s+1
7
+
s+2
(s + 1)2 + 4
The method for breaking this H(s) into a sum of lower order terms, called Partial
Fraction Expansion (PFE), will be covered later in the Course (in Chapter 4).
For now, note that the two terms in the above equation are entries #4 & #8 in
Table 8. Thus, using the linearity property,
h(t) = 7 e2t u(t) + et cos(2t) u(t) .

Note that the ROC of H(s) is consistent with the table entries and with the
linearity property. That is, Rh is the intersection of Re{s} > 2 and
Re{s} > 1.
Also, you can prove the decomposition of H(s) shown above is correct by recombining it (i.e. by nding the common denominator, a procedure you need to be
familiar with).
180
3.4.4
The Unilateral Laplace Transform (ULT)
The Laplace transform is a powerful tool for the design and analysis of CT LTI systems.
Although the BLT provides much of what is needed, it does not provide for the ability to
account for system initial conditions. The ULT does. Here we introduce the ULT, and in
Chapter 4 of the Course Notes we will show how it can be used to represent initial conditions.
Given a signal x(t), the ULT is dened as
X (s) =
x(t) est dt .
(50)
The lower bound on the integral, t = 0 , means that the we integrate across t = 0, so that
if there is an impulse at t = 0, it will be included in the integral.
Example 3.31: The signals x(t) = et u(t) and y(t) = e|t| have that same ULT
since x(t) = y(t); t 0 . That is,
X (s) = Y(s) = X(s) =
1
s+1
ROC = Re{s} > 1 ,
where X(s) is the BLT of x(t) (from Table 8). Note that, with the ULT, the
ROC is not needed in uniquely determining the inverse LT. Basically, with the
ULT, left sided signals are not considered.
Example 3.32: Determine the ULT of x(t) = e2t sin(10t) u(t)
Solution: Note that right-sided signal entries in a BLT tables can be used to
determine ULTs and inverse ULTs.
X (s) =
3.4.5
10
.
(s + 2)2 + 100
Some Useful Rules-of-Thumb on When to Use Dierent Transforms
1. Use the generalized CTFT instead of the CTFS when representing both energy and
periodic signals. Note, however, that for periodic signals the CTFS must still be used
to determine the weights of the impulses at the harmonic frequencies in the CTFT
representation.
2. Use the CTFT when analyzing signals and when interested in frequency response of
CT LTI systems.
3. Use the Laplace transform when dealing general system analysis issues.
4. Use the ULT when specically dealing with initial conditions. Otherwise use the BLT.
3.5
181
Continuous Time Transform Properties
We have already studied the linearity, delay, symmetry and Parsevals Theorem properties
of the CTFS. We saw that the linearity and delay properties provided, in conjunctions with
Table 6 of CTFS pairs, an ability to identify CFTS coecients, without directly computing
the coecients, for any linear combination of delayed signals represented in Table 6. Similarly, these properties and Table 6 provide a means for identifying signals given their CTFS
coecients. While introducing the CTFT and Laplace transform, we mentioned that similar
properties exist (e.g. linearity) and can be useful.
The utility of transform properties goes well beyond that of helping us derive transforms
and inverse transforms. The CTFS properties weve studied so far (e.g. delay, Parsevals
theorem) have provided us with insight into the CTFS and the signals they represent. Other
properties will provide us with additional insight. Transform properties are also useful in
solving engineering problems. In fact we will see that many of the transform properties we
consider represent important engineering applications. With this in mind, in the following
lectures we will investigate continuous time transform properties.
For several reasons we will cover the properties of the CTFS, CTFT and Laplace transform
together. First and foremost, the approach we take emphasizes the similarity between the
transforms. (Remember that the Laplace transform can be viewed as a generalization of
the CTFT, and the generalized CTFT encompasses the CTFS.) Additionally, there is an
eciency in covering properties for all three transforms together. Instead of covering sixteen
transform properties three dierent times, we will cover them all at once. So we will be able
to focus on each property a little longer, hopefully making sure we understand it.
Since the transforms are dierent (e.g. they represent dierent signal classes), any given
property may look slightly dierent for each transform. Additionally, a property that is
important within the context of one transform, may be irrelevant or unenlightening for
another. Thus for a given property we may choose to focus more on one transform than
another.
Table 9 is a list of CTFS, CTFT and Laplace transform properties. Those which are discussed in these Course Notes are indicated with an EC (Explicitly Covered). Subsequently
in these Notes, we will sometimes refer to these properties using the number designated in
Table 9. For example, Property #11 is the convolution property. Of the properties listed
in Table 9, Properties #1 to #7 are relevant on a basic level, though they may additionally
suggest applications. For example, Properties #6 & #7 (time and frequency domain scaling) are important for a basic understanding of audio pitch shifting. Properties #8 through
#11 are fundamental to basic signal processing functions which we will overview later: multiplication for spectrum estimation, radio and sampling; modulation for radio; Parsevals
theorem for spectrum estimation; and convolution for CT LTI system analysis (of course).
The last ve properties are important, but for topics that we will not focus on in this Course
(e.g. Property #16 is important for controls), so we will only mention these properties in
the Notes.
182
Table 9: Continuous Time Transform Properties.

#
Property
CTFS
CTFT
Laplace transform
Linearity
EC
EC
EC
Time shift (delay)
EC
EC
EC
Conjugation
EC
EC
EC
Symmetry
EC
EC
NA
Time reversal
EC
NC
NC
Time scale
EC
EC
EC
Frequency scale
NC
EC
NA
Multiplication
EC
EC
NA
Modulation
EC
EC
NC
10 Parsevals
EC
EC
NA
11 Convolution
NC
EC
EC
12 Dierentiation (in t)
NC
EC
EC
13 Integration (in t)
NA
EC
EC
14 Duality
NA
EC
NA
15 Dierentiation (in s)
Initial & nal
16 value theorems
NA
NA
EC
NA
NA
EC
EC Explicitly Covered in these Course Notes.

NA Not applicable - either irrelevant or very unenlightening.
NC Property exists and is interesting, but it is not covered in these Course Notes.
183
We now proceed to discuss each Property listed in Table 9.

Property #1: Linearity
Given
x(t) = c1 x1 (t) + c2 x2 (t)
(51)
for the CTFT

X(j) = c1 X1 (j) + c2 X2 (j) ;
(52)
for the Laplace transform

X(s) = c1 X1 (s) + c2 X2 (s) ;
Rx = Rx1 Rx2
(53)
and for the CTFS, assuming that x1 (t) and x2 (t) are periodic with the same period,
ax = c1 ax1 + c2 ax2
k
k
k
(54)
We studied linearity while covering the CTFS, so refer back to that discussion. We also
covered linearity, informally, when introducing the CTFT (see Examples 3.18 & 3.21) and
the Laplace transform (see Example 3.27 and Eq (49)).
x(t) = 10 e5t cos(20t + /3) u(t) .
Solution:
x(t) = 5 e5t (ej(20t+/3) + ej(20t+/3) ) u(t)
= 5 ej/3 e(5j20)t u(t) + 5 ej/3 e(5+j20)t u(t)
Thus,
5 ej/3
5 ej/3
+
(5 j20) + j
(5 + j20) + j
5 ej/3
5 ej/3
=
+
(5 + j) j20
(5 + j) + j20
10(5 + j) cos(/3) 20 sin(/3)
=
(5 + j)2 + (20)2
X(j) =
Check out entries 8 & 9 of Table 8. As noted earlier, even though Table 8 is a
Laplace transform pairs table, it is useful for CTFTs since the Laplace transform
can be viewed as a generalization of the CTFT. To use these entries for this
example, you could make use of the trigonometric identity
cos(0 t + ) = cos() cos(0 t) sin() sin(0 t) .
184
x(t) = 5 e3t u(t) 4 t e2t u(t) .
Solution:
5
4
5
4
2
3 + j
(2 + j)
3 + j
4 + 4(j) + (j)2
5(j)2 + 16(j) + 8
=
.
(j)3 + 7(j)2 + 16(j) + 12
X(j) =
Note that X(j) is a rational function of j. As noted earlier, similar rational

functions occur frequently when dealing with transforms, so we need to learn how
to deal with (i.e. identify inverse transforms for) them.
Example 3.35: Determine the inverse Laplace transform of
H(s) =
s2 + 11
s3 4s2 + s + 6
Rh = Re{s} > 3 .
Solution: This H(s) does not appear in Table 8 on Laplace transform pairs.
However, note that
H(s) =
5
5
1
+
.
s+1
s2
s3
You can prove this by recombining the 1-st order terms to form the original H(s).
Now, using linearity, we have
h(t) = et u(t) 5 e2t u(t) + 5 e3t u(t) .
The technique for breaking the rational function H(s) into a sum of lower order
rational terms is called partial fraction expansion. We will focus on this technique
later, in Chapter 4 of these Notes when considering CT LTI systems.
185
Property #2: Time shift (delay)

Given
y(t) = x(t )
for the CTFT
Y (j) = X(j) ej
(55)
;
(56)
for the Laplace transform

Y (s) = X(s) es ;
and for the CTFS
ay = ax ejk0
k
k
Ry = Rx
(57)
(58)
We studied this property for the CTFS, so refer back to that discussion. Note that for
both the CTFS and the CTFT, the eect on the frequency domain representation (i.e. the
eect on the weight applied to each complex sinusoid that constitutes the signal x(t)) is a
phase shift which is proportional to both the frequency and delay. Make sure that this makes
sense to you!
Example 3.36: Let x(t) be periodic with period T = 4, and let y(t) = x(t + 8).
Determine the CTFS coecients of y(t) in terms of those of x(t).
Solution:
ay = ax ejk(2/4)8 = ax ejk(4)
k
k
k
x
= ak
Since the CTFS coecients for y(t) are equal to those of x(t), we have that
y(t) = x(t). This is because y(t) is x(t) delayed by exactly two periods.
Example 3.37: Consider x(t) = e3t u(t 3): a) Determine the CTFT of x(t).
b) Let y(t) = 6 x(t 5). Determine the phase shift between the 0 =
3
components of x(t) and y(t).
Solution: a)
x(t) = e9 e3(t3) u(t 3)
X(j) = e9
e(9+j3)
ej3
=
.
j + 3
j + 3
b) The gain of 6 between x(t) and y(t) does not eect relative phases between
these two signals. The delay of 5 does. Since by the delay property Y (j) =
X(j) ej , in general the phase shift for frequency and delay is . For
0 = and = 5 we have phase shift 5 .
3
3
186
Property #10: Parsevals theorem (energy and power)
For energy signals:
E =
|x(t)|2 dt =
1
2
|X(j)|2 d .
(59)
For periodic power signals:

P =
1
T
<T >
|x(t)|2 dt =
k=
|ak |2 =
A2 ,
k
(60)
k=
where ak = Ak ejk .
Recall that we covered Parsevals theorem for the CTFS, so refer back to that discussion.
Parsevals theorem quanties the idea of content as a function for frequency. Note that
power and energy are not eected by the phase of the frequency description, but only the
magnitude. Thus, for example, delaying a signal does not eect its energy or power.
For the CTFT, Parsevals theorem establishes that for a signal x(t) with total energy E,
2
the function |X(j)| plotted vs. gives the distribution of total energy over frequency. Thus
2
we call this plot the energy spectrum. Similarly, for periodic power signals, |ak |2 vs. k (or
) is called the power spectrum.
2
Example 3.38: Let x(t) = t sin(100t). Determine the % of energy over the
frequency band 50 50.
Solution:
187
1
.
1.8ej120t
Determine the % power in the DC term.
Solution: First note that x(t) is periodic, so it is a power signal. Its fundamental
1
period is T = 60 , so 0 = 120. Its CTFS coecients are
0.8k
0
ax =
k
k0
k<0
(For practice, derive this expression directly, or write x(t) as a geometric series and observe the CTFS coecients by inspection.) Let PT and P0 denote,
respectively, the total and DC powers.
PT =
k=
Thus
P0
PT
|ax |2 =
k
(0.64)k = 2.777...
k=0
P0 = |ax |2 = 1 .
0
100% = 36%.
Parsevals theorem does not say that the energy (or power) of a sum of signals is equal to
the sum of the energies (or powers) of the individual signals being summed. In general, this
is not true. It is only true if the signals do not overlap either in time or in frequency. Why?
Example 3.40: Let x1 (t) = cos(10t) and x2 (t) = cos(10t + ). Determine the
power of x(t) = x1 (t) + x2 (t).
Solution: P x1 = P x2 =
1
2
since
axi =
k
1
2
k = 1
otherwise
Note that P x = P x1 + P x2 = 1 since these two signals overlap in frequency. In

fact, P x = 0 since x(t) = 0.
Example 3.41: Find the energy of x(t) =
1
.
4+t2
Solution: Consider Table 7, entry # 7. Although not directly applicable, the

2a
term in that entry is similar to x(t) (for a = 2). The main dierence
a2 + 2
is that t and domains are switched. However, considering the CTFT and
ICTFT equations, they are similar, except that the domains are switched. So it
is reasonable to postulate that
1
K e2||
(61)
4 + t2
for some constant K. If true, then using the ICTFT, we will have
K
e2|| ejt d
x(t) =
2
0
K
=
e2 ejt d +
e2 ejt d
2
0
These are easy enough integrals to solve. So try it, and show that K = provides
the required x(t). Eq (61), with K = is entry #13 of Table 7 for c = 2. We
have eectively derived that entry. Later we will consider the duality property,
Property # 14, which formalizes this idea of switching domains for entries in
Table 7 in order to derive new CTFT pairs.
188
Magnitude and phase Spectra: We have already dened, for periodic power and for
energy signals, respectively, the power spectrum |ak |2 vs. k (or ) and energy spectrum
|X(j)|2
vs. . Also of interest are the:
2
1. Magnitude spectrum Ak vs. k (or ) for periodic power signals and
|X(j)| vs. for energy signals; and
2. Phase spectrum k vs. k (or ) for periodic power signals and
X(j) vs. for energy signals.
Example 3.42: Plot the magnitude and phase spectra of x(t) = (t 5).
Solution:
Example 3.43: Given the following magnitude and phase plots, determine x(t).
Solution:
189
Property #4: Symmetry

Symmetry is exhibited in CTFS and CTFT in several ways. Of these, the most useful
symmetry characteristics are captured by the following statement: real-valued in one domain
results in complex symmetry in the other. This statement means that, for real-values x(t),
for the CTFT
X(j) = X (j) ,
(62)
and for the CTFS,
ak = a .
k
(63)
We have already considered this property for the CTFS. This statement also means that, if
for the CTFT X(j) is real-valued or if for the CTFS the ak are real-valued, then
x(t) = x (t) .
(64)
As examples, check out some of the previous examples and some of the entries in Tables 6
& 7.
Example 3.44: Given a signal x(t) with CTFT X(j) =
Solution:
4
,
4+ 2
is x(t) real valued?
190
Property #9: Modulation (frequency shift, e.g. radio and cell phones)
Modulation means to multiply by a sinusoidal signal.
For the CTFT, given a constant frequency 0 ,
y(t) = x(t) ej0 t
Y (j) = X(j( 0 )) .
(65)
Proof: Although proving any of the transform properties we are considering is

relatively straightforward, we have not and will not be emphasizing property
proofs in this Course. Nonetheless, it is useful to see at least one proof, to see
how they typically proceed. So, well look at one here.
Y (j) =
=
y(t) ejt dt =
x(t) ej0 t ejt dt
x(t) ej(0 )t dt
= X(j( 0 )) .
That is, multiplication by a complex sinusoid of frequency 0 in the time domain results
in a shift by 0 in the frequency domain. So, to shift a signals energy to a dierent band of
frequencies, we modulate it. Using the above result for complex sinusoidal modulation, plus
Eulers identities and linearity, we have that
x(t) cos(0 t)
1
{X(j( 0 )) + X(j( + 0 ))}
2
(66)
x(t) sin(0 t)
1
{X(j( 0 )) X(j( + 0 ))}
2j
(67)
(W
Example 3.45: Determine the CTFT of x(t) = 2 sinW tt/2) cos(0 t), where
2
0 > W .
Solution: We start with entry #12 of Table 7. (We will derive this entry a little
later using the multiplication property.)
1
sin2 (Wt/2)
2
CTFT
W t2
W
1/2
sin2 (Wt/2)
2
W t2
cos (0 t)
CTFT
0+W
+W
0
Figure 69: CTFT of an AM modulated sinc2 signal.
191
For the CTFS version of the modulation property, assume a periodic signal x(t) with
fundamental frequency 0 , and let m be an integer. Then
x(t) ejm0 t
ax
km .
Example 3.46: Derive the CTFS modulation properties for x(t) cos(m0 t) and
x(t) sin(m0 t). (Assume x(t) is periodic with fundamental period T = 2 .)
0
Solution:
Example 3.47: Determine the CTFS coecients of

0.36
cos(50 t)
x(t) =
1.64 1.6 cos(0 t)
Solution: Let v(t) = 1.64 0.36cos(0 t) , which is periodic with fundamental fre1.6
quency 0 since cos(0 t) (the only part of it that is a function of t) is. Note
that
0.8ej0 t
1
+
v(t) =
1 0.8ej0t
1 0.8ej0 t
=
=
j0 tk
0.8 e
k=0
0.8k ej0 tk
k=1
0.8|k| ejk0 t
k=
Dont worry about how the steps above were arrived at. The point is that v(t)
is now expressed as a CTFS expansion, where the coecients are av = 0.8|k|.
k
Now, using the modulation property of the CTFS,
1 v
1
1
1 v
ak5 +
ak+5 =
0.8|k5| +
0.8|k+5|
ax =
k
2
2
2
2
(68)
192
Property #8: Multiplication (e.g. windowing, modulation, sampling)

For the CTFT,
x(t) y(t)
1
1
X(j) Y (j) =
2
2
Example 3.48: Modulation again Solution:

sin2 (W t)
x(t) =
(t)2
Solution:
X(j) Y (j( )) d
(69)
193
Example 3.50: Let x(t) have CTFT as illustrated below. Its important feature,
for this example, is that its frequency content is bandlimited to W W .
Determine the CTFT of
xT (t) = x(t) p(t) ;
Assume that T <
p(t) =
n=
(t nT )
.
W
Solution:
X(j )
x(t)
P(j )
p(t)
(2 /T)
(1)
...
T
...
2T
3T
...
...
0
2 0
3 0
XT (j )
x (t)
T
(x(0))
...
...
T
T
2T
(x(T))
3T
...
...
A/T
0
2 0
3 0
Figure 70: Illustration of the multiplication property of the CTFT.
In Example 3.50, note that since T < W is assumed, we have that 0 > W , and there is
2
1
no overlap in Xp (j) of the shifted images of X(j). Since the impulse rate is fs = T , we
can say that the impulse rate is fast enough, relative to the highest frequency W of x(t),
to avoid overlapping of the shifted images of X(j). This has very important consequences
related to the sampling and reconstructions of CT signals.
194
Property #6, #7: Time & frequency scale
Beyond their general utility in signal analysis, these two transform properties, especially
as they apply to the CTFT and the CTFS, build and reinforce an intuitive understanding.
They state that, concerning the time and frequency domain representations of signals, if you
spread a signal out in one domain you contract it in the other. This should make sense to
you. For example, if you spread a signal out in time, it changes at a slower rate, which
means it has become more lowpass.
The time & frequency scale property is, for real-valued > 0,
x(t)
X j
(70)
If > 1, x(t) is a contraction of x(t) (since the argument t changes more quickly than t)
and thus an expansion of X(j). Conversely, if 0 < < 1, x(t) is an expansion of x(t) and
thus a contraction of X(j).
Example 3.51: Starting with the CTFT pair
x(t) =
1
sin(10t)
t
p20 ()
the CTFT pair of x(10t) is

x(10t) =
1
sin(100t)
10t
p40
10
10
1
p200 () .
10
The new signal oscillates at a 10 times faster rate, and its bandwidth has expanded by a factor of 10.
You may have noticed that these properties are not that useful for expanding the set of
signals that you can take CTFTs and inverse CTFTs. However the intuition, that expansion
in one domain means contraction in the other, is very important.
For the CTFS, the combined time & frequency scale property is simply
y(t) = x(t) =
ax ejk(0 )t
k
(71)
k=
That is
ay = ax .
k
k
The only change is in the fundamental frequency.
(72)
195
Property #5: Time reversal (i.e. fold)

Given
x(t)
X(j)
(73)
x(t)
ax
k
(74)
or
,
Let y(t) = x(t). Then,

Y (j) = X(j)
or
ay = ax
k
k
(75)
(76)
Combining the time & frequency scale and the time reversal properties, we have that, for
any nonzero real number ,
x(t)
X j
||
(77)
and, for the CTFS

x(t)
ax
sign()k
where the original fundamental frequency 0 is changed to ||0.

Example 3.52: Determine the CTFT of x(t) = e|t| , with Re{} > 0.
Solution: Note that x(t) = et u(t) + et u(t). From Table 7 and from the
time reversal property we have, respectively,
et u(t)
1
+ j
et u(t)
1
j
So,
X(j) =
1
2
1
+
= 2
.
j
+ j
+ 2
This derives Entry #7 of Table 7.
(78)
196
Property #3: Conjugation
For the CTFT,
x (t)
X (j) .
(79)
a .
k
(80)
For the CTFS,

x (t)
Example 3.53: Let x(t) be a real-valued energy signal. By the conjugation property
x (t)
X (j) .
However, by the symmetry property of real-valued signals,

X(j) = X (j)
so
X (j) = X(j)
meaning that
x (t)
X(j) ,
which is of course true since x(t) is assumed real-valued in this Example.
197
Property #12: Dierentiation in time (e.g. for LCC di. equs.)

This property is of fundamental importance in CT LTI system analysis and design, since
such systems involve dierentiation of input and output signals.
First, for the CTFT, we have that
dn
x(t)
dtn
(j)n X(j) .
Example 3.54: Starting with the CTFT pair

1 + pT (t)
2() + T sinc(T /2) ,
d
determine the CTFT of y(t) = dt x(t), rst using the dierentiation property
and second by direct evaluation of y(t).
Solutions:
First:
Y (j) = j X(j)
= j 2 () + jT
sin(T /2)
T /2
ejT /2 ejT /2
= 2j
= ejT /2 ejT /2
2j
Second:
T
T
d
x(t) = (t + ) (t )
dt
2
2
jT /2
jT /2
Y (j) = e
e
y(t) =
Example 3.55: Given the I/O LCC dierential equation

d
d
y(t) + 6 y(t) = 3
x(t) ,
dt
dt
determine Y (j) in terms of X(j).
5
Solution:
5 (j) Y (j) + 6 Y (j) = 3 (j) X(j)
Y (j) [5 (j) + 6] = 3 (j) X(j)
Y (j) =
3 (j)
X(j)
5 (j) + 6
(81)
198
Now, for the bilateral Laplace transform, the dierentiation in time property is
dn
x(t)
dtn
sn X(s) .
(82)

5
d
d
y(t) + 6 y(t) = 3
x(t) ,
dt
dt
determine Y (s) in terms of X(s).

Solution:
Comparing examples Examples 3.55 & 3.56, we see that the Example 3.55 result is a special
case of the Example 3.56 result (i.e. for s = j). So the results from these two examples are
basically the same, though the result from Example 3.56 may be a little easier to work with
since it involves a rational function of s, i.e.
3s
5s + 6
as opposed to a rational function of j, i.e.
3 (j)
5 (j) + 6
Comparing the dierentiation properties for the bilateral Laplace transform and the CTFT,
we see that the property for the CTFT is a special case (for s = j) of the property for the
Laplace transform.
199
For the unilateral Laplace transform, the dierentiation property is a little more involved
We have that
d
x(t)
s X (s) x(0 )
(83)
x(1) (t) =
dt
x(2) (t)
s2 X (s) s x(0 ) x(1) (0 )

(84)
x(3) (t)
x(n) (t)
s3 X (s) s2 x(0 ) s x(1) (0 ) x(2) (0 )
sn X (s) sn1 x(0 ) sn2 x(1) (0 ) x(n1) (0 )
(85)
(86)
where x(k) (0 ) is the k th derivative of x(t) evaluated at incrementally less than t = 0.

The dierentiation property of the Laplace transform, in this unilateral form, allows us to
deal with initial conditions of CT LTI systems.
5 y (1) (t) + 6 y(t) = 3 x(1) (t) ,
determine Y(s) in terms of X (s).
Solution:
5 (s Y(s) y(0 )) + 6 Y(s) = 3 (s X (s) x(0 ))
Y(s) [5 s + 6] 5 y(0 ) = 3 s X (s) 3 x(0 ))
Y(s) =
3s
3 x(0 ) + 5 y(0 )
X (j) +
5s + 6
5s + 6
Note that in Example 3.57 the output is composed of an input term and an initial condition
term. In general, for CT LTI systems, these two terms will result from the unilateral Laplace
transform of the LCC dierential equation.
Property #11: Integration
For the CTFT,
t
x( ) d
1
X(j) + X(0) ()
j
Example 3.58: Determine the CTFT of the step function u(t). That is, derive
entry # 3 of Table 7.
Solution: Starting with the CTFT pair (t) 1, we have that
u(t) =
1
(1) + (1) ()
j
1
=
+ () .
j
( ) d =
(87)
200
Property #13: Convolution (CT LTI systems!)
Let x(t) and h(t) be two signals (i.e. an input to and the impulse response of a DT LTI
system). Then, for the CTFT
y(t) = x(t) h(t)
Y (j) = X(j) H(j) .
(88)
A convolution in the time domain is a product in the frequency domain. Do you not like
convolution but love Fourier transforms? If so, then this property provides an alternative
to the time domain convolution calculation an alternative to the folding and shifting and
producting and integrating.
Proof of the Convolution Property of the CTFT: As mentioned earlier, you are
not responsible for proofs of properties, but it is good to see a few.
Y (j) =
=
x( ) h(t ) d
x( ) ej
ejt dt
h(t ) ej(t ) dt
Noting that
h(t ) ej(t ) dt = H(j)
(89)
since with the integral over t it is independent of , we have

Y (j) = H(j)
x( ) ej d
= H(j) X(j)
Since convolution is central to CT LTI systems, this property is central to CT LTI system
analysis and design. We will stress this in Chapter 4. Specically, Within the context of CT
LTI systems, H(j) is the CTFT of the impulse response h(t) we will see that H(j) is
an important characterization of a CT LTI system.
Considering the relationship between the Laplace transform and the CTFT, it should be
no surprised that for the bilateral Laplace transform,
y(t) = x(t) h(t)
Y (s) = X(s) H(s) .
(90)
The ROC for y(t) will be at least the intersection11 of the ROCs of x(t) and h(t). For the
CTFS,
y(t) = x(t) h(t)
ay = ax H(jk0 ) .
(91)
k
k
11
It could be larger if there is pole/zero cancellation a topic of Chapter 4.

CTFT of y(t) = x(t) h(t).
201
1
t
sin(2W t) and h(t) =
1
t
sin(W t). Determine the
Solution:
Example 3.60: Given the H(j) plotted below and the signal x(t) = A cos(0 t),
determine y(t) = x(t) h(t).
Solution:
202
Example 3.61: Using the Laplace transform, determine the CT LTI system output
when
h(t) = 4 e2t u(t) (t) ,
x(t) = sin(2t) u(t) ,
and the initial conditions are all zero.
Solution: From Table 8 and linearity, we have
2
+ 4
4
H(s) =
1
s+2
s + 6
=
s+2
X(s) =
s2
Rx : Re{s} > 0
Rh : Re{s} > 2
Then, by the convolution property,

Y (s) = X(s) H(s) =
s2
2
s 6
+ 4 s+2
Ry : Re{s} > 0
2s 12
(s2 + 4)(s + 2)
2
2
2s
=
+ 2
+ 2
.
s + 2
s + 4
s + 4
=
The last line was derived using partial fraction expansion, a topic we will cover
soon. For now note that you can prove the last line is correct by deriving the
second line from it by nding the common denominator.
Taking the inverse BLT, using Table 8 and linearity, we have
y(t) = 2 e2t u(t) + sin(2t) u(t) 2 cos(2t) u(t) .
203
Property #14: Duality

If you compare the CTFT and inverse CTFT equations,
X(j) =
x(t) ejt dt
(92)
1
X(j) ejt d
(93)
2
you notice that they are very similar, but not identical. Also, referring to Table 7, if you
compare entries #8 & #9, or entries #2 & #12, or entries #7 & #11 with itself, you will
notice a duality between domains. If you have a transform pair
x(t) =
x(t)
X(j) ,
(94)
then the CTFT of X(t) will look like x(j) and the inverse CTFT of x(j) will look like
X(t).
The duality property states that, given a CTFT pair
x1 (t)
X1 (j) ,
(95)
then we also have the pair

x2 (t) =
1
X1 (t)
2
X2 (j) = x1 () .
(96)
Note also that there is a duality of CTFT properties.
Weve seen just the right number of CT transform properties for this Course, so as the
saying goes, well quite while were ahead. Two other properties which are note covered here,
and you will not be responsible for learning in this Course:
Property #15: Dierentiation in s
Property #16: Initial and nal value theorems
On the following pages we provide Tables of the CTFT properties (Table 10) and the Laplace
transform properties (Table 11) covered in this Section of the Course.
204
Table 10: Continuous Time Fourier Transform (CTFT) Properties.

Property
Time Domain
Frequency Domain
Symmetry
real-valued x(t)
X(j) = X (j)
Delay
x(t )
Time Scale/Fold
x(at)
Linearity
a1 x1 (t) + a2 x2 (t)
a1 X1 (j) + a2 X2 (j)
Convolution
x(t) h(t)
X(j) H(j)
Dierentiation
d
x(t)
dt
jX(j)
Integration
Parsevals Theorem
Parsevals Theorem
P =
1
T
1
X(j)
j
|x(t)|2 dt
T
0
E=
|x(t)|2 dt
Multiplication
x(t) w(t)
Modulation
x(t) cos(0 t)
+ X(0)()
P =
|X(j)|2 d
|ak |2
k=
1
2
x(t) ej0 t
Modulation
1
2
X(j) ]
j
a
1
X
|a|
x() d
E=
X(j) ej = |X(j)| ej[
X(j)W (j( ))d

X(j( 0 ))
1
[X(j(
2
0 )) + X(j( + 0 ))]
205
Table 11: Bilateral Laplace Transform (BLT) Properties.

Property
Time Domain
s Domain
ROC R
Delay
x(t )
X(s) es
R = Rx
Time Scale/Fold
x(at)
Linearity
a1 x1 (t) + a2 x2 (t)
a1 X1 (s) + a2 X2 (s)
R at least Rx1 Rx2
Convolution
x(t) h(t)
X(s) H(s)
R at least Rx Rh
Dierentiation
d
x(t)
dt
s X(s)
R = Rx
Integration
Multiplication by t
x() d
t x(t)
1
X
|a|
1
s
s
a
X(s)
d
ds X(s)
if s Rx ,
s
a
R at least Rx {Re{s} > 0}

R = Rx
206
3.6
Practicum 3
Synthesis of Audio Signals

Reporting Requirements: Follow report instructions for Practicum 1.
Objective: to gain experience with and an understanding of the representation of a signal
in terms of its frequency content. Critical to this is the idea of synthesizing signals as a
linear combination (i.e. weighted sum) of sinusoids of dierent frequencies.
Background:
Sampling was discussed on an elementary level (adequate for this Practicum) during the
second lecture, so review your Section 1 Notes. Basically, the ideal A/D converter assumed
here forms a discrete time signal x[n] from an underlying continuous time signal x(t) as
x[n] = x(nTs )
(1)
where Ts is the sampling interval in seconds (so fs = 1/Ts is the sampling frequency in
samples/second).
A D/A converter operates to reconstruct x(t) by eectively interpolating between the
samples (in x[n]). As we shall see later in the course, x(t) can be exactly reconstructed only
if fs > 2fmax where fmax is the maximum frequency component of x(t).
MIDI Files: A MIDI le is an ecient representation of an audio CT signal. It contains
harmonic, harmonic level, and duration information from which the signal can be synthesized.
See, for example, www.midi.org/about-midi/tutorial/tutor.shtml for more information.
Music Score: Attached is a discussion on reading a music score, and on converting notes to
their fundamental frequencies. Review this before the practicum sessions.
Depending on the rst letter of your last name, as assigned below, you will synthesize one
of the four attached music scores.
1. If you last name begins with A,B,C,D,E,F or G, work on Fur Elise.
2. If you last name begins with H,I,J,K,L,N or M, work on Jesu, Joy of Mans Desiring.
3. If you last name begins with O,P,Q,R,S or T, work on Minuet in G.
4. If you last name begins with U,V,W,X,Y or Z, work on Beethovens Fifth Symphony.
Alternatives: As an alternative you may obtain permission from the instructor to either:
work with another score that you select and provide; or generate a sequence of voice sounds
(e.g. vowels a,e,i,o and u).
207
Procedures: This practicum follows Lab C.3 in the reference text DSP First: A Multimedia Approach, by McClellan, Schafer and Yoder, Prentice Hall, 1998.
This Lab consists of 5 procedures, the rst 3 of which are to be performed the 1-st week.
Procedures 4 and 5 should be preformed the 2-nd week. Read the Background above before
the week-1 lab. Procedure 1 should also be complete before the week-1 lab session. Procedures 2 & 3 should be completed at the week-1 lab session. Procedure 4 should be completed
before the week-2 lab session. Procedure 5 should be completed at the week-2 lab session.
1. sumcos - a siusoid linear combiner: Write an m-le that will synthesize a waveform of
the form
M
x(t) =
Ak cos(2fk t + k )
(2)
k=1
where M is the number of sinusoidal components, and Ak , wk = 2fk and k are,

respectively, the amplitude, angular frequency and phase of the k th sinusoidal component. The rst few lines of the m-le should look like:
function
xx = sumcos(f, X, fs, dur)
% SUMCOS
Function to synthesize a sum-of-cosine signal
% usage:
%
xx = the synthesized signal
%
f = vector of frequencies (in Hz)
%
X = vector of complex exponentials: (Amp*exp{j*phase})
%
fs = the sampling rate in samples per second
%
dur = time duration of the signal
%
%
Note: f and X must be the same length
%
X(1) corresponds to frequency f(1)
%
X(2) corresponds to frequency f(2), etc.
Within the m-le, you can use the command length(f) to determine the number
of sinusoidal components. You should use a for loop over the dierent frequencies.
Within this loop use the Matlab abs and angle functions to strip the magnitude
and phase o of each X(k). Do not generate complex-valued sinusoids.
Test your sumcos m-le with the following cases:
(a) xx = sumcos([20], [1], 200, 0.25);
(b) xx = sumcos([20 40], [1 1/2], 200, 0.25);
(c) xx = sumcos([20 40 60 80], [1 -1 1 -1], 200, 0.25).
Plot the results using the proper independent axis scale.
208
2. Generating, plotting and listening to tones:
(a) Compute a vector x1 of samples of a sinusoid with A = 100, 0 = 2(1100) and

= 0. Use a sampling rate of 8000 samples/second, for a duration of 2 seconds.
Using the Matlab command soundsc( ) (in Matlab, type help soundsc for
an explanation of this command), play the resulting vector through the D/A
converter of your computer. Listen to the output (you may need headphones).
(b) Compute a vector x2 as samples, over a 2 second duration, of a sinusoid with
A = 100, 0 = 2(1650) and = /3. Again, listen to the reconstructed signal.
How does it compare to the signal from 2.(a)? Concatenate the signal as follows:
xx = [x1 zeros(1, 2000) x2];
(3)
and listen to this signal. Explain what you hear.

(c) Now send the xx generated in (b) above to the D/A converter, but double the
sampling rate used in soundsc( ) to 16,000 samples/second. Observe how the
duration and pitch of the signal changed. Explain.
(d) Plot (use the plot command) x1 for 0 t 0.02 seconds. Also, plot the
spectrogram of xx using the spectrogram command. (Type help spectrogram
to determine its usage. Make sure your axes are properly labeled with time and
frequency, e.g. use spectrogram(xx,256,250,512,fs, yaxis).)
3. Generating, plotting and listening to musical notes:
(a) Write an m-le, called note.m, which is a function that generates a desired note
for a desired duration. A note will be a periodic signal consisting of harmonically
sinusoidal components. This function should have the following form:
function
tone = note(keynum,fs,harmonic nums,harmonic vals, dur)
% NOTE
Produces a sinusoidal signal corresponding to a
%
piano key number
% usage:
%
tone = output signal
%
keynum = the piano keyboard number of the desired note
%
fs = specied sampling frequency
%
harmonic nums = array containing note harmonic numbers
%
harmonic vals = array containing note harmonic amplitudes
%
dur = time duration of the signal
%
fund freq =
tone =
For the fund freq line, use the 21/12 formula that appears in the class handout.
Start with the reference note (middle C (261.6 Hz.) or A 440 is recommended),
and solve for the fundamental frequency based on this reference. For tone,
use your sumcos function to superimpose the harmonic components specied by
harmonic nums at the levels specied by harmonic vals.
209
(b) Generate a note of 2 second duration, at 11,025 samples/sec. rate, consisting of

the 1-st ve harmonics at levels [1 .5 .3 .2 .3], that represents the note E5 (key
56) above A 440 (key 49; 440 Hz.). Listen to this note and plot its spectrogram.
(c) Complete the following m-le to play scales:
function
play scale.m
%
keys = [40 42 44 45 47 49 51 52]
% Notes C D E F G A B C
% key # 40 is middle C
%
dur = 0.25 * ones(1,length(keys));
fs = 11025;
harmonic nums = 1:5;
harmonic vals = [1 .5 .3 .2 .3];
xx = zeros(1,sum(dur)*fs + 100);
n1 = 1;
for kk = 1:length(keys);
keynum = keys(kk);
tone =
n2 = n1 + length(tone) - 1;
xx(n1:n2) = xx(n1:n2) + tone;
n1 = n2+1;
end
soundsc(xx, fs)
For tone, use your note.m to generate the signal for the keynum.
(d) Plot the spectrogram of xx, and provide your note.m and play scale.m les.
210
Use a sampling frequency of fs = 11025 samples/sec. for all steps of procedure 4. and 5.
4. Determine the frequencies vs. time for the audio you are to synthesize. Record this
information in a table. (For example, if you are generting a musical score, before the
practicum session determine the keynumber, start time and time duration needed for
each note. Record these in a table.)
5. Synthesize your musical score or sequence of voice sounds:
(a) In Matlab, synthesize your musical score as a sequence of superimposed notes.
Use your sumcos or note function. Their are various approaches to this. For
example, each track (e.g. base and treble) can be generated in a separate array.
For each track, concatenate the notes using an m-le similar to play scale.m.
Then mix (sum) the separate tracks. If you choose to generate a sequence of voice
sounds, alter this procedure as needed.
(b) Using the Matlab command soundsc, play your audio signal for the instructor
or TA, so that he can verify the completion of this task.
(c) Generate the spectrogram of the resulting signal.
(d) Compare and comment on the relationships between the audio signal, the spectrogram and what you hear.
You might achieve a more pleasing sounding signal by applying a separate amplitude
to each note or vowel, and/or by changing the note amplitude over its duration (e.g.
tapering the amplitude so its intensity decreases over time). Changing harmonic content can result in sounds similar to other musical instruments or speakers. Feel free to
experiment with these and other tweaks to the basic signal you generated.
211
Practicum 3
Instructor/TA Sign O Sheet
Students Name:
1. Pract. 3, Procedure 1: plots for (a,b,c)

2. Pract. 3, Procedure 2(d): spectrogram plot
3. Pract. 3, Procedure 3(d): spectrogram plot
4. Pract. 3, Procedure 5(b,c): audio and spectrogram
212
.

Reading music and frequency mapping:
213
214
Musical scores:
3.7
215
Problems
Chapter Topics:
3.1-25 (introductory CTFS and CTFS properties);
3.26-34 (introductory CTFT and the generalized CTFT);
3.35-38 (introductory Laplace transforms and ROC);
3.39-60 (CT transform properties);
1. Consider two sinusoids x1 (t) = 8 cos(4t 0.3) and x2 (t) = cos(7t). Determine
the fundamental frequency and the period of the signal x(t) = x1 (t) + x2 (t).
2. For each of the following the signals, determine the fundamental frequency, and sketch
its power spectrum, labeling the frequencies (in radians/second) and the complex coecient of each frequency component.
(a) x(t) =
2
k=0
0.5k cos((210k)t + (/2)).
(b) x(t) = cos(260t) + 4 cos(270t) 2 cos(290t).

3. Consider the signal x(t) = cos(2t). Since x(t) is periodic with period 1, it is also periodic with period N where N is any positive integer. Determine the CTFS coecients
of x(t) under the assumption that the fundamental period is N = 3. Generalizing this,
what is a basic characteristic of the CTFS coecients computed assuming that the
fundamental period is positive integer N times what it actually is?
4. Consider a signal x(t) that consists of an innite number of impulses, one at each time
t = 0.01 n for all integer n.
a) What is the fundamental period T and fundamental frequency 0 of x(t)?
b) Plot the magnitude spectrum of x(t) over 1100 1100.
c) Plot the phase spectrum of x(t) over 1100 1100.
d) Describe the 2-nd harmonic component of x(t). The answer to this is a sinusoid.
5. Given that a signal x(t) is periodic with period T = 3, and that its exponential CTFS
coecients are
ak =
.9k k 0
0 k<0
Determine x(t).
6. Using the CTFS analysis equation, derive the expression for the CTFS coecients for
of the sawtooth wave given in the CTFS Pair Table of the Course Notes.
7. Consider the CT periodic signal
x(t) =
n=
x1 (t 2n) ;
x1 (t) = e|t| p2 (t) .
216
(a) Plot x1 (t) and x(t).
(b) For x(t), nd the DC CTFS coecient ax .
0
8. Consider the CT periodic signal
x(t) =
n=
x1 (t 0.1n) ;
x1 (t) = p.02 (t) .
(a) Plot x1 (t) and x(t).

(b) Using the CTFS analysis equation, determine the CTFS coecient ax as well as
0
the rest of the ax .
k
(c) Determine the total power in x(t).
(d) Determine the power in the DC component (this is |ax |2 ). What is the percentage
0
of the total power in the DC component?
(e) Plot the magnitude spectrum of x(t) over the range of frequency 80 80.
9. Use the CTFS table and the linearity & delay properties to determine the CTFS
coecients ay for
k
y(t) =
n=
1
1
y1 (t) = (t ) + (t + ) .
4
4
y1 (t n) ;
10. Consider the periodic continuous-time signal
y(t) =
n=
y1 (t 4n) ;
y1 (t) = e2t p4 (t) .
(a) Sketch y1 (t) and y(t).

(b) Using the analysis equations, determine the CTFS coecient ay , and the ay .
0
k
(c) Determine the total power, and the power in DC (zeroth harmonic).
(d) What percentage of the total power is in the DC component?
(e) Plot the Spectrum of y(t) over the range of frequency 4 4.
11. Consider the following periodic signal with fundamental frequency 0 = 2:
x(t) =
sin (N + 1 )0 t
2
sin
1
t
2 0
ax =
k
1
0
|k| N
otherwise
Let N = 20. Determine the percentage of total power in the frequency band
11 11.
12. Consider the signal x(t) =
n=
10 (t 0.5n).
(a) Determine its fundamental period T and fundamental frequency 0 .
217
(b) Determine its CTFS coecients, the ax .

k
(c) Sketch its power spectrum (i.e. |ax |2 vs. ) for 10 10.
x
(d) Determine the power in the frequency range 200 200.
(e) What is the % of total power in the frequency range 200 200?
13. Consider the following CT periodic signal:

x(t) =
0.75
1.25 cos(0 t)
(a) For 0 = 120, what is the fundamental frequency 0 and the period T ?
(b) What are the CTFS coecients?
(c) What is the power of the DC (i.e. 0-th harmonic) component of this signal?
(d) What fraction of the total power of the signal is in its DC component?
(e) Now let 0 = 200. Let x1 (t) be the part of x(t) which is in the frequency
range 500 500. What is the fundamental period T and fundamental
frequency 0 of x(t)? Sketch the magnitude spectrum of x1 (t). Determine the
percentage of the total power in x(t) that is in x1 (t).
14. Consider a periodic signal x(t) with period T = 0.01 and CTFS coecients
(.5)|k|
0
ak =
k = 0, 1, 2, 3, 4
otherwise
(a) Determine the total power PT .

(b) If x(t) is processed with a lter that passes without any change all input sinusoidal
components in the frequency range |f | 150 while completely attenuating sinusoidal components outside this range, determine the % of the total input power
that passes through the lter.
15. Consider the periodic CT signal x(t), with period T = 0.01, that has CTFS coecients
|k|1/2
0
ak =
k = 0, 1, 2, , 20
otherwise
a) Determine the total power PT of x(t).

b) Determine the percentage of PT in the frequency band 1100 1100.
x(t) =
n=
x1 (t 4n) ;
(a) Sketch x1 (t) and x(t) for 4 t 8.
x1 (t) = t p4 (t 2) .
(b) Sketch the magnitude spectrum for 2 2 (i.e. for 1 f 1).
218
(c) What is the power of x(t)?
(d) What percentage of the power of x(t) is NOT in the DC plus 1-st harmonic terms?
17. Consider a fullwave rectied cosine wave x(t) = |4 cos(120t)|.
(a) What is its fundamental period and frequency?
(b) Carefully plot x(t) and x(t) over two of their periods, where x(t) is the sum of
the DC and rst harmonic components of x(t).

(c) What is the total power PT of x(t).
(d) Plot the magnitude spectrum of x(t) over the frequency range 360 360.
(e) What is the percentage of x(t) power in its DC plus rst harmonic?
18. A DC source can be generated from an AC source x(t) (e.g. a 120 Volt AC wall signal)
by rst full wave rectifying the AC signal to form y(t) = |x(t)|, and then removing (via
a lter) all but the DC component. Let x(t) = 170 cos(120t).
(a) Sketch both x(t) and y(t).
(b) What are the fundamental periods of x(t) and y(t)?
(c) What are the powers of x(t) and y(t)? Are they the same?
(d) What percentage of the power of x(t) is in the DC component of y(t)? This is
the percentage of power of the original source x(t) preserved in the DC source (ie.
the DC component of y(t))?
19. A DC source can be generated from an AC source x(t) (e.g. a 120 Volt AC wall signal)
by rst half wave rectifying the AC signal to form
y(t) =
x(t) 0
x(t) < 0
x(t)
0
and then removing (via a lter) all but the DC component. Let x(t) = 170 cos(120t).
(a) Sketch both x(t) and y(t).
(b) What are the powers of x(t) and y(t)? (Px is easy. Py is simply related to Px .)
(c) Describe the power of y(t) as a function of frequency .
(d) What percentage of the power of x(t) is in the DC component of y(t)?
z(t) =
n=
z1 (t 4n) ;
z1 (t) = (2t 4) p4 (t 2) .
(a) Sketch z1 (t) and z(t).

(b) Using Table 6 of the Course Notes and linearity, for z(t) determine the CTFS
coecient a0 , and the ak .
219
(c) Determine the total power, and the power in DC (zeroth harmonic).
(d) Plot the Spectrum of z(t) over the range of frequency 2 f 2.
21. Consider the signal
x(t) =
x1 (t + 4n) ;
x1 (t) = (4 + 2t) p4 (t) .
n=
(a) Sketch x(t).

(b) Determine its power Px .
(c) Using Table 6 and CTFS properties, determine its CTFS coecients, the ax .
k
(d) Determine the percentage of power of x(t) not in its DC component.
(e) Determine the power of the signal z(t) with CTFS coecients az = ax ej 2 (k/6) .
k
k
22. Consider the half-wave rectied cosine with CTFS coecients given in the CTFS Pair
Table of the Course Notes. For A = 1 and T0 = 1, denote the signal x(t).
(a) What are the CTFS coecients ax , ax , ax , ax and ax ?
0
1
1
2
2
1
(b) Consider the signal y(t) = x(t 2 ). What are the CTFS coecients ay , ay ,
0
1
ay , ay and ay ?
1
2
2
(c) Consider z(t) = x(t) + y(t). What are the CTFS coecients az , az , az , az and
0
1
1
2
az ?
2
(d) Plot z(t); 0 t 2.
23. Consider the following CT periodic signal x(t)

x(t)
2
0.1
0.1
0.2
0.3
0.4
a) What is its fundamental period T and frequency 0 ?

b) To nd the CTFS coecients (i.e. the ax s) using Table 6 of the Course Notes
k
and properties,what properties do you need? What Table 6 entries (i.e. what
signal types) do you need?
c) Determine ax , ax and ax .
o
1
2
d) Determine the power in the DC component of x(t).
24. Consider the signal x(t) = 3 + | cos(200t) |.
a) Sketch x(t) over two periods. What is the fundamental frequency 0 and the
fundamental period T ?
b) Using the CTFS Pairs table and the linearity property, determine expressions for
the CTFS coecients, the ax s. Plot ax vs. for 1000 1000.
k
k
220
c) Determine the combined power in the DC and 1-st harmonic components of x(t).
d) Given y(t) = x(t 1), determine expressions for the CTFS coecients, the ay
k
s.
x(t) =
n=
x1 (t n6) ;
x1 (t) =
2
3
t
0
3 < t 3
otherwise
(a) Plot x1 (t) and x(t), each over the range 3 < t 9.
(b) What it the fundamental frequency 0 of x(t)?

(c) What is the total power of x(t)?
(d) Using Table 6 of the Course Notes, and CTFS properties, determine the CTFS
coecients of x(t) for all harmonics k.
(e) Determine the percentage of total power in the DC component.
(f) Determine the percentage of total power in the rst harmonic (positive & negative
frequency).
26. Using the CTFT analysis equation, determine the CTFT of x(t) = (t+3) (t3).
27. Consider the energy signal x(t) = e5t u(t 3). Sketch x(t). By direct evaluation of
the CTFT integral, determine X(j).
28. Given any constant c and any CTFT pair
x(t)
it is true that
c x(t)
X(j) ,
c X(j)
is a CTFT pair. Determine the CTFT of y(t) =
sin(t)
t
and sketch it for 2 2.
29. a) By evaluating the inverse CTFT integral for t = 0, determine x(0) given that
X(j) = 5e|2| .
b) Using the CTFT synthesis equation, determine the ICTFT of P (j) = e|10| .
30. Using the CTFT analysis equation, determine the CTFT of the signal
19
x(t) =
k=0
0.9k (t k5) .
31. Using the CTFT analysis equation, determine the CTFT of

x(t) =
k=0
Assume |c| < 1.
ck (t kT ) .
221
32. Consider a signal x(t) which is periodic with fundamental period T = 8. Over the
range 4 t 4 it is
x(t) = (t + 1) + (t 1)
4t4 .
Determine and plot its generalized CTFT.

33. Determine the generalized CTFT of the signal
z(t) = 6 cos(2t) +
1
cos(3t + /3) .
2
34. Transform Pair Problems:

(a) Find the CTFT of x1 (t) = (t + 1) (t 1).
(b) Find the CTFT of x2 (t) = e2t u(t + 1).

(c) Find the BLT of x3 (t) = cos(3t) (t).
(d) Find the ULT of x4 (t) = et u(t + 1).
(e) Find the inverse CTFT of X5 (j) = ej5 .

(f) Find the causal inverse BLT of X6 (s) =
signal.
j
s+j3
j
sj3
explicitly as a real-valued
(g) Determine the CTFT X(j) for x(t) = (1 + cos(10t)) cos(100t). What is the
fundamental frequency of this periodic signal?
(h) Determine the CTFT Y (j) of y(t) = t et u(t) et u(t).
(i) Find the energy Ez of z(t) = 7
sin(10(t4))
(t4)
ej100t .
(j) Determine the BLT W (s), with ROC Rw , for w(t) = cos(2, 009t) u(t).
35. Determine the bilateral Laplace transform and the associated ROC for the signals.
Express your X(s) as a single rational function of s.
(a) x(t) = e2t u(t) + e3t u(t).
(b) x(t) = 4 e2t u(t) 6 e3t u(t).
36. Determine the Bilateral Laplace Transform (BLT), including the Region of Convergence
(ROC), of the signal
x(t) = 6 e5|t| + 4 cos(1, 000, 000, 000t) u(t) .
37. Consider the following CT signal:
x(t) = 3 e2t u(t) + 7 e5t u(t) .
a) Determine its BLT X(s), including the ROC Rx .
222
b) Your X(s) should be a rational function of s (i.e. a numerator polynomial in

s over a denominator polynomial in s). Express your X(s) in rational function
form.
38. Find the inverse BLT of the following:
(a) Y4 (s) =
s+3
.
(s+2)2
Assume y4 (t) is causal.
(b) Determine the inverse Bilateral Laplace Transform for

H(s) =
2
5
2
+
+
,
s + (1 + j)
s + (1 j)
s7
where the ROC includes Re{s} = 0. Express your h(t) explicitly as a real-valued
signal.
(c) Determine the BLT X(s) and ROC of the signal
x(t) = 4 e3t u(t) t e5t u(t) .
Express X(s) as a rational function of s (i.e. nd the common denominator).
39. Determine the CTFT of x(t) = u(t) u(t 3).
40. Basic CTFS and CTFT problems:
(a) Consider a periodic signal x(t) with period T0 = 0.001 and CTFS coecients
ak = (0.9)k u[k].
i. Is x(t) real-valued? Why?
ii. Determine the total power of x(t).
iii. Determine the percentage of x(t) power in the frequency band
10, 000 10, 000.
41. For the following four CT convolution problems, solve one using graphical convolution,
one using the convolution table and properties, one using Laplace transforms and one
using which ever method you wish. For each, show all work.
(a) x1 (t) = e4t u(t) and h1 (t) = e4t cos(8t) u(t) .
(b) x2 (t) = (t 10) + u(t) + ej100t and h2 (t) = e5t u(t) .
(c) x3 (t) = p5 (t) and h3 (t) = p2 (t 1) .
(d) x4 (t) = ej30t and h4 (t) = e10|t| .
42. Determine the generalized CTFT the signal

x(t) =
n=
x1 (t 3n) ,
x1 (t) = p3 (t 1.5) + p1 (t 1.5) .
223
43. Your are given two periodic signals, each with period T0 = 0.01 seconds. Specically,
one is the sinusoid x1 (t) = cos(400t), and the other is the square wave
x2 (t) =
n=
x3 [t n(.01)] ;
x3 (t) = p.005 (t) .
For the signal x(t) = 2x1 (t) + 5x2 (t), determine the CTFS coecients. Determine the
power of x(t) in the frequency range 250 250.
44. Basic CT transforms & properties:
(a) Using the BLT integral equation, derive the BLT (including the ROC) of
x1 (t) = p3 (t 4).
(b) Using tables and properties, determine the CTFT of the following x2 (t)).
x (t)
2
4
2
10
10
(c) You are given that x3 (t) is a CT periodic signal with fundamental frequency
0 = 100 and CTFS coecients ak = .9k u[k]. Plot the ak s as a function of .
Using the geometric series equation, determine x3 (t).
(d) Consider the saw tooth signal x4 (t) from the CFTS table in the Course Notes,
with A = 1 and To = 1. Using the delay property, determine the CTFS coecients
of x5 (t) = x4 (t 10).
45. Consider the following signal
x(t) =
sin2 (2t)
1 + 0.25
(t)2
cos(40t) .
a) Determine and sketch the CTFT X(j).

b) Identify on your plot the power and the energy portions of X(j).
c) Concerning the energy portion only, what % of the energy is in the frequency
band 0 40.
46. Consider the CT signal x(t) =
sin(20t)
.
t
a) Determine its CTFT X(j). Plot its CTFT magnitude |X(j)|. Determine its
energy Ex .
b) Let y(t) = x(t 2). Plot |Y (j)|. Determine its energy Ey .
c) Let z(t) = x(t) ej30t . Plot |Z(j)|. Determine its energy Ez .
224
d) Let v(t) = z(t 2). Plot |V (j)|. Determine its energy Ev .

Label and show axis details on each plot.
47. The CTFT:
(a) Let
x(t) =
n=
y(t) = x(t) ej2t .
(t n)
Determine the power of y(t) in the frequency band 6 6.
(b) Determine y(t) and Y (j) given
sin(20(t 1))
sin2 (10t)
y(t) =
(t 1)
(t)2
48. Parsevals theorem problems:
(a) Given an input x(t) = e|t| to a CT LTI system with ideal lowpass lter frequency
response H(j) = p2 (), determine the percentage of input energy that makes it
to the output. You may have use for the indenite integral
1
1
d =
+ tan1 ()
2 )2
2)
(1 +
2(1 +
2
and the facts that tan1 (1) =
and tan1 (0) = 0.
(b) Consider the signal v(t) = x(t) ej10t , where x(t) = e3t cos(2t)u(t). What
percentage of x(t) energy is in v(t)?
(c) Given that 1 < c < 1, a periodic signal x(t) has CTFS coecients ax = c|k| .
k
This signal is the input to a CT LTI system with impulse response h(t) = (t5).
Determine the percentage of the input power that makes it to the output.
49. Consider a signal x(t) with exponential CTFS expansion is
x(t) =
ax ej(n/2)t
n
n=
where
ax
n
1
2
2
2 n2
n=0
n = 1, 3, 5, ...
otherwise
Find the exponential CTFS coecients for:

(a) v(t) = x(t 2)
(b) w(t) = x(t 4)
225
(c) y(t) = x(t)ej(/2)t

Simplify the coecient expressions as much as possible.
50. CT Transforms and Properties: The following problems are independent of one another.
(a) Consider the signal xa (t) =
n= (t 4). Determine its total power, and the
power in the frequency band 6 6.
1
2
(b) Consider the signal xb (t) = t sin(20t) t sin(10t). Determine the percentage of its energy in the frequency band 10 10.
(c) Let ya (t) =

i.
ii.
iii.
iv.
1
.
1+t2
Determine the energy of:
ya (t).
yb (t) = ya (t 5).
yc (t) = ya (t) ej4t .
yd (t) = 6 ya (t 5) ej4t .
51. Determine the ICTFT for

X(j) = |X(j)| ej
X(j)
|X(j)| = || p2 ()
X(j) = 3 .
52. Determine the energy of the signal x(t) which has CTFT
X(j) =
2
+ sinc() ej4 .
j + 1
Note that this will be much more easily done in one domain rather than the other.
53. Consider the signal x(t) with CTFT X(j) = X1 ( 1) + X1 ( + 1) where
X1 () = e|| p2 ().
a) Sketch X(j).
b) Determine the percentage of the total energy thats in the band 1 1.
54. Let v(t) = x(t) y(t) where x(t) = 1 + cos(3t) and Y (j) =
determine the CTFS coecients cv and cv .
0
5
55. Consider two signals, y(t) =
3
4+t2
and z(t) = ej10t y(t).
a) Determine Y (j) and Ey (the energy of y(t)).

b) Determine Z(j) and Ez .
c) What percentage of Ez is in the frequency band > 10?
d) What percentage of Ey is in the frequency band || > 0.25?
n=
.4|n| ( n),
226
56. Consider the spectrum analysis problem of computing the CTFT of a signal x(t) given
only a window of it, x(t); T t T . A standard practice is to just compute the
CTFT of
xw (t) = x(t) w(t)
where w(t) = p2T (t) is called the window. (That is, we compute the CTFT of the
part of the signal which is available.) Let x(t) = ej8t and T = 1. Determine and
sketch Xw (j) and compare this approximate spectrum to the true spectrum X(j).
57. Consider the CT signals x(t) =
sin(100t)
,
t
z1 (t) = ej200t and z2 (t) = cos(200t).
(a) Plot the CTFT of x(t) and determine the percentage of x(t) energy in the frequency band 0 .
(b) Plot the CTFT of v1 (t) = x(t) z1 (t) and determine the percentage of v1 (t)
energy in the frequency band 200 200.
(c) Plot the CTFT of v2 (t) = x(t) z2 (t) and determine the percentage of v2 (t)
energy in the frequency band 200 200.
58. Transform property problems:

(a) Find the CTFT of y1 (t) = sinc(5t) sinc(10t).
(b) Find y2 (t) = cos(4t) h(t), where H(j) = e|| .
(c) Find the input x3 (t) = h(t) y3 (t) where h(t) = (t 1) and output CTFT
3
Y3 (j) = j+2 .
59. The deterministic autocorrelation function cx (t) of a CT signal x(t) is dened as

cx (t) = x(t) x(t) .
That is, it is the convolution of signal x(t) with the folded signal x(t). Use CTFTs
and properties to determine cx (t) for x(t) = et p1 (t). Verify your result using
graphical convolution.
60. Consider the signal
v(t) =
2
1
sin(10t) +
sin2 (5t) .
t
(t)2
(a) Determine and sketch its CTFT.

(b) Determine and sketch the CTFT of x(t) = 2 v(t) cos(100t).
(c) Determine the percent energy of x(t) in the frequency band || < 50.
(d) Determine the percent energy of x(t) in the frequency band || < 100.
227
Applications of CT Transforms
1. In Section 4.1 we then explore how CT transforms can be used to characterize, analyze,
implement and design CT LTI systems.
2. In Section 4.2 we use CT transforms to facilitate descriptions of several higher level
signal processing function.

Be familiar with the various CT LTI system descriptions discussed in this Chapter,
including: the impulse response, the frequency response, the transfer function and
system poles/zeros. Understand the relationships between these system descriptions.
Know how to solve CT LTI system I/O problems using dierent system descriptions.
Understand the relationship between dierent CT LTI system descriptions and stability/causality. At least understand how the Laplace transform can be used to deal with
CT LTI system initial conditions.
Be comfortable with how the CT transforms covered in this Chapter are applied to the
advanced topics discussed in Section 3.7.
4.1
Continuous Time LTI Systems
Recall the discussion in Section 3.1 on the response of a CT LTI system to an exponential
input. We showed that for an exponential input, of the form x(t) = est , where s = + j is
a complex valued constant of arbitrary value, the output is
y(t) = H(s) est = |H(s)|et ej(t+
where
H(s) =
H(s))
(1)
h(t) est dt
(2)
and h(t) is the system impulse response. Note that the output is an exponential with the
same argument s. The system alters only the magnitude and phase of the exponential, as
governed by H(s). We now recognize H(s) as the bilateral Laplace transform of the impulse
response h(t). So, what if any is the role of the ROC of h(t)? Is the Bilateral Laplace
representation of the system H(s), termed the transfer function, a useful representation of
the system? These are questions we explore in this Section.
Additionally, note that since the system is linear and we can use the Laplace transform
to write and input x(t) as a linear combination of exponentials, i.e.
x(t) =
1
2j
+j
j
X(s) est ds
Rx
(3)
we have a way to use the Laplace transform to compute the output y(t) due to any input
x(t). Is this a useful approach to computing system outputs?
228
We can make similar observations and ask similar questions concerning the CTFT. Considering the relationship between the CTFT and the bilateral Laplace transform, Eq (1)
shows that the output of a CT LTI system due to input x(t) = ejt is
y(t) = H(j) ejt = |H(j)|ej(t+
where
H(j) =
H(j))
(4)
h(t) ejt dt ,
(5)
as long as the ROC of h(t) includes the j axis of the s-plane, which essentially means
that h(t) is an energy signal. We see that H(j), the CTFT of the impulse response h(t),
dictates how the system operates on complex sinusoids. We get a sinusoid out, of the same
frequency, with magnitude and phase altered as indicated by H(j). We thus term H(j)
the frequency response.
As with the Laplace transform, since the system is linear and we can use the CTFT to
write and input x(t) as a linear combination of complex sinusoids, i.e.
x(t) =
1
2
X(j) ejt d
(6)
we have a way to use the CTFT to compute the output y(t) due to any input x(t). Is this
useful?
The objective of Section 3.6 is to explore the application of Laplace transforms and the
CTFT to CT LTI systems, and to use the Laplace transform and CTFT to further develop our
system analysis capability. The Figure 71 illustrates the CT LTI input/output relationship
in terms of the CT transforms we are studying.
st + j
st
x(t) =
x(t) =
CT LTI system
h(t); H(s)
1
2j
1
2
x(t) =
+j
j
X( + j) e(+j)t d
X(j) ejt d
k=
ax
k
jk0 t
H(s) e st = | H(s) | e
y(t) =
y(t) =
1
2j
1
2
y(t) =
+j
j
H(s)
X( + j) H( + j) e(+j)t d
X(j) H(j) ejt d
ax H(jk0) ejk0t
k
k=
Figure 71: Transform based I/O representations of a CT LTI system.
229
Recalling that, for a CT LTI system with impulse response h(t), the output y(t) due to
input x(t) is the convolution y(t) = x(t) h(t). Recalling the convolution properties for the
CTFT and bilateral Laplace transform, we have the following transform pairs:
With the bilinear Laplace transform
BLT
y(t) = x(t) h(t)
Y (s) = X(s) H(s)
(7)
where H(s) = BLT {h(t)}, the BLT of h(t), is the system transfer function.
With the CTFT
CT F T
y(t) = x(t) h(t)
Y (j) = X(j) H(j)
(8)
where H(j) = CT F S{h(t)}, the CTFT of h(t), is the system frequency response.
With the CTFS
ay = ax H(jk0 ) ;
k
k
CT F S
y(t) = x(t) h(t)
k = 0, 1, 2,
(9)
where H(jk0 ) is the CTFT of h(t), is the system frequency response, evaluated at = k0 .
4.1.1
CT LTI System I/O Descriptions
Frequency response: H(j)

Transfer function: H(s)
Impulse response: h(t)
LCC dierential equation:
M
ak y
(k)
bk x(k) (t)
(t) =
(10)
k=0
k=0
If we take the bilateral Laplace transform of the LCC dierential equation, using the
dierentiation and linearity properties, we get
N
ak sk Y (s) =
k=0
bk sk X(s)
(11)
k=0
N
Y (s)
ak sk = X(s)
k=0
bk sk
(12)
k=0
Y (s)
=
X(s)
M
k=0
N
k=0
bk sk
.
ak sk
(13)
From the convolution property of the BLT, then, we have

H(s) =
Y (s)
=
X(s)
M
k=0
N
k=0
bk sk
.
ak sk
(14)
230
So, the transfer function of a CT LTI system is a rational function of s. Note from Eq
(10) that the transfer function H(s) can be written down, by inspection, from the I/O LCC
dierential equation the input side of the dierential equation determines the numerator of
H(s) and the output side determines the denominator. Conversely, the I/O LCC dierential
equation can be obtained by inspection from the transfer function. Similarly, taking the
CTFT of Eq (10) using the dierentiation and linearity properties, we get
Y (j)
H(j) =
=
X(j)
M
k=0
N
k=0
bk (j)k
= H(s)|s=j
ak (j)k
(15)
The frequency response of a CT LTI system is a rational function of (j).

Figure 72 depicts how any one CT LTI system description can be obtained from another.
LCC Diff. Eq.
by inspection
H(s); H(j )
h(t)
BLT; CTFT
Figure 72: Converting between dierent CT LTI system I/O descriptions.

Example 4.1: Consider a CT LTI system described by LCC dierential equation
y (3) (t) 4 y (2) (t) + y (1) (t) + 6y(t) = x(2) (t) + 11 x(t) .
Find the system impulse response assuming that the system is causal.
Solution:
The technique for decomposing H(s) into a sum of low order rational functions
is called Partial Fraction Expansion (PFE). We will learn this technique in Sub1
section 4.1.2. Recall that each s+a term of H(s) has two possible inverse bilateral
Laplace transforms. Since the system is assumed causal, of the two, the inverse is
right sided. Finally, note that, since |h(t)| dt = , this system is unstable.
231
Poles and Zeros:

Consider the representation shown below of the transfer function H(s), obtained by factoring the numerator and denominator polynomials, N(s) and D(s) respectively, into products
of 1-st order sections (one for each polynomial root):
N(s)
H(s) =
= G
D(s)
M
i=1
N
i=1
(s zi )
(s pi )
(16)
where G = bM is a gain term. The numerator polynomial roots, the zi ; i = 1, 2, M, are

aN
called the zeros of the system. The denominator polynomial roots, the pi ; i = 1, 2, , N,
are called the poles of the system.
Example 4.2: Determine the zeros and poles of the system in Example 4.1.
Solution:
poles: p1 = 1, p2 = 2, p3 = 3
zeros: z1 = j 11, z2 = j 11, (i.e. s2 + 11 = (s + j 11)(s j 11))

Example 4.3: Consider the CT LTI system with impulse response
h(t) = et cos(2t /4) u(t) t u(t) .
Find the transfer function, the I/O LCC dierential equation, and the poles/zeros.
Solution: We could break h(t) up using Eulers identity, but instead well use a
trig identity here.
1
1
h(t) = et cos(2t) u(t) + et sin(2t) u(t) t u(t)
2
2
Using the BLT table in the Course Notes, and linearity, we have
2
s+1
1
+
H(s) =
2+4
(s + 2)2 + 4
2 (s + 2)
1
s3 + 1.12s2 2s 5
2
=
.
s4 + 2s3 + 5s2
By inspection,
1
s2
1
y (4) (t) + 2 y (3) (t) + 5 y (2) (t) = x(3) (t) + 1.12 x(2) (t) 2 x(1) (t) 5 x(t) .
2
poles (roots of (s4 + 2s3 + 5s2 )): p1 = p2 = 0, p3,4 = 1 j2
1
zeros (roots of ( 2 s3 +1.12s2 2s5)): z1 = 1.89, z2,3 = 1.7372j0.8498
Considering h(t), note that, due to the t u(t) term, this system is nonstable. This term
corresponds to the 2 poles at s = 0. This suggests that the poles may somehow be related
to stability. We will explore this relationship in Subsection 4.1.3.
232
4.1.2
Partial Fraction Expansion (PFE)
Here we study the PFE of a transfer function of the form

H(s) =
bM sM + bM 1 sM 1 + + b1 s + b0
aN sN + aN 1 sM 1 + + a1 s + a0
(17)
Our objective is to decompose H(s) into a sum of lower order rational functions so that,
using linearity and Laplace transform tables, we can identify the inverse Laplace transform
h(t) = L1 {H(s)}.
Note that if we wish to take the inverse CTFT of a frequency response
H(j) =
bM (j)M + bM 1 (j)M 1 + + b1 (j) + b0

aN (j)N + aN 1 (j)M 1 + + a1 (j) + a0
(18)
we can just replace the js with ss and use the techniques described below. Several
techniques are required, depending on whether H(s) is strictly proper (i.e. M < N) or not,
and on whether or not there are repeated poles.
1. Strictly proper transfer function (M < N), with distinct poles: Let pi ; i = 1, 2, , N
be the system poles, which are assumed distinct. Then H(s) can be expressed as a
weighted sum of rst order rational terms as follows,
N
H(s) =
i=1
ri
s pi
(19)
where the ri , termed the residues, are computed as

ri = (s pi ) H(s)|s=pi
(20)
Then, for example assuming a causal h(t),

N
ri e
h(t) =
pi t
u(t) .
(21)
i=1
Regardless on whether or not H(s) is strictly proper or not, and on whether there are
repeated roots or not, we will be able to expand H(s) as a linear combination of low order
rational functions. The coecients of the expansion, which we denote as the ri s, are called
residues.
Each {pi , ri } pair corresponds to an exponential component of the inverse transform. It
may be causal or not, depending on the ROC. Note that there is no requirement that the
poles or residues be real-valued. If the ai and bi coecients of H(s) are real-valued, then
the poles and corresponding residues will either be real-valued or exist in complex conjugate
pairs. Any complex-valued poles corresponds to a complex exponential component (i.e. an
oscillating and perhaps decaying/expanding component) of the impulse response. We will
explore this in detail in an example below.
Partial fraction expansion (PFE) is not restricted to transfer functions. Any rational
function can be expressed this way. So PFE is a generally useful procedure in determining
the inverse transform (e.g. of input and output signals).
233
Example 4.4: Find the causal impulse response for

H(s) =
10s3 + 80 s2 + 200s + 154

(s + 1)(s + 2)(s + 3)(s + 4)
Solution:
r1 =
r2
r3
r4
10s3 + 80 s2 + 200s + 154

(s + 2)(s + 3)(s + 4)
10s3 + 80 s2 + 200s + 154

=
(s + 1)(s + 3)(s + 4)
= 2
= 1 .
10 + 80 200 + 154
= 4
(1)(2)(3)
80 + 320 400 + 154

= 3
(1)(1)(2)
s=1
s=2
Thus,
H(s) =
4
3
2
1
+
+
+
.
s+1
s+2
s+3
s+4
Since the system is causal,

h(t) = 4 et u(t) + 3 e2t u(t) + 2 e3t u(t) + e4t u(t) .
Note that this system is stable. From this example can you anticipate the characteristic of the poles that is required for a casual system to be stable? We will
address this issue later.
Matlab Demo: Matlab contains many functions which are useful for system analysis. We will explore a number of these later, in Subsection 3.7.1 and Practicum
5. For now, try the following:
B = [10 80 200 154];
A = [1 10 35 50 24];
Z = roots(B);
P = roots(A);
[r,p,k] = residue(B,A);
%
%
%
%
%
transfer function numerator polynomial coefficient vector

transfer function denominator polynomial coefficient vector
zeros
poles
residues, poles, k=[] since H(s) is strictly proper
234
Example 4.5 - Complex conjugate poles: Determine the impulse response h(t)
for the following transfer function. Express it explicitly as a real-valued signal.
Assume the system is causal.
1
1
=
s2 + 4s + 5
(s + 2 + j)(s + 2 j)
H(s) =
Solution: Note that the two poles, p1 = 2 j, p1 = 2 + j, are complex

conjugates. Their residues will be too.
r1 =
1
s+2j
=
s=2j
1
1
=
j
2j
2
So r2 = 1 j, and
2
1 j
j
2
+
H(s) =
s+2+j
s+2j
1
2
and, since the system is causal,

1
1
j e(2j)t u(t)
j e(2+j)t u(t)
2
2
1
{ejt ejt } u(t)
= e2t
2j
= e2t sin(t) u(t)
h(t) =
Note that
H(s) =
1
(s + 2)2 + 1
h(t) = e2t sin(t) u(t)
is entry #9 of Table 8, with 0 = 1 and a = 2. So this entry, and entry #8 of

Table 8, are useful for complex conjugate poles.
Since we know that the contribution to h(t) due to a complex conjugate pair of poles will
be real-valued, we can use the following short cut. Let p1 = Re{p1 } + jIm{p1 }, p2 = p ,
1
r1 = |r1 | ej r1 and r2 = r1 . Assume the system is causal. Then these two poles contribute
|r1 | ej
r1
e(Re{p1 }+jIm{p1 })t u(t) + |r1 | ej
|r1 | eRe{pi }t
ej(Im{p1 }t+
r1 )
r1
+ ej(Im{p1 }t+
e(Re{p1 }jIm{p1 })t u(t) =
r1 )
u(t) =
2 |r1 | eRe{pi }t cos(Im{p1 }t + r1 ) u(t)

to the impulse response. The point is, given p1 = Re{p1 } + jIm{p1 } and r1 = |r1 | ej
can just write down the impulse response contribution as in Eq (22).
(22)
r1
, we
235
2. Strictly proper transfer function (M < N), repeated poles: The following transfer
function is used to illustrate the general problem of repeated poles. It shows only a
single repeated pole, but in a multiple repeated pole case, each repeated pole is handled
as illustrated below.
H(s) =
N(s)
(s p1 )(s p2 ) (s pm1 )(s pm )
m1
=
i=1
ri
+
s pi
k=1
(23)
rm,k
(s pm )k
(24)
where = N m + 1,
ri = (s pi ) H(s)|s=pi
i = 1, 2, , m 1
(25)
and
(k)
1
(s pm ) H(s)
( k)! s(k)
rm,k =
s=pm
k = 1, 2, ,
(26)
Example 4.6: Determine the causal inverse Laplace transform of

X(s) =
5s2 13s + 10
s3 5s2 + 8s 4
(27)
Solution: p1 = 1, p2 = p3 = 2
X(s) =
r1
r2,1
r2,2
+
+
s1
s2
(s 2)2
You can show that r1 = 2.

r2,2 =
r2,1 =
5s2 13s + 10
s1
= 4
s=2
1 d 5s2 13s + 10
1 ds
s1
s=2
5s2 13s + 10
10s 13
+
=
s1
(s 1)2
=
s=2
7
4
= 3
1
1
So, since x(t) is assumed causal,

x(t) = 2 et u(t) + 3 e2t u(t) + 4t e2t u(t) .
236
3. Non strictly proper transfer function (M N): Use long division as follows
H(s) =
R(s)
N(s)
=
+ P (s)
D(s)
D(s)
(28)
where
P (s)
D(s)
(29)
N(s)
.
.
.
R(s)
order less than N
Example 4.7: Determine the causal inverse Laplace transform of

X(s) =
s3 + 2s 4
s2 + 4s 2
Solution: Since M > N we must rst perform long division until R(s) has
order less than D(s)
s + 4s + 2
s4
s3
+ 2s 4
3
2
s + 4s 2s
4s2 + 4s 4
4s2 16s + 8
20s 12
Thus,
X(s) =
20s 12
+ s 4 .
s2 + 4s 2
The causal inverse BLT of this is

x(t) = L1
20s 12
s2 + 4s 2
d
(t) 4 (t) .
dt

4.1.3
237
Stability, Causality, the j Axis and ROC
Consider, for discussion purposes, a strictly proper H(s) with distinct poles pi ; i = 1, 2, N
N
H(s) =
i=1
Each
ri
spi
r1
r2
rN
ri
=
+
++
.
s pi
s p1
s p2
s pN
(30)
contributes either
ri
ri
epi t u(t)
epi t u(t)
(31)
(32)
to the inverse Laplace transform. The ROC indicates which:

If pi is to the left of the ROC, then Eq (31)
If pi is to the right of the ROC, then Eq (32).
These ROC rules also apply to the repeated poles and non strictly proper cases. Figure 73
illustrates this, where the poles at s = 2 1 and s = 1 correspond to causal impulse
response components, and the poles at s = 1 and s = 2 correspond to noncausal components.
Im
1
0
causal poles
ROC
.....
.....
splane
1
0
1
0
1
0
1
0
11111111
00000000
1
0
11111111
00000000
1
0
11
00 x
11
00
11111111
1 00000000
1
0
11111111
00000000
111111 11111111
000000 00000000
1
0
noncausal poles
111
000
1111
0000
111111 11111111
000000 00000000
1
0
111
000
1111
0000
111111 11111111
000000 00000000
1
0
111
000
1111
0000
111111 11111111
000000 00000000
1
0
111
000
1111
0000
111111 11111111
000000 00000000
1
0
111
000
1111
0000
111111 11111111
000000 00000000
1
0
111
000
111111 00000000
000000x
1
0
x
x
111 11111111111111
000 00000000000000
1
0
11111111
1
0
111
000
1
0
11111111
00000000
1
0
Re
111
000
11111111
00000000
2
1
1
2
1
0
111
000
11111111
00000000
1
0
111
000
11111111
00000000
1
0
111
000
11111111
00000000
1
0
111
000
11111111
00000000
1
0
11111111
00000000
x
1
0
11111111
00000000
1
0
11
00
11111111
00000000
1
0
1 00000000
11111111
1
0
11111111
00000000
1 1
0 0
11111111
00000000
1 1
0 0
1 1
0 0
1
0
1
0
Figure 73: Illustration of the relationship between pole positions, ROC and causality of
impulse response components.
Example 4.8: Consider the Laplace transform
H(s) =
10s3 + 80 s2 + 200s + 154

4
3
2
1
=
+
+
+
(s + 1)(s + 2)(s + 3)(s + 4)
s+1
s+2
s+3
s+4
with ROC Rh = 2 < Re{s} < 1. Determine h(t). Is the system stable?
Based on this example can you identify a condition on the ROC that guarantees
stability, whether or not the system is causal?
Solution:
238
ROC Properties: Let h denote the maximum of the real parts of all the causal poles. Let
l denote the minimum of the real parts of all the non causal poles.
#1 In general the ROC is a vertical strip: h < Re{s} < l .
#2 The ROC contains no poles.
#3 A nite duration x(t) has Rx = s.
#4 Right sided (causal) x(t) means right sided Rx (i.e. Rx = h < Re{s} < ).
#5 Left sided x(t) means left sided Rx (i.e. Rx = < Re{s} < l ).
#6 Two sided x(t) means either Rx = h < Re{s} < l , or Rx = (the empty set) if
l h .
Stability and the j Axis: Stated for distinct pole impulse response components, but
true for repeated pole components as well.
#1 Right sided signal:
h(t) = ept u(t)
1
sp
Rh = Re{s} > Re{p}
(33)
For stability, Re{p} < 0. The ROC must include the j axis.
#2 Left sided signal:
h(t) = ept u(t)
1
sp
Rh = Re{s} < Re{p}
(34)
For stability, Re{p} > 0. The ROC must include the j axis.
#3 Generally, for stability, the ROC must include the j axis.
Causality and the ROC: h(t) must be right sided. So the ROC must be right sided.
Causality & Stability and the j Axis & ROC: The ROC must be right sided and
include the j axis. All poles must be in the left half plane, i.e.
Re{pi } < 0
i = 1, 2, , N .
(35)
239
Example 4.9: Consider the transfer function

H(s) =
2s + 1
1
1
=
+
(s 2)(s + 3)
s2
s+3
Determine all possible impulse responses, and for each comment on stability and
causality.
Solution: The gure below shows the dierent regions of the s-plane that could
be the ROC, depending on the causality/stability of the system.
Im{s}
splane
Re{s}
Figure 74: s-plane illustration for Example 4.9.

Case 1: both poles are causal the ROC is Re{s} > 2
h(t) = e2t u(t) + e3t u(t) .
The 1-st impulse response component is non stable.
Case 2: both poles are noncausal the ROC is Re{s} < 3
h(t) = e2t u(t) e3t u(t) .
The 2-nd impulse response component is non stable.
Case 3: the s = 3 pole is causal, the s = 2 pole is noncausal the ROC is
3 < Re{s} < 2
h(t) = e2t u(t) e3t u(t) .
Both impulse response components are stable.
Case 4: the s = 3 pole is noncausal, the s = 2 pole is causal the ROC does
not exist
h(t) = e2t u(t) e3t u(t) .
Both impulse response component are non stable.
240
4.1.4
Frequency Response and Pole/Zero Locations
In Subsection 4.1.1 we established that the frequency response H(j) of a CT LTI system
is the transfer function H(s) evaluated on the j axis i.e. H(j) = H(s)|s=j . Thus,
for any CT LTI system, the frequency response can be expressed in terms of its poles and
zeros as
M
M
k
N(j)
Y (j)
k=0 bk (j)
i=1 (j zi )
=
.
(36)
H(j) =
=
= G N
N
k
X(j)
D(j)
k=0 ak (j)
i=1 (j pi )
By properties of magnitudes and phases of sums and products of complex numbers, we have
that the magnitude response of a CT LTI system can be written as
|H(j)| = |G|
M
i=1
N
i=1
|j zi |
|j pi |
(37)
and the phase response as

M
(j zi )
H(j) =
k=1
k=1
(j pi ) .
(38)
These two equations show how pole and zero locations eect the magnitude and phase
response of an CT LTI. We will focus on the magnitude response below.
The magnitude response is obtained be evaluating the magnitude of the transfer function
|H(s)| on the s = j axis for . Consider a zero zi which is close to the j axis
for some 0 . For close to 0 , the numerator term |j zi | will be small, so |H(j)| will
be close to zero. Thats why the zi are called zeros. On the other hand, consider a pole pi
which is close to the j axis for some 0 . For close to 0 , the denominator term |j pi |
will be small, so |H(j)| will be large. Thats why the Pi are called poles because the plot
of |H(j)| looks like a high pole close to 0 .
In Subsection 4.2.1 and Practicum 5a we will explore this relationship between CT LTI
frequency response and pole/zero locations using Matlab CT lter design capabilities. Below
we consider using pole/zero placement to realize a notch lter.
Example 2.30: Figure 75 illustrates this magnitude response, pole/zero location
relationship for a notch lter with transfer function
H(s) =
(a)
(s j0 )(s + j0 )
(s (0.1 j0 ))(s (0.1 + j0 ))
Im
xo
(b)
H ( j )
Re
0.1
xo
Figure 75: (a) Pole/zero locations; and (b) frequency response for a CT notch lter.

4.1.5
241
The Unilateral Laplace Transform and System Initial Conditions
Consider the dierentiation property of the unilateral Laplace transform

dn
x(t) = sn X (s) { sn1 x(0 ) + sn2 x(1) (0 ) + + sx(n2) (0 ) + x(n1) (0 )} . (39)
dtn
Using this and taking the unilateral Laplace transform of the CT LTI I/O LCC dierential
equation of a causal system, for the 1-st order case,
y (1) (t) + a0 y(t) = b0 x(t) ,
(40)
s Y(s) y(0 ) + a0 Y(s) = b0 X (s)
(41)
we get
or, after moving things around (using algebra rules),
Y(s) =
C(s)
N(s)
X (s) +
.
D(s)
D(s)
(42)
The rst term on the right is due to the input and looks just like the bilateral Laplace
b0
transform output expression (i.e. H(s) = N (s) = s+a0 ). The second term on the right is due
D(s)
to initial conditions. D(s) = s + a0 is the transfer function denominator polynomial, and
C(s) = y(0 ).
For the 2-nd order case, assume x(0 ) = 0., as in the Course test. We have
y (2) (t) + a1 y (1) (t) + a0 y(t) = b1 x(1) (t) + b0 x(t) .
(43)
Taking the unilateral Laplace transform, we get

s2 Y(s) {sy(0 ) + y (1) (0 )} + a1 s Y(s) a1 y(0 ) + a0 Y(s) = b1 sX (s) + b0 X (s) (44)
or, after moving things around (using algebra rules)
Y(s) =
N(s)
C(s)
X (s) +
.
D(s)
D(s)
(45)
Again, the rst term on the right is due to the input and looks just like the bilateral Laplace
b1 s+b0
transform output expression (i.e. H(s) = N (s) = s2 +a1 s+a0 ). The second term on the right
D(s)
is due to initial conditions. D(s) = s2 + sa1 s + a0 is the transfer function denominator
polynomial, and C(s) = y(0 )(s + a1 ) + y (1) (0 ).
The extension to an N-th order transfer function is straightforward.
242
Example 4.10: Find the output of the causal CT LTI system
y (2) (t) + 3y (1) (t) + 2y(t) = x(t)
given input x(t) = 2u(t), y(0 ) = 3 and y (1) (0 ) = 5.
Solution:
Y(s) =
s2
1
3(s + 3) + (5)
X (s) +
.
+ 3s + 2
s2 + 3s + 2
The 1-st term on the right is due to the input. The 2-nd term is due to initial
conditions. X (s) = 2 . Performing PFE on each term,
s
Y(s) =
1
1
2
+
s
s+2
s+1
2
1
+
s+2
s+1
Since the system is causal,

y(t) =
u(t) + e2t u(t) 2et u(t)
2e2t u(t) + et u(t)
where again, the 1-st term is due to the input and the 2-nd term is due to initial
conditions. Combining these, we have
y(t) = u(t) + 3e2t u(t) et u(t) .

4.1.6
243
CT LTI System Block Diagrams
#1 Parallel subsystems: H(s) = H1 (s) + H2 (s) (LT linearity property)

H(j) = H1 (j) + H2 (j) (CTFT linearity property)
#2 Cascade subsystems: H(s) = H1 (s) H2 (s) (LT convolution property)
H(j) = H1 (j) H2 (j) (CTFT convolution property)
#3 Feedback subsystems:
Derivation:
Example 4.11: Consider the following block diagram depicting an interconnection

of causal CT LTI subsystems.
X(s)
Y(s)
+
H1 (s)
H2 (s)
H3 (s)
Figure 76: Block diagram for Example 4.11.
where H1 (s) =
s+1
;
s2
H2 (s) = 1; and H3 (s) =
sa
.
sb
Select the constant b so that the overall system is stable. With this b, select the
constant a so that the overall system transfer function is H(s) = 1 .
2
Solution:
s+1
H1 (s) H2 (s)
s2
=
1 + H2 (s) H3 (s)
1 + sa
sb
1
(s + 1) (s b)
=
.
2 (s 2) (s (a + b)/2)
H(s) =
There is a pole at s = 2, which is a problem since it can not be both stable and
causal. However, we can use the zero at s = b to cancel it. So, set b = 2. Now
the transfer function is
1 (s + 1)
.
H(s) =
2 s ( a + 1)
2
With a = 4, the remaining pole and zero cancel, and H(s) =
completes the design.
1
.
2
This
244
Example 4.12: Consider a causal CT LTI system with transfer function

H(s) =
2s2 + 5s + 2
.
s3 + 2s2 + s
For the following parallel three subsystem conguration, determine r1 , r2 , r3 , p1 , p2

and p3 .
r1
s p1
x(t)
r2
2
(s p )
2
y(t)
r3
s p3
Figure 77: Block diagram for Example 4.12.

Solution: This structure looks like it could be a direct parallel implementation of
the PFE of H(s), where the pi are the poles and the ri the corresponding residues
of the system. Rooting the denominator polynomial, we have that p1 = p2 = 1
and p3 = 0. With the repeated root, we have that
H(s) =
r1
r2
r3
+
+
.
2
s p1
(s p1 )
s p3
So the structure does implement the system. Performing the PFE, we get that
r1 = 0, r2 = 1 and r3 = 2. Thus the top path of this parallel structure is not
required.
Note that, in general, the PFE of a transfer function H(s) describes a parallel implementation of the system. On the other hand, by factoring H(s) into a product of lower order terms,
using the pole/zero product factorizations of the numerator/denominator polynomials, we
can generate cascaded realizations of the system.
4.2
245
Signal Processing Functions
This Course primarily concerns basic signal & systems tools, specically transforms and convolution. In the study of signal processing, what else is there? With a general understanding
of convolution and several CT transforms we now have, its a good time to step back, look
at the bigger picture, address this question, and consider a few more-advanced topics.
There are various ways to categorize signals and systems issues. A useful one follows:
Basic tools: Within the context of signals and systems, convolution and the transforms
covered in this and the next Chapter are the basic tools. Of course, there are more
basic mathematical tools we commonly use such as: algebra, linear algebra, calculus
and dierential equations. Beyond this, we get into advanced tools and functions.
More advanced tools: there are plenty of these, including: more advanced transforms
(e.g. wavelets, time-frequency); stochastic modeling (dealing with randomness, which
we overview in Chapter 7); and optimization (used, for example, in signal and system
design).
Signal processing functions: these are the more advanced thing we do, across various
applications, including: modulation, ltering, detection and classication, parameter
estimation, spectral estimation, control, and sampling/reconstruction. These functions
all apply to one dimensional signals (e.g. a single sensor output), multidimensional
signals (e.g. the output of multiple sensors), images and video.
System design: this concerns the use of basic and more advanced tools to create systems
that eectively perform the desired signal processing function.
System implementation: we mentioned this category of signals and systems topics in
the introduction (see Section 1.1). Implementation topics include: circuits, electronics,
digital hardware and software. Concerning all of these, we are interested in accuracy,
reliability, exibility, sustainability and cost.
Applications: we also mentioned this category of signals and systems topics in the introduction. Spend a few minutes reminding yourself about some of these. As engineers,
applications are the reason we are interested in signals and systems.
You can learn more about these topics in senior and graduate level courses either: here or
at another educational institution; or less formally but perhaps more directedly in industry.
Below we overview a few topics that cut across these categories.
246
4.2.1
Filtering
In a broad context, ltering implies the separation of something into components, keeping
some and discarding others. Filters are prevalent both in nature and in the systems humans
create. In the context of signal processing a lter is a system which separates a desired signal
component from a superimposed undesired component. For example, the signal might be the
output of a sensor or sensor preamp, consisting of a desired signal superimposed with noise
and interfering signals. The lter operates on an input signal to pass the desired component
of the input with amplication or unit gain while attenuating undesired components. To
lter the desired signal from the noise and interference, we must exploit dierences between
them. Although, depending on the application, there are dierent signal characteristics
that may be exploited to dierentiate desired signal from noise and interference, a very
common characteristic used is frequency content. If this is the case, then a LTI system
can be eectively employed since, as we know from previous discussions, a LTI system has
frequency selectivity as determined by its frequency response.
We have already considered many examples of DT and CT LTI system frequency response.
This Subsection provides a brief introduction to the practical implementation and design of
CT frequency selective lters. We will consider DT frequency selective lters in Chapter 4 of
this Course. Frequency selective lter design and implementation is a principle topic several
senior and graduates level courses (e.g. on microwave systems, digital signal processing,
optimum and adaptive ltering).
From discussions earlier in this Chapter, we know that a CT LTI system eects any
frequency component of an input signal x(t) as indicated by the systems frequency response,
denoted H(j). That is, for any input signal
x(t) =
1
2
X(j) ejt dt ,
(46)
Y (j) ejt dt ,
(47)
the CT LTI system output is

y(t) =
1
2
where Y (j) = X(j) H(j). An important consequence of this is that we can consider
how a LTI system operates on each frequency component separately. Additionally, a LTI
system does not output any frequency component that is not already in the input.
Considering the magnitudes and phases of the dierent signal components, we have that
the output frequency components are weighted by the output CTFT
Y (j) = |Y (j)| ej
Y (j)
= |X(j)| |H(j)| ej(
X(j)+ H(j))
(48)
This shows how the magnitude and phase of the input are eected by the CT LTI system,
and suggests that concerning frequency selective ltering we are in general interested in:
the magnitude response - |H(j)| vs. , or equivalently
the magnitude-squared response, in dB - 10 log10 |H(j)|2 vs. ; and
the phase response - H(j) vs. , or equivalently
247
d
the group delay - () = d H(j).
Although in many applications the systems phase response H(j) and group delay ()
are important (e.g. in high delity audio phase distortion is undesirable), in this introductory
discussion we will focus on magnitude |H(j)| and magnitude-squared |H(j)|2 responses
since these functions indicate frequency selectivity (i.e. they tell us which frequencies are
amplied and which are attenuated). The magnitude-squared response is often plotted in dB,
i.e. 10 log10 |H(j)|2, because it displays details of the response over a wide dynamic range.
We begin with a sequence of examples which consider lters which have ideal frequency
characteristics. This will motivate a subsequent discussion on ler specications and design.
Example 4.13: Consider an ideal lowpass lter with frequency response
Hlp (j) =
1
0
|| W
|| > W
where W is called the cuto frequency. Determine the impulse response hlp (t),
the magnitude response |Hlp (j)|, the phase response Hlp () and the group
delay lp (). Given these results, comment on the practicality of this lter.
Solution: From the CTFT table developed in Section 3.3, we have
hlp (t) = ICT F T {Hlp(j)} =
sin(W t)
t
By inspection, |Hlp (j)| = Hlp (j), Hlp (j) = 0, and lp () = 0.
This system has ideal frequency response characteristics since it passes low frequencies without magnitude or phase distortion while perfectly attenuating high
frequencies. However the system is not realizable because it is noncausal, with
an impulse response that suggests that the current output is a function of the
input innitely far into the future. Furthermore, although it may not be obvious
given the foundation we have established in this Course, this system does not
correspond to a LCC dierential equation of nite order (since the frequency
response is not a rational function of (j)). Thus, even if causal, it would not
be implementable as, for example, an RLC circuit with a nite number of components.
248
Example 4.14: Consider the following lter with delay added in an attempt to
deal with the causality problem uncovered in Example 4.13:
h(t) = hlp (t 1 ) .
Determine its magnitude and phase responses and its group delay. Comment on
its realizability.
Solution: using the delay property of the CTFT, we have
H(j) = Hlp (j) ej1 .
(49)
The delay introduces a phase term only, so does not eect the magnitude response. Thus,
|H(j)| = Hlp (j);
H(j) =
Hlp (j) 1 = 1 ,
and
() =
d
d
H(j) =
d
(1 ) = 1 .
d
Note that this system is still noncausal. So it is not realizable, although for large
positive 1 most of the energy of the impulse response is in the causal range,
which is perhaps an improvement. H(j) = 1 , which again implies no phase
distortion. (Constant group delay implies pure delay, which for example does
not eect our perception of an acoustic signal.) This example provides some
experience with the consideration of group delay ().
Example 4.15: Consider an ideal highpass lter Hhp (j) with zero phase response
and cuto frequency W . Sketch this frequency response. Sketch the block diagram of an implementation for this lter as a parallel of a ideal lowpass lter
and a short circuit. Considering this implementation, and linearity, determine
the impulse response hhp (t). Comment on its realizability.
Solution:

Example 4.16: Consider an ideal bandpass lter Hbp (j) with zero phase response
and passband W1 || W 2. Sketch this frequency response. Sketch the block
diagram of an implementation for this lter as a parallel of two ideal lowpass
lters. Considering this implementation, and linearity, determine the impulse
response hbp (t). Comment on its realizability.
Solution:
Example 4.17: Consider an ideal bandstop lter Hbs (j) with zero phase response
and stopband W1 || W 2. Sketch this frequency response. Sketch the block
diagram of an implementation for this lter as a parallel of two ideal lowpass
lters and a short circuit. Considering this implementation, and linearity, determine the impulse response hbs (t). Comment on its realizability.
Solution:
249
250
Example 4.13-17 demonstrate that ideal frequency selective lters are not realizable. This
fact has motivated the development of a considerable volume of results on frequency selective
CT lter design. Basically, the frequency selective design procedure begins with a frequency
response specication and results with a lter transfer function and then a realization description. We now illustrate a frequency response specication and provide a Matlab transfer
function design example. The topics of lter design and implementation are covered in senior
and graduate level courses.
Since ideal frequency selective lters are not realizable, we can realize only an approximation of a desired ideal response. In CT lter design, we start with a set of specications that
reect the degree of accuracy which we require of this approximation. Figure 78 illustrates
CT lowpass lter design specications. Only positive frequencies are shown since we are targeting a real-valued coecient transfer function, so that the frequency response is complex
symmetric. p and s are, respectively, the passband and stopband cuto frequencies. p and
s are, respectively, the passband and stopband tolerances. The frequency band p s
is called the transition band. Since no desired phase response characteristics are provided,
the implication is that in this example we are not concerned with phase distortion.
Highpass, bandpass and bandstop lters frequency response specications are similarly
described.
| H( j ) |
1 + p
1 p
stopband
passband
......
transition band
Figure 78: Typical CT lowpass lter design specications.
Matlab Demo: Matlab provides a wide variety of lter design functions, including a number
for frequency selective lters. In this demo, a CT lowpass Butterworth is designed and
evaluated. This lter will have a lowpass 3 dB cut-o frequency of = 1 rad/sec.. The
Matlab function freqs computes frequency response and can automatically plot its magnitude
and phase. In Matlab, type help freqs for more details and options for its use.

echo on
%************************************************************
%
%
Continuous-Time Filter Design & Evaluation
%
%
Butterworth lowpass: order N = 6,
%
cutoff wn = 1 radian
%
%************************************************************
%
%
Design Butterworth Filter w/ butter.m
%
(try cheby1, cheby2 and ellip)
%
%
B = vector of transfer function numerator coefficients
%
A = vector of transfer function denominator coefficients
N=6;
wn=1;
[B,A]=butter(N,wn,s)
pause
%
Plot Pole Locations using zplane
%
(no zeros for analog Butterworth lowpass)
%
p=roots(A);
zplane([],p)
pause
%
Partial Fraction Expansion of H(s) - You can determine
%
h(t) from results. P is the vector of poles.
%
R is the vector of respective residues.
%
K is the vector of long-devision coefficients if the transfer
%
function is not strictly proper.
%
[R,P,K]=residue(B,A)
pause
%
Compute and Display the Frequency Response using freqs
%
freqs(B,A)
251
252
1
0.8
0.6
Imaginary Part
0.4
0.2
0
0.2
0.4
0.6
0.8
1
1
0.5
0
Real Part
0.5
Figure 79: Pole plot (there are no zeros) for a Butterworth CT lowpass lter.
10
Magnitude
10
10
10
10
10
Frequency (rad/s)
10
Phase (degrees)
200
100
100
200
1
10
10
Frequency (rad/s)
10
Figure 80: Frequency response (magnitude-squared in db) and phase response for a Butterworth CT lowpass lter.

4.2.2
253
Sampling
This Subsection of the Course addresses sampling of a continuous-time signal to form a corresponding discrete-time signal, and the inverse operation of reconstruction on the underlying
continuous-time signals from its samples. We rst consider issues associated with sampling
sinusoids. We then take a more formal approach to the study of sampling, employing the
CTFT to derive the sampling theorem.
Preliminary View: Sampling and Aliasing of Sinusoids
Sampling Sinusoids - A Temporal View
The ideal Continuous-time to Discrete-time (C-to-D) converter, illustrated in Figure 81,
was introduced Chapter 1. As used here, the term C-to-D converter implies an idealization
of a realistic A/D (analog/digital) converter in that the output is an ideal sampling on the
input, i.e. x[n] = xc (nT ). So amplitude quantization, timing jitter and other practical
issues are not considered12
x c (t)
Ideal CtoD
converter
x[n] = xc (nT s )
T s ; fs = 1/Ts
Figure 81: An idealistic C-to-D converter.
Given a sinusoidal input
xc (t) = A cos(t + )
(50)
where is the continuous frequency in radians/second, the C-to-D converter output is

x[n] = xc (nTs ) = A cos((nTs ) + ) = A cos(n + )
(51)
where = Ts is the discrete frequency, often referred to as the normalized radian frequency. Its units are:
radians/second seconds/sample = radians/sample .
(52)
Recall that in Subsection 1.2.4 of these Notes we established the fact that DT frequency
is ambiguous that is, any two sinusoids with frequencies diering by an integer multiple of
2 will appear exactly the same point-by-point in time. This ambiguity issue turns out to
be central to our understanding sampling. It claries the problem that must be overcome to
implement eective sampling. We now illustrate this problem with an example.
12
Practical issues should be covered to some extent in senior level DSP courses. Basically, well up into the
GHz sampling rate range, one can get closer and closer to an ideal C-to-D converter by putting more and
more resources into an A/D converter.
254
Example 4.18: Consider the following two CT sinusoids:
xc (t) = cos(t) ,
sampled at a rate fs =
1
Ts
yc (t) = cos(3t)
= 1 to form two DT signals
x[n] = cos(n) ,
y[n] = cos(3n) .
From our discussion in Subsection 1.2.3 on the Course on DT sinusoids, we know

that these two sinusoids are ambiguous (i.e. they are the same sequence). This
is illustrated below.
Illustration:
Example 4.18 demonstrates the fundamental problem with sampling: that two distinct CT
signals can result in the same DT signal. For sinusoids specically, since for any integer k
cos(n) = cos(( + k2)n) ,
(53)
for a sampling interval Ts , the set of all continuous sinusoidal frequencies

{ + k2 ; all integer k} are ambiguous after sampling.
Ts
Recall from Subsection 1.2.4 of the Course that the range of unambiguous DT sinusoidal
frequencies is
< .
(54)
For sampling rate Ts this corresponds to the range
<
(rad./sec.)
Ts
Ts
or
fs
fs
< f
(Hertz)
2
2
(55)
of unambiguous CT sinusoidal frequencies. The sampled signal due to any sinusoid of frequency outside this range will look exactly like the sampled signal of a sinusoid of frequency
within this range. So, given a DT sinusoid of frequency < , it could correspond to
a CT sinusoid with any frequency
( + k2)
(56)
Ts
for any integer k.
255
In practice, given a DT sinusoid x[n] = A cos(n + ), it is standard to assume that it
corresponds to a CT sinusoid with the lowest possible frequency magnitude = Ts , i.e. the
k = 0 frequency. A standard D-to-C or A/D converter, makes this assumption.
Ideal Reconstruction (i.e. ideal C-to-D Conversion): an ideal D-to-C converter is depicted
in Figure 82.
x[n] = xc (nT s )
x c (t)
Ideal DtoC
converter
T s ; fs = 1/Ts
Figure 82: An idealistic D-to-C converter.

Later in this Subsection, when discussing the sampling theorem, we will explore how C-to-D
and D-to-C converters function ideally. For now, assume that for each sinusoidal component
of x[n] a sinusoidal component is generated in the reconstructed xc (t) according to the DT
to CT frequency mapping described above.
Aliasing: The term aliasing is the formal name given to the result of this assumption that,
for a given and Ts , the is assumed to be the lowest possible frequency magnitude Ts (i.e.
the k = 0 frequency). Basically, in terms of sinusoids, the idea is as follows:
Aliasing is the mapping of DT sinusoidal frequencies that are outside the
range { < } to the corresponding frequencies inside the range
{ < } (i.e. at a distance k2).
In terms of underlying CT sinusoidal frequency, aliasing the mapping of that
are outside the range { Ts < Ts } to the corresponding frequencies
inside the range { Ts < Ts }.
Antialiasing Filter: To eliminate the frequency ambiguity that leads to aliasing, it is common
practice to restrict the range of frequencies of a CT signal before sampling it. For a sample
1
rate of fs = Ts , the frequency range of x(t) should be restricted to
<
(rad./sec.);
Ts
Ts
fs
fs
< f
(Hz.) .
2
2
As illustrated in Figure 83, an antialiasing lter implements this bandlimiting.

x (t)
Antialiasing
Filter
passes only
fmax < f < f max
x c (t)
Ideal CtoD
converter
x[n] = xc (nTs )
Ts ; fs = 1/Ts
fs > 2 fmax
Figure 83: An antialiasing lter as a preprocessor before a C-to-D converter.
(57)
256
Sampling Sinusoids - A Spectral View
We just established the fact that any DT sinusoid with frequency outside the range
< is the same, sample by sample, as a DT sinusoid in this range, and that
convention dictates that we then consider the frequency of any DT sinusoid to be in this
range. We illustrate this in Figure 84(a) using a spectral plot of a DT frequency component
at = + .4. . We say that the frequency = + 0.4 is aliased to frequency + 0.4
which is in the range < . For this illustration, if the DT signal is the result of
1
sampling a CT complex sinusoid at the sampling rate fs = Ts , then the aliasing in terms of
CT frequency is illustrated in Figure 84(b). This shows that a CT complex-valued sinusoid
1
of frequency = +.4 , sampled at a rate fs = Ts , will be aliased to frequency = +.4
Ts
Ts
when it is reconstructed. For multiple complex or real-valued sinusoids, the gures above
illustrate what happens to each sinusoidal component.
aliased
(a)
+.4
+.4
aliased
(b)
+.4
Ts Ts
+.4
Ts Ts
Figure 84: A spectral illustration of aliasing.

Example 4.19: Consider the DT signal
x[n] = cos(0.5n) ,
which was ideally sampled from a CT signal as x[n] = xc (nTs ). If aliasing is
possible, determine three possible xc (t) consistent with this x[n].
Solution:
257
Example 4.20: Let

xc (t) = cos(2t) + cos(12t) ,
and
x[n] = xc (nTs ) ;
Ts = 0.125 ;
s =
2
= 16 .
Ts
What is the reconstructed signal? Making no assumption on whether or not

aliasing has occurred, discuss what the original signal xc (t) could have been.
Solution Plot:
Given just x[n] and Ts , we dont know if the cos(0.5n) component of x[n] came
from a CT component cos(12t) or perhaps cos(4t). For that matter, we dont
know that the DT sinusoidal component cos(0.25n) came from the CT sinusoid
cos(2t). It could have come from, for example, cos(18t).
However, if we assume that there is no aliasing (i.e. if we assume that xc (t) was
properly bandlimited for the sample rate fs = 8 before sampling), then we would
conclude with condence that the xc (t) that corresponds to the x[n] given above
is
xc (t) = cos(2t) + cos(4t) .
We would be wrong because we assumed that there is no aliasing whereas there
actually is.
258
A Formal Treatment of Sampling
The objective of this discussion is to formally identify what specication on sampling,

if any, result in the ability to exactly reconstruct a CT signal from its samples. As an
introductory example, we discussed sampling and reconstruction back in Subsection 1.1.3.
At the time it seemed counterintuitive that we would be able to discard all information
about a CT signal between its sample values an then be able to recover this thrown away
information. Here will see that under certain conditions exact reconstruction is theoretically
possible! Even though it may seem that this is, at least in part, a DT signal issue, we can
eectively address it here using only the CTFT. Specically, all we need is the multiplication
and convolution properties of the CTFT. The result derived in this Subsection is extremely
important since DT processing of CT signals has become so prevalent.
Consider a signal x(t) and corresponding CTFT X(j) illustrated in Figure 85(a). The
X(j )
x(t)
(a)
PT (j )
p(t)
(2 /T)
(1)
...
...
T
0 T
3T
(b)
...
2T
2
T
4
T
A/T
...
(x(T))
T
2
T
X (j )
x (t)
T
(x(0))
...
...
0 T
...
...
W
2
T
(c)
X (j )
x (t)
T
(x(0))
A/T
...
...
T
(d)
2T
(x(T))
...
...
W
W 2
T
4
T
Figure 85: A visualization of sampling, as modeled using an impulse train multiplication

and studied using the CTFT.
shape of x(t) is unimportant, except that it is bandlimited in frequency to the range W
W as shown. Also consider the periodic the impulse train
p(t) =
n=
(t nT )
2
PT (j) =
T
l=
( l
2
) ,
T
(58)
which is periodic with fundamental period T sec. and fundamental frequency 0 = 2 . Figure
T
85(b) depicts this signal and it CTFT. Let xT (t) = x(t) p(t). (You may recall that we
259
considered this set of signals before, in Example 3.50 of Section 3.5, when considering the
multiplication property of the CTFT.). Figure 85(c) illustrates xT (t). A key point here is
that xT (t) represents sampling x(t) with a sample duration T since xT (t) contains exactly
the same information as the samples of x(t) no more and no less. Multiplication of a CT
signal by an impulse train is a model of sampling. From this point, we will refer to the
impulses in xT (t) as the samples of x(t).
From the multiplication property of the CTFT,
1
X(j) PT (j)
2
2
2
1
X(j)
( l )
=
T
2
T
l=
XT (j) =
1
T
1
T
l=
l=
X(j) ( l
X(j( l
(59)
(60)
2
)
T
(61)
2
)) .
T
(62)
This is a key result. The CTFT XT (j) of our model of sampling is a periodic extension of
the CTFT of x(t) periodically extended every integer multiple of 0 = 2 .
T
Figure 85(c) depicts this spectrum for the case where T > W . Note that the higher
frequency images of X(j() (i.e. in the periodic extension XT (j)) do not overlap with
X(j(). This is because T > W which is the same as the condition that
fs > 2fmax
(63)
1
W
where fs = T is the sampling frequency and fmax = 2 is the highest frequency component
of x(t) (in Hz.). That is, weve sampled at a rate greater than twice the highest frequency
of the signal.
To explore the possibility of exactly reconstructing x(t) from xT (t) (i.e. from its samples)
consider processing xT (t) as shown in Figure 86.
Hlp (j )
xT (t)
Hlp (j )
y(t)
T
Figure 86: Reconstruction of x(t) from xT (t) (i.e. from its samples) .
The reconstruction lter, Hlp (j), is an ideal lowpass lter that perfectly passes X(j)
with gain T while completely attenuating the higher frequency images of X(j(). Thus,
Y (j) = XT (j) Hlp (j) = X(j) .
(64)
260
That is, y(t) = x(t). We have perfectly reconstructed x(t) from its samples. The requirements are fs > 2fmax (which implies that x(t) is bandlimited), and the use of an ideal
lowpass reconstruction lter.
Figure 85(d) illustrates what happens when fmax > f2s . The higher frequency images of
X(j) in the periodic extension XT (j) overlap with X(j). So X(j) can not be separated
from these other images, and thus x(t) can not be reconstructed from its samples. If this
is the case, and we attempt to reconstruct x(t) as illustrated in Figure 86, we will not
produce the original x(t). The frequency content we will get is as illustrated by XT (j) in
Figure 85(d). After lowpass ltering, we will have the part of XT (j) over T T .
We observe that in this range that some of the higher frequencies in X(j) are mapped to
lower frequencies. For example, if x(t) and y(t) are audio signals, some of the higher pitch
components in x(t) will sound like lower pitch components in y(t). As with the sampling of
sinusoids, we call this eect aliasing. Aliasing is bad. Very, very bad!
The Sampling Theorem: So, to reconstruct a signal from its samples, one must
sample the signal at a sample rate fs which is at least twice the highest frequency fmax of
the signal. fs = 2 fmax is termed the Nyquist rate. If a signal is sampled at at least its
Nyquist rate, we have shown that it can be reconstructed by ideal lowpass ltering. In other
words,
x(t) = xT (t) hlp (t)
(65)
where
hlp (t) = T
sin( T t)
t
(66)
is the ICTFT of Hlp (j). So,

x(t) = xT (t) hlp (t) =
=
=
n=
n=
x(nT )
x(nT )
n=
x(nT ) (t nT )
(t nT ) T
sin( T (t nT ))
.
(t nT )
T
sin( T t)
t
sin( T t)
T
t
(67)
(68)
(69)
Eq (69) shows that x(t) can be reconstructed from its samples {x(nT ); n = 0, 1, 2, }
as an interpolation between the samples using the interpolation function hlp (t). This is called
the sampling theorem.
261
2
(50t)
Example 4.21: Let xc (t) = A sin(t)2 . Determine the CTFT of the re50
constructed signal for sampling rates of fs = 100 samples/second and fs =
75 samples/second.
Solution:
262
4.2.3
Modulation
We begin this Subsection with a discussion on Amplitude Modulation (AM), which explores
on a basic level how AM radio is implemented. This discussion motivates a set of basic signal
processing tasks which are commonly found in many applications. These tasks (moving
signals around in frequency, selecting certain frequency components, and quantifying the
selected signal) correspond to the modulation, ltering and Parsevals theorem properties
of the CTFT. We close this Subsection with several examples which employ these three
properties.
Amplitude Modulation
The modulation property of the CTFT tells us that if multiply a signal by a sinusoid of
frequency 0 , then we shift its frequency content by 0 . Mathematically, we have already
established that
y(t) = x(t) ej0 t
Y (j) = X(j( 0 )) ,
(70)
x(t) cos(0 t)
1
{X(j( 0 )) + X(j( + 0 ))} ,
2
(71)
x(t) sin(0 t)
1
{X(j( 0 )) X(j( + 0 ))} .
2j
(72)
and
In applications such as broadcast radio (e.g. AM and FM), cell phones, RADAR, SONAR,
astrophysical exploration and ultrasound imaging, to name just a few, we wish to move the
frequency content of a signals of interest to dierent frequencies. So some form of modulation
is a central function in many applications. Beyond this, modulation is very commonly used
in signal processing systems as part of a strategy to reduce implementational complexity,
thereby reducing cost and/or improving performance.If you look into the detailed description
of a spectrum analyzer, speech processor or audio system, for example, you may very well
observe that one or more modulators are employed. So if you enroll in an advanced signal
processing or signal processing application course, you will likely see a lot of modulation. In
this Subsection we briey overview one important application, Amplitude Modulation (AM)
radio.
Synchronous AM Communications
Consider the simple modulation scheme shown in Figure 87(a). In the context of communications (e.g. radio), x(t) is called the information signal which is bandlimited to
m m (e.g. for speech, m = 24000 is assumed). cos(c t) is termed the carrier
where c is the carrier frequency. The resulting signal y(t), the modulated or transmission
signal, would be amplied and transmitted over a communication channel. We call this
amplitude modulation because the information signal is transmitted as the amplitude of the
transmitted sinusoidal carrier.
Figure 87(b) illustrates the information and modulated signal frequency content. Since the
carrier is a real-valued sinusoid, the information signal frequency content is translated both
up and down in frequency by c . (Note that, as illustrated, X(j) is complex symmetric
since we will assume x(t) is real-valued. Then y(t) is also real-valued, and Y (j) is complex
263
y(t) = x(t) cos( c t)
x(t)
y(t)
w(t)
Hlp (j )
x1 (t)
cos( c t + c )
cos( c t)
Hlp (j )
2
(a)
(c)
X(j )
Y(j )
A
A/2
cm c
(b)
W(j )
A
2
2 c
cos(
2 c
(d)
Figure 87: AM modulation.

symmetric.) The motivation of this modulation scheme is that it is necessary (either by
allocation and/or performance considerations) for the information signal to be transmitted
over the frequency range c m || c + m .
Now consider the demodulation scheme shown in Figure 87(b). It consists of another
modulator (to translate the information signal in frequency back to its original location) and
a lowpass lter whose purpose is identied below. The received signal is assumed to be y(t).
Thus we are ignoring possible channel distortion and noise. (Although in practice channel
distortion and noise may require a more sophisticated receiver, the demodulation principle
depicted in this scheme still represents the primary receiver function.) The output of the
receiver modulator is
w(t) = x(t) cos(c t) cos(c t + c )
1
1
= x(t)
cos(c ) + x(t)
cos(2c + c )
2
2
(73)
.
(74)
The phase c represents the fact that the receiver local oscillator in general may not be
phase synchronous with the received carrier. Again as dictated by the modulation property
of the CTFT, the frequency content of w(t) will be as illustrated in Fig 87(d). It consists of
the information signals original spectrum, scaled by 1 cos(c ), plus a term modulated up to
s
264
frequency 2c . After lowpass ltering to remove the 2c component of w(t), we have

x1 (t) = x(t) cos(c ) .
(75)
For c = 0, i.e. for the receiver oscillator synchronized with the transmitter oscillator,
x1 (t) = x(t). Regardless of synchronization, the output is a scaled version of the input, so
that in the absence of channel distortion and noise, we get exactly what we want.
Asynchronous AM Communications
The problem with the AM receiver depicted in Figure 87(b) is that if c = 0, x1 (t)
might be a signicant attenuation of x(t). In fact, for c = , for example, the information
2
signal is completely absent in w(t) and x1 (t). In the presence of channel noise, cos(c ) 0
would render the received signal useless. There are receiver circuits, such as phased-locked
loops, that can be used to assure that c 0. However for commercial AM radio these
circuits are considered to be too expensive to implement in a radio receiver. Thus, in
practice, an alternative approach to AM radio reception is employed. Below, very briey
describe the idea behind the practical receiver commonly used for AM broadcast reception.
Although there is not much to this receiver in terms of demonstration of transform theory,
the value of a transform domain analysis of AM modulation/demodulation has nonetheless
been established.
As an alternative to the synchronous AM scheme discussed earlier, consider the transmitter
shown below in Figure 88(a). The multiplier value A is selected such that A > |x(t)| for
all t for any expected x(t). The transmitted signal y(t) is the carrier with an amplitude
(termed the envelope) which has the shape of the information signal. An example y(t) is
shown in Figure 88(b). Its spectrum is illustrated in Figure 88(c). Figure 88(d) shows an
inexpensive circuit that is commonly used to demodulate the received signal, extracting the
envelope from the carrier, and the resulting signal x1 (t) is illustrated in Figure 88(b). Besides
a DC component that can be subsequently removed, and some undesirable fast changes
(high frequency components) that can be smoothed out (lowpass ltered), this output is an
acceptable received signal for (low delity) AM radio.
265
y(t) = (A+x(t)) cos( ct)
x(t)
+
A
cos( c t)
(a)
A+x(t)
x1 (t)
y(t)
(b)
Y(j )
X(j )
A
A/2
cm c
(c)
x1 (t)
y(t)
(d)
Figure 88: A practical asynchronous AM transmitter and receiver.
Modulation, Filtering and Parsevals Theorem Examples
Example 4.22: Consider the system below.
a) Sketch the CTFTs of x(t), v(t) and y(t), showing peak magnitudes and
cut-o frequencies.
b) How can m(t) be recovered from y(t)?
This is just another look a synchronous AM radio.
m(t)
1 + m(t)
x(t)
v(t)
cos(50 t)
y(t)
cos(50 t)
M (j )
H (j )
10
H (j )
10
30
30
Figure 89: System for Example 4.22.
266
Solution:
267
Example 4.23: Consider the following system with input x(t).

a) Sketch, showing frequency cutos and peak values, Y (j), Z(j) and V (j).
b) Determine the energy of v(t).
x(t)
H1 (j )
y(t)
z(t)
e
X(j )
v (t)
H2 (j )
j 10 t
H1 (j )
H2 (j )
10
1
10
10

Solution:
10
268
Example 4.24: Consider the modulation/ltering system shown below.

a) Let 0 = 100. Determine the energy Ey of the output y(t).
b) Now let 0 sweep between 200 0 200. Plot the output energy Ey
vs. 0 .
This problem illustrates the basic idea behind an analog spectrum analyzer.
x(t)
H (j )
j
e
X(j )
H (j )
2
1
100
80
y (t)
80 100

Solution:
269
Example 4.25: Consider the system below.

a) Sketch the CTFT of v(t), showing center & cuto frequencies, and amplitudes.
b) Specify 0 and the lter frequency response H(j) so y(t) = x1 (t) (i.e. so
that x1 (t) is received). For this 0 , sketch the CTFT of r(t) and y(t).
c) Instead of b), let 0 = 12, 500 (i.e. the receiver is mis-tuned), and assume
that the lter H(j) is ideal lowpass with a gain of one over the band
between 2500. Sketch the CTFT of r(t) and y(t).
This example illustrates the basic idea behind multiuser broadcast radio.
x1 (t)
v1 (t)
v(t) = v1 (t) + v2 (t)
cos(10,000 t)
x2 (t)
r(t)
H (j )
y(t)
cos( 0 t )
v2 (t)
cos(15,000 t)
X2 (j )
X1 (j )
1
2,500
2,500
2,500
2,500

Solution:
270
4.2.4
RLC Circuits
In your basic circuits course, you learned the following voltage/current relationships for the
three basic circuits elements:
v(t) = R i(t)
d
i(t)
v(t) = L
dt
t
1
v(t) =
i( ) d
C
(resistor)
(76)
(inductor)
(77)
(capacitor) .
(78)
Taking the bilateral Laplace transform of these equations, we have:

V (s) = R I(s)
V (s) = Ls I(s)
1
V (s) =
I(s)
Cs
(resistor)
(inductor)
(79)
(80)
(capacitor) .
(81)
This leads to the Laplace domain circuit diagram representations shown in Figure 93.
+
R
+
Ls
(a)
(b)
+
1
sC
(c)
Figure 93: Laplace transform representation of: (a) a resistor; (b) an inductor; (c) a capacitor.
The next two Examples illustrate the usefulness of these representations in directly analyzing the circuit transfer function. In following these Examples keep in mind how the
extension of the analysis procedure to higher order (more energy storage element) circuits
wound be systematic. It essentially reduces to analysis to that of a pure resistor circuit, so
loop and node dierential/integral equations are replaced with algebraic equations.
271
Example 4.26: Determine the transfer function of the following serial RLC circuit.
R
Ls
1
Cs
X(s)
Y(s)
Figure 94: Serial RLC circuit for Example 4.26.

Solution: Using the voltage divider equation, we have
Y (s) =
(1/Cs)
(1/LC)
X(s) = 2
X(s) .
Ls + R + (1/Cs)
s + (R/L)s + (1/LC)
Thus,
H(s) =
(1/LC)
Y (s)
= 2
.
X(s)
s + (R/L)s + (1/LC)
Example 4.27: Determine the transfer function of the following RLC circuit.
R
+
X(s)
L1s
Y(s)
L2s
1
C1 s
1
C2 s
Z(s)
Figure 95: RLC circuit for Example 4.27.

Solution: First note that the impedance of the right portion of the circuit is
Z(s) =
L2 C2 s2 + 1
(1/C1s)(L2 s + (1/C2 s)
=
.
(1/C1s) + L2 s + (1/C2 s)
L2 C1 C2 s3 + (C1 + C2 )s
Using the voltage divider equation, we have that the voltage across C1 is
VC1 (s) =
Z(s)
X(s) ,
L1 s + R + Z(s)
so that
Y (s) =
VC1 (s)
(C1 s Z(s))
=
X(s) .
(1/C1s)
L1 s + R + Z(s)
Thus, plugging in the expression for Z(s) and manipulating the expression into
rational function form, we get
H(s) =
L2 C1 C2 s3 + C1 s
Y (s)
=
.
X(s)
L1 L2 C1 C2 s4 + RL2 C1 C2 s3 + [L1 (C1 + C2 ) + L1 C2 ]s2 + R(C1 + C2 )s + 1
272
4.2.5
Spectrum Estimation
See Practicums 4(a) & 4(b) for an introductory experience on this topic. If you take the
Biomedical Signal Processing Course, you will learn much more about this topic.
4.3
273
Practicum 4a
Continuous Time Fourier Series (CTFS) & an Initial Consideration of Spectrum Analysis
Introduction
In this practicum you will study the representation of CT periodic signals using the CTFS.
Specically, you will see, via Matlab, how periodic signals can be built as weighted sums of
sinusoids. You will examine the eect of representing a periodic signal using only some of
the terms of its CTFS expansion. Finally, you will study the eect that modication of the
CTFS coecients has on a periodic signal.
To simplify coding and interpretation, in this practicum we will consider a real-valued
signal and its CTFS in cosine (i.e. trigonometric) form. For a real-valued periodic signal
with period T and fundamental frequency 0 = 2/T, the CTFS can be written as
x(t) = a0 + 2
k=1
|ak | cos(k0 t + ak )
(1)
where the ak as the exponential CTFS coecients

ak =
1
T
T
0
x(t)ejk0 t dt .
(2)
Specically, we will consider a periodic signal x(t) with period T = 4 which is dened over
one period as
x(t) = (4 t)
0 t 4
(3)
Its CTFS coecients are
ak =
2
j2
k
2
k
ej/2
k=0
otherwise
(4)
274
Procedures: Before the session, complete Procedure 1.(a), and use CTFS properties to
identify the time domain relationship between x(t) and y(t) in Procedure 2.(a).
1. CTFS Representation of a periodic signal:
(a) With pencil and paper, determine the power of x(t) as
P =
1
T
T
0
x2 (t) dt
(5)
Compare this power to the sum of the magnitude-squares of the ak s. This result illustrates a property of the CTFS (and CTFT) called Parsevals theorem.
Comment on the result, and discuss how Parsevals theorem can be useful.
(b) Consider the approximation of x(t)
N
xN (t) = a0 + 2
k=1
4
cos(kt/2 /2) .
k=1 k
|ak | cos(k0 t + ak ) = 2 +
(6)
Using Matlab, plot x(t) for 2 t 8. Now plot xN (t) for N = 1, 3, 7, 11 and 51,
and compare these to x(t). In your own words, explain why the xN (t) are both
similar and dierent from x(t).

(c) Plot the power spectrum of x(t), |ak |2 vs. k and vs. for 6 6. How
much of the power of x(t) is not represented on this plot?
2. Modication of the CTFS Coecients:
(a) Now consider a new signal y(t) formed by modifying the CTFS coecients of the
x(t) above. Specically, let ax = ak , and let the CTFS coecients of y(t) be
k
2 T
4
ay = ax ejk T
k
k
= ax ejk 2
k
(7)
Use Matlab to plot the approximation of y(t)

y51 (t) =
ay
0
51
+2
k=1
|ay | cos(k0 t + ay ) .
k
k
(8)
Compare your estimate of y(t) to the original signal x(t). What eect has this
particular change of CTFS coecients had on the original x(t)? In your own
words, why does this make sense?
(b) Repeat 2(a) for z(t) formed using CTFS coecients
az = ax ej 2 (k/6)
k
k
(9)
275
3. The data le pract2 data.mat found in

v:/Electrical Computer/Buckley/ece3240
contains N1 = 2048 samples of a CT periodic signal taken at a sample rate of
fs = 48, 000 samples/sec.
(a) Copy the data le into your working directory or otherwise assure that you have
access to it within Matlab. Load the le into Matlab. In Matlab, what is the
data array name? Plot it. How many periods of the signal are present? What is
the period N of the signal (in samples)? What is the period T of the signal (in
seconds)?
(b) What is the fundamental frequency 0 of this CT periodic signal (in radians/sec.).
(c) By periodically extending the available data, create 1 sec. of the signal. Use
the Matlab soundsc function to listen to this 1 sec. segment. Considering your
experience from Practicum III, what does this signal sound like?
(d) Process the rst N samples of the data array with the Matlab t function13 . The
rst 64 values of the result will give the DC and rst 63 positive frequency CTFS
coecients. Specically, letting X[k]; k = 0, 1, , 63 denote these values, then
ak =
X[k]
;
N
k = 0, 1, , 63 .
(10)
Plot the magnitude of these values vs. frequency over the range 0 630 .
How many non-zero coecients are there over this range? What harmonic frequencies (in Hz.) do they correspond to? Where have you seen these harmonics
and harmonic levels before? What musical note and note frequency is this sound
closest in frequency to?
13
You will learn how this t Matlab function works, and how to use it in solving other Digital Signal
Processing (DSP) problems, in ECE5790 if you take it!
276
Practicum 4a
Students Name:
1. Procedure 1(a): derive the power of x(t)
2. Procedure 1(b): plots of xN (t); N = 1, 3, 7, 11, 51
3. Procedure 1(c): plots of |ax |2 vs. k and |ax |2 vs.

k
k
4. Procedure 2(a): write down the equation relating y(t) and x(t)
5. Procedures 2(a,b): plots of y51 (t) and z51 (t)

.
277
278
4.4
Practicum 4b
Spectrum Analysis
Introduction
In this practicum you will investigate both the CTFS and the CTFT. Specically, you will
study the frequency content of several CT signals. For each, you will numerically derive and
plot spectra. In eect, you will be designing, implementing and testing a digital spectrum
analyzer (i.e. a spectrum analyzer which computes spectrum estimates using digital hardware
and/or software). These days, the vast majority of spectrum analyzers are digital.
First you will develop a spectrum analyzer for CT periodic signals. That is, you will design
and implement a Matlab function for computing the CTFS coecients from samples of one
period of a CT periodic signal. You will test this spectrum analyzer on a signal with known
frequency content an Amplitude Modulated (AM) signal. You will then use it to analyze
an ElectroCardoGram (ECG) signal.
Second, you will develop a spectrum analyzer for CT energy signals. As with the CTFS,
you will rst test this spectrum analyzer on a signal with known frequency content an
energy pulse. You will then use it to analyze an Acoustic Emissions (AE) signal. For each
of these signals, you will draw conclusions from its spectrum.
1. Periodic Amplitude Modulated (AM) Sinusoid: In AM communications, the
transmitted information is embedded as the carrier sinusoids amplitude.
Consider the signal
x(t) = (1 + 0.25 cos(m t)) cos(c t)
(11)
where c (in radians/second) is called the carrier frequency. cos(c t) is the carrier
signal that controls what band of frequencies the transmitted signal occupies. m is
the frequency of the sinusoid used here to simulate the message signal. The constant
0.25 is called the modulation index. It controls the amount of variation in the carrier
sinusoids amplitude. In this practicum, let m = 2 and c = 20 m . (In radio
applications, c would be much higher, and cos(m t) would be replaced with speech
or music.)
(a) Before the rst practicum session, determine o and T , the fundamental frequency
and period of x(t)?
(b) Before the rst practicum session, analytically determine and sketch the CTFS
coecients ax .
k
(c) Recall that in Part 3 of Practicum 4(a) you took a black box approach to analysis. That is, you used the Matlab t function to compute CTFS coecients. Now
develop your own generally applicable spectrum estimation algorithm for CT periodic signals. Specically, starting with the exponential CTFS coecient integral
equation, use numerical integration to compute the ax s for k = 0, 1, 2, ..., 30.
k
Using the Matlab stem function, plot the power spectrum |ax |2 vs. . Referring
k
to this plot, describe the frequency content of x(t). Is it what you expected?
279
2. An ECG Signal: Now that you have tested and veried your spectrum analyzer on
data with know frequency content, you will proceed to use it to analyze a signal with
unknown frequency content. An ECG signal usually consists of repeated heart beat
signals, called QRST complexes. Although the resulting ECG signal is not exactly
periodic, because no two QRST complexes are exactly alike and because the rate of
these beats changes over time, we will treat the ECG signal as periodic here because
it can be an accurate approximation.
Consider data le, at
v:/Electrical Computer/ece3240/ECGdat
2
which contains samples of one period of an ECG signal x(t), covering time 0 t 3 ,
in seconds.
(a) What are o and T , the fundamental frequency and period of x(t)?
(b) What is the sample interval Ts and the sample rate fs ?
(c) Plot x(t) for 0 t 2.
(d) Using the spectrum analyzer you developed in Procedure 1, numerically compute
the CTFS coecients ax s for k = 0, 1, 2, ..., K. For reasons that will become
k
clear later when we discuss sampling, assume that x(t) contains frequency components up to f2s but not beyond. Select K with this in mind. Plot the power
spectrum |ax |2 vs. . Referring to this plot, describe the frequency content of
k
x(t). (Relating characteristics of this plot to the hearts condition is something
a Cardiologist does, assuming he or she understands signal frequency content.
A signal processing engineer would design the equipment based on specications
which assure that the Cardiologist is presented the spectrum to the level of accuracy she/he needs.)
3. Rectangular Energy Pulse: We now turn to the analysis of CT energy signals using
the CTFT. Consider a signal x(t) = p5 (t 2.5), i.e. a pulse of width 5 startingat t = 0.
Its CTFT is denoted X(j).
(a) Before coming to the practicum session, analytically derive X(j).
(b) Choose a sampling interval Ts , and generate and plot N samples of this x(t). Using
numerical integration to approximate the CTFT equation, compute estimates of
X(j) for M values of over the frequency range 0 W , and plot the
energy spectrum |X(j)|2. Your algorithm should treat Ts , N, M and W as
input parameters (so your program is exible enough to use with a variety of
signals). In eect, you are designing a general purpose digital spectrum analyzer.
Since x(t) is real-valued, its CTFT is complex symmetric. So, only compute
estimates of X(j) for nonnegative . However, plot its energy spectrum for
both negative and positive frequency over range W W.
280
The idea is to compute an accurate estimate of X(j) without resorting to excessive computation. Select Ts , N, M and W (the analysis bandwidth) so as to
obtain an accurate estimate of X(j). Justify your choices of the design parameters, and compare your plot to the known |X(j)|2.
In eect, you have employed your general purpose digital spectrum analyzer to
perform analysis on a specic signal of interest.
(c) In retrospect, can the general purpose spectrum analyzer you developed in this
procedure for CT energy signals be used to analyze CT periodic signals?
4. AE Signal: The data le

v:/Electrical Computer/ece3240/AEdat
contains a dense sampling of an AE signal y(t) over its duration 0 t .8 msec..
This signal was measured using a small acoustic sensor attached to a piece of metal as
it was bent so as to cause cracking.
(a) What is the spacing Ts between samples.
(b) Plot y(t) vs. time. Select values of Ts , N, M and W .
(c) Using the analyzer and analysis procedure you developed in Procedure 3., numerically compute samples of its CTFT Y (j), and plot the energy spectrum
|Y (j)|2 . As in Procedure 3., the plot should be over both positive and negative
frequency. Justify your choice of analysis parameters. From the plot, describe
the frequency content of the signal. Compare characteristics of the spectrum (e.g.
frequency positions of spectral peaks) to characteristics of the AE signal (e.g. the
frequency of oscillation).

.
281
282
Practicum 4b: Spectrum Analysis

Students Name:
For this Practicum, attach all plots requiring sign o below.
1. Procedure 1(a,b): write down T and 0 . Analytically derive the ax s (e.g. using the
k
CTFS equation or Table 6 of the Course Notes), and sketch them vs. .
2. Procedure 1(c): numerically computed |ax |2 vs. plot

k
3. Procedure 1(c): Are you convinced that your spectrum analyzer is functioning correctly?
4. Procedure 2(a): Record 0 and T for this signal
sampling interval Ts and rate fs for this data
. Also, record the

.
5. Procedure 2(a): For reasons that will become clear to you later in the Course when
we study sampling, assume that x(t) contains frequency components up to f2s but no
higher. With this in mind, determine the highest harmonic, K, that you will compute
a CTFS coecient for
.
6. Procedure 2(b,c): x(t), and numerically computed |ax |2 vs. plots
k
7. Procedure 2(c): Comment on the frequency content of the ECG signal.
283
8. Procedure 3(a): Analytically derive (e.g. using the CTFT equation or Table 7 of
the Coruse Notes), and sketch X(j).
9. Procedure 3 (a,b): write down your selected values for T, N, M and W

Explain your choices.
10. Procedures 3(b): plot of numerically computed |X(j)|2

11. Procedures 3(b): Are you convinced that your spectrum analyzer is functioning correctly?
. Are you comfortable with your selected values of T, N, M and W ,
and your rationale for them?
12. Procedures 4(b): Write down and explain your choice for values of T, N, M and W .
13. Procedures 4(c): plots of y(t) and numerically computed |Y (j)|2

Discuss your time-domain vs. frequency-domain comparison (i.e. how are characteristics in one domain manifested in the other).
284
4.5
Problems
Chapter Topics:
4.1-21 (CT transforms and LTI systems);
4.22-26 (ideal lters);
4.27-41 (sampling sinusoids & the sampling theorem);
4.42-51 (modulation, ltering and Parsevals theorem);
4.52-59 (circuits)
1. Consider a CT LTI system with impulse response
h(t) =
sin(10t)
t
1
and input x(t) =
n= p1/6 (t 3 n). Determine the percentage of input power that
makes it to the output.
2. Consider a CT LTI system with zero initial conditions and impulse response
h(t) =
and input x(t) =
n=
sin(10t)
cos(20t)
t
(t 0.25n). Determine the output power Py .
3. Consider the CT LTI system

y (2) (t) + 9 y(t) = x(1) (t) + 2 x(t)
(assume h(t) is right sided), and input
x(t) = 4 e2t u(t) .
(a) Determine the system transfer function H(s).
(b) Determine the impulse response h(t).
(c) Determine the output y(t) (assuming zero initial conditions).
4. Consider the causal DT LTI system with I/O LCC dierential equation
y (2) (t) + 3 y (1) (t) + 2 y(t) = x(2) (t) + 4 x(1) (t) + 3 x(t) .
This system has zeros z1 = 1 and z2 = 3 and poles p1 = 1 and p2 = 2.
a) Determine the transfer function H(s). Simplify it as much as possible.
b) From H(s), determine the impulse response h(t). (Hint: Consider the BLT linearity and dierentiation properties. It is important to complete a) properly rst.)
5. Consider a CT LTI system with transfer function
H(s) =
1
.
(s + 2)(s + 4)(s 1 j5)(s 1 + j5)
285
(a) If the system is causal, what is the ROC? Is the system stable? Explain your
answers.
(b) If the system is stable, what is the ROC? Is the system causal? Explain your
answers.
(c) Given that s = 3 is in the ROC, is the system causal? Is the system stable?
Explain your answers.
6. A LTI system has impulse response h(t) = u(t) + e2t cos(3t)u(t). Determine the
dierential equation that relates input x(t) and output y(t).
7. The following problems are independent of each other.
(a) Determine the CTFTs of x1 (t) = 2 p10 (t 10) and x2 (t) = 2
more energy? Why?
sin(5t)
.
t
Which has
(b) Determine the inverse BLT of

X(s) =
2s
6
4
=
s2 + 5s + 6
s+3
s+2
Rx = Re{s} > 2 . (12)
(c) Determine the BLT, Y (s) and ROC Ry , of y(t) = 6 e3t u(t) + 4 e2t u(t).
Write Y (s) as a single rational function.
8. Consider a causal CT LTI system for which input x(t) = (t) + 2 e4t u(t) results
in output y(t) = e4t u(t) e6t u(t). Determine the system transfer function H(s)
and the impulse response h(t).
9. Basic CT LTI Systems and Transforms:
(a) Consider the CT LTI system described by the following LCC dierential equation
2y (3) (t) 5y (1) (t) + y(t) = x(3) (t) + x(2) (t) + x(1) (t) .
Determine H(j).
(b) Consider a CT LTI system with transfer function
H(s) =
1
.
(s + 2)(s 4)
i. If the system is causal, what is the ROC? Is the system stable? Explain your
answers.
ii. If the system is stable, what is the ROC? Is the system causal? Explain your
answers.
iii. Given that s = 3 is in the ROC, is the system causal? Is the system stable?
Explain your answers.
10. Consider a CT LTI system with transfer function
H(s) =
s + 10
(s + 2)(s + 4)
Rh = Re{s} > 2 .
286
(a) What is the system frequency response?
(b) What is the system I/O dierential equation?
(c) What are the system poles and zeros?
(d) Derive the partial fraction expansion of H(s). Show all steps and intermediate
results, or get no credit.
(e) What is the system impulse response?
(f) Is the system stable? Why? Causal? Why?
11. Consider the following causal CT LTI system
y (2) (t) + 6 y (1) (t) + 9 y(t) = 8 x(t) .
Given zero initial conditions, its output is y(t) =
x(t). (Note that s2 + 6s + 9 = (s + 3)2 .)
40
24
t4 e3t u(t). Determine its input
12. Consider a CT LTI system with the following I/O dierential equation:
y (3) (t) y (1) (t) = 2 x(2) (t) 8 x(t) .
(13)
This system is causal with poles p1 = 0, p2 = 1 and p3 = 1.

(a) Is this system stable? Why?
(b) Determine the transfer function H(s), along with its ROC.
(c) Determine the impulse response h(t). Show all work in forming the PFE.
13. Consider a causal CT LTI system with I/O dierential equation
y (2) (t) + y (1) (t) 20 y(t) = x(1) (t) 58 x(t) .
(14)
This system has poles at s = 4 and s = 5.

(a) Determine its transfer function H(s).
(b) Determine its impulse response h(t). Show all work for the required PFE.
(c) Is this system stable? Explain your answer in terms of the impulse response. If
unstable, which pole causes the instability? What is the characteristic of this pole
that results in the instability?
14. A CT LTI system has transfer function H(s) =
20
.
100s2
(a) What is its frequency response H(j)?

(b) Given input x1 (t) = 4 cos(10t), determine the output y1 (t).
(c) Given output y2 (t) = 4 cos(10t), determine the input x2 (t).
287
15. Consider a CT LTI system with input

x(t) = 3 e3t u(t) + 4 e4t u(t)
and resulting output
y(t) = et u(t) + 2 e2t u(t) .
Determine the system transfer function, input/output dierential equation and impulse
response.
16. Consider a CT LTI system. You know that an input x(t) with Bilateral Laplace
Transform (BLT)
X(s) =
4s + 10
;
(s + 2)(s + 1)
ROC = Re{s} > 1
results in output y(t) = e2t u(t).

(a) Derive the partial fraction expansion of X(s). Show all steps and intermediate
(b) Determine the input x(t).
(c) Determine the BLT of y(t).
(d) Determine the systems transfer function H(s) and ROC.
(e) Determine the systems I/O dierential equation.
(f) Is the system stable? Why?
(g) (3-pts. extra credit) Determine the impulse response h(t).
17. Consider a causal CT LTI system with I/O dierential equation
y (1) (t) + 2 y(t) = x(1) (t) + 4 x(t) .
Given zero initial conditions, the output is observed to be
y(t) = (t) 2 e3t u(t) .
What is the input x(t)?
18. Consider a causal LTI CT system with I/O dierential equation
d
d
y(t) + 4y(t) =
x(t) + 3x(t)
dt
dt
with input x(t) = 3e3t u(t) 2e2t u(t) and initial conditions x(0 ) = 0 and
y(0 ) = 2. Determine the output for t 0.
19. Consider a CT LTI multipath communications channel
y(t) = x(t 0.5) + x(t 2.5) .
288
(a) What is the impulse response h(t) of this channel (i.e. y(t) = h(t) for x(t) = (t))?
(b) What is the frequency response H(j) of the channel? (Hint: to assist with part
(c), think of using the delay property and an Euler identity.)
(c) Plot the magnitude and phase of the frequency response over 2 2.
(d) If x(t) = cos( t) is transmitted through the channel, what is the output y(t)?
2
What do your answers to part (b) and (c) suggest about the problem with communications through multipath channels?
(e) If the output is y(t) = 5 cos(2t), what is the input x(t)?
20. Consider the following CT LTI system composed of CT LTI subsystems with transfer
functions
H1 (s) =
1
;
s+3
H2 (s) =
1
;
s+5
H3 (s) =
1
;
s + (1 + j)
H4 (s) =
1
.
s + (1 j)
H3 (s)
x(t)
(a)
(b)
H1 (s)
H2 (s)
(c)
y(t)
H4 (s)
(a) Determine the transfer function from node (b) to node (c).
(b) Assuming that there is no pole/zero cancellation when combining these subsystems into the overall system (from node (a) to node (c)), what are the poles of
the overall system?
(c) If the overall system is causal, is it stable? Why?
21. For each system description on the left, determine the one system description on the
right which is equivalent to it.
289
h(t) = ej10t u(t)
H(j) = 2 ( 10)
y (2) (t) 100 y(t) = 20 x(t)
h(t) = cos(10t)
y (1) (t) = x(t)
h(t) = ej10t
y (1) (t) j10 y(t) = x(t)
y (2) (t) + 10 y(t) = x(t)
h(t) = sin(10t) u(t)
H(j) = ( + 10) + ( 10)
H(s) =
1
;
s2 +10
Rh = Re{s} > 0
H(s) =
1
;
s10
Rh = Re{s} < 0
poles : p1,2 = 10; Rh = Im{s} > 5
H(j) =
20
2 +100
h(t) = u(t)
H(j) =
10
100 2 2
22. Given an input x(t) = e|t| to a CT LTI system with ideal lowpass lter frequency
response H(j) = p2 (), determine the percentage of input energy that makes it to
the output. You may have use for the indenite integral
1
1
d =
+ tan1 ()
(1 + 2 )2
2(1 + 2 )
2
and the facts that tan1 (1) =
and tan1 (0) = 0.
23. Given a LTI system with frequency response H(j) = p12 (), determine the output
y(t), and its energy & power, for the following inputs:
(a) x(t) = cos(12t).
(b) x(t) = cos(12t) u(t).
(c) x(t) = cos(12t) sinc(6t/).
24. Use the CTFT synthesis equation to derive the impulse response of each of the following
ideal frequency selective lters, and compare the result to that given in the example:
(a) the highpass lter of Example 4.15;
(b) the bandpass lter of Example 4.16;
(c) the bandstop lter of Example 4.17.
25. Starting with the ideal lowpass lter impulse response derived in Example 4.13, use the
modulation property of the CTFT to derive the impulse response of an ideal bandpass
lter. Compare your result to that of Example 4.16.
290
26. Consider the following CT lowpass lter I/O dierential equation:
y (2) (t) + 2 y (1) (t) + y(t) = x(t) .

(a) Determine the frequency response. Use Matlab to plot the magnitude and phase
responses.
(b) Determine if this lter meets the following lowpass ltering specication:
|H(j3)| < .1.
(c) Determine if this lter meets the following lowpass ltering specication:
|H(j 0.5)|db > 3dB.
27. Consider an ideal C-to-D converter, operating at sample rate fs = 300, that samples a
signal
x(t) = 5 cos(100t) + 2 cos(400t)
to generate x[n] = x(nTs ) where Ts = f1s . The signal x[n] is then processed by an ideal
D-to-C converter, also operating at fs = 300, to derive output y(t).
(a) Determine and plot the spectrum for x(t), x[n] and y(t).
(b) Determine y(t).
28. Sampling Sinusoids: Consider an ideal C-to-D converter, operating at sample rate
fs = 300, that samples the complex-valued signal
x(t) = 5 e
j100t
+ 5e
j500t
D-to-C converter, also operating at fs = 300, to derive complex-valued output y(t).
(a) Determine and plot the spectrum for x(t), x[n] and y(t). (b) Determine y(t).
29. An ideal C-to-D converter, operating at sampling rate fs = 8, 000 samples/sec., has
input x(t) and output x[n]. Say that x[n] = cos( n), and you know that the input is
4
of the form x(t) = cos(2f0 t), but you dont know f0 . Given that you know only that
f0 10, 000, give three possible values of f0 . Justify your answers.
30. Sampling Sinusoids:
(a) A CT signal xc (t) = 6 + 3 cos(12, 000t) + 4 cos(20, 000t) is sampled at rate
fs = 10, 000 samples/sec. with an ideal C-to-D converter to form x[n]. An
ideal D-to-C converter operating at rate fs is then used to form yc (t) from x[n].
Accurately sketch the frequency spectra of xc (t) and yc (t). Determine yc (t).
(b) A CT audio signal x(t) = 10 cos(12, 000t) was ideally sampled at a rate of 8000
samples/second to form samples x[n]; n = 0, 1, , 999 which were loaded into
the Matlab array xx. By mistake, Shane plays this audio signal using the Matlab
command soundsc(xx,6000) (i.e. he plays it back assuming a sample rate of 6000
samples/second). Let y(t) denote this CT signal. Sketch the spectra of x(t), x[n]
and y(t).
291
31. Consider an ideal C-to-D converter, operating at a typical speech sampling rate of
fs = 10, 000 samples/sec.. It samples the audio signal
x(t) = 2 cos(1, 600t) + 2 cos(12, 000t)
audio D-to-C converter, also operating at fs = 10, 000 samples/sec., to generate audio
output y(t).
(a) Determine and plot the spectra for x(t), x[n] and y(t).
(b) Determine y(t).
(c) If you listen the the input x(t), what frequencies (in Hertz) will you hear?
(d) If you listen the the output y(t), what frequencies (in Hertz) will you hear?
32. Consider an ideal C-to-D converter, operating at the high delity audio sampling rate
fs = 44, 100 samples/sec., that samples the audio signal
x(t) = 5 cos(30, 000t) + 2 cos(50, 000t)
audio D-to-C converter, also operating at fs = 44, 100 samples/sec., to generate audio
output y(t).
(a) Determine and plot the spectra for x(t), x[n] and y(t).
(b) Determine y(t).
33. Consider a real-valued signal xc (t) which is a musical note, middle C with fundamental
frequency f0 = 260 Hz, that is composed of the rst ve harmonics with CTFS
coecients ax ; k = 1, 2, 3, 4, 5 = {1, .5, .3, .2, .1}
k
(a) Write the equation for xc (t) and carefully sketch the power spectrum |ak |2 vs.
(in radians/second) showing all values.
1
xc (t) is sampled with an ideal C-to-D converter with sample rate fs = Ts = 1, 600 samples/sec.
to form x[n] = xc (nTs ). yc (t) is then reconstructed from x[n] via an ideal D-to-C
converter operating at the same sample rate fs .
(b) Which harmonics of xc (t) are aliased when yc (t) is formed. Why? Demonstrate
this aliasing using a spectrum plot.
(c) Write the equation for yc (t).
34. A real-valued CT signal x(t) consisting of two sinusoidal components is sampled as at
rate of fs = 8, 000 samples/second and later ideally reconstructed assuming the same
sample rate. The reconstructed signal is y(t). The sinusoidal components of x(t) are
3k Hz and its 5-th harmonic.
(a) What are the two frequencies in x(t)?
292
(b) Of these, which are the same in both x(t) and y(t)? Why?
(c) Which frequencies in x(t) are not the same in y(t)? What are they in y(t)? Why?
35. Consider a real-valued xc (t) which is the musical note, middle C (with fundamental frequency f0 = 260 Hz). It is composed of four harmonics those with CTFS
coecients ax ; k = 1, 3, 5, 10 = {2, 2, 1, .5}.
k
(a) Write the equation for xc (t) and carefully sketch the power spectrum |ak |2 vs.
(in radians/second) showing all values.
1
xc (t) is sampled with an ideal C-to-D converter with sample rate fs = Ts = 2, 000 samples/sec.
to form x[n] = xc (nTs ). yc (t) is then reconstructed from x[n] via an ideal D-to-C
converter operating at the same sample rate fs .
(b) Which harmonics of xc (t) are aliased when yc (t) is formed. Why? Demonstrate
this aliasing by drawing the power spectrum plot of yc (t), labeling all frequencies.
(c) Write the equation for yc (t).
x(t) =
0.75
.
1.25 cos(100t)
(a) You wish to sample this signal at the minimum sample rate fs required such that
90% of the signal power is unaliased. Determine this rate.
(b) For the sample rate fs you determined in part (a), say you process x(t) with the
proper antialiasing lter to eliminate all components which would otherwise be
aliased. You then ideally sample this ltered signal at rate fs and reconstruct
it using an ideal D-to-C converter operating at rate fs . Call this reconstructed
signal ya (t). Determine an expression for ya (t).
37. The musical note A above middle C has fundamental frequency 0 = 880. An organ
generates this note with the rst few odd harmonics as follows
x(t) = cos(0 t) + 0.3 cos(30 t) + 0.3 cos(50 t) .
Say x(t) is ideally sampled at rate fs = 3000 samples/second and then ideally reconstructed to form y(t). Find y(t).
38. Consider the CT signal xp (t) = x(t) p(t), where
x(t) =
1
sin(10, 000t) ;
t
p(t) =
n=
(t n
1
)
8, 000
1
(i.e. x(t) is eectively sampled at a rate fs = T = 8, 000 samples/second). x(t) is
later reconstructed as y(t) = xp (t) h(t), where
H(j) =
T
0
|| < 8, 000
otherwise
293
a) Sketch the CTFTs of xp (t) and y(t), over 16, 000 16, 000, showing
important frequencies and levels. Is there aliasing? Why?
b) What is the energy of x(t) and of y(t)?
39. Consider the CT signal xp (t) = x(t) p(t), where
x(t) = p1 (t) ;
p(t) =
(i.e. x(t) is eectively sampled at a rate fs =

reconstructed as y(t) = xp (t) h(t), where
n=
1
=
T
(t n)
1 sample/second). x(t) is later
||
otherwise
1
0
H(j) =
(a) Sketch x(t), xp (t) and y(t).

(b) Sketch the CTFTs of x(t), xp (t) and y(t), over 2 2, showing important
frequencies and levels. Is there aliasing? Why?
40. Consider the following system, with x(t) =
H(j) =
Let T =
1
.
100
1
0
1
t
sin(100t), pT (t) =
400 || 600
otherwise
n=
(t nT ),
Determine/sketch XT (j). Determine xT (t) and y(t).

x(t)
H(j )
y(t)
p (t)
T
41. In this problem we consider ideal sampling (i.e. C-to-D conversion), ideal reconstruction (i.e. D-to-C conversion) and aliasing. Assume a sampling rate of fs = 10, 000
samples/sec. and corresponding sampling interval Ts = f1s .
(a) Consider the signal
x2 (t) = 2 cos(1, 000t) 2 cos(19, 000t) .
The C-to-D forms x2 [n] = x2 (n/fs ) and the D-to-C reconstructs y2 (t) from
x2 [n] as discussed in class. Plot X2 (j) and Y2 (j).
(b) Let x3 (t) be a signal with CTFT as shown below. Let xTs (t) = x3 (t) p(t)
mathematically represent the ideal sampling of x3 (t), where p(t) =
n= (t
nTs ). We then lowpass lter xTs (t) as shown to exactly reconstruct x3 (t) if there
is no aliasing.
Accurately plot P (j), XTs (j) and Y3 (j). Why is y3 (t) not equal to x3 (t) ?
294
X 3 (j )
x (t)
y (t)
H(j )
H(j )
Ts
1
20,000
20,000
10,000
10,000
1
42. Let X(j) = cos 40 p40 ().
Let v(t) = x(t) cos(100t). Let w(t) = v(t) cos(80t). Let Y (j) = W (j) p40 ().
(a) Carefully sketch X(j), V (j), W (j), and Y (j).

(b) What percentage of energy of w(t) is in y(t)? (Hint: can you determine this by
inspection of the sketches?)
43. Consider the system shown below which consists of a modulator and a frequency selective lter. The input is x(t) = sin(10t) .
t
x(t)
v(t)
CT LTI
H(j )
y(t)
H(j )
1
e
j 30 t
20
40
a) Sketch V (j) showing pertinent axis values.

b) Sketch Y (j) showing pertinent axis values.
c) Determine y(t) and its energy.
44. Consider the CT system shown below. Let x(t) =
1
and hhp (t) = (t) t sin(50t).
adder
x(t)
1
t
sin(10t), 0 = 10, 1 = 50
multiplier
r(t)
v(t)
CT LTI
y(t)
h hp(t)
cos ( t)
0
cos ( t)
1
(a) Carfully sketch the CTFTs of r(t), v(t) and y(t). Show frequency and CTFT
values.
(b) What is the energy of y(t)? Is y(t) an energy signal?
45. Consider a CT system in which input x(t) is rst multiplied by a complex sinusoid
ej10t and then lowpass ltered by H(j) = p2 (). For input with CTFT
X(j) = 1 + p20 (), determine the energy Ey of the output signal y(t).
295
46. The CT system illustrated below in part shows the ideal C/D and D/C converter
model (i.e. the sinusoid sampling and aliasing results established in this course apply).
Let xc (t) = cos(45t), and assume p0.1 (t) is an impulse train with impulse spacing
T = 0.1 seconds (i.e the C/D converter operates at a fs = 10 sample/second rate).
The ideal lowpass reconstruction lter has cuto frequency W = 10.
Ideal DtoC and CtoD converter model
x c (t)
0.1
(t)
DT LTI
modulator section
y(t)
h hp(t)
h lp (t)
ideal lowpass
reconstruction
filter
(t)
0.1
x c (t)
CT LTI
r(t)
ideal highpass
filter
cos ( t)
1
(a) Sketch the CTFT of y(t) over the range || 50.
(b) The purpose of the modulator section of the gure above is to recover xc (t) from
y(t). What values of modulator frequency 1 and highpass lter cuto frequency
W1 will accomplish this? Justify your choices with a brief discussion and/or
sketches.
(c) Sketch the CTFT of x0.1 (t) over the range || 50.
47. Consider the following system. The input has CTFT X(j) = 1 + A p2 ( 0 ).
Denote the energy of the output as Ey (1 ). Note that it depends (i.e. is a function of)
the modulation frequency 1 .
H(j )
x(t)
v(t)
CT LTI
H(j )
j 1 t
y(t)
1
1
(a) For what range of 1 will Ey (1 ) NOT be eected by the pulse in X(j)? What
is Ey (1 ) for this case? Denote this value E0 .
(b) As we sweep through values of 1 , we want to monitor Ey (1 ) to detect the
presence of a pulse in X(j). We do this by comparing Ey (1 ) to a threshold
level T . If Ey (1 ) > T we detect the pulse. Set T = 2E0 (so we dont get a
false detection if something minor goes wrong with our circuit). Determine the
minimum value of A required to detect the pulse (when 1 = 0 ).
(c) Plot Ey (1 ) vs. 1 for 0 = 10.
48. Consider the common model, shown in Figure (a) below, for ideal sampling a CT signal
1
xa (t), where pT (t) =
n= (t nT ) and T = 125 is the sampling interval.
Let xa (t) = cos(100t) + cos(200t) + cos(300t). Determine y(t).
296
x (t)
x (t)
y(t)
CT LTI
x (t)
a
x (t)
CT LTI
H(j )
p (t)
p (t)
(a)
y(t)
CT LTI
H (j )
H(j )
(b)
H(j )
Now consider Figure (b) above which is the same as Figure (a) except it includes a
lowpass lter Ha (j) = p2W (j). This lter is called an anti-aliasing lter since its
cuto frequency W is selected to eliminate all frequency components of xa (t) that
would be aliased, but no other frequency components. Determine W in terms of the
sampling interval T .
49. Consider the following CT system. The input x(t) has CTFT as shown, and it energy
1
is Ex = 4 . The CT LTI system has impulse response h(t) = sin(30t) .
t
X(j )
1
v(t) = x(t) e
x(t)
j 30 t
CT LTI
y(t) = v(t) * h(t)
h(t)
/2
/2
j 30 t
a) Sketch V (j), and determine its energy Ev .

b) Sketch H(j).
c) Sketch Y (j), and determine its energy Ey .
Label and show axis details on each plot.
50. Consider the following system, where x1 (t) = cos(5t) and x2 (t) =
x1 (t)
1
t
sin(5t).
v1 (t)
v(t) = v1 (t) + v2 (t)
cos(100 t)
x2 (t)
v2 (t)
r(t)
H (j )
y(t)
cos( 0 t )
cos(122 t)
a) Sketch the CTFT of v(t).

b) Determine the energy of v(t) over the frequency band 10 30.
c) Select 0 and the lter frequency response H(j) so y(t) = x1 (t).
297
d) For your part c) selections, sketch the CTFT of r(t) and y(t).
51. Consider the modulation/ltering system illustrated below. The input x(t) has CTFT
X(j) = |X(j)| ej3 , where |X(j)|2 is shown.
X(j )
H(j )
x(t)
v(t)
CT LTI
y(t)
H(j )
j 0 t
(a) Is x(t) real-valued? Why?

(b) What is Ex , the energy of x(t)?
(c) Determine the value of the modulation frequency 0 that maximizes the energy
Ey in the output y(t).
(d) For your 0 , what is the energy in y(t)?
52. Consider the following capacitor circuit.
(a) Determine the transfer function H(s), frequency response H(j) and impulse
response h(t).
(b) Is the system stable? Give two reasons why.
(c) Determine the step response s(t).
(d) Determine the output y(t) due to input x(t) = cos(0.1t) u(t).
+
x(t)
y(t)
53. Consider the following circuit. Let R = 1 M and C = 1 F . Determine the

Y (s)
transfer function H(s) = X(s) .
R
+
C
x(t)
y(t)
298
54. Consider the following current divider.
(a) Determine the transfer function H(s), I/O dierential equation, and poles/zeros.
(b) For L1 = L2 = 1 H, R1 = 1 and R2 = 7, nd the impulse response h(t).
(c) For the inductor/resistors in (b), nd the frequency response and sketch |H(j)|2.
(d) Determine 3dB such that
|H(j3dB )|2
{|H(j)|2 }max
= 0.5.
y(t)
R1
R2
L1
L2
x(t)
55. For the following RL circuit, let R = 1 k and L = 10 mH.

(a) Determine the frequency response H(j).
(b) Determine y(t) for x(t) = 1 volt.
(c) Determine y(t) for x(t) = cos(109 t) volts.
(d) Based parts (b,c) results, what type of lter would you say this is? Why?
L
+
+
R
x(t)
y(t)
56. Consider the following circuit.

y(t)
x(t)
(a) Determine the transfer function H(s).
299
(b) Let R = 1 , L = 2.5 H, and C = 0.4 F . Find the step response y(t) = s(t).
(c) For L = R2 C, determine the response to x(t) = sin2 (10t) u(t).
57. For the following circuit, let R = 1, C1 = 2 and C2 = 6.
R
+
C1
x(t)
C2
y(t)
(a)
(b)
(c)
(d)
(e)
Determine the transfer function H(s) and the impulse response h(t).
Let x1 (t) = 2 et/2 . Find the output y1 (t).
Let x(t) = 2 et/2 . Find the output yzs (t); t 0.
Let x(t) = 0, y(0 ) = 1 and y (1) (0 ) = 0. Determine the output yzi (t); t 0.
Sketch |H(j)|2 and determine H(j).
58. Consider a parallel RL circuit, shown below, with input current source x(t). Let the
output y(t) be the voltage across the components.
(a) Determine H(s), H(j), h(t), the I/O dierential equation, and the poles/zeros.
(b) For stability, what are the restrictions on R and L?
(c) For R = 100 and L = 0.1, determine the output for input x(t) = cos(1, 000t).
+
L
x(t)
y(t)
59. Consider the following circuit, for which C = 1 and R = 2.

R
+
+
C2
x(t)
y(t)
(a) Determine the circuit impulse response h(t).

(b) Given that the output voltage is y(t) = (t)
determine the input voltage x(t).
3
2
e2t u(t) so that Y (s) =
s+ 1
2
,
s+2
300
301
Discrete Time Transforms
In the introduction to Chapter 3, we considered the table of the transforms that are covered
in this Course. This Table, Table 5, is reproduced below. In Chapter 3 we covered the
transforms in column 2 of this table. That is, we covered CT transform. In this Chapter
we consider the DT transforms which are most commonly used for DT signal and system
analysis. We cover the three transforms in column 1 of Table 5: the Discrete-Time Fourier
Series (DTFS); the Discrete-Time Fourier Transform (DTFT); and the z-transform.
Table 5: Transforms
P
e
r
i
o
d
i
c
E
n
e
r
g
y
G
e
n
e
r
a
l
Discrete Time
Discrete Time Fourier Series
(DTFS): Period N
x[n] =
X[k] =
k=<N >
1
N
X[k] ej(2/N )kn
n=<N >
x[n] ej(2/N )nk
Discrete Time Fourier Transform

(DTFT)
x[n] =
1
2
X(ej ) =
X(ej ) ejn d
n=
x[n] ejn
z-Transform
x[n] =
1
2j
X(z) =
X(z) z n1 dz
n=
x[n] z n
Continuous Time
Continuous Time Fourier Series
(CTFS): Period T0 ; 0 = 2
T0
k=
x(t) =
ak =
1
T0
ak ejk0t
x(t) ejkot dt
<T0 >
Continuous Time Fourier Transform

(CTFT)
x(t) =
1
2
X(j) =
X(j) ejt d
x(t) ejt dt
Laplace Transform
x(t) =
1
2j
X(s) =
c+j
cj
X(s) est ds
x(t) est dt
Recall that for each transform in Table 5 a synthesis and an analysis equation is presented.
The synthesis equation, listed rst, shows a signal represented (or synthesized) as a linear
combination of the basic signals used for that transform. This synthesis equation is also called
the inverse transform. The second equation is called the analysis or transform equation. It
analyzes the signal to generate the weightings for the linear combination representation. For
example, as we will formalize later, note that with the DTFT (Table 5, 1-st column, second
row), DT energy signals are synthesized as linear combination of complex-valued sinusoid
over the continuous range of frequency < .
Because of the close DT/CT analogy between these two sets of three transforms in Table
5, the material in this Chapter and Chapter 6 is very similar to that of Chapters 3 & 4. This
302
should be expected since the transforms considered in this Chapter are the DT analogies of
those introduced in Chapter 3 (i.e. they use the same basic signals to represent the same
classes of signals, except that in this Chapter they are DT). Plus, after all, a transform
is a transform. Therefore, we will see that many of the properties and the applications
of these DT transforms will parallel those of CT transforms. This Chapter can be covered
independent of Chapters 3 & 4, although it is recommended that at least the introduction and
CTFS Sections of Chapter 3 be discussed rst, since developing a qualitative understanding
of the frequency content of a signal is probably most easily accomplished with the CTFS
(i.e. for CT periodic signals). Whichever set of Chapters is covered rst, 3 & 4 or 5 & 6,
take the second set of Chapters covered as an opportunity to reinforce and validate what
youve already learned in the rst, and to extend the basic idea of transforms to new classes
of signals.
The organization of this Chapter closely parallels that of Chapter 3. In Section 5.1 of this
Chapter we provide an introduction which formalizes the fact that any exponential input
to a DT LTI system results in a exponential output with the same exponent. (Recall that
we have already established this in Chapter 2 through multiple examples.) In Sections 5.2
and 5.3 we introduce, respectively, the DTFS and the DTFT. In Section 5.4 we generalize
the DTFT to include the DTFS. In Section 5.5 we consider some of the most important
properties of the DTFT. In Section 5.6 we focus on those aspects of the z-transform which
are most relevant to DT LTI system representation and analysis.
Understand, on a basic level, what a transform is.
Understand specically what class of signals are represented by and what basic signals
are employed for the DTFS, the DTFT and the z-transform.
Be able to use the DTFS, DTFT and z-transform synthesis/analysis equations and
transform tables.
Be familiar with the transform properties covered in this Chapter, understanding which
are more important in this Course and why.
5.1
303
The Frequency Response & Transfer Function of a DT LTI

System
This Section parallels the CT discussion in Section 3.1 on CT LTI system response to complex
exponential inputs. Let x[n] be a DT signal for example an impulse [n], a step u[n], an
exponential an u[n], a complex sinusoid ej0 n ; < 0 < (in radians/sample), or a realvalued sinusoid A cos(0 n + ). Recall that for DT sinusoidal signals we restrict frequency
to the range < , or sometimes 0 < 2, because outside a range of width 2
sinusoidal frequencies are ambiguous. As depicted in Figure 96, if x[n] is the input to a DT
LTI system with impulse response h[n], then the output is
y[n] = x[n] h[n] =
x[k] h[n k] ,
(1)
i.e. as established in Chapter 2, the output is the convolution of the input and the impulse
response.
[n]
x[n] =
x[k]
h[n]
DT LTI system
impulse resp. h[n]
[nk]
y[n] =
x[k] h[nk]
k
Figure 96: A DT LTI system and the convolution sum.

Let the input to a DT LTI system be an exponential z n (where z = rej is any complex
number with magnitude r and phase ). Then the output is
y[n] =
=
k=
h[k] x[n k]
(2)
h[k] z (nk)
(3)
k=
= zn
k=
h[k] z k
(4)
304
So, for an exponential input x[n] = z n to a DT LTI system, the output is an exponential
with the same z. Weve seen this behavior before for CT LTI systems. So, for this specic
(exponential) type of signal,
y[n] = H(z) z n
(5)
where
H(z) =
h[n] z n
(6)
n=
is called the transfer function. If we specically look at values of z on the unit circle, i.e.
z = ej , we have that for complex sinusoidal inputs, of the form x[n] = ej0 n , the output is
y[n] = H(ejo ) ej0 n
(7)
where H(ej0 ) is
j
H(e ) =
h[n] ejn
n=
(8)
=0
which is called the frequency response function, evaluated at = 0 .

In general, H(ej ) is a complex-valued function. If we write it in terms of its magnitude
and phase, i.e.
j
H(ej ) = |H(ej )| ej H(e ) ,
(9)
then for input x[n] = A ej(0 n+) , the output is
y[n] = A |H(ej )| ej(0 n++
H(ej ))
(10)
This shows that for a LTI system, for a sinusoidal input, the output is sinusoid with the
same frequency, with magnitude and phase altered by the system as dictated by the frequency
response H(ej ). Using the linearity property of the DT LTI system and Eulers identity,
we have that the output for a real-valued sinusoidal input A cos(0 n + ) is
y[n] = A |H(ej0 )| cos(0 n + + H(ej0 )) .
(11)
This last result requires that H(ej0 ) = H (ej0 ), a symmetry property which we will cover
later that holds only for real-valued h[n]. (See, for example, Example 2.26 in Chapter 2.)
305
Example 5.1: Consider the impulse response h[n] = [n] + [n 1] for a DT LTI
system. Determine the frequency response and plot its magnitude and phase.
Solution:
H(ej ) =
h[n] ejn
n=
= 1 + ej = ej/2 (ej/2 + ej/2 )
= ej/2 2 cos( )
2
|H(ej )| = 2 cos( )
2
H(ej ) =
||
| H(e j ) |
H(ej )
/2
....
....
1/2
/2
Figure 97: Frequency response for Example 5.1.

1
Example 5.2: Let the impulse response for a DT LTI system be h[n] = 10 p10 [n],
i.e. a 10-point averager. Determine the frequency response and identify its
magnitude and phase.
Solution:
306
Example 5.3: The impulse response for a DT LTI system is h[n] = [n] [n 1].
Determine the frequency response and plot its magnitude and phase. Determine
the output when the input is x[n] = x1 [n] + x2 [n] = 1 + ejn .
Solution:
H(ej ) =
h[n] ejn
n=
= 1 ej = ej/2 (ej/2 ej/2 )
= jej/2 2 sin( )
2
|H(ej )| =
H(e ) =
2 sin( )
2
2 sin( )
2
0<
< 0
0<
< 0
| H(e j ) |
H(ej )
/2
....
....
/2

From this we see that H(ej0) = 0 and H(ej ) = 2. So y[n] = 2 ejn .
307
Exploiting the linearity property of the DT LTI system, results we just established for
exponential and sinusoidal inputs extends to basically any input as shown below.
j 0 n
N 1
x[n] =
DT LTI system
h[n]; H(e
ax ejk0n ; 0 =
k
k=0
x[n] =
1
2
x[n] =
1
2j
2
N
X(ej ) ejn d
H(e
X(z) z n1 dz
j 0
) e
j 0 n
H(z) z n
y[n] =
N 1
ax H(ejk0 ) ejk0n
k
k=0
y[n] =
1
2
y[n] =
1
2j
X(ej ) H(ej ) ejn d
X(z) H(z) z n1 dz
Figure 99: Transform based I/O representations of a DT LTI system.

This in part motivates the following transforms:
1. The DTFS: this is the representation of a DT periodic signal as a linear combination of harmonically related complex sinusoids. Unlike the CTFS, in general only N
frequencies are required, essentially because of the ambiguity of DT frequency.
2. The DTFT: This is the representation of a DT energy signal as a linear combination
of complex sinusoids over a continuum of frequencies. Unlike the CTFT, only a nite
range of frequencies are required (i.e. < ). This is because of the ambiguity
of DT frequency.
3. The z-transform: This is a representation of a almost any DT signal as a linear combination of z n signals over certain continuous sets of z.
Of course, as pointed out earlier on several occasions, since transforms provide a decomposition of a signal as a linear combination of simple signals (e.g. complex sinusoids), the study
of transforms needs no further motivation.
308
5.2
The DTFS Representation of DT Periodic Signals
This Section parallels the CT discussion in Section 3.2 on the CTFS. Let x[n] be periodic
with fundamental period N, i.e.
x[n] = x[n + N] ;
all n .
(12)
The fundamental frequency is 0 = 2 , in radians/sample. The harmonics of 0 , over the

N
the unambiguous frequency range 0 < 2 are
k 0 =
2
k ; k = 0, 1, . . . , N 1 .
N
(13)
(Remember that all harmonics outside this range will be identical, sample-by-sample in
time, to a harmonic in this range.) The k th harmonic is 2 k. DC is the 0th harmonic. It
N
is important to understand that, because all frequencies separated by an integer multiple of
2 correspond to the same sinusoidal signal, the (N k)th harmonic, 2 (N k) is the same
N
as the k th harmonic 2 (k). This is illustrated in Figure 100.
N
2
(k)
N
2
(1)
N
N+k
2
N
2
(Nk)
N
Nk
2
(N1)
N
N1
Figure 100: Illustration of the ambiguity of DTFS frequencies.

DT complex sinusoids of frequency k0 ; k = 0, 1, , N 1 are periodic with period
N = 2 (i.e. there will be k periods over N samples, so N is the fundamental period for
0
k = 1 only). Any linear combination of these will also be periodic with period N. This leads
to the DTFS, which states that any periodic signal x[n] with period N can be synthesized
as
x[n] =
N 1
ak ejk0n =
k=0
N 1
ak ej(2/N )kn
all n .
(14)
k=0
This equation, the transform synthesis equation, called the inverse DTFS. The transform
analysis equation, which is called the DTFS equation, is
ak =
1
N
N 1
x[n] ejk0 n =
n=0
1
N
N 1
x[n] ej(2/N )kn
all k .
(15)
n=0
Note that the ak are periodic in k with period N. (This is also a result of the ambiguity
of DT frequency.) Because x[n] and ak are periodic with period N, as are the ej(2/N )kn , it
does not matter what period we sum over. So,
ak ej(2/N )kn
all n
x[n] ej(2/N )kn
x[n] =
all k
(16)
k=<N >
ak =
1
N
n=<N >
(17)
309
where < N > denotes any period. Try proving that the inverse DTFS equation, with the ak s
given by the DTFS equation, is true. To do this, plug Eq (17) into Eq (16) and simplify.
From a theoretical perspective, the DTFS is not a transform that we are primarily interested in for two reasons. First, except for sinusoids, we dont often encounter DT periodic
signals in signal processing problems, and it is easy to identify the frequency content of DT
sinusoids without the DTFS formalization (see Example 5.6 below). Second, as established
below in Section 5.4, we can use the DTFT to represent DT periodic signals (just as we can
use the CTFT to represent CT periodic signals). However, from a implementational perspective, the DTFS is very important. This is because, as opposed to the other transforms we
consider in this Course, the DTFS can be computed with digital hardware or software (i.e.
we can implement the DTFS). This is because the DTFS equations, Eqs (14,15), require only
nite length summations, whereas the other transforms require either innite summations
or integrals.
The DTFS equations, Eqs (14,15), play a major role in many DSP implementations. (In
fact, in Practicums 4a,b, when using the Matlab t function, you were implementing these
equations. You will also be using them in Practicum 5 when you use the Matlab freqz
command.) Because of this, Eqs (14,15) constitute a primary topic of a senior level DSP
course. Below, we will simply consider a few examples and build a table of transform pairs.
Later we will show how the DTFS can be incorporated into the DTFT.
k=
Example 5.4: Let x[n] =
[n kN]. Determine the DTFS.
Solution: This derives Table 12, Entry # 5.

1
=
N
ak
N 1
x[n] ej(2/N )kn
n=0
N 1
1
[n] ej(2/N )kn
N n=0
1
=
x[0] ej(2/n)k0
N
1
k
=
N
=
....
1
N
....
0
2
N
4
N
6
N
....
N1
(N1)
2
N
Figure 101: DTFS coecients for Example 5.4.

So this x[n], a DT impulse train, has equal content at all its possible frequency
components. In the gure, note that ak is periodic with period N (in k) or,
equivalently, 2 in .
310
Example 5.5: Let x[n] =
Determine the DTFS.
k= x1 [n k16]
where x1 [n] = 2[n + 1] + 2[n 1].
Solution:
8
1
x[n] ej(2/16)kn
16 n=7
1
=
ej(/8)k + ej(/8)k
8
1
cos
k
k
=
4
8
ak =
ak
x[n]
....
....
DTFS; N=16
....
....
7
....
8
k

Example 5.6: Let x[n] = A cos(3 2 n); N = 10. Determine the ak s.
10
Solution: This derives Table 12, Entry # 3.
A j(2/10)3n
A j(2/10)3n
x[n] =
e
+
e
2
2
Since this is already expressed in the DTFS expansion form, we can pick out the
DTFS coecients by inspection. With N + 10 we see that x[n] consists of k = 3
and k = 3 harmonics. So,
A
((k + 3) + (k 3))
2
A
=
((k 3) + (k 7))
2
k = 4, 3, , 4, 5
ak =
k = 0, 1, , 8, 9
A
2
....
4
....
0
or
a
A
2
....
....
0

Note that in this example we describe ak over the two ranges commonly used,
i.e. 0 k N 1 and, assuming N is even, (N/2) + 1 k N/2.
311
Table 12: Discrete Time Fourier Series (DTFS) Pairs.
Signal
14
DTFS 15
(k = 0, 1, , N 1)
[k]
exp{j 2 ln}; 0 l N 1
N
[k l]
cos
2
ln
N
; 1l<
N
2
1
2
[[k l] + [k (l N)]]
sin
2
ln
N
; 1l<
N
2
1
2j
[[k l] + [k (l N)]]
1
N
[n lN]
l=
x1 [n lN]
1
N
1 + cos
2
k
N
l=
x1 [n] = .5[n + 1] + [n] + .5[n 1]
x1 [n lN];
l=
1
sin( 2 (N1 + 2 )k )
N
N sin( N k )
x1 = u[n + N1 ] u[n (N1 + 1)]; N > 2N1 + 1
15 x[n] is assumed periodic with period N samples. Some of the basic signals below (e.g.
the DC signal and pure sinusoids) may also have shorter periods than N.
16 The DTFS coecient sequence, the ak , is periodic with period N where N is also the
period of x[n]. For a given k, ak corresponds to the k th harmonic, 2 k. In this table
N
we describe ak over the one period k = 0, 1, , N 1. The DTFS is the periodic
extension of this over all k.
312
5.3
The DTFT Representation of DT Energy Signals
This Section parallels the CT discussion in Section 3.3. Let x[n] be an energy signal, i.e.
E =
n=
|x[n]|2 < .
(18)
Then, according to the DTFT synthesis equation (i.e. the inverse DTFT) in Table 5,
x[n] =
1
2
<2>
X(ej ) ejn d
(19)
where, using the DTFT analysis equation (i.e. the DTFT),

X(ej ) =
x[n] ejn
(20)
n=
Note that, as with the DTFS, the frequency content is periodic with period 2. That is,
the DTFT X(ej ) is periodic in frequency with period 2. This is because ejn in Eq (20),
if considered a function of , is periodic with period 2. The synthesis equation, Eq (19),
indicates that any energy signal x[n] can be represented as a linear combination of complexvalued sinusoids. The frequencies of these sinusoids is any continuous range of width 2.
All this is a direct consequence of the frequency ambiguity of DT sinusoids. We usually
represents DT signals using frequencies over the range . So the inverse DTFT
is usually written as
1
X(ej ) ejn d .
(21)
x[n] =
2
Notice that the frequency response function H(ej ), introduced in Section 5.1, is the DTFT
of the impulse response h[n]. (Refer to Examples 5.1-2 in Section 5.1.)
313
Example 5.7: Find the DTFT of x[n] = an u[n]. Assume 0 < a < 1. Find the
magnitude-squared and phase of the DTFT. (This derives Table 13, Entry #3.)
Solution:
X(ej ) =
x[n] ejn =
n=
an ejn =
(a ej )n
n=0
n=0
1
1
=
j
1 ae
(1 a cos()) + j a sin()
1
1
=
|X(ej )|2 =
(1 a cos())2 + (a sin())2
(1 + a2 ) 2a cos()
a sin()
X(ej ) = tan1
1 a cos()
=
X(ej )
2
2
1
1a
....
....
2
1
1+a
Figure 104: DTFT for Example 5.7.

Example 5.8: Let x[n] = [n k] for some integer k. Find X(ej ). (This derives
Table 13, Entry #1.)
Solution:
314
Example 5.9: For the given X(ej ), nd x[n]. (This derives Table 13, Entry #8.)
Solution:
Example 5.10: Let h[n] = [n] + [n 1]. Find H(ej ).

Solution: See Example 5.1.
H(ej ) = 1 + ej = ej/2 (ej/2 + ej/2 ) = ej/2 2 cos(/2)
| H(e j ) |
H(ej )
/2
....
....
1/2
/2
315
Example 5.11: Find x[n] for X(ej ) = 2(); < . (This derives Table
13, Entry #13, for 0 = 0.)
Solution:
Note that this x[n] is not an energy signal. The DTFT is supposed to be for
energy signals, so thats going on here? If youve covered the CTFT Section,
then does this remind you of our initial sequence of DTFT examples?
Example 5.12: Find x[n] for X(ej ) = 2( 2 k); 0 < 2 where k is an
N
integer over the range k = 0, 1, , N 1. (This derives Table 13, Entry #13.)
Solution:
Example 5.13: Find X(ej ) for

x[n] =
N 1
ak ej( N k)n
k=0
Use results from Example 5.12 and the linearity of the DTFT equation integral,
X(ej ) =
N 1
k=0
ak 2
(This derives Table 13, Entry #14.)
2
k
N
0 < 2 .
316
Table 13: Discrete Time Fourier Transform (DTFT) Pairs

Signal
(n)
[n k]
DTFT16
( )
ejk
1
1
(1+n2 )
e||
an u[n]; |a| < 1
1
1aej
(n + 1)an u[n];
|a| < 1
1
(1aej )2
(n+r1)! n
a u[n];
n!(r1)!
|a| < 1
1
(1aej )r
6
7
ej
pN [n] = u[n] u[n N]

u[n + N1 ] u[n (N1 + 1)]
sin(W n)
;
n
sin( N )
2
sin( )
2
1
sin((N1 + 2 ))
sin( )
2
1 0 || W
0 W < ||
0<W
[n] 2
N1
2
0
0
sin2 ( n)
2
(n)2
10
a|n|
1a2
(1+a2 )2a cos
11
an cos(0 n) u[n]
1 [a cos(0 )] ej
1 [2a cos(0 )] ej + a2 ej2
12
an sin(0 n) u[n]
[a sin(0 )] ej
1 [2a cos(0 )] ej + a2 ej2
13
ej0 n ; 0
2 ( 0 )
14
N 1
k=0
ak ej(2/N )nk
N 1
k=0
2 ak
2
k ;
N
0 < 2
17 The DTFT, X(ej ), is periodic in with period 2. In the rst 7 table entries, the
X(ej ) expressions are valid for all . For the nal 4 entries, X(ej ) is described for
one period, and the DTFT is its periodic extension over all .
5.4
317
The DTFT Representation of Periodic Signals
This Section parallels the CT discussion in Subsection 3.3.1 on the generalized CTFT. Examples 5.11-13 suggest that, even though the DTFT as dened is for the representation of
DT energy signals, it can be used to represents the frequency content of periodic signals as
well. In fact, Example 5.13 shows how this is done for any periodic signal. The leads to
what is referred to as the generalized DTFT.
Let x[n] be a periodic DT signal with period N. We know from Section 5.2 of these Notes
that this signal consists of the weighted sum of N harmonic components, i.e.
x[n] =
N 1
ak ej N kn
(22)
k=0
where the ak s are the DTFS coecients. Example 5.13 establishes that the DTFT is
j
X(e ) =
N 1
k=0
2 ak ( k
2
)
N
0 < 2 .
(23)
To nd the DTFT of periodic x[n], nd the coecients of its complex-valued sinusoidal

expansion and plug them into Eq (23).
Example 5.14: Find the DTFT of x[n] = A cos( 3 n).
10
Solution: N = 20; harmonics at k = 3 over 9 k 10
A j(3/10)n
e
+ ej(3/10)n
2
2
2
X(ej ) = A ( 3 ) + ( + 3 )
20
20
x[n] =
<
X(ej )
(A )
(A )
3 1 0 1 2 3
32
20
3 2
20
10
318
Example 5.15: Find the DTFT of x[n] = A cos(n).
Solution: Referring back to Example 5.12, which established that the DTFT of
a complex-valued sinusoid is an impulse, note that there is no restriction on the
frequency in the sinusoid (i.e. it does not have to be of the form k 2 for some
N
integer N). Thus,
A jn
e + ejn
2
X(ej ) = A (( 1) + ( + 1))
x[n] =
X(e
(A )
<
(A )
l=
Example 5.16: Find the DTFT of x[n] =

x1 [n] = .5[n + 1] + [n] + .5[n 1].
x1 [n 10l] where
Solution:
1
10
1
=
10
1
=
10
x[n] ej(2/10)kn
ak =
n=4
.5ej(2/10)k + 1 + .5ej(2/10)k
2
k
10
1 + cos
Using Table 13. Entry #11, we get the X(ej ) shown below.
X(ej )
( 2 )
10
....
....
2
10
2
10

Recall that an impulse for a continuous independent variable (in this case ) is
innitely high. In the gure above impulses are drawn with dierent heights just
to provide a visualization of the dierent multiplication factor applied to each
impulse.
319
Example 5.17: Find the DTFT of x[n] =
l=
[n 10l].
Solution: From Table 12, Entry #5, we have that ak =

13 entry #13, we get the X(ej ) shown below.
X(e
1
10
for all k. Using Table
( 2 )
10
....
....
2
10
0 2
10

Example 5.18: A DT LTI system has frequency response H(ej ) shown below
over the range .
H(e j )
2
/3
/3

a) Sketch H(ej ) over the range 2 2. b) Determine the impulse
response h[n]. c) What is the output y[n], and its DTFT Y (ej ), due to input
x[n] = ej(/2)n + 2 cos((/5)n + (/3))
Solution:
320
5.5
Properties of the DTFT
If you have already covered Chapter 3 of these Notes, on CT transforms, then you are already
familiar with transform property concepts, and this Section will serve to reinforce these
concepts and to ne tune them for DT applications. If you have not yet covered transform
properties in Chapter 3, then even though this Section may be your rst formal treatment
of transform properties, you should nd the topic somewhat familiar. Upon reection, you
will see that several properties are exhibited in the Examples we have already considered
in this Chapter. As we proceed through this Section we will reect back on some of these
previous examples.
There are several reasons that we consider transform properties. One is that these properties enable us to use transform tables to an extended realm of synthesis and analysis
problems. So, for example, we will be able to use DTFT properties along with the DTFT
pairs in Table 13 to identify DTFTs and IDTFTs for a broad range of signals. However,
the utility of transform properties goes well beyond this. Properties will provide us with
insight into the transforms themselves and into the signals and systems they represent. The
delay property and Parsevals theorem for both the DFTS and DTFT are examples of this.
Transform properties are also useful in solving engineering problems. Finally, and most
importantly, we will see that many of the transform properties we consider represent important engineering applications. So with this in mind, in the following Subsections we will
investigate transform properties.
Our coverage of DT transform properties in this Chapter will proceed as follows. First,
in this Section, we consider some of the more useful and important properties of the DTFT.
Since the DTFT encompasses DT periodic signals, through the generalized DTFT, these
properties are also DTFS properties. Then, in Section 5.7, after we introduce the z-transform,
we will consider only those z-transform properties which are particularly useful for DT LTI
system analysis. The intent here is not to exhaustively list, prove, exemplify and motivate
every possible transform property. That would be exhausting. Rather it is to focus on a
few of the most basic and important properties, and to nourish a sense of the relevance of
this topic. You should keep in mind that there are other transform properties that down the
road you may nd useful. After completing this Chapter, you might skim through Section
3.5, on CT transform properties, to expand and reinforce what you learn here.
5.5.1
Periodicity of X(ej )
The DTFT X(ej ) of a signal x[n] is periodic with period 2. That is,
X(ej ) = X(ej(+2) )
(24)
This is a consequence of the fact that DT sinusoids are unique only over a 2 range of
frequency. This property is also explained by looking at the DTFT equation
X(ej ) =
x[n] ejn
(25)
n=
Each ejn term in the summation is periodic over with period 2. So the weighted sum
is periodic too.
321
Example 5.19: Show that the DTFT of h[n] = [n] + [n 1] is periodic in

with period 2.
Solution: This is the signal from Example 5.1. Back then, we had not yet dened
the DTFT. Nonetheless, we did in eect take the DTFT of this h[n]. Below we
reproduce the results.
H(ej ) =
h[n] ejn
n=
= 1 + ej = ej/2 (ej/2 + ej/2 )
= ej/2 2 cos( )
2
j
j
Note that H(e ) = 1 + e
is periodic with period 2 since both 1 and ej
j
are. Since H(e ) is periodic, its magnitude and phase must also be.
Concerning the magnitude, |H(ej )| = |2 cos( )| since the ej/2 term in H(ej )
2
does not eect magnitude. From the plot below, it is clear that |H(ej )| =
|H(ej(+2) )| for all .
| H(e j ) |
2
....
....
3
H(ej )
/2
....
3
....
1/2
/2

Concerning the phase, the plot above probably does not seem obvious. To explain
this, rst note from the expression H(ej ) = ej/2 2 cos( ) derived above,
2
clearly
|| ,
H(ej ) =
2
since 2 cos( ) is real-valued and positive over that range.
2
Over the range < 3, 2 cos( ) is negative, so it contributes a + to
2
the phase (i.e. 1 = ej ). Thus, as shown in the phase plot, over this range,
H(ej ) = + .
2
Similarly, over the range 3 < , 2 cos( ) is negative, so it contributes
2
a to the phase (i.e. 1 = ej ). Thus, as shown in the phase plot, over this
range, H(ej ) = .
2
Over the range 3 < 5, 2 cos( ) is again positive, so H(ej ) = + 2
2
2
(i.e. ej2 = 1). Etc.
322
5.5.2
Symmetry of X(ej )
For real-valued x[n],

X(ej ) = X (ej )
(26)
Remember that complex symmetry means

|X(ej )| = |X(ej )|,
X(ej ) = X(ej ),
Re{X(ej )} = Re{|X(ej )}, and

Im{X(ej )} = Im{|X(ej )}.
Proof: Although we will not be proving all for these properties, lets do this one.
It wont hurt, but it will give us a some condence in properties, and following
the proof is an exercise in mathematical exercise.
X(ej ) =
=
n=
n=
x[n] ejn
x[n] cos(n) j
x[n] sin(n) .
n=
Since, by assumption, x[n] is real-valued,
Re{X(ej )} =
x[n] cos(n) ,
n=
which is even symmetric, and

Im{X(ej )} =
x[n] sin(n) ,
n=
which is odd symmetric. Thus X(ej ) is complex symmetric.

Combining this property with the periodicity property, note that for a real-valued signal x[n],
to describe its DTFT X(ej ) over all , we need only describe it over the range 0 .
323
Example 5.20: Consider Example 5.19. Note that h[n] is real-valued. From the
plots presented in Example 5.19, we see that |H(ej )| = |H(ej )| (i.e. even
symmetric) and H(ej ) = H(ej ) (i.e. odd symmetric). So H(ej ) is
complex symmetric.
The phase plot shown in Example 5.19 and referred to in Example 5.20 is called
wrapped because 2 is added as needed to keep the plot looking periodic and
odd symmetric, with | H(ej )| . In the 2-nd gure below, the equivalent
unwrapped phase plot is shown.
H(e
7/2
5/2
3/2
....
....
/2
7
/2
3/2
5/2
7/2
H(ej )
....
7/2
5/2
3/2
/2
7
/2
3/2
5/2
7/2
....
Figure 112: Phase response for Example 5.20.

This 2-nd gure is just a plot of when 2 cos(/2) is positive. When 2 cos(/2)
2
is negative, either a is added (positive ) or subtracted (negative ).
In the 2-nd gure, the phase is still shown to be odd symmetric. Although in
its unwrapped form the phase does not appear to be periodic with period 2,
we know from Example 5.19 that it is. It only appears periodic when plotted as
wrapped (as in the 1-st gure above).
In Examples 5.19 & 5.20 we focus on the phase wrapping issue only so that you are not
confused when you see similar plots in subsequent Examples and Matlab based Practicums.
You need not worry about phase wrapping and unwrapping in this Course. The focus in this
Course will be more on the magnitude of the DTFT.
324
5.5.3
Time Delay
Given a DTFT pair
x[n]
X(ej )
(27)
the delay property says that, for any integer k,

x[n k]
X(ej ) ejk = |X(ej )| ej(
X(ej ) k)
(28)
It is important to the general understanding on what a transform is to step back and make
sure you understand what this property suggests. Recall that for a given x[n] and corresponding X(ej ), the complex-sinusoidal component of x[n] at frequency is
X(ej ) ejn .
(29)
If we delay the signal by k samples, then we delay each of its sinusoidal components by k
samples. Then you delay a sinusoid, you dont eect its magnitude, only its phase. You shift
its phase by k, which is proportional to the frequency and to the amount of delay. Thus,
after the delay, the component of x[n k] at frequency is
X(ej ) ejk ejn = |X(ej )| ej(n
X(ej ) k)
(30)
The idea is that a delay in the time domain is a phase shift in the frequency domain. This
phase shift is proportional to both the frequency of the sinusoidal component and the length
of the delay.
Example 5.21: Determine the DTFT of: (a) p11 [n 5]; and (b) p11 [n + 5].
Solutions: From Table 13, entry #6,
p11 [n]
j5
sin
sin
11
2
(a) Then, for delay k = 5, we have that

p11 [n 5]
j10
sin
(b) For delay k = 5 we have that

p11 [n + 5]
sin
sin
sin
11
2
11
2
We see that the application of the delay property is pretty straightforward.

In Example 5.21, note that p11 [n + 5] is symmetric around n = 0 and that its DTFT is
real-valued. This looks like the symmetry property introduced in Subsection 5.5.2, except
that the domains are switched.
More Symmetry Properties: If X(ej ) is real-valued, then x[n] is complex symmetric.
If x[n] is real-valued and symmetric, then so is X(ej ).
Just remember, concerning Fourier transforms, real-valued in one domain, means complex
symmetric in the other.

5.5.4
325
Linearity
Given DTFT pairs xi [n] Xi (ej ); i = 1, 2, , N and constants ci ; i = 1, 2, , N,

N
ci xi [n]
i=1
ci Xi (ej )
(31)
i=1
In Example 5.13, where we derived the DTFT of a weighted sum of complex-valued sinusoids,
we eectively used the linearity property of the DTFT. At that time we invoked it as the
linearity property of the DTFT integral.
For the DFTS, let xi [n]; i = 1, 2, , N all be periodic with period N, with corresponding
DTFS coecients axi ; i = 1, 2, , N. Let ci ; i = 1, 2, , N be arbitrary constants. Then
k
N
ci xi [n]
i=1
ci axi
k
(32)
i=1
Example 5.21: Determine the signal corresponding to the DTFT shown below.
X(ej )
5
....
....

Solution:
326
Example 5.22: Determine the IDTFT of
X(ej ) =
(r1 + r2 ) (p2 r1 + p1 r2 ) ej
.
1 (p1 + p2 ) ej + p1 p2 ej2
Solution: Assuming that we are given that

r1
r2
X(ej ) =
+
,
j
1 p1 e
1 p2 ej
then by the linearity property and Table 13, Entry #3, we have that
x[n] = r1 pn u[n] + r2 pn u[n] .
1
2
5.5.5
Convolution
The convolution property of the DTFT states that a convolution in the time domain is a
product in the frequency domain, i.e.
y[n] = x[n] h[n]
Y (ej ) = X(ej ) H(ej ) .
(33)
Since for a DT LTI system the input is convolved with the system impulse response to
derive the output, the implication of the DTFT convolution property for DT LTI systems
is signicant. Note, however, that the utility of this property goes beyond DT LTI system
analysis. In Section 5.6 we will explore the application on the DTFT to DT LTI systems in
some depth. For now, just consider the following proof and a few straightforward examples.
Proof: Following the following proof may be a test of fortitude, but not fortitude
required for a test.
y[n] =
=
k=
x[k] h[n k]
1
2
k=
1
=
2
X(ej )
X(ej ) ejk d
1
H(ej )
2
1
2
k=
H(ej ) ej(nk) d
ej()k
ejn d .
j()k
Using the fact that, for any < . ,
= 2( )
k= e
(this is a result from the theory of generalized functions), we have that
1
y[n] =
H(ej ) ( ) d ejn d .
X(ej )
2
But, H(ej ) ( ) d = H(ej ) (i.e. the integral of a function times an

impulse yields the function evaluated at the position of the impulse). Thus,
1
X(ej ) H(ej ) ejn d
y[n] =
2
1
=
Y (ej ) ejn d
2
where Y (ej ) = X(ej ) H(ej ). This completes the proof.
327
Example 5.23: Determine x[n] = an u[n] convolved with itself 10 times.

Solution: Let y[n] denote x[n] convolved with itself 10 times. Using Table 13,
Entry #3 and the linearity property, we have that
Y (ej ) =
X(ej )
10
1
.
(1 a ej )10
Then, by Table 13, Entry #5, we have

y[n] =
(n + 9)! n
a u[n] .
n! 9!
Example 4.24: Determine y[n] = x[n] h[n] where h[n] =

x[n] = 3 cos(0.2n) 4 cos(0.6n).
Solution:
sin(0.5n)
n
and
328
Example 5.25: Determine y[n] = x[n] h[n] where x[n] =

h[n] =
sin(0.5n)
.
n
where 0 < W <

Solution:
You can make use of the DTFT pair

0
(W/) /(2)
(W/) + /(2)
sin2 (W n)
(n)2
sin2 (0.4n)
(n)2
is assumed.
2W
0 2W
2W 0
2W
and

5.5.6
329
Parsevals Theorem
For DT energy signals

E =
n=
|x[n]|2 =
1
2
<2>
|X(ej )|2 d .
(34)
|ak |2 .
(35)
For DT periodic power signals

P =
1
N
n=<N >
|x[n]|2 =
k=<N >
If you have already covered Parsevals theorem for the CTFS and/or CTFT, refer back to
that discussion. Either way, Parsevals theorem quanties the idea of content as a function
for frequency. Note that power and energy are not eected by the phase of the frequency
description, but only the magnitude. Thus, for example, delaying a signal does not eect its
energy or power.
For the DTFT, Parsevals theorem establishes that for a signal x[n] with total energy E,
j 2
the function |X(e )| plotted vs. gives the distribution of total energy over frequency. Thus
2
we call this plot the energy spectrum. Similarly, for periodic power signals, |ak |2 vs. k (or
) is called the power spectrum.
Spectra:
Energy spectrum: |X(ej )|2 vs.
Magnitude spectrum: |X(ej )| vs.
Magnitude spectrum in dB: 20 log10 {|X(ej )|} = 10 log10 {|X(ej )|2 } vs.
Phase spectrum: X(ej ) vs.
Power spectrum: |ak |2 vs. or k
Example 5.26: Determine the percentage of energy of the signal x[n] = 5
over the frequency band .
3
3
Solution:
sin( n)
2
n
330
Example 5.27: Consider the signal x[n] =

k= [n 5k] which is the input
to a DT LTI system with impulse response h[n] = sin(0.5n) . What percentage
n
of the input power is present in the output?
Solution:
5.5.7
Multiplication
In Subsection 5.5.5 we saw that, for the DTFT, convolution in the time domain corresponds
to multiplication in the frequency domain. It turns out to be more general than this. For
either the CTFT or the DTFT, a convolution in one domain corresponds to a multiplication
in the other domain. Any specic example of this could be proved by paralleling the proof
in Subsection 5.5.5.
The multiplication property for the DTFT is
y[n] = x1 [n] x2 [n]
Y (ej ) =
1
2
<2>
X1 (ej ) X2 (ej() ) d .
(36)
This is a convolution, of X1 (ej ) and X2 (ej ), in the frequency domain. For each , as
a function of dummy variable one is folded and shifted by , then producted with the
other, and the result integrated. There is a new twist here (compared to the convolution sum
and integral covered in Chapter 2). When folding and shifting the one signal (in frequency),
you must remember that in frequency it is of innite duration and periodic with period
2. Also note that the integral is over one period only. This type of convolution, which is
applicable to periodic functions, is called periodic or circular convolution.
331
2
Example 5.28: Determine the DTFT of sin (W2n) where 0 < W <
(n)
derive the DTFT pair given in Example 5.25.
.
2
That is,
Solution:
Example 5.29: Determine the DTFT of

y[n] = sinc(.2n) sinc(.3n) .
sin(.2n)
.2n
Solution: Let x[n] = sinc(.2n) =
|| .5
.2 < ||
5
0
X(ej ) =
and v[n] = sinc(.3n) =

V (ej ) =
X(e
10/3
0
|| .3
.3 < ||
....
sin(.3n)
.
.3n
....
.2
.2
V(ej() ); = 0
10/3
1.7
.3
.3
1.7
For any shift < <

this image does not overlap
j
nonzero X(e )
over the range < <
For any shift < <

this image does not overlap
j
nonzero X(e )
over the range < <
Y(e
10/3
....
....
.5
.1 .1
.5
Figure 114: Frequency domain circular convolution for Example 5.29.
332
Example 5.30: Determine the DTFT of
y[n] = sinc(.5n) sinc(.6n) .
Solution: Let x[n] = sinc(.5n) =
X(ej ) =
sin(.5n)
.5n
|| .5
.5 < ||
2
0
sin(.6n)
.
.6n
and v[n] = sinc(.6n) =
10/6
0
V (ej ) =
|| .6
.6 < ||
X(e )
....
....
.5
V(e
j()
.5
); = 0
10/6
1.4
.6
For some shifts < <

this image does overlap
j
nonzero X(e )
over the range < <
10/6
....
.9
Y(e
.6
1.4
For some shifts < <

this image does overlap
j
nonzero X(e )
over the range < <
....
2/6
.1 .1
.9

Example 5.31: Circularly convolve the following two functions of .
X(ej ) =
k=
Solution:
p/10 ( k2)
V (ej ) =
k=
2 (
k2)
2

5.5.8
333
Modulation
Modulation is a special case of multiplication where one of the signals is a real-valued or

complex-valued sinusoid. For example, let x[n] be a signal with with DTFT X(ej ).
x[n]ejo n
1
2
<2>
X(ej ) 2 ( o )) d = X(ej(0 ) )
|| . (37)
Similarly,
x[n] cos(o n)
1
2
X(ej(0 ) + X(ej(+0 ) )
|| ,
(38)
x[n] cos(o n)
1
2j
X(ej(0 ) X(ej(+0 ) )
|| .
(39)
and
Example 5.33: Determine the frequency response of a FIR with impulse response
1
h[n] = N ej0 n pN [n].
Solution:
334
Example 5.34: Determine the DTFT of v[n] = x[n] w[n] where

x[n] = A1 ej1 n + A2 ej2 n and w[n] = pN [n]. Assume < 1 , 2 < .
Solution: From Table 13 entry #6, plus the DTFT linearity and modulation
properties, we have
j
j(N 1)/2
W (e ) = e
sin
sin
N
2
0 ||
X(ej ) = A1 2 ( 1 ) + A2 2 ( 2 )
0 ||
The exact solution will depend on N, A1 , 1 , A2 and 2 . The gure below shows
a visualization of the result.
W(ej )
21
X(ej() ); = 0
2 2 2
superposition of these
Y(e j )
2 2
21
range of interest
335
Table 14: Discrete Time Fourier Transform (DTFT) Properties.

Property
Time Domain
Frequency Domain
Periodicity
x[n]
X(ej ) = X(ej(+2) )
Symmetry
real-valued x[n]
X(ej ) = X (ej )
Delay
x[n k]
Linearity
a1 x1 [n] + a2 x2 [n]
a1 X1 (ej ) + a2 X2 (ej )
Convolution
x[n] h[n]
X(ej ) H(ej )
Parsevals Theorem
E=
|x[n]|2
X(ej ) ejk = |X(ej )| ej[
E=
n=
Parsevals Theorem
P =
1
N
N 1
n=0
|x[n]|2
Multiplication
x[n] w[n]
Modulation
x[n] ej0 n
1
2
<2>
P =
|X(ej )|2 d
N 1
k=0
1
2
<2>
X(ej )k]
|ak |2
X(ej )W (ej() )d
X(ej(0 ) )
336
5.6
The z-Transform
The z-transform is the last transform considered in these Course Notes. It appears in our
Table 5 of Transforms as the column 1, row 3 Entry. As such, it is used to represent
basically any DT signal as linear combinations of DT complex-valued exponential signals.
The z-transform is the DT analogy of the Laplace transform covered in Chapter 3. As such,
it has the following attributes:
generally it is very useful for the representation and analysis of DT LTI systems;
it leads to the transfer function representation of DT LTI systems;
its bilateral version, also called the two-sided z-transform, can be considered as a
generalization of the DTFT, and thus the DT LTI system transfer function is closely
related to the system frequency response;
its unilateral version, also known as the one-sided z-transform, can be used to handle
DT LTI system initial conditions.
Below, we will focus on the bilateral z-transform.
In engineering, the principal utility of the z-transform is for DT LTI system analysis and
design. We will see that the z-transform is essentially a generalization of the DTFT. As
such, it provides us with the frequency response of a DT LTI system. Additionally it is
useful in the study of: system stability, transient & steady-state response behavior, lter
implementation structures, and many other characteristics of DT LTI systems.
5.6.1
The Bilateral z-Transform and its Region of Convergence
Let z = rej be a complex variable. The z-transform of a DT signal x[n] is

X(z) =
x[n] z n
(40)
n=
This is the z-transform analysis equation. For a given signal x[n], the Region Of Convergence
(ROC), denoted Rx , is the region of z for which the function X(z) exists (i.e. all the values
of z such that the z-transform summation can be performed).
The inverse z-transform is
x[n] =
1
2j
X(z) z n1 dz
(41)
where the integral is a contour integral over a closed contour in the z-plane which encompasses
the origin (z = 0) and is in the ROC17 . Eq (41) is the z-transform synthesis equation. We
represent a z-transform pair as
x[n]
17
X(z) ;
Rx .
(42)
Contour integration in the complex plane is a typical topic in a course on Complex Variables. You will
not be responsible for contour integration in this Course. We typically use tables (e.g. Table 15 below) and
properties (e.g. Subsection 5.7.2 below) to perform inverse transforms.
337
It is important to remember what inverse transform equations suggest. The inverse ztransform indicates that a signal x[n] can be represented as a linear combination the basic
signals z n1 for z along a closed contour C. The weighting function for the linear combination is the z-transform function X(z). Since C can be any closed contour in the region
of convergence which surrounds to origin, the synthesis equation suggests that there are
an innite number of ways to represent x[n] as a linear combination of DT complex-valued
exponential functions.
Example 5.35: Find the z-transform, including the ROC, for x[n] = [n n0 ]
where n0 is an integer delay. This Example derives Entries #1,2 of Table 15.
Solution:
Example 5.36: Determine the inverse z-transform of the function

X(z) = 1 + 2z 1 + 3z 2 + z 3 .
Solution: Comparing this X(z) to the z-transform expression
X(z) =
n=
x[n] z n =
+ x[2]z 2 + x[1]z 1 + x[0]z 0 + x[1]z 1 +
we can pick out the nonzero values of x[n] as the coecients of the polynomial
(in z 1 ) given in this problem. We see that there are nonzero values of x[n] only
for n = 0, 1, 2, 3. We get,
x[n] = [n] + 2[n 1] + 3[n 2] + [n 3] .
The ROC for this signal will be Rx = z.
Example 5.37: Determine the z-transform, including the ROC, of x[n] = u[nn0 ]
where n0 is an integer delay.
Solution:
338
Example 5.38: Determine the z-transform, including the ROC, of x[n] = an u[n].
Solution:
Example 5.39: Determine the z-transform, including the ROC, of

x[n] = an u[n 1]. This Example derives Entry #4 of Table 15.
Solution:
Note that the signals in Examples 5.38,5.39 have the same z-transform function. What
dierentiates them is their ROCs. This is reminiscent of the bilateral Laplace transform.
In these z-transform examples the ROCs are, respectively, the exterior and the interior of a
circle in the z-plane.
339
Example 5.40: For the z-transform, determine h[n].

H(z) =
2
1
3 1
z
4
1 2
z
8
Rh = |z| >
1
.
2
Solution: Although there are some rational functions of z 1 in Table 15, none
match this H(z). Using PFE on H(z), for which we will describe a detailed
procedure a little later, we have
4
H(z) =
1 1
z
2
2
.
1
1 4 z 1
(Check this by combining the two terms using the common denominator). Anticipating the linearity property of the z-transform, and recognizing that since
the ROC is the exterior of a circle the inverse will be right-sided, we have
1
2
h[n] = 4
u[n] 2
1
4
u[n]
If h[n] is the impulse response of a DT LTI system, the system would be both
causal (h[n] = 0 for n < 0) and stable (h[n] decays as n so since it is a sum
of exponentials it is absolutely summable).
1
Example 5.41: Same as Example 5.40, but with Rh = |z| < 4 .
Solution: As opposed to the Example 5.40 solution, we now have that h[n] will
be left-sided since the ROC is the interior of a circle. Thus,
h[n] = 4
1
2
u[n 1] + 2
1
4
u[n 1]
This h[n] is neither causal nor stable.
Example 5.42: Same as Example 5.40, but with Rh =
1
4
1
< |z| < 2 .
Solution: As opposed to the Example 5.40 solution, we now have that the 1-st
term of H(z), with the denominator root z = 1 , will be contribute a left-sided
2
1
term to h[n]. The 2-nd term, with the denominator root z = 4 , will be contribute
a right-sided term to h[n]. So
h[n] = 4
1
2
u[n 1] 2
This h[n] is neither causal nor stable.
1
4
u[n]
340
As noted in the Example 5.40 solution, in solving Examples 5.40-42 we are anticipating
the linearity property of the z-transform. So, for any of these problems, we see that the
z-transform function is the sum of the z-transforms of the individual additive components of
the signal. Also note that the ROC is the intersection of the ROCs of the individual additive
components of the signal. This makes sense since it would seem that to take the z-transform
of a sum of signals you would have to be able to take the z-transform of each component
of that signal separately. The region of z for which you can take the z-transform of every
component is the intersection of the individual ROCs.
Table 15 is a list of some commonly used bilateral z-transform pairs. Note that in this
Table all z-transform functions, i.e. all the X(z), are rational functions of z 1 . Since our
primary interest with the z-transform is DT LTI system analysis and design, this suggests
that rational z-transform functions are most commonly encountered when dealing with such
systems. This is true, as we will see in Chapter 6. Also note that in Table 15 the z-transform
is described by both the function X(z) and the ROC Rx .
Table 15: z-Transform Pairs.
Signal
z-Transform
z
[n]
1;
[n n0 ]
z n0 ;
an u[n]
1
1 a z 1
|z| > |a|
an u[n 1]
1
1 a z 1
|z| < |a|
n an u[n]
a z 1
(1 a z 1 )2
|z| > |a|
n an u[n 1]
a z 1
(1 a z 1 )2
|z| < |a|
r n cos(0 n) u[n]
1 [r cos(0 )] z 1
1 [2r cos(0 )] z 1 + r 2 z 2
|z| > r
r n sin(0 n) u[n]
[r sin(0 )] z 1
1 [2r cos(0 )] z 1 + r 2 z 2
|z| > r
r n cos(0 n) u[n 1]
1 [r cos(0 )] z 1
1 [2r cos(0 )] z 1 + r 2 z 2
|z| < r
10
r n sin(0 n) u[n 1]
[r sin(0 )] z 1
1 [2r cos(0 )] z 1 + r 2 z 2
|z| < r

5.6.2
341
Selected Properties of the z-Transform
As pointed out earlier in the Course, there are many similarities between the properties of
dierent transforms. Because of this you can anticipate that since there are many properties
for any other transform we have considered, there will be numerous z-transform properties.
If you go through a table of properties for some other transform which we covered in some
depth earlier, say Table 14 in this Chapter on DTFT properties, you can anticipate that for
almost every entry you can expect that there will be an analogous z-transform property. So,
with your experience with properties of other transforms, you eectively already know a lot
about z-transform properties. That said, because of time constraints, in this Subsection we
describe on only those z-transform properties which are most pertinent to the study of DT
LTI systems. Let
h1 [n]
H1 (z) ;
Rh1
(43)
H2 (z) ;
h2 [n]
H1 (z) z n0 ;
Rh2
(44)
be two z-transform pairs. The following z-transform properties are most relevant to DT LTI
system analysis and design.
Time Delay: Given
h1 [n n0 ]
Rh1
(45)
Linearity:
c1 h1 [n] + c2 h2 [n]
ROC includes Rh1 Rh2 (46)
c1 H1 (z) + c2 H2 (z) ;
Convolution:
h1 [n] h2 [n]
H1 (z) H2 (z) ;
ROC includes Rh1 Rh2
Table 16: z-Transform Properties.

Property
Time Domain
z Domain
ROC
Delay
x[n n0 ]
z n0 X(z)
Rx
Linearity
a1 x1 [n] + a2 x2 [n]
a1 X1 (z) + a2 X2 (z)
includes Rx1 Rx2
Convolution
x[n] h[n]
X(z) H(z)
includes Rx Rh
(47)
342
Consider Examples 5.40-42 as illustrations of the linearity property.

Example 5.43: Determine the causal inverse z-transform of
b + c z 1
.
1 + d z 1
X(z) =
Solution:
b
c z 1
X(z) =
+
.
1 + d z 1
1 + d z 1
In the 2-nd term, the z 1 in the numerator represents a delay of one sample.
Using the linearity property and Entry #3 of Table 15 (since we are looking for
a causal solution), we have that
x[n] = b (d)n u[n] + c (d)n1 u[n 1] .
Example 5.44: Given the input
x[n] = 6
1
2
n1
u[n]
and output
y[n] = 4
1
2
u[n] 2
1
4
u[n]
to a causal DT LTI system, determine the system impulse response.

Solution:
This is an example of an important signal processing function, systems identication.

A note on ROC: The ROC will always be an annular ring in the z-plane. Let the inner
and outer radii of this ring be denoted as rin and rout , respectively. A right-sided signal
will have rout = . A left-sided signal will have rin = 0. A nite duration signal will have
rin = 0 and rout = .
343
Example 5.45: Determine the causal inverse z-transform of

H(z) =
Solution:
4z 2
2z 1
2z 1 + 5z 2 z 3
=
3
1
1 4 z 1 + 1 z 2
1 1 z 1
1 4 z 1
8
2
344
5.7
Problems
Chapter Topics:
5.1-16 (frequency response of simple DT LTI systems);
5.17-24 (DTFS and DTFT);
5.25-37 (DTFT properties);
5.38-41 (basic z-transform and properties)
1. Consider the pure delay system
y[n] = x[n n0 ] .
(a) Determine its frequency, magnitude and phase responses.
(b) For n0 = 10, determine the response to x[n] = 4 ej(4/10n) .
2. Consider the simple rst-dierence system y[n] = x[n] x[n 1], which has impulse
response h[n] = [n] [n 1].
(a) Determine the frequency response H(ej ) and sketch its magnitude over the range
.
(b) Determine the response y[n] to input x[n] = 3 + 5 cos(n).

3. Consider a DT LTI system with input/output equation
y[n] = x[n + 1] x[n 1] .
Determine the frequency response H(ej ) and plot its magnitude (or magnitude squared)
for . Determine the output for input
x[n] = 5 + 2 sin(n) .
4. Consider the DT LTI system with dierence equation y[n] = 4 x[n + 1] 4 x[n 1],
which has impulse response h[n] = 4 [n + 1] 4 [n 1].
(a) Determine the frequency response H(ej ) and sketch its magnitude over the range
.
(b) Determine the response y[n] to input x[n] = 3 + 6 cos( n) + 5 cos(n).

2
5. Consider the DT LTI system
y[n] = x[n] x[n 2] .
Determine its frequency response H(ej ). Determine the output for input
x[n] = 1 + 2 ej 2 n + 3 ejn .
Determine and plot |H(ej )|2 .
345
6. For the DT LTI system with impulse response h[n] = [n] + [n 2], determine the
frequency response H(ej ) and sketch its magnitude over the range .
y[n] = 4 x[n + 1] + 4 x[n 1] .
Determine the frequency response H(ej ) and plot its magnitude (or magnitude squared)
for . Determine the output for input
x[n] = 3 + 5 cos(n) .
y[n] = 2x[n + 1] + 3x[n] + 2x[n 1] .
(a) Determine its frequency response H(ej ).
(b) Plot |H(ej )| for .
(c) Determine the response y[n] for input x[n] = 2 +
1
2
cos( n) + 4 cos(n).
2
9. Consider the simple system

y[n] = x[n] 0.5 x[n 1] .
(a) Determine its frequency response, and plots its magnitude and phase.
(b) Determine the response to input
x[n] = 4 + cos(n) .
10. Frequency response of a DT LTI system:
(a) For a DT LTI system, given input x[n] = cos( n) and output y[n] = 2 cos( n+ ),
2
2
3
determine H(ej/2), the frequency response evaluated at = .
2
(b) You know that a DT LTI system has a real-valued impulse response, and for input
x[n] = ejn/2 the output is y[n] = 3 ej(n/2 /3) . Determine the output due
to input x1 [n] = 5 cos(n/2).
11. Consider a 10-point averager. Determine its frequency response, its magnitude response, its phase response, and its response to input
x[n] = 4 + cos
n+
10
2
+ cos
n+
10
12. Consider an 8 point averager which is a LTI system with impulse response
1
h[n] = 8 p8 [n]. Remember that the output of a LTI system, when the input is a
sinusoid, is a sinusoid of the same frequency. Determine the output sinusoid magnitude
and phase, and write down the equation for the output, for the following inputs:
346
(a) x1 [n] = 5
(b) x2 [n] = ej3n/8
(c) x3 [n] = 4 cos(n/4) .
13. For a DT LTI system, the frequency response is
H(ej ) =
0
+ 0
Sketch H(ej ). Determine the response y[n] to input x[n] = 3 + 5 cos( n).
2
14. Determine the frequency response for the causal DT LTI system
y[n] = x[n] + 0.5 y[n 1] .
Determine its magnitude and plot its magnitude-squared responses.
15. Sinusoidal inputs to DT LTI systems:
(a) Consider the DT LTI system with impulse response h[n] = .5n u[n]. The input
is x[n] = 2 ejn . The output is of the form y[n] = A ej(0 n+) . Determine A, 0
and . Justify your answer.
(b) Consider the DT LTI system and input
y[n] = x[n + 2] + x[n 2] ,
x[n] = 3 cos( n + ) = 5 cos( n + ) .

4
6
2
7
Plot the magnitude & phase of the frequency response. Determine output y[n].
16. Consider a noncausal DT LTI system with impulse response h[n] = .5|n| .
(a) Determine the frequency response H(ej ). Sketch its magnitude and its phase
over .
(b) Determine the output y1 [n] for input x1 [n] = ej0 n with 0 = 0.
(c) Determine the output y2 [n] for input x2 [n] = 3 cos(n).
17. Determine the DTFS coecients of x[n] = sin
2
n
3
cos
n
2
18. Consider a DT signal x[n] which is periodic with period N = 10 and has discrete-time
Fourier series coecients X[k] = .9k ; k = 0, 1, , 9.
(a) Describe the DTFT of x[n] over the range .
(b) Determine the power of x[n].
(c) If x[n] is the input to a DT LTI ideal lowpass lter with frequency response
H(ej ) = p (); , what is the power of the lter output y[n]?
19. Given the DTFS coecients shown below, determine the signal x[n].
347
a
x
k
....
....
.5
.25
3
20. Determine the DTFT or inverse DTFT for each of the following.
a) w[n] = p64 [n]. Also determine W (ej0 ).
b) X(ej ) = p () + ( /2); || .
c) y[n] = (0.5)n cos((/2)n) u[n].
d) z[n] = n (0.2)n u[n] (hint: use linearity and two table entries).
21. Determine the IDTFT of the following functions:
(a) X1 (ej ) =
|| 3
4
4
3
||
4
1
0
0 ||
(b) X2 (ej ) = 1 + 3ej + 2ej2 4ej3 + ej10 .

(c) X3 (ej ) = ej/2 for .
22. Consider a DT signal x[n] with DTFT X(ej ) = 3p () + 2p2 (); .

Sketch X(ej ) for 3 3 and determine x[n].
23. Given the DTFT
Z(ej ) =
0.5
0.5
+
j/3 ej
1 0.9 e
1 0.9 ej/3 ej
(48)
determine z[n] expicitely as a real-valued signal. What entry in Table 13 does this
correspond to?
24. DTFT/IDTFT Problems:
(a) Given the DTFT
X(ej ) =
2
ej2
2 + ej + ej2 0.5ej3
=
+
1 0.25ej2
1 0.5 ej
1 + 0.5 ej
,
(49)
determine x[n].
(b) Given
y[n] = 2 (0.9)n u[n] (0.8)n1 u[n 1] ,
determine its DTFT Y (ej ) and express it as a rational function of ej .
(50)
348
25. Determine the DTFT V (ej ) of the signal v[n] = x[n] w[n] where x[n] =
w[n] =
1
n
sin
3n
5
1
n
sin
n
3
and
.
sin(n)
.
n
26. Sketch the DTFT of x4 [n] = ej(2/10)n 2
27. Solve the following using the DTFT pair table and DTFT properties:
(a) Determine the DTFT Y (ej ) of the signal y[n] = an u[n 3].
(b) Determine the DTFT Z(ej ) of the signal z[n] = .5|n2| .

28. Determine the energy of x3 [n] for
1 + ej
X3 (e ) =
.
1 0.3ej
j
29. DTFT
(a) Using the DTFT pairs Table, determine the convolution y[n] = x[n] h[n] where
h[n] = .9n u[n] and x[n] = (n + 1) (.9)n u[n].
(b) Determine the energy of the signal s[n] =
sin( n)
2
n
sin( n)
3
.
n
30. Consider the DT signal x[n] with DTFT

X(ej ) =
0
0
(a) Plot X(ej ) over the range .
(b) Determine x[n].
(c) Determine the percentage of x[n] energy in the frequency band .

2
2
31. Given a DT signal x[n] with DTFT
X(ej ) = 1 + p/2 ()
determine the percentage of the total energy in the frequency band 3

4
32. Determine the DTFT V (ej ) of the signal v[n] = x[n] w[n] where x[n] = cos
w[n] =
1
n
sin
n
4
3
.
4
n
4
and
33. Consider a signal x[n]. Say you know it is real-valued, and that its DTFT is 1 over the
range 0 and 0 over the range < .
2
2
(a) Accurately sketch X(ej ) over the range 2 2.
(b) Determine x[n].
(c) Determine and sketch Y (ej ); for y[n] = x[n] ej(/4)n .
349
(d) For the y[n] in part (c), determine the percentage of energy in the frequency band
.
4
4
34. For each signal described below, state whether it is a power or energy signal. Determine
its power or energy.
(a) X1 (ej ) = 6 cos()
(b) x2 [n] = 3
sin( n)
4
n
(c) x3 [n] = 3 cos(n) 2
(d) X4 (ej ) =
1
1 .5 ej
.5 ej
1 .5 ej
35. DTFT pairs and properties:

(a) Given
X(ej ) =
1 + 3 ej2
,
1 + .5 ej
determine x[n].
(b) Given
z[n] =
1
1
,
(1 + n2 )
determine the fraction of its energy in the band .

2
2
sin2 (n/6)
sin(2n/3)
(c) Sketch the DTFT of the signal x[n] = 4
24 (n)2 . Determine
n
the percentage of its total energy in the band 3 2 .

3
(d) Given a signal x[n] with DTFT

X(ej ) =
1
0
0
2
otherwise over
use an entry in the DTFT table and the DTFT modulation property to determine
x[n]. What percentage of the total energy of x[n] is in the frequency band
0 10 .
36. Determine the DTFT and energy of the following signals:
(a) a[n] =
sin((/8)n)
;
n
(b) b[n] = a[n] cos((/2)n);

(c) c[n] = b[n] (n 5);
(d) d[n] = c[n]
sin2 ((/8)n)
.
(n)2
37. Determine the DTFT (or IDTFT) and energy of the following signals:
350
a) A(ej ) =
1
;
1 0.9 ej
b) B(ej ) = A(ej ) 2 ej4 ;

c) c[n] = b[n] ej(/2)n ;
d) D(ej ) =
A(ej )
;
1 0.9 ej
e) E(ej ) = A(ej ) (1 0.9 ej ).

38. Determine the z-transform pair for:
(a) X(z) = 3 2 z 2 + 4 z 5 .
(b) Y (z) =
(c) V (z) =
3
. Assume y[n] is causal.
1 + 0.25 z 1
6 z
. Assume v[n] is causal.
z .6
(d) x2 [n] = 4 cos(n) u[n 1] (Hint: use Eulers identities and linearity).
(e) Use the delay property to determine the right-sided inverse z-transform of
1
H(z) = z 4 0.25 z 5 .
(f) A causal y[n] has z-transform Y (z) =
(not necessarily in that order).
1
.
(10.8z 1 )3
Determine y[n] and its DTFT
39. Find the z-transform, including the ROC, for x[n] = 0.4n u[n] + 1.25n u[n].
40. Using and Eulers identity, linearity and Entry #4 of Table 15 of the Course Notes,
determine the z-transform, including the ROC, of x[n] = 4n cos( 2 n + ) u[n 1].
6
2
41. Consider a DT signal x[n] with z-transform
X(z) =
3
3 z 1
4 7z 1 + 3.68z 2 0.128z 3
= 1 +
.
1 1.6z 1 + 0.8z 2 0.128z 3
1 0.8z 1
(1 0.4z 1 )2
Given that the ROC is Rx = .4 < |z| < .8, determine x[n].
42. Consider two signals x1 [n] = [n]+[n1]+[n2] and x2 [n] = [n][n1]+[n2],
along with x[n] = x1 [n] x2 [n].
(a) Determine the z-transforms X1 [z] and X2 [z].
(b) From X1 [z] and X2 [z], determine the z-transform, X(z), of x[n].
(c) From X[z], determine x[n].
43. Determine the inverse z-transform (or z-transform) and energy for the following:
a) A(z) =
1
;
1 0.9 z 1
5
b) B(z) = 2 A(z) z
Ra = |z| > 0.9;
; Rb = Ra ;
c) C(z) = A(z) ; Rc = |z| < 0.9;
d) D(z) = A(z) (1 0.9 z 1 )(1 + 0.9 z 1 ).
351
Applications of DT Transforms
1. In Sections 6.1 & 6.2 we then explore how DT transforms can be used to characterize,
analyze, implement and design DT LTI systems.
2. In Section 6.3 we use DT transforms to facilitate descriptions of several higher level
signal processing function.

Be familiar with the various DT LTI system descriptions discussed in this Chapter,
including: the impulse response, the frequency response, the transfer function and
system poles/zeros. Understand the relationships between these system descriptions.
Know how to solve DT LTI system I/O problems using dierent system descriptions.
Understand the relationship between dierent DT LTI system descriptions and stability/causality.
Be comfortable with how the DT transforms covered in this Chapter are applied to
the advanced topics discussed in Section 5.9.
352
6.1
The DTFT and DT LTI Systems
In this Section we employ a few DTFT properties to expand on our ability to work with DT
LTI systems. The fact that this Section is brief should not lead you to conclude that the
DTFT does not provide much benet in dealing with DT LTI systems. The facts are that:
1. At the beginning of Chapter 5 we spent a signicant amount of eort using the function
H(ej ), the DTFT of the impulse response, to identify DT LTI system outputs for
sinusoidal inputs. We just didnt call H(ej ) a DTFT back then.
2. In Section 6.2 we will consider, in depth, the application of both the DTFT and ztransform to the analysis of DT LTI systems.
Here we simply formalize and exemplify two concepts.
6.1.1
The Frequency Response
Figure 117 illustrates the DTFT convolution property as applied to DT LTI systems.
x[n]
h[n]; H(e
X(e
y[n] = x[n] * h[n]
DT LTI system
j
Y(e
) = X(e
) H(e
Figure 117: The DTFT convolution property and DT LTI systems.

As in Section 5.1, H(ej ) is called the frequency response. We now recognize it as being the
DTFT of h[n]. So, the output DTFT is the product of the input DTFT and the frequency
response. The most important consequence of the fact that Y (ej ) = X(ej ) H(ej ) is that
it indicates that the frequency response characterizes the frequency selectivity of a DT LTI
system. To see this, consider some frequency 0 . X(ej0 ) is the weight on the 0 frequency
component of the input. Y (ej0 ) = X(ej0 ) H(ej0 ) is then weight on the 0 frequency
component of the output. If |H(ej0 )| << 1, then that frequency component of the input
is attenuated by the system. If |H(ej0 )| > 1, then that component is amplied. Similarly,
H(ej ) dictates input phase shift as a function of frequency.
Another use of the DTFT for system analysis is derived by simply rearranging the DTFT
I/O relationship as follows:
Y (ej )
X(ej )
Y (ej )
.
X(ej ) =
H(ej )
H(ej ) =
(1)
(2)
Back in Chapter 2 we developed a method, called convolution, for computing the output of
a DT LTI system given the input x[n] and the impulse response h[n]. However, determining
x[n] given y[n] and h[n], or h[n] given y[n] and x[n] was not addressed. In other words, we
could not do deconvolution. Well, now we can.
353
Finding H(ej ), or equivalently h[n], is called the system identication problem. It is an

important problem is many applications, ranging from cellular communications to RADAR
to geophysical exploration. Eq (1) suggest that, given the input and output of a DT LTI
system, we can use the DTFT to identify the system.
Finding X(ej ), or equivalently x[n], is called the communications problem. It is an
important problem is many applications, ranging from cellular communications to biomedical
engineering. Eq (2) suggest that, given the description and output of a DT LTI system, we
can use the DTFT to identify the system input. Referring back to Example 5.3, the DC
component of the input was completely attenuated while the = component was amplied.
We will consider these system identication and communications problems in some depth
in Section 6.2.
6.1.2
Bringing in the I/O Dierence Equation
In Chapter 1 of this Course we established that any DT LTI system can be represented by
a Linear Constant-Coecient (LCC) dierence equation of the form
N
ak y[n k] =
k=0
k=0
bk x[n k] .
(3)
Taking the DTFT of this equation, using the delay and linearity properties, we get
N
jk
ak e
bk ejk X(ej )
Y (e ) =
k=0
(4)
k=0
N
Y (ej )
ak ejk = X(ej )
k=0
Y (ej )
=
X(ej )
bk ejk
k=0
M
jk
k=0 bk e
N
jk
k=0 ak e
(5)
(6)
From the convolution property of the DTFT, then, we have

H(ej ) =
Y (ej )
=
X(ej )
M
k=0
N
k=0
bk ejk
.
ak ejk
(7)
So, the frequency response of a DT LTI system is a rational function of ej . Note from
Eq (4) that the frequency response H(ej ) can be written down, by inspection, from the
I/O LCC dierence equation the input side of the dierential equation determines the
numerator of H(ej ) and the output side determines the denominator. Conversely, the I/O
LCC dierential equation can be obtained by inspection from the transfer function.
If you have already covered Chapter 4, then you may recall that back in Section 4.1 of
these Notes we saw similar results that the frequency response of a CT LTI system could
be obtained by inspection from the I/O LCC dierential equation, and that this frequency
response was a rational function (in the CT case, it is rational in s = j). In general you
can expect transforms to be used in very similar ways for DT and CT systems.
354
Example 6.1: Consider the DT LTI system described by the following LCC difference equation.
y[n] 0.25y[n 1] + y[n 2] = x[n] + x[n 2] .
Determine the frequency response H(ej ) and the output due to input
x[n] = A cos((/2)n).
Solution:
355
Example 6.2: Given the following DT LTI system frequency response, nd the
impulse response and the LCC dierence equation.
H(ej ) =
2 0.5ej
.
1 0.9ej
Solution:
Example 6.3: Determine the impulse response for the system described by the
LCC dierence equation
y[n]
1
3
y[n 1] + y[n 2] = x[n] .
4
8
Solution:
From Example 6.3 it should be clear that it is important to be able to break any order
rational function of ej into a sum of lower order rational functions. As is also the case
for CT LTI systems, we can use partial fraction expansion to do this. We defer to later in
Section 6.2 of these Notes for coverage of the details for doing this.
356
6.2
The z-Transform and the LTI System Transfer Function
Consider the DT LTI system block diagram shown below.

z
H(z) z
DT LTI system
h[n]; H(z )
X(z)
Y(z) = X(z) H(z)
H(z) =
h[n] z n
(8)
n=
Figure 118: A DT LTI system and its transfer function.

By the convolution property of the z-transform, we have that Y (z) = X(z) H(z). The
z-transform of the system impulse response, H(z), impulse response, is called the transfer
function. It is also referred to as the system function.
Example 6.4: Determine the transfer function for the following FIR lter
y[n] = x[n] + 2 x[n 1] + x[n 2].
Solution:
Example 6.5: For a general FIR lter, with impulse response

M
h[n] =
k=0
h[k] [n k] =
k=0
bk [n k] ,
determine the I/O dierence equation and the transfer function.

Solution:
357
Now consider a DT LTI system represented generally by the Linear Constant-Coecient

(LCC) dierence equation
N
k=0
ak y[n k] =
bk x[n k] .
k=0
(9)
If we use the delay and linearity properties of the z-transform and take the z-transform of
this LCC dierence equation, we get
N
ak z k Y (z) =
k=0
bk z k X(z) .
(10)
k=0
Factoring out Y (z) on the left side and X(z) on the right, we have
N
Y (z)
ak z k = X(z)
bk z k .
(11)
k=0
k=0
From the convolution property of the z-transform in Table 16, the LTI system output has
z-transform
Y (z) = X(z) H(z) ,
(12)
so that the transfer function is
H(z) =
Y (z)
=
X(z)
M
k=0
N
k=0
bk z k
ak z k
(13)
As emphasized earlier, the transfer function H(z) can be obtained by inspection form the
system LCC dierence equation. The numerator polynomial roots, zi ; i = 1, 2, , M, are
the system zeros. The denominator polynomial roots, pi ; i = 1, 2, , N, are the system
b
poles. G = a0 is a gain term.
0
Since the DTFT and z-transform are closely related (i.e. the DTFT can be considered a
special case of the z-transform), the frequency response and transfer function of a DT LTI
system are closely related. Below, we rst explore this relationship.
Since H(z) is rational, we can nd the impulse response h[n] (the inverse z-transform) using
partial fraction expansion and standard z-transform pair table entries. We next describe how
this is accomplished.
Since the inverse z-transform of a rational H(z) is not unique, we need additional information to determine the impulse (e.g. the ROC; that the system is causal; that the
system is stable). To close, we discuss the relationship between the system poles/zeros and
causality/stability. We also look a DT LTI system block interconnections.
358
Consider again a DT LTI system represented by the Linear Constant-Coecient (LCC)

dierence equation
N
k=0
ak y[n k] =
k=0
bk x[n k] .
(14)
and corresponding transfer function

H(z) =
M
k=0
N
k=0
bk z k
= z N M
k
ak z
M
k=0
N
k=0
bk z M k
ak z N k
(15)
Factoring the numerator and denominator polynomials of the transfer function into rst
order terms, we see that H(z) can be expressed in terms its the numerator and denominator
roots as
M
M
1
k=1 (1 zk z )
k=1 (z zk )
N M
H(z) = G N
= Gz
,
(16)
N
1 )
k=1 (1 pk z
k=1 (z pk )
b
where G = a0 is a gain term. The zk ; k = 1, 2, , M, which are the numerator polynomial
0
roots, are termed the zeros of the system. The pk ; k = 1, 2, , N, which are the denominator
polynomial roots, are called the poles of the system.
The reason the zk are termed zeros is that H(zk ) = 0. If we were to plot H(z) vs.
z, we would see the plot go to zero at the zk . The reason the pk are termed poles is that
H(zk ) = . If we were to plot H(z) vs. z, we would see the plot go to innity at the pk .
That is, we would see innitely high poles in the plot at the pk . This is illustrated below
in Figure 119.
| H(z) |
Im {z}
11111
00000
11111
00000
11111
00000
11111
00000
111111
000000
1
11111
00000
p0
k
111111111111
000000000000
11111
00000
11111
00000
11111
00000
Re {z}
unit circle in the zplane
Figure 119: The eect of poles and zeros on the transfer function H(z).
For N > M we say that there are N M zeros at the origin (i.e. at z = 0). For M > N
we say that there are M N poles at the origin (i.e. at z = 0).

6.2.1
359
Frequency Response
Consider a general DT LTI system transfer function

H(z) =
h[n] z n
(17)
n=
Recall that z is a complex variable. H(z) exists for the ROC of the system. Let
z = ej ; , i.e. consider z on the unit circle of the z-plane. The frequency
response of the system, i.e. the DTFT of the impulse response, is
H(ej ) = H(z)|z=ej =
h[n] ejn .
(18)
n=
Note the assumption here that the ROC of h[n] includes the unit circle, so that H(z) exists
for z on the unit circle. So, generally the DTFT can be considered as the z-transform
evaluated on the unit circle, assuming that H(z) exists on the unit circle. Specically, the
frequency response of a DT LTI system is the transfer function evaluated on the unit circle.
Figure 120(a) below illustrates this important relationship. Figure 120(b) shows how any
one DT LTI system description can be obtained from another.
Im {z}
zplane
(a)
Re {z}
unit circle ( z = e j
LCC Diff. Eq.

inspection
H(e
inspection
z = e j
ztransform
DTFT
(b)
H(z)
h[n]
Figure 120: (a) evaluation of the z-plane on the unit circle (illustrating the DTFT/ztransform relationship); (b) relationships between DT LTI system descriptions.
360
Now, letting z = ej (i.e. evaluating the transfer function on the unit circle of the z-plane)
we get the frequency response18
M
k=0
N
k=0
bk ejk
= G
ak ejk
M
k=1 (1
N
k=1 (1
M
k=1
N
k=1
(ej zk )
.
(ej pk )
(19)
More generally, the z-transform of any signal, evaluated on the unit circle, is the DTFT
of the signal. This assumes that the ROC includes the unit circle. If not, then strictly
speaking19 , the DTFT does not exist.
Now consider the system magnitude and phase responses. First, for the magnitude response, noting that the magnitude of a product is the product of the magnitudes and the
magnitude of a ratio is the ratio of the magnitudes, we have that
H(ej ) =
zk ej )
= G ej(N M )
j )
pk e
|H(ej )| = |G|
M
k=1
N
k=1
|ej zk |
|ej pk |
(20)
Note that, for any value of , |ej zk | is the distance from the zero zk to the point ej
on the unit circle. So, as we evaluate the frequency response magnitude, that is as we travel
around the unit circle, when we get close to a zero zk , |ej zk | gets small, which tends to
drive the magnitude response down. Conversely, |ej pk | is the distance from the pole pk
to the point ej on the unit circle. As we travel around the unit circle, when we get close to
a pole pk , the magnitude response increases. The magnitude response is the multiplicative
contribution due to all the pole and zero distances to the unit circle.
For the phase response, noting that the angle of a product is the sum of the magnitudes
and the angle of a ratio is the dierence of the angles, we have that
M
H(ej ) =
G + (N M) +
k=1
(ej zk )
k=1
(ej pk )
(21)
These equations, which relate the magnitude and phase responses with the pole and zero locations, are very useful in understanding DT LTI systems. Below we focus on the magnitude
response.
Consider a DT LTI system which has a zero on the unit circle at zk = ej0 . At = 0 , the
k th numerator term of the magnitude response is zero, so |H(ej0 )| = 0. Thats why zeros
are called zeros. Conversely, with a pole on the unit circle at pk = ej0 , at = 0 , the k th
denominator term is zero, so |H(ej0 )| = (and the system is unstable). So, zeros on or
near the unit circle tend to pull down the magnitude response at frequencies corresponding to
the angles of the zeros. Conversely, poles near the unit circle tend to pull up the magnitude
response at frequencies corresponding to the angles of the poles.
As an example, refer to results of Practicum 5(b). Also, considering results of Practicum
5(a), we see that for CT LTI systems its the location of the poles and zeros relative to the
j axis that dictates characteristics of the magnitude response.
18
Since H(z) = n= h[n] z n , setting z = ej , we get H(ej ) =

DTFT of the impulse response, and thus the frequency response.
19
Not considering the generalized DTFT.
n=
h[n] ejn , which is the

Example 6.6: Given the LCC I/O dierence equation of a DT LTI system
y[n] + 0.81 y[n 2] = 5 x[n] 5 x[n 2] ,
determine H(z), the poles and zeros, and sketch |H(ej )|.
Solution:
361
362
Example 6.7: Find the poles/zeros of the 10-point averager with impulse response
h[n] =
1
10
9
k=0
[n k] .
Plot the pole/zero diagram. From this, roughly sketch the magnitude response.
Solution: We have seen this problem before. It is basically Example 5.2. In the
solution to that problem, we eectively took the DTFT of h[n] to determine the
following frequency response:
H(ej ) =
sin(5)
1 ej10
= ej(9/2)
.
j )
10 (1 e
10 sin(/2)
Substituting ej with z, or directly taking the z-transform of h[n], we get

1 z 10 1
1 z 10
= 9
.
H(z) =
10 (1 z 1 )
z 10 (z 1)
There is a pole at z = 1 and nine poles at z = 0. The zeros are the ten roots
of z 10 1, which are the ten roots of unity zk = ej(2/10)k ; k = 0, 1, , 9.
These zeros are uniformly distributed around the unit circle. The pole/zero plot
is shown in the rst gure below.
We can use the H(ej ) expression above to plot the magnitude of the frequency
response, as we did in Example 5.2. Alternatively, we can roughly sketch |H(ej )|
by considering the pole/zero placements relative to the unit circle in the z-plane.
We do this in the second gure below.
Im {z}
H(e
zplane
1
Re {z}
....
2/10 2/10
2/10
2/10
....
0
4/10
8/10
2/10
6/10
Figure 121: Pole/zero plot, and corresponding magnitude response, for Example 6.7.
Note that the nine poles at z = 0 have no eect on the shape of the magnitude
response since their distance to the unit circle is constant for varying . Note also
that the eect of the pole at z = 1 is canceled by the eect of the zero at z = 1.
This is called pole/zero cancellation the pole and zero cancel each other in the
rational H(z) expression. All the other zeros, uniformly distributed around the
unit circle, pull the magnitude response to zero at their corresponding angles.
363
Example 6:8: The gure below illustrates the relationship between magnitude
response and pole/zero location for two DT lowpass lters. These lters were
designed using Matlab. In Practicums 5a,b you will design and analyze DT lters
using a variety of Matlab function.
Chebyshev Type I Design
90 2
120
60
1
150
30
180
180
0
330
210
240
270
330
240
0
20
40
60
270
300
0
20
40
60
1
2
(radians/sample)
1
2
(radians/sample)
0.5
1
1.5
(radians/sample)
0
|H(ej )| (dB)
0
|H(ej )| (dB)
210
300
|H(ej )| (dB)
|H(ej )| (dB)
Chebyshev Type II Design

90 2
120
60
1
150
30
0.5
1
1.5
0.5
1
1.5
0.5
1
1.5
(radians/sample)
Figure 122: Pole/zero locations and frequency responses for two DT lowpass lters.
Example 6:9: The gure below illustrates this magnitude response, pole/zero
location relationship for a DT lter with transfer function
(1 z 1 ej0 )(1 z 1 ej0 )
(1 z 1 .9 ej0 )(1 z 1 .9 ej0 )
H(z) =
Note that for this lter two pole/zero pairs are positioned so as place notches
(a.k.a. nulls) in the magnitude response at = 0 .
Im
(a)
H (e
(b)
o
x
Re
x
o
Figure 123: Pole/zero plot, and corresponding magnitude response, for Example 6.9.
364
6.2.2
Partial Fraction Expansion (PFE) of Rational H(z)
From our discussion of the transfer function of a DT LTI system, and its relationship to the
systems LCC I/O dierence equation, we have come to realize that the transfer function is
always a rational function of z (or equivalently z 1 ). We are motivated to take the inverse
z-transfer function of the transfer function H(z) because this inverse is the system impulse
response h[n]. Thus we need to be able to take the inverse z-transform of any rational
function of z.
How do we take inverse z-transforms? Thus far we have relied on the use of z-transform
pair tables, and perhaps the linearity and time-delay properties. This approach, as it stands,
is applicable to only H(z) closely related to the few entries we have in our transform pair
table, Table 15. Alternatively, long division20 of a rational H(z) can provide the impulse
response, though as a general approach this is limited since it does not provide a closed
form expression for h[n]. Below, we introduce a general, systematic approach to determining
the inverse z-transform of any rational function of z. The approach is to break the rational
transfer function into a linear combination of simple rational terms that are in our table.
The method for breaking up the rational transfer function is called partial fraction expansion
(PFE). This discussion is a slightly modied version of the PFE discussion in Subsection 4.1.2
of these Notes. So if you have previously covered the application of the Laplace transform to
CT LTI systems, this will serve basically as a review, though you will see that the suggested
procedure is slightly dierent.
Here we study the PFE of a transfer function of the form
b0 + b1 z 1
H(z) =
a0 + a1 z 1
b0 z M
= z N M
a0 z N
+ + bM 1 z (M 1) + bM z M
+ + aN 1 z (N 1) + aN z N
+ b1 z M 1 + + bM 1 z + bM
.
+ a1 z N 1 + + aN 1 z + aN
(22)
Our objective is to decompose H(z) into a sum of lower order rational functions so that, using
linearity and z-transform tables, we can identify the inverse z-transform h[n] = Z 1 {H(z)}.
Note that if we wish to take the inverse DTFT of a frequency response
H(ej ) =
b0 + b1 ej + + bM 1 ej(M 1) + bM ejM
a0 + a1 ej + + aN 1 ej(N 1) + aN ejN
(23)
we can just replace the ej s with z 1 s and use the techniques described below. Several
techniques are required, depending on whether H(z) is proper (i.e. M N) or not, and on
whether or not there are repeated poles.
1. Proper transfer function (M N), with distinct poles: Let pi ; i = 1, 2, , N be the
system poles, which are assumed distinct. Then H(z) can be expressed as a weighted
sum of rst order rational terms as follows,
H(z)
=
z
20
N
i=1
ri
z pi
Long division of rational functions is illustrated below in Example 6.14.
(24)
365
where the ri , termed the residues, are computed as

ri = (z pi )
H(z)
z
(25)
z=pi
The reason we are expanding H(z) instead of H(z) is that the 1-st order terms of the
z
ri
expansion, the zpi are not in our z-transform table. However, multiplying Eq (24) by
z, we have
N
N
ri
zri
=
.
(26)
H(z) =
1
i=1 1 pi z
i=1 z pi
The
ri
1pi z 1
terms are in the Table 15.
Then, assuming for example a causal21 h[n], taking the inverse z-transform, we have
N
ri pn u[n] .
i
h[n] =
(27)
i=1
Regardless on whether or not H(z) is proper or not, and on whether there are repeated
roots or not, we will be able to expand H(z) as a linear combination of low order rational
functions. The coecients of the expansion, which we denote as the ri s, are called residues.
Each {pi , ri } pair corresponds to an exponential component of the inverse transform. It
may be causal or not, depending on the ROC. Note that there is no requirement that the
poles or residues be real-valued. If the ai and bi coecients of H(z) are real-valued, then
the poles and corresponding residues will either be real-valued or exist in complex conjugate
pairs. Any complex-valued poles corresponds to a complex exponential component (i.e. an
oscillating and perhaps decaying/expanding component) of the impulse response. We will
explore this in detail in an example below.
Partial fraction expansion (PFE) is not restricted to transfer functions. Any rational
function can be expressed this way. So PFE is a generally useful procedure in determining
the inverse transform (e.g. of input and output signals).
1
Recall that 1az1 has two inverse z-transforms, an u[n] and an u[n 1], depending of the ROC.
Here, we will assume an u[n]. That is, we will assume the casual inverse transform.
21
366
Example 6.10: Determine the impulse response of the causal DT LTI system
with LCC dierence equation
y[n]
7
1
5
y[n 1] +
y[n 2] = 3x[n] x[n 1] .
12
12
6
Solution:
H(z) =
5
3 6 z 1
3 5 z 1
6
=
1
7
1 12 z 1 + 12 z 2
(1 1 z 1 )(1
3
3z 5
H(z)
6
=
1
z
(z 3 )(z
r1 =
3z
z
5
6
1
4 z=(1/3)
1
)
4
= 2,
2z
z
1
3
h[n] = 2
z
z
1
3
r1
z
1
4
1
3
1
z
1
4
2
a
u[n] +
1 1
z
3
1
4
r2
z
1
3
3z
z
r2 =
2
H(z)
=
z
z
H(z) =
1 1
z )
4
z(3z 5 )
6
.
1
(z 3 )(z 1 )
4
1
4
5
6
1
3 z=(1/4)
= 1 .
1
1
1 1
z
4
u[n] .
Note that this system is stable, i.e. h[n] is absolutely summable because each
exponential component is decaying. From this example, can you anticipate the
characteristic of the poles that is required for a casual system to be stable? We
will address this issue later.
367
Example 6.11: Determine the impulse response of the causal DT LTI system with
LCC I/O dierence equation
y[n] 0.3 y[n 1] 0.4 y[n 2] = x[n] 2.1 x[n 1] .
Solution:
Matlab Demo - Matlab contains many functions which are useful for system
analysis. We will explore a number of these later, in Practicums 5a,b. For now,
try the following:
B = [1 -2.1];
A = [1 -0.3 -0.4];
Z = zeros(B);
P = zeros(A);
[r,p,k] = residue(B,A);
%
%
%
%
%
transfer function numerator polynomial coefficient vector

transfer function denominator polynomial coefficient vector
zeros
poles
residues, poles, k=[] since H(z)/z is strictly proper
This provides the PFE of the H(z)/z for this example.
368
Example 6.12: Determine the impulse response of the causal DT LTI system with
LCC I/O dierence equation
y[n] 0.8 y[n 1] + 0.64 y[n 2] = 4.0 x[n 1] .
Solution: Note that the two poles, p1 = 0.8 ej/3 , p2 = 0.8 ej/3 , are complex
conjugates. Their residues will be too. This is a necessary result of the coecients
of H(z) being real-valued (i.e. the coecients of the dierence equation are realvalued).
H(z) =
4z 1
4z
4z
= 2
=
.
1 + 0.64z 2
j/3 )(z 0.8 ej/3 )
1 0.8z
z 0.8z + 0.64
(z 0.8 e
H(z)
4
r1
r2
=
=
+
.
j/3 )(z 0.8 ej/3 )
j/3
z
(z 0.8 e
z 0.8 e
z 0.8 ej/3
4
H(z)
=
(z 0.8 ej/3 )
z
z 0.8 ej/3 z=(0.8ej/3 )
z=(0.8ej/3 )
4
4
=
=
= j 2.88675 2.89j = 2.89 ej/2 .
j/3 0.8 ej/3
0.8e
j 0.4 sin(/3)
r1 =
Similarly, r2 2.89 ej/2 .
H(z)
2.89 ej/2
2.89 ej/2
=
+
.
z
z 0.8ej/3
z 0.8ej/3
2.89 ej/2
2.89 ej/2
+
.
1 0.8ej/3 z 1
1 0.8ej/3 z 1
Note that the two additive terms in H(z) form a complex conjugate pair. Again,
this is the necessary result of the coecients of H(z) being real-valued.
H(z) =
h[n] = 2.89 ej/2
0.8ej/3
u[n] + 2.89 ej/2
0.8ej/3
u[n] ,
which is also composed of complex conjugate pairs. Combining these terms, using
an Eulers identity,
h[n] = 5.78 (0.8)n sin((/3)n) u[n] = 5.78 (0.8)n cos((/3)n (/2)) u[n] .
Note also that this is, basically, Entry #8 of Table 15.
Reecting on Example 6.12 note that since the coecients of H(z) are real-valued, the two
complex conjugate pair poles will contribute, to H(z), a term of the form
r
r
=
,
1 p z 1
1 p z 1
with corresponding impulse response component (assuming causality)
(r pn + r (p )n ) u[n] = |r| |p|n ej
ej
p n
+ ej
ej
= 2 |r| |p|n cos( p n + r) u[n] .
p n
(28)
(29)
So we need to nd the residue of just one of the poles, then we can simply write down the
impulse response component, Eq (29), due to both poles.
369
2. Proper transfer function (M < N), repeated poles: The following transfer function is
used to illustrate the general problem of repeated poles. It shows only a single repeated
pole, but in a multiple repeated pole case, each repeated pole is handled as illustrated
below.
N(z)
H(z)
=
z
(z p1 )(z p2 ) (z pm1 )(z pm )
m1
ri
rm,k
=
+
k
i=1 z pi
k=1 (z pm )
(30)
(31)
where = N m + 1,
ri = (z pi )
H(z)
z
i = 1, 2, , m 1
z=pi
(32)
and
H(z)
(k)
1
(z pm )
(k)
( k)! z
z
rm,k =
z=pm
k = 1, 2, ,
Example 6.13: Determine the partial fraction expansion of

X(z) =
5z 3 13z 2 + 10z
.
z 3 5z 2 + 8z 4
Solution: Note that p1 = 1, p2 = p3 = 2.

X(z)
5z 2 13z + 10
r1
r2,1
r2,2
= 3
=
+
+
z
z 5z 2 + 8z 4
z1
z2
(z 2)2
Which is strictly proper, so we can do a PFE. You can show that r1 = 2.
r2,2 =
r2,1
5z 2 13z + 10
z1
= 4
z=2
1 d 5z 2 13z + 10
=
1 dz
z1
z=2
10z 13
5z 13z + 10
+
z1
(z 1)2
=
z=2
7
4
= 3 .
1
1
So,
X(z) =
3
4 z 1
2
+
+
1 z 1
1 2z 1
(1 2z 1 )2
In taking the inverse transform of this, for the last term we can use either
Entry #5 or #6 of Table 15, depending on whether or not the inverse is
causal.
(33)
370
3. Non proper transfer function (M > N): Use long division as follows
H(z)
N(z)
R(z)
=
=
+ P (z)
z
D(z)
D(z)
(34)
where
P (z)
(35)
N(z)
.
.
.
D(z)
R(z)
order less than N
Example 6.14: Determine the inverse z-transform of

X(z) =
z 4 + 2z 2 4z
.
z 2 + 4z 2
Solution: We wish to expand

X(z)
z 3 + 2z 4
= 2
.
z
z + 4z 2
Since the numerator is of order greater than or equal to the denominator
(i,e, since M > N), we must rst perform long division until R(z) has order
less than D(z)
2
z + 4z + 2
z4
z3
+ 2z 4
3
2
z + 4z 2z
4z 2 + 4z 4
4z 2 16z + 8
20z 12
Thus,
X(z)
20z 12
= 2
+ z 4 ,
z
z + 4z 2
20 12z 1
+ z 2 4z .
1 + 4z 1 2z 2
The inverse z-transform is
20 12z 1
x([n] = Z 1
+ [n + 2] 4[n + 1] ,
1 + 4z 1 2z 2
X(z) =
where for inverse z-transform transform of the rational term, follow the M
N case solutions described earlier.

6.2.3
371
Stability, Causality, the Unit Circle and ROC
Consider, for discussion purposes, a proper H(z) with distinct poles pi ; i = 1, 2, N

N
H(z) =
i=1
Each
ri
1pi z 1
r1
r2
rN
ri
=
+
++
.
1
1
1
1 pi z
1 p1 z
1 p2 z
1 pN z 1
(36)
contributes either
pn u[n]
i
pn u[n 1]
i
ri
ri
(37)
(38)
to the inverse z-transform. The ROC indicates which:

If pi is inside the ROC, then Eq (37)
If pi is outside the ROC, then Eq (38).
These ROC rules also apply to the repeated poles and non strictly proper cases. Figure 124
illustrates this, where the poles inside the circle of radius 0.9 correspond to causal impulse
response components, and the poles outside the circle of radius 1.1 correspond to noncausal
components.
Im {z}
zplane
ROC
x
x
x
causal
poles
Re {z}
noncausal poles
0.9
1.1
Figure 124: Illustration of the relationship between pole positions, ROC and causality of
impulse response components.
Example 6.15: Given the system LCC dierence equation
y[n] + y[n 1] 1.75y[n 2] + 0.5y[n 3] = x[n 1] 2x[n 2] ,
Is the corresponding causal system stable? Is the stable system causal?
Solution:
372
ROC Properties: Let rh denote the maximum of the magnitudes of all the causal poles.
Let rl denote the minimum of the magnitudes of all the noncausal poles.
#1 In general the ROC is an annular ring: rh < |z| < rl .
#2 The ROC contains no poles.
#3 A nite duration x[n] has Rx = z.
#4 Right sided (causal) x[n] means Rx is the exterior of a circle (i.e. Rx = rh < |z| < ).
#5 Left sided x[n] means Rx is the interior of a circle (i.e. Rx = 0 |z| < rl ).
#6 Two sided x[n] means either Rx = rh < |z| < rl , or Rx = (the empty set) if rl rh .
Stability and the Unit Circle: Stated for distinct pole impulse response components,
but true for repeated pole components as well.
#1 Right sided signal:
h[n] = pn u[n]
1
1 p z 1
Rh = |z| > |p|
(39)
For stability, |p| < 1. The ROC must include the unit circle.
#2 Left sided signal:
h[n] = pn u[n 1]
1
1 p z 1
Rh = |z| < |p|
(40)
For stability, |p| > 1. The ROC must include the unit circle.
#3 Generally, for stability, the ROC must include the unit circle.
Causality and the ROC: h[n] must be right sided. So the ROC must be the exterior of a
circle.
Causality & Stability and the Unit Circle & ROC: The ROC must be the exterior of
a circle and include the unit circle. All poles must be inside the unit circle, i.e.
|pi | < 1
i = 1, 2, , N .
(41)
373
Example 6.16: Consider the transfer function

11 + 7z 1
.
1 + 1.5z 1 z 2
Determine all possible impulse responses, and for each comment on stability and
causality.
H(z) =
Solution: The gure below shows the dierent regions of the z-plane that could
be the ROC, depending on the causality/stability of the system.
Im {z}
zplane
.5
Re {z}
Figure 125: Pole/zero plot, and possible ROCs, for Example 6.16.
H(z) =
Try this PFE yourself.
6
5
+
.
1
1 0.5z
1 + 2z 1
Case 1: both poles are causal the ROC is |z| > 2
h[n] = 5 (.5)n u[n] + 6 (2)n u[n]
The 2-nd impulse response component is non stable.

Case 2: both poles are noncausal the ROC is |z| < .5
h[n] = 5 (.5)n u[n 1] 6 (2)n u[n 1]
The 1-st impulse response component is non stable.

Case 3: the z = 0.5 pole is causal, the z = 2 pole is noncausal the ROC is
0.5 < |z| < 2
h[n] = 5 (.5)n u[n] 6 (2)n u[n 1]
Both impulse response components are stable.

Case 4: the z = 0.5 pole is noncausal, the z = 2 pole is causal the ROC does
not exist
h[n] = 5 (.5)n u[n 1] 6 (2)n u[n]
Both impulse response components are non stable.
374
6.2.4
DT LTI System Block Diagrams
#1 Parallel subsystems: H(z) = H1 (z) + H2 (z) (z-transform linearity property)

H(ej ) = H1 (ej ) + H2 (ej ) (DTFT linearity property)
#2 Cascade subsystems: H(z) = H1 (z) H2 (z) (z-transform convolution property)
H(ej ) = H1 (ej ) H2 (ej ) (DTFT convolution property)
#3 Feedback subsystems:
Derivation:
Note that, in general, the PFE of a transfer function H(z) describes a parallel implementation
of the system. On the other hand, by factoring H(z) into a product of lower order terms,
using the pole/zero product factorizations of the numerator/denominator polynomials, we
can generate cascaded realizations of the system.
Example 6.17: Consider the causal DT LTI feedback system, where for the causal
feed forward subsystem
y[n] 2y[n 1] = v[n] ,
and for the feedback subsystem
H2 [z] = 1.5z 1
a) Is the feed forward subsystem H1 (z) stable? Why?

b) What is the overall system transfer function H(z)?
c) Is the overall system stable? Why?
Solution:
a) The feed forward subsystem has transfer function H1 (x) = 1 1 z 1 . It has
2
one pole, at p1 = 2. With this pole, since the system is causal, it is unstable.
1
b)
H1 (z)
1 2 z 1
H(z) =
=
1
1 + H1 (z) H2 (z)
1 + 1 1.5 2z z 1
1
1
=
=
.
1 2 z 1 + 1.5 z 1
1 0.5 z 1
c) The overall system has one pole, at p1 = 0.5. With this pole, since the system
is causal, it is stable.
375
Example 6.18: Consider two causal LTI subsystems connected in parallel. Let
their transfer functions be
H1 (z) =
1 + z 1
,
1
1 4 z 2
H2 (z) =
3
1
1 + 2 z 1
Determine the overall system transfer function. Determine the impulse response
of the overall system, both directly from the overall system transfer function and
by combining the individual impulse response results. Compare the two results.
Solution:
376
Example 6.19: Consider the cascade of two DT LTI systems. The rst has
transfer function H1 (z) = 1 (0.9)10 z 10 . This subsystem has zeros at
zk = 0.9 ej2k/10 ; k = 0, 1, , 9. The second has transfer function H2 (z) =
1
.
1+0.9z 1
1. Determine the impulse response of each subsystem.
2. Draw the pole/zero diagram for each subsystem, and for the cascade of the
two subsystems.
3. Determine the transfer function of the cascade of the two subsystems.
4. Roughly sketch the frequency response of the cascade of the two subsystems.
Solution:
6.3
6.3.1
377
Signal Processing Functions & Implementation

Channel Equalization
Back in Subsection 1.2.3 of these Course Notes we introduced the channel equalization, and
we have used this problem from time to time in this Course to demonstrate Course topics.
For example, in Practicum 1 we used this problem, and in particular discrete time models
of the digital communications channel and equalizer, as an example of simple DT systems.
In Chapter 2 of this Course, when introducing cascaded system interconnections, we saw
that the combined impulse response of a cascaded DT LTI channel and equalizer is the
convolution of the individual impulse responses of the channel and of the equalizer. The
realization makes it somewhat easy to evaluate the eectiveness of a particular equalizer at
compensating for the distortion of a given channel. However, until now we have not had a
tool that could be used to design an equalizer for a given channel. With the z-transform,
and the associated concept of transfer function, we now have an eective tool.
Figure 126 illustrates the digital communication channel equalization problem, where
Hc (z) and He (z) are the transfer functions of, respectively, the DT LTI channel and DTLTI
equalizer. Each channel input x[n] represents a symbol (that represents a set of bits). Each
channel output v[n] is our received version of x[n]. Because of multipath propagation, a
transmitted symbol appears at the channel output spread over time. A general model for a
LTI channel is the FIR lter model,
M
v[n] =
k=0
hc [k] x[n k] .
(42)
For many realistic digital communication system this channel model can be rigorously justied, and in cases where it con not be justied, it at least represents a good starting point
for a more realistic representation. Anyway, it makes sense, since it shows that at any
given symbol time n we receive a superposition of the transmitted symbol an some previous
symbols (which makes sense because of the channel multipath i.e. memory).
x[n]
Channel
Hc (z)
v[n]
Equalizer
He (z)
y[n]
Figure 126: DT LTI model of the digital communication channel equalization problem.
The equalizer objective to to make its output equal to the transmitted symbol, i.e. y[n] =
x[n], or at least y[n] = x[n m] for some integer m 0. The reason for allowing a delay of
m symbols is that there is inevitably a channel propagation delay. So, ideally, we would like
hc [n] he [n] = [n m], or equivalently, Hc (z) he (z) = z m . In the following examples
we explore this equalization problem.
378
Example 6.19: Consider the equalizer problem rst considered in Practicum 1.

1 2z 1
Let the channel transfer function be H(z) = 1 .5z 1 . Denote as G(z) the
equalizer transfer function.
a) Consider G(z) = H 1 (z). Show that this exactly equalizes the channel.
What is the problem with this solution?
1 .5z 1
b) Consider G(z) = 2 + z 1 . Identify a positive and a negative aspect of this
solution.
Solution:
Example 6.20: Two LTI subsystems, a channel and an equalizer, have impulse
responses hc [n] = [n] + 0.9[n 1] and he [n] = (.9)n u[n].
1. Identify, on a z-plane diagram, the locations of the poles/zeros of these two
individual subsystems.
2. Consider the cascade of these two subsystems. Identify the locations of the
resulting systems poles/zeros, and use these to roughly sketch the magnitude response |H(ej )|.
Solution:
379
Example 6.21: Consider two cascaded causal DT LTI subsystems, a channel and
an equalizer. The channel has dierence equation
v[n] = x[n 4] 0.25x[n 6]
where x[n] is the input and v[n] is the output. The equalizer has transfer function
z
.
He (z) =
(z 0.5)(z + 0.5)
1. Determine the impulse responses hc [n] and he [n] of the two subsystems.
2. Determine the impulse response of the cascade of these two subsystems.

Solution:
Example 6.22: Consider a DT LTI channel with transfer function Hc (z) =

1 0.81 z 2 . Consider the cascade of Hc (z) and a DT LTI equalizer He (z)
which exactly inverts the eect of the channel.
1. Design the equalizer. That is, determine its transfer function He (z).
2. Can this equalizer be realized as a causal/stable system? Why?
3. Starting with the PFE of He (z), determine the causal impulse response
he [n].
Solution:
380
Example 6.23: Consider a DT LTI communications channel with impulse response hc [n] = [n] + 2[n 1]. At the receiver, we use a DT LTI equalizer (as
in Practicum 1) to counteract the eect of the channel. Determine the transfer
function, He (z), of the ideal equalizer (i.e. which will have an output equal to
the channel input for any channel input). Assuming that this equalizer is causal,
what is its impulse response? Explain the problem with implementing this equalizer, and identify the characteristic of the channel poles/zeros that causes this
problem.
Solution:
Examples 6.20-6.23 are relatively simple channel equalization problems. Example 6.21
illustrates the typical need to design for delay. What do you think would happen if you
attempted to design the equalizer without allowing for a delay in the overall channel/equalizer
response? Try designing and evaluating without the delay. Example 6.23 suggests a very
important realistic consideration equalizer stability. In part because of stability concerns,
in practice channel equalizers are almost always FIR lters. Even for the simple channels
considered in the examples above, do you think an FIR equalizer can be eective?
Equalizer design starts with the understanding developed in the examples above. A senior
level DSP course would typically develop some more advanced design tools to apply to the
problem of designing an equalizer for a DT LTI channel. In a graduate digital communications course you would learn how to tackle this problem for situations when the channel
is unknown and time varying. Think about a cellular mobile phone application. Can you
expect the channel transfer function from the cell phone to a base station to be known? No.
Can you expect that channel transfer function to remain constant over time? No. Even so,
the solution to this more realistic channel equalization problem starts with the understanding
developed in this Subsection.

6.3.2
Sampling: A DT Perspective
6.3.3
Filter Banks
6.3.4
Spectrum Estimation
6.3.5
381
Real-Time DT Systems
Presentations by:
Brian McCarthy: chairman and founder, Adaptive Digital Technologies Inc., Plymouth
Meeting, PA; BSEE Villanova Un. 1978; VU ECE Dept. Adjunct Professor. Brian is
the instructor for ece7710, Real-Time Digital Signal Processing.
Sean Gallagher: Senior DSP Specialist, Xilinx; MSEE Villanova Un. 1992; VU ECE
Dept. Adjunct Professor. Sean is the instructor for ece7711, Hardware Digital Signal
Processing.
382
6.4
Practicum 4c
Spectrum Analysis of DT Energy Signals

Introduction
In this practicum you will investigate the DTFT. Specically, you will study the frequency
content of several DT signals. For each, you will numerically derive and plot spectra. In
eect, you will be designing, implementing and testing a digital spectrum analyzer (i.e.
a spectrum analyzer which computes spectrum estimates using digital hardware and/or
software). These days, the vast majority of spectrum analyzers are digital.
You will develop a spectrum analyzer for DT energy signals. You will rst test this
spectrum analyzer on a signal with known frequency content. You will then use it to analyze
an Acoustic Emissions (AE) signal. For each of these signals, you will draw conclusions from
its spectrum.
Background
The DTFT analysis equation, shown below, calculates the frequency content of an DT
energy signal:
X(ej ) =
x[n] ejn ;
< .
n=
(1)
We know that the range of is usually restricted to < because of the

ambiguity of discrete-time frequency. We also know that if x[n] is derived as samples of an
1
underlying CT signal xc (t), i.e. if x[n] = xc (nTs ) where fs = Ts is the sample rate, then
discrete-time frequency (in radians/sample) corresponds to continuous-time frequency
(in radians/second) as = Ts . Thus, if xc (t) is sampled without aliasing, then xc (t) is
bandlimited to the frequency range Ts < Ts . In Subsection 4.2.2 of the Course

Notes it is shown that, with no aliasing, the DTFT of x[n] is related to the Continuous-Time
Fourier Transform (CTFT) of xc (t) as
Xc (j) =
1
X(ejTs ) ;
Ts
<
,
Ts
Ts
(2)
where Xc (j), the CTFT of xc (t), is the content of xc (t) (i.e. similar to what X(ej ) is
for x[n]). By Parsevals theorem for the DTFT, |X(ej )|2 vs. is the energy spectrum of
X(ej ). Similarly,
|Xc (j)|2 =
1
|X(ejTs )|2 ;
Ts2
<
Ts
Ts
(3)
is the energy spectrum of xc (t) (again assuming no aliasing).

In application, there are two practical problems associated with computing the DTFT.
First, in general a signal x[n] has innite duration, so that the DTFT (Eq (1)) can not
be computed. In practice, only a nite duration of x[n] is processed, say the N samples
x[n]; n = N1 , N1 + 1, N1 + 2, , N2 where N = N2 N1 + 1. Then
X(ej ) =
N2
n=N1
x[n] ejn ;
<
(4)
383
is computed, which will in general be an approximation. The second practical problem is

that X(ej ) is a function of continuous . With a computer, we can compute this for only
a nite number of s, say for the M samples
k =
2W
k;
M
k=
M
M
+ 1, , 1, 0, 1, 2, ,
2
2
(5)
(in Eq (5) it is assumed that M is an even integer), where 0 < W denes the lowpass
range of you wish to observe. So, we compute only samples of X(ej ), which in turn are
j
in general approximate samples of the true X(e ).
In designing a spectrum analyzer for DT energy signals, or for an underlying CT signal from which the DT signal was derived, you must choose N1 , N2 , W and M. Your
implementation should have these as selectable analyzer parameters.
Procedures
1. Code up you spectrum analyzer keeping N1 , N2 , W and M as selectable parameters.
2. Select a signal x1 [n] with known DTFT X1 (ej ) (e.g. from a table of DTFT pairs),
keeping in mind that this is to be a test signal for your spectrum analyzer. Select
N1 , N2 , W and M. In Matlab, use the plot function to view |X1 (ej )|2 vs. for
W < W . Make sure you plot a dense enough sampling of |X1 (ej )|2 to get
an accurate representation of it.
Generate x1 [n]; n = N1 , N1 + 1, N1 + 2, , N2 and use the stem function to plot
it. Using your spectrum analyzer, compute the M samples of X1 (ej ). Use the Matlab
1 (ej )|2 . Compare your |X1 (ej )|2 and |X1 (ej )|2 plots.
plot function to view |X
You have now tested you spectrum analyzer.

3. The data le
v:/Electrical Computer/ece3240/AEdat
contains y[n] = yc (nTs ), a dense sampling of an AE signal yc (t) over its duration
0 t .8 msec.. This signal was measured using a small acoustic sensor attached to
a piece of metal as it was bent so as to cause cracking.
(a) What is the spacing Ts between samples. Plot yc (t) vs. time (in seconds).
(b) Use the spectrum analyzer you developed and tested in Procedure 2. to process
y[n]. What values of N1 and N2 should you use? Select and justify values of
W and M. (You may have to try several joint values.) Plot your estimate of
|Y (ej )|2 vs. and the corresponding estimate of |Yc (j)|2 vs. . Considering
your |Yc (j)|2 estimate plot, describe the frequency content of the signal yc (t).
Compare characteristics of this spectrum (e.g. frequency positions of spectral
peaks) to characteristics of the AE signal (e.g. the frequency of oscillation).
384
.
385
Practicum 4c
Instructor/TA Sign O Sheet
Students Name:
1. Procedure 2: Write down the equations for x1 [n] and corresponding X1 (ej ) you selected to test your analyzer. Explain why you choose this pair.
2. Procedure 2: List the values of N1 , N2 , W and M you selected for analyzing x1 [n].
Briey explain why you choose these values.
3. Procedure 2: Plot of |X1 (ej )|2 .
4. Procedure 2: Plots of x1 [n] and |X1 (ej )|2 .
5. Procedure 2: Comment on your comparison between your |X1 (ej )|2 and |X1 (ej )|2
plots, explaining any noticeable dierences. Explain why you are convinced that your
spectrum analyzer is working properly.
386
6. Procedure 3(a): What is Ts ?
Plot of yc (t).
Procedure 3(b): What are your selected values of N1 , N2 , W and M? Explain how
you arrived at your value of M.
7. Procedure 3(b): Plots of your estimates of |Y (ej )|2 and |Yc (j)|2 .
8. Procedure 3(b): Compare characteristics of this spectrum to characteristics of the AE
signal.
6.5
387
Practicum 5a
CT & DT Filter Design & Analysis

Introduction
In this practicum you will investigate CT & DT lter design and evaluation. Filter design
and analysis is easily accomplished with Matlab, which provides numerous CT and DT lter
design and analysis functions. In Matlab, type help signal to view a list that includes
available lter design options.
Procedures: Procedure 1 should be completed before the 1-st practicum session. Procedures 2-4, on CT lter design and evaluation, should be completed the 1-st practicum session.
Procedures 5-7, on DT lter design and evaluation, should be completed at the 2-nd session.
1. Before the practicum session, study the accompanying Matlab program. This program
designs and analyzes a 6-th order CT lowpass Butterworth lter with a lowpass cuto
frequency of 1 rad./sec.. It generates a pole/zero plot, as well as plots of magnitude
and phase responses. Note the use of the butter, roots, zplane, residue and freqs Matlab
functions. Using the Matlab help resource, become familiar with these. Implement this
program and consider the resulting plots.
2. Following the Butterworth lter example discussed above, design and analyze in a similar manner a Chebyshev Type II low pass lter of order 6, with stop band attenuation
of 40 dB and with cut o frequency o = 120. First design a low pass prototype that
has a cut o of 1 rad./sec. and the correct stop band attenuation using the command
[B,A]=cheby2(6,40,1,s). Write down the transfer function.
3. Now convert the prototype from Procedure 2. to have a cuto of o = 120 using
the lp2lp function (type help lp2lp for information on how to use this function). For
this 2-nd lter, provide a pole/zero plot, and frequency response magnitude and phase
plots. Write down the impulse response h(t) explicitly as a real-valued function.
4. Comment on the relationship between the lters pole and zero locations in the s-plane
and characteristics of its frequency response.
5. Using the Matlab ellip command, design a 6-th order (i.e. N = 6) discrete time
lowpass lter with a cuto frequency 0 = , passband peak-to-peak ripple of 0.5dB,
3
and stopband attenuation of 40dB. Write down its input/output dierence equation,
the transfer function, and the ROC (assuming the lter is causal).
6. Plot the poles and zeros using zplane, and use the Matlab freqz command to compute
and plot, over the range , the phase response and the magnitude response
in dB. (The magnitude response, in dB, is 20 log10 {|H(ej )|} = 10 log10 {|H(ej )|2 }.)
388
7. Generate 300 samples, for n = 0, 1, 2, , 299, of two seperate real-valued sinusoidal

signals, one with frequency in the passband and one in the stopband. What frequencies
did you choose? By analyzing the frequency response plots you generated in Procedure
6, record the magnitude and phase responses for the two frequencies you selected. Use
the Matlab lter function to process both of your sinusoidal signals with the lter
analyzed in Procedure 6. For each signal use Matlab to plot the input and output on
the same gure, and in steady state carefully compare the input and output sinusoid
amplitudes and phases to observe that systems eect on the input sinusoid. Do these
observed results match the frequency responses you recorded from the Procedure 6
plots? Why?
echo on
%************************************************************
%
%
Continuous-Time Filter Design & Evaluation
%
%
Butterworth lowpass: order N = 6,
%
cutoff wn = 1 radian
%
%************************************************************
%
%
Design Butterworth Filter w/ butter.m
%
(try cheby1, cheby2 and ellip)
%
%
B = vector of transfer function numerator coefficients
%
A = vector of transfer function denominator coefficients
N=6;
wn=1;
[B,A]=butter(N,wn,s)
pause
%
Plot Pole Locations using zplane
%
(no zeros for analog Butterworth lowpass)
%
p=roots(A);
zplane([],p)
pause
%
Partial Fraction Expansion of H(s) - You can determine
%
h(t) from results. P is the vector of poles.
%
R is the vector of respective residues.
%
K is the vector of long-division coefficients if the transfer
%
function is not strictly proper.
%
[R,P,K]=residue(B,A)
pause
%
Compute and Display the Frequency Response using freqs
%
freqs(B,A)
389
390
.
391
Practicum 5a
Instructor/TA Sign-O and Comment Sheet
Students Name:
1. Procedure 2: write down the prototype Chebychev lter transfer function
2. Procedure 3: pole/zero, Mag. and phase response plots

3. Procedure 3: write down the impulse response h(t) as a real-valued function
4. Procedure 4: relate the pole/zero locations to the frequency response plots
5. Procedure 5: write down the Elliptic lter dierence equation, transfer function and
ROC.
6. Procedure 6: pole/zero and mag./phase plots
7. Procedure 7: compare theoretical and observed magnitude/phase responses
392
.
6.6
393
Practicum 5b
DT Filters
Introduction
In Week 1 of this Practicum you will observe the frequency response of an FIR lter, by
analyzing the impulse and frequency responses and by observing their responses to sinusoidal
inputs and speech. You will learn how to evaluate the frequency response, rst directly by
computing the DTFT of the impulse response and then using the Matlab function freqz. You
will also learn to lter an input signal (specically a sinusoid and speech) with an FIR lter.
In previous Practicums you had done some ltering directly. Here you will learn to use the
Matlab function rlt for some simple FIR lters. (Note: the rlt.m le can be found in the
Course Folder .../electrical computer/ece3240.) You will then use a few specialized plotting
functions in the Course Folder to analyze and implement some more interesting FIR lters
operating on a speech signal
In Week 2 you will investigate discrete time lter design and evaluation. Filter design is
easily accomplished with Matlab, which provides numerous CT and DT lter design functions. In Matlab, type help signal to view a list that includes available lter design options.
Also browse help specically for several of the lter design functions.
Procedures:
Week 1: Do Procedure 1(a) before the Week 1 lab session.
1. A 3-pt. averager:
y[n] =
1
3
2
k=0
x[n k] .
(1)
(a) On paper, show that the frequency response of the 3-pt. averager is
H(ej ) =
2 cos() + 1 j
e
3
(2)
Sketch the magnitude of the frequency response. Determine the magnitude and
phase of the frequency response at = 0.25. Determine the lter transfer
function H[z], and the system poles & zeros.
394
(b) Use the Matlab function freqz to compute the frequency response. Specically,
implement the following Matlab code:
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
bb = (1/3)*ones(1,3);
ww = -pi:(pi/200):pi;
H = freqz(bb,1,ww);
subplot(211)
plot(ww,abs(H))
xlabel(normalized frequency (rad/sample))
ylabel(abs(H))
subplot(212)
plot(ww,angle(H))
xlabel(normalized frequency (rad/sample))
ylabel(angle(H))
Type help freqz to see how, in general, to use this function. (Note that magnitude
and phase are plotted. Plots of the real and imaginary parts of the frequency
response are not very interesting.)
(c) Compare the frequency response results derived in (a,b) above, both over
and specically at = 0.25.
(d) Generate 50 samples of the DT cosine signal x[n] = 7 cos(0.25n + (/3)); n =

0, 1, 49 and store them in the vector xx.
(e) Using the provided function rlt, generate the corresponding output yy. (Note
that xx and yy are not the same length.)
(f) Using subplot, plot the rst 50 values of xx and yy on the same gure. Use the
stem function and label the time axis 0 n 49. From the xx plot, measure
(and record) the input magnitude, frequency and phase. How were these values
identied from the plot?
(g) Except for the rst few values, yy appears to be a sinusoid. As in (f) above,
measure (and record) its magnitude, frequency and phase.
(h) Characterize the lters performance, for this input, by computing the magnitude
gain and phase shift using your input/output measurements from (f) & (g) above.
Compare these results to the frequency response of this lter derived in (a-b)
directly above, evaluated at = 0.25.
2. Filtering Speech Samples:

(a) Make sure you have access to les lab6dat.mat, inout.m and striplot.m which can
be found in the Course Folder. To understand the Matlab functions you will use,
type help lename for the lenames inout and striplot.
(b) Load the data le lab6dat.mat. Get familiar with the data that this loads (e.g.
type who, look at the lter coecient vectors h1, h2, and listen to the speech
signal x2).
395
(c) Filter the speech signal with the lter h1 and observe results using the commands
>> y1 = rlt(h1,x2);
>> inout(x2,y1,3000,1000,3);
Compare and comment on the input and output signals. For example, is the
output rougher or smoother than the input?
(d) Use freqz to plot the frequency response of lter h1 as a function of frequency
over . Why is h1 called a lowpass lter? Use the stem plot to
plot h1. What value of n are these lter impulse response coecients symmetric
about?
(e) Repeat (c) and (d) using the highpass lter h2. Denote the lter output y2.
(f) Make an A-B-C listening comparison using the command
>> soundsc([x2;y1;y2],8000);
Comment on your perception of the ltered outputs vs. the original signal.
(g) Make an A-D listening comparison using the command
>> soundsc([x2;(y1+y2)],8000);
Comment on your perception of the composite signal vs. the original. Did you
expect this result? Why?
Week 2: Do Procedure 2(a) and review the attached lter design & analysis example before
the Week 2 practicum session.
1. DT Filter Analysis:
(a) Consider the LTI discrete time lter with input/output dierence equation
y[n] 1.556 y[n 1]
+ 1.272 y[n 2] 0.398 y[n 3]

= 0.0798 (x[n] + x[n 1] + x[n 2] + x[n 3]) .
(b) Use the Matlab freqz command to compute the frequency response of this lter over the range . Plot the phase response, the magnitude response, and the magnitude response in dB. (The magnitude response, in dB, is
20 log10 {|H(ej )|}.)
(c) What type of lter is this? What is the cuto frequency?
(d) Using this Butterworth lter, and the Matlab lter function, lter the x2 signal
in the lab6dat.mat le used in Practicum V. Use the inout function to plot the
input and output. Comment on the eect of the lter on the input. Using the
soundsc function, listen to the input and output, and again comment on the eect
of the lter.
396
2. Filter Design:
Using Matlab, design a discrete time lowpass lter. This lter should be the minimum
order lter that meets the specication given below.
Specications:
i. Type: if your last name begins with a letter A through G - Chebychev Type I (use
the Matlab command cheby1); if your last name begins with a letter H through M
- Chebychev Type II (use the Matlab command cheby2); if your last name begins
with a letter N through Z - Elliptical (use the Matlab command ellip).
ii. Passband ripple: 2dB.
iii. Stopband attenuation: 50dB.
iv. Passband cuto p =
your grade on T est 2

.
80
v. Stopband cuto: s = 1.5 p .

Procedures:
(a) On a piece of paper, note your lter type. Write down and sketch your design
specications. Bring this to the Practicum session.
(b) Determine the minimum lter order which meets these specication. Discuss how
you selected this order.
(c) Determine the transfer function, and the input/output dierence equation.
(d) Determine the lter impulse response (explicitly as a real-valued function). Describe this derivation starting with the partial fraction expansion, generated using
the Matlab residue command, of the transfer function.
(e) Using your lter analysis experience gained by conducting the CT ltering Practicum
(Practicum 5a) and by reviewing the DT lter design & analysis Matlab example
referred to above, conduct your own procedures to analyze your DT lter. Use
Matlab to verify that your lter meets specications, and that its characteristics
are consistent with your understanding of the relationships between these lter
characteristics and the frequency response specications. Gather plots and other
results to be attached to your sign o sheet.
(f) By direct implementation of the I/O dierence equation, implement your lter
in Matlab. (Consider extending your simple IIR lter code from Practicum 1.)
Using a technique explored in Practicum 5a, verify that your lter is functioning
properly by analyzing outputs for one or more inputs.
echo on
%************************************************************
%
%
Discrete-Time Lowpass Filter Design & Evaluation
%
%************************************************************
%
%
We design a discrete-time lowpass filter, and then:
%
%
1) look at its numerator & denominator coefficients;
%
2) compute, list and plot its poles and zeros;
%
3) determine its partial fraction expansion from which
%
its impulse response can be written;
%
4) plot the frequency response.
%
pause
%
%
The specifications for the filter are:
%
%
1) 8-th order elliptic lowpass;
%
2) cutoff .4*pi radians/sample;
%
3) passband ripple of 2dB; and
%
4) minimum stopband attenuation of 40dB.
%
N=8;
PB=2;
SB=40;
CFREQ=.4;
[B,A]=ellip(N,PB,SB,CFREQ);
pause
[B,A] % list transfer function numerator/denominator coefficients
pause
%
%
Determine zeros & poles from Transfer function coefficient
%
Z=roots(B);
P=roots(A);
zang=angle(Z);
zamp=abs(Z);
pang=angle(P);
pamp=abs(P);
pause
[zamp,zang,pamp,pang] % list pole/zero magnitude/phase
pause
397
398
%
%
Plot zeros and Poles in the z-plane (try using the "zplane" function)
%
subplot(221)
polar(zang,zamp,o)
hold on
pause
polar(pang,pamp,x)
text(-1.45,0.2,Hugo Chavez)
pause
%
% Partial fraction expansion of H(z)
%
[res,pol,dr]=residue(B,A);
rang=angle(res);
ramp=abs(res);
[pamp,pang,ramp,rang] % list residues/poles
pause
%
%
Compute the Frequency Response (try "freqz" without an output)
%
Npts=512;
[hfreq,w]=freqz(B,A,Npts,whole);
mag=abs(hfreq);
phase=angle(hfreq)*180/pi;
%
%
Plot the Frequency Response
%
%
a) phase vs. angular freq. (from 0 to 2*pi)
%
subplot(223)
plot(w,phase)
xlabel(Freq. (radians/sample))
ylabel(Phase Response (degrees))
text(2.7,-150,Felipe Calderon)
pause
%
%
b) magnitude vs. freq. (in radians/sample from 0 to 2*pi)
%
subplot(222)
plot(w,mag)
ylabel(Magnitude Response)
text(2.5,.8,Angel Cabrera)
pause

%
%
c) magnitude (dB) vs. freq. (in radians/sample from -pi to pi)
%
(try using "fftshift" to convert w ang mag to -pi to pi range)
%
w1(1:Npts/2)=w(1+Npts/2:Npts)-(2*pi);
w1(1+Npts/2:Npts)=w(1:Npts/2);
mag1(1:Npts/2)=mag(1+Npts/2:Npts);
mag1(1+Npts/2:Npts)=mag(1:Npts/2);
subplot(224)
plot(w1,20*log10(mag1))
ylabel(Magnitude Response (dB))
text(-1,-40,Lula da Salva)
399
400
.
401
Practicum 5b
Students Name:
1. Week 1, Procedure 1(a,b): Mag. & phase response plots

2. Week 1, Procedure 1(h): Computed Mag. & phase response, comparison
3. Week 1, Procedure 2(f,g): soundsc results
4. Week 2, Procedure 1(b): three frequency response plots
5. Week 2, Procedure 1(d): input/output plots
6. Week 2, Procedure 2(a): with a paper clip, attach the page with your lter frequency
response specications drawing and other required information
7. Week 2, Procedure 2(b): below, record your lter order and state how you arrived at
this value
8. Week 2, Procedure 2(c): record your I/O dierence equation (which will be used for
Procedure 2(f)).
402
9. Week 2, Procedure 2(d): record your poles/residues, and write down your impulse
response explicitly as a real-valued function.
10. Week 2, Procedure 2(e): with a paper clip, attach the plots and other information
resulting from your lter analysis. State the goals of your analysis. Using your
plots/results as a reference, discuss your observations that verify that your lter characteristics are consistent with your specications.
11. Week 2, Procedure 2(f): with a paper clip, attach your Matlab IIR lter implementation code, along with any plots/results used to verify the your lter is properly
functioning. Provide your argument that these plots/results verify proper functionality.
6.7
403
Problems
Chapter Topics:
6.1-30 (DTFT and DT LTI systems);
6.31-51 (z-transform and DT LTI systems);
6.52-63 (stability, causality and ROC);
6.64-70 (DT LTI system interconnections)
6.71-75 (applications)
1. Consider a DT LTI system with input x[n] = p5 [n] and impulse response
h[n] = sin(0.25n) . Determine Y (ej ), the frequency content of the output y[n]. Plot
n
|Y (ej )| over .
2. Consider a DT LTI system with input x[n], impulse response h[n] and output y[n] =
x[n] h[n].
(a) Let h[n] = sin(.5n) sin(.25n) and x[n] = 5[n 1]. Determine and plot H(ej ).
n
n
Determine Y (ej ). Using Y (ej ) determine the output energy Ey .
(b) Let x[n] = [n] 1.8[n 1] + .81[n 2] and h[n] = (n + 1) .9n u[n]. Determine
H(ej ) and X(ej ). From these, determine Y (ej ) (simplify the expression as
much as possible). From this determine y[n].
(c) Consider a DT LTI system with frequency response
H(ej ) = [2 + 2 cos(2)] ej2 ;
|| .
(3)
Consider input x[n] = 1 + 2 cos((/8)n) + 3 cos((/4)n) + 4 cos((/2)n).

Determine the output y[n].
(d) Consider a DT LTI system with impulse response h[n] = n an u[n]. Using entries
#3 and #4 of the DTFT table, and linearity, determine the frequency response
H(ej ) and the I/O dierence equation.
3. Consider a DT LTI system with |H(ej )| = 2 and H(ej ) = sin(). Its input is
x[n] = .5n u[n]. What is the energy of the output y[n]?
h[n] =
1
(.9)n u[n]
10
and input
x[n] =
k=
[n 5k] ,
Determine Y (ej ), the DTFT of the output. Determine the input power Px and the
output power Py .
404
5. The DTFT of a|n| is
1 a2
(1 + a2 ) 2a cos
Consider a DT LTI system with impulse response h[n] = 0.5|n| . Determine the frequency response H(ej ), and sketch it for . Determine the output for
input
x[n] = 4 + 5 cos(n/2) 3 cos(n) .
h[n] =
sin(.25n)
.
n
Determine the systems frequency response H(ej ). Plot H(ej ) for 2 2.

Determine its output for input
x[n] = 3 + 2 cos(.1n + ) + 5 cos(n) .
7. Given the DT LTI lter frequency response
H(ej ) =
1
0
|| 2
3
otherwise over
Determine the impulse response h[n]. (The answer should be written explicitly as a
real-valued signal.)
8. Consider a DT LTI system with frequency response
H(ej ) =
sin(2.5 )
.
sin(0.5 )
Determine and plot the impulse response h[n]. What frequencies over the range
are completely attenuated by this system.
Now consider a DT LTI system with frequency response

H(ej ) = ej10
sin(2.5 )
.
sin(0.5 )
Determine and plot the impulse response h[n]. What is the output y[n] for input
x[n] = 5 + cos( 2 n)?
5
9. Consider a DT LTI system with input x[n], impulse response h[n] and output y[n].
sin( n)
2
(a) Let h[n] = n . Determine & sketch the frequency response H(ej ) for
.
405
(b) Let x[n] = [n + 2] + 2[n] + [n 2]. Determine its DTFT X(ej ).
(c) Using an Eulers identity, your X(ej ) from (b) can be expressed in the form
X(ej ) = c1 + c2 cos(2), where c1 and c2 are constants. Determine these
constants and then sketch X(ej ); .
(d) Determine and sketch the DTFT of y[n]. From your sketches, determine the
percentage of the energy of x[n] in y[n].
10. A DT LTI system has impulse response h[n] = [n] + 3[n 1] + n(.5)n u[n 1].
Determine a I/O dierence equation for this system.
11. Consider a DT LTI system with input x[n] = [n + 2] + [n] + [n 2] and impulse
sin( n)
2
response h[n] = n .
a) Determine X(ej ), and plot it over || < (Hints: X(ej ) is real-valued. Think
Eulers identities.)
b) Determine H(ej ) and Y (ej ), and plot them over || < .
c) What percentage of input energy Ex makes it to the lter output?
12. DT LTI Systems:

(a) Given a DT LTI system with impulse response h[n] = n 0.9n u[n], determine
the system input/output dierence equation.
(b) Consider a DT LTI system with input/output dierence equation
y[n] 0.9 y[n 2] = x[n] + 0.9 x[n 2] .
The input is x[n] = s[n] + n[n], where s[n] = ejn is a signal of interest, and
n[n] = 4 ej(/2)n is noise. SNR is dened as signal power divided by noise power.
What is the input SNR, SNRin ? What is the output SNR, SNRout ? Determine
the gain G = SN Rout .
SN Rin
13. Consider a DT LTI lowpass lter with frequency response H(ej ) = p (); || < .
Consider input x[n] = s[n] + n[n], where s[n] = [n] is a desired signal and
1
1
n[n] = 2 [n + 2] + [n] + 2 [n 2] is noise. Signal-to-Noise Ratio (SNR) is dened as
signal energy over noise energy.
(a) Determine the lter input SNR, SNRin .
(b) Determine the lter output SNR, SNRout .
(c) The lter gain is dened as G = SN Rout . Describe a lter that has greater gain
SN Rin
than H(ej ) = p (); || < . Justify your answer.
14. Consider the third-dierence DT LTI system
y[n] = x[n] x[n 3] .
(a) Determine the impulse response h[n] and the frequency response H(ej ).
406
(b) Plot the magnitude and phase of the frequency response.
(c) Determine the output for input x[n] = 8 3 cos(n + (/10)).
(d) Determine the output for input x[n] = 2 cos((/2)n).

15. Consider the LTI lter with impulse response
h[n] = 2[n + 1] + [n] 2[n 1] .

(a) Determine H(ej ).
(b) Plot |H(ej )| for .
(c) Determine output y[n] for input x[n] = 3 cos(n).
16. Consider a DT LTI system with I/O dierence equation

y[n] 0.5 y[n 1] = x[n 2] + x[n 3] .
a) Determine the frequency response H(ej ).
b) Determine the output for input x1 [n] = [n 3].
c) Determine the output for input x2 [n] = 3 cos(n + ).

3
d) [Extra credit] Determine the output for input x3 [n] = cos( n). (1 + 0.5j =
2
1.118 ej(.1476) )
17. Consider a ideal DT lowpass lter with frequency response H(ej ) = 2 p () over the
frequency range .
(a) Given an input signal with DTFT X1 (ej ) = e|| ; < , determine the
percentage of input energy that makes it to the output.
(b) Given input signal x2 [n] =
l= [n l5], determine the percentage of input
power that makes it to the output.
18. Consider a DT LTI system with input x[n], impulse response h[n] and output y[n].
a) h[n] = 1 {[n + 2] + [n] + [n 2]}. Determine the frequency response. Use an
2
Eulers identity to express it as a real-valued function. Sketch it over .
sin( n)
2
b) x[n] = n . Determine and sketch its DTFT, and the DTFT of y[n]. Determine
Ex , the energy of x[n].
c) Without integrating over |Y (ej )|2 , determine the output energy Ey .

h[n] = 2 n 0.5n u[n] .
a) Use linearity and the DTFT pairs table to determine the system frequency response.
b) What is the DC gain (i.e. |H(ej0)|) ?
407
c) What is the gain for = (i.e. |H(ej )|) ?

H(ej ) =
(1 .9 ej )
(1 .5 ej )(1 + .5 ej )
a) What is the system dierence equation?

b) Noting that Y (ej ) = X(ej ) H(ej ) (by the convolution property of the DTFT),
determine Y (ej ) then y[n] due to input with DTFT
X(ej ) =
(1 + .5 ej )
(1 .9 ej )
c) Now nd the output y[n] due to input x[n] = [n] + .5 [n 1] .

21. Given the impulse response
h[n] = .5n u[n] n .9n u[n]
determine the system input/output dierence equation.
h[n] = 5 (.8)n u[n] + 2 (.5)n u[n] .
Determine the system frequency response and LCC dierence equation.
23. Given a DT LTI system with input/output dierence equation
2 y[n 1] y[n 2] = x[n] .5 x[n 1] .
where x[n] is the input and y[n] is the output, determine the system impulse response
h[n].
24. Given a system with input/output dierence equation
y[n] .2y[n 1] + .2y[n 3] = x[n 1] + x[n 2] ,
determine the output y[n] for input x[n] = 5.
25. Consider the DT LTI system which, for input x[n] = .5n u[n], has output y[n] with
DTFT
Y (ej ) =
2
(1 + .9
ej )(1
.5 ej )
a) Determine the system frequency response H(ej ).

b) Determine the impulse response h[n].
408
c) Now that you have identied the system, nd the output y1 [n] due to input
x1 [n] = 3 + ejn .
d) Based on your results in c), would you say that this system is a highpass or
lowpass lter?
H(ej ) =
(1 .9 ej )
.
(1 .5 ej )(1 + .5 ej )
(a) What is the system dierence equation?

(b) Given output y[n] = 3 (.5)n1 u[n 1], determine the input x[n].
y[n] .9y[n 1] = x[n] .7x[n 1] .
We know that the output is y[n] = .9n u[n]. Find the frequency response of the
system, and the DTFT of the output. Use these to determine the input x[n] that
resulted in this output y[n].
28. Given input
5
(.5)n1 u[n 1]
2
x[n] = [n] +
and output
y[n] = (.5)n u[n]

to a DT LTI system, determine the system impulse response h[n]. Is the system stable?
Why?
29. We have a DT LTI system with unknown impulse response h[n]. For this system, we
have the following two input/output (I/O) pairs:
x1 [n] = (1)n u[n] ;
y1 [n] = .7n u[n]
x2 [n] = cos(n) ;
y1 [n] = 0.
Determine the impulse response h[n]. Do you need both I/O pairs to solve this problem? Why? Can you solve this problem with either I/O pair? Why?
30. System Identication:
(a) Determine what you can about the DT LTI system frequency response given
input/output pair
x1 [n] = cos(n(/2)) ;
y1 [n] = 2 cos(n(/2) + (/3)) .
409
(b) Determine what you can about the DT LTI system frequency response given
input/output pair
x2 [n] = [n] 0.5 [n 1] ;
y2 [n] = [n] [n 1] .

h[n] = 2 (0.5)n u[n] 0.5 (0.25)n1 u[n 1].
(a) Determine the system transfer function H(z).
(b) Determine the system input/output dierence equation.
(c) Determine the system poles and zeros.
(d) Determine the system frequency response H(ej ).
32. DT Transform Pairs and DT LTI System Descriptions:
(a) Consider a causal DT LTI system with transfer function
Ha (z) =
z+4
z + 0.5
(4)
Determine its impulse response ha [n].

(b) Consider a DT LTI system with impulse response
hb [n] = [n] + 10 (0.5)n u[n 1]
(5)
Determine its dierence equation and the ROC of its transfer function.
(c) Consider a DT LTI system with frequency response
Hc (ej ) = sin(2)
(6)
Determine its impulse response hc [n].

(d) Consider a DT LTI system with impulse response
hd [n] = 2n cos(0.5n) u[n 1] .
(7)
Determine its frequency response Hd (ej ).

(e) Consider a DT LTI system with frequency response
He (ej ) = ej0.5
Determine its impulse response he [n].

33. Given a causal DT LTI system with transfer function
H1 (z) =
2(1 + j) z
2(1 j) z
+
,
j/4
z .4 e
z .4 ej/4
determine its impulse response h1 [n]. Write it explicitly as a real-valued signal.
(8)
410
34. Transfer functions of DT LTI systems:
(a) Given transfer function H1 (z) =
1
10.25z 1
1
1 with ROC {|z| > 4 }, nd h1 [n].
(b) Given impulse response h2 [n] = cos(0.5n) u[n], determine H2 (z) and the ROC.
Find the poles/zeros.
(c) Given impulse response h3 [n] = 0.5n u[n] + n (2)n u[n 1], determine the
system dierence equation.
(d) Consider a DT LTI system with impulse response
h[n] = 4 (1/2)n2 u[n 2] 2 (1/4)n1 u[n 1] .
Determine the z-transform function H(z), expressed as a rational function (i.e.
with a common denominator). Sketch the region-of-convergence Rh .
35. Consider a causal DT LTI system with I/O dierence equation
ya [n] 0.5 ya [n 1] = xa [n] + 6 xa [n 1]
n
u[n]. Determine ya [n]. (You need to do a PFE. You must

and input xa [n] = 1
8
show all steps of this to receive signicant credit.)
36. Consider a DT LTI system with impulse response hc [n] = an u[n] and measured
output yc [n] = 2 (a)|n| . Determine the corresponding input xc [n].
37. Consider a DT LTI system with transfer function
H2 (z) =
(z ej/2 )(z ej3/2 )

.
(z 0.95)(z + 0.95)
Sketch the pole/zero diagram. Considering only the pole/zero diagram, roughly sketch
the magnitude of the frequency response. What is H(ej ) ?
38. Consider input x[n] = 2n .5n u[n] + .3n cos( n) u[n] to a DT LTI system with impulse
2
response h[n] = [n] [n 1] + .25[n 2], determine the output z-transform Y (z).
Simplify your expression for Y (z) and then determine y[n].
39. Given a DT LTI system with impulse response
h[n] = 2 2n u[n 1] 4 .5n cos( n) u[n] ,

2
determine the transfer function H(z), the ROC Rh , the frequency response H(ej ),
the I/O dierence equation, and the poles. How would you nd the zeros?
40. Consider a causal DT LTI system with I/O dierence equation
y[n] = x[n] .25 x[n 2] .
This system has zeros at z1 = 0.5 and z2 = 0.5. Determine the input x[n] that
produces output y[n] = p2 [n]. (x[n] will be causal.) To do this use z-transforms.
Theres a PFE required. For this, show all calculations required to derive the residues.
411
41. Consider a DT LTI system with transfer function of the form

H(z) =
(z z1 )(z z2 )
.
(z p1 )(z p2 )
Consider the following dierent sets of system parameters:

i) z1 = 0, z2 = 0, p1 = 0.9j, p2 = 0.9j
ii) z1 = 0.9, z2 = 0.9, p1 = 0.9, p2 = 0.9
iii) z1 = 0.9, z2 = 0.9, p1 = 0.9, p2 = 0.9
iv) z1 = 0.9, z2 = 0.9, p1 = 0.9, p2 = 0.9

v) z1 = 1, z2 = 1, p1 = 0, p2 = 0
Answer the following:

(a) Which set of parameters corresponds to a lowpass lter? Why?
(b) Which set of parameters corresponds to a highpass lter? Why?
(c) Which set of parameters corresponds to a lter which has a zero response at some
frequencies? Why?
(d) Which set of parameters corresponds to a lter which has a maximum frequency
response at frequencies = ? Why?
2
(e) Which set of parameters corresponds to a lter which has constant gain at all
frequencies? Why?
Roughly sketch the magnitude of the frequency response for one of these set of parameters. Clearly indicate which set you are considering.
42. Consider a DT LTI system with transfer function of the form
H(z) =
(z z1 )(z z2 )
.
(z p1 )(z p2 )
Consider the following dierent sets of system parameters:

i) z1 = 0, z2 = 0, p1 = 0.9j, p2 = 0.9j
ii) z1 = 0.9, z2 = 0.9, p1 = 0.9, p2 = 0.9
iii) z1 = 0.9, z2 = 0.9, p1 = 1.1, p2 = 1.1
iv) z1 = 0.9, z2 = 0.9, p1 = 0.9, p2 = 0.9

v) z1 = j, z2 = j, p1 = .9j, p2 = .9j
Answer the following:

(a) Which set of parameters corresponds to a lowpass lter? Why?
(b) Which set of parameters corresponds to a highpass lter? Why?
412
(c) Which set of parameters corresponds to a lter which has a zero response at some
frequencies? Why?
(d) Which set of parameters corresponds to a lter which has a maximum frequency
response at frequencies = ? Why?
2
(e) Which set of parameters corresponds to a lter which has constant gain at all
frequencies? Why?
Roughly sketch the magnitude of the frequency response for one of these set of parameters. Clearly indicate which set you are considering.
h[n] = 2 (0.5)n u[n] + 1.96 (0.7)n1 cos(
n) u[n] .
2
Determine the transfer function H(z). Sketch the pole/zero diagram. What is the system ROC? Determine the system dierence equation. Based on the pole/zero locations,
roughly sketch the overall system frequency response magnitude.
44. A DT LTI system is described by the LCC dierence equation
y[n] + 0.3y[n 1] 0.4y[n 2] = x[n] + 6x[n 1] .
The poles of this system are p1 = 0.8 and p2 = 0.5. Determine the system impulse
response. Explicitly write it as a real-valued signal.
45. Given a causal system with transfer function,
H(z) =
7
1 + 0.1 z 1 0.12 z 2
7
(1 0.3 z 1 ) (1 + 0.4 z 1 )
nd the partial fraction expansion of H(z) and the corresponding impulse response
h[n].
46. Consider the following causal DT LTI system with dierence equation
y[n] 1.5y[n 1] y[n 2] = x[n] + 8x[n 1] .
(a) Find the impulse response. (The poles of the system are p1 = 2 and p2 = 0.5.)
(b) Find the response to x[n] = [n] + 0.5[n 1].
47. A causal DT LTI system is described by the LCC dierence equation

y[n] .9y[n 1] + .81y[n 2] .729y[n 3] = 4x[n] 1.8x[n 1] + 1.62x[n 2] .
The poles of this system are p1 = .9, p2 = .9j and p3 = .9j. Determine the system
impulse response. Write it explicitly as a real-valued signal.
413
48. Consider a causal DT LTI system with transfer function

H(z) =
1 0.9 z 1
.
(1 0.5 z 1 )(1 + 0.5 z 1 )
Given output y[n] = 3 (.5)n1 u[n 1], determine the input x[n].
49. Consider a DT LTI system which, for input x[n] = (0.2)n u[n], produces output
y[n] = (0.4)n cos( n) u[n]. Determine the real-valued systems impulse response. A
2
partial fraction expansion is required. Show all steps in its derivation.
50. Consider a causal DT LTI system with output z-transform
Y (z) =
3 z 1
;
1 0.5 z 1
Ry = |z| > 0.5
and input
x[n] = 3 (0.9)n1 u[n 1] + 1.5 (0.9)n2u[n 2] .
(a) Determine X(z) in rational function form.
(b) Determine the system transfer function H(z), simplied as much as possible.
(c) Sketch the system pole/zero diagram.
(d) Determine the system I/O dierence equation.
(e) Determine the impulse response h[n].
51. Consider a DT LTI system with impulse response h[n] = [n] 3 n (0.5)n u[n] and
input
x[n] = [n] [n 1] + 0.25 [n 2].
a) Determine the system transfer function H(z) and write it in rational function
form.
b) What is the system I/O dierence equation?
c) What is the system frequency response H(ej )?
d) Determine the z-transform X(z) of the input.
e) Determine the output z-transform Y (z) and from that the output y[n].
52. Consider a DT LTI system with I/O dierence equation
9
y[n] =
k=0
2k x[n k] .
(a) Determine the transfer function. Use the geometric series to express it as a rational
function of z.
(b) Determine its poles and zeros. (Note that the roots of z N aN are
a ej(2/N )k ; k = 0, 1, , N 1. Plot these on the z-plane.
414
(c) Is this system stable? Why?
h2 [n] =
1
3
u[n]
1
2
u[n] .
(a) Find its transfer function H2 (z) and ROC.

(b) Determine its I/O dierence equation.
(c) What are its poles and zeros?
(d) Is the system stable? Explain why in terms of the ROC. Explain why in terms of
the pole/zero locations.
54. Consider a stable DT LTI system with transfer function
H(z) =
1 + 4.8 z 1
1 + 1.2 z 1 1.6 z 2
which has poles p1 = 0.8 and p2 = 2.

(a) Is the system causal? Why?
(b) Determine the PFE of H(z). To receive any credit, you must show all calculations.
(c) Determine the impulse response h[n].
55. Consider a DT LTI system with transfer function
H(z) =
9 z2
(z + 0.5)(z 0.25)
ROC = 0.25 < |z| < 0.5 .
(a) What is the system frequency response?

(b) What is the system I/O dierential equation?
(c) What are the system poles and zeros?
(d) Derive the partial fraction expansion of H(z). Show all steps and intermediate
(e) What is the system impulse response?
(f) Is the system stable? Why? Causal? Why?
56. Consider a causal DT LTI system with transfer function
H(z) =
1 1.05 z 1
1 1.05 z 1
=
.
1 2.1 z 1 + 0.2 z 2
(1 2 z 1 ) (1 0.1 z 1 )
a) Find the partial fraction expansion of H(z) & the corresponding impulse response
h[n].
b) What is the system dierence equation?
415
c) What is the system frequency response?

d) What are the system poles and zeros?
e) What is the ROC, Rh , of h[n]?
f) Is the system stable? In terms of h[n],why? In terms of Rh ,why?
57. Consider the following causal DT LTI system with dierence equation
y[n 1] .5 y[n 2] .5 y[n 3] = x[n 1] + 4 x[n 2] .
The poles of the system are p1 = 1 and p2 = 0.5. Determine the transfer function
H(z) and the impulse response h[n] of this system. Is the system stable?
58. The z-transform and DT LTI systems:
(a) Given a DT LTI system with impulse response h[n] = (.8)n u[n] with output
1
z-transform Y (z) = 1.7z 1 , determine the input x[n].
1+.8z
(b) A stable DT LTI system is described by the LCC dierence equation
y[n] 1.5y[n 1] y[n 2] = 5 x[n 1] .
The poles of the system are p1 = .5 and p2 = 2. Determine the system impulse
response h[n].
59. Consider a DT LTI system with impulse response h[n] which has z-transform
H(z) =
5
4
1 + 12z 1
=
.
1 z 2
1
1 + 1.5z
1 0.5z
1 + 2z 1
(a) If the ROC is Rh = {0.5 < |z| < 2}, what is h[n]?
(b) If h[n] is stable (i.e. decays as n and n ), what is it?
(c) Determine the h[n] which would be consistent with this H(z) except that it has
no ROC (i.e. it has no z-transform).
y[n] 1.4 y[n 1] + 0.4 y[n 2] = 2 x[n] + 0.6 x[n 1]
(a) Provide a pole/zero diagram for this system.

(b) Can this system be stable? To get credit you must explain your answer.
(c) If the system is causal, what is its impulse response h[n]? A partial fraction expansion (PFE) is required. To get full credit you must show all work in performing
this PFE.
416
61. Consider a DT LTI system with input
x[n] = [n] 4(.8)n u[n 1]
X(z) =
1 + 4z 1
1 + .8z 1
and corresponding output

y[n] = 5(.5)n u[n] 4(.8)n u[n]
Y (z) =
1 + 2z 1
.
(1 + .8z 1 )(1 .5z 1 )
(a) Determine the system transfer function H(z), the poles, and the dierence equation.
(b) Assume that the systems is stable. Determine the system region of convergence
and impulse response.
62. The following three parts are independent of one another.
(a) Consider the following two transfer functions:
Ha (z) =
1
;
1 0.9z 1
Hb (z) =
1
.
1 1.1z 1
i. Which one, if any, can be stable/causal. If any, give the corresponding ROC.
ii. Which one, if any, can be stable/non-causal. If any, give the corresponding
ROC.
iii. Which one, if any, can be non-stable/causal. If any, give the corresponding
ROC.
iv. Which one, if any, can be non-stable/non-causal. If any, give the corresponding ROC.
(b) Given the following DT LTI system transfer function, sketch the pole/zero diagram. Sketch the magnitude of the frequency response in enough detail to illustrate the eect of each pole and zero.
H(z) =
(z 1)(z + 1)(z 0.9j)(z + 0.9j)

.
(z 0.95j)(z + 0.95j)
(c) Given a causal DT LTI system with transfer function

H1 (z) =
2j z
2j z
+
,
j3/4
z .9 e
z .9 ej3/4
determine its impulse response h1 [n]. Write it explicitly as a real-valued signal.

63. Consider the following Matlab code:
[B,A] = butter(2,[0.25 0.75],stop)
Z = roots(B)
[R,P,K] = residue(B,A)
417
Note that B is the array of transfer function numerator polynomial coecients. Say
this code generates the following results:
i. Z1 = Z2 = ej2 , Z3 = Z4 = ej2 .
ii. P1 = 0.8 ej2.3 , P2 = 0.8 ej2.3 , P3 = 0.7 ej1.6 , P4 = 0.7 ej1.6.
iii. R1 = 0.2 ej1.1, R2 = 0.2 ej1.1 , R3 = 0.24 ej1.4 , R4 = 0.24 ej1.4 .
iv. K = 0.6.
Consider the following questions/tasks.
(a) Is the designed system stable? Explain your answer.
(b) Sketch the pole/zero diagram.
(c) Roughly sketch the magnitude of the system frequency response. What kind of
lter is this (e.g. lowpass, highpass, bandpass, bandstop)?
64. DTFT and DT LTI systems:
(a) Recall that the impulse response of the cascade of two systems is the convolution
of the individual impulse responses. Consider the cascade of a rst DT LTI system
with LCC dierence equation
v[n] = 0.7 v[n 1] + x[n]
(where x[n] is the input and v[n] is the output) with a second DT LTI system
with impulse response
h2 [n] = [n 1] + 0.7[n 2] .
Determine the frequency responses of the rst, second and combined systems.
What is the impulse response of the combined system?
(b) Given a DT LTI system with impulse response
h[n] = (n + 1) 0.2n u[n] ,
determine the frequency response and LCC dierence equation.
65. Consider two DT LTI systems connected in parallel. One has impulse response
h1 [n] = n (0.7)n1 u[n 1], and the other h2 [n] = [n]. Determine the input/output
LCC dierence equation of the overall system.
66. Consider two LTI subsystems connected in parallel. Let their system functions be
H1 (z) =
1 1 z 1
2
,
1
1 8 z 1
H2 (z) =
3
1
1 + 8 z 1
Determine the causal impulse response of the overall system.
418
67. Two LTI subsystems have impulse responses h1 [n] = [n] 0.5[n 1] and
h2 [n] = 0.5n u[n].
(a) Identify, on a z-plane diagram, the locations of the poles/zeros of these two individual subsystems.
(b) Consider the cascade of these two subsystems. Identify the locations of the resulting systems poles/zeros, and use these to roughly sketch the magnitude response
|H(e )|.
68. Consider two causal DT LTI subsystems connected in cascade. The rst subsystem
has I/O dierence equation
y[n] + 0.9 y[n 1] = x[n] 0.5 x[n 1] .
The second subsystem has impulse response
h2 [n] = (0.5)n u[n] .
(a) First determine the transfer function H(z) of the overall system.
(b) From your H(z) above, determine the overall impulse response h[n].
69. Consider two causal DT LTI subsystems in cascade. The rst has I/O dierence
equation
yb [n] 0.7 yb [n 1] = 2 xb [n] + 1.2 xb [n 1] .
Determine the I/O dierence equation of the second subsystem such that its output is
equal to the input of the rst subsystem.
70. Consider the following feedback interconnection of DT LTI systems.
X(z)
Y(z)
H1 (z)
H2 (z)
H
This system has overall transfer function H(z) = 1 + H11 (z)H2 (z) . Let H1 [z] =
(z)
and H2 (z) = g where g is a constant gain term. Assume that H1 (z) is causal.
1
z+2
(a) Determine the range of gain g for which the the overall system is stable.
(b) For values of g within the range identied in part (a), which value maximizes the
DC gain?
71. Consider the following channel/equalizer block diagram. Assume both subsystems are
DT LTI.
419
x[n]
v[n]
Channel
Hc (z)
y[n]
Equalizer
He (z)
(a) First let Hc (z) = 1 + 0.9 z 1 and He (z) = 1 0.9 z 1 + 0.81 z 2 0.729 z 3 .
Determine the overall impulse response from x[n] to y[n]. For random process
input x[n], with correlation function Rx [m] = 2 [n], determine the output
correlation function Ry [m].
(b) Again let Hc (z) = 1 + 0.9 z 1 . Using DT transforms, design a better equalizer
than that used in (a). For your equalizer, determine the overall transfer function
and impulse from x[n] to y[n]. Is your equalizer causal and stable? Why?
(c) Now let Hc (z) = 1 + 2 z 1 . Using your previous experience with equalizers,
design an equalizer. Analyze it (i.e. look at causality, stability, eectiveness).
72. The following system is called a linear predictor structure when the FIR (nite impulse
response) DT LTI lter is designed to minimize the power of the signal e[n]. Then the
FIR lter output x[n] is a prediction of input x[n] based on past values of this input.
x[n]
(a)
(b)
e[n]
D sample
delay
FIR filter
^
x[n]
h fir [n]
x[nD]
Assume that the FIR lter has impulse response hf ir [n] =
1
10
p10 [n].
(a) What is Hf ir (ej ), the frequency response of the FIR lter? What is its response
to a DC input (e.g. to x[n D] = 1)?
(b) What is the impulse response of the entire structure (i.e. from (a) to (b))? What
is its response to a DC input (e.g. to x[n] = 1)?
73. Consider a digital communications channel modeled as a DT LTI system with transfer
function
Hc (z) =
(z 0.9)(z + 0.1)
z 3 + 0.5 z 2
which you wish to equalize.

(a) Denote the channel input and output as, respectively, x[n] and v[n]. Determine
the channel I/O dierence equation. Without solving for the channel impulse
response, determine if the channel can be both causal and stable. Identify the
characteristics of Hc (z) and the dierence equation which justify your answer.
(b) Determine the transfer function He (z) of the equalizer which exactly equalizes
this channel.
420
(c) For your He (z), using PFE, determine the right-sided impulse response he [n]. To
get any credit for this part you must show all steps of the PFE.
(d) Is your he [n] causal? Stable? (Note that it should not be both.) Suggest how you
can modify your he [n] so that you have an eective causal and stable equalizer.
74. Consider a DT LTI channel that given input x[n] = [n] + [n 1] has output
v[n] = [n] + 3[n 1] + 2[n 2].
(a) Determine the channel transfer function Hc (z) and corresponding impulse response hc [n].
(b) Determine the impulse response, denoted he1 [n], of the causal equalizer which
exactly inverts this channel. Is there a problem with this equalizer? Why?
(c) Consider as an alternative an equalizer with impulse response
he2 [n] = 0.0625[n] + 0.125[n 1] 0.25[n 2] + 0.5[n 3] .
Determine the overall channel/equalizer impulse response ht2 [n]. Is this equalizer
eective? How could it be improved?
75. Consider the channel/equalizer problem. For each of the following channels determine
the exact equalizer transfer function and impulse response, and using system characteristics discuss each equalizers practicality (i.e. its stability and causality).
(a) hc1 [n] = (0.5)n u[n].
(b) hc2 [n] = [n] [n 1].
(c) hc3 [n] = [n 3].
For the following channel just identify the ideal equalizer transfer function in rational
function form.
10
k
(d) hc4 [n] =
k=0 0.8 [n k]. (Hint: use the geometric series equation on the
transfer function.)
421
Introduction to Random Processes
The signals we have considered to this point in this Course have been both known and
deterministic. That is, we assume they are determined (or described) by equations and we
know what the equations are. We have seen that very eective tools and practices exist
for analyzing and processing signals under these assumptions. However, most signals we
observe in nature have some random characteristics, meaning that there is some uncertainty
about them. Either we have no equation to describe them (they are not deterministic), or
we have an equation to describe them but there is something unknown about this equation
in which case they are deterministic but unknown. An example of an unknown deterministic
signal is a sinusoid with unknown frequency, magnitude and/or phase. Signals with random
characteristics are called random processes (a.k.a. random signals, stochastic processes).
In this Chapter of the Course we briey describe and mathematically characterize random
processes.
It would certainly be advantageous if the tools and practices we have learned in this Course
could be eectively applied to random processes. It turns out that they can. Our objective
here is to show at the most basic level how the theory considered up to this point in the Course
(i.e. concerning transforms and LTI systems) applies to random processes. Specically, we
will learn how Fourier transforms can be used to characterize the frequency content of a
random signal, and how LTI systems eect random signals. Since we have only two lectures
to introduce this topic, we will focus on DT random processes. The representation and
processing of CT random processes closely parallels our discussion below.
Some of this material is a review of probability, which is included here to refresh the
understanding you acquired, in your engineering probability course, of the representation
and analysis you random phenomena. This constitutes Section 7.1, which is a review of
random variables. The probabilistic representation of random processes is a direct extension
of that of a random variable and of multiple random variables. We develop this representation
in Sections 7.2, and use it in Section 7.3 in conjunction with Fourier transforms to develop a
representation of the frequency content of a random process. In Section 6.4 we will see how
LTI systems eect random signals.
You will see this topic again in future communications courses. It may also covered as
a brief overview in a senior level DSP course. It may also be studied in some depth and
employed extensively in some other senior level course (e.g. Biomedical DSP elective, since
biomedical signals are almost always random). If you are really interested in a more indepth understanding of this topic, nd a way to take a level graduate course on Stochastic
Processes.
For testing purposes, you will only be responsible for a few basic topics that closely relate
to material already covered in this Course. These topics are the contents of Sections 7.2,3,4.
Learn what a random process is, and basically how to characterize it.
Learn what a DT wide-sense stationary random process is, and how to characterize it.
Learn how a DT LTI system eects a DT wide-sense stationary random process.
422
7.1
Review of Random Variables
Recall that, strictly speaking, a random variable is a mapping from the outcomes of a random
experiment to the set of numbers. In signal processing problems, the outcomes of a random
experiment are often already numbers (e.g. the voltage at the output of a sensor), so we tend
to think of random variables as the data itself, and forget about any mapping. Remember
that a random variable is characterized by the numerical values it can take on, and by the
probabilities of these values.
A Single Random Variable
Let X denote a real-valued random variable which generally takes on values x over the
range x . Its probability distribution function (PDF) is dened as
FX (x) = P (X x) .
(1)
That is, for a given value x, the PDF FX (x) of the random variable X is the probability that
the random variable X will be less than or equal to the value x. The PDF is a function of
values x. The PDF has the following properties:
1. FX () = 0.
2. FX () = 1.
3. FX (x) is a nondecreasing function of x.
4. P (a < X b) = FX (b) FX (a).
The probability density function (pdf ) of a random variable X is dened as
fX (x) =
d
FX (x) .
dx
(2)
The pdf has the following properties:

1. fX () = fX () = 0.
2. fX (x) 0 ; x.
3.
fX (x) dx = 1.
4. P (a < X b) =
b
a
fX (x) dx.
This last property is the reason that fX (x) is referred to as a probability density function probabilities are computed by integrating over it (the area under the curve is the probability).
Note, from calculus or by property 4., that
FX (x) =
fX () d .
(3)
423
Example 7.1: A uniform random variable - consider the following pdf

1
ba
fX (x) =
a<xb
otherwise
f (x)
X
1
ba
x1
x2
x (the values of X)
Figure 127: The uniform pdf.

Assume a < x1 < x2 < b. Determine the expression for P (x1 X x2 ).
Solution:
P (x1 X x2 ) =
x2 x1
b a
Example 7.2: A Gaussian (normal) random variable - consider the following pdf
fX (x) =
1
2
2
e(x) /2
2 2
2 > 0
f (x)
X
x (the values of X)
Figure 128: The Gaussian pdf.

Assume x1 < x2 . Determine the expression for P (x1 X x2 ).
Solution:
P (x1 X x2 ) =
x2
x1
1
x1
2
2
e(x) /2 dx = Q
2
x2
The Q-function is dened numerically as

Q(x) =
1
2
ex /2 dx .
2
Although it is necessary to nd probabilities of Gaussian and other random variables in

many applications, you will not be required to do this in this course.
424
Multiple Random Variables
Let Xi ; i = 1, 2, , N be N real-valued random variables. Let X = [X1 , X2 , , XN ]T

be an N-dimensional column vector representation of these. Let x = [x1 , x2 , , xN ]T the
N-dimensional column vector of values of X. The joint pdf of X is denoted fX (x). It is an
N-dimensional function of the values x.
The joint pdf has the following properties:
1. fX (x) = 0 if any element of x is either or .
2. fX (x) 0 ; x.
3.
fX (x) dx = 1.
4. P (a < X b) =
b
a
fX (x) dx.
As was the case for a single random variable, note that property 4 indicates why fX (x) is
termed a probability density - probabilities are computed by integrating over it. For example,
for the N = 2 random variable case, the volume under the fX (x) surface is the probability.
In general, we say the hyper volume under fX (x) over a certain range of x is the probability
that the random variables in X are jointly in that range.
Example 7.3: Let N = 2 and X = [X1 , X2 ]T . Let
fX (x) =
0 x1 1,
otherwise
8x1 x2
0
fX
x2
1 ,X 2
0 x2 x1
(x1 , x2 )
1
x2
x1
1
1
Region of Support
x1
Figure 129: A 2 dimensional pdf.

The region-of-support is the range of values x for which the joint pdf is nonzero
(i.e. the range of possible values of X).
Determine P (0.5 X1 1, 0.5 X2 1).
Solution:
P (0.5 X1 1, 0.5 X2 1) =
x1
1
.5
= 8
= 8
.5
1
.5
1
.5
x1
x1
.5
x2 dx2
x3
x1
1
2
8
dx1 = 8
dx1 =
x4
1
x1
fX1 ,X2 (x1 , x2 ) dx2 dx1 =
.5
1
.5
x2
1
2
.5
x1
=
.5
8x1 x2 dx2 dx1

x2 1
1
2
8
9
16
dx1
425
Marginalization:
Let X = [X1 , X2 , , XN ]T be a set of random variables, and consider some partitioning
of these, for example X 1 = [X1 , X2 , , XP ]T and X 2 = [XP +1 , X2 , , XN ]T . To obtain the
joint pdf fX 1 (x1 ) we marginalize over X2 as follows:
fX 1 (x1 ) =
fX (x) dx2
(4)
That is, you integrate over x2 to eliminate it.

Statistical Independence:
Consider X = [X1 , X2 , , XN ]T partitioned, for example, as X 1 = [X1 , X2 , , XP ]T and
X 2 = [XP +1 , X2 , , XN ]T . Then, by denition, the random variable sets X 1 and X 2 are
statistically independent of each other if and only if
fX (x) = fX 1 (x1 ) fX 2 (x2 ) .
(5)
Example 7.4: For X = [X1 , X2 ]T , let

fX (x) = e(x1 +x2 ) u(x1 ) u(x2 ) .
Are X1 and X2 statistically independent?
Solution:
fX (x) = ex1 u(x1 ) ex2 u(x2 ) = fX1 (x1 ) fX2 (x2 )
(i.e. the joint pdf factors into the product of the two single-dimensional pdf s).
Try marginalizing over X1 to get fX2 (x2 ), and over X2 to get fX1 (x1 ), So X1 and
X2 are statistically independent.
Expectation
Let X be a single real-valued random variable with pdf fX (x). The expected value of X
is dened as
E{X} =
x fX (x) dx .
(6)
E{} =
fX (x) dx is the expectation operator. (In this case we are simply considering
the expected value of X.) Evaluating this equation, observe that E{X} is a weighted
average of the values x, where the weighting function is the pdf. This probabilistic weighting
emphasized values x which are more probable. That makes sense.
It is useful to remember that the expectation operator is linear. Thus, for constants a and
b, and random variable X,
E{aX + b} = a E{X} + b .
(7)
426
Moments
Consider, for positive integer , the class of functions g(X) = X . The moments about
the origin are dened as:
= E{X } =
x fX (x) dx .
(8)
For example, the 1-st moment about the origin of X, 1 = E{X}, is termed the mean of X.
The mean of X is denoted x , i.e. 1 = x . It is useful to think of the 2-nd moment about
the origin, 2 = E{X 2 }, as the energy (or power) of the random variable.
Again for positive integer , consider the class of functions g(X) = (X ) . The central
moments are dened as:
m = E{(X x ) } =
(x ) fX (x) dx .
(9)
The most commonly considered central moment is the 2-nd order central moment
m2 = E{(X x )2 } =
(x x )2 fX (x) dx .
(10)
2
2
We term this moment the variance. The variance of X is denote x , i.e. m2 = x . It
is useful to think of the variance as the energy (or power) of the variation of the random
variable from its mean. Note that
2
x = 2 2 .
x
(11)
You can use the linearity of the expectation operator to prove this.
Example 7.5: Determine the mean and variance of the uniform random variable
X considered in Example 7.1.
Solution: For the mean,
b
x =
1
1 x2
x dx =
ba
ba 2
=
a
1 a2 b2
a+b
=
ba
2
2
For the variance, let q = b a be the width of the density function. Then
2
x
1
=
q
b
a
a+b
x
2
1
dx =
q
q/2
q/2
1 x3
x dx =
q 3
q/2
The variance of a uniform random variable is 2 =

over twelve.
q2
,
12
=
q/2
1
q
q3
q3
+
24 24
i.e. the width squared
The mean of a random variable with symmetric fX (x) is always the point of symmetry.
q2
12
427
Joint Moments of Two Random Variables

Consider two random variables X and Y , and let Z = g(X, Y ) be some function of them.
The expectation of Z is
E{Z} = E{g(X, Y )} =
g(x, y) fX,Y (x, y) dx dy
(12)
This generalizes to g(Xi ; i = 1, 2, , N) is an obvious manner.

Given two random variables X and Y , the ij th moment about the origin is
ij = E{X i Y j } =
xi y j fX,Y (x, y) dx dy
(13)
For example, the mean of X, is

x = 10 = E{X 1 Y 0 } = E{X} =
=
=
x fX,Y (x, y) dx dy
fX,Y (x, y) dy dx
x fX (x) dx .
Correlation, an important joint moment about the origin, is dened for random variables
X and Y as
Corr{X, Y } = rxy = 11 = E{XY } .
(14)
We say that X and Y are uncorrelated if rxy = x y . We say that X and Y are orthogonal
if rxy = 0. Note the mixed terminology.
Given two random variables X and Y , the ij th joint central moment is
mij = E{(X x )i (Y y )j } =
(x x )i (y y )j fX,Y (x, y) dx dy
. (15)
For both joint central moments ij and joint moments about the origin mij , the order of the
moment is i + j.
The covariance between X and Y , the 2-nd order central moment, is
Cov{X, Y } = cxy = m11 = E{(X x )(Y y )} .
(16)
Note that if both X and Y are zero mean, then cxy = rxy , the correlation is equal to the
variance.
428
7.2
DT Random Processes
A DT random process is a sequence of random variables. Let X[n] denote a random process,
where the independent variable n typically represents sample time. Then, for each integer
value n, X[n] is a random variable. We denote a realization of X[n] as x[n]. Given a
realization x[n], we can treat it as a signal just as we have done throughout the Course to
this point. For example, we can take a DTFT of the realization to determine its frequency
content, or we can lter it with a frequency selective DT LTI system. However, we are
usually more interested in the characterization or processing of all possible realizations, then
just one that we may have already observed. After all, the one we observe may not be
representative on a lot of other ones we may observe.
In this Section we characterize in a useful way the probabilistic nature of a random process.
We begin with a few examples
Example 7.6: Discrete-time white noise You have likely heard the expression
white noise before. Qualitatively, this term suggest totally random in some sense.
The gure below illustrates one possible realization of a white noise random
process N[n].
n[n]
....
....
n
Figure 130: Illustration of a realization of DT white noise.

It is drawn to give a visual sense of randomness. We will see below that, by
2
denition, white noise means that E{N[n]} = 0; n, E{|N[n]|2 } = n , and
E{N[n] N [m]} = 0; n = m. That is, all the random variables that constitute
the random process are zero-mean and they all have the same variance, and all
pairs of these random variables are uncorrelated.
So why the term white? We will answer this question as an Example in Section
7.3.
Example 7.7: A complex sinusoidal random process A complex sinusoidal random process X[n] has the form
X[n] = A ej(n+)
where in general A, and are random variables. A realization of X[n] will be a
complex sinusoid, whose magnitude, frequency and phase will be some realization
of, respectively, A, and .
429
Consider, for example, the case where A and are constant (i.e. known, nonrandom), and is uniformly distributed with pdf
f () =
1
2
0 < 2
otherwise
The mean of each random variable that constitutes this random process is
x[n] = E{X[n]} = E{A ej(n+) } = Aejn E{ej }
2
1
= Aejn
ej d = 0 .
2 0
In general, the mean of the random variables is dierent at dierent times. This
one has constant (zero) mean for all time, i.e. x[n] = x = 0.
The correlation between any two random variables, say at times n and m, is
E{X[m] X [n]} = E{Aej(m+) Aej(n+) } = A2 ej(mn) E{ej ej }
= A2 ej(mn) E{1} = A2 ej(mn)
We see from this expression that for m = n, i.e. when we are correlating a random
2
sample with itself, we just get the variance of that random variable, x[n] = A2 .
2
2
Note that this is not a function of n, i.e. x[n] = x = A2 . Also note that
the correlation between two random variables, at times n and m, is a function of
only the distance in time m n between them. It is not a function of where the
samples are in time.
Partial Characterizations of DT Random Processes: The Mean & Correlation Functions

A complete probabilistic description of a DT random process consists of the set of all
joint pdf s of all combinations of the random variables that constitutes the random process.
In many situations all of this information is not available, and in most situations it is not
necessary for the eective processing of the random process. An eective, common and
somewhat general representation of random processes is in terms of moments. Although
higher order moments are sometimes, in the vast majority of applications the 1-st and 2-nd
order moments of a random process can be eectively employed. Here we describe the 1-st
and 2-nd order moment descriptions of DT random processes.
The Mean Function: The mean function of a DT random process X[n] is dened as
x[n] = E{X[n]} =
x[n] fX[n] (x[n]) dx[n] ;
n .
(17)
It is the function of means of all the random variables that constitute the random
process. In general, as the notation x[n] implies, the mean is time varying.
430
The Correlation Function: The correlation function of a DT random process X[n] is

dened as
Rxx [m, n] = E{X[m] X [n]} =
x[m] x [n] fX[m],X[n] (x[m], x[n]) dx[m] dx[n] ; m, n.
(18)
It is the function of all correlations between the random variables that constitute the
random process. It is a two dimensional function of the times m and n of the two
samples which are being correlated.
The Covariance Function: The covariance function of a DT random process X[n] is
dened as
Cxx [m, n] = E{(X[m] x[m] ) (X[n] x[n] ) }
= Rxx [m, n] x[m]
x[n] .
(19)
(20)
Eq (20) can be derived from Eq (19) using the linearity property of the expectation.
Note that if the random process is zero-mean for all time, then Cxx [m, n] = Rxx [m, n].
One way that random processes are characterized is in terms of properties of their mean and
correlation functions. We now identify the most common category of DT random processes.
Wide-Sense Stationary DT Processes
Qualitatively, stationarity of a random process means that its probabilistic characteristics
do not change with time. There are dierent types of stationarity corresponding to dierent
characteristics. Stationarity in the mean means
x[n] = E{X[n]} = x
(21)
That is, the mean is not a function of time n. Wide-sense stationarity means stationarity in
the mean plus
Rxx [n + l, n] = E{X[n + l] X [n]} = Rxx [l] .
(22)
That is, in addition to the mean is not being function of time n, the autocorrelation function
is not a function of time n, but only a function of the dierence in time l between samples
being correlated. This distance, l, is termed the lag.
Example 7.8: Discrete-Time White Noise In Example 7.6 DT white noise
was described as a DT random process, say N[n], with E{N[n]} = 0; n,
2
E{|N[n]|2 } = n , and E{N[n] N [m]} = 0; n = m. We now recognize that
with these properties, white noise is zero mean, i.e.
n[n] = n = 0 ,
with correlation function
2
Rnn [n + l, n] = Rnn [l] = n [l] .
So, white noise is wide-sense stationary.
431
Example 7.9: A complex sinusoidal random process In Example 7.7 we considered the DT random process
X[n] = A ej(n+)
where A and are constant, and is uniformly distributed over values 0 <
2.
We observed that x[n] = 0, i.e. we now say that the random process has zero
mean. We also concluded that E{X[m] X [n]} = A2 ej(mn) . That is, the
correlation function is Rxx [n + l, n] = Rxx [l] = A2 ejl . So the correlation
function is a function of only the lag (the distance in time between the random
variables).
As in Example 7.8, this random process is wide-sense stationary.
Example 7.10: Another complex sinusoidal random process As in Example 7.9,
consider
X[n] = A ej(n+)
where is still constant and is uniformly distributed over values 0 < 2,
2
but now let A be a Gaussian random variable with zero mean and variance a .
Assume that A and are statistically independent. Determine the mean and
correlation functions. Is this random process wide-sense stationary?
Solution: Note that fA, (a, ) = fA (a) f (), since A and are statistically
independent. The mean function is
x[n] =
= ejn
1
2
x[n] fA, (a, ) da d = ejn

2
0
1
2
2a
2 /2 2
a
ea
ej f () d
a fA (a) da
= 0.
That is, both integrals in the last line are zero. This random process is zero-mean.
The correlation function can be shown to be
2
Rxx [n + l, n] = Rxx [l] = a ejl
So this DT sinusoidal random process is also wide-sense stationary.
432
Example 7.11: A real-valued sinusoidal random process Consider the random

process
X[n] = A cos(n + )
where A and are constants and is uniformly distributed over values
0 < 2. It can be shown that
x[n] = x = 0 ,
and
Rxx [n + l, n] = Rxx [l] = A2 cos(l) .
So, yet again, the sinusoidal random process is wide-sense stationary.
Not all sinusoidal random processes are wide-sense stationary. The above examples were
selected because we are only interested in wide-sense stationary processes in this overview.
Example 7.12: Additive White Gaussian Noise (AWGN) A very common type
of random process, often observed at the sampled output of a sensor (or the
sampled output of a preamplier connected directly to a sensor) when there is
no signal received by the sensor, is AWGN.
White means that all the sample are zero-mean, uncorrelated with one another,
and all have the same variance (i.e. see Example 7.8).
Gaussian means that each sample is Gaussian distributed. Let N[n] denote the
AWGN process. Then,
fN [n] (n[n]) =
1
2
2n
en
2 [n]/2 2
n
(23)
and since uncorrelated Gaussian random variables are statistically independent22 ,

the joint pdf of any set of samples is the product of their individual pdf s, each of
the Eq (23) form. So for this random process we know the complete statistical
description.
Additive implies that, in the presence of a signal, the noise is added to it. It also
typically implies that the signal, if also a random process, is uncorrelated with
the noise. Let X[n] = S[n] + N[n] be the sampled sensor output, where S[n] is a
signal and N[n] is AWGN. Then it is easy to show that, since S[n] and N[n] are
uncorrelated,
Rxx [l] = Rss [l] + Rnn [l]
l ,
2
where we already know that Rnn [l] = n [l].
22
Hopefully you recall this from your Probability course. If not, trust me on this one.
433
Signal-to-Noise Ratio (SNR): For wide-sense stationary random processes, as with other
power signals, SNR is dened as the ration of the signal power to the noise power. For a
random process X[n] = S[n] + N[n], consisting of wide-sense stationary signal S[n] and noise
N[n], the SNR is
SNR =
Rss [0]
;
Rnn [0]
SNRdb = 10 log10 (SNR) .
(24)
Example 7.13: Let X[n] = S[n] + N[n], where the signal S[n] is a complex
sinusoidal process as described in Example 7.10, and N[n] is AWGN with variance
2
2
2
n . Let a = 20 and n = 10. The SNR is
2
20
a
Rss [0]
= 2 =
= 2;
SNR =
Rnn [0]
n
10
7.3
SNRdb 3 dB .
Power Spectral Density
Let X[n] be a wide-sense stationary random process. Let x[n] be a realization. Denote the
DTFT of a 2N + 1 sample window of the random process as
N
j
x[n] ejn
XN (e ) = XN () =
(25)
n=N
The power spectral density is dened as

Sxx (ej ) =
lim
1
E{|XN (ej )|2 } .
2N + 1
(26)
The power spectral density is the expected value of the magnitude-squared of the DTFT of
a window of the random process, as the window width approaches innity. This denition
captures what we want as a measure of the frequency content of a DT random process.
Lets take an alternative view of Sxx (ej ). First consider the term on the right of the Eq
(26), without the limit and expectation:
1
|XN (ej )|2
2N + 1
1
2N + 1
=
=
1
2N + 1
1
=
2N + 1
x[n] ejn
n=N
x [l] ejl
l=N
x[n] x [l] ej(nl)

n=N
l=N
n+N
n=N
m=nN
x[n] x [n m] ejm .
Taking the expected value, we have

1
1
E{|XN (ej )|2 } =
2N + 1
2N + 1
n+N
Rxx [m] ejm .

n=N
m=nN
(27)
434
Now, taking the limit as N , we have
1
E{|XN (ej )|2 }
N 2N + 1
N
1
Rxx [m] ejm
= lim
N 2N + 1
n=N m=
Sxx (ej ) =
=
=
lim
m=
jm
Rxx [m] e
lim
1
2N + 1
1
n=N
Rxx [m] ejm .
m=
Thus, the power spectral density and the autocorrelation function of a wide-sense stationary
random process form the DTFT pair
Sxx (ej ) =
Rxx [l] ejl
(28)
l=
Rxx [l] =
1
2
Sxx (ej ) ejl d
2
Example 7.14 - Given the correlation function Rxx [l] = x 0.5|l|, determine and
sketch the power spectral density Sxx (ej ).
Solution:
(29)

Example 7.15 - Consider the complex sinusoidal random signal described in Example 7.10 embedded in AWGN. Determine the mean function, the correlation
function and the power spectral density.
Solution:
435
436
7.4
DT Random Processes & DT LTI Systems
Consider a wide-sense stationary random process X[n] with mean x and autocorrelation
function Rx [l], and a DT LTI system with impulse response h[n]. For x[n], a realization of
X[n], the input/output relationship is the convolution sum
y[n] =
k=
h[k] x[n k]
(30)
1. The output mean function is:
E{y[n]} =
k=
h[k] E{x[n k]}
y = x
h[k]
(31)
k=
2. The output correlation function is:

Ryy [l] = E{Y [n + l] Y [n]}
=
m= k=
m=
= h[l]
h [m]
m=
h[k] h [m] E{x[n + l k] x [n m]}
k=
h[k] Rxx [(l + m) k]
h [m] (h[l] Rxx [l + m])
h [m] Rxx [l + m]
m=
= h[l]
i=
h [i l] Rxx [i]
= h[l] h [l] Rxx [l]
(32)
3. The output power spectral density: From the above result on DT LTI system input/output autocorrelation functions, and DTFT properties, it can be shown that
Syy (ej ) = Sxx (ej ) |H(ej )|2 .
(33)

2
Example 7.16: Consider a zero-mean AWGN input X[n], with variance n , and
a DT LTI system with impulse response h[n] = [n] + [n 1]. Determine the
correlation function and power spectral density of the output Y [n].
Solution:
437
438
Example 7.17: Let input X[n] be as in Example 7.15, and let the DT LTI system
impulse response be h[n] = an u[n], where |a| < 1. Determine the correlation
function and power spectral density of the output Y [n].
Solution:
1
Example 7.18: Let h[n] = N (u[n] u[n N]), and let X[n] be a wide-sense
2
stationary complex sinusoidal process with correlation function Rxx [l] = x ej0 l
where |0 | . Determine the correlation function and power spectral density
of the output Y [n].
Solution:
7.5
439
Problems
1. Correlation function, power spectral density problems (feel free to use Fourier Transform tables):
(a) Given Rxx [l] = [l + 1] + 2[l] [l 1], determine Sxx (ej ).
(b) Given Rxx [l] = A2 cos(0 l), describe Sxx (ej ); < . Assume < 0
.
(c) Given the following power spectral density, determine to autocorrelation function
Rxx [l]:
Sxx (ej ) =
1
0
||
2
< ||
2
(d) Given the following power spectral density, determine to autocorrelation function
Rxx [l]:
Sxx (ej ) =
0
1
||
2
< ||
2
(e) Given power spectral density Syy (ej ) = 2 + 2 cos(2), determine correlation
function Ryy [m].
2. Assume random processes are DT and wide-sense stationary:
(a) Given X[n] with correlation function Rxx [l] = 3[l] ([l + 1] + [l 1]), determine
and plot the power spectral density Sxx (ej ). What percentage of X[n] power is
in the frequency band 0 ?
(b) Given Y [n] with power spectral density

Syy (ej ) =
1 + 2
1 2
0
0
determine the correlation function Ryy [l]. What is the power of Y [n]? (Hint: use
the DTFT pair table and linearity.)
3. Assume random processes are DT and wide-sense stationary, and systems are DT LTI:
(a) By performing the required convolutions, determine and plot the correlation function of the output Y [n] of a system with impulse response h[n] = p3 [n] and input
X[n] with correlation function Rxx [l] = 2[l] + [l + 1] + [l 1].
(b) Consider the following system:
y[n] .5 y[n 2] = x[n] + .5 x[n 2] .

i. For input correlation function Rxx [l] = 4 [l], determine the expression for
the output power spectral density.
440
ii. For input correlation function Rxx [l] = 3 ej(/2)l , determine the output
correlation function.
4. Given the 3-point averager considered in the rst week of Practicum 5, i.e. with I/O
dierence equation
y[n] =
1
(x[n] + x[n 1] + x[n 2]) ,
3
with a wide-sense stationary random process input. For each of the following inputs,
determine the output correlation function and power spectral density.
(a) X1 [n] with correlation function Rx1 ,x1 [l] = [l] + 0.5 [l 1] + 0.5 [l + 1].
(You must rst compute Ry1 ,y1 [l] directly using graphical convolution to receive
credit for this problem.)
(b) X2 [n] with correlation function Rx2 ,x2 [l] = ej(/4)l . (You must rst compute
Sy2 ,y2 () using the frequency response H(ej ) to receive credit for this problem.)
(c) X3 [n] with correlation function Rx3 ,x3 [l] = cos( 2 l). (You can do this whichever
3
way you wish.)
5. A DT LTI system has impulse response h[n] = p4 [n]. The input is zero-mean white
2
noise with variance x .
(a) What is the output mean function y [n]?
(b) What is the correlation function of the output?
6. Correlation functions and LTI systems:
(a) Consider a wide-sense stationary random process X[n], with correlation function
Rxx [l] = 0.5 [l+1] + 1.25 [l]0.5 [l1]. It is the input to a LTI system with
impulse response h[n] = 0.5n u[n]. Determine the output signals correlation
function and power spectral density.
(b) Consider a wide-sense stationary input X[n] to a LTI system with frequency
response H(ej ) = ej/2 . Let the autocorrelation function of the input be
Rxx [l] = 5[l] + a|l| . Determine the autocorrelation function Ryy [l] of the output
Y [n].
7. Consider a WSS autoregressive signal X[n] which is the output of the all-pole lter
X[n] = .5X[n 1] + W [n]
with zero-mean, unit-variance, uncorrelated input W [n]. This signal X[n] is the input
to a DT linear time invariant system described by the following dierence equation
Y [n] + .9Y [n 1] = X[n] .25X[n 2] .
Determine the power spectral density of the output Y [n].
441
8. A zero-mean random process X[n] with autocorrelation function Rxx [l] = 0.5|l| is applied to a DT LTI system with impulse response h[n] = [n] + [n 1].
(a) What is the power spectral density of the input?
(b) What is the correlation function of the output?
(c) What is the power spectral density of the output?
9. Consider a DT LTI system with impulse response h[n] = .5n u[n] and let wide-sense
stationary random process X[n], with autocorrelation function Rxx [m] = 3 [m], be
the input. Determine the output autocorrelation function Ryy [m].
10. The power of any wide-sense stationary random process X[n] is
Px = Rxx [0] =
1
2
Sxx (ej ) d .
Let such an X[n], with autocorrelation function Rxx [m] = 1 [m+2]+[m]+ 1 [m2],
2
2
be the input to an ideal lowpass lter with frequency response
H(ej ) =
1
0
||
2
< ||
2
Determine the power of the output wide-sense stationary random process Y [n].
11. Determine the mean and the autocorrelation function for the random process
X[n] = V [n] + 3V [n 1]
where V [n] is a sequence of independent random variables with mean v and variance
2
v . Is X[n] wide-sense stationary?
2
12. Consider a zero-mean AWGN input X[n], with variance n = 1, and a DT LTI system
with impulse response h[n] = 1n p4 [n]. Determine the correlation function and power
spectral density of the output Y [n].
13. Consider a wide-sense stationary input with correlation function Rxx [l] = |l| with
|| < 1, and a DT LTI system with impulse response h[n] = [n] [n1]. Determine
the correlation function and power spectral density of the output Y [n]. Also nd the
output mean function.
14. In all parts below, the input is wide-sense stationary random process & the system is
some LTI system.
a) Given zero-mean input X1 [n], with correlation function Rx1 x1 [l] = 5 (0.9)|l| , can
the output correlation function be
(a) Ry1 y1 [l] = Rx1 x1 [l], or
(b) Ry1 y1 [l] = Rx1 x1 [l 5]?
Why?
442
b) Given zero-mean input X2 [n], with correlation function Rx2 x2 [l] = 4 ej(/2)l , to
a system with impulse response h2 [n] = (0.8)n u[n], is the output power density
spectrum
(a) Sy2 y2 (ej ) = 0; , or
8
(b) Sy2 y2 (ej ) = 1.64 ( ); ?
2
Why?
c) Given input X3 [n], with mean x3 = 5 and correlation function Rx3 x3 [l] = 25 +
(0.9)|l| , to a system with impulse response h3 [n] = [n] [n 1], determine
Ry3 y3 [0], the zero lag of the output correlation function. (Hint: Find h[n] h[l].
Then convolve it, as much as needed, with Rx3 x3 [l].)
15. Consider a DT LTI system with impulse response h[n] = [n] [n1], with random
process input X(t) resulting in random process output Y [n].
(a) Determine and plot h[n] h[n] and |H(ej )|2 .
(b) Given that Rxx [l] = 4 [l], determine Ryy [l] and Syy (ej ).
(c) Given that
j
Sxx (e ) =
4 () + 5 ( )
2
0
<
otherwise over <
determine Ryy [l] and Syy (ej ).

(d) Given that Rxx [l] = 6
sin( l)
2
,
l
determine Syy (ej ).
16. Consider a DT LTI system frequency response

H(ej ) =
1
0
||
2
< ||
2
(a) Given that the input is the wide-sense stationary process X[n] from Example
6.6,8 in the Course Notes, determine the autocorrelation function and the power
(b) Given that the input is the wide-sense stationary process X[n] from Examples
6.7,9 the Course Notes, with frequency 0 = , determine the autocorrelation
3
function and the power spectral density of the output Y [n].
(c) Given that the input is the wide-sense stationary process X[n] from Example 6.11
the Course Notes, with frequency 0 = 2 , determine the autocorrelation function
3
and the power spectral density of the output Y [n].
17. Consider a DT LTI system with impulse response h[n] = [n] + [n 1].
(a) Given that the input is the wide-sense stationary process X[n] from Example
6.6,8 in the Course Notes, determine the autocorrelation function and the power
443
(b) Given that the input is the wide-sense stationary process X[n] from Examples
6.7,9 the Course Notes, with frequency 0 = , determine the autocorrelation
3
(c) Given that the input is the wide-sense stationary process X[n] from Examples
6.11 the Course Notes, with frequency 0 = 2 , determine the autocorrelation
3
18. DT Random Processes:
a) Consider a DT LTI system with impulse response h[n] = 2[n 2] and a random
process input with correlation function Rxx [l] = 0.9|l| . Find the output signals
correlation function Ryy [l] and power spectral density Syy (ej )
b) Consider a DT LTI system with random process input with correlation function
Rxx [l] = 4 [l], and frequency response H(ej ) = 3 e|/2| ; . Find the
output signals correlation function Ryy [l] and power spectral density Syy (ej ).
19. Similar to the DT case, given a DT wide-sense stationary random process X(t), its
autocorrelation function Rxx ( and power spectral density Sxx (j) are related as a
CTFT pair. The power spectral density of a CT wide-sense stationary process is
Sxx (j) = 1/( 2 + 9). Determine the autocorrelation function.
20. Let X[n] be a wide-sense stationary DT random process with correlation function
Rxx [l]. If X[n] is obtained by sampling a wide-sense stationary CT random process
Xa (t) with CT correlation function Rxa xa ( ), then it is well know that
Rxx [l] = Rxa xa (lT )
where T is the sample interval (i.e. the DT correlation function is the CT correlation
1
function sampled at rate fs = T ).
Let Rxa xa ( ) = 9 e| | , and T = 1. Determine Rxx [l] and the corresponding power
spectral density Sxx (ej ).
21. Modulation, Ideal CT Bandpass Filtering, and CT Random Processes: Consider the
CT system illustrated below, for which the input signal x(t) has CTFT X(j) as
given. The additive noise N(t) is a CT random process with autocorrelation function
Rnn ( ) = 4 ( ).
a) Determine Xs (j), and the CTFTs of the signal components of v(t) and y(t).
(Since the lter is linear, you can ignore the noise for this question).
b) For CT random processes, the power spectral density is the CTFT of the autocorrelation function. Also, for a CT LTI system with frequency response H(j)
the output power spectral density is the the input power spectral density times
|H(j)|2. Also, for a random process Z(t) with power spectral density Szz (j),
the power spectral density of ej0 t Z(t) is Szz (j( 0 )). Determine the power
spectral densities of the noise components of v(t) and y(t). (Since the lter is
linear, you can ignore the signal for this question).
444
x(t)
X(j )
x s (t)
x s (t) + N(t)
500
500
cos(2000 t)
N(t)
H(j )
v(t)
y(t)
H(j )
j2000 t
1000
3000
22. Consider a DT LTI system with impulse response h[n] = 2 [n + 3] + 2 [n 3]

and wide-sense stationary random process input.
(a) Determine the system frequency response H(ej ), and plot its magnitude and
phase for .
(b) If the input X1 [n] has correlation function Rx1 x1 [l] = 1+2 ejl/6 +3 ejl/3 +4 ejl/2 ,
determine the power spectral density Sy1 y1 (ej ) and correlation function Ry1 y1 [l]
of the output y1 [n].
(c) Using graphical convolution in the time domain, determine the output correlation
function Ry2 y2 [l] for input correlation function Rx2 x2 [l] = [l + 1] + 2 [l] + [l 1].
23. Consider a DT LTI system with unknown I/O characteristics that we wish to identify.
The following questions are independent of one another.
(a) Given that we know the input is x1 [n] = [n] and the corresponding output has
1
DTFT Y1 (ej ) = 1 + 0.9 ej , determine all you can about h[n].
(b) Given that we know the input is x2 [n] = [n] and the corresponding output has
1
z-transform Y2 (z) = 1 + 0.9 z 1 , determine all you can about h[n].
(c) Given that we know the input is x3 [n] =
has DTFT
Y3 (ej ) =
1
1 + 0.9 ej
sin(n/2)
n
and the corresponding output
|| /2
/2 < ||
determine all you can about H(ej ).

(d) Say you know that the input is a wide-sense stationary random process with
correlation function Rxx [l] = 0.9 [l + 1] + 1.81 [l] + 0.9 [l 1], and that the
correlation function of the output is Ryy [l] = [l], determine all you can about
H(ej ).
24. DT random Processes and DT LTI Systems:
(a) Given a DT LTI system with impulse response ha [n] = (1)n p5 [n] with
wide-sense stationary input xa [n] with correlation function Rxa xa [l] = [l] +
445
0.5 [[l + 1] + [l 1]], determine and accurately plot the output correlation
function Rya ya [l]. Solve this problem using graphical convolutions, showing all
work.
(b) Given a DT LTI system with impulse response hb [n] = [n] + [n 2] with
wide-sense stationary input xb [n] with correlation function Rxb xb [l] = 5 [l] +
3 cos(0.5l), determine Ryb yb [l] and Syb yb (ej ).
25. Consider a DT LTI system with impulse response h[n] = [n] [n 2] + [n 4] with
input wide sense stationary process X[n] and output Y [n].
(a) Graphically determine g[n] = h[n] h[n]. Also determine G(ej ) = |H(ej )|2 .
2
(b) Given that the input correlation function is Rxx [l] = x [l], determine the output
correlation function Ryy [l] and corresponding power spectral density Syy (ej ).
(c) Given that the input correlation function is Rxx [l] = 7ejl , determine the output
correlation function Ryy [l] and corresponding power spectral density Syy (ej ).
26. Consider an FIR lter with transfer function H(z) = 0.5z + 1 0.5z 1 .
a) What is the lter impulse response h[n]? Is this lter stable? Causal? Why?
b) What is the lter frequency response H(ej )? Using an Eulers identity, write
H(ej ) explicitly as a real-valued function. Sketch H(ej ).
c) Determine the output y1 [n] due to input x1 [n] = 6.
d) Determine the output y2 [n] due to input x2 [n] = 4 ej(/2)n .
e) Determine the output correlation function Ry3 y3 [l] due to a random input with
correlation function Rx3 x3 [l] = 3 ejn .
27. A DT LTI system with impulse response h[n] = 0.9n u[n 2] has an output which is
a random signal with correlation function Ryy [l] = 5 [l].
a) Using graphical convolution, determine f [l] = h[l] h[l]. Note that f [l] is symmetric (i.e. f [l] = f [l]).
b) Using your results from part a), and the DTFT pair table, determine |H(ej )|2
(i.e. the DTFT of f [l]).
c) What is Syy (ej ), the output power spectral density? Using this and your result
from part b), determine the input power spectral density Sxx (ej ).
d) Using your result from part c), determine the input correlation function Rxx [l].
h[n] = [n] 0.4 [n 1] + 0.2 [n 2]
(34)
and random process input x[n]. For each of the following input correlation functions,
determine the output correlation function and output power spectral density.
(a) Rxx [l] = 0.5 [l + 1] + [l] + 0.5 [l 1].
(b) Rxx [l] = 4 cos((/2)l).

Call 3220

Uploaded by

Copyright:

Available Formats

Call 3220

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Call 3220

Uploaded by

Copyright:

Available Formats

What are some examples of signal processing applications discussed in the document?

What are some examples of signal processing applications discussed in the document?

What are the basic properties of discrete-time and continuous-time signals and operators discussed?

What are the basic properties of discrete-time and continuous-time signals and operators discussed?

c Kevin Buckley - 2013

ECE3220 Signal Processing

yc (t), y[n] = yc (nT)

Magnitude Response (dB)

c Kevin Buckley - 2013

c Kevin Buckley - 2013

2 Time Domain Analysis of LTI Systems

c Kevin Buckley - 2013

CT LTI System Issues . . . . . . . . .

3 Continuous Time Transforms

5 Discrete Time Transforms

c Kevin Buckley - 2013

The DTFT Representation of Periodic Signals .

7 Introduction to Random Processes

c Kevin Buckley - 2013

c Kevin Buckley - 2013

A DT LTI system and the convolution sum. . . . . . . . . . . . . . . . . . . 81

c Kevin Buckley - 2013

c Kevin Buckley - 2013

Pole/zero plot, and corresponding magnitude response, for Example 6.7. . .

c Kevin Buckley - 2013

c Kevin Buckley - 2013

c Kevin Buckley - 2013

This Course is an introduction to the theory and methods of:

c Kevin Buckley - 2013

Chapter 1 Objective Checklist

c Kevin Buckley - 2013

Motivation & Background

c Kevin Buckley - 2013

c Kevin Buckley - 2013

Signal Processing Examples

yc (t), y[n] = yc (nT)

Figure 1: An illustration of sampling.

c Kevin Buckley - 2013

Figure 2: Ideal sampling and reconstruction block diagrams.

Figure 3: A simple parallel RC circuit.

c Kevin Buckley - 2013

Consider the problem of transmitting a continuous-time signal xc (t) over a communication

Figure 4: A multipath communication channel.

c Kevin Buckley - 2013

An FIR Filter Example

We will explain this terminology later.

c Kevin Buckley - 2013

A Simple Discrete-Time (DT) System

Figure 6: A simple DT system.

c Kevin Buckley - 2013

c Kevin Buckley - 2013

d) Find the response yd [n] to input

when the initial condition is yd [1] = 5.

Course Objective, Linear Combinations & Two Basic Concepts

c Kevin Buckley - 2013

c Kevin Buckley - 2013

c Kevin Buckley - 2013

Robert Pirsig, Ayn Rand