Communications S&S

LEVEL 3 3015 COMMUNICATIONS, SIGNALS
AND SYSTEMS 2003

Peter H. Cole
January 5, 2004
Contents
1 INTRODUCTION 1
1.1 Course Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.4 Access to Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.5 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.6 Essential Knowledge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.7 Making Use of the Lectures . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.8 Other Goodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 REVISION 3
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2.1 SI Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2.2 Quantities derived by dierentiation . . . . . . . . . . . . . . . . . 3
2.2.3 Frequency and angular frequency . . . . . . . . . . . . . . . . . . . 4
2.3 RC and RL Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3.1 Things not possible . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3.2 Establishing a capacitor voltage . . . . . . . . . . . . . . . . . . . . 4
2.3.3 Discharging a capacitor . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3.4 Establishing an inductor current . . . . . . . . . . . . . . . . . . . . 5
2.3.5 De-energising an inductor . . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Phasor Analysis for Linear a.c. Circuits . . . . . . . . . . . . . . . . . . . . 7
2.4.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4.2 Particular cautions . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4.3 Peak value and r.m.s. value phasors . . . . . . . . . . . . . . . . . . 8
2.4.4 Properties of a sinusoid . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4.5 Unusual waveforms . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4.6 Resistance and resistivity . . . . . . . . . . . . . . . . . . . . . . . . 9
2.5 Characterisation of N-Port Networks . . . . . . . . . . . . . . . . . . . . . 9
2.5.1 Denitions of port variables . . . . . . . . . . . . . . . . . . . . . . 9
2.5.2 Denition of impedance matrix . . . . . . . . . . . . . . . . . . . . 10
2.5.3 Denition of admittance matrix . . . . . . . . . . . . . . . . . . . . 11
2.5.4 Relations between impedance and admittance parameters . . . . . . 12
2.5.5 Pathological cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5.6 The reciprocity theorem . . . . . . . . . . . . . . . . . . . . . . . . 12
2.6 Thevenins and Nortons Theorems . . . . . . . . . . . . . . . . . . . . . . 12
i
ii CONTENTS
2.6.1 Thevenin and Norton networks . . . . . . . . . . . . . . . . . . . . 12
2.6.2 Determination of Thevenin and Norton parameters . . . . . . . . . 14
2.6.3 Pathological cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.6.4 Networks with non-linear elements . . . . . . . . . . . . . . . . . . 16
2.6.5 Frequency domain equivalent circuits . . . . . . . . . . . . . . . . . 16
2.6.6 Warning against undue generalisation . . . . . . . . . . . . . . . . . 18
2.7 Tuned Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.7.1 Ideal series resonant circuit . . . . . . . . . . . . . . . . . . . . . . 19
2.7.2 Ideal parallel resonant circuit. . . . . . . . . . . . . . . . . . . . . . 20
2.7.3 Practical parallel resonant circuit . . . . . . . . . . . . . . . . . . . 22
2.7.4 Common features . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.7.5 Practical parallel to ideal parallel circuit transformation . . . . . . 24
2.8 Maximum Power Transfer Theorem . . . . . . . . . . . . . . . . . . . . . . 24
2.8.1 Available source power . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.8.2 Logarithmic expression of power ratios . . . . . . . . . . . . . . . . 25
2.8.3 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.8.4 Insertion gain and insertion loss . . . . . . . . . . . . . . . . . . . . 26
3 INTRODUCTION TO PROBABILITY 27
3.1 Aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Set Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.1 Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.2 Belonging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.3 Subset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.4 Set equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.5 Universal set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.6 Null set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.7 Venn diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.8 Set operations and concepts . . . . . . . . . . . . . . . . . . . . . . 29
3.2.9 Mutual exclusivity . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.10 Collective exhaustion . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.11 Theorems on sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 Applying set theory to probability . . . . . . . . . . . . . . . . . . . . . . . 33
3.3.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3.2 Some denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3.3 Comparisons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3.4 Further comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3.6 Fundamental remark . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3.7 Theorem on event spaces . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3.8 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3.9 A modest proposal . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3.10 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.4 Probability Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.5 Probability Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.5.1 Probabilities with the union of disjoint sets . . . . . . . . . . . . . . 36
3.5.2 Probabilities with the union of mutually exclusive sets . . . . . . . 37
CONTENTS iii
3.5.3 Probabilities of outcomes combined to form an event . . . . . . . . 37
3.5.4 Theorem on equally likely outcomes . . . . . . . . . . . . . . . . . . 37
3.5.5 Theorem on no outcome . . . . . . . . . . . . . . . . . . . . . . . . 37
3.5.6 Theorem on complementary outcomes . . . . . . . . . . . . . . . . 37
3.5.7 Theorem on arbitrary sets . . . . . . . . . . . . . . . . . . . . . . . 37
3.5.8 Theorem on partially ordered sets . . . . . . . . . . . . . . . . . . . 37
3.5.9 Theorem on adding probabilities of intersections . . . . . . . . . . . 37
3.5.10 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.6 Conditional Probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.6.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.6.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.6.3 Fundamental formula for conditional probability . . . . . . . . . . . 39
3.6.4 Law of total probability . . . . . . . . . . . . . . . . . . . . . . . . 39
3.6.5 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.7 Bayes theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.8 Independence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.8.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.9 Sequential Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.9.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.9.2 Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.10 Counting Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.10.1 Fundamental Principle of Counting . . . . . . . . . . . . . . . . . . 42
3.10.2 Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.10.3 Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.10.4 Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.11 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4 RANDOM VARIABLES 45
4.1 Review of Probability Theory . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.1.1 Probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.1.2 Conditional probability . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.1.3 Bayes rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.1.4 The Binomial Distribution . . . . . . . . . . . . . . . . . . . . . . . 46
4.2 Random Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.2.2 Cumulative Distribution Function . . . . . . . . . . . . . . . . . . . 46
4.2.3 Probability Density Function . . . . . . . . . . . . . . . . . . . . . 47
4.2.4 Joint Distribution and Density Functions . . . . . . . . . . . . . . . 48
4.2.5 Conditional Density Function . . . . . . . . . . . . . . . . . . . . . 49
4.3 Expectation and Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.1 Mean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.2 Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.3.3 Central moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.3.4 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.4 The Gaussian Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.4.1 One Dimensional Gaussian Distribution . . . . . . . . . . . . . . . 51
4.4.2 Two Dimensional Gaussian Distribution . . . . . . . . . . . . . . . 52
iv CONTENTS
4.5 Transformation of Random Variables . . . . . . . . . . . . . . . . . . . . . 53
4.5.1 Single Variable Transformations . . . . . . . . . . . . . . . . . . . . 53
4.5.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.5.3 Two-variable transformations . . . . . . . . . . . . . . . . . . . . . 54
4.5.4 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.5.5 Application to Gaussian variables . . . . . . . . . . . . . . . . . . . 56
4.6 Characteristic Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.6.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.6.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.7 Second Characteristic Function . . . . . . . . . . . . . . . . . . . . . . . . 58
4.7.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.7.2 Cumulants of the Gaussian distribution . . . . . . . . . . . . . . . . 58
4.7.3 Further properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.8 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5 RANDOM PROCESSES 61
5.1 General Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.2 First and Second Order Statistics . . . . . . . . . . . . . . . . . . . . . . . 62
5.2.1 First order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.2.2 Second order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.3 Mean, Autocorrelation & Autocovariance . . . . . . . . . . . . . . . . . . . 63
5.3.1 Mean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.3.2 Autocorrelation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.3.3 Autocovariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.3.4 Variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.3.5 Commentary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.4 Poisson Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.4.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.4.2 How the argument will unfold . . . . . . . . . . . . . . . . . . . . . 64
5.4.3 The Poisson point process . . . . . . . . . . . . . . . . . . . . . . . 66
5.4.4 The X process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.4.5 Additive properties of intervals . . . . . . . . . . . . . . . . . . . . 69
5.4.6 The Y process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.4.7 The P process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.4.8 The Poisson impulse process . . . . . . . . . . . . . . . . . . . . . . 72
5.4.9 Return to the X process . . . . . . . . . . . . . . . . . . . . . . . . 73
5.4.10 Random Walk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.4.11 Wiener-Levy Process . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.4.12 The reection principle . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.5 Sine wave and Random Noise . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.6 Cross-correlation and Covariance . . . . . . . . . . . . . . . . . . . . . . . 79
5.6.1 Real processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.6.2 Complex processes . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.7 Stationary Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.7.1 Denition: Stationary in the strict sense . . . . . . . . . . . . . . . 80
5.7.2 First order statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.7.3 Second order statistics . . . . . . . . . . . . . . . . . . . . . . . . . 80
CONTENTS v
5.7.4 Denition: Stationary in the wide sense . . . . . . . . . . . . . . . . 81
5.8 Transformation of Random Processes . . . . . . . . . . . . . . . . . . . . . 81
5.8.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.8.2 Memoryless Transformations . . . . . . . . . . . . . . . . . . . . . . 81
5.8.3 Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.8.4 Example: Dierentiation operator . . . . . . . . . . . . . . . . . . . 84
5.8.5 Further example: Poisson impulse process . . . . . . . . . . . . . . 84
5.8.6 Simplied results for stationary processes . . . . . . . . . . . . . . . 85
5.9 Ergodicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.9.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.9.2 Time averages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.9.3 Time and ensemble statistics . . . . . . . . . . . . . . . . . . . . . . 86
5.9.4 Denition of an ergodic system . . . . . . . . . . . . . . . . . . . . 86
5.9.5 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.10 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6 POWER SPECTRA OF STATIONARY PROCESSES 91
6.1 Correlation Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.1.1 Recall of denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.1.2 Properties of correlation functions . . . . . . . . . . . . . . . . . . . 91
6.1.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.2 The Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2.1 Example: Rectangular pulse . . . . . . . . . . . . . . . . . . . . . . 93
6.2.2 Example: Exponential pulse . . . . . . . . . . . . . . . . . . . . . . 95
6.2.3 Example: Triangular pulse . . . . . . . . . . . . . . . . . . . . . . . 95
6.2.4 Hermitian property . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.2.5 The Rayleigh energy theorem . . . . . . . . . . . . . . . . . . . . . 96
6.2.6 Properties of Fourier transforms . . . . . . . . . . . . . . . . . . . . 96
6.2.7 Example: Double exponential pulse . . . . . . . . . . . . . . . . . . 97
6.2.8 Another example . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.3 Singularity Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.3.1 Example: Application to periodic signals . . . . . . . . . . . . . . . 99
6.4 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.5 Power Spectral Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.5.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.5.2 Wiener Khinchine theorem . . . . . . . . . . . . . . . . . . . . . . . 101
6.5.3 Application of the theorem . . . . . . . . . . . . . . . . . . . . . . . 101
6.6 Cross Power Spectral Density . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.6.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.6.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.6.3 Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.7 Linear Time Invariant Systems . . . . . . . . . . . . . . . . . . . . . . . . 103
6.7.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.8 Cyclostationary Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.8.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.8.2 Example: sine wave . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.8.3 Example: random binary signal . . . . . . . . . . . . . . . . . . . . 106
vi CONTENTS
6.8.4 Filtering property . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.9 Chapter summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
7 RANDOM NOISE SIGNALS 111
7.1 White Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.1.1 Pure white noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.1.2 Bandlimited white noise . . . . . . . . . . . . . . . . . . . . . . . . 111
7.2 Thermal Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.2.1 General principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.2.2 Lumped circuit representations . . . . . . . . . . . . . . . . . . . . 114
7.3 Shot Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.3.1 Origin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.3.2 Simple model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.3.3 More accurate model . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.3.4 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
7.4 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
8 ANALOG FILTER DESIGN 117
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
8.1.1 Uses of lters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
8.1.2 Transfer function . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
8.1.3 Impulse response . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
8.1.4 Frequency response . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
8.1.5 Finding the transfer function . . . . . . . . . . . . . . . . . . . . . 119
8.1.6 Minimum phase lters . . . . . . . . . . . . . . . . . . . . . . . . . 120
8.1.7 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
8.2 Classication of Filter Responses . . . . . . . . . . . . . . . . . . . . . . . 120
8.3 Impedance and Frequency Scaling . . . . . . . . . . . . . . . . . . . . . . . 121
8.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8.3.2 Impedance scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8.3.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
8.3.4 Frequency scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
8.3.5 Special note . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
8.4 The Approximation Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 126
8.5 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
9 LOW PASS PROTOTYPES 129
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
9.1.1 General approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
9.1.2 Signicance of the phase response . . . . . . . . . . . . . . . . . . . 129
9.1.3 Normalised low pass lters . . . . . . . . . . . . . . . . . . . . . . . 129
9.2 Getting the Element Values . . . . . . . . . . . . . . . . . . . . . . . . . . 130
9.3 Butterworth Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
9.3.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
9.3.2 Achieving a lesser passband attenuation . . . . . . . . . . . . . . . 131
9.3.3 Further frequency scaling . . . . . . . . . . . . . . . . . . . . . . . . 133
9.3.4 Achieving a required stopband attenuation . . . . . . . . . . . . . . 134
CONTENTS vii
9.3.5 Phase characteristic . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
9.3.6 Pole-zero conguration . . . . . . . . . . . . . . . . . . . . . . . . . 134
9.4 Chebychev Type I Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
9.4.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
9.4.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
9.4.3 Pole positions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
9.5 Chebychev Type II Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
9.5.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
9.5.2 Illustration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
9.5.3 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
9.6 All Pass Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
9.6.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
9.6.2 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
9.6.3 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
9.7 Bessel-Thompson Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
9.7.1 Dening property . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
9.7.2 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
9.7.3 Graphical expression of responses . . . . . . . . . . . . . . . . . . . 145
9.8 Elliptic Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
9.8.1 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
9.8.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
9.8.3 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
9.8.4 Illustration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
9.8.5 Magnitude response . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
9.8.6 Phase response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
9.8.7 Pulse response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
9.9 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
10 FILTER DESIGN AND TRANSFORMATIONS 153
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
10.1.1 General approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
10.1.2 Notation for unit angular frequency . . . . . . . . . . . . . . . . . . 153
10.2 Low Pass Filter Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
10.2.1 Design steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
10.2.2 Frequency transformation of specications . . . . . . . . . . . . . . 154
10.2.3 Design of normalised lter . . . . . . . . . . . . . . . . . . . . . . . 154
10.2.4 Scaling elements for frequency . . . . . . . . . . . . . . . . . . . . . 155
10.2.5 Scaling elements for frequency and impedance . . . . . . . . . . . . 155
10.2.6 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
10.3 The High Pass Transformation . . . . . . . . . . . . . . . . . . . . . . . . . 156
10.3.1 Structure of low pass lter . . . . . . . . . . . . . . . . . . . . . . . 156
10.3.2 Structure of a high pass lter . . . . . . . . . . . . . . . . . . . . . 156
10.3.3 Illustration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
10.3.4 Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
10.3.5 Design steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
10.3.6 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
10.4 The Band Pass Transformation . . . . . . . . . . . . . . . . . . . . . . . . 162
viii CONTENTS
10.4.1 Structure of a band pass lter . . . . . . . . . . . . . . . . . . . . . 162
10.4.2 Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
10.4.3 Design equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
10.4.4 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
10.5 The Band Stop Transformation . . . . . . . . . . . . . . . . . . . . . . . . 166
10.6 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
11 REALISATION OF PASSIVE FILTER CIRCUITS 167
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
11.2 Some important concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
11.2.1 Aspects of notation . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
11.2.2 Available source power . . . . . . . . . . . . . . . . . . . . . . . . . 167
11.2.3 Scattering Parameters . . . . . . . . . . . . . . . . . . . . . . . . . 168
11.3 Lumped L-C Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
11.3.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
11.3.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
11.3.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
11.3.4 Generalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
11.4 Admittance Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
11.4.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
11.4.2 Example: Third Order Butterworth Filter . . . . . . . . . . . . . . 175
11.5 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
12 ACTIVE FILTERS 177
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
12.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
12.1.2 Realisation strategies . . . . . . . . . . . . . . . . . . . . . . . . . . 177
12.2 First Order Transfer Functions . . . . . . . . . . . . . . . . . . . . . . . . . 177
12.2.1 Passive circuits realisations . . . . . . . . . . . . . . . . . . . . . . . 177
12.2.2 Eect of circuit loading . . . . . . . . . . . . . . . . . . . . . . . . . 178
12.2.3 Inverting amplier circuits . . . . . . . . . . . . . . . . . . . . . . . 179
12.2.4 Non-inverting amplier circuits . . . . . . . . . . . . . . . . . . . . 180
12.2.5 More complex functions . . . . . . . . . . . . . . . . . . . . . . . . 181
12.3 Second Order Transfer Functions . . . . . . . . . . . . . . . . . . . . . . . 181
12.3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
12.3.2 Low pass Sallen and Key circuit . . . . . . . . . . . . . . . . . . . . 181
12.3.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
12.3.4 Band pass Sallen and Key circuit . . . . . . . . . . . . . . . . . . . 183
12.3.5 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
12.3.6 Inverting Band Pass Circuit . . . . . . . . . . . . . . . . . . . . . . 184
12.3.7 Biquadratic Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
12.3.8 Case 1:
z
<
o
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
12.3.9 Case 2:
z
>
o
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
12.4 Switched Capacitor Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
12.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
12.4.2 Low pass lter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
12.4.3 Circuit properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
CONTENTS ix
12.5 Frequency Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
12.6 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
A REFERENCES 191
B FORMULAE AND TABLES 193
B.1 Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
B.2 Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
B.3 Trigonometric Identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
B.4 Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
B.4.1 The direct transform . . . . . . . . . . . . . . . . . . . . . . . . . . 194
B.4.2 The reverse transform . . . . . . . . . . . . . . . . . . . . . . . . . 195
B.4.3 Theorems and transforms . . . . . . . . . . . . . . . . . . . . . . . 195
C PROOFS 201
C.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
C.2 Bounds on the Autocorrelation Function . . . . . . . . . . . . . . . . . . . 201
C.3 The Wiener Khinchine Theorem . . . . . . . . . . . . . . . . . . . . . . . . 202
C.4 Analysis of the Bessel-Thompson Filter . . . . . . . . . . . . . . . . . . . . 204
D ADVICE ON STUDYING FOR THE EXAMINATION 209
E NOTES ON STUDENT ERRORS 213
E.1 Common Errors in Quiz 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
F STANDARD INTERNATIONAL TERMINOLOGY AND UNITS 217
F.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
F.2 Informal discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
F.3 The Base Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
F.4 Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
F.5 The Fundamental Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
F.6 List of Standard Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
G STOP PRESS 225
G.1 Thevenins and Nortons Theorems . . . . . . . . . . . . . . . . . . . . . . 225
G.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
G.1.2 Thevenins theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
G.1.3 Nortons theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
G.1.4 Relations between the circuits . . . . . . . . . . . . . . . . . . . . . 226
G.1.5 Commentary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
G.1.6 Cautions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
H SOME NOTES ON PHASOR ANALYSIS 229
H.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
H.2 A Simple Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
H.3 Analysis by Dierential Equations . . . . . . . . . . . . . . . . . . . . . . . 229
H.4 Disadvantageous Features . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
H.5 Representation of Sinusoids . . . . . . . . . . . . . . . . . . . . . . . . . . 231
x CONTENTS
H.6 Analysis Leading to Algebraic Equations . . . . . . . . . . . . . . . . . . . 232
H.7 Application to Circuit Analysis . . . . . . . . . . . . . . . . . . . . . . . . 232
H.8 The Physical and Mathematical Systems . . . . . . . . . . . . . . . . . . . 232
H.8.1 The physical system . . . . . . . . . . . . . . . . . . . . . . . . . . 233
H.8.2 The mathematical system . . . . . . . . . . . . . . . . . . . . . . . 233
H.8.3 Deletion of the time function . . . . . . . . . . . . . . . . . . . . . . 233
I MODULATION SYSTEMS 237
I.1 Acknowledgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
I.2 Introduction and Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
I.3 Modulation Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
I.3.1 Denition of modulation . . . . . . . . . . . . . . . . . . . . . . . . 237
I.3.2 Classes of modulation . . . . . . . . . . . . . . . . . . . . . . . . . 238
I.4 Bandpass Signals and Systems . . . . . . . . . . . . . . . . . . . . . . . . . 238
I.5 Linear modulation systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
I.5.1 Amplitude Modulation . . . . . . . . . . . . . . . . . . . . . . . . . 238
I.5.2 AM circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
I.5.3 Double sideband suppressed carrier modulation . . . . . . . . . . . 240
I.5.4 Single sideband modulation . . . . . . . . . . . . . . . . . . . . . . 241
I.5.5 Vestigial sideband modulation . . . . . . . . . . . . . . . . . . . . . 242
I.5.6 Detection systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
I.6 Receivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
I.6.1 AM superheterodyne receiver . . . . . . . . . . . . . . . . . . . . . 244
I.7 Angle modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
I.7.1 Phase modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
I.7.2 Frequency modulation . . . . . . . . . . . . . . . . . . . . . . . . . 246
I.7.3 Spectral analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
I.7.4 Bandwidth requirements . . . . . . . . . . . . . . . . . . . . . . . . 249
I.7.5 Detectors of angle modulation . . . . . . . . . . . . . . . . . . . . . 250
I.8 Performance of modulation systems . . . . . . . . . . . . . . . . . . . . . . 250
I.8.1 A general communication system . . . . . . . . . . . . . . . . . . . 251
I.8.2 Illustration of modulation eects . . . . . . . . . . . . . . . . . . . 252
I.8.3 PM post detection noise spectrum . . . . . . . . . . . . . . . . . . . 254
I.8.4 FM post detection noise spectrum . . . . . . . . . . . . . . . . . . . 255
I.8.5 Pre-emphasis and de-emphasis in frequency modulation . . . . . . . 255
I.8.6 Parameters for broadcast FM . . . . . . . . . . . . . . . . . . . . . 256
I.9 Destination signal to noise ratios . . . . . . . . . . . . . . . . . . . . . . . 256
I.9.1 FM threshold eect and mutilation . . . . . . . . . . . . . . . . . . 256
I.9.2 FM threshold extension (advanced topic to be omitted) . . . . . . . 256
I.10 Comparison of CW modulation systems . . . . . . . . . . . . . . . . . . . . 257
List of Figures
2.1 Charging a capacitor though a resistor. . . . . . . . . . . . . . . . . . . . . 4
2.2 Waveforms in charging a capacitor through a resistor. . . . . . . . . . . . . 4
2.3 Discharging a capacitor though a resistor. . . . . . . . . . . . . . . . . . . 5
2.4 Waveforms in discharging a capacitor through a resistor. . . . . . . . . . . 5
2.5 Energising an inductor though a resistor. . . . . . . . . . . . . . . . . . . . 5
2.6 Waveforms in energising an inductor through a resistor. . . . . . . . . . . . 6
2.7 De-energising an inductor though a resistor. . . . . . . . . . . . . . . . . . 6
2.8 Waveforms in de-energizing an inductor through a resistor. . . . . . . . . . 6
2.9 Polar representation of a complex number. . . . . . . . . . . . . . . . . . . 7
2.10 An N-port network. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.11 A nice linear one port network. . . . . . . . . . . . . . . . . . . . . . . . . 13
2.12 A linear terminal voltage current relation. . . . . . . . . . . . . . . . . . . 13
2.13 Basic forms of Thevenin and Norton circuits. . . . . . . . . . . . . . . . . . 14
2.14 A network with no Norton equivalent. . . . . . . . . . . . . . . . . . . . . . 15
2.15 A network with no Thevenin equivalent. . . . . . . . . . . . . . . . . . . . 15
2.16 Simplication of a Thevenin network. . . . . . . . . . . . . . . . . . . . . . 16
2.17 A network with a non-linear element. . . . . . . . . . . . . . . . . . . . . . 16
2.18 Another network with a non-linear element. . . . . . . . . . . . . . . . . . 17
2.19 A further network with a non-linear element. . . . . . . . . . . . . . . . . . 17
2.20 Thevenin representation of an automobile battery. . . . . . . . . . . . . . . 18
2.21 Norton representation of the same battery. . . . . . . . . . . . . . . . . . . 18
2.22 Ideal series resonant circuit. . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.23 Variation of series circuit current vs frequency. . . . . . . . . . . . . . . . . 20
2.24 Ideal parallel resonant circuit . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.25 Variation of parallel circuit voltage vs frequency . . . . . . . . . . . . . . . 21
2.26 Practical parallel resonant circuit. . . . . . . . . . . . . . . . . . . . . . . . 22
2.27 Practical parallel to ideal parallel circuit transformation . . . . . . . . . . . 24
2.28 Context for maximum power transfer theorem. . . . . . . . . . . . . . . . . 25
2.29 Context for denition of insertion gain and insertion loss. . . . . . . . . . . 26
3.1 Venn diagram representing A contained within B. . . . . . . . . . . . . . . 28
3.2 Venn diagram representing the union of A and B. . . . . . . . . . . . . . . 29
3.3 Venn diagram representing the intersection of A and B. . . . . . . . . . . . 30
3.4 Venn diagram representing the complement of A. . . . . . . . . . . . . . . 30
3.5 Venn diagram representing the dierence between A and B. . . . . . . . . 31
3.6 Venn diagram representing mutually exclusive sets. . . . . . . . . . . . . . 32
3.7 Venn diagram representing collectively exhaustive sets. . . . . . . . . . . . 32
xi
xii LIST OF FIGURES
3.8 Tree diagram for resistor production. . . . . . . . . . . . . . . . . . . . . . 42
4.1 Distribution functions of continuous and discrete variables. . . . . . . . . . 47
4.2 Probability density function of a continuous variable. . . . . . . . . . . . . 48
4.3 Probability density function of a discrete variable. . . . . . . . . . . . . . . 49
4.4 Gaussian probability density function. . . . . . . . . . . . . . . . . . . . . 51
4.5 The Q function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.6 Functional relation between random variables. . . . . . . . . . . . . . . . . 53
4.7 General functional relation. . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.8 Cartesian to polar conversion. . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.1 Conceptual representation of a random process. . . . . . . . . . . . . . . . 61
5.2 Probability density function of a random process. . . . . . . . . . . . . . . 62
5.3 Conditional probability density function. . . . . . . . . . . . . . . . . . . . 63
5.4 Poisson point process. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.5 Probabilities of number of points in an interval. . . . . . . . . . . . . . . . 67
5.6 Probability density function of time between successive pulses. . . . . . . . 68
5.7 X variable for Poisson point process. . . . . . . . . . . . . . . . . . . . . . 69
5.8 Y variable for Poisson point process. . . . . . . . . . . . . . . . . . . . . . 70
5.9 Overlapping and non-overlapping intervals. . . . . . . . . . . . . . . . . . . 70
5.10 Correlation function of a Poisson pulse process. . . . . . . . . . . . . . . . 73
5.11 Poisson impulse process. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.12 Random walk process. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.13 Probabilities in a random walk. . . . . . . . . . . . . . . . . . . . . . . . . 75
5.14 The reection principle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.15 Phasor diagram of sine wave and random noise. . . . . . . . . . . . . . . . 78
5.16 Probability density functions of r and . . . . . . . . . . . . . . . . . . . . 79
5.17 Illustration of a linear system. . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.18 Examples of random processes. . . . . . . . . . . . . . . . . . . . . . . . . 87
6.1 Correlation functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.2 Autocorrelation function of a sine wave. . . . . . . . . . . . . . . . . . . . 93
6.3 Rectangular pulse and its Fourier transform. . . . . . . . . . . . . . . . . . 94
6.4 The sinc function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.5 Fourier transform of exponential pulse. . . . . . . . . . . . . . . . . . . . . 95
6.6 Fourier transform of triangular pulse. . . . . . . . . . . . . . . . . . . . . . 96
6.7 Fourier transform of double exponential pulse. . . . . . . . . . . . . . . . . 98
6.8 Fourier transform of a constant by limiting process. . . . . . . . . . . . . . 99
6.9 Fourier transform of a periodic signal. . . . . . . . . . . . . . . . . . . . . . 100
6.10 Convolution of two time functions. . . . . . . . . . . . . . . . . . . . . . . 101
6.11 Autocorrelation and power spectrum of Poisson pulse process. . . . . . . . 102
6.12 Linear time invariant system. . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.13 Meaning of power spectral density. . . . . . . . . . . . . . . . . . . . . . . 105
6.14 Random binary waveform. . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
6.15 Autocorrelation function of random binary waveform. . . . . . . . . . . . . 107
6.16 Average autocorrelation and power spectrum. . . . . . . . . . . . . . . . . 108
7.1 Power spectrum and autocorrelation function of white noise. . . . . . . . . 111
LIST OF FIGURES xiii
7.2 Bandlimited white noise. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.3 White and narrowband noise. . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.4 Field containment structure with single port access. . . . . . . . . . . . . . 113
7.5 Thermal noise equivalent circuits. . . . . . . . . . . . . . . . . . . . . . . . 114
7.6 Power spectrum of shot noise. . . . . . . . . . . . . . . . . . . . . . . . . . 116
8.1 Context for denition of transfer function. . . . . . . . . . . . . . . . . . . 117
8.2 Poles and zeros of H(s)H(s) in the s plane. . . . . . . . . . . . . . . . . 119
8.3 Poles and zeros of example lter. . . . . . . . . . . . . . . . . . . . . . . . 120
8.4 Zero selection for minimum and non-minimum phase lters. . . . . . . . . 121
8.5 Denition of various ideal lter responses. . . . . . . . . . . . . . . . . . . 122
8.6 Denition of parameters of realistic low pass lter response. . . . . . . . . 123
8.7 Example circuit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
8.8 Normalised circuit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
8.9 Impedance scaled circuit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
8.10 Response of original circuit. . . . . . . . . . . . . . . . . . . . . . . . . . . 125
8.11 Response of frequency scaled circuit. . . . . . . . . . . . . . . . . . . . . . 126
8.12 Impedance and frequency scaled circuit. . . . . . . . . . . . . . . . . . . . 127
8.13 Ideal low pass lter response. . . . . . . . . . . . . . . . . . . . . . . . . . 127
8.14 Specication of required response. . . . . . . . . . . . . . . . . . . . . . . . 128
9.1 Amplitude responses of Butterworth lters. . . . . . . . . . . . . . . . . . . 132
9.2 Phase responses of Butterworth lters. . . . . . . . . . . . . . . . . . . . . 134
9.3 Pulse response of an n = 3 Butterworth lters. . . . . . . . . . . . . . . . . 135
9.4 Poles of Butterworth lter for n = 5. . . . . . . . . . . . . . . . . . . . . . 136
9.5 Magnitude responses of Chebychev I lters. . . . . . . . . . . . . . . . . . 138
9.6 Phase responses of Chebychev I lters. . . . . . . . . . . . . . . . . . . . . 139
9.7 Pulse response of an n = 3 Chebychev I lter. . . . . . . . . . . . . . . . . 140
9.8 Poles of an n = 5 Chebychev I lters. . . . . . . . . . . . . . . . . . . . . . 142
9.9 Magnitude responses of Chebychev II lters. . . . . . . . . . . . . . . . . . 143
9.10 Poles and Zeros of an all pass lter. . . . . . . . . . . . . . . . . . . . . . . 144
9.11 Frequency responses of Bessel-Thompson lters. . . . . . . . . . . . . . . . 145
9.12 Phase responses of Bessel-Thompson lters. . . . . . . . . . . . . . . . . . 146
9.13 Pulse response of n = 3 Bessel-Thompson lter. . . . . . . . . . . . . . . . 147
9.14 Magnitude responses of elliptic lters. . . . . . . . . . . . . . . . . . . . . . 149
9.15 Phase responses of elliptic lters. . . . . . . . . . . . . . . . . . . . . . . . 150
9.16 Pulse responses of an n = 3 elliptic lter. . . . . . . . . . . . . . . . . . . . 151
10.1 Specication of the required response. . . . . . . . . . . . . . . . . . . . . . 155
10.2 Filter design. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
10.3 Typical low pass lter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
10.4 Typical high pass lter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
10.5 High pass response specications. . . . . . . . . . . . . . . . . . . . . . . . 157
10.6 Specications of required response. . . . . . . . . . . . . . . . . . . . . . . 157
10.7 Filter design. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
10.8 Coaxial line lter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
10.9 Specication of required response. . . . . . . . . . . . . . . . . . . . . . . . 160
xiv LIST OF FIGURES
10.10Normalised low pass lter. . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
10.11The required lter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
10.12Typical band pass lter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
10.13Band pass transformation of elements. . . . . . . . . . . . . . . . . . . . . 163
10.14Band pass transformation. . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
10.15Band pass lter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
11.1 Thevenin equivalent circuit of a source. . . . . . . . . . . . . . . . . . . . . 168
11.2 Two-port network with interconnected transmission lines. . . . . . . . . . . 169
11.3 Two port circuit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
11.4 Realisation of third order Butterworth lter. . . . . . . . . . . . . . . . . . 172
11.5 Realisation of third order elliptic lter. . . . . . . . . . . . . . . . . . . . . 173
11.6 A two port circuit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
11.7 An example circuit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
11.8 Realisation of lter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
12.1 First order RC circuits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
12.2 First order RC circuit with buer amplier. . . . . . . . . . . . . . . . . . 178
12.3 Inverting operational amplier circuit. . . . . . . . . . . . . . . . . . . . . 179
12.4 Non-inverting operational amplier. . . . . . . . . . . . . . . . . . . . . . . 180
12.5 Low pass Sallen and Key circuit. . . . . . . . . . . . . . . . . . . . . . . . 182
12.6 Band pass Sallen and Key circuit. . . . . . . . . . . . . . . . . . . . . . . . 183
12.7 Inverting band pass circuit. . . . . . . . . . . . . . . . . . . . . . . . . . . 185
12.8 Bi-quadratic band pass circuit. . . . . . . . . . . . . . . . . . . . . . . . . . 186
12.9 Switched capacitor circuit. . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
12.10Switch waveforms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
12.11Switched capacitor low pass lter. . . . . . . . . . . . . . . . . . . . . . . . 189
H.1 A simple circuit for analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . 229
H.2 Sign conventions for circuit variables. . . . . . . . . . . . . . . . . . . . . . 230
H.3 Phasors, rotating arms and projections. . . . . . . . . . . . . . . . . . . . . 231
I.1 Fourier spectrum for a single tone message. . . . . . . . . . . . . . . . . . . 238
I.2 Illustration of a simple amplitude modulated signal. . . . . . . . . . . . . . 239
I.3 Fourier spectrum with amplitude modulation by a single tone. . . . . . . . 239
I.4 Amplitude modulation using a balanced mixer. . . . . . . . . . . . . . . . 240
I.5 Circuit for envelope detection. . . . . . . . . . . . . . . . . . . . . . . . . . 240
I.6 Illustration of a double sideband suppressed carrier modulated signal. . . . 241
I.7 Single sideband modulation using sideband lter. . . . . . . . . . . . . . . 241
I.8 Phase shift single sideband modulator. . . . . . . . . . . . . . . . . . . . . 242
I.9 Illustration of vestigial sideband ltering. . . . . . . . . . . . . . . . . . . . 243
I.10 Synchronous detection of a modulated signal. . . . . . . . . . . . . . . . . 243
I.11 Block diagram of a superheterodyne receiver. . . . . . . . . . . . . . . . . . 244
I.12 Passbands in a superheterodyne receiver. . . . . . . . . . . . . . . . . . . . 245
I.13 Narrow band phase modulator using balanced mixer. . . . . . . . . . . . . 246
I.14 Bessel functions of various orders. . . . . . . . . . . . . . . . . . . . . . . . 248
I.15 Amplitude spectrum of an angle modulated signal. . . . . . . . . . . . . . 248
I.16 Magnitude spectrum of an angle modulated signal. . . . . . . . . . . . . . 248
LIST OF FIGURES xv
I.17 Number of signicant sidebands in a frequency modulated signal. . . . . . 249
I.18 A balanced discriminator circuit. . . . . . . . . . . . . . . . . . . . . . . . 250
I.19 Derivation of balanced FM discriminator transfer function. . . . . . . . . . 251
I.20 A general CW communication system. . . . . . . . . . . . . . . . . . . . . 251
I.21 Signal and noise in various modulation systems. . . . . . . . . . . . . . . . 253
I.22 PM post detection noise spectrum. . . . . . . . . . . . . . . . . . . . . . . 254
I.23 FM post detection noise spectrum. . . . . . . . . . . . . . . . . . . . . . . 255
I.24 FM noise performance as a function of gamma. . . . . . . . . . . . . . . . 257
I.25 FMFB receiver block diagram. . . . . . . . . . . . . . . . . . . . . . . . . . 257
xvi LIST OF FIGURES
Chapter 1
INTRODUCTION
1.1 Course Denition
The course has the formal title: 3015 COMMUNICATIONS, SIGNALS AND SYS-
TEMS, and occupies approximately 22 lectures at Level 3. In addition, it is expected
that there will be, in times set aside in the timetable for lectures, two brief examinations,
which will supplement the main examination at the end of the semester.
1.2 Objectives
The course seeks to introduce
Basic probability theory and applications thereof.
The statistical properties of signals, the noise which can accompany those signals,
and what we can do about it.
The behaviour and design of circuits for ltering signals.
Elementary communication theory and simple communication ciruits.
1.3 Acknowledgments
Chapter 3 of these notes has drawn heavily on the book listed as Reference1 3 in Appen-
dix A. The notes also draw extensively on material originally written for other courses by
Dr. B. R. Davis. An emphasis on the concept of essential knowledge has been introduced
by the current lecturer.
1.4 Access to Notes
The current version of the lecture notes, homework and tutorial problems, and some
solutions, will be progressively available at the current Lecturers web page which is at
http://www.eleceng.adelaide.edu.au/Personal/peter/peter.
1
2 CHAPTER 1. INTRODUCTION
1.5 Bibliography
A list of useful reference books appears in Appendix A. Some of those books are concerned
with professional practice beyond the limits of this introductory course.
1.6 Essential Knowledge
An attempt will be made in this course to connect the material to that of other courses,
and in particular to courses on basic circuit theory, and to descriptions in engineering
terms of practical hardware.
The presentation will also be based on the belief that the understanding of advanced
work is facilitated by having a rm and correct grasp of elementary and fundamental
material.
These two factors will lead to the identication of a number of fundamental concepts,
recorded in Chapter 2, of which all students should be able to demonstrate clear under-
standing without recourse to notes.
A signicant part of the assessment in the subject will be devoted to testing whether
such skill has been acquired.
1.7 Making Use of the Lectures
Not all of the material to be presented in this course will be treated in detail in these notes.
Students are therefore advised to attend all lectures, and to come to lectures prepared to
take notes on occasion.
1.8 Other Goodies
Advice on how to study for the examination is contained in Appendix D.
Chapter 2
REVISION
2.1 Introduction
This chapter provides a listing of what is considered to be essential knowledge in this
course. The rst assessment exercise will be set on what may be found here.
2.2 Units
2.2.1 SI Units
In this course and presumably other courses, quantities will be expressed using terminol-
ogy and units dened in SI units and recommendations for the use of their multiples and
of certain other units, International Standard ISO 1000 (1992), International Organisa-
tion for Standardisation, Case postale 56, CH-1211, Genevre 20, Switzerland.
You are expected to know the names and SI units for the circuit theory quantities
commonly denoted by R, X, Z, G, B, Y, L, and C, for the eld quantities of voltage,
current, and power.
While we are discussing terminology and units, we take the opportunity to remind
ourselves that the names and units of the electromagnetic eld quantities denoted by E,
H, D, B and J are as shown in the Table below.
E E Electric eld intensity Vm
1
H H Magnetic eld intensity Am
1
D D Electric ux density Cm
2
B B Magnetic ux density Wbm
2
J J Volume current density Am
2
Table 2.1: Names and units for eld quantities.
2.2.2 Quantities derived by dierentiation
An important point to remember is that when one quantity is derived from another by
dierentiating with respect to a dimensioned variable, for example position or time, a
change of dimensions and thus of units results.
3
4 CHAPTER 2. REVISION
2.2.3 Frequency and angular frequency
It is also important to distinguish between the concepts of frequency and angular fre-
quency. Something called a frequency is measured in Hz. Normally such a quantity if
denoted by the symbol f. Something called an angular frequency is measured in rad/s.
Normally such a quantity if denoted by the symbol . The relation between the two is
= 2f (2.1)
It is very common that students lose marks by not observing the distinction.
The quantity known as bandwidth, and denoted by BW, in also measured in Hz. A
related quantity, known as angular bandwidth, and greater than BW by a factor 2, is
measured in rad/s. Not observing the distinction is a further common source of error.
2.3 RC and RL Circuits
2.3.1 Things not possible
It should be remembered that it is not possible to have a step change in capacitor voltage,
not is it possible to have a step change in inductor current. We will focus below on what
is possible, and review some standard results, which should be committed to memory.
2.3.2 Establishing a capacitor voltage
Figure 2.1: Charging a capacitor though a resistor.
Figure 2.2: Waveforms in charging a capacitor through a resistor.
2.3. RC AND RL CIRCUITS 5
2.3.3 Discharging a capacitor
Figure 2.3: Discharging a capacitor though a resistor.
Figure 2.4: Waveforms in discharging a capacitor through a resistor.
2.3.4 Establishing an inductor current
Figure 2.5: Energising an inductor though a resistor.
2.3.5 De-energising an inductor
Figure 2.6: Waveforms in energising an inductor through a resistor.
Figure 2.7: De-energising an inductor though a resistor.
Figure 2.8: Waveforms in de-energizing an inductor through a resistor.
2.4. PHASOR ANALYSIS FOR LINEAR A.C. CIRCUITS 7
2.4 Phasor Analysis for Linear a.c. Circuits
2.4.1 Notation
Much of the time in this course we will be dealing with variables which directly express
the values of the physical quantities, such as, for example, voltage or current. If those
physical quantities have a time variation, so do the variables of our equations.
In such a case, when the quantities represented are scalars, as in the example of
voltage or current just mentioned, we use lower case Roman or sometimes Greek letters
to represent them. Sometimes the time variation is shown, and sometimes it is not, as for
example in the equation for a real time-varying voltage
v = v(t). (2.2)
In many cases it will be convenient to restrict the time variation of all physical quan-
tities to be either constant or sinusoidal, or more explicitly to be of cosine form.
In such cases, the behaviour of each time-varying quantity is known for all time if we
know the frequency, the amplitude and the phase of the cosine function of time. In a single
context, all such quantities are assumed to have the same frequency, which is stated once
as a xed part of that context, but the dierent variables representing dierent quantities
can have various amplitudes and phases.
In this situation, it is convenient to introduce a variable called a complex phasor which
while not itself being a function of time, does represent in the manner described below
a time-varying quantity. What we do is to introduce a complex number, constructed so
that the magnitude of the complex number is the amplitude of the cosine waveform, and
the angle of the complex number in its polar representation, as shown for example in
Figure 2.9, is the phase angle of the cosine waveform.
Figure 2.9: Polar representation of a complex number.
Thus for the sinusoidally varying quantity
v = V
m
cos(t + ) (2.3)
which has an amplitude V
m
and a phase angle , we construct using V
m
and the
time invariant complex phasor V given by
V = V
m
e
j
(2.4)
It may be noted that the relation between the time invariant complex phasor V and
the time-varying variable v(t) which it represents is
v(t) =
_
Ve
jt
_
(2.5)
In words, this relation says that to recover the time function from the complex phasor,
we multiply the complex phasor by e
jt
and take the real part.
A graphical interpretation of the mathematical operation just dened is that to recover
the time function from the complex phasor, we can rst represent as shown in Figure 2.9
the phasor in the Argand diagram, and take its projection on the horizontal axis as the
expression of the value of the time function at the time t = 0. To visualise the behaviour
of the physical quantity as a function of time, we must construct a rotating arm whose
position at t = 0 is that of the complex phasor, and rotate it in a counter-clockwise
direction on the Argand diagram at an angular frequency , starting at time t = 0 at the
position illustrated in Figure 2.9, and watch the values of the projection on the horizontal
axis of the rotating arm which results.
Notice in the above exposition that we have not said that the phasor rotates. To do
so would contradict the denition of the phasor as a time invariant quantity. We have
given a dierent name, namely rotating arm, to the thing that rotates.
In establishing the notation described above, we have been able, because it is available,
to use dierent calligraphy, namely v (italic) and V (upright Roman), to distinguish the
real time-varying variables directly representing the physical quantities, and the time
invariant complex phasors indirectly representing them. The dierence in notation is
helpful in avoiding misunderstandings.
You are expected to have a facility for conversion of time domain variables to frequency
domain variables and vice versa.
2.4.2 Particular cautions
There can be no j in a time domain expression.
There can be no t in a frequency domain expression.
2.4.3 Peak value and r.m.s. value phasors
There are two common conventions in which sinusoidal quantities are expressed. In the
rst, known as the peak value phasor convention, the phasors have magnitudes that are
equal to the peak values of the sinusoidal quantities expressed. Thus the relation between
the phasors and the real time varying quantities represented is given by
v(t) =
_
Ve
jt
_
(2.6)
In the second convention, known as the r.m.s. value phasor convention, the phasors
have magnitudes that are equal to the r.m.s. values of the sinusoidal quantities expressed.
those r.m.s. values are less by a factor of

2 than the peak values. Thus the relation
between the phasors and the real time varying quantities represented is given by
2.5. CHARACTERISATION OF N-PORT NETWORKS 9
v(t) =
_
2Ve
jt
_
(2.7)
How can you tell which convention is in use? Well, in a perfect world, whenever
an r.m.s. phasor is used, it should have r.m.s. after it. Another thing which helps
is knowing that power engineers generally employ r.m.s. phasors, and communication
engineers generally employ peak value phasors. Do you seem to live in a perfect world?
2.4.4 Properties of a sinusoid
For the sinusoidal voltage given in peak value phasor terms by
V = 10j (2.8)
you should be able to persuade yourself that the maximum value is 10 V, the minimum
value is -10 V, the average value is 0 V, the value at time t = 0 is 0 V, and the r.m.s.
value is approximately 7.07 V.
When a sinusoidal current ows in a resistor, the power dissipated is time-varying,
but is at all points of time either positive or zero. It oscillates, at a rate of twice the
frequency of the sinusoidal current, between a value of zero and twice the average value.
The average value of the power dissipated is the same as would be dissipated by d.c.
current equal to the r.m.s. value of the sinusoid.
2.4.5 Unusual waveforms
In calculating r.m.s. values for unusual waveforms, please remember that r.m.s means the
(root (of the mean (of the square))).
2.4.6 Resistance and resistivity
You should be familiar with the formula
R =
l
A
(2.9)
for the resistance R of a rectangular bar of material of length l, resistivity and cross
sectional area A, and be able to work practical exercises thereon.
2.5 Characterisation of N-Port Networks
2.5.1 Denitions of port variables
For characterisation of an N-port network such as is illustrated in Figure 2.10, we must
rst dene senses for the voltage and currents at each of the ports, a port being a pair of
terminals.
The convention universally chosen is that the current I is dened as that which enters
a selected terminal of the pair, while the voltage V is simultaneously dened as the voltage
of the selected terminal with respect to the other. It is also assumed for the purpose of
Figure 2.10: An N-port network.
characterisation that if current I enters the selected terminal, a current I emerges from
the other terminal.
This condition might or might not be a natural property of the network, but it can
be certainly be imposed from outside by ensuring that each of the signal generators or
passive impedances which are connected externally to the ports are oating i.e. have
no interconnections with one another. In line with this picture of oating external port
connections, we make no inquiry as to what voltages may exist between one pair of
terminals considered as a port and another pair of terminals considered as another port.
The lack of interest in this matter can be excused in some cases because the external
connections are truly oating, in which case the number of independent currents which
can ow between the 2N terminals is limited to just N, and can be excused in some
other cases because there is at each port a common ground connection, in which case
the number of independent voltages which can exist between the 2N terminals is limited
to just N. In some other cases in which neither of these situations applies, we must be
aware that the viewpoint we are adopting is not capable of representing the full range of
behaviour of a network with 2N terminals. Whatever be the case, we are going to persist
with our notion that the variables shown in Figure 2.10 are sucient for our purpose.
2.5.2 Denition of impedance matrix
With the variables so dened, we can take the view that we could in an empirical sense
impose, by means of ideal current generators at each port, whatever currents we please,
2.5. CHARACTERISATION OF N-PORT NETWORKS 11
and leave it to the network to determine the voltages at the ports, which voltages we
could then measure. If we assume that the network contains only linear passive circuit
elements, we can conclude from this view that the network is describable by the set of
equations below, in which the N N array of parameters is called the impedance matrix
for the network.
V
1
V
2
.
.
.
V
N
Z
11
Z
12
... Z
1N
Z
21
Z
22
... Z
2N
.
.
.
.
.
.
.
.
.
.
.
.
Z
N1
Z
N2
... Z
NN
I
1
I
2
.
.
.
I
N
(2.10)
It is clear from the above equation that the element Z
ij
may be determined by injecting
a current into port j, ensuring that no currents are injected into the other ports by leaving
them open circuit, and studying the voltage developed at port i, taking care that the
process of observing that voltage is conducted via an ideal voltmeter, i.e. one which
draws no current from the port.
In the light of this observation, the set of impedance parameters of a network are
sometimes given the alternative name of the open circuit parameters of that network.
Of course the ideal voltmeter contemplated in the paragraph above does not exist,
and some care must be taken to correct for the nite impedance of real voltmeters used
in the observation.
2.5.3 Denition of admittance matrix
An alternative view which we could in an empirical sense take is that we could impose,
by means of ideal voltage generators at each port, whatever port voltages we please, and
leave it to the network to determine the currents which then ow at the ports, which
currents we could then measure. If we continue to assume that the network contains
only linear passive circuit elements, we can conclude from this view that the network is
describable by the set of equations below, in which the N N array of parameters is
called the admittance matrix for the network.
I
1
I
2
.
.
.
I
N
Y
11
Y
12
... Y
1N
Y
21
Y
22
... Y
2N
.
.
.
.
.
.
.
.
.
.
.
.
Y
N1
Y
N2
... Y
NN
V
1
V
2
.
.
.
V
N
(2.11)
It is clear from the above equation that the element Y
ij
may be determined by placing
a known voltage across port j, ensuring that no voltages appear across the other ports
by placing short circuits across them, and studying the current owing in the reference
direction at port i when a short circuit is placed across that port, taking care that the
process of observing that current is conducted via an ideal ammeter, i.e. one which places
no burden on the port, i.e. one which has no series resistance.
In the light of this observation, the set of admittance parameters of a network are
sometimes given the alternative name of the short circuit parameters of that network.
Of course the ideal ammeter contemplated in the paragraph above does not exist, and
some care must be taken to correct for the non-zero burden of real ammeters used in the
observation.
2.5.4 Relations between impedance and admittance parameters
It is easy to show that when both sets of parameters exist, the impedance matrix and the
admittance matrix are inverses of one another.
2.5.5 Pathological cases
While it is true that most networks have both impedance and admittance matrices, there
are special cases in which one or other of the matrices (or even both) may not exist. For
example a network with an internal short circuit across one port may have an impedance
matrix, but cannot have an admittance matrix, and a network with one port internally
open circuited and not otherwise internally connected may have an admittance matrix,
but cannot have an impedance matrix.
2.5.6 The reciprocity theorem
A fundamental theorem which may be applied in the network theory context to linear pas-
sive networks and in the electromagnetic theory context to screened enclosures containing
media which exhibit appropriate behaviour, is the reciprocity theorem, which states, in the
network theory context, that under appropriate conditions the impedance and admittance
matrices are symmetric.
A set of sucient conditions for the reciprocity theorem to apply is that the net-
work may contain resistors, inductors and capacitors, but contains no voltage or current
generators.
The most enlightening proof of the reciprocity theorem is that provided for the fully
general version of the theorem which is established in the electromagnetic theory context
at Level 4 of the course. That fully general treatment illuminates both the assumptions
underlying lumped element network theory and the restrictions on material constitutive
parameters which are required for the theorem to be valid.
2.6 Thevenins and Nortons Theorems
2.6.1 Thevenin and Norton networks
An illustration of a fairly general linear network is provided in Figure 2.11 below. The
network has many internal connections, but only a single pair of external terminals on
which we will shortly focus attention. We expect that a set of liner equations, deriving
from the equations describing each element, will determine the behaviour of the circuit,
and in particular the relation between the current and voltage at the external terminals.
It is important to note that this expectation of a linear relation between the current
and voltage at the external terminals derives from the linearity of the circuit elements em
internal to the network, and does not depend upon whether we plan to connect either
linear elements or non-linear elements externally.
The linear relation between external terminal current and voltage can be expressed
graphically as in Figure 2.12 below. The gure reminds us that the most general linear
relation can be described in terms of two parameters, which might be chosen as the slope
2.6. THEVENINS AND NORTONS THEOREMS 13
Figure 2.11: A nice linear one port network.
and the intercept on the vertical axis, or might alternatively be chosen as the intercepts
on the two axes.
Figure 2.12: A linear terminal voltage current relation.
We now ask the question of whether the circuit, viewed solely in terms of the relation
between its external terminal voltage and current, can be represented by some simpler
network in which there are fewer elements. The obvious answer is that it can, and in more
that one way. The two networks shown in Figure 2.13 are each capable of representing
a relation such as shown in Figure 2.12. The resistance appearing in both circuits is the
negative of the slope in Figure 2.12, while the voltage and current generators have values
equal to the intercepts on the axes in Figure 2.12.
These circuits are called respectively the Thevenin and Norton equivalent circuits of
Figure 2.13: Basic forms of Thevenin and Norton circuits.
the network.
We must emphasise, and will do so repeatedly, that for the Thevenin and Norton
circuits to exist, there is a need for linearity of the internal circuit elements, at least those
which have the possibility of inuencing the external terminal voltage and current, but
no constraint is passed on what we may connect externally.
2.6.2 Determination of Thevenin and Norton parameters
The rules for determining for a particular network the parameters of the Thevenin and
Norton circuits are quite straightforward. They are as follows.
The voltage generator of the Thevenin equivalent circuit is the voltage which appears
across the terminals of the given network when its external terminals are open
circuited.
The current generator of the Norton equivalent circuit is the current which ows in
a short circuit placed across the terminals of the given network.
The internal impedance of both the Thevenin and Norton circuits is the impedance
seen looking back into the external terminals of the given network when all sources
inside that network have been reduced to zero.
In interpreting the last rule it is important to note that reducing a voltage source to
zero is equivalent to replacing it by a short circuit, while reducing a current source to
zero is equivalent to replacing it by an open circuit.
A simple consequence of the above rules and the equivalence of the Thevenin and
Norton circuits to one another as well as to the given network is that the Thevenin and
Norton parameters are related by the equation
I
sc
=
V
oc
R
(2.12)
2.6.3 Pathological cases
Not all networks have both Thevenin and Norton circuits. In fact an attempt to construct
a Norton equivalent for the ideal voltage source shown in Figure 2.14 leads to an innite
current in parallel with a zero resistance, and no sensible calculations can be performed
with that result. We say the Norton circuit does not exist.
Figure 2.14: A network with no Norton equivalent.
Similarly an attempt to construct a Thevenin equivalent for the ideal current source
shown in Figure 2.15 leads to an innite voltage in series with an innite resistance, and
no sensible calculations can be performed with that result. We say the Thevenin circuit
does not exist.
Figure 2.15: A network with no Thevenin equivalent.
Some networks lend themselves to immediate simplication in the construction of a
Thevenin or Norton circuit. For example an ideal voltage source in parallel with some
other component, as shown in Figure 2.16, produces exactly the same eect as that ideal
voltage source alone, so the other component may be immediately omitted, ie replaced
by an open circuit.
Figure 2.16: Simplication of a Thevenin network.
A similar simplication occurs when an ideal current source is in series with some
other component. A little thought reveals that the other component may be omitted
from the circuit, ie replaced by a short circuit.
2.6.4 Networks with non-linear elements
Figure 2.17: A network with a non-linear element.
Although the presence of a nonlinear element such as is shown in Figure 2.17 below
will produce a nonlinear relation between terminal current and voltage, and thus preclude
the existence of a Thevenin or Norton circuit, there are pathological cases, such as are
illustrated in Figure g:netw:anne or Figure g:netw:fnne, in which a non-linear element
is present in such a way as to produce no non-linearity of the external terminal voltage
current relation, and a Thevenin or Norton circuit may still exist.
2.6.5 Frequency domain equivalent circuits
Although we have conducted the above discussion in the implied context of d.c. networks,
the results are also valid in the context of sinusoidally varying voltages, represented by
complex phasors, in networks consisting of complex impedances. That this is true follows
from the homomorphormism of the equations describing the above two contexts.
Figure 2.18: Another network with a non-linear element.
Figure 2.19: A further network with a non-linear element.
2.6.6 Warning against undue generalisation
It is most important to note that the Thevenin and Norton equivalent circuits do not
represent the details of the internal behaviour of the network. They do not indicate what
components are actually inside the network, nor do they give details of properties other
than the relation between external terminal voltage and current, for example the internal
power dissipation.
A good example of the above statement is provided by the representation of an au-
tomobile battery, which has an open circuit voltage of about 12 V d.c. and which drops
only slightly when quite large currents are drawn from it. A Thevenin equivalent circuit
for a typical battery is shown in Figure 2.20.
12 V
0.01 W
A
B
Figure 2.20: Thevenin representation of an automobile battery.
From the terminal voltage and current point of view the battery may equally well be
represented by the Norton equivalent circuit of Figure 2.21. Either circuit will give correct
results for the terminal voltage and current when any particular load, linear or non-linear,
is connected to the terminals.
A
B
0.01 W
1200 A
Figure 2.21: Norton representation of the same battery.
2.7. TUNED CIRCUITS 19
If we make the mistake of imagining that the equivalent circuit goes beyond the repre-
sentation of the terminal voltage and current to the representation of the internal opera-
tions of the circuit, for example to the calculation of the internal power dissipation, we are
immediately advised of our error by an obvious and large contradiction. The Thevenin
circuit appears to imply that with no external load, the battery has no internal power
dissipation, while the Norton equivalent circuit appears to imply that under the same
conditions the internal power dissipation is 14.4 kW! Of course as neither the Thevenin
nor Norton circuits represent other than terminal voltage and current relations, neither
conclusion can properly be drawn.
2.7 Tuned Circuits
The impedance vs frequency behaviour of simple tuned circuits is summarised in the
sections below. All of these details should be committed to memory.
2.7.1 Ideal series resonant circuit
A simple series resonant circuit excited by an ideal voltage generator is shown in Fig-
ure 2.22. In that circuit the resonant frequency is given by
Figure 2.22: Ideal series resonant circuit.
f
0
=
1
2
LC
(2.13)
The impedance of the circuit is a minimum at resonance and is just equal to the
resistance r. This minimum impedance at resonance is known as the dynamic impedance.
For excitation with a constant voltage over a range of frequencies the variation of the
current owing versus frequency is as shown in Figure 2.23.
The quality factor Q of all resonant circuits is dened as
Q = 2
(energy stored at resonance)
(energy dissipated per cycle)
(2.14)
Figure 2.23: Variation of series circuit current vs frequency.
It may be shown for a simple series resonant circuit that
Q =
2f
0
L
r
(2.15)
At resonance the voltage appearing across the inductor (or, alternatively, the capac-
itor) is just Q times the source voltage. For that reason, Q is sometimes known as the
voltage magnication factor.
The bandwidth BW is dened as the dierence (in Hz) between the frequencies at
which the series circuit current is
1
2
of the value at resonance. At those so-called half-
power points the power dissipated in the circuit is just half the value at resonance. It may
be shown that the bandwidth and Q are related by
BW =
f
0
Q
(2.16)
2.7.2 Ideal parallel resonant circuit.
A simple parallel resonant circuit excited by an ideal current generator is shown in Fig-
ure 2.24 below. In that circuit the resonant frequency is again given by
f
0
=
1
2
LC
(2.17)
The impedance of the circuit is a maximum at resonance and is just equal to the
resistance R. This maximum impedance at resonance is known as the dynamic impedance.
For excitation with a constant current over a range of frequencies the variation of the
voltage appearing across the circuit versus frequency is as shown in Figure 2.25 below.
The quality factor Q is again dened as
Figure 2.24: Ideal parallel resonant circuit
Figure 2.25: Variation of parallel circuit voltage vs frequency
Q = 2
(2.18)
It may be shown for an ideal parallel resonant circuit that
Q =
R
2f
0
L
(2.19)
The bandwidth BW is dened as the dierence (in Hz) between frequencies at which
the parallel circuit voltage is
1
2
of the value at resonance. At those so-called half-power
points the power dissipated in the circuit is just half the value at resonance. It may be
shown that the bandwidth and Q are related by
BW =
f
0
Q
(2.20)
2.7.3 Practical parallel resonant circuit
A alternative parallel resonant circuit excited by an ideal current generator is shown in
Figure 2.26 below. This circuit is known is the practical parallel resonant circuit because
in a parallel resonant circuit, the signicant losses generally occur within the inductor in
the manner indicated by the circuit.
Analysis of that circuit is algebraically more complex than for each of the two circuits
considered previously, and we in the below statements provide the results of an approx-
imate analysis which is valid when the quality factor is signicantly greater than unity
(about ve is good enough for most purposes).
Figure 2.26: Practical parallel resonant circuit.
In the practical parallel resonant circuit the resonant frequency is again given by
f
0
=
1
2
LC
(2.21)
The impedance of the circuit is again a maximum at resonance and may be shown to
be equal to he expression (
0
L)
2
/r. This maximum impedance at resonance is known as
the dynamic impedance. We should note carefully that this impedance is denitely not a
single one of the impedances appearing in the circuit, but is a function of all of them.
For excitation with a constant voltage over a range of frequencies the variation of the
voltage appearing across the circuit versus frequency resembles that of the ideal parallel
resonant circuit, which was shown in Figure 2.25 above, and will match it when we set
the dynamic impedance R of the ideal parallel resonant circuit equal to the dynamic
impedance given in the paragraph above for the practical parallel resonant circuit, that
is
R =
(
0
L)
2
r
(2.22)
The quality factor Q is, as always, dened as
Q = 2
(2.23)
It may be shown for the practical parallel resonant circuit that
Q =
2f
0
L
r
(2.24)
Using this result we may derive the alternative expression for the dynamic impedance
of a practical parallel resonant circuit
R = Q
2
r (2.25)
The bandwidth BW is dened as the dierence (in Hz) between frequencies at which
the parallel circuit voltage is
1
2
of the value at resonance. At those so-called half-power
points the power dissipated in the circuit is just half the value at resonance. It may be
again be shown that the bandwidth and Q are related by
BW =
f
0
Q
(2.26)
2.7.4 Common features
In the above sections we can recognise the common features that resonant frequency is
always given by
f
0
=
1
2
LC
(2.27)
and quality factor is always given by
Q = 2
(2.28)
Bandwidth is always related to quality factor by
BW =
f
0
Q
(2.29)
Quality factor in terms of circuit elements is either
Q =
2f
0
L
r
(2.30)
or
Q =
R
2f
0
L
(2.31)
the choice being easily made on the basis of whether to reduce the losses, ie increase
the Q, we should seek either a lower or higher resistance.
2.7.5 Practical parallel to ideal parallel circuit transformation
Because of the similarities between the ideal and practical parallel circuits, we often when
faced with a practical parallel circuit, simplify our thinking by replacing it with an ideal
parallel circuit, with the same inductor and capacitor values, and a parallel resistor R
given by
R =
(
0
L)
2
r
= Q
2
r (2.32)
This transformation is illustrated in Figure 2.27 below.
R C L
L
C
r
R = Q r where Q =
( L) w
r
2
o
Figure 2.27: Practical parallel to ideal parallel circuit transformation
2.8 Maximum Power Transfer Theorem
The maximum power transfer theorem of linear a.c. networks is concerned with the situ-
ation depicted in Figure 2.28, in which a sinusoidal source of internal voltage represented
by the peak value phasor V
S
and internal impedance Z
S
feeds an external load Z
L
.
The theorem states that the load impedance which causes the maximum power to be
extracted from the source is the complex conjugate of the source impedance. You should
prove the theorem as an exercise.
2.8. MAXIMUM POWER TRANSFER THEOREM 25
Figure 2.28: Context for maximum power transfer theorem.
2.8.1 Available source power
The concept of the available source power is closely related to the above theorem. It is, for
a particular source, simply the greatest power which can be extracted from that source.
It is easy to show from the above result that the available source power for the source
of which the Thevenin equivalent circuit is shown in Figure 2.28 is given by
Available source power =
|V
S
|
2
8R
S
(2.33)
where R
S
is the real part of the source impedance Z
S
. You should also derive this
result as an exercise, as experience shows that students frequently misunderstand and
misremember this result.
It is important to realise that any power thought to be dissipated in the source re-
sistance R
S
is not included in the denition of available source power, nor is such power
physically meaningful.
2.8.2 Logarithmic expression of power ratios
In communication systems it is frequently useful to express power ratios logarithmically.
This is because, as a signal progresses through a communication system, it is multiplied
or divided by various factors expressing gain or loss. Multiplying such a string of factors
mentally is burdensome, but if the gain or loss is expressed logarithmically, the expressions
of gain or loss need merely to be added, and mental calculations become simpler.
The basic logarithmic expression of a power ratio is the Bel, dened as
Power ratio in Bels = log
10
_
One power
Another power
_
(2.34)
The Bel is too coarse a unit for common use, so most power ratios are expressed in
decibels, (dB) which are one tenth of a Bel. Thus
Power ratio in dB = 10 log
10
_
One power
Another power
_
(2.35)
It is important to note the factor of 10 in the above result, and also that dB are
here dened in terms of power ratios, not voltage ratios. Using the denitions given here
will always result in correct calculations. Sadly, experience shows that use of alternative
formulae generally results in incorrect calculations.
2.8.3 Exercise
Two resistors of resistance R and 2R are placed in parallel and are connected to a sin-
gle voltage source. Calculate in dB the ratio of the power dissipated in the resistor of
resistance 2R to the power dissipated in the resistor of resistance R.
2.8.4 Insertion gain and insertion loss
The related quantities Insertion gain and Insertion loss tell us what eect a network has
on the transmission of power between a source and a load. They are dened with the aid
of Figure 2.29 below, in which an a.c. source of internal voltage V
S
and source impedance
R
S
is connected to a resistive load R
L
via a two port network for which we are going to
dene the insertion gain or insertion loss.
Figure 2.29: Context for denition of insertion gain and insertion loss.
We are concerned with two values of the load voltage phasor V
L
. The rst, denoted
by V
L1
is the load voltage when the network is not present and the source is connected
directly to the load. The second, denoted by V
L2
, is the load voltage when the network
is inserted between the source and the load, as shown in Figure 2.29.
The insertion gain is a measure, in dB terms, of how much the load power is increased
by the insertion of the network. Thus
Insertion gain = 10 log
10
V
L2
V
L1
2
(2.36)
.
We should note the taking of magnitudes and the squaring.
When the network increases the power to the load, the insertion gain is positive, and
when it decreases the power to the load, the insertion gain is negative. It is common in
the latter case to speak of the Insertion loss, dened as
Insertion loss = 10 log
10
V
L1
V
L2
2
(2.37)
.
Clearly for a given network, the insertion loss = - the insertion gain.
Chapter 3
INTRODUCTION TO
PROBABILITY
3.1 Aims
The aims of this Chapter are
To introduce students to probability theory.
To develop in students an intuition of how the theory applies to practical situations.
To teach students how to apply probability theory to solving engineering problems.
3.2 Set Theory
3.2.1 Concept
We consider sets, generally denoted by capital letters, of elements, frequently denoted by
lower case letters, unless the elements are themselves sets, in which case upper case letters
are generally used.
Sets can be dened in various ways. One way is just to list the elements, e.g.
A = {my house, my car, my computer} (3.1)
Another is to give a rule for generating the elements of the set, e.g.
B = {x
2
|x = 1, 2, 3, } (3.2)
More generally we provide a test for determining whether something is or is not an
element of the set, e.g.
A = {everyone who was alive at some time in 2000} (3.3)
An important point is that for each set there must be a test as to whether anything
conceivable is or is not a member of the set.
27
28 CHAPTER 3. INTRODUCTION TO PROBABILITY
3.2.2 Belonging
x A means x is an element of the set A. x / A means x is not a member of the set A.
3.2.3 Subset
A B means the set A is a subset of the set B, i.e. every element of A is an element of
B.
3.2.4 Set equality
A = B means A B and B A.
3.2.5 Universal set
The universal set, by convention denoted by S, is the set of all things we can possibly
consider at a given moment.
3.2.6 Null set
The null set, for which the notation is , has by denition no elements. Also by denition,
it is a subset of every set, that is, for every set A, A.
3.2.7 Venn diagrams
Figure 3.1: Venn diagram representing A contained within B.
3.2. SET THEORY 29
3.2.8 Set operations and concepts
Union
The union of two sets A and B is denoted by A B and is the set which contains all
of the elements which are in either the set A or the set B. An illustration of the union
relation is provided in Figure 3.2.
Figure 3.2: Venn diagram representing the union of A and B.
Intersection
The intersection of two sets A and B is denoted by A B and is the set which contains
those elements which are in both the set A and the set B. An illustration of the intersection
relation is provided in Figure 3.3.
Complement
The complement of a set A, denoted by A
c
, is the set of all elements in S that are not in
A. The complement relation is illustrated in Figure 3.4.
Dierence
The dierence between a set A and a set B is denoted by A B, and is the set which
contains all the elements of A that are not elements of B.
The dierence relation is illustrated in Figure 3.5.
3.2.9 Mutual exclusivity
A collection of sets is mutually exclusive if and only if none of the elements of A are
elements of B, and vice versa. Symbolically,
Figure 3.3: Venn diagram representing the intersection of A and B.
Figure 3.4: Venn diagram representing the complement of A.
3.2. SET THEORY 31
Figure 3.5: Venn diagram representing the dierence between A and B.
A
i
A
j
= for all i = j (3.4)
The mutually exclusive relation is illustrated for two sets in Figure 3.6. When two
sets A and B are mutually exclusive, we say they are disjoint.
3.2.10 Collective exhaustion
A collection of sets is collectively exhaustive if and only if all of the elements of S are
elements of at least one of them. Symbolically, the sets A
1
to A
n
are collectively exhaustive
if and only if
A
1
A
2
A
2
A
3
A
n
= S (3.5)
The condition of collective exhaustion is illustrated in Figure 3.7.
3.2.11 Theorems on sets
De Morgans theorem
De Morgans theorem states that
(A B)
c
= A
c
B
c
(3.6)
In the proof, which we will not reproduce in detail here, we show that (A B)
c
A
c
B
c
,
and also that A
c
B
c
(A B)
c
. Hence the equality. We may recognise here the parallel
to some theorems of digital logic.
Figure 3.6: Venn diagram representing mutually exclusive sets.
Figure 3.7: Venn diagram representing collectively exhaustive sets.
3.3. APPLYING SET THEORY TO PROBABILITY 33
3.3 Applying set theory to probability
We will require the basic concepts of an experiment, which contains a procedure and
observations. We will also have models of actual physical experiments.
3.3.1 Example
As an illustration, we consider the tossing of a coin. We might have in this context
The procedure. Toss the coin and let it land upon a table.
The observation. Observe which side faces up after it lands.
The model. Head or tails are equally likely. The result of each toss is unrelated
to the result of previous tosses.
3.3.2 Some denitions
An outcome of an experiment is any possible observation of that experiment.
An event is a set of outcomes of an experiment.
An event space is a collectively exhaustive, mutually exclusive, set of events.
The sample space of an experiment is the nest-grain, mutually exclusive, collec-
tively exhaustive set of all possible outcomes. Finest grain means that all possible
distinguishable outcomes are identied separately.
3.3.3 Comparisons
An event space and a sample space have a lot in common. The members of both are
mutually exclusive and collectively exhaustive. They dier in the nest-grain property
that applies to a sample space but not to an event space. Because it possesses the nest-
grain property, a sample space contains all the details of an experiment. The members
of a sample space are outcomes. By contrast, the members of an event space are events.
The event space is a set of events (sets), while the sample space is a set of outcomes
(elements). Usually, a member of an event space contains many outcomes.
3.3.4 Further comparison
The following table relates the terminology of probability to set theory.
SET ALGEBRA PROBABILITY
element outcome
set event
universal set sample space
Dening the sample space and its outcomes are key elements of the solution of any
probability problem. A probability problem arises from some practical situation that can
be modelled as an experiment. To work on the problem, it is necessary to dene the
experiment carefully, and then derive the sample space. Getting this right is a big step
toward solving the problem.
3.3.5 Examples
Example 1
Suppose we roll a six sided die and observe the number of dots on the side facing upwards.
We can label these outcomes i = 1, , 6 where i denotes the outcome that i dots appear
on the up face. The sample space is S = {1, 2, , 6}. Each subset of S is an event.
Examples of events are
The event E
1
= {Roll 4 or higher} = {4, 5, 6}.
The event E
2
= {The roll is even} = {2, 4, 6}.
The event E
3
= {The roll is the square of an integer} = {1, 4}.
Example 2
We wait for someone to make a phone call, and observe the duration of the call in minutes.
The outcome x is a non-negative real number. The sample space is S = {x|x > 0}. The
event the phone call lasts longer than ve minutes is {x|x > 5}.
3.3.6 Fundamental remark
We are used in other branches of study to resolving a complex construct into a number of
simpler constructs which have some sort of independence or orthogonality between them.
As one example, for vectors in a three dimensional space, we may resolve an arbitrary
vector into the sum of three vectors which are mutually orthogonal. As another example,
in the study of periodic waveforms, we may resolve a periodic waveform of arbitrary
shape into a number of Fourier components which have an orthogonality relation between
them. In a similar way, it is possible to resolve a complex event into the union of a set
of mutually exclusive events. In fact, the entire theory of probability can be said to be
based on unions of mutually exclusive events.
3.3.7 Theorem on event spaces
For an event space B = {B
1
, B
2
, } (which will be both mutually exclusive and collec-
tively exhaustive), and any set A of events in the sample space, let C
i
= A B
i
. For
i = j, the events C
i
and C
j
are mutually exclusive and
A = C
1
C
2
(3.7)
3.3. APPLYING SET THEORY TO PROBABILITY 35
3.3.8 Example
As an example, we consider the experiment of tossing the four coins: ve cents, ten cents,
twenty cents and fty cents (we are lucky to have so many coins), and of observing the
results of the tossing, in the order of the coins just listed. The sample space consists of
sixteen four-letter words, such as hhhh, hhht, down to tttt, if we list them in an orderly
way.
In this example, we choose A to be the set of outcomes with less than three heads.
There are eleven such outcomes shown below.
A = {tttt, httt, thtt, ttht, ttth, hhtt, htht, htth, tthh, thth, thht} (3.8)
We also let the set {B
0
, B
1
, B
2
, B
3
, B
4
, } denote the event space in which B
i
= {outcomes with i heads
The above theorem then states that
A = (A B
0
) (A B
1
) (A B
2
) (3.9)
In this example, B
0
, B
1
B
2
are all subsets of A. Therefore A B
i
= B
i
for i = 0,1,2.
Also, for i = 3 and i = 4, A B
i
= , so that A = B
0
B
1
B
2
, which is a union of
disjoint sets. In words, this example states that the event less than three heads is the
union of zero heads, one head, and two heads.
The experiment in this example refers to what could be called a toy problem, one
that is easily visualised but is not something we would do in the course of our work.
Mathematically, however, it is equivalent to many real engineering problems. As a rst
example, we might observe a pair of modems which transmit four bits via a telephone
line from one computer to another. For each bit, we might observe whether the receiving
modem detects the bit correctly (c), or makes an error (e). As a second example, we
might consider the testing of four integrated circuits. For each one, we might observe
whether the circuit is acceptable (a), or a reject (r).
In all of these examples, the sample space contains 16 four-letter words formed with
an alphabet containing two letters. If we are only interested in the number of times one of
the letters occurs, it is sucient to refer only to the event space B, which does not contain
all of the information about the experiment, but does contain all of the information we
need.
The event space is simpler to deal with than the sample space because it has fewer
members. There are ve events in the event space, and 16 outcomes in the sample space.
The simplication is much more signicant when the complexity of the experiment
is higher; for example, testing ten circuits. The sample space has 2
10
= 1024 members,
while the corresponding event space has only 11 members.
The concept of an event space is useful because it allows us to express any event as a
union of mutually exclusive events. We have already remarked that the entire theory of
probability is based on unions of mutually exclusive events.
3.3.9 A modest proposal
Be sure to understand the theorem in equation 3.7 and its illustration in the above example
thoroughly, as it forms the basis of the solution to many practical problems.
3.3.10 Exercise
Monitor three consecutive phone calls going through a telephone switching oce. Classify
each one as a voice call (v) if someone is speaking, or a data call (d) if the call is carrying a
modem or fax signal. Your observation is a sequence of three letters (each letter is either
v or d). For example, two voice calls followed by one data call corresponds to vvd.
Write the elements of the following sets
(a) A
1
= {rst call is a voice call} (b) B
1
{rst call is a data call}
(c) A
2
= {second call is a voice call} (d) B
2
{second call is a data call}
(e) A
3
= {all calls are the same} (f) B
3
{voice and data alternate}
(g) A
4
= {one or more voice calls} (h) B
4
{two or more data calls}
For each pair of events A
1
and B
1
, A
2
and B
2
and so on, identify whether the pair of
events is either mutually exclusive or collectively exhaustive.
3.4 Probability Axioms
A probability measure P[] is a function that maps events in the sample space to real
numbers such that
Axiom 1
For any event A, P[A] 0. (3.10)
Axiom 2
P[S] = 1. (3.11)
Axiom 3
For any countable collection A
1
, A
2
, of mutually exclusive events,
P[A
1
A
2
] = P[A
1
] +P[A
2
] + (3.12)
The entire theory of probability is built upon the three axioms above.
3.5 Probability Theorems
3.5.1 Probabilities with the union of disjoint sets
Given events A and B such that A B = then
P[A B] = P[A] +P[B]. (3.13)
3.5. PROBABILITY THEOREMS 37
3.5.2 Probabilities with the union of mutually exclusive sets
If B = B
1
B
2
B
m
and B
i
B
j
= for i = j then
P[B] =
m
i=1
P[B
i
]. (3.14)
.
3.5.3 Probabilities of outcomes combined to form an event
For a set B = {s
1
, s
2
, , s
m
} of outcomes, not necessarily complete, of an experiment,
the probability of the event B = {s
1
, s
2
, , s
m
} is the sum of the probabilities on the
outcomes contained in the event.
P[B] =
m
i=1
P[{s
i
}]. (3.15)
3.5.4 Theorem on equally likely outcomes
For an experiment with sample space S = {s
1
, , s
n
} in which each outcome s
i
is equally
likely,
P[s
i
] = 1/n for 1 i n. (3.16)
3.5.5 Theorem on no outcome
P[] = 0. (3.17)
3.5.6 Theorem on complementary outcomes
P[A
c
] = 1 P[A]. (3.18)
3.5.7 Theorem on arbitrary sets
For any arbitrary A and B which are not necessarily disjoint,
P[A B] = P[A] +P[B] P[A B]. (3.19)
3.5.8 Theorem on partially ordered sets
If A B, then P[A] P[B]. (3.20)
3.5.9 Theorem on adding probabilities of intersections
This theorem states that for an event A, we can nd its probability by adding the prob-
abilities of the parts of A which are in the separate components of an event space. Thus
for any event A, and event space B
1
, B
2
, , B
m
, (which we know will be both mutually
exclusive and collectively exhaustive)
P[A] =
m
i=1
P[A B
i
]. (3.21)
This is also a very useful theorem.
3.5.10 Exercise
We monitor a phone call, and classify the call as a voice call (V ), if someone is speaking,
or as a data call (D) if the call is carrying a modem or fax signal. We also separately
classify the call as long (L) if the call lasts for more than three minutes, otherwise the
call is classied as brief (B). Based on data collected by the telephone company, we have
available the following probability model: P[V ] = 0.7, P[L] = 0.6, P[V L] = 0.35.
Find the following probabilities.
(a) P[DL] (b) P[D L]
(c) P[V B] (d) P[V L]
(e) P[V D] (f) P[LB]
3.6 Conditional Probability
Consider an experiment that consists of testing two integrated circuits that come from
the same silicon wafer, and observing in each case whether a circuit is accepted (a) or
rejected (r). There are four outcomes rr, ra, ar, and aa of the experiment. Let A denote
the event that the second circuit is a failure. Mathematically, A = {rr, ar}.
The circuits come from a high-quality production line. Therefore the a priori proba-
bility of A is very low. In advance, we are pretty certain that the second circuit will be
acceptable. However, some wafers become contaminated by dust, and these wafers have
a high proportion of defective chips. Let B = {rr, ra} denote the event that the rst chip
tested is rejected.
Given the knowledge that the rst chip was rejected, our knowledge of the quality
of the second chip changes. With the rst chip a reject, the likelihood that the second
chip will also be rejected is higher than it was a priori. Thus we would assign a higher
probability to the rejection of the second chip when we know that the rst chip is a reject.
Such an idea is formalised in the next section.
3.6.1 Denition
We denote the probability of an event A given that the event B is known to have occurred
by P[A|B].
3.6.2 Properties
Note that P[A|B] is a probability measure, relative to the sample space that consist of all
the outcomes in B. Thus P[A|B] has the three properties, corresponding to the axioms
of probability, set out below.
3.6. CONDITIONAL PROBABILITY 39
Positivity
P[A|B] 0. (3.22)
Unity with certainty
P[B|B] = 1. (3.23)
Summing probabilities for mutually exclusive events
If A = A
1
A
2
with A
i
A
j
= for i = j, then
P[A|B] = P[A
1
|B] +P[A
2
|B] + (3.24)
3.6.3 Fundamental formula for conditional probability
P[A|B] =
P[AB]
P[B]
. (3.25)
3.6.4 Law of total probability
If B
1
, B
2
, , B
m
is an event space and P[B
i
] > 0 for i = 1, , m, then
P[A] =
m
i=1
P[A|B
i
]P[B
i
]. (3.26)
3.6.5 Example
A company has three machines B1, B2, and B3 for making 1k resistors. It has been
observed that 80% of resistors produced by B1 are within 50 of the nominal value.
Machine B2 produces 90% of resistors within 50 of the nominal value. The percentage
for machine B3 is 60%. Each hour, machine B1 produces 3000 resistors, B2 produces
4000 resistors and B3 produces 3000 resistors. All of the resistors are mixed together at
random in one bin, and packed for shipment. What is the probability that the company
ships a resistor that is within 50 of the nominal value?
Let A = {resistor is within 50 of the nominal value} . Using the resistor accuracy
information to formulate a probability model, we write
P[A|B
1
] = 0.8, P[A|B
2
] = 0.9, P[A|B
3
] = 0.6 (3.27)
The production gures state that 3000 + 4000 + 3000 = 10,000 resistors per hour
are produced. The fraction from machine B1 is P[B1] = 3000/10000 = 0.3. Similarly,
P[B2] = 0.4 and P[B3] = 0.3. Now it is a simple matter to apply the law of total
probability to nd the accuracy probability for all resistors shipped by the company.
P[A] = P[A|B
1
]P[B
1
] +P[A|B
2
]P[B
2
] +P[A|B
3
]P[B
3
] (3.28)
= (0.8)(0.3) + (0.9)(0.4) + (0.6)(0.3) (3.29)
= 0.78 (3.30)
For the whole factory, 78% of resistors are within 50 of the nominal value.
3.7 Bayes theorem
In many situations, we have advance information about P[A|B], and need to calculate
P[B|A]. To do this we have the following formula
P[B|A] =
P[A|B]P[B]
P[A]
. (3.31)
The proof follows from the result that P[BA] = P[AB], and the fundamental theorem
of conditional probability.
Bayes theorem has a name because it is extremely useful for making inferences about
phenomena that cannot be observed directly. Sometimes these inferences are described
as reasoning about causes when we observe eects.
For example, let B
1
, , B
m
be an event space that includes all possible states of
something that interests us, but we cannot observe directly (for example, the machine that
made a particular resistor). For each possible state, B
i
, we know the a priori probability
P[B
i
] and P[A|B
i
], the probability that an event A occurs (the resistor meets a quality
criterion) if B
i
is the actual state. Now we observe the actual event (either the resistor
passes or fails a test), and we ask about the thing we are interested in (the machines that
might have produced the resistor). That is, we use the Bayes theorem to nd P[B
1
|A],
P[B
2
|A], ,P[B
m
|A]. In performing the calculation, we use the law of total probability
to calculate the denominator in Bayes theorem. Thus for state B
i
P[B
i
|A] =
P[A|B
i
]P[B
i
]
m
i=1
P[A|B
i
]P[B
i
]
. (3.32)
3.8 Independence
3.8.1 Denition
Events A and B are independent events if and only if
P[AB] = P[A]P[B]. (3.33)
The following formulas are equivalent to the denition of independent events A and
B.
P[A|B] = P[A]; P[B|A] = P[B]. (3.34)
To interpret independence, consider probability as a description of our knowledge of
the result of the experiment. P[A] describes our prior knowledge (before the experiment
is performed) that the outcome is included in event A. The fact that the outcome is in
B is partial information about the experiment. P[A|B] reects our knowledge of A when
we learn B occurs. P[A|B] = P[A] states that learning that B occurs does not change
our information about A. It is in this sense that the events are independent.
Keep in mind that independent and disjoint are not synonyms. In some contexts these
words can have similar meanings, but this is not the case in probability. Disjoint events
have no outcomes in common, and therefore P[AB] = 0. In most situations independent
events are not disjoint! Exceptions occur only when P[A] = 0 or P[B] = 0.
3.9. SEQUENTIAL EXPERIMENTS 41
3.9 Sequential Experiments
Many experiments consist of a sequence of sub experiments. The procedure followed for
each sub experiment may depend on the results of the previous sub experiments. We can
use a tree diagram to represent the sequential nature of the sub experiments. Following
the procedure and recording the observations of the experiment is equivalent to following
a sequence of branches from the root of the tree to a leaf. Each leaf corresponds to an
outcome of the experiment.
It is natural to model conditional probabilities in terms of sequential experiments
and to illustrate them with tree diagrams. At the root of the tree, the probability of a
particular event is described by our prior knowledge. If the possible results of the rst sub
experiment are described by the events B
1
, , B
m
, then {B
1
, , B
m
} is an event space.
From the root, we draw branches to each event B
i
. Following a branch from the root
corresponds to observing the result of the rst sub experiment. We label the branches
with the a priori probabilities P[B
1
], , P[B
m
].
For each event B
i
, we have conditional probabilities describing the results of the second
sub experiment. Thus from each of the rst set of branches, we draw a new branch and
label it with the conditional probability. Following a sequence of branches from the root to
a leaf, (a right endpoint of the tree), species the result of each sub experiment. Thus the
leaves represent outcomes of the complete experiment. The probability of each outcome
is the product of the probabilities on the branches between the root of the tree and the
leaf corresponding to the outcome. Generally, we label each leaf with the corresponding
outcome and its probability. We see this procedure illustrated in the following example.
3.9.1 Example
For the resistors of the previous example, we have used A to denote the event that a
randomly chosen resistor is within 50 ohms of the normal value. This could mean
acceptable. Let us use the notation N to be the complement of A i.e. not acceptable.
The experiment of testing a resistor can be viewed as a two-step procedure. First we
identify which machine (B1, B2, or B3) which produced the resistor. Second, we nd out
if the resistor is acceptable. We provide a sequential tree for this experiment in Figure 3.8.
To use the tree to nd the probability of the event B
2
N, a non-acceptable resistor from
Machine B2, we start at the left and nd the probability of reaching B
2
is P[B
2
] = 0.4.
We then move to the right to B
2
N, and multiply P[B
2
] by P[N|B
2
] = 0.1 to obtain
P[B
2
N] = (0.4)(0.1) = 0.04
3.9.2 Observations
We observe in this example a general property of all tree diagrams that represent sequen-
tial experiments. The probabilities on the branches leaving any node add up to 1. This is
a consequence of the law of total probability, and the property of conditional probabilities
that correspond to Axiom 3.
Figure 3.8: Tree diagram for resistor production.
3.10 Counting Methods
3.10.1 Fundamental Principle of Counting
If experiment A has n possible outcomes, and experiment B has k Possible outcomes,
then there are nk possible outcomes when you perform both experiments.
3.10.2 Theorem
The number of permutations of k objects, chosen from n distinguishable objects, is
P
n
k
= (n)
k
= n(n 1)(n 2) (n k + 1) =
n!
(n k)!
(3.35)
.
The two formulae on the left are two common notations for the concept dened. The
two formulae on the right are alternative expressions of the result.
3.10.3 Theorem
The number of combinations of k objects, chosen from n distinguishable objects, is
C
n
k
=
_
n
k
_
=
n(n 1)(n 2) (n k + 1)
k(k 1)(k 2) 1
=
n!
(n k)!k!
(3.36)
.
The two formulae on the left are two common notations for the concept dened. The
formula on the right is the result. The result may be derived from the result in the previous
section, when we take note that when considering combinations, as we are this section,
we do not distinguish the order of the chosen objects, but that for the k distinguishable
3.11. CHAPTER SUMMARY 43
chosen objects, there are k! ways in which they could be ordered to become a permutation
as considered in the previous section.
3.10.4 Theorem
Given n distinguishable objects, there are n
k
ways to choose, with replacement, an ordered
sample of k objects. While the replacement of an object, after it is chosen, may produce
some objects which are not distinguishable, the objects remain distinguishable in the
sample by their positions.
3.11 Chapter Summary
This chapter introduced the model of an experiment consisting of a procedure and
observations.
Outcomes of the experiments are elements of a sample space.
Probability is a number assigned to every set in the sample space.
Three axioms contain the fundamental properties of probability. These axioms are
used to develop methods of working on practical problems.
Sample space, event and outcome are the probability theory terms for the set theory
concepts of universal set, set and element.
A probability measure P[A] is a function that assigns a number between 0 and 1 to
every event A in the sample space. The assigned probabilities conform to the three
axioms.
A conditional probability P[A|B] describes the likelihood of A given that B has
occurred. The conditional probability P[A|B] also satises the three axioms of
probability.
Tree diagrams illustrate experiments that consist of a sequence of steps. The labels
on the tree branches can be used to calculate the probabilities of outcomes of a
combined experiment.
Counting methods can be used to determine the number of outcomes of complicated
experiments.
Chapter 4
RANDOM VARIABLES
4.1 Review of Probability Theory
4.1.1 Probability
We review in this section some of the important results of the previous chapter.
If A
1
, A
2
, . . . , A
n
are possible outcomes of an experiment or measurement, then the
probability of A
i
is:
P[A
i
] = lim
N
N
i
N
(4.1)
where N is the total number of experiments, and N
i
is the number of experiments in
which A
i
occurred. Note that
n
i=1
P[A
i
] = 1. (4.2)
4.1.2 Conditional probability
Similarly, if B
1
, B
2
, . . . , B
m
are also possible outcomes of the same experiment, then the
joint probability of A
i
and B
j
is
P[A
i
, B
j
] = lim
N
N
ij
N
(4.3)
where N
ij
is the number of experiments in which both A
i
and B
j
occurred.
If we consider only those experiments in which B
j
occurred, then the conditional
probability of A
i
given B
j
is
P[A
i
| B
j
] = lim
N
N
ij
N
j
(4.4)
where N
j
is the number of experiments in which B
j
occurred.
The | symbol means that the quantities after it are known (or given) and are not
random variables.
45
46 CHAPTER 4. RANDOM VARIABLES
4.1.3 Bayes rule
This leads to Bayes Rule which we will use quite often.
P[A
i
, B
j
] = P[A
i
] P[B
j
| A
i
] = P[B
j
] P[A
i
| B
j
] (4.5)
If P[A
i
| B
j
] = P[A
i
] for all i, j, then all the A
i
, B
j
are independent and P[A
i
, B
j
] =
P[A
i
] P[B
j
].
4.1.4 The Binomial Distribution
The binomial distribution arises when we consider n identical objects each of which can
be placed either in box A with probability p or in box B with probability q = 1 p. We
want the probability P(r) that there are r objects in box A.
We may regard the experiment as performing a series of n trials. For each trial we
allocate the outcome with probability p to box A and probability q = 1p to box b. The
event in which we are interested is where r outcomes are allocated to box a.
The sample space is a series of outcomes like aababb with n elements in the string.
The event space consists of all strings of that form with r factors of a. All such strings
are equi-probable, with probability p
r
q
nr
. We ask how many of them are there?
Now there are C
n
r
ways of arranging n objects with r in box A and n r in box B,
where
C
n
r
=
_
n
r
_
=
n!
r!(n r)!
. (4.6)
Each of these arrangements has probability p
r
q
nr
, so we have
P(r) =
_
n
r
_
p
r
q
nr
. (4.7)
The variable r is called a random variable.
4.2 Random Variables
4.2.1 Notation
Random variables arise when a number can be assigned to the outcome of an experi-
ment. If the outcome of an experiment is designated , and to every outcome we assign
a number X(), then X is called a random variable. We will normally use upper case
letters, e.g. X, for random variables, and lower case letters, e.g. x, for the values which
they may take.
4.2.2 Cumulative Distribution Function
(a) Denition
We dene the cumulative distribution function F
X
(x) as the probability of obtaining
an outcome for which X() x where x is a number. In an equation we write this as
4.2. RANDOM VARIABLES 47
F
X
(x) = P[; X() x] (4.8)
or simply as
P[X() x] (4.9)
(b) Properties
For any discrete random variable with range S
X
= {x
1
, x2, } satisfying x
1
x
2
,
we have the properties
Going from left to right on the x-axis, F(x) starts at zero and ends at 1.
The cumulative distribution function never decreases as it goes from left to right.
For each discrete random variable X there is a jump, (i.e. a discontinuity) at each
value of x S
X
. The height of the jump at x
i
is P
X
(x
i
).
Between jumps, the graph of the cumulative distribution function of a discrete
random variable X is a straight line.
These properties are illustrated in Figure 4.1. The cumulative distribution function
may be written simply as F(x) if there is no confusion over which random variable it
belongs to.
If the random variable X takes on only discrete values, then F
X
(x) is a series of steps
as shown in Figure 4.1, each step being the probability of that particular value.
Figure 4.1: Distribution functions of continuous and discrete variables.
If X takes on a continuous range of values then F
X
(x) is a continuous function. It
is also possible to have random variables which are a mixture of discrete and continuous
values.
4.2.3 Probability Density Function
(a) Denition
The probability density function (pdf) of a continuous random variable is the derivative
of the distribution function and is shown in Figure 4.2.
p
X
(x) =
dF
X
(x)
dx
= lim
dx0
P{x X x +dx}
dx
. (4.10)
This may also be written p(x) if there is no confusion over which random variable it
refers to.
(b) Properties
We should be able to see from the above equation that p
X
dx is probability that the
outcome of the experiment will produce an X in the range x X x +dx
The area under p
X
(x) is unity, because
_
+
p
X
(x) dx = F
X
() F
X
() = 1. (4.11)
Also because F
X
(x) is non-decreasing, we must have p
X
(x) 0.
(c) Appearance
The general appearance of a probability density function of a continuous variable is shown
in Figure 4.2.
Figure 4.2: Probability density function of a continuous variable.
We can also dene the pdf of a discrete random variable by use of the delta function
(x) as the derivative of a unit step u(x). Hence for a discrete random variable X taking
n possible discrete values x
i
with probabilities P
i
, we have
p
X
(x) =
n
i=1
P
i
(x x
i
). (4.12)
The general appearance of a probability density function of a discrete variable is shown
in Figure 4.3.
4.2.4 Joint Distribution and Density Functions
The joint distribution function F
XY
(x, y) of two random variables X and Y is dened
as
F
XY
(x, y) = P{X x, Y y}. (4.13)
The joint pdf is
4.3. EXPECTATION AND MOMENTS 49
Figure 4.3: Probability density function of a discrete variable.
p
XY
(x, y) =

2
F
XY
(x, y)
xy
(4.14)
= lim
dx,dy0
P{x X x +dx, y Y y +dy}
dx dy
(4.15)
The joint density function is related to the one dimensional density functions by the
relations
p
X
(x) =
_
+
p
XY
(x, y) dy (4.16)
p
Y
(y) =
_
+
p
XY
(x, y) dx (4.17)
If X and Y are independent, then p
XY
(x, y) = p
X
(x) p
Y
(y).
4.2.5 Conditional Density Function
This is written as p
X|Y
(x | y), where the vertical bar means that the quantities after it
are known.
p
X|Y
(x| y) = lim
dx0
P{x X x +dx| Y = y}
dx
(4.18)
=
p
XY
(x, y)
p
Y
(y)
(Bayes Rule for pdf s) (4.19)
4.3 Expectation and Moments
4.3.1 Mean
If X is a random variable, then its average, mean or expected value is given by
X
= E{X} =
_

x p
X
(x) dx (4.20)
or in the case of a discrete random variable
X
= E{X} =
n
i=1
x
i
P(X = x
i
). (4.21)
In general, if Y = f(X) we have
E{Y } =
_

f(x) p
X
(x) dx. (4.22)
4.3.2 Moments
The moments m
k
of a random variable are dened as
m
k
= E{X
k
} =
_

x
k
p
X
(x) dx. (4.23)
4.3.3 Central moments
The central moments
k
are the moments of the zero mean random variable X
X
.
k
= E{(X
X
)
k
} =
_

(x
X
)
k
p
X
(x) dx. (4.24)
Note that:
m
1
=
X
= mean
1
= 0
2
= m
2
2
X
=
2
X
= variance
4.3.4 Example
The mean of the binomial distribution may be obtained from the identity
(p +q)
n
=
n
r=0
_
n
r
_
p
r
q
nr
(4.25)
Dierentiating both sides, but the right hand side explicitly, gives
(p +q)
n
p
=
n
r=0
_
n
r
_
r p
r1
q
nr
(4.26)
Now dierentiating the left hand side explicitly gives
n(p +q)
n1
=
n
r=0
_
n
r
_
r p
r1
q
nr
(4.27)
Multiplication of both sides by p then gives
np(p +q)
n1
=
n
r=0
_
n
r
_
r p
r
q
nr
(4.28)
The right hand side may be recognised as E(r) as given by equation 4.20, and the left
hand side may be simplied, in the case in which we are interested, by putting p +q = 1,
so
E{r} = np. (4.29)
4.4. THE GAUSSIAN DISTRIBUTION 51
4.4 The Gaussian Distribution
4.4.1 One Dimensional Gaussian Distribution
The Gaussian (or normal) distribution arises in many situations because it happens to
be the limiting distribution when a large number of arbitrary random variables are added
together. This fact is called the central limit theorem).
The Gaussian distribution is dened by two parameters, its mean and its variance
2
. The probability density function of a Gaussian random variable X is shown in Fig-
ure 4.4and is given by
p
X
(x) =
1
2
e
(x)
2
/2
2
. (4.30)
Figure 4.4: Gaussian probability density function.
The distribution function F
X
(x) cannot be expressed in terms of elementary functions,
but is of considerable interest because it gives the probability of error in digital systems
subject to additive Gaussian noise. It can be expressed in terms of a standard function
Q(x) dened as
Q(x) =
1
2
_

x
e
t
2
/2
dt. (4.31)
The Q function is generally available from tables, such as are provided in Appendix B
of thee notes. Its form is shown in Figure 4.5.
A little bit of elementary calculus gives us the result for the cumulative distribution
function
F
X
(x) =
_
x
p
X
() d = 1 Q
_
x
_
(4.32)
We recall that the cumulative distribution function F
X
(x) gives the probability that X
will be less than x. Since errors occur in a digital system when the noise random variable
X exceeds some threshold value a, the Q function, which gives for a Gaussian probability
distribution the probability that X will exceed some x, is generally tabulated rather than
F
X
(x). The probability of error is therfore given by
Figure 4.5: The Q function.
P
e
= P{X > a} = Q
_
a
_
(4.33)
Usually = 0 in such applications.
4.4.2 Two Dimensional Gaussian Distribution
In two dimensions, the joint distribution of two Gaussian variables with zero means is
p(x
1
, x
2
) =
1
2
1
1
2
exp
_
1
2(1
2
)
_
x
2
1
2
1

2x
1
x
2
2
+
x
2
2
2
2
__
(4.34)
where
2
i
= E{x
2
i
} = variance (4.35)
(4.36)
=
E{x
1
x
2
}
2
= correlation coecient (4.37)
The correlation coecient satises 1 1, with = 1 corresponding to full
correlation (e.g.. x
2
= ax
1
), and = 0 corresponding to zero correlation.
If x
1
and x
2
are uncorrelated, then they are independent, since putting = 0 in the
above gives
p(x
1
, x
2
) =
1
2
1
2
exp
_
1
2
_
x
2
1
2
1
+
x
2
2
2
2
__
(4.38)
(4.39)
= p(x
1
) p(x
2
). (4.40)
For non-zero means, we simply replace x
1
by x
1

1
and x
2
by x
2

2
in the above
formulae.
4.5. TRANSFORMATION OF RANDOM VARIABLES 53
4.5 Transformation of Random Variables
4.5.1 Single Variable Transformations
If X = X() is a random variable, and Y = f[X], where f[.] is a deterministic function
as shown in Figure 4.6, then Y = f[X()] = Y () is also a random variable.
Figure 4.6: Functional relation between random variables.
We note that the values of x and y which the random variables actually take also
satisfy the same functional relational dependence y = f(x).
We can nd the cumulative distribution function and the probability density function
of Y as follows
F
Y
(y) = P{Y y} (4.41)
= P{X x} (4.42)
= F
X
(x). (4.43)
p
Y
(y) =
dF
Y
(y)
dy
(4.44)
=
dF
X
(x)
dx
dx
dy
(4.45)
= p
x
(x)
dx
dy
. (4.46)
Hence p
Y
(y) dy = p
X
(x) dx (4.47)
A more general situation shown in Figure 4.7. Here there are several values of X which
lead to the same value Y , so each of these must be included in the calculation of p
Y
(y).
Note that reverse calculation cannot be done. That is, if the relationship between X
and Y is as shown, then a knowledge of p
Y
(y) will not uniquely dene p
X
(x), since the
value of X which corresponds to a particular value of Y is not unique.
Figure 4.7: General functional relation.
Taking note that in this case
F
Y
(y) = F
X
(x
1
) F
X
(x
2
) +F
X
(x
3
) (4.48)
we see that
p
Y
(y) = p
X
(x
1
)
dx
1
dy
p
X
(x
2
)
dx
2
dy
+p
X
(x
3
)
dx
3
dy
. (4.49)
We note that, as dx
2
/dy is negative, all terms give positive contributions. Generalising
the result, we say
p
Y
(y) =

i
p
X
(x
i
)
dx
i
dy
(4.50)
4.5.2 Example
Y = X
2
(4.51)
X =
Y (4.52)
dX
dY
=
1
2
Y
(4.53)
p
Y
(y) =
1
2
y
[p
X
(
y) +p
X
(
y)] (4.54)
4.5.3 Two-variable transformations
If X
1
and X
2
are random variables which together transform to variables Y
1
and Y
2
by
relations of the form
Y
1
= Y
1
(X
1
, X
2
) (4.55)
Y
2
= Y
2
(X
1
, X
2
) (4.56)
4.5. TRANSFORMATION OF RANDOM VARIABLES 55
or expressed in another way
X
1
= X
1
(Y
1
, Y
2
) (4.57)
X
2
= X
2
(Y
1
, Y
2
) (4.58)
then the joint probability density function of Y
1
and Y
2
is given by
p
Y
(y
1
, y
2
) = J p
X
(x
1
, x
2
) (4.59)
where J =
x
1
/y
1
x
1
/y
2
x
2
/y
1
x
2
/y
2
(4.60)
=
(x
1
, x
2
)
(y
1
, y
2
)
(4.61)
= Jacobian of transformation. (4.62)
4.5.4 Example
In the Cartesian to polar conversation shown in Figure 4.8, given p
XY
(x, y), nd p
R
(r, ).
Figure 4.8: Cartesian to polar conversion.
X = R cos (4.63)
Y = R sin (4.64)
J =
(x, y)
(r, )
(4.65)
=
cos r sin
sin r cos
(4.66)
= r. (4.67)
Hence p
R
(r, ) = r p
XY
(r cos , r sin ). (4.68)
4.5.5 Application to Gaussian variables
In particular, if X and Y are independent Gaussian variables of zero mean and equal
variance
2
, then
p
XY
(x, y) =
1
2
2
e
(x
2
+y
2
)/2
2
(4.69)
p
R
(r, ) =
r
2
2
e
r
2
/2
2
(4.70)
p
R
(r) =
r
2
e
r
2
/2
2
(Rayleigh distribution) (4.71)
p
() =
1
2
(Uniform distribution). (4.72)
The extension to multivariable transformations follows the same procedure.
4.6 Characteristic Function
4.6.1 Denition
The characteristic function
X
(u) of a random variable X is dened as
X
(u) = E{e
juX
} =
_

p
X
(x) e
jux
dx. (4.73)
This formula is slightly dierent from but can be related to the Fourier transform.
The inverse relation is
p
X
(x) =
1
2
_

X
(u) e
jux
du. (4.74)
Since
e
jux
=

k=0
(jux)
k
k!
(4.75)
we have
X
(u) =

k=0
(ju)
k
k!
m
k
(4.76)
provided the series converges. If X and Y are independent, and Z = X +Y , and we
require
F
Z
(z) = P{X +Y z} (4.77)
we can allow X to range over all values, provided we restrict Y to satisfy the inequality
X +Y z, and we obtain
F
Z
(z) =
_

x=
_
zx
y=
p
X
(x) p
Y
(y) dxdy (4.78)
4.6. CHARACTERISTIC FUNCTION 57
Recalling that a denite integral is a really a function of its limits we obtain
p
Z
(z) =
F
Z
(z)
z
(4.79)
=
_

p
X
(x) p
Y
(z x) dx (4.80)
This last expression may be recognised as a convolution, and as the Fourier transform
of a convolution is the product of the individual Fourier transforms, we have
Z
(u) =
X
(u)
Y
(u). (4.81)
4.6.2 Examples
(a) The binomial distribution
For the binomial distribution
r
(u) =
n
r=0
_
n
r
_
e
jur
p
r
q
nr
(4.82)
= (pe
ju
+q)
n
(4.83)
= 1 +junp (npq +n
2
p
2
)
u
2
2
+ (4.84)
This then gives m
1
= E{r} = np and m
2
= E{r
2
} = npq + n
2
p
2
. The variance is
therefore
2
= E{(r np)
2
} = npq.
(b) The Gaussian distribution
For the Gaussian distribution
p
X
(x) =
1
2
e
(x)
2
/2
2
(4.85)
X
(u) =
_

2
e
(x)
2
/2
2
e
jux
dx (4.86)
To make progress we merge the two exponential factors, complete the square in the
exponent, and nd we have, in the exponent, a constant left over, which, after placement
as a separate exponential factor, can be removed from the integral sign. The result is
X
(u) = e
(juu
2
2
/2)
_

2
e
(xju
2
)
2
/2
2
dx (4.87)
The integral reduces to a constant of value unity, so we have
X
(u) = e
(juu
2
2
/2)
(4.88)
4.7 Second Characteristic Function
4.7.1 Denition
There is also a second characteristic function dened as
X
(u) = ln
X
(u) =

k=1
(ju)
k
k!
c
k
(4.89)
where c
k
are called the cumulants. The rst four cumulants are
c
1
= m
1
= mean
c
2
=
2
= variance
c
3
=
3
= skewness
c
4
=
4
3
2
2
= excess.
(4.90)
4.7.2 Cumulants of the Gaussian distribution
X
(u) = ln
X
(u) = ju
1
2
u
2
2
. (4.91)
Hence for a Gaussian random variable, only the rst two cumulants are non-zero, so
for an arbitrary distribution the third and fourth cumulants are used as an indication of
how close the probability density function is to Gaussian.
4.7.3 Further properties
The second characteristic function has the property that it is additive for the sum of in-
dependent variables. This property therefore also applies to the cumulants. For example,
if X and Y are independent and Z = X +Y , then
Z
(u) =
X
(u) +
Y
(u). (4.92)
In particular, the means and variances also add.
Z
=
X
+
Y
(4.93)
2
Z
=
2
X
+
2
Y
(4.94)
This property extends to the sum of any number of independent variables.
4.8 Chapter Summary
In this chapter we have
Made a review of probability theory.
Introduced the idea of a random variable through the binomial distribution.
Dened the cumulative distribution function and the probability density distribution
for both discrete and continuous random variables.
Dened joint and conditional density functions for a pair of random variables.
Dened the expected value and the moments and central moments for a random
variable.
Dened the Gaussian distribution in one and two dimensions.
Dened the Q function and established its relation to the distribution function for
a Gaussian random variable.
Considered a transformation of a random variable and derived a formula for the
probability density distribution of the transformed variable.
Stated the corresponding formula for two-variable transformations.
Applied that formula to the case of a rectangular to polar transformation.
Dened the characteristic function and established its utility in the calculation of
moments.
Introduced the second characteristic function and stated some of its properties.
Chapter 5
RANDOM PROCESSES
5.1 General Concepts
A random variable results when a number is assigned to each outcome of an experi-
ment or measurement. If to each outcome we assign a time function X(t, ) according
to some rule, then this family of time functions is called a random process. The functions
are usually dened for t , but this is not necessary. An illustration is provided
in Figure 5.1.
Figure 5.1: Conceptual representation of a random process.
For a particular outcome
j
, the time function X(t,
j
) is a single deterministic time
function, and is called a realisation of the random process. For a specied time t
i
, X(t
i
, )
is a random variable, taking on dierent values as varies. Finally, X(t
i
,
j
) is a xed
number.
Usually we omit the functional dependence of X on , but note that this can lead to
61
62 CHAPTER 5. RANDOM PROCESSES
ambiguities as X(t) may represent
(i) X(t, ) the random process
(ii) X(t,
j
) a particular realisation
(iii) X(t
i
, ) a random variable
(iv) X(t
i
,
j
) a xed number.
Usually the context will make it clear which interpretation is to be used. Two random
processes are said to be identical if X(t,
j
) = Y (t,
j
) for all t and
j
.
5.2 First and Second Order Statistics
5.2.1 First order
The distribution function F
X
(x; t) is dened as
F
X
(x; t) = P{ | X(t, ) x} = P{X(t) x} (5.1)
and the probability density function is
p
X
(x; t) =
F
X
(x; t)
x
. (5.2)
In general these are functions of both x and t as shown in Figure 5.2.
5.2.2 Second order
Given two instants of time t
1
and t
2
, we have two random variables X(t
1
) and X(t
2
).
Their joint distribution function is
F
X
1
X
2
(x
1
, x
2
; t
1
, t
2
) = P{X(t
1
) x
1
, X(t
2
) x
2
}. (5.3)
Figure 5.2: Probability density function of a random process.
5.3. MEAN, AUTOCORRELATION & AUTOCOVARIANCE 63
Joint and conditional density functions are dened in the same way as for random
variables, i.e.
p
X
1
X
2
(x
1
, x
2
; t
1
, t
2
) =

2
F
X
1
X
2
(x
1
, x
2
; t
1
, t
2
)
x
1
x
2
. (5.4)
p
X
1
|X
2
(x
1
; t
1
| x
2
; t
2
) =
p
X
1
X
2
(x
1
, x
2
; t
1
, t
2
)
p
X
2
(x
2
; t
2
)
. (5.5)
The latter is interpreted as the density function of X(t
1
) for those realisations of X(t)
for which X(t
2
) = x
2
, as is illustrated in Figure 5.3.
Figure 5.3: Conditional probability density function.
5.3 Mean, Autocorrelation & Autocovariance
5.3.1 Mean
The mean
X
(t) is dened as
X
(t) = E{X(t)} =
_

xp
X
(x; t) dx. (5.6)
This is a deterministic function of time and is not a random variable.
5.3.2 Autocorrelation
The autocorrelation function R
XX
(t
1
, t
2
) is dened as
R
XX
(t
1
, t
2
) = E{X(t
1
) X(t
2
)} =
_

x
1
x
2
p
X
1
X
2
(x
1
, x
2
; t
1
, t
2
) dx
1
dx
2
. (5.7)
This is also a deterministic function of the times t
1
and t
2
, and is not a random
variable.
5.3.3 Autocovariance
The autocovariance function C
XX
(t
1
, t
2
) is the autocorrelation function of the zero mean
process X(t)
X
(t). Thus
C
XX
(t
1
, t
2
) = E{ [X(t
1
)
X
(t
1
)] [X(t
2
)
X
(t
2
)] } (5.8)
= R
XX
(t
1
, t
2
)
X
(t
1
)
X
(t
2
). (5.9)
This too is a deterministic function of the times t
1
and t
2
, and is not a random variable.
5.3.4 Variance
The variance
2
X
(t) is
2
X
(t) = C
XX
(t, t) = R
XX
(t, t)
2
X
(t) (5.10)
This is a deterministic function of the time t and is not a random variable.
5.3.5 Commentary
We see that all of the recently dened functions are ensemble averages, and depend on
the way the ensemble evolves with time. Sometimes there is an evolution, and sometimes
there is not. We will encounter later the class of stationary systems which appear not
to evolve. Those systems will be found to have constant (i.e. independent of time) rst
order statistics, but have second order statistics (autocorrelation, autocovariance) which
do depend upon the time dierence t
1
t
2
.
5.4 Poisson Processes
5.4.1 Objective
We begin the study here of a number of Poisson processes which are concerned with the
random arrival of events in time. Our eventual objective is to determine the statistics of
a Poisson impulse process which is dened in Section 5.4.8. The result will be and is used
in Section 7.3 of a later chapter for calculation of statistics of what is called shot noise.
5.4.2 How the argument will unfold
(a) The Poisson point process
We begin with the denition of The Poisson point process, which is dened as the random
arrival of some unspecied events with uniform probability per unit time, and develop the
probability density function and the statistics of that process, focusing on the probability
that various numbers of events n may occur in a time interval t.
5.4. POISSON PROCESSES 65
(b) Detour
We will make a detour to calculate a probability density function for the time between
pulses, but this is not part of the main thread of the argument.
(c) The X process
Returning to the main thread of the argument, we provide the denition of The X process
where X is the number of events which have occurred over time from zero to t. The X
process is thankfully unrelated to The X les. The process is substantially the same
process as the Poisson point process originally introduced, but the focus in on now on
how the number of events X observed evolves over time. The viewpoint is in accord with
that introduced in Section 5.1.
(d) Relevant observations
We will take note of the fact that a process which considers the number of pulses which
occur over a time t, starting at any point to in time, will be the same as for the X
processes where the time t starts at time t = 0.
We will also take note of an additive property of time intervals and also of the X
process, which states that not only is the total duration of two adjacent time intervals
the sum of the individual time intervals, but the total number of events in the sum of two
time intervals is the sum of the numbers of events in the two time intervals.
(e) The Y process
We next dene The Y Process which consists of considering the number of events which
occur over a period of time a, starting at time t, i.e. over the period (t, t +a).
The rst order statistics will be found to be the same as for the X process, if the
interval a in the Y process is identied with the interval t in the X process. In the
calculation of the second order statistics of the Y process, we will be considering intervals
which begin at dierent starting points t
1
and t
2
. We will nd in the investigation that we
need to take account of whether or not the two intervals (t
1
, t
1
+a) and (t
2
, t
2
+a) overlap
one another. This is because when the two intervals do not overlap, events within them
are completely independent, but when they do overlap, partially or completely, events
within them are not completely independent.
(f) The Z process
We will be aided in this process of considering possibly overlapping intervals by the
denition of The Z Process, which is closely related to the X process, but which considers
the statistics of the number of events which occur in an interval of adjustable duration
and adjustable starting point. Fortunately, because all the intervals for which we consider
the Z process statistics either do not overlap at all or overlap completely, we nd that
we need only consider the mean and second moment of that process, and we will not be
concerned with its full autocorrelation function.
(g) The P process
After establishing the statistics of the Y process we will proceed to the The Poisson pulse
process, which we will call The P process, which is the same as the Y process, but is
simply scaled in amplitude, so that it is in a form to make, soon after the denition, the
transition to the Poisson impulse process which we call The I process.
(h) The I process
This process makes the events which occur at each Poisson point into impulses. The
transition is done by allowing the interval a, involved in the denition of the Y and P
processes, to tend to zero.
(i) Return to the X process
With the aid of the insights developed along the way, we return to consider the statistics
of the Poisson X process, sometimes called on account of its appearance The Poisson step
process, and derive the mean and correlation function for that process.
5.4.3 The Poisson point process
(a) Denition
Suppose we have points in time t
1
, t
2
, . . . , t
n
. . . distributed randomly and indepen-
dently, such that the average density is per second, as is shown in Figure 5.4.
Figure 5.4: Poisson point process.
Let P(n; t) be the probability of nding n points in a time interval of length t.
(b) Probability density function
We will derive the probability density function by a limiting procedure. Suppose that we
distribute N points at random along a time interval of length T. The average density is
therefore = N/T.
The probability of each point being within a particular interval of length t is p = t/T =
t/N, and the probability of that point being outside that interval is q = 1p = (1
t
N
).
The probability of n points being in that interval is therefore given by the binomial
distribution
P(n; t) =
_
N
n
__
t
N
_
n
_
1
t
N
_
Nn
(5.11)
=
N(N 1) (N n + 1)
n!
_
t
N
_
n
_
1
t
N
_
Nn
. (5.12)
Now we let N and T tend to innity, but keep constant. We obtain
P(n; t) =
(t)
n
n!
e
t
(5.13)
where we have used the relations
lim
N
_
1
x
N
_
N
= e
x
(5.14)
and
lim
N
_
1
x
N
_
n
= 1. (5.15)
(c) Illustration
The probabilities P(n; t) are shown in Figure 5.5 for t 2.5. The maximum occurs at
n t. we note that possible outcomes include nding any number of points, including
zero, in the interval. The probability of nding a large number of points in the interval
becomes small, but is never zero.
Figure 5.5: Probabilities of number of points in an interval.
(d) Mean and variance
The random variable n has a mean
= t (5.16)
and variance
2
= t (5.17)
Both results are easily shown by computing the characteristic function
(u) = E{e
jun
} (5.18)
=

n=0
e
t
(t e
ju
)
n
n!
(5.19)
= e
t (e
ju
1)
(5.20)
= 1 +jut +
(ju)
2
2!
(t +
2
t
2
) + (5.21)
Inspection of this function gives the rst and second moments
m
1
= t (5.22)
m
2
= t +
2
t
2
. (5.23)
from which we obtain the above results for mean and variance.
(e) Time between pulses
The probability density function of the separation time T
s
between successive points can
be derived as follows. The cumulative probability distribution that the time T
s
between
successive points will be less than some time t
s
is
P[T
s
< t
s
] = P[at least one point in an interval t
s
] (5.24)
= P[no points in an interval t
s
] (5.25)
= 1 P(0; t
s
) = 1 e
t
s
. (5.26)
Thus the probability density function of the time t
s
between successive pulses is
p
s
(t
s
) = e
ts
. (5.27)
This function is illustrated in Figure 5.6.
Figure 5.6: Probability density function of time between successive pulses.
5.4.4 The X process
(a) Denition
We dene The X process where X(t) = the number of points in the interval (0, t). An
illustration of the X process is provided in Figure 5.7.
Figure 5.7: X variable for Poisson point process.
(b) Statistics
For this process we may derive from Section 5.4.3 the statistics
m
1
= t (5.28)
m
2
= t +
2
t
2
(5.29)
Hence
2
= t (5.30)
(c) Remarks
We will encounter later in Section 5.7 the concept of stationary processes. One of the
several important properties of stationary processes is that both the mean and variance
are independent of time. Clearly the X process, in which both the mean and second
moment depend upon time, is not stationary. But we will be able to derive from the X
process a process which is stationary, and we will do so in Section 5.4.6.
5.4.5 Additive properties of intervals
We point out here a property which will be useful in the further development. If we have
two intervals of lengths a and b and they adjoin, the sum of the intervals is a +b, and the
number of points in the whole interval is the sum of the number of points in the separate
intervals. So if X(t) is the number of points in the interval (0, t), then the number of
points in an interval (t, t +a) is X(t +a) X(t).
5.4.6 The Y process
(a) Denition
We now dene The Y process as the process in which Y (t) is the number of points in the
interval (t, t +a).
(b) Illustration
Depending on the spacings of the points and the interval size a, we may have zero, one
or more points in the interval, so the Y process appears in one realisation as shown in
Figure 5.8.
Figure 5.8: Y variable for Poisson point process.
(c) First order statistics
We have already noted that Y (t) = X(t + a) X(t), so the expected value of the Y
process is
E{Y (t)} = E{X(t +a)} E{X(t)} = (t +a) (t) = a (5.31)
(d) Second order statistics
We now proceed to the calculation of the correlation function E{Y (t
1
)Y (t
2
)}. The in-
tervals (t
1
, t
1
+ a) and (t
2
, t
2
+ a) involved in this calculation are shown below, in two
dierent congurations, one where they overlap, and one where they do not.
Figure 5.9: Overlapping and non-overlapping intervals.
In the diagram, we simplify the identication of the intervals (t
1
, t
1
+a) and (t
2
, t
2
+a)
by calling them intervals P and Q respectively.
In the case where there is overlap, we divide the original P and Q intervals into sub-
intervals A and B and C and D respectively. The intervals B and C, which are the same
interval, are the region of overlap.
(c) The Z process
As the Y process is concerned with intervals with xed length a, and we are about to
consider the properties of processes where the intervals are adjustable in size as well as
in position, we introduce the Z process, closely related to the X and Y processes, where
the variable Z is the number of points found in an interval in which both the start and
end points can be selected by ourselves.
We will be greatly aided in our development by the two facts that: (i) the mean and
second moment of the Z process are the same as for the X process of the same duration,
and (ii) we will require only those statistics of the process rather than having to derive
the full correlation function of the Z process.
Now the variable E{Y (t
1
)Y (t
2
)} we are seeking can be expressed in the alternative
notation E{Y (P)Y (Q)}. In the case of overlap, we can perform the subdivision of the
intervals, and take note of the additive property of the Y variables for separate intervals,
to see that the quantity we are seeking is E{[Z(A) +Z(B)][(Z(C) +Z(D)]}.
In the case of no overlap, the number of points in the intervals P and Q are inde-
pendent, and the expectation of the product is the product of the means, which are both
known to be a, so the result is
2
a
2
.
Alternatively, in this case of overlap, we consider the result of E{[Z(A)+Z(B)][Z(C)+
Z(D)]}. In the expansion of product, we will have four terms, in which there are three
cases of non-overlapping intervals, and one case of an interval (B) overlapping with itself
(C).
For the non-overlapping intervals, we make use of the independence just mentioned to
say the result is the product of the means of the two intervals. Thus for these terms we
have
E{Z(A)Z(C)} =
2
(t
2
t
1
)(t
1
+a t
2
) (5.32)
E{Z(A)Z(D)} =
2
(t
2
t
1
)(t
2
t
1
) (5.33)
E{Z(B)Z(D)} =
2
(t
1
+a t
2
)(t
2
t
1
). (5.34)
For the fourth term, we have to calculate E{Z(B)Z(C)}. As explained above, this is
just the second moment of the Z function when its duration is (t
1
+a t
2
). This is same
as for X function for this time duration, so is (t
1
+a t
2
) +
2
(t
1
+a t
2
)
2
.
Summing all four terms then gives
E{Y (P)Y (Q)} =
2
(t
2
t
1
)(t
1
+a t
2
) (5.35)
+
2
(t
2
t
1
)(t
2
t
1
)
+
2
(t
1
+a t
2
)(t
2
t
1
)
+ (t
1
+a t
2
) +
2
(t
1
+a t
2
)
2
.
After some algebraic simplication we arrive at
E{Y (P)Y (Q)} +
2
a
2
+ (t
1
+a t
2
). (5.36)
Now this last term can be expressed as t
toverlap
, where the overlap is between the
P and Q segments. If there is no overlap, we already have
2
a
2
as our result. Thus we
can regard
2
a
2
+ t
overlap
as our formula for both the overlap and no overlap cases.
We still have the assumption that t
2
> t
1
in place. When t
2
< t
1
, we could rst change
our formula to one where new variables t
3
and t
4
replace t
1
and t
2
, and then set t
1
= t
3
and t
2
= t
4
so that now t
2
< t
1
when t
4
> t
3
. Our formula
2
a
2
+
2
(t
3
+ a t
4
)
2
for
the case of overlap will become
2
a
2
+
2
(t
2
+a t
1
)
2
, which can again be translated as
2
+ t
overlap
. Thus we will regard the formula
R
Y Y
(t
1
, t
2
) =
2
a
2
+ t
overlap
(5.37)
as covering all cases, whatever the relation between t
1
and t
2
, and whether or not there
is overlap.
5.4.7 The P process
We now introduce the P process, which is an amplitude rescaled version of the Y process,
by dening
P =
Y
a
. (5.38)
What we are doing is converting the variable to something which by its area represents
the same thing as the Y process represents, at least for intervals containing single events,
by it amplitude. When only a single point is found in a time interval a, we are producing
a pulse (not yet an impulse) whose area represents the number (one) of pulses in the
interval. The statistics of this process are easily derived from those of the Y process to
be
E{P} =
a
a
= (5.39)
R
PP
=

2
a
2
+ t
overlap
a
2
=
2
+
t
overlap
a
2
(5.40)
An illustration of the correlation function is provided in Figure 5.10.
It is notable that the correlation function depends only on the time dierence t
1
t
2
.
We will see later that this is a general property of stationary processes, of which this
process is an example.
5.4.8 The Poisson impulse process
In the limit as a 0, we obtain the derivative of the original Poisson process. This is a
Poisson impulse process. An illustration is provided in Figure 5.11.
We may summarise its statistics as
Figure 5.10: Correlation function of a Poisson pulse process.
Figure 5.11: Poisson impulse process.
I(t) = lim
a0
P(t) =

i
(t t
i
) (5.41)
E{I(t)} = (5.42)
R
II
(t
1
, t
2
) =
2
+ (t
2
t
1
). (5.43)
These results are the natural evolution of the results of the last section, and are in
accord with Figure 5.10, wherein we see that in the limit as a 0, the triangular part of
the autocorrelation function becomes, since its area is , the delta function (t
2
t
1
) .
5.4.9 Return to the X process
With the aid of the insights developed along the way, we return to consider the statistics
of the Poisson X process, sometimes called on account of its appearance illustrated in
Figure 5.7 The Poisson step process, and derive the mean and correlation function for
that process. We leave it as an exercise to the reader to derive
X
(t) = t (5.44)
and
R
XX
(t
1
, t
2
) =
2
t
1
t
2
+ min(t
1
, t
2
), (5.45)
These results will be useful later in Section 5.8, wherein it is shown how the statistics
of a process, which is the output of a linear transformation applied to an input process of
which the statistics are known, may be derived from suitable transformations applied to
the statistics of the input process. We will in that section be re-deriving the statistics of
the impulse process by recognising it as the derivative of the step process, and recognising
also that dierentiation is an example of a linear transformation.
5.4.10 Random Walk
(a) Denition
Every T seconds we increment or decrement X(t) by an amount s according to the toss
of a coin, starting with X(0) = 0. This is the random walk process shown in Figure 5.12.
Note that, as stated, the step size is s. We are interested in the rst and second order
statistics of this process.
Figure 5.12: Random walk process.
(b) Analysis
If t is in interval n, dened as (n 1)T < t < nT, and there have been a total of n
increments and decrements. We dene the number of increments to be k, so the number
of decrements is n k. The value of X(t) in this interval, designated X
n
is given by
X
n
= (2k n)s = r s (5.46)
where r = 2k n can take the values n, n + 2, . . . , n 2, n, as the number of
increments is k = (r +n)/2, X
n
has the binomial distribution dened below and as shown
for a particular case in Figure 5.13.
P{X
n
= r s} =
_
n
k
_
_
1
2
_
n
(5.47)
where
_
n
k
_
=
n!
k! (n k)!
. (5.48)
Figure 5.13: Probabilities in a random walk.
It is easy to list the step sequences which lead to the results above. Doing so would
form a useful albeit simple, exercise.
(c) Alternative development
Taking another view, and making preparation for the study of the Wiener-Levy process
presented in the next section, we can express the total displacement after n steps as
X
n
= u
1
+u
2
+. . . +u
n
(5.49)
where the u
i
are independent random variables equal to s with equal probability.
The statistics of the u variables are
E{u
i
} = 0 (5.50)
E{u
2
i
} = s
2
(5.51)
from which we derive the statistics of the X variable as
E{X
n
} =
n
i=1
E{u
i
} = 0 (5.52)
E{X
2
n
} =
n
i=1
n
j=1
E{u
i
u
j
} = ns
2
. (5.53)
In the above equations, we have substituted for the X variables in terms of the u
variables, and have interchanged the order of summation and expectation. In the last
line, we have also used the independence of the dierent steps.
5.4.11 Wiener-Levy Process
(a) Denition
The Wiener-Levy process arises if we let T 0 and s 0, such that the variance of X
n
remains nite.
(b) Analysis
Let t = nT and let s
2
= T. Putting w(t) = X
n
, and letting T 0, we obtain w(0) = 0
and
E{w(t)} = 0 (5.54)
E{w
2
(t)} = ns
2
=
t
T
T = t. (5.55)
These results derive directly from the last results of the previous section. In addition,
as w(t) consists of the sum of a large number of independent variables u
i
, by the central
limit theorem, the probability density function of w(t) is Gaussian, so we may write
p(w; t) =
1
2t
e
w
2
/2t
(5.56)
wherein we have used the known values of mean and variance just derived.
[Note that we have used the same symbol for the random process as for the values it
takes. This is common practice and usually causes no confusion, but to be strictly correct
we should use W(t) for the random process and w for the values that it takes.]
(c) Autocorrelation
The autocorrelation function is
R
ww
(t
1
, t
2
) = E{w(t
1
)w(t
2
)} (5.57)
= E{w(t
1
)[w(t
2
) w(t
1
) +w(t
1
)]} (5.58)
= E{w(t
1
)[w(t
2
) w(t
1
)]} +E{w
2
(t
1
)} (5.59)
= t
1
for t
1
t
2
(5.60)
= min(t
1
, t
2
). (5.61)
In the above equations, the rst line provides the standard denition of the autocor-
relation function. The second line introduces a clever breakup of one of the factors, to
provide either independence or complete dependence in the multiplication to follow. The
third line performs that multiplication. The fourth line introduces the assumption t
1
< t
2
,
and then uses the fact that w(t
1
) and w(t
2
) w(t
1
) are independent for t
1
t
2
, and also
uses the known value for variance given in equation 5.55. The fth line rewrites the result
of the fourth line in a way that is consistent with the assumption that t
1
< t
2
, and is also
consistent with the parallel result which can be obtained by permuting the variables t
1
and t
2
, and is thus regarded as the general result.
If t
1
< t
2
and w(t
1
) is known, then we can write
w
2
= w
1
+w
d
(5.62)
where w
2
= w(t
2
) and w
d
= w(t
2
) w(t
1
). This may be regarded as saying that where
we are at t
2
is where we are at t
1
plus the further displacement w
d
. In this view, and
regarding w(t
1
) as known, and noting that w
d
has zero mean, the conditional mean of
w(t
2
) is w
1
and its conditional variance is that of w
d
, namely (t
2
t
1
).
5.5. SINE WAVE AND RANDOM NOISE 77
p(w
2
| w
1
) =
1
_
2(t
2
t
1
)
exp
_
(w
2
w
1
)
2
2(t
2
t
1
)
_
. (5.63)
5.4.12 The reection principle
We note that this conditional probability density function is symmetrical about w
1
. This
is a form of the reection principle for a Wiener-Levy process which states that:
If for some experimental outcome
1
, the process w(t,
1
) has the value w = w
1
at
time t = t
1
, then there exists another outcome
2
for which the realisation w(t,
2
) is
identical to w(t,
1
) for t t
1
, and is the mirror image about w = w
1
for t > t
1
.
Figure 5.14: The reection principle.
An interpretation of the diagram of Figure 5.14 is provided by the statement that
although we got here by a particular path, we are just as likely to go backward as forward.
Is this particularly interesting? Perhaps, not yet.
5.5 Sine wave and Random Noise
It can be shown that a narrow band noise signal n(t) can be expressed as
n(t) = x(t) cos
o
t y(t) sin
o
t (5.64)
where
o
is the angular frequency at the centre of the noise spectrum and x(t) and y(t)
are random signals which vary at a rate comparable to the bandwidth of the signal. The
negative sign has been chosen before the second term in equation 5.64 so that x(t) will
represent the in-phase component of n(t) and y(t) will represent the quadrature component
of n(t), in the usual senses of those terms. In the particular case when n(t) is Gaussian,
x(t) and y(t) are independent equal variance Gaussian processes of zero mean and joint
probability density function,
p(x, y) =
1
2
2
e
(x
2
+y
2
)/2
2
(5.65)
where
2
is the variance of n(t) and also of x(t) and y(t). If we add a sine wave to
this, then we have a signal
v(t) = a cos
o
t +n(t) (5.66)
= a cos
o
t +x(t) cos
o
t y(t) sin
o
t. (5.67)
The phasor diagram of v(t) is as shown below. We see that we can dene the envelope
and phase of v(t) from this phasor diagram. As the variables are related by
Figure 5.15: Phasor diagram of sine wave and random noise.
r cos = a +x (5.68)
r sin = y (5.69)
the joint probability distribution in terms of r and is
p(r, ) =
r
2
2
exp
_
(r cos a)
2
+ (r sin )
2
2
2
_
(5.70)
=
r
2
2
exp
_
(r
2
2ar cos +a
2
)
2
2
_
. (5.71)
In deriving the above results we have used the results of Section 4.5. The Jacobian of
the transformation is again r.
This joint density function can with some diculty be integrated with respect to either
or r to give the probability density functions of r and respectively. The results are
p(r) =
r
2
I
0
_
ar
2
_
exp
_
a
2
+r
2
2
2
_
(5.72)
p() =
1
2
exp
_
a
2
2
2
_
+
_
a cos
2
_
exp
_
a
2
sin
2
2
2
_
Q
_
a cos
_
(5.73)
where I
0
(x) is the modied Bessel function of order zero, and Q(x) is the Gaussian
probability integral
Q(x) =
1
2
_

x
e
t
2
/2
dt. (5.74)
5.6. CROSS-CORRELATION AND COVARIANCE 79
The distribution of r is called the Rician distribution and it becomes the Rayleigh
distribution for a = 0. It is the distribution often assumed for the envelope of a received
signal which consists of a dominant component plus a number of multipath components
of random amplitude and phase.
Figure 5.16: Probability density functions of r and .
5.6 Cross-correlation and Covariance
5.6.1 Real processes
The cross correlation function of two processes x(t) and y(t) is dened as
R
xy
(t
1
, t
2
) = E{x(t
1
)y(t
2
)}. (5.75)
The two process are said to be orthogonal if R
xy
(t
1
, t
2
) = 0.
The covariance function of two processes x(t) and y(t) is dened as
C
xy
(t
1
, t
2
) = E{[x(t
1
)
x
(t
1
)][y(t
2
)
y
(t
2
)]} (5.76)
= R
xy
(t
1
, t
2
)
x
(t
1
)
y
(t
2
). (5.77)
This is just the cross correlation function of the zero mean processes derived from x(t)
and y(t).
5.6.2 Complex processes
If the processes x(t) and y(t) are complex, then the second term in the above denitions
is replaced by its complex conjugate, i.e.
R
xy
(t
1
, t
2
) = E{x(t
1
) y
(t
2
)}. (5.78)
The denitions of this section should be compared with those of Section 5.3.
5.7 Stationary Processes
5.7.1 Denition: Stationary in the strict sense
A random process x(t) is stationary in the strict sense if its statistics are invariant
under time translation. This means that
p(x
1
, . . . , x
n
; t
1
, . . . , t
n
) = p(x
1
, . . . , x
n
; t
1
+ , . . . , t
n
+ ) (5.79)
for all values of the variables and time shift .
5.7.2 First order statistics
For a random process which is stationary in the strict sense, the rst order probability
density function will be time invariant. Thus
p(x; t) = p(x; 0) = p(x) (5.80)
In the second equation above, we have simplied the notation, as there is no need in
the rst order probability density function for a time variable.
5.7.3 Second order statistics
For a random process which is stationary in the strict sense, the second order probability
density function must be a function only of = t
1
t
2
. The result is
p(x
1
, x
2
; t
1
, t
2
) = p(x
1
, x
2
; t
1
t
2
, 0) = p(x
1
, x
2
; ). (5.81)
The above result is obtained by putting = t
2
in equation 5.79 for the particular case
of a rst order probability density function, and in the second of the equations above
introducing simplied notation which recognises the fact that the result depends only on
the time dierence = t
2
t
1
. The autocorrelation function, the denition of which
involves the second order probability density function, is therefore also a function of
alone. The result is
R
xx
(t
1
, t
2
) = R
xx
(t
1
t
2
, 0) (5.82)
= R
xx
() (5.83)
The above result may be obtained by starting with the denition of the autocorrelation
function in equation 5.7, making use of equation 5.79 for the particular case of a second
order probability density function, putting = t
2
in the right hand side thereof, and in
the second of the equations above, again introducing simplied notation which recognises
the fact that the result depends only on the time dierence = t
1
t
2
.
In terms of our expectation notation, we have
R
xx
(t
1
, t
2
) = E{X(t
1
)X(t
2
)} (5.84)
from which we note there is a symmetry property about R
xx
so that R
xx
(t
1
, t
2
) =
R
xx
(t
2
, t
1
), so that
5.8. TRANSFORMATION OF RANDOM PROCESSES 81
R
xx
(t
1
, t
2
) = R
xx
(t
2
t
1
, 0) (5.85)
= R
xx
(). (5.86)
Thus R
xx
() = R
xx
(), i.e. the autocorrelation function of a strict sense stationary
process is an even function.
5.7.4 Denition: Stationary in the wide sense
A process is stationary in the wide sense if
E{x(t)} = = constant (5.87)
E{x(t)x(t ))} = R() ie. independent of t (5.88)
As expected from the terminology, we see from the denitions that a process which is
stationary in the strict sense is also stationary in the wide sense.
5.8 Transformation of Random Processes
5.8.1 Denition
Given a random process, when we can assign to each of its realisations x(t) a new
function y(t) according to some transformation rule
y(t) = T[x(t)] (5.89)
the process y(t) is called a transformation of x(t). The transformation may be de-
terministic or random, but here we will only consider deterministic transformations. In
principle, the statistics of y(t) can be found from those of x(t) for any transformation.
In practice, however, the computations involved may become very complicated except for
the two cases considered below.
5.8.2 Memoryless Transformations
Here y = y(x), i.e. the instantaneous value of x(t) determines the instantaneous value
of y(t) according to some functional dependence. For example, if y = x
2
then
p
y
(y; t) =

i
p
x
(x
i
; t)
dx
dy
(5.90)
=
1
2
y
[p
x
(
y; t) +p
x
(
_
(y); t)], (5.91)
p
y
1
y
2
(y
1
, y
2
; t
1
, t
2
) = p
x
1
x
2
(x
1
, x
2
; t
1
, t
2
)
(x
1
, x
2
)
(y
1
, y
2
)
(5.92)
=

all signs
p
x
1
x
2
(
y
1
,
y
2
; t
1
, t
2
)
4
y
1
y
2
. (5.93)
The autocorrelation function is
R
yy
(t
1
, t
2
) =
_

x
2
1
x
2
2
p(x
1
, x
2
; t
1
, t
2
) dx
1
dx
2
. (5.94)
5.8.3 Linear Systems
(a) Concept
The concept of a linear system is illustrated in Figure 5.17 and in the denitions below.
Figure 5.17: Illustration of a linear system.
We write a linear transformation as
y(t) = L[x(t)] (5.95)
where L is linear if
L[a
1
x
1
(t) +a
2
x
2
(t)] = a
1
L[x
1
(t)] +a
2
L[x
2
(t)] (5.96)
for any a
1
, a
2
, x
1
(t), x
2
(t). The essential ideas are that if the input is scaled by a factor,
so is the output, and if the input is the sum of two separate time functions, the output is
the sum of the outputs which would result for these two inputs being applied separately.
It can be seen that a transformation of the form
y(t) =

i
h(t,
i
) x(
i
) (5.97)
is a linear transformation. In this transformation, an input x(
i
) applied at time
i
generates a time function which is the product of that input and the function h(t,
i
).
That time function can be considered to be the response to a unit input applied at that
time. Note that inputs applied at dierent times may generate dierent time functions
which may not be simply delayed versions of the time response to an input applied at
t = 0. Thus the system might not be time invariant. It does however have the property
that if an inputs at time
i
is scaled, so will be the output, and if several inputs are applied
at time
i
, the output will be the required sum of the separate outputs to those inputs
applied separately. All these properties are preserved if what is considered to be the input
are signals applied a number of dierent times
i
, and the summation in equation 5.97 is
operative. We may view the linearity we are dening here as akin to the linearity of the
distributive law of multiplication over addition, and deriving in part therefrom.
5.8. TRANSFORMATION OF RANDOM PROCESSES 83
(b) First order
The mean value (averaged over the ensemble, not over time) of the output is given by
E{y(t)} = E{L[x(t)]} (5.98)
= E{
i
h(t,
i
) x(
i
)} (5.99)
=

i
h(t,
i
) E{x(
i
)} (5.100)
= L[E{x(t)}]. (5.101)
Again we note that the mean value of the input is an ensemble mean, not an average
over time. What we have shown here is that the order of the L and E operators may be
interchanged.
(c) Second order
We investigate the properties of the various correlation functions below. For convenience,
we start with the cross correlation function R
xy
.
R
xy
(t
1
, t
2
) = E{x(t
1
) y(t
2
)} (5.102)
= E{x(t
1
)

i
h(t
2
,
i
) x(
i
)} (5.103)
=

i
h(t
2
,
i
) R
xx
(t
1
,
i
) (5.104)
= L
t
2
[R
xx
(t
1
, t
2
)] (5.105)
where L
t
2
means we operate with respect to t
2
, with t
1
being regarded as a constant.
In the rst line above, we have the denition of the cross correlation function. In the
second line we have substituted for y(t
2
) using the linear transformation. In the third
line, we have used the distributive law, and the denition of R
xx
for times x and i. In
the fourth line, we are using the denition of the transform of a function of t
2
, not t, and
recognise that change by adding a subscript to L, as has been explained at the head of
this paragraph.
For the other cross correlation function
R
yx
(t
1
, t
2
) = E{y(t
1
) x(t
2
)} (5.106)
= L
t
1
[R
xx
(t
1
, t
2
)]. (5.107)
The steps of this derivation are not shown, but are similar to those above. Finally, for
the autocorrelation function of the output
R
yy
(t
1
, t
2
) = E{y(t
1
) y(t
2
)} (5.108)
= E{L
t
1
[x(t
1
) y(t
2
)]} (5.109)
= L
t
1
[E{x(t
1
) y(t
2
)}] (5.110)
= L
t
1
[R
xy
(t
1
, t
2
)] (5.111)
= L
t
1
L
t
2
[R
xx
(t
1
, t
2
)]. (5.112)
In the rst line above, we have the denition of the autocorrelation function. In the
second line we have substituted for y(t
1
) using the linear transformation. In the third
line, we have used the interchangeability of L and E, which rests on the understanding
that L
t
1
means we operate with respect to t
1
, with t
2
being regarded as constant. In the
fourth line, we use the denition of R
xy
(t
1
, t
2
), and in the fth line we use the result in
equation 5.107 above.
5.8.4 Example: Dierentiation operator
If our operator is that of dierentiation, L = d/dt so y(t) = dx(t)/dt = x(t), we nd that
the rst and second order statistics of the output satisfy, as particular examples of the
formulae of the last section
x
(t) =
x
(t) (5.113)
R
x x
(t
1
, t
2
) =
R
xx
(t
1
, t
2
)
t
2
(5.114)
R
xx
(t
1
, t
2
) =
R
xx
(t
1
, t
2
)
t
1
(5.115)
R
x x
(t
1
, t
2
) =

2
R
xx
(t
1
, t
2
)
t
1
t
2
. (5.116)
5.8.5 Further example: Poisson impulse process
As a further example, we apply the results just derived to the Poisson step process, which
becomes on dierentiation the Poisson impulse process, to obtain the rst and second
order statistics of that process. Applying the formulae of the last section to the Poisson
step process X(t) for which
X
(t) = t (5.117)
and
R
XX
(t
1
, t
2
) =
2
t
1
t
2
+ min(t
1
, t
2
), (5.118)
we obtain the mean and autocorrelation of I(t) =

X(t) as
I
(t) = (5.119)
R
II
(t
1
, t
2
) =

t
1
[
2
t
1
+ u(t
1
t
2
)] (5.120)
=
2
+ (t
1
t
2
). (5.121)
In the second of the equations above, we have already performed the required dier-
entiation with respect to t
2
of the expression in equation 5.118, having noted that the
5.9. ERGODICITY 85
function min(t
1
, t
2
) is t
2
when t
2
< t
1
and becomes t
1
when t
2
> t
1
. Thus the derivative
of this function with respect to t
2
is 1 for t
2
< t
1
, and drops to zero for t
2
> t
1
. This is
just the behavior of the step function u(t
1
t
2
).
The values in the three equations above agree with our previous results, derived in
Section 5.4.8. [Note that (t) is an even function].
5.8.6 Simplied results for stationary processes
For a stationary process x(t) with mean and autocorrelation R
xx
() the results of Section
5.8.4 can be restated in the simplied form
x
= 0 (5.122)
(5.123)
R
x x
() =
dR
xx
()
d
(5.124)
R
xx
() =
dR
xx
()
d
(5.125)
R
x x
() =
d
2
R
xx
()
d
2
. (5.126)
These results are obtained when we put = t
1
t
2
, simplify the notation because
only the time dierence is involved, and employ, for the same reason, total derivatives
instead of partial derivatives.
5.9 Ergodicity
5.9.1 Overview
Up to the present, we have calculated all of our statistics, including time dependent ones,
as ensemble averages, not time averages. However many systems have the property that
each single realisation of the system exhibits, over time, the full range of variation of
behavior which is found between the dierent realisations in an ensemble.
A random process is said to be ergodic if all its statistics can be determined from a
single realisation x(t,
i
) of the process. For a single realisation, we will use a time average
to determine the statistics. We will examine how this is done below.
5.9.2 Time averages
The time averaging operation is dened as
x(t) = lim
T
1
T
_
T/2
T/2
x(t) dt. (5.127)
It can be shown that the limit exists if the random variable represented by the aver-
aging for nite T has a variance which tends to zero as T .
Clearly since the time average eliminates the variable t, a process can only be ergodic
if it is stationary, but not all stationary processes are ergodic.
Similarly the autocorrelation function of an ergodic process can be determined as
R
xx
() = x(t) x(t ) (5.128)
5.9.3 Time and ensemble statistics
We can form the two sets of statistics discussed below.
(a) Time Statistics
x
(
i
) = x(t,
i
) (5.129)
= lim
T
(1/T)
_
T/2
T/2
x(t,
i
) dt (5.130)
R
xx
(,
i
) = x(t,
i
) x(t ,
i
). (5.131)
In general, these statistics will depend on the particular realisation, but are indepen-
dent of time, apart from the dependence of R
xx
on the time dierence . If the statistics
do not depend on the particular realisation, the process is called regular. An illustration
of both regular and not regular, stationary and not stationary, processes is provided in
Figure 5.18.
(b) Ensemble Statistics
x
(t) = E{x(t)} =
_

xp
x
(x; t) dx (5.132)
R
xx
(t, t ) = E{x(t)x(t )} (5.133)
=
_

x
1
x
2
p
x
1
x
2
(x
1
, x
2
; t, t ) dx
1
dx
2
(5.134)
where x
1
= x(t) and x
2
= x(t ).
In general, the ensemble statistics may be functions of time, but if not,( apart from
the dependence of R
xx
on the time dierence ),the process is said to be stationary. An
illustration of both stationary and not stationary, regular and not regular, processes is
provided in Figure 5.18.
5.9.4 Denition of an ergodic system
If a process is both regular and stationary, it is called ergodic. Most of the processes we
will consider will be ergodic, or nearly so. As stated above, the advantage of having an
ergodic process is that the statistics can be calculated by either time or ensemble averages,
whichever is the most convenient.
5.9. ERGODICITY 87
Figure 5.18: Examples of random processes.
5.9.5 Challenges
The question of whether a random process is ergodic or not is usually dicult to answer
rigorously. On must of course begin with a mathematical model of the process. But to
see the sorts of further diculties involved, we consider rstly the mean of a process.
Let us dene a random variable
v =
1
T
_
T/2
T/2
x(t) dt (5.135)
This has a mean value equal to
x
, the mean of x(t).
E{v} =
v
(5.136)
=
1
T
_
T/2
T/2
E{x(t)} dt (5.137)
= E{x(t)} =
x
(5.138)
If T then we have
lim
T
v = x(t) (5.139)
This limit approaches a constant value if the variance of v tends to zero as T .
The variance of v is
2
v
= E{(v )
2
} (5.140)
=
1
T
2
_
T/2
T/2
_
T/2
T/2
E{[x(t
1
)
x
][x(t
2
)
x
]} dt
1
dt
2
(5.141)
=
1
T
2
_
T/2
T/2
_
T/2
T/2
C
xx
(t
1
, t
2
) dt
1
dt
2
(5.142)
=
1
T
_
T
T
_
1
_
C
xx
() d (5.143)
where C
xx
() is the autocovariance function of x(t).
If
2
v
tends to zero as T then the process is ergodic in the mean. In general, if
C
xx
() 0 as , which will be true in most cases, then
2
v
0.
To rigorously establish the ergodicity of the autocorrelation function requires consid-
eration of the fourth order statistics of x(t).
Introduced the idea of a random process and its dependence on time.
Dened the rst order probability density distribution at time t for a random
process.
dened the second order probability density distribution at times t
1
and t
2
for a
random process.
Dened the mean, variance, autocorrelation function and autocovariance function
for a random process.
Considered a range of Poisson processes including the Poisson point process, the
Poisson interval process, the Poisosn pulse process and the Poisson impulse prpcess,
deriving for the latter the autocorrelation function.
Analysed the random walk and found its mean and variance.
Dened the Wiener-Levy process, and derived its probability density distribution
for time t, and its autocorrelation function for times t
1
and t
2
.
Considered the statisitcs of a sine wave plus random noise, and derived the probably
density distribution of the magnitude and phase.
Dened the cross correlation and cross covariance functions for a pair of random
processes.
Dened the concept of a stationary process in the strict and wide senses.
Establised that for a stationary process the autocorrelation function depends only
on the time dierence = t
1
t
2
.
Introduced the idea of a linear system, and derived for such a system formulae for
the cross correlation functions between input and output, and the autocorrelation
function of the output, all in terms of the autocorrelation function of the input and
the properties of the transformation.
Specialised those formulae to the case when the transformation is that of dieren-
tiation.
Applied those formulae the case of the Poisson step process, to provde an alternative
derivation of the autocorrrelation function of the Poisson impulse process.
Dened the concept of a regular system, reviwed the concepts of a stationary system,
and dened the concept of an ergodic system.
Shown how for an ergodic systems time averges can be used instead of ensemble
averages in the calculation of rst and second order statistics, i.e. the mean and
autocorrelation function.
Chapter 6
POWER SPECTRA OF
STATIONARY PROCESSES
6.1 Correlation Functions
6.1.1 Recall of denitions
For a stationary random process x(t), the auto and cross correlation functions R
xx
()
and R
xy
() are given by
R
xx
() = E{x(t) x(t )} (6.1)
R
xy
() = E{x(t) y(t )} (6.2)
We will mainly be concerned with real time functions, and in what follows this will be
assumed, unless specically stated otherwise.
6.1.2 Properties of correlation functions
.
1. The central peak of the autocorrelation function give the total power of the wave-
form.
R
xx
(0) = E{x
2
(t)} = P. (6.3)
The P on the right hand side may be thought of as the total power, if the X variable
is a voltage or current in a one resistor.
2. The correlation function is bounded in magnitude by that total power.
|R
xx
()| R
xx
(0). (6.4)
We will not provide a proof of this result here. Proofs may be found in the references,
and also in Appendix C. The result is illustrated in Figure 6.1.
91
92 CHAPTER 6. POWER SPECTRA OF STATIONARY PROCESSES
3. The correlation function is an even function.
R
xx
() = R
xx
() (6.5)
We have already derived this result in Section 5.7. This result is also illustrated in
Figure 6.1.
4. The cross correlation functions R
yx
() and R
xy
() exhibit reection symmetry.
R
yx
() = R
xy
() (6.6)
The proof proceeds from the denition of the cross correlation function in Section 5.6
and makes use of the stationary property in Section 5.7. This result is also illustrated
in Figure 6.1.
R ( )
t
t
t
XX
xy
R(0)
-R(0)
R ( )
yx
R ( )
t
t
Figure 6.1: Correlation functions.
5. R
xx
() and R
xy
() are functions only of the time dierence.
A consequence of the last property is that
E{x(t +a) y(t +b)} = R
xy
(a b) (6.7)
6.1.3 Example
For a sine wave x(t) = A cos(
o
t +), where we assume is a random variable uniformly
distributed over to ,
R
xx
() = E{A
2
cos(
o
t + ) cos(
o
t
o
+ )} (6.8)
=
A
2
2
E{cos(
o
) + cos(2
o
t
o
+ 2)} (6.9)
=
A
2
2
cos(
o
). (6.10)
This result is illustrated in Figure 6.2
6.2. THE FOURIER TRANSFORM 93
R ( )
xx
t
A/2
2
t
Figure 6.2: Autocorrelation function of a sine wave.
6.2 The Fourier Transform
The Fourier transform is the basic connection between the time domain and the fre-
quency domain for signals which exist for < t < . The Laplace transform is of use
only in dealing with events for t 0.
Here we will use f in Hertz as the frequency variable rather than , since this has
many advantages. Also frequency f is associated with the complex exponential e
j2ft
,
and f may be positive or negative. All real signals contain both positive and negative
frequencies.
A non-periodic signal cannot be represented as a Fourier series for all t. However, it
can be represented by use of the Fourier transform, which can be thought of as the limit
of a periodic signal as the period tends to innity.
The Fourier transform X(f) of a time signal x(t) is dened as
X(f) =
_

x(t) e
j2ft
dt. (6.11)
The Fourier transform X(f) is often referred to as the spectrum of x(t).
The reverse transform is
x(t) =
_

X(f) e
j2ft
df. (6.12)
The forward and reverse transforms are the same operation, except that one has a
positive exponential and the other has a negative exponential. This leads to a great deal
of symmetry in the relations. The only place 2 appears is in the exponent. These are
good things to remember.
6.2.1 Example: Rectangular pulse
Consider a rectangular pulse as shown in Figure 6.3. The rectangular pulse can be de-
scribed in terms of a standard function rect(t) which is a rectangular function of unit
height for
1
2
< t <
1
2
and zero elsewhere. Since the pulse considered has a height A and
a width , the standard rect function must be scaled in both amplitude and time.
x(t) = Arect(t/) (6.13)
X(f) =
_
/2
/2
Ae
j2ft
dt (6.14)
= A
_
e
j2ft
j2f
_
/2
/2
(6.15)
= A
e
jf
e
jf
j2f
(6.16)
= A
sin(f)
f
(6.17)
= A sinc(f). (6.18)
The Fourier transform is shown below.
Figure 6.3: Rectangular pulse and its Fourier transform.
Here we have encountered another new function sinc(x) which is dened as
sinc(x) =
sin(x)
x
(6.19)
The sinc function is illustrated in Figure 6.4.
The sinc(f) function arises whenever rectangular pulses of width occur, so will be
encountered whenever digital pulse signals using rectangular pulses are considered.
Because of the symmetry in the Fourier transform, we nd that a sinc pulse in the
time domain has a Fourier transform which is a rectangular function of frequency.
x(t) = Asinc(t/T) X(f) = AT rect(fT). (6.20)
This means that a sinc pulse in the time domain has its frequency components uni-
formly distributed over the frequency range
1
2T
< f <
1
2T
. It also has the property that
it is the minimum bandwidth pulse which goes through zero at uniformly spaced instants
(except for t = 0).
Figure 6.4: The sinc function.
6.2.2 Example: Exponential pulse
The exponential pulse has the form x(t) = u(t) e
at
as shown in Figure 6.5. Its Fourier
transform is
X(f) =
_

0
e
at
e
j2ft
dt =
1
a +j2f
. (6.21)
and is also illustrated in Figure 6.5.
Figure 6.5: Fourier transform of exponential pulse.
6.2.3 Example: Triangular pulse
The triangular pulse has the form x(t) = A(t/T) as shown in Figure 6.6. The function
(x) is another new function, the triangle function. The basic triangle function (x) is
of unit height, extends from 1 to +1, and has unit area. The Fourier transform of the
triangular pulse is
X(f) =
_
T
T
A
_
1
|t|
T
_
e
j2ft
dt (6.22)
= 2A
_
T
0
_
1
t
T
_
cos(2ft) dt (6.23)
= AT sinc
2
(fT). (6.24)
Details of the integration by parts have been omitted. This example provides an
illustration of the equivalence of multiplication in the frequency domain to convolution in
the time domain. The convolution process will be discussed in Section 6.4.
Figure 6.6: Fourier transform of triangular pulse.
6.2.4 Hermitian property
For a real signal, the Fourier transform is hermitian i.e. X(f) = X
(f), so the magnitude

|X(f)| is even and the phase spectrum arg X(f) is odd.
6.2.5 The Rayleigh energy theorem
Fourier transforms exist for signals of nite energy E. We may express the energy E as
E =
_

x
2
(t) dt (6.25)
=
_

x(t)
__

X(f) e
j2ft
df
_
dt (6.26)
=
_

X(f)
__

x(t) e
j2ft
dt
_
df (6.27)
=
_

X(f) X
(f) df (6.28)
=
_

|X(f)|
2
df (6.29)
This is Rayleighs Energy Theorem, but is often called Parsevals theorem. Since
|X(f)|
2
integrated over all frequencies gives the total energy, then |X(f)|
2
is called the
energy density spectrum. It is real, even and non-negative.
6.2.6 Properties of Fourier transforms
There is a sheet of Fourier transforms provided in Appendix B of these notes. You
should note the various theorems, relations and special functions which are contained on
this sheet. In practical applications of the Fourier transform, the relations on this sheet
should be adequate to perform any of the Fourier transforms required (i.e. you should
not have to calculate any integrals).
Some useful properties of the Fourier transform are shown below. You must understand
and be able to apply them. Some of the more important relations are also shown below.
Operation Time domain Frequency domain
Time scale x(t/T) |T| X(fT)
Frequency scale |f
o
| x(f
o
t) X(f/f
o
)
Time delay x(t T) X(f) e
j2fT
Frequency shift x(t) e
j2fot
X(f f
o
)
Inversion x(t) X(f)
Dierentiation dx(t)/dt j2f X(f)
Integration
_
t
x() d X(f)/j2f + (1/2)X(0) (f)

Multiply by t t x(t) (1/j2) dX(f)/df
To these properties should be added the Hermitian property
X(f) = X
(f) (6.30)
for Fourier transforms of real time functions stated in Section 6.2.4. Other useful
relations are
x(0) =
_

X(f) df = area of X(f) (6.31)

X(0) =
_

x(t) dt = area of x(t) (6.32)

_

x
2
(t) dt =
_

|X(f)|
2
df. (6.33)
The rst two lines above are particular, and interesting, cases of the direct and reverse
transforms. The last result restates the Rayleigh energy theorem.
6.2.7 Example: Double exponential pulse
We provide an example of use of the theorems by deriving the Fourier transform of a
double exponential pulse x(t) = e
a|t|
as shown in Figure 6.7.
x(t) = u(t) e
at
+u(t) e
at
(6.34)
(6.35)
X(f) =
1
a +j
+
1
a j
=
2a
a
2
+
2
. (6.36)
Figure 6.7: Fourier transform of double exponential pulse.
6.2.8 Another example
We provide another example of use of the theorems by solving the problem below.
Find y(t) if Y (f) = X
(f) e
j2ft
o
x(t) X(f) (6.37)
x
(t) X
(f) (6.38)
x
(t +t
o
) X
(f) e
j2ft
o
(6.39)
y(t) = x
(t
o
t) (6.40)
= x(t
o
t) if x(t) is real. (6.41)
In the above equations, the rst line denes notation, the second line is obtained by
conjugating the direct transform, the third line by recognising the eect of a time delay
t
0
, the fourth line by recognising that we have already obtained on the right hand side
the answer, and the fth line depends on the condition shown.
6.3 Singularity Functions
The main ones of interest are the delta function and its derivatives. They arise when
we form the Fourier transforms of signals with innite energy (such as periodic signals).
WE seek to evaluate the Fourier transform of the constant function
x(t) = 1 X(f) =
_

e
j2ft
dt. (6.42)
This integral does not converge in the ordinary sense. However if we consider the
rectangular function x
T
(t) = rect(t/T), which has nite energy, we have already seen it
has the Fourier transform X
T
(f) = T sinc(fT) as shown in Figure 6.8. It is hopefully
clear from the basic properties of the Fourier transform, that for the given time function,
which is equal to one at t = 0, the area under the Fourier transform is unity, as shown in
that gure.
As T then x
T
(t) x(t) for all nite t, and X
T
(f) tends to a large narrow pulse
of unit area. This is the delta function (f). The delta function is dened by the relation
_

(x) (x) dx = (0) (6.43)

6.4. CONVOLUTION 99
Figure 6.8: Fourier transform of a constant by limiting process.
where (x) is any function continuous at t = 0. The delta function is actually a
distribution (or generalised function), which is an entity which can only be dened by
an integral. For our purposes it is sucient to regard the delta function as the limit of
a large narrow pulse of unit area, and to recognise it as the derivative of the unit step
function. Thus we have the direct and reverse transforms
_

e
j2ft
dt = (f) (6.44)
x(t) =
_

(f) e
j2ft
df = 1 (6.45)
6.3.1 Example: Application to periodic signals
Let x(t) be periodic of period T. Dening f
o
= 1/T gives
x(t) =

n=
X
n
e
j2nf
o
t
(6.46)
(6.47)
X
n
=
1
T
_
T/2
T/2
x(t) e
j2nf
o
t
dt (6.48)
(6.49)
X(f) =

n=
X
n
(f nf
o
) (6.50)
where f
o
= 1/T. (6.51)
In the above equations, the top two lines are hopefully familiar expressions for a
Fourier series and its coecients. The third line is obtained by performing the Fourier
transform on the top line, and interchanging the order of integration and summation. An
illustration of the right hand side is provided in Figure 6.9.
6.4 Convolution
If two frequency functions are multiplied together, what is the corresponding operation
in the time domain? We pursue this question below.
Figure 6.9: Fourier transform of a periodic signal.
x(t) X(f) and y(t) Y (f) (6.52)
z(t) X(f) Y (f) (6.53)
z(t) =
_

X(f) Y (f) e
j2ft
df (6.54)
=
_

__

x() e
j2f
d
_
Y (f) e
j2ft
df (6.55)
=
_

x()
__

Y (f) e
j2f(t)
df
_
d (6.56)
=
_

x() y(t ) d = x(t) y(t) (6.57)

In the above equations, the rst line establishes notation, and the second line denes
z(t). The third line expresses z(t) as the reverse transform of the product XY . The
fourth line is obtained by substituting for X(f), and the fth line by interchanging the
order of integration with respect to frequency and time, with as our time variable, and
re-arranging factors. The sixth line recognises the quantity in square brackets above it as
a reverse transform with variable t , and also introduces the symbol for convolution.
As noted, the operation is called convolution. The order of the terms may be
interchanged.
Similarly we have convolution in the frequency domain
x(t) y(t) X(f) Y (f) =
_

X() Y (f ) d. (6.58)
The convolution operation is illustrated in Figure 6.10.
6.5 Power Spectral Density
6.5.1 Denition
In Section 6.2.5 on the Rayleigh energy theorem, we noted that for a variable x(t) of which
X(f) is the Fourier transform, |X(f)|
2
could be interpreted as a power spectral density.
We now formally dene it to be so, and next prove an important relation between the
power spectral density and the autocorrelation function of the variable x(t).
6.5. POWER SPECTRAL DENSITY 101
Figure 6.10: Convolution of two time functions.
6.5.2 Wiener Khinchine theorem
The Wiener Khinchine theorem states that the power spectral density and the autocor-
relation function of a stationary random process are Fourier transform pairs.
We use the notation S
xx
(f) to denote the power spectral density, and already have
the notation R
xx
() to denote the autocorrelation function. Thus we have the Fourier
transform pairs
S
xx
(f) =
_

R
xx
() e
j2f
d (6.59)
R
xx
() =
_

S
xx
(f) e
j2f
df. (6.60)
Because there are some mathematical diculties involved in the denition of the power
sepctral density, many text books simply state the theorem without providing a proof,
and often take the Fourier transform of the autocorrelation function as a denition of
power spectral density.
What we have done in these notes is to provide a proof, but have relagated that proof
to Appendix C
6.5.3 Application of the theorem
For the Poisson Pulse Process y(t),
R
yy
() =
2
+

(/) (6.61)
so S
yy
(f) =
2
(f) + sinc
2
(f). (6.62)
These results are illustrated in Figure 6.11.
e
l/e
l
2
t
f
0
-2/e -1/e 1/e 2/e
R ( ) t
yy
l
l d( )
2
f S ( )
yy
f
-e
Figure 6.11: Autocorrelation and power spectrum of Poisson pulse process.
6.6 Cross Power Spectral Density
6.6.1 Denition
The cross power spectral density S
xy
(f) of two stationary random processes x(t) and y(t)
is dened as the Fourier transform of the cross correlation function R
xy
.
S
xy
(f) =
_

R
xy
() e
j2f
d (6.63)
6.6.2 Properties
The cross power spectral density is usually complex. Because R
xy
() = R
yx
()
S
yx
(f) = S
xy
(f). (6.64)
Deriving this result makes use of the inversion and Hermitian properties of Fourier
transforms.
6.6.3 Application
We will not encounter cross spectral densities very often, except to note that they arise
when we add signals.
v(t) = x(t) +y(t) (6.65)
(6.66)
R
vv
() = E{v(t) v(t )} (6.67)
= E{[x(t) +y(t)][x(t ) +y(t )]} (6.68)
= R
xx
() +R
xy
() +R
yx
() +R
yy
() (6.69)
S
vv
(f) = S
xx
(f) + 2 Re[S
xy
(f)] +S
yy
(f). (6.70)
We have used equation 6.64 in obtaining the last result.
6.7. LINEAR TIME INVARIANT SYSTEMS 103
6.7 Linear Time Invariant Systems
These systems may be described by an impulse response h(t) or its Fourier transform
the transfer function H(f).
Figure 6.12: Linear time invariant system.
The transfer function relates input and output in the frequency domain by
Y (f) = X(f) H(f). (6.71)
We have seen in Section 5.8 an input x(t) applied to a linear system (which is not
necessarily time invariant) produces an output y(t) given by
y(t) = L[x(t)] =
_

h(t, ) x() d (6.72)

The function h(t, ) is the response to an impulse applied at time . If the system is
time invariant, its impulse response (for an impulse applied at time t = 0) is known as
h(t). Because of the time invariance, the impulse response for an impulse applied at time
t = is h(t ). In this case h(t, ) = h(t ), and the linear operator L is the operation
of convolution.
y(t) = L[x(t)] =
_

x() h(t ) d = h(t) x(t) (6.73)

Applying the results of the previous chapter, and in particular that the operations L
ane E can be interchanged, we obtain
E{y(t)} = L[E{x(t)}] (6.74)
y
= h(t)
x
= H(0)
x
(6.75)
In the second of these results, we are to form the convolution of the impulse response
h(t) and a constant. Inspection of the convolution integral shows that the result is to
form the product of that constant and the integral over all time of the impulse response
h(t). By the properties of the Fourier transform, that integral is just the zero frequency
component H(0) of the transfer function H(f). Thus we have shown that since
x
is
constant, then so is
y
. For the cross correlation function we have
R
xy
(t
1
, t
2
) = L
t
2
[R
xx
(t
1
, t
2
)] (6.76)
=
_

h(t
2
, ) R
xx
(t
1
, ) d (6.77)
=
_

h(t
2
) R
xx
(t
1
) d (6.78)
R
xy
() =
_

h() R
xx
( ) d (6.79)
= h() R
xx
() (6.80)
where we have put t
2
= 0 and t
1
= without loss of generality since x(t) and y(t) are
both stationary. In the above algebra, the rst line is a result from Section 5.8.3 and is true
for all linear transformations and systems, the second line expands on the notation L
t
2
,
the third line makes use of the stationary property of R and h, fourth line uses t
1
t
2
=
and the property R
xy
(t
1
, t
2
) = R
xy
(t
1
t
2
), while the fth line can be derived by putting
= and x() = h() and comparing the result with the convolution integral derived
in Section 6.4.
Similarly
R
yx
() = h() R
xx
(). (6.81)
For the autocorrelation function
R
yy
(t
1
, t
2
) = L
t
1
[R
xy
(t
1
, t
2
)] (6.82)
=
_

h(t
1
, ) R
xy
(, t
2
) d (6.83)
=
_

h(t
1
) R
xy
( t
2
) d (6.84)
R
yy
() =
_

h( ) R
xy
() d (6.85)
= h() R
xy
(). (6.86)
In the above equations, the rst line is a result from Section 5.8.3 and provides a
good starting point. The second line expands on the notation L
t
1
, and selects t as t
1
and
regards t
2
as a constant. The third line makes use of the stationary property of R and
h. The fourth line puts t
2
= 0 and t
1
= , and makes use of the stationery property R
yy
,
while the fth line is obtained by recognising that we have a convolution in the line above.
Taking Fourier transforms of the several results above gives
S
xy
(f) = H
(f) S
xx
(f) (6.87)
S
yx
(f) = H(f) S
xx
(f) (6.88)
S
yy
(f) = |H(f)|
2
S
xx
(f). (6.89)
We see that all of the second order statistics of the output can be related, by aspects
of the transfer function, to the second order statistics of the input, i.e. to the input power
spectral density.
6.8. CYCLOSTATIONARY PROCESSES 105
6.7.1 Example
Suppose the lter is the ideal bandpass lter dened by
H(f) =
_
1 ; f
1
< f < f
2
0 ; elsewhere
(6.90)
Figure 6.13: Meaning of power spectral density.
The output power spectral density will be given by
S
yy
(f) = |H(f)|
2
S
xx
(f) (6.91)
R
yy
(0) =
_

S
yy
(f) df (6.92)
E{ |y(t)|
2
} =
_
f
2
f
1
S
xx
(f) df (6.93)
In the rst line above, we use a result from equation 6.89. The second line is just
the reverse Fourier transform, taken at the value t = 0. The third line comes from
equation 6.3, and also recognises that the shape of the lter function allows the limits of
integration on the right hand side to be contracted.
But R
yy
(0) = E{ |y(t)|
2
} is the power of the output signal y(t), so our view that S
xx
(f)
is a measure of the power density at various frequencies is conrmed.
Note that a real lter would have, because of the hermitian property of H(f) if h(t) is
real, a passband from f
2
to f
1
in addition to the passband shown. For the lter above,
h(t) and hence y(t) would be complex. The example is therefore somewhat articial.
6.8 Cyclostationary Processes
6.8.1 Denition
There are certain random processes which are not stationary, but have statistics which
are periodic. These processes are called cyclostationary, since we nd that if the statistics
are averaged over a time equal to the periodicity, then they are time invariant.
6.8.2 Example: sine wave
The simplest cyclostationary signal is a deterministic sine wave. Consider the following
signal
v(t) = A cos(
o
t + ) (6.94)
where A and are xed values. Clearly this signal is not stationary, since its statistics
are time dependent, as we see below.
E{v(t)} = A cos(
o
t + ) (6.95)
R
vv
(t
1
, t
2
) = A
2
cos(
o
t
1
+ ) cos(
o
t
2
+ ) (6.96)
=
A
2
2
{cos[
o
(t
1
t
2
)] + cos[
o
(t
1
+t
2
) + 2]}. (6.97)
However, if these statistics are averaged over one period of the sine wave, keeping
t
1
t
2
constant, we obtain
E{v(t)} = 0 (6.98)
R
vv
(t
1
, t
2
) =
A
2
2
cos[
o
(t
1
t
2
)]. (6.99)
This is the same result as would be obtained by considering to be a random variable
uniformly distributed over (, ). The power spectral density is
S
vv
(f) =
A
2
2
{(f f
o
) + (f +f
o
)}. (6.100)
This last result draws on the results for the Fourier transforms of periodic signals
obtained in Section 6.3.
Although we have provided here an example of a deterministic time function, there are
other examples of cyclostationary signals which are not deterministic. Such an example
is provided in the next section.
6.8.3 Example: random binary signal
Consider a random binary signal x(t). This is a signal which is constant over intervals
of length T, and takes the values of A with equal probability, and independently of the
values in other intervals.
The mean
x
= 0, so this is a constant. However the autocorrelation function is not
time invariant.
R
xx
(t
1
, t
2
) =
_
A
2
; t
1
, t
2
in same interval
0 ; t
1
, t
2
in dierent intervals
(6.101)
6.8. CYCLOSTATIONARY PROCESSES 107
Figure 6.14: Random binary waveform.
Figure 6.15: Autocorrelation function of random binary waveform.
The periodic structure of the autocorrelation function is obvious, so if we consider a
xed = t
1
t
2
, and average the result as t
2
spans an interval of length T, we obtain an
average autocorrelation function
R
xx
() = A
2
(/T). (6.102)
By taking the Fourier transform of R
xx
(), we get the power spectral density of x(t),
which is interpreted as the power density at various frequencies, averaged over a time T.
S
xx
(f) = A
2
T sinc
2
(fT) (6.103)
R ( )
-T 0 T
A
2
AT
2
0 1/T 2/T f -2/T -1/T
t
f
S ( )
t
xx
xx
Figure 6.16: Average autocorrelation and power spectrum.
6.8.4 Filtering property
An important property of cyclostationary processes is that when they are ltered by a
linear time invariant lter, the output is also cyclostationary. We will not prove this
result.
6.9 Chapter summary
Reviewed for stationary random processes the denitions of auto and cross correla-
tion functions.
Established some properties, e.g. bounds and reection, of the cross and auto
correlation functions.
Introduced the Fourier transform and derived some transforms of common wave-
forms.
Proven the Rayleigh Energy Theorem.
Introduced the delta function.
Applied the delta function to the derivation of the Fourier transform of a periodic
signal.
Established the convolution property of time waveforms whose Fourier transforms
have been multiplied.
Introduced the idea of power spectral density.
Proven the Wiener-Khinchine theorem for stationary random processes.
Dened the concept of cross power spectral density, and provided a simple illustra-
tion in the context of the sum of two random processes.
Introduced the concept of linear time invariant systems.
Derived for such systems formulae relating the cross correlation functions between
input and output, and the autocorrelation function of the output, to the autocorre-
lation function of the input.
Took Fourier transforms of those results to derive formulae relating the cross power
spectral density between input and output, and the power spectral density of the
output, to the power spectral density of the input.
Dened cyclostationary processes, and considered some of their properties.
Chapter 7
RANDOM NOISE SIGNALS
7.1 White Noise
7.1.1 Pure white noise
In most systems there are unwanted signals which we call noise. A random signal n(t)
which has S
nn
(f) = (where is a constant) is called white noise.
White noise has an autocorrelation function
R
nn
() = () (7.1)
These are shown in Figure 7.1.
S (f)
nn
f 0 0
t
a
R ( )
nn
t
Figure 7.1: Power spectrum and autocorrelation function of white noise.
Hence n(t
1
) and n(t
2
) are uncorrelated except for t
1
= t
2
. True white noise does not
exist in practice since it has innite bandwidth and innite power.
7.1.2 Bandlimited white noise
In practice we have bandlimited white noise, where the noise has a constant power spectral
density at all frequencies to which a particular system responds, but may fall to zero
outside this range as shown in Figure 7.2.
In calculating the noise response of systems, the input noise may be assumed to be
white noise, since the region where this is not true is where the system does not respond
anyway.
111
112 CHAPTER 7. RANDOM NOISE SIGNALS
Figure 7.2: Bandlimited white noise.
If the power spectral density of the noise is not constant over the range of frequencies
of interest, then the noise is called coloured.
The character of bandlimited white noise depends on its relative bandwidth. In Fig-
ure 7.3 is shown wideband white noise of bandwidth 2 MHz centred on 1 MHz and also
narrowband noise of bandwidth 100 kHz centred on 1 MHz.
Figure 7.3: White and narrowband noise.
7.2. THERMAL NOISE 113
7.2 Thermal Noise
7.2.1 General principles
Thermal noise arises from the thermal agitation of the charge carriers and from thermally
generated electromagnetic elds which may be present in an electromagnetic eld con-
tainment structure, inside of which all elements are in thermal equilibrium at an absolute
temperature T degrees Kelvin.
When such a structure has a single access port, such as is shown in Figure 7.4, it
may be shown that, for all frequencies up to the frequency for which quantum eects
become important, the available source power P
a
due to thermal noise in the structure
and available through the port and over a positive bandwidth B is given by
P
a
= kTB (7.2)
where k = Boltzmanns constant = 1.381 10
23
J/K. In this formula, we use in the
Bandwidth B only frequencies on the positive half of the axis.
Figure 7.4: Field containment structure with single port access.
Thus over the above dened frequency range, there is a uniform distribution of avail-
able noise power per unit bandwidth of amount kT.
This is a general property of thermal noise, and can be used in situations where object
sizes and frequencies of interest are such that the concepts of lumped circuit theory are
valid, and in also situations where they are not.
The frequency range for which the above result holds is given by hf << kT, where h =
6.626 10
34
Js is Plancks constant. This frequency corresponds to optical frequencies
at room temperature.
Thermal noise can be reduced by cooling, and this is often done where low noise
ampliers are required, such as in radio astronomy or space communications.
The concept of available source power, used in the above statements, means the power
which can be extracted from a network into an optimally matched load. It is unrelated to
anything which may be imagined to be happening inside the network.
7.2.2 Lumped circuit representations
If we are working in a context where lumped circuit concepts are valid, and the general
device considered above is an instantaneous voltage v(t) in series with a resistor R, or is a
device with a sinusoidal voltage represented by an r.m.s. phasor V in series with complex
impedance R+jX of which the real part is R, such as are shown in Figure 7.5, you should
show as an exercise that the available source powers are
P
a
=
|v|
2
4R
(7.3)
in the time domain case or
P
a
=
|V |
2
4R
(7.4)
in the frequency domain case.
Figure 7.5: Thermal noise equivalent circuits.
Thus when the internal source produces just thermal noise, and the source impedance
is just a resistance R, that source may be represented by a random variable v which has
a white spectrum given by
S
vv
(f) = 2kTR volt
2
/Hz (7.5)
where R is the source resistance in ohms. In this formula, to correctly calculate the
total noise, we need to take into account frequencies on both halves of the frequency axis.
For the more complicated impedance Z(f), it can be shown that
S
vv
(f) = 2kT Re{Z(f)} (7.6)
In addition to the Thevenin noise equivalent circuits shown in Figure 7.5, the corre-
sponding Norton equivalent circuits also exist.
7.3. SHOT NOISE 115
A common formula found in many text books for the noise voltage v
r.m.s.
back of
resistor R is
v
r.m.s.
=
4kTB (7.7)
This formula may be reconciled with that of equation 7.5 when it is recalled that in
equation 7.5 contributions from both halves of the frequency axis must be used, but in
equation 7.7, only positive frequencies are considered.
7.3 Shot Noise
7.3.1 Origin
This arises from the discrete nature of current ow in electronic devices. The eect
is not observed in conductors where the electrons all move in a correlated way when an
electric eld is applied. However, where the current ow consists of individual electrons
crossing a eld region, such as electrons emitted from the cathode in a vacuum tube, or
in the current ow in a bipolar transistor when holes or electrons cross the base depletion
region at random instants of time, shot noise is found in the electron current.
7.3.2 Simple model
Shot noise can be modelled as current pulses occurring at random instants of time. This
then corresponds to a Poisson impulse process with current pulses of area q = 1.60210
19
coulomb occurring at an average rate per second. Hence the autocorrelation function is
R
ii
() =
2
q
2
+ q
2
() (7.8)
= I
2
+qI () (7.9)
where I = q is the average DC current. The Fourier transform is
S
ii
(f) = I
2
(f) +qI (7.10)
The rst term corresponds to the d.c component and the qI term corresponds to shot
noise. The current pulses are of course not perfect impulses (due to the nite transit time
across the eld region) so the spectrum is not white at all frequencies.
7.3.3 More accurate model
If T is the transit time across the eld region, then it is found that the shot noise is white
up to frequencies of the order of 1/T. It usually occurs that this frequency corresponds
to the upper useful operating frequency of the device.
If the current pulse is p(t), then the shot noise power spectral density is, ignoring the
d.c. term
S
ii
(f) = |P(f)|
2
(7.11)
where P(f) is the Fourier transform of p(t).
7.3.4 Example
Suppose the current pulse p(t) = 2qt/T
2
for 0 t T. The Fourier transform of p(t) is
P(f) =
2q
(j2fT)
2
{1 (1 +j2fT)e
j2fT
} (7.12)
This gives the power spectrum shown in Figure 7.6. Unity on the vertical axis corre-
sponds to qI, and on the horizontal axis to fT.
Figure 7.6: Power spectrum of shot noise.
7.4 Chapter Summary
Dened white noise and provided illustrations of bandlimited derivations thereof.
described the properties of thermal noise, and derived formulae for the available
source power from thermal sources.
Dened shot noise, and used the statistics of the Poisson impulse processes to derive
its autocorrelation function and power spectral density.
Derived the power spectral density of a modied form of shot noise.
Chapter 8
ANALOG FILTER DESIGN
8.1 Introduction
8.1.1 Uses of lters
Filters are used to modify the frequency spectrum of signals or to reject unwanted
signals. The most common application which you see in practice is the selection of radio
or TV channels. This is done by lters which pass only the signals which are in a specied
location in the frequency spectrum.
In other applications, lters may be used to shape the response in the time domain.
For instance, some electronic devices (such as operational ampliers or thyristors) will
only work properly if the slope of the input signal is less than a specied value. This
leads to problems if a signal with a step change is introduced, so a lter could be used to
limit the slope of the signal to an acceptable value.
8.1.2 Transfer function
The lter circuits we will consider will have a transfer function H(s) obtained by stan-
dard circuit analysis techniques using either the complex frequency concept or Laplace
transforms. The transfer function will usually be a voltage transfer function expressing
the ratio of the output voltage V
o
and the input voltage V
i
, as shown in Figure 8.1.
Figure 8.1: Context for denition of transfer function.
The transfer function is
H(s) =
V
o
(s)
V
i
(s)
(8.1)
117
118 CHAPTER 8. ANALOG FILTER DESIGN
where V
i
(s) and V
o
(s) are the generalised phasor or Laplace domain representations of
the corresponding time domain quantities. The output is sometimes called the response
and the input the excitation. Expressing lter characteristics in terms of H(s) is referred
to as the transform domain or frequency domain representation.
Notice that in this denition the source voltage V
s
does not appear.
8.1.3 Impulse response
In the time domain, the lter characteristic can be expressed in terms of its impulse
response h(t), which is the response when the input is a delta function (t). For causal
lters (i.e. lters which are physically realisable), h(t) = 0 for t < 0 and its Laplace
transform H(s) will have poles only on the left half s plane. The relation between the
impulse response and the frequency response is
H(s) =
_

0
h(t) e
st
dt (8.2)
Given an input signal v
i
(t), the output of the lter can be found either from the
transform domain as
V
o
(s) = V
i
(s) H(s) (8.3)
or in the time domain using convolution as
v
o
(t) =
_

h() v
i
(t ) d (8.4)
8.1.4 Frequency response
Of particular interest is the frequency response which is given by H(j). It is known that
if the input signal is a sine wave of angular frequency radians/sec, then the steady state
response is also a sine wave of the same frequency.
v
i
(t) = A cos(t + ) (8.5)
v
o
(t) = A|H(j)| cos(t + + arg H(j)) (8.6)
The magnitude response |H(j)| will often be what interests us most, but in many
applications the phase response arg H(j) may also be of interest. It should be noted that
if H(s) is a voltage transfer function, then |H(j)| is dimensionless and may be expressed
in dB terms as 10 log
10
|H(j)|
2
dB (decibels) and the phase response is usually expressed
in degrees (but sometimes in radians). |H(j)| is also called the amplitude response.
You should know how to obtain H(j) from the pole-zero plot in the s plane, as this
can be used both as a computational technique and for gaining insight into how certain
lter responses are obtained.
A causal lter must have phase shift. In fact the phase response can be found from
the magnitude response, so the two are not independent.
8.1. INTRODUCTION 119
8.1.5 Finding the transfer function
Given the magnitude response |H(j)|
2
, and some knowledge of the positions of the poles
and zeros which could create that magnitude response, it is possible to nd the poles and
zeros of the transfer function H(s) by the following method.
We recall rst that H(s) is a quotient of polynomials, with real coecients, in s, i.e.
H(s) =
P(s)
Q(s)
(8.7)
The poles and zeros of H(s) are mirrored in the horizontal axis.
In each of those polynomials, when s is changed in sign, the even powers of s will
remain unchanged, and the odd powers will change in sign. Thus when s = j, change
of sign of s will leave the even powers, which make up the real part of the polynomial
unchanged, but will change the signs of the odd powers, which make up the imaginary
part of the polynomials. The result is that each polynomial P(s) or Q(s) becomes its
complex conjugate when s = j is changed in sign. Thus when s = j
H(s) H(s) = H(j) H
(j) = |H(j)|
2
(8.8)
Now |H(j)|
2
is an even function of (and hence also of s), so its poles and zeros are
now seen to be symmetrical about both axes in the s plane as shown in Figure 8.2.
Figure 8.2: Poles and zeros of H(s)H(s) in the s plane.
We know that a realisable lter must have its poles on the left half s plane, so the
poles of |H(j)|
2
which lie there must belong to H(s) and those on the right half s plane
belong to H(s). There is no restriction on the zeros, except that if a zero s
o
is assigned
to H(s) then s
o
must be assigned to H(s).
8.1.6 Minimum phase lters
Choosing all the zeros on the left half s plane as belonging to H(s) gives a minimum
phase lter.
8.1.7 Example
Consider a lter with poles and zeros as shown in Figure 8.3.
|H(j)|
2
=
2
2
(4 +
2
)
1 +
4
=
2s
2
(4 s
2
)
1 +s
4
(8.9)
This has zeros at 0, 0, 2, 2 rad/s and poles at 0.7071 j 0.7071 rad/s.
Figure 8.3: Poles and zeros of example lter.
In our rst choice of poles and zeros of H(s), we put the poles of H(s) at s = 0.7071
j0.7071 rad/s, but now select zeros at s = 0 and s = 2 rad/s, to obtain
H(s) =
2 s(2 s)
s
2
+
2 s + 1
(8.10)
In our second choice of poles and zeros of H(s), we put the poles of H(s) at s =
0.7071 j0.7071 rad/s, and the zeros at s = 0 and s = 2 rad/s, to obtain
H(s) =
2 s(s + 2)
s
2
+
2 s + 1
(8.11)
The two possible choices are shown in Figure 8.4. The rst is minimum phase, the
second is not.
8.2 Classication of Filter Responses
It is common to classify lters into one of the groups low pass, high pass, band pass and
band stop, for which idealised transfer functions are illustrated in Figure 8.5. In these
diagrams for the low pass and high pass lters a cut o frequency f
c
can be recognised,
8.3. IMPEDANCE AND FREQUENCY SCALING 121
Figure 8.4: Zero selection for minimum and non-minimum phase lters.
and for the band pass and band stop lters a centre frequency f
o
and a bandwidth BW
can be recognised.
In practice neither the at tops nor vertical steps of these lters are physically real-
isable, so we have illustrated for the low pass lter the more realistic response shown in
Figure 8.6.
This gure shows that the concept of cut o frequency f
c
can be expanded to the
two concepts of pass band edge frequency f
b
and stop band edge frequency f
s
, and new
parameters passband ripple
p
and stop band attenuation
s
have been introduced. The
latter two parameters are normally expressed in dB, using the d.c. response level as
reference, and detaching the negative sign.
In addition to the responses illustrated, there are also the all pass lter responses,
which have in the ideal case an amplitude response which is independent of frequency,
but which may be used to provide some frequency-independent attenuation, or may be
used to provide, in the transfer function, some desired frequency-dependent phase shift.
8.3 Impedance and Frequency Scaling
8.3.1 Introduction
In designing and analysing circuits, to avoid having to consider very large or very
small numbers, it is convenient to have element values and angular frequencies of the or-
der of unity. Many standard circuit and lter designs are based on an impedance level of
1 ohm and an angular frequency of 1 rad/s. These are often referred to as normalised cir-
cuits. However practical circuits usually involve element values, impedances and angular
frequencies which are much dierent from unity.
By impedance and frequency scaling, we can extend the application of normalised
circuits to more useful impedance and frequency ranges.
8.3.2 Impedance scaling
Consider a circuit with a voltage transfer function, such as shown in Figure 8.7. Its
voltage transfer function is
Figure 8.5: Denition of various ideal lter responses.
Figure 8.6: Denition of parameters of realistic low pass lter response.
H(s) =
V
o
(s)
V
i
(s)
. (8.12)
Figure 8.7: Example circuit.
If the impedance of every element is multiplied by the same constant k
z
, then we have
resistors are multiplied by k
z
;
inductors are multiplied by k
z
;
capacitors are divided by k
z
;
voltages are not aected;
currents are divided by k
z
.
In particular, the voltage transfer function H(s), which is a dimensionless function of
various impedance ratios, is not aected.
8.3.3 Example
Consider the circuit shown in Figure 8.8. It has a transfer function
H(s) =
V
o
(s)
V (s)
=
1
s + 1
. (8.13)
Figure 8.8: Normalised circuit.
If we impedance scale this by 1000, then the transfer function is unaected, but the
circuit becomes as shown in Figure 8.9.
Figure 8.9: Impedance scaled circuit.
8.3.4 Frequency scaling
By the process of frequency scaling we can, by scaling element values, change a low pass
response covering one frequency range to a low pass response covering another frequency
range. By more complex element changes we can change a lter with a low pass response
to one with a high pass response. With even more complex element changes, in which a
single reactive element becomes a pair of series-connected or parallel-connected reactive
elements, we can change a network with a low pass response to one with a band pass or
band stop response.
These processes are known as frequency scaling because in each case there is a func-
tional relation between a point on the frequency axis of the new lter and a point on the
frequency axis of the original lter (at the same value of the transfer function).
When a low pass lter is frequency scaled to produce another low pass lter, the
relation between elements and frequencies is one of simple proportionality. In the other
cases a non linear functional relation results.
All of the frequency scaling methods we use are however based on the general principle
that the changed reactive elements have at the scaled frequency the same impedance as
the original elements had at the original frequency.
We will study in this chapter only the simple low pass to low pass frequency scaling,
and leave the more complex cases to Chapter 10.
If we multiply, for a low pass lter, the frequency scale by a factor k
f
(i.e. the response
at an angular frequency of rad/s now becomes the response at an angular frequency of
k
f
rad/sec), then we have
resistances are unchanged;
inductances and capacitances are divided by k
f
;
s in transfer functions is replaced by s/k
f
;
pole and zero positions are multiplied by k
f
;
time scales are divided by k
f
;
t in time domain expressions is replaced by k
f
t.
You should note that in accord with the above, frequency scaling by a factor k
f
also
scales time responses by a factor 1/k
f
.
Note also that in the transfer function, s or are divided by k
f
, not multiplied. This
is so that, when s or are k
f
times larger, the transfer function remains the same.
For our previous example, the frequency response is
H(j) =
1
1 +j
. (8.14)
In the time domain, the step response is
v
o
(t) = u(t) (1 e
t
). (8.15)
These responses are shown in Figure 8.10.
Figure 8.10: Response of original circuit.
If we scale the corner frequency (originally an angular frequency of 1 rad/s) to 1 kHz,
the scaling factor is k
f
= 21000. The frequency response is now
H(j) =
1
1 +j/6283
. (8.16)
The step response is now scaled by 1/k
f
, so 1 sec becomes 1/(21000) = 159 s. The
step response is
v
o
(t) = u(t) (1 e
6283t
). (8.17)
8.3.5 Special note
The impulse response does not scale in the same way. The amplitude of the impulse
response of the scaled lter is multiplied by the frequency scaling factor k
f
. This is
because while a step function is the same when scaled in time, u(k
f
t) = u(t), a delta
function scales according to k
f
(k
f
t) = (t). Hence the original impulse response is
u(t) e
t
whereas the frequency scaled circuit has an impulse response 6283 u(t) e
6283t
.
This result can also be derived from the properties of the Laplace transform. It is
known that the Fourier transform of h(t) is H(s), and the frequency scaling theorem for
Laplace transforms gives the relation that k h(kt) transforms to H(s/k).
The new frequency and step responses are shown in Figure 8.11.
Figure 8.11: Response of frequency scaled circuit.
Finally after both impedance and frequency scaling, the circuit becomes as shown in
Figure 8.12.
8.4 The Approximation Problem
An ideal low pass lter would have a response as shown in Figure 8.13.
In practice we cannot achieve this response exactly, but we can approximate it as
closely as we wish. Circuits built from lumped elements have transfer functions which
Figure 8.12: Impedance and frequency scaled circuit.
are rational functions of s, so we approximate the response H(s) by a rational function
with a denominator degree n. The larger the value of n, the closer the approximation. In
practice this means that the lter complexity (number of elements or building blocks) is
proportional to n, so we do not wish to have n any larger than necessary.
Figure 8.13: Ideal low pass lter response.
Of course our specications are not dened in terms of s, but in terms of angular
frequency , so we need some means of converting a required frequency response into a
transfer function H(s) which we can realise in circuit form. We therefore set specications
which dene how accurately we require the approximation to be. Such specications are
shown in Figure 8.14.
We see that the important parameters are
p
= passband ripple = 10 log |H
1
|
2
/|H
2
|
2
e.g. 1 dB
s
= stopband attenuation = 10 log |H
1
|
2
/|H
3
|
2
e.g. 25 dB
p
= passband angular frequency
s
= stopband angular frequency.
8.5 Chapter Summary
Mentioned briey the uses for lters in engineering technology.
Clarifed denitions of lter respnses.
Figure 8.14: Specication of required response.
Described techniques for nding pole and zero conguratins from algebraic expres-
sions for lter responses.
Introduced the ideal low pass, high pass, band pass or band stop classications of
lter responses.
Explained how to perform impedance and frequency scaling of low pass lters.
Described how practial lter responses may be specied.
Chapter 9
LOW PASS PROTOTYPES
9.1 Introduction
9.1.1 General approach
Filter design is usually done by placing the poles and zeros of the transfer function
so that the required response specications are met. In this section we will consider the
responses of some standard low pass lters. Later we will try to obtain some insight
into how these lter structures are obtained.
We will mainly be concerned with the magnitude response |H(j)|, but in some cases
the phase response may also be of interest.
9.1.2 Signicance of the phase response
If a lter has an amplitude response which is unity at all frequencies, and the phase
response is linear with frequency, the lter produces only a pure time delay equal to
the derivative of the phase with respect to angular frequency. After they have all been
delayed by a common amount of time, all frequency components maintain their original
relative phases with respect to one another. If a pulse is input to a practical lter with
a limited passband amplitude response but a liner phase response, the output pulse will
retain its general shape and properties, but sharp corners will be smoothed out due to
the attenuation of the high frequency components.
If the phase response is non-linear, then the relative phase of various frequency com-
ponents will not be preserved, and the pulse shape may be considerably modied.
9.1.3 Normalised low pass lters
We introduce here normalised low pass lters which are low pass lters expressed in terms
of a normalised angular frequency variable
=

n
(9.1)
where
n
is a normalising angular frequency which is chosen to be somewhere close
to the transition region between the pass band and stop band of the lter, the exact
positioning being discussed below.
129
130 CHAPTER 9. LOW PASS PROTOTYPES
The benet of expressing the transfer function in this way is that various numerical
constants which appear in the denition of the frequency response and which, because it
is not normal to insert units in the middle of a formula, appear to be dimensionless, are
in fact dimensionless.
Just where the normalising frequency is placed depends on the particular lter design,
among a range of standard designs, being discussed. Common choices are at the pass
band edge frequency f
p
, the stop band edge frequency f
s
, or the 3 dB down point in
the transition region. The guiding principle is that the transfer function in terms of the
normalised angular frequency variable, should aord a simple description. In fact the
lter denitions normally originate in terms of the normalised angular frequency variable
, but where the point = 1 stands in the transition region between the pass band and
stop band depends on the lter type.
Those types will be dened in more detail later, but we can state in advance that
for the Butterworth lters the point = 1 occurs at the pass band edge, and for the
most basic type of lter this edge is placed at the 3 dB down point of the response,
but is sometimes shifted so that a dierent value of response at the passband edge is
obtained. For the Chebychev type I and elliptic lters the point = 1 occurs at the
upper frequency extremity of a speciable pass band ripple, for the Chebychev type II
lters it occurs at the lowest frequency for a speciable stop band attenuation, while for
the Bessel-Thompson lters it occurs at a level which is too complex to state here.
The general approach to the design of low pass prototypes is to nd a function K()
such that
|H(j)|
2
=
1
1 +K
2
()
. (9.2)
The function K() should be such that |K()| << 1 for < 1 and |K()| >> 1 for
>> 1.
9.2 Getting the Element Values
For each of the normalised low pass lters the element values, when the cut o frequency
c
is at 1 rad/s, the element values, in a 1 ohm system, are all of the order of 1F or 1H,
and may be obtained from the transfer function using a variety of techniques outlined in
Chapter 10.
For practical applications those element values need to be scaled for impedance and
frequency as described in Chapter 8.
Mostly the scaling is simple and concerned simply with the ratio of a pass band edge
angular frequency in the scaled lter to a corresponding pass band edge of 1 rad/s in the
unscaled lter.
Sometimes, however, a slightly more complex scaling as outlined in the next section
is required.
9.3. BUTTERWORTH FILTERS 131
9.3 Butterworth Filters
9.3.1 Denition
A Butterworth lter of order n and a 3 dB passband edge has a normalised angular
frequency response
|H(j)|
2
=
1
1 +
2n
(9.3)
We must in the below analysis distinguish between the use of n as a subscript to
denote a normalised variable, and as a superscript to denote the order of a lter or the
raising of a variable to a power.
The equation above corresponds to having K() =
n
. Introducing the normalised
complex angular frequency variable s
n
= s/
n
and replacing by js
n
on the right hand
side and using equation 8.8 we have
H(s
n
)H(s
n
) =
1
1 + ()
n
s
2n
n
. (9.4)
The Butterworth lter is characterised by the single parameter n and has a magnitude
response which is said to be maximally at. This means that the rst 2n 1 derivatives
of |H(j)|
2
are zero at = 0. This is easily seen by expanding |H(j)|
2
as given in
equation 9.3 as a Taylor series
|H(j)|
2
= 1
2n
+
4n
+ (9.5)
At normalised angular frequency = 0 the magnitude response is 0 dB. In the pass-
band, the magnitude response provides a smooth approximation to |H(j)|
2
= 1 (i.e.
without any ripples in the response). At normalised angular frequency = 1 the magni-
tude response is 0.7071 or 3 dB, and for normalised angular frequency > 1 the response
falls o at 20n dB/decade. Hence the value of n required is determined by how rapidly
the response needs to fall in this region. The magnitude responses for Butterworth lters
with a 3 dB attenuation at the passband edge are shown in Figure 9.1.
9.3.2 Achieving a lesser passband attenuation
For a lter of given order n, if it is unsatisfactory that the attenuation at the edge of the
pass band is 3dB, we need to nd rst the value of
p
corresponding to the desired pass
band attenuation
p
. In that calculation we employ equation 9.3. We also calculate, via
techniques outlined in Chapter 10, the element values for a lter with cut o frequency
1 rad/s in a 1 system. That lter will have the desired pass band attenuation at an
angular frequency of
p
=
p
rad/s.
To achieve the same pass band edge attenuation at an angular frequency of 1 rad/s in
the scaled lter, the scaling factor k
f
is given by
k
f
=
1 rad/s
p
rad/s
(9.6)
Normalised angular frequency W
Figure 9.1: Amplitude responses of Butterworth lters.
For instance,
p
= 1 dB and n = 3 requires that
p
= 0.79835 on the normalised
frequency scale and produces, via the techniques outlined in Chapter 10, a lter with
p
= 1 dB at a frequency of
p
= rad/s = 0.79835 rad/s.
To make the lter a low pass prototype with a passband angular frequency
p
of 1
rad/s then requires that the original lter be frequency scaled by a factor of
k
f
=
1 rad/s
0.79835 rad/s
= 1.2526. (9.7)
The pole locations are then multiplied by k
f
, and inductances and capacitances are
divided by k
f
.
In the scaling process, the frequency response, originally given by
|H(j)|
2
=
1
1 +
2n
(9.8)
becomes modied, with in transfer functions being replaced by /k
f
, so that the
above equation becomes, after frequency scaling,
|H(j)|
2
=
1
1 +k
2n
f

2n
(9.9)
If we continue to use a normalising angular frequency of
n
of 1 rad/s to dene our
normalised frequency variable , the transfer function of the frequency scaled lter is
given in terms of the normalised frequency variable as
|H(j)|
2
=
1
1 +k
2n
f

2n
(9.10)
In the particular example being illustrated here, which has k
f
= 1.2526, this becomes
|H(j)|
2
=
1
1 + 0.25893
2n
(9.11)
This change of transfer function under this form of frequency scaling, in which the
normalising frequency does not change, is to be contrasted with the dierent behaviour
of the transfer function under the frequency scaling described in the next section.
9.3.3 Further frequency scaling
Any further frequency scaling, for example for the purpose of achieving a signicantly
dierent passband edge, or of converting for a low pass lter to a band pass lter, will
leave the normalised transfer function unchanged, provided that
for a low pass lter the normalising frequency is changed at the same time as the
element values are changed, or
for a band pass lter the nonlinear transformation between the normalised frequency
of the low pass response and the operating frequency band pass response is employed.
9.3.4 Achieving a required stopband attenuation
To achieve with the standard lter an attenuation of at least
s
at some normalised
angular frequency
s
requires that the order n satisfy
10 log(1 +
2n
s
) >
s
. (9.12)
9.3.5 Phase characteristic
The phase characteristic tends to be somewhat non-linear in the vicinity of = 1 rad,
and this results in pulse distortion. The phase responses of Butterworth lters are shown
in Figure 9.2.
Figure 9.2: Phase responses of Butterworth lters.
The pulse response of a lter of order n = 3 with a 1 dB attenuation at = 1 and a
normalising frequency of f
n
to a rectangular pulse of duration T = 1/(2f
n
) is shown in
Figure 9.3. The asymmetric response is due to the non-linear phase characteristic.
9.3.6 Pole-zero conguration
There are no zeros, and the 2n poles of H(s
n
)H(s
n
) are uniformly distributed around a
circle of unit radius in the s plane, as shown for n = 5 in Figure 9.4. The Butterworth
lter is an example of an all pole lter. The pole positions are given by
Figure 9.3: Pulse response of an n = 3 Butterworth lters.
s
nk
= e
j
k
where
k
=

2
+ (2k 1)

2n
(9.13)
i.e. s
nk
= sin
_
(2k 1)
2n
_
+j cos
_
(2k 1)
2n
_
k = 1, . . . 2n. (9.14)
All the poles of H(s
n
)H(s
n
) are generated by putting k = 1, 2, . . . , 2n. The poles on
the left half s
n
plane are those of H(s
n
). There are poles on the real axis only for odd
order lters.
Figure 9.4: Poles of Butterworth lter for n = 5.
9.4 Chebychev Type I Filters
9.4.1 Denition
Chebychev type I lters provide an equiripple response in the region 0 < < 1 and a
monotonically decreasing response for > 1.
The normalised frequency response of the Chebychev I lter of order n is
|H(j)|
2
=
1
1 +
2
T
2
n
()
(9.15)
where T
n
() is the Chebychev polynomial of degree n, of which the rst few are
T
o
() = 1 (9.16)
T
1
() = (9.17)
T
2
() = 2
2
1 (9.18)
T
3
() = 4
3
3 (9.19)
T
4
() = 8
4
8
2
+ 1 (9.20)
T
5
() = 16
5
20
3
+ 5 (9.21)
9.4. CHEBYCHEV TYPE I FILTERS 137
Note that all have the value T
n
() = 1 at = 1. A recurrence relation dening the
remainder of the sequence is
T
n
() = 2T
n1
() T
n2
() (9.22)
This relation conrms that all the polynomials retain the property that T
n
() = 1 at
= 1. The Chebychev polynomials are also given by the following relations.
For < 1,
T
n
() = cos n (9.23)
where = cos (9.24)
i.e. T
n
() = cos(narccos ). (9.25)
For > 1 rad,
T
n
() = cosh n (9.26)
where = cosh (9.27)
i.e. T
n
() = cosh(n arcosh ) (9.28)
9.4.2 Properties
Chebychev I lters have K() = T
n
() and are characterised by two parameters n and
. Whereas with the Butterworth lter the desired passband attenuation can be achieved
simply by frequency scaling, in the Chebychev lter the parameter sets the passband
attenuation.
The Chebychev polynomials oscillate between 1 for 0 < < 1, are equal to 1 at
= 1, and increase monotonically for > 1. Hence the magnitude response |H(j)|
oscillates between 1 and A for 0 < < 1, where
A =
1
1 +
2
. (9.29)
Hence in the passband the response is equiripple of amplitude
p
, where
p
= 10 log(1 +
2
) (9.30)
and the passband angular frequency is at = 1.
The required value of n is determined by the required attenuation at some normalised
stopband angular frequency
s
. The attenuation increases at higher frequencies.
s
< 10 log(1 +
2
cosh
2
(n)) (9.31)
where
s
= cosh (9.32)
The magnitude responses of Chebychev I lters are shown in Figure 9.5 for = 0.5,
which corresponds to
p
= 1.0 dB.
Figure 9.5: Magnitude responses of Chebychev I lters.
9.4. CHEBYCHEV TYPE I FILTERS 139
For the same value of n and allowable attenuation
p
in the passband, Chebychev I
lters have a higher attenuation for an angular frequency > 1 than Butterworth lters.
However the phase response is more non-linear, and this can be a problem if pulse shape
is to be maintained.
That the phase response of Chebychev I lters is less linear than for Butterworth lters
is shown in Figure 9.6.
Figure 9.6: Phase responses of Chebychev I lters.
The normalised pulse response of an n = 3 lter with 1 dB attenuation at = 1 is
shown in Figure 9.7. Again an input pulse duration of T = 1/(2f
n
) is used, and again
the time scale is expressed in units of T.
9.4.3 Pole positions
Chebychev I lters are further examples of all pole lters. Calculation of the pole positions
of H(s
n
)H(s
n
) can be done as follows. Substituting = cos with = a + jb in the
equation for pole positions gives
1 +
2
T
2
n
() = 0 (9.33)
1 +
2
cos
2
(n) = 0 (9.34)
cos(na +jnb) = j
1
(9.35)
Figure 9.7: Pulse response of an n = 3 Chebychev I lter.
9.5. CHEBYCHEV TYPE II FILTERS 141
cos(na) cosh(nb) j sin(na) sinh(nb) = j
1
(9.36)
cos(na) cosh(nb) = 0 (9.37)
sin(na) sinh(nb) =
1
. (9.38)
Hence
a =
(2k 1)
2n
; k an integer (9.39)
b =
1
n
sinh
1
(1/) (9.40)
s
n
= j cos() = j cos(a +jb) (9.41)
= sin(a) sinh(b) +j cos(a) cosh(b) (9.42)
= sin
_
(2k 1)
2n
_
sinh
_
1
n
sinh
1
(1/)
_
(9.43)
+j cos
_
(2k 1)
2n
_
cosh
_
1
n
sinh
1
(1/)
_
(9.44)
and all the poles are generated by putting k = 1, . . . , 2n.
These poles lie on ellipse in the s
n
plane as shown in Figure 9.8 for n = 5. Those of
H(s
n
) are in the left half plane. Again we note that there are poles on the real axis only
for odd order lters.
9.5 Chebychev Type II Filters
9.5.1 Denition
Type II Chebychev lters provide an equiripple response in the region > 1 and a
monotonically decreasing response for < 1. The normalised frequency response of the
Chebychev II lter of order n is
|H(j)|
2
=
1
1 +a
2
/T
2
n
(1/)
(9.45)
where T
n
(x) is the Chebychev polynomial of degree n as before and a is a parameter
which sets the stop band attenuation at angular frequency
s
= 1.
s
= 10 log(1 +a
2
). (9.46)
The attenuation at the normalised passband angular frequency
p
is calculated as
p
= 10 log[1 +a
2
/T
2
n
(
s
/
p
)]. (9.47)
Figure 9.8: Poles of an n = 5 Chebychev I lters.
9.6. ALL PASS FILTERS 143
9.5.2 Illustration
The Chebychev II lter above is normalised to a stopband frequency of = 1. The
magnitude responses for n = 1 to 5 and
s
= 20 dB (corresponding to a = 9.95) are
shown in Figure 9.9.
9.5.3 Design
Figure 9.9: Magnitude responses of Chebychev II lters.
Chebychev II lters have both poles and zeros, and the zeros lie on the imaginary axis.
The details of calculating the poles and zeros are not considered in this course. The design
of such lters is easily achieved using MATLAB.
9.6 All Pass Filters
9.6.1 Denition
An all pass lter is one for which |H(j)| = constant for all . All pass lters are
constructed by having poles on the left half s
n
plane accompanied by mirror image zeros
in the right half s
n
plane as shown in Figure 9.10.
Figure 9.10: Poles and Zeros of an all pass lter.
9.6.2 Purpose
The main purpose of an all pass lter is to introduce phase shift without changing the
magnitude response. By cascading an all pass lter with a Butterworth or Chebychev
lter, the overall phase response can be made much closer to linear, while the desirable
magnitude response features of those lters are retained.
9.6.3 Design
The problem of phase compensation is not very tractable and we will not deal with it
here. It is sucient to say that computer search techniques often yield a suitable solution
more rapidly than an analytical approach.
9.7 Bessel-Thompson Filters
9.7.1 Dening property
The Butterworth lters have a response which has a maximally at magnitude response.
This is achieved by having as many as possible derivatives of |H(j)|
2
at = 0 equal to
zero.
In contrast, the Bessel-Thompson lters achieve a phase response which is as linear as
possible by having a maximally at delay response in which as many as possible derivatives
of arg H(j) at = 0 are equal to zero.
9.7. BESSEL-THOMPSON FILTERS 145
9.7.2 Analysis
The algebraic analysis of the Bessel-Thompson lter is too tedious to include here, so has
been relegated to Appendix C, where it will not do us much harm. The principal results
are stated below.
9.7.3 Graphical expression of responses
The magnitude responses of Bessel-Thompson lters are shown in Figure 9.11. The re-
sponses are all for a normalised group delay of unity.
Figure 9.11: Frequency responses of Bessel-Thompson lters.
The phase responses are shown in Figure 9.12.
It can be seen that the phase responses are reasonably linear in the passband, but
the attenuation characteristics in the stopband are not as good as for Butterworth or
Chebychev lters.
The pulse response of an n = 3 lter with a normalised angular frequency 1 dB
bandwidth = 1 is shown in Figure 9.13 and it can be seen that this is substantially
better than for Butterworth or Chebychev lters.
Figure 9.12: Phase responses of Bessel-Thompson lters.
9.7. BESSEL-THOMPSON FILTERS 147
Figure 9.13: Pulse response of n = 3 Bessel-Thompson lter.
9.8 Elliptic Filters
9.8.1 Denition
These lters are also called Cauer-Chebychev lters. The transfer function H(s
n
) has
poles on the left half s
n
plane and zeros on the imaginary axis.
An elliptic lter uses a rational expression for K() and has an equiripple response
in both the passband and stopband. Of all the lters considered so far, for a given n the
elliptic lter has the fastest transition from the passband to the stopband, but it also has
the worst phase linearity.
The magnitude response of an elliptic lter is given by
|H(j)|
2
=
1
1 +
2
R
2
n
()
(9.48)
where
R
2n
() =
(
2
1
2
) (
2
n
2
)
(1
2
1
2
) (1
2
n
2
)
(9.49)
R
2n+1
() =
(
2
1
2
) (
2
n
2
)
(1
2
1
2
) (1
2
n
2
)
(9.50)
9.8.2 Properties
The function R() has the following properties
The zeros
1
, . . . ,
n
are all less than unity.
The poles are the reciprocals of the zeros and are all greater than unity.
R(1/) = 1/R().
The zeros are chosen so that R() is equiripple for 0 < <
p
where
p
< 1 rad.
i.e. |R()| a in this region. In the region >
s
where
s
= 1/
p
, we also have
an equiripple behavior |R()| 1/a.
The value of determines the tradeo between passband ripple
p
and stopband
attenuation
s
.
9.8.3 Design
The design of elliptic lters is rather complex, and involves the use of elliptic functions.
MATLAB provides a means of obtaining a low pass prototype with
p
= 1 rad/sec using
the command [z,p,k]=ellipap(n,
p
,
s
). This gives the poles and zeros and gain constant
of the transfer function in the form
H(s
n
) = k
(s
n
z
1
) (s
n
z
n
)
(s
n
p
1
) (s
n
p
n
)
. (9.51)
9.8. ELLIPTIC FILTERS 149
9.8.4 Illustration
The poles and zeros for
p
= 1,
p
= 1.0 dB and
s
= 20.0 dB are shown in the table
below.
Order n Gain k Zeros Poles
1 1.96523 1.96523
2 0.10001 j3.20564 0.51479 j0.94239
3 0.32057 j1.43993 0.16150 j1.00335
0.64358
4 0.099978 j2.03918 0.40026 j0.65087
j1.12433 0.05162 j1.00365
5 0.28302 j1.24945 0.16106 j0.89691
j1.03847 0.01658 j1.00146
0.57200
9.8.5 Magnitude response
The magnitude responses of elliptic lters are shown in Figure 9.14 for
p
= 1.0 dB and
s
= 20.0 dB. The very sharp cut o behavior is evident.
Figure 9.14: Magnitude responses of elliptic lters.
9.8.6 Phase response
The phase responses are shown in Figure 9.15. The transfer function has zeros on the
imaginary axis and the phase jumps by 180
o
at the frequencies corresponding to those
zeros.
Figure 9.15: Phase responses of elliptic lters.
9.8.7 Pulse response
The normalised pulse response of an n = 3 lter with a 1 dB bandwidth = 1 and
s
= 20 dB is shown in Figure 9.16 and it can be seen that this is the worst of all the
lters so far considered.
9.9 Chapter Summary
Introduced the idea of describing lter responses in terms of a normalised dimen-
sionless frequency variable.
Dened the class of Butterworth low pass lters.
Shown how for those lters the frequency scale and the elements can be adjusted to
achieve various attenuations at the edge of the passband.
Figure 9.16: Pulse responses of an n = 3 elliptic lter.
Illustrated the responses of those lters.
derived the pole and zero congurations of those lters.
Introduced the Chebychev Type I lters, and established their principal properties
including analytical description of the frequency response, the pole congurations,
and have provided a graphical illustration the responses in the time and frequency
domains.
Introduced the Chebychev Type II lters, and established their principal properties
domains.
Introduced the Chebychev type 1 lters, and established their principal properties
domains.
Made brief mention of a class of all pass lters and their uses.
Introduced briey the Bessel Thompson lters, and their uses, and have provided a
graphical illustration the responses in the time and frequency domains.
Introduced the Elliptic or Cauer-Chebychev lters, and established their principal
properties including analytical description of the frequency response, the pole and
zero congurations, and have provided a graphical illustration the responses in the
time and frequency domains.
Chapter 10
FILTER DESIGN AND
TRANSFORMATIONS
10.1 Introduction
10.1.1 General approach
Filters are usually designed as
low pass prototypes;
impedance level of 1 ohm;
cut o angular frequency of 1 rad/s.
The lters resulting are called prototype low pass lters. They can be impedance and
frequency scaled as required. We will also see later that they can be transformed into
high pass, band pass or band stop lters.
10.1.2 Notation for unit angular frequency
The use of dimensioned constants, such as 1 rad/s, in equations makes the checking that
each side of an equation has the same units dicult. As such checking is important, we
will introduce the identier
u
for the unit angular frequency of 1 rad/s, and will use it
wherever possible in algebraic equations.
10.2 Low Pass Filter Design
10.2.1 Design steps
The design is accomplished through the following steps
Establish the specications.
Transform the frequency scale to derive the specications of an equivalent nor-
malised lter.
153
154 CHAPTER 10. FILTER DESIGN AND TRANSFORMATIONS
Design the normalised lter.
Rescale the elements to the frequency range desired.
Further rescale the elements to the impedance level desired.
Each of these steps is further illustrated in the sections which follow.
10.2.2 Frequency transformation of specications
We rst establish the lter specications as actually required. The lter specication
is then converted to a low pass prototype lter, which is done by scaling the desired cuto
angular frequency
c
to the normalised value
c
given by
c
=

c
n
(10.1)
Most often, we dene the cut-o angular frequency to be the passband edge
n
and
n
is chosen so that the passband edge has a normalised frequency
p
of 1, i.e. we set
n
=
p
. However, in the case of a Butterworth lter, if the passband edge is specied at
a value of
p
which is not 3 dB, we rst determine the value of
p
corresponding to that
value of
p
, and then choose
n
so as to obtain that value of
p
. This leads to
p
=

p
n
i.e.
n
=

p
p
(10.2)
We can also write
s
n
=
s
n
(10.3)
where s
n
= j is the normalised complex angular frequency and s = j is the actual
complex angular frequency.
An illustration is provided in Figure 10.1. The normalised low pass specications are
also shown. These have the same magnitude characteristics, but the frequency scale is
normalised to a passband angular frequency near 1 rad/s.
The standard designs derived in Chapter 11 all have a cut-o frequency
c
of 1 rad/s,
i.e.
c
=
u
, but our desired lter may have a cut-o frequency of
c
=
p
/
p
. Thus the
frequency scaling ratio required is
k
f
=

p
u
(10.4)
10.2.3 Design of normalised lter
This will be considered later, in Chapter 11.
10.2. LOW PASS FILTER DESIGN 155
Figure 10.1: Specication of the required response.
10.2.4 Scaling elements for frequency
R
n
R
n
(10.5)
L
n

L
n
k
f
(10.6)
C
n

C
n
k
f
(10.7)
s = k
f
s
n
(10.8)
We can use the last expression to transform any poles and zeros.
10.2.5 Scaling elements for frequency and impedance
R
n
R = k
z
R
n
(10.9)
L
n
L =
k
z
L
n
k
f
(10.10)
C
n
C =
C
n
k
z
k
f
(10.11)
10.2.6 Example
An example of the lter design is shown in Figure 10.2.
Figure 10.2: Filter design.
10.3 The High Pass Transformation
10.3.1 Structure of low pass lter
A low pass lter usually consists of aladder type structure with inductors in the
series arms and capacitors in the shunt arms as shown in Figure 10.3.
Figure 10.3: Typical low pass lter.
At low frequencies the inductors are low impedance and the capacitors are high im-
pedance so most of the current injected at the left hand end reaches the output at the
right hand end. As the frequency increases however, the inductors oer an increasing im-
pedance to the ow of current, while the capacitors tend to shunt the current to ground
as their impedance falls.
10.3.2 Structure of a high pass lter
A high pass lter can be made by interchanging the positions of the inductors and the
capacitors as shown in Figure 10.4.
10.3.3 Illustration
A high pass lter has a transfer function as shown in Figure 10.5, and an illustration of
idealised low pass and high pass lter responses is provided in Figure 10.6
10.3. THE HIGH PASS TRANSFORMATION 157
Figure 10.4: Typical high pass lter.
Figure 10.5: High pass response specications.
Figure 10.6: Specications of required response.
10.3.4 Transformations
A transformation of both the elements and frequency scale which makes, at the corre-
sponding frequencies for the two lters, corresponding reactances have the same values
(and may be applied to any low pass lter to produce a corresponding high pass lter) is
L
h
=
1
2
x
C
l
(10.12)
C
h
=
1
2
x
L
l
(10.13)
h
=
2
x
l
(10.14)
In the above,
x
is the crossover frequency, i.e. the frequency at which the transformed
frequency response intersects the original low pass frequency response.
Normally, we will make
x
=
p
, so the two lters will have the same passband edge
frequency.
The change of sign of the frequency variable will have the eect of producing, for
the transformed lter, the response for negative frequencies, but as this response is the
complex conjugate of the response for positive frequencies, and we are designing for a
magnitude response, this fact will have no consequence.
The resulting response is as shown in Figure 10.5.
When the corresponding low pass lter has been designed to have elements R
n
, L
n
and C
n
, the corresponding elements for the high pass lter, scaled for impedance and
frequency, are given by
R
n
R = k
z
R
n
= k
z
R
n
(10.15)
C
n
L =
k
z
k
f
2
x
C
n
=
k
z
k
f
2
p
C
n
(10.16)
L
n
C =
k
f
2
x
k
z
L
n
=
k
f
2
p
k
z
L
n
(10.17)
where in the second equation in each line we have put
p
=
x
.
again we note that if the passband is dened for a value of
p
other than unity, the
frequency scaling ratio required is given by
k
f
=

p
u
(10.18)
10.3.5 Design steps
The design steps are as outlined above. The nal design is shown in Figure 10.7.
10.3.6 Example
A TV receiver suers from interference from a nearby 27 MHz CB transmitter. To reduce
the interference, a lter which attenuates the 27 MHz signal by at least 15 dB is required,
while not attenuating the TV frequencies from 45 MHz upwards by not more than 1 dB.
Figure 10.7: Filter design.
The lter can be inserted in series with the antenna cable as shown in Figure 10.8.
Figure 10.8: Coaxial line lter.
We dene the eect of the lter by its insertion gain H
i
(s) dened as
H
i
(s) =
V
o
(s) with lter
V
o
(s) without lter
(10.19)
=
V
o
(s)
V (s)/2
=
2V
o
(s)
V (s)
(10.20)
This denition applies when the source and load impedances are equal, which is usually
the case. (For unequal source and termination impedances, the design should proceed from
a consideration of scattering coecients.)
1. Specications These are shown in Figure 10.9.
Figure 10.9: Specication of required response.
The passband edge
p
of 2 45 MHz of the desired lter becomes the passband
edge = 1 in the normalised low pass lter. As frequency ratios are inverted in the
transformation the stop band edge in the normalised low pass lter is at = 45/27
as shown in Figure 10.9.
The frequency scaling factor is
k
f
=

p
u
=
2 45 10
6
1 1
= 2 45 10
6
(10.21)
2. Design normalised lter
A suitable lter design is shown in Figure 10.10. It is a Chebyshev I lter of order 3
with 1 dB passband ripple. The method for nding the element values has not yet
been explained. It will be covered in Chapter 11
Figure 10.10: Normalised low pass lter.
H
i
(s
n
) =
1
2s
3
n
+ 2s
2
n
+ 2.5s
n
+ 1
(10.22)
|H
i
(j)|
2
=
1
(1 2
2
)
2
+ (2.5 2
3
)
2
(10.23)
=
1
4
6
6
4
+ 2.25
2
+ 1
(10.24)
We now check that we have achieved the specied level of attenuation.
= 0 : |H
i
|
2
= 1 10 log(1) = 0.0 dB
= 1 : |H
i
|
2
= 1/1.25 10 log(1/1.25) = 1.0 dB
= 45/27 : |H
i
|
2
= 1/46.69 10 log(1/46.69) = 16.7 dB
3. Transform to high pass and frequency and impedance scale
The equations for transforming to a high pass and impedance and frequency scaling
are
R = k
z
R
n
= k
z
R
n
(10.25)
L =
k
z
k
f
2
x
C
n
=
k
z
k
f
2
p
C
n
(10.26)
C =
k
f
2
x
k
z
L
n
=
k
f
2
p
k
z
L
n
(10.27)
With k
f
= 2 45 10
6
and
p
= 2 45 10
6
rad/s and K
z
= 75 these lead to
R = 75 (10.28)
C = 1/(445 10
6
75) = 23.6 pF (10.29)
L = 75/(245 10
6
) = 265 nH (10.30)
The nal design is shown in Figure 10.11.
Figure 10.11: The required lter.
10.4 The Band Pass Transformation
10.4.1 Structure of a band pass lter
To make a band pass lter, the series elements must be low impedance at some fre-
quency
o
and the shunt elements high impedance at
o
. Hence we use a series L-C in
the series branches and a parallel L-C in the shunt branches as shown in Figure 10.12.
Figure 10.12: Typical band pass lter.
A band pass lter is characterised by its centre frequency f
0
and bandwidth BW.
At an angular frequency
0
which we call band centre, the pairs of series elements
may be made be resonant so as to provided zero series impedance, and the pairs of shunt
elements may also be made resonant so as to provide zero shunt admittance. This situation
will correspond to zero frequency in a low pass lter.
At frequencies above the band centre frequency in the band pass lter, the net im-
pedance of the pairs of series elements will then become inductive, and the net admittance
of the pairs of shunt elements will become capacitive. This behaviour parallels the be-
haviour of the low pass lter at positive frequencies.
10.4.2 Transformation
It is easy to show that if we replace a single inductor L
n
in a normalised low pass lter
by a combination of a scaled inductor
L =

u
L
n
(10.31)
in series with its resonating capacitor, and we replace a single capacitor C
n
in a
normalised low pass lter by a combination of a scaled capacitor
C =

u
C
n
(10.32)
in shunt with its resonating inductor, and employ the frequency transformation
=

0
0

0
_
(10.33)
10.4. THE BAND PASS TRANSFORMATION 163
then we achieve a situation in which, at corresponding frequencies, the impedances
of the series and shunt branches of the band pass lter are equal to the impedances of
corresponding series and shunt branches of the normalised low pass ler.
The element transformations are as shown in Figure 10.13.
Figure 10.13: Band pass transformation of elements.
There are two values of for every value of as shown in Figure 10.14.
Figure 10.14: Band pass transformation.
Notice also that
=
p2
p1
= angular bandwidth (10.34)
o
=

p1
p2
= centre angular frequency (10.35)
Note from the above that thecentre frequency is the geometric mean of the cut
o frequencies, not the arithmetic mean. The resulting band pass lter will also have
geometric symmetry about
o
. This means that the magnitude response at a frequency
will be the same as that at a frequency
2
o
/. Points c and h correspond to = 0.
Points d and i correspond to = 1. Points g and b correspond to = 1.
10.4.3 Design equations
These have already been given as
L
n
L =

u
L
n
(10.36)
in series with its resonating (at band centre) capacitor, and
C
n
C =

u
C
n
(10.37)
in shunt with its resonating (at band centre) inductor.
If we wish to nd the frequencies of the actual lter which correspond to particular
values of normalised frequency of the normalised lter, the inverse of frequency transform
is
f =
(f)
2

_
2
(f)
2
4
+f
2
o
(10.38)
10.4.4 Example
Transform the low pass lter in our previous example to a band pass lter from 80 kHz
to 100 kHz and impedance 50 ohm.
f
o
=

80 100 = 89.44 kHz (10.39)
f = 100 80 = 20.00 kHz (10.40)
2 H L = 2 50/(220 10
3
) = 795.8 uH + resonating C (10.41)
1 F C = 1/(220 10
3
50) = 159.2 nF + resonating L (10.42)
Resonating C = 1/[(289.44 10
3
)
2
795.8 10
6
] = 3.979 nF (10.43)
Resonating L = 1/[(289.44 10
3
)
2
159.2 10
9
] = 19.89 uH (10.44)
This gives the lter shown in Figure 10.15.
Insertion gain at 85 kHz and 120 kHz
1
=
89.44
20.00
_
85
89.44

89.44
85
_
= 0.4556 (10.45)
2
=
89.44
20.00
_
120
89.44

89.44
120
_
= 2.667 (10.46)
10.4. THE BAND PASS TRANSFORMATION 165
Figure 10.15: Band pass lter.
|H
i
(j
1
)|
2
= 1/1.244 = 0.9 dB at 85 kHz (10.47)
|H
i
(j
2
)|
2
= 1/1153 = 30.6 dB at 120 kHz (10.48)
At what frequencies is the insertion loss 10 dB?
|H
i
(j)|
2
= 10
1
= 0.1 (10.49)
4
6
6
4
+ 2.25
2
+ 1 = 10 (10.50)
2
= 1.852 (10.51)
= 1.361 (10.52)
Now the bandpass transformation of frequencies is
=

o
o

o
_
(10.53)
=
f
o
f
_
f
f
o

f
o
f
_
(10.54)
0 = f
2
(f) f f
2
o
(10.55)
f =
(f)
2

_
2
(f)
2
4
+f
2
o
(10.56)
For this example we have
(f)
2
=
(20)(1.361)
2
= 13.61 (10.57)
f = 13.61
13.61
2
+ 89.44
2
(10.58)
= 13.61 90.47 (10.59)
= 76.86 kHz, 104.08 kHz (10.60)
Note that in the above equations all the frequencies except for have been expressed
in kHz. Of course the equation may be solved in rad/s or in Hz if desired, but since only
the ratios of frequencies are involved, any consistent set of units may be used.
10.5 The Band Stop Transformation
To make a band stop lter, the series elements must be high impedance at some
frequency
o
and the shunt elements low impedance at
o
. Hence we use a parallel L-C
in the series branches and a series L-C in the shunt branches.
The required transformations are
s
n
=

o
_
s
o
+

o
s
_
(10.61)
=

o
_

o

o
_
(10.62)
We will not further consider band stop lters.
Introduced notation for a unit angular frequency.
Extended our treatment of frequency scaling and impedance scaling for low pass
lters.
Introduced scaling to produce high pass responses from low pass responses.
Introduced scaling to produce band pass responses from low pass responses.
Stated a transformation which will produce band stop responses.
Chapter 11
REALISATION OF PASSIVE
FILTER CIRCUITS
11.1 Introduction
Once the lter transfer function H(s) is obtained, the next step is to realise it with an
appropriate circuit. In this section we will consider realisation as a passive circuit, which
is one which contains no active devices such as operational ampliers. These circuits are
important at frequencies or power levels where the use of active lters is precluded because
of frequency response limitations or power handling capacity of the active devices.
The realisation of passive circuits takes us into the area of circuit synthesis, which is
much too extensive to cover fully in this course. However we will look at some of the
principles involved.
11.2 Some important concepts
11.2.1 Aspects of notation
In this chapter we will, as elsewhere in these notes, employ peak value phasors, and thus
will nd some factors of 2 present in power calculations. In this respect the notes may
dier from the year 2000 version. Such factors would be absent if r.m.s. phasors were
used.
We will also be calculating the responses of lters which are designed for an angular
cut o frequency of 1 rad/s. The result will be that the responses will be expressed in
terms of a dimensioned complex angular frequency variable s of units rad/s, rather than
a normalised complex angular frequency variable s
n
of no units.
A result will be that many of the expressions describing transfer functions will con-
tain dimensioned numerical constants, and where those constants occur in the interior of
expression, the dimensions will not be explicitly recognised. This usage is traditional.
11.2.2 Available source power
Although this topic has been treated elsewhere in the course, and indeed in these notes,
its importance justies a re-presentation.
167
168 CHAPTER 11. REALISATION OF PASSIVE FILTER CIRCUITS
Figure 11.1 shows the equivalent circuit of a voltage source of which the internal
voltage is represented by the peak value phasor V
S
and whose internal impedance is the
resistor R.
Figure 11.1: Thevenin equivalent circuit of a source.
By the maximum power transfer theorem of circuit theory, the load impedance R
L
which will extract maximum power from that source is equal to the complex conjugate of
the source impedance. When, as is the case here, the source impedance has the real value
R, to maximum power extracting load is also R, and at that load the voltage V
L
which
appears across the load is given by
V
L
=
1
2
V
S
(11.1)
The power then delivered to the load is called the available source power and is given
by
P =
|V
S
|
2
8R
(11.2)
11.2.3 Scattering Parameters
Scattering parameters nd their natural application in distributed circuits, or where net-
works need to be interconnected by transmission lines, which are generally assumed to
be lossless. In such cases it is much easier to keep track of forward and backward waves
on the interconnecting transmission lines than it is to work in terms of total voltage and
current thereon. An illustration of a two-port network with interconnecting transmission
lines is provided in Figure 11.2.
The lines connected to ports 1 and 2 have characteristic impedances Z
01
and Z
02
respectively. On those lines we have, as shown in the gure, travelling waves a
1
and a
2
incident on ports 1 and 2 respectively, and we have travelling waves b
1
and b
2
emerging
from ports 1 and 2 respectively. Scattering parameters will be dened by forming ratios
between the a and b variables, when those variables have been dened.
There are a number of possible choices for the a and b variables. When the two
transmission lines have the same characteristic impedance, it is possible, from the point of
view of dening the scattering parameters, to choose a and b to be the phasors representing
the incident and emergent wave components of the voltage waves on the line. When the
11.2. SOME IMPORTANT CONCEPTS 169
Figure 11.2: Two-port network with interconnected transmission lines.
two transmission lines have dierent characteristic impedances, the a and b variables are
chosen to be proportional to and in phase with the incident and emergent voltage waves,
but are rescaled in magnitude (and also changed in units) so that they bear a simple and
common relation to the powers incident on and emergent from the ports.
In that case, if V
+
1
and V
+
2
represent the incident voltage waves at ports 1 and 2
while V
1
and V
2
represent the emergent voltage waves for ports 1 and 2 respectively, we
formally dene
a
1
= V
+
1
/
_
Z
01
(11.3)
a
2
= V
+
2
/
_
Z
02
(11.4)
b
1
= V
1
/
_
Z
01
(11.5)
b
2
= V
2
/
_
Z
02
(11.6)
We will consider in these notes mostly the case when the transmission lines all have a
common characteristic impedance Z
0
.
The scattering parameters are dened as the coecients S
ij
(s) in the following linear
equations
b
1
(s) = S
11
(s)a
1
(s) +S
12
(s)a
2
(s) (11.7)
b
2
(s) = S
21
(s)a
1
(s) +S
22
(s)a
2
(s) (11.8)
which may be interpreted to say that if we send incident waves a
1
and a
2
on trans-
mission lines 1 and 2 toward the network, the network will send emergent waves b
1
and
b
2
back to us. Examining these equations, we see that the signicance of the scattering
coecients is
S
11
(s) = reection coecient at port 1 when port 2 has a matched load
S
21
(s) = forward scattering from port 1 to port 2 when port 2 has a matched load
S
12
(s) = reverse scattering from port 2 to port 1 when port 1 has a matched load
S
22
(s) = reection coecient at port 2 when port 1 has a matched load.
Now we consider a passive two port circuit, connected to sources with Thevenin equiva-
lent circuits of voltages V
g1
(s) and V
g2
(s) in series with resistances R
1
and R
2
respectively.
For the purpose of denition of scattering parameters, we take the transmission lines which
could be connected to ports 1 and 2 to have characteristic impedances Z
01
and Z
02
to be
equal to R
1
and R
2
respectively.
Figure 11.3: Two port circuit.
We will be interested in the situation where port 1 is the input, and port 2 is the
output, and is terminated in just a passive load R
2
. Then since V
g2
(s) = 0 we have a
2
= 0
and
If we put s = j, then we have
1
2
|a
1
(j)|
2
=
|V
g1
(j)|
2
|
8R
1
(11.9)
= power available from source 1
1
2
|b
2
(j)|
2
=
|V
2
(j)|
2
2R
2
(11.10)
= power in the load impedance R
2
We dene the insertion power gain of the network as
Insertion power gain of the network =
Load power
Available source power
(11.11)
from which we obtain
Insertion power gain of the network = |S
21
|
2
(11.12)
To realise a lter we make S
21
(s) = H(s), where H(s) is the transfer function of one
of the lters considered in the previous sections.
11.3 Lumped L-C Circuits
11.3.1 Objective
Here we wish to realise the transfer function as the insertion gain function of an L-C
circuit terminated at both ends in resistors. We will consider low pass prototype lters
in which the source impedance is 1 ohm. For transfer functions which have a maximum
11.3. LUMPED L-C CIRCUITS 171
value of unity at zero frequency, the load resistance will also be 1 ohm. In cases where
the d.c. gain is not unity (such as occurs in even order Chebyshev or elliptic lters) the
terminating impedance will not be 1 ohm.
We will illustrate the realisation procedure with an example.
11.3.2 Example
Consider a third order Butterworth lter with
H(s)H(s) =
1
1 s
6
(11.13)
H(s) =
1
s
3
+ 2s
2
+ 2s + 1
(11.14)
We want a circuit with S
21
(s) = H(s). The L-C circuit is lossless, so we have
S
11
(s)S
11
(s) = 1 S
21
(s)S
21
(s) (11.15)
=
s
6
1 s
6
(11.16)
=
_
s
3
s
3
+ 2s
2
+ 2s + 1
_
.
_
s
3
s
3
+ 2s
2
2s + 1
_
(11.17)
S
11
(s) =
s
3
s
3
+ 2s
2
+ 2s + 1
(11.18)
Here we have assigned the poles on the left half s plane to S
11
(s) and the zeros can
be assigned arbitrarily but symmetrically.
We can now determine Z
1
(s), the input impedance at port 1.
Z
1
(s) =
1 +S
11
(s)
1 S
11
(s)
(11.19)
=
2s
3
+ 2s
2
+ 2s + 1
2s
2
+ 2s + 1
(11.20)
For an all pole lter (such as Butterworth, Chebyshev or Bessel-Thompson lters) we
can simply nd circuit elements by repeated synthetic division.
Z
1
(s) = s +
s + 1
2s
2
+ 2s + 1
(11.21)
= s +
1
2s +
1
s + 1
(11.22)
This is called a continued fraction expansion and yields the circuit shown in Fig-
ure 11.4.
Figure 11.4: Realisation of third order Butterworth lter.
Taking the sign for S
11
(s) will yield a dierent circuit with the same transfer func-
tion.
With elliptic lters, we can proceed in the same way and obtain Z
1
(s). However the
circuit elements cannot then be found from Z
1
(s) by simple synthetic division since we
have to realise the transfer function zeros, and a more subtle approach is required. The
technique is straightforward, but more complicated than simple synthetic division and
the algebra is tedious (as shown in the example below). Such examples will not form part
of this course.
11.3.3 Example
A third order elliptic lter with
p
= 1 rad/sec, 1 dB passband ripple and 20 dB stopband
attenuation has a transfer function H(s) = S
21
(s) (see the table in Section 9.8.4)
S
21
(s) =
0.32057(s
2
+ 1.4399
2
)
(s + 0.64357)(s
2
+ 0.32300s + 1.0328)
(11.23)
After considerable algebraic labour we are able to derive the two results below.
S
11
(s) =
s(s
2
+ 0.82491)
(s + 0.64357)(s
2
+ 0.32300s + 1.0328)
(11.24)
Z
1
(s) =
2s
3
+ 0.96657s
2
+ 2.0656s + 0.66468
0.96657s
2
+ 0.41576s + 0.66468
(11.25)
= 1.5538s +
1
1.2966s
s
2
+ 1.4399
2
+
1
1.5539s + 1
(11.26)
The expansion of Z
1
(s) in the form shown requires recognition of the zero in S
21
(s)
at s = j1.4399. The rst inductance extracted must have an impedance equal to Z
1
(s)
at that value of s, followed by a shunt element which is series resonant at that frequency.
The method also works for higher order elliptic lters.
This leads to the circuit shown in Figure 11.5 which has S
21
(s) = H(s). The imaginary
zeros of H(s) are realised by the series L-C combination in the central shunt arm.
11.4. ADMITTANCE PARAMETERS 173
Figure 11.5: Realisation of third order elliptic lter.
11.3.4 Generalisation
The lters shown are low pass lters, but of course can be transformed to high pass or
band pass lters by the transformation techniques considered in Chapter 10.
11.4 Admittance Parameters
A two port circuit as shown in Figure 11.6 can also be characterised by its admittance
(or y) parameters.
Figure 11.6: A two port circuit.
These parameters are dened by the equations
I
1
(s) = V
1
(s)y
11
(s) +V
2
(s)y
12
(s) (11.27)
I
2
(s) = V
1
(s)y
21
(s) +V
2
(s)y
22
(s) (11.28)
where y
ij
are the admittance parameters (with the dimensions of admittance) and
y
ji
= y
ij
.
If the two port contains only inductors and capacitors, it can be shown that the
admittance parameters are all of the form
y
ij
(s) =
Even(s)
Odd(s)
or
Odd(s)
Even(s)
(11.29)
where Even(s) is a polynomial with only even powers of s and Odd(s) is a polynomial
with only odd powers of s.
11.4.1 Example
For the circuit in Figure 11.7 nd the y parameters.
Figure 11.7: An example circuit.
Nodal analysis give the equations
I
1
= V
1
_
2
3s
_
+ V
x
_
2
3s
_
0 = V
1
_
2
3s
_
+ V
x
_
8
3s
+
4s
3
_
+ V
2
_
2
s
_
I
2
= V
x
_
2
s
_
+ V
2
_
2
s
_
(11.30)
Eliminating V
x
from these equations gives :
I
1
= V
1
_
2s
2
+ 3
3s
3
+ 6s
_
+ V
2
_
1
s
3
+ 2s
_
I
2
= V
1
_
1
s
3
+ 2s
_
+ V
2
_
2s
2
+ 1
s
3
+ 2s
_
If port #2 is terminated in a 1 resistor, we have I
2
= V
2
, and substituting this
into the second y equation gives
H(s) =
V
2
V
1
=
y
21
1 +y
22
(11.31)
Now if we have a transfer function H(s) and it can be realised by an LC circuit
terminated in 1 , then we need to rearrange H(s) into the form above.
11.4.2 Example: Third Order Butterworth Filter
H(s) =
1
s
3
+ 2s
2
+ 2s + 1
(11.32)
=
1
s
3
+ 2s
1 +
2s
2
+ 1
s
3
+ 2s
(11.33)
In choosing this re-arrangement over all other possible re-arrangements, we are guided
by the principle set out in equation 11.29. We now see that we have
y
22
(s) =
2s
2
+ 1
s
3
+ 2s
(11.34)
=
1
0.5000s +
1
1.333s +
1
1.500s
(11.35)
This leads to the realisation shown in Figure 11.8.
Figure 11.8: Realisation of lter.
Note that the form of the lter realised here is dierent from that realised by the
scattering parameter method. In the scattering parameter method, the lter was realised
with equal source and load impedances, and H(s) = S
21
(s) = 2V
2
(s)/V
g1
(s).
The y parameter method realises a lter with zero source resistance, and H(s) =
V
2
(s)/V
1
(s). Not all transfer functions can be realised by an LC lter terminated in a
resistor, but Butterworth, Chebyshev, Bessel-Thompson and Elliptic lters all can.
Claried some aspects of phasor notation.
Revised the concept of available source power.
Introduced scattering parameters and their relation to travelling waves.
Shown how a polynomial expression for a forward scattering parameter may be
manipulated to produce a polynomial expression for an input impedance.
Shown how the lter elements can be extracted from that expression.
Dened admittance parameters for a network.
Stated relations which must apply between admittance parameters for a lossless
ladder network.
Shown how for such a network fed from a zero source impedance admittance para-
meters can be derived from the transfer function.
Shown how the network elements can then be extracted from the admittance para-
meters.
Chapter 12
ACTIVE FILTERS
12.1 Introduction
12.1.1 Motivation
Inductors are dicult and expensive to manufacture accurately and with low loss.
They also do not allow fabrication as integrated circuits. However, resistors, capacitors
and ampliers are easily made in both discrete and integrated form.
Without active devices, resistors and capacitors alone can only realise circuits with real
poles, and this is much too restrictive to be of any practical use. However by including
active devices, it is possible to realise both real and complex poles. The use of active
devices also allows a building block approach by which complex transfer functions can
be realised as a cascade of low order transfer functions. This is because of the buering
capability of active devices, which enables the individual building blocks to be designed
without having to worry about the loading eects of later sections.
12.1.2 Realisation strategies
There is also a good practical reason for realising transfer functions by simple building
blocks. The roots of high order polynomials tend to be very sensitive to the accuracy
of the coecients of the polynomial, and it can be very dicult to set the coecients
accurately enough to realise the desired transfer function. Hence in practical circuits the
building blocks are usually no more than second order. The same is true of digital lters.
The subject of active lter design is very extensive and we will only consider a small
sample of possible circuits.
12.2 First Order Transfer Functions
12.2.1 Passive circuits realisations
The simplest R-C circuits are shown in Figure 12.1. These have transfer functions
given by
H
lp
(s) =
1
1 +sCR
(12.1)
177
178 CHAPTER 12. ACTIVE FILTERS
Figure 12.1: First order RC circuits.
H
hp
(s) =
sCR
1 +sCR
(12.2)
These circuits should be driven from a low impedance source (|Z
s
| << R) and have
a buer amplier at the output. The ideal buer is a unity gain operational amplier
circuit.
12.2.2 Eect of circuit loading
If the buer input impedance is not innite (e.g. an inverting operational amplier cir-
cuit), then the input impedance must be taken into account in the calculation of the
transfer function. For example, the transfer function of the circuit shown in Figure 12.2 is
H(s) =
R
3
R
1
+R
2
+sCR
1
R
2
(12.3)
Figure 12.2: First order RC circuit with buer amplier.
12.2. FIRST ORDER TRANSFER FUNCTIONS 179
12.2.3 Inverting amplier circuits
However, once it is recognised that an operational amplier is needed, the operational
amplier can be integrated into the circuit. The simplest conguration is the inverting
amplier as shown in Figure 12.3.
Figure 12.3: Inverting operational amplier circuit.
The transfer function of this circuit is
H(s) =
Z
2
(s)
Z
1
(s)
(12.4)
By choosing Z
1
(s) and Z
2
(s) as various series and parallel combinations of Rs and
Cs, a whole variety of transfer functions can be realised. In generating the following
table, the operational ampliers will be assumed ideal. This means that
The gain is assumed large (ideally innite) and independent of frequency.
The inputs are assumed to take no current (ie innite input impedance).
The output impedance is assumed negligible (ideally zero).
In practice none of these assumptions is exactly true, and the non-idealities of the
operational ampliers have to be taken into account. However, here we will assume that
the circuits are realised at frequency and impedance levels where the assumptions above
are reasonably accurate. With these assumptions, we can realise the following transfer
functions.
Z
1
Z
2
Transfer function Type
R
1
R
2

R
2
R
1
Inverting amplier
R
1
sC

1
sCR
Integrator
1
sC
R sCR Dierentiator
R
1
+
1
sC
1
R
2

sC
1
R
2
1 +sC
1
R
High pass lter
R
1
R
2
//
1
sC
2

R
2
R
1
(1 +sC
2
R
2
)
Low pass lter
R
1
R
2
//(R
3
+
1
sC
3
)
R
2
(1 +sC
3
R
3
)
R
1
[1 +sC
3
(R
2
+R
3
)]
Phase lag circuit
R
1
//(R
2
+
1
sC
2
) R
3

R
3
[1 +sC
2
(R
1
+R
2
)]
R
1
(1 +sC
2
R
2
)
Phase lead circuit
Note that // means in parallel with.
12.2.4 Non-inverting amplier circuits
The non-inverting conguration can also be used. For the circuit shown in Figure 2 the
transfer function is
H(s) =
V
o
(s)
V
i
(s)
=
Z
3
(s) +Z
4
(s)
Z
3
(s)
(12.5)
Figure 12.4: Non-inverting operational amplier.
12.3. SECOND ORDER TRANSFER FUNCTIONS 181
Typical congurations include:
Z
3
Z
4
Transfer function Type
R
1
R
2
R
1
+R
2
R
1
Non-inverting amp
0 1 Unity gain buer
R
1
R
2
+
1
sC
2
1 +sC
2
(R
1
+R
2
)
sC
2
R
1
Low freq integrator
R
1
R
2
//(R
3
+
1
sC
3
) 1 +
R
2
(1 +sC
3
R
3
)
R
1
[1 +sC
3
(R
2
+R
3
)]
Phase lag circuit
R
1
//(R
2
+
1
sC
2
) R
3
1 +
R
3
[1 +sC
2
(R
1
+R
2
)]
R
1
(1 +sC
2
R
2
)
Phase lead circuit
The non-inverting conguration is more limited in the types of response it can realise,
since with R and C elements it can never produce a transfer function with a gain of less
than unity at any frequency.
12.2.5 More complex functions
More complicated transfer functions can be realised by connecting the circuits in cascade,
because the buering eect of the operational amplier (i.e. its low output impedance)
means that the overall transfer function is simply the product of the individual transfer
functions. You should note that this is NOT the case with circuits using only passive
components.
While both the inverting and non-inverting congurations shown above can realise
transfer functions with more than one pole or zero, they can only realise real poles and
zeros.
12.3 Second Order Transfer Functions
12.3.1 Overview
This section will consider active RC circuits which can realise transfer functions with
two poles which may be either both real or a conjugate complex pair.
It can be shown that passive circuits made up of resistors and capacitors alone can
only realise real poles. To realise complex poles, such as required in many of the lters
considered previously, either inductors or active devices must be included.
12.3.2 Low pass Sallen and Key circuit
One conguration which realises a transfer function with no zeros and two poles is the
Sallen and Key circuit shown in Figure 12.5. The amplier of gain K can be realised by a
non-inverting operational amplier with appropriate resistor values. A particularly stable
conguration is obtained when K = 1 using the unity gain buer circuit.
Figure 12.5: Low pass Sallen and Key circuit.
Analysing this by nodal analysis gives:
V
a
_
1
R
1
+
1
R
2
+sC
1
_
V
b
_
KsC
1
+
1
R
2
_
= V
i
_
1
R
1
_
V
a
_
1
R
2
_
+ V
b
_
1
R
2
+sC
2
_
= 0
Hence:
H(s) =
V
o
(s)
V
i
(s)
= K
V
b
(s)
V
i
(s)
(12.6)
(12.7)
H(s) =
K
R
1
R
2
C
1
C
2
s
2
+s
_
1
R
1
C
1
+
1
R
2
C
1
+
1 K
R
2
C
2
_
+
1
R
1
R
2
C
1
C
2
(12.8)
This transfer function is of the form
H(s) =
K
2
o
s
2
+s
o
/Q+
2
o
(12.9)
and K is the gain at zero frequency.
By comparing coecients we obtain the relations
2
o
=
1
C
1
C
2
R
1
R
2
(12.10)
o
Q
=
1
R
1
C
1
+
1
R
2
C
1
+
1 K
R
2
C
2
(12.11)
If Q > 1/2 then this circuit will realise complex poles. The design can be simplied by
choosing R
1
= R
2
= R and K = 1.
12.3.3 Example
Design a lter to have a transfer function with f
o
= 1 kHz and Q = 10.
Choose R
1
= R
2
= 10 k and K = 1. Hence we need
C
1
C
2
=
1
4
2
f
2
o
R
1
R
2
= 2.533 10
16
F
2
(12.12)
1
C
1
_
1
R
1
+
1
R
2
_
=

o
Q
= 628.3 (12.13)
C
1
= 3.183 10
7
F (12.14)
C
2
= 7.958 10
10
F (12.15)
12.3.4 Band pass Sallen and Key circuit
To realise two poles and a zero at the origin we can use the circuit in Figure 12.6.
Figure 12.6: Band pass Sallen and Key circuit.
This has a transfer function
H(s) =
sK
R
1
C
1
s
2
+s
_
1
R
1
C
1
+
1
R
3
C
1
+
1
R
3
C
2
+
1 K
R
2
C
1
_
+
R
1
+R
2
R
1
R
2
R
3
C
1
C
2
(12.16)
2
o
=
R
1
+R
2
R
1
R
2
R
3
C
1
C
2
(12.17)
o
Q
=
1
R
1
C
1
+
1
R
3
C
1
+
1
R
3
C
2
+
1 K
R
2
C
1
(12.18)
H(j
o
) =
KQ
o
R
1
C
1
(12.19)
12.3.5 Example
Realise a band pass lter with
o
= 1 rad/sec and Q = 2.
We will choose R
1
= R
2
= R
3
= 1 and C
1
= C
2
= C.
o
=
2
C
rad/s (12.20)
C = 1.4142 F (12.21)
o
Q
=
4 K
C
(12.22)
K = 3.2929 (12.23)
H(j
o
) = 4.6569 (12.24)
This design can be frequency and impedance scaled as required.
12.3.6 Inverting Band Pass Circuit
Another conguration which produces a response with a zero at s = 0 and two (usually
complex) poles is shown in Figure 12.7.
This circuit uses the operational amplier as an integral part of the circuit. It can be
analysed using the virtual earth concept. The virtual earth concept requires that V
b
= 0
in the circuit, so nodal analysis gives:
V
a
_
1
R
1
+sC
1
+sC
2
_
V
o
(sC
1
) = V
i
_
1
R
1
_
V
a
(sC
2
) + V
o
_
1
R
2
_
= 0
This gives a transfer function
H(s) =
V
o
(s)
V
i
(s)
=

s
R
1
C
1
s
2
+s
_
1
R
2
C
1
+
1
R
2
C
2
_
+
1
R
1
R
2
C
1
C
2
(12.25)
Figure 12.7: Inverting band pass circuit.
2
o
=
1
R
1
R
2
C
1
C
2
(12.26)
o
Q
=
1
R
2
C
1
+
1
R
2
C
2
(12.27)
H(j
o
) =
R
2
C
2
R
1
(C
1
+C
2
)
(12.28)
12.3.7 Biquadratic Filters
Of particular interest here is the conguration which can be used to realise zeros on the
imaginary axis as required by elliptic lters. A normalised circuit is shown in Figure 12.8.
The values next to the resistors are their resistances in ohms, and the values next to the
capacitors are their capacitances in farads. The amplier is assumed ideal, with constant
voltage gain K, innite input impedance, and zero output impedance. The number m is
a scaling constant which controls the spread of component values. Typically m = 1.
It has a transfer function
H(s) =
K
o
(s
2
+
2
z
)
s
2
+s
o
Q
+
2
o
(12.29)
The choice of the admittance Y shown in Figure 12.8 depends on the relative values
of
z
and
o
.
12.3.8 Case 1:
z
<
o
We choose Y = 1/R (ie. the element is a resistor of resistance R ohms). The transfer
function is then
Figure 12.8: Bi-quadratic band pass circuit.
H(s) =
K
_
s
2
+
1
C
2
_
s
2
+s
_
m+ 1
C
__
1
R
+
2 K
m
_
+
1
C
2
_
1 +
m+ 1
R
_ (12.30)
Matching coecients gives
K
o
= K (12.31)
z
=
1
C
(12.32)
2
o
=
1
C
2
_
1 +
m+ 1
R
_
(12.33)
o
Q
=
_
m+ 1
C
_ _
1
R
+
2 K
m
_
(12.34)
The last three equations can be solved for C, R and K.
C =
1
z
(12.35)
(12.36)
R =
m+ 1
_
2
o
2
z
1
_
(12.37)
(12.38)
K = 2 +
m
m+ 1
_
2
o
2
z
1

o
Q
z
_
(12.39)
12.4. SWITCHED CAPACITOR FILTERS 187
12.3.9 Case 2:
z
>
o
We choose the admittance Y = sCr where r is some ratio (ie. this element is a capacitor
of value rC farads). The transfer function is then
H(s) =
KA
_
s
2
+
1
C
2
_
s
2
+sA
_
m+ 1
C
__
r +
2 K
m
_
+
A
C
2
(12.40)
A =
1
1 + (m+ 1)r
(12.41)
Again, matching coecients gives
K
o
=
K
1 + (m+ 1)r
(12.42)
z
=
1
C
(12.43)
2
o
=
1
C
2
{1 + (m+ 1)r}
(12.44)
o
Q
=
m+ 1
C{1 + (m+ 1)r}
_
r +
2 K
m
_
(12.45)
The last three equations can be solved for C, r and K.
C =
1
z
(12.46)
r =
_
2
z
2
o
1
_
m+ 1
(12.47)
K = 2 +
m
m+ 1
_
2
z
2
o
1

z
Q
o
_
(12.48)
12.4 Switched Capacitor Filters
12.4.1 Introduction
One of the diculties encountered with active lter circuits is in tuning the lters to the
correct frequency. With discrete components this can be done either by using components
of suciently high accuracy or by providing trimming adjustments.
However with integrated circuit realisations it is often dicult to realise component
values accurately, although the components on one integrated circuit will closely track one
another. That is, two capacitors which are equal on one circuit will be equal on another,
but the absolute values may dier from circuit to circuit. This leads to a problem since
unless the resistances and capacitances vary such that the product RC is the same from
circuit to circuit, the frequency responses of the circuits will be dierent.
One way of overcoming this problem is to realise resistors using capacitors. Consider
the circuit shown in Figure 12.9.
Figure 12.9: Switched capacitor circuit.
The two switches are usually realised as FET switches. The two switches SW1 and
SW2 are turned on and o periodically at a rate f
c
such that their on periods do not
overlap. This is shown in Figure 12.10.
Now suppose the circuit is connected to two voltages v
1
and v
2
which do not vary
signicantly over one clock period. During the time SW1 is on, the capacitor C charges
to the voltage v
1
, assuming the switch resistance is low enough that the charging time
constant is much less than T
c
= 1/f
c
.
Figure 12.10: Switch waveforms.
When SW2 is on, the capacitor is charged to v
2
. Hence the capacitor is alternately
charged to voltages v
1
and v
2
. During the time SW1 is on, a charge q = C(v
1
v
2
) ows
from v
1
to the capacitor. During the time SW2 is on the same charge then ows out to
v
2
. Hence we have an average current ow from v
1
to v
2
which is
i =
q
T
c
= Cf
c
(v
1
v
2
) (12.49)
12.4. SWITCHED CAPACITOR FILTERS 189
This is the same as if a resistance R was connected between v
1
and v
2
, where R is
R =
1
Cf
c
(12.50)
12.4.2 Low pass lter
Hence we can make a low pass lter with the circuit shown in Figure 12.11.
Figure 12.11: Switched capacitor low pass lter.
The transfer function of this lter is
H(s) =
1
1 +
sC
2
Cf
c
(12.51)
and it can be seen that the frequency response is determined by f
c
and the ratio of
the capacitors.
12.4.3 Circuit properties
This circuit has the following advantages.
The only elements required are capacitors and active elements (as ampliers or
switches).
The lter can be tuned simply by altering the clock frequency f
c
.
Since all capacitor ratios will be preserved in an integrated circuit realisation, this
means the RC products remain constant from circuit to circuit.
The clock frequency has to be chosen much higher than any of the frequencies to be
ltered (perhaps ten times higher at least), but low enough such that T
c
is much greater
than the charge and discharge time constants for the switched capacitor.
Switched capacitor lters can be congured exactly the same as the active lters we
have considered so far, simply replacing all the resistors by switched capacitors. However,
switched capacitor lters really operate by sampling all signals at a rate f
c
, so strictly
speaking they are discrete time lters. These are usually called digital lters.
12.5 Frequency Limitations
Active lter resonant circuits usually work as expected up to frequencies which are about
0.05 times the unity gain frequency of the operational amplier although problems may
occur at lower frequencies than this if the poles have high Q. At higher frequencies the
non-ideality of the frequency response of the operational amplier needs to be taken into
account. This can be done by replacing the ideal operational amplier with one with a
transfer function
H
a
(s) =
A
0
1 +
sA
0
b
(12.52)
where A
0
= DC gain and
b
is the unity gain angular frequency in rad/s.
Explained why active lters are needed in current communications technology.
Cataloged the lter responses obtainable from single operational amplier circuits.
Introduces two Sallen and Key circuits for producing second order transfer functions.
Introduced biquadratic lters for producing second order transfer functions.
Briey described switched capacitor lters.
Appendix A
REFERENCES
The following books are useful as references.
1. SI units and recommendations for the use of their multiples and of certain other
units, International Standard ISO 1000 (1992), International Organisation for Stan-
dardisation, Case postale 56, CH-1211, Genevre 20, Switzerland.
2. Peyton Z. Peebles: Probability, Random Variables and Random Signal Principles,
Mc Graw Hill, 2001.
3. R. D. Yates and D. J. Goodman: Probability and Stochastic Processes: A Friendly
Introduction for Electrical and Computer Engineers, Wiley 1999.
4. P Cooke: Noise and Modulation in Communication Eelctronics, Cooke, 2000.
5. S. Hayken and B. van Veen: Signals and Systems, Wiley, 1999.
6. R. G. Brown and P. Y. C. Hwang: Introduction to Randon Signals and Applied
Kalman Filtering, Wiley 1992.
7. A. Papoulis: Probability, Random Variables and Stochastic Processes, Mc. Graw
Hill, 1984.
8. Gordon E. Carlson Signal and Liner System Analysis, Houghton Miin, 1992.
9. A. Bruce Carslon: Communication Systems, Mc. Graw Hill, 1975.
10. W. Davenport and W. Root: An Introduction to the Theory of Random Signals
and Noise, Mc. Graw Hill, 1958.
11. A. Zverev: Handbook of Filter Synthesis, Wiley, 1967.
12. Matthei, L. Young, E.M.T. Jones, Microwave Filters, Impedance-Matching Net-
works, and Coupling Structures, Reprint of the edition published by McGraw-Hill,
p. 203, 207, New York, 1964, Artech House, Norwood, 1980.
191
192 APPENDIX A. REFERENCES
Appendix B
FORMULAE AND TABLES
B.1 Acknowledgement
The information in this Appendix has been kindly supplied by Dr. B. R. Davis.
B.2 Complex Numbers
j =

1
z = x +jy = r e
j
z
= x jy = r e
j
= arg z = arg(x, y)
|z| = r =
_
x
2
+y
2
z z
= r
2
= x
2
+y
2
= |z|
2
x = Re{z} =
1
2
(z +z
)
y = Im{z} =
1
2j
(z z
)
193
194 APPENDIX B. FORMULAE AND TABLES
B.3 Trigonometric Identities
cos(A +B) = cos Acos B sinAsin B
cos(AB) = cos Acos B + sinAsinB
sin(A +B) = sinAcos B + cos AsinB
sin(AB) = sinAcos B cos Asin B
2 cos Acos B = cos(A+B) + cos(AB)
2 cos AsinB = sin(A +B) sin(AB)
2 sin Acos B = sin(A +B) + sin(A B)
2 sinAsinB = cos(AB) cos(A +B)
cos A+ cos B = 2 cos
_
A +B
2
_
cos
_
AB
2
_
cos A cos B = 2 sin
_
A+B
2
_
sin
_
B A
2
_
sinA + sinB = 2 sin
_
A+B
2
_
cos
_
AB
2
_
sin AsinB = 2 cos
_
A +B
2
_
sin
_
AB
2
_
cos 2 = 2 cos
2
1 = 1 2 sin
2
= cos
2
sin
2
sin 2 = 2 sin cos

cos 3 = 4 cos
3
3 cos
sin 3 = 3 sin 4 sin
3
A cos +B sin =

A
2
+B
2
cos{ arg(A, B)}
arg(A, B) =
arctan(B/A) ; A > 0
arctan(B/A) + ; A < 0, B > 0
arctan(B/A) ; A < 0, B < 0
e
j
= cos j sin
cos =
1
2
(e
j
+e
j
)
sin =
1
2j
(e
j
e
j
)
B.4 Fourier Transforms
B.4.1 The direct transform
X(f) =
_
+
x(t)e
j2ft
dt
B.4. FOURIER TRANSFORMS 195
B.4.2 The reverse transform
x(t) =
_
+
X(f)e
+j2ft
df
B.4.3 Theorems and transforms
Theorems Transforms
x(t) X(f) u(t)e
at
1
a +j2f
; a > 0
x(t/T) |T| X(fT) e
a|t|
2a
a
2
+ (2f)
2
; a > 0
x(t T) X(f) e
j2fT
1
a
2
+t
2
a
e
|2fa|
; a > 0
x(t) e
j2Ft
X(f F) (t) 1
x(t) X(f) 1 (f)
dx(t)
dt
j2f X(f) u(t)
1
j2f
+
1
2
(f)
_
t
x() d
X(f)
j2f
+
1
2
X(0) (f) sgn(t)
1
jf
t x(t)
1
j2
dX(f)
df
1
t
j sgn(f)
X(t) x(f) rect(t/T) |T| sinc(fT)
rep
T
x(t) |F| comb
F
(f) X(f) sinc(t/T) |T| rect(fT)
|T| comb
T
(t) x(t) rep
F
X(f) (t/T) |T| sinc
2
(fT)
x(t) y(t) X(f) Y (f) comb
T
(t) |F| comb
F
(f)
x(t) y(t) X(f) Y (f) e
t
2
/2T
2
|T|
2 e
1
2
(2fT)
2
x
(t) X
(f) sgn(t) rect(t/T)

1 cos(fT)
jf
[ Note that a, F and T are real constants, with a > 0 and F T = 1. ]
Denitions Relations
sinc(x) =
sin(x)
x
x(0) = area of X(f)
rep
P
f(x) =
+
f(x nP) X(0) = area of x(t)

comb
P
(x) =
+
(x nP) X(f) = X
(f) if x(t) is real

u(x) =
_
1 ; x > 0
0 ; x < 0
X(f) = real & even if x(t) real & even
(x) = unit impulse X(f) = imag. & odd if x(t) real & odd
sgn(x) =
_
1 ; x > 0
1 ; x < 0
_
+
x(t) y
(t) dt =
_
+
X(f)Y
(f) df
rect(x) =
_
1 ; |x| < 0.5
0 ; |x| > 0.5
_
+
|x(t)|
2
dt =
_
+
|X(f)|
2
df
(x) =
_
1 |x| ; |x| < 1
0 ; |x| > 1
f(x) g(x) =
_
+
f() g(x ) d
[ Unless stated otherwise, the relations are valid for x(t) real or complex. ]
TABLE OF THE Q FUNCTION
Q(x) =
1
2
_

x
e
t
2
/2
dt
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0.0 5.000E-01 4.960E-01 4.920E-01 4.880E-01 4.840E-01 4.801E-01 4.761E-01 4.721E-01 4.681E-01 4.641E-01
0.1 4.602E-01 4.562E-01 4.522E-01 4.483E-01 4.443E-01 4.404E-01 4.364E-01 4.325E-01 4.286E-01 4.247E-01
0.2 4.207E-01 4.168E-01 4.129E-01 4.090E-01 4.052E-01 4.013E-01 3.974E-01 3.936E-01 3.897E-01 3.859E-01
0.3 3.821E-01 3.783E-01 3.745E-01 3.707E-01 3.669E-01 3.632E-01 3.594E-01 3.557E-01 3.520E-01 3.483E-01
0.4 3.446E-01 3.409E-01 3.372E-01 3.336E-01 3.300E-01 3.264E-01 3.228E-01 3.192E-01 3.156E-01 3.121E-01
0.5 3.085E-01 3.050E-01 3.015E-01 2.981E-01 2.946E-01 2.912E-01 2.877E-01 2.843E-01 2.810E-01 2.776E-01
0.6 2.743E-01 2.709E-01 2.676E-01 2.643E-01 2.611E-01 2.578E-01 2.546E-01 2.514E-01 2.483E-01 2.451E-01
0.7 2.420E-01 2.389E-01 2.358E-01 2.327E-01 2.296E-01 2.266E-01 2.236E-01 2.206E-01 2.177E-01 2.148E-01
0.8 2.119E-01 2.090E-01 2.061E-01 2.033E-01 2.005E-01 1.977E-01 1.949E-01 1.922E-01 1.894E-01 1.867E-01
0.9 1.841E-01 1.814E-01 1.788E-01 1.762E-01 1.736E-01 1.711E-01 1.685E-01 1.660E-01 1.635E-01 1.611E-01
1.0 1.587E-01 1.562E-01 1.539E-01 1.515E-01 1.492E-01 1.469E-01 1.446E-01 1.423E-01 1.401E-01 1.379E-01
1.1 1.357E-01 1.335E-01 1.314E-01 1.292E-01 1.271E-01 1.251E-01 1.230E-01 1.210E-01 1.190E-01 1.170E-01
1.2 1.151E-01 1.131E-01 1.112E-01 1.093E-01 1.075E-01 1.056E-01 1.038E-01 1.020E-01 1.003E-01 9.853E-02
1.3 9.680E-02 9.510E-02 9.342E-02 9.176E-02 9.012E-02 8.851E-02 8.692E-02 8.534E-02 8.379E-02 8.226E-02
1.4 8.076E-02 7.927E-02 7.780E-02 7.636E-02 7.493E-02 7.353E-02 7.215E-02 7.078E-02 6.944E-02 6.811E-02
1.5 6.681E-02 6.552E-02 6.426E-02 6.301E-02 6.178E-02 6.057E-02 5.938E-02 5.821E-02 5.705E-02 5.592E-02
1.6 5.480E-02 5.370E-02 5.262E-02 5.155E-02 5.050E-02 4.947E-02 4.846E-02 4.746E-02 4.648E-02 4.551E-02
1.7 4.457E-02 4.363E-02 4.272E-02 4.182E-02 4.093E-02 4.006E-02 3.920E-02 3.836E-02 3.754E-02 3.673E-02
1.8 3.593E-02 3.515E-02 3.438E-02 3.362E-02 3.288E-02 3.216E-02 3.144E-02 3.074E-02 3.005E-02 2.938E-02
1.9 2.872E-02 2.807E-02 2.743E-02 2.680E-02 2.619E-02 2.559E-02 2.500E-02 2.442E-02 2.385E-02 2.330E-02
2.0 2.275E-02 2.222E-02 2.169E-02 2.118E-02 2.068E-02 2.018E-02 1.970E-02 1.923E-02 1.876E-02 1.831E-02
2.1 1.786E-02 1.743E-02 1.700E-02 1.659E-02 1.618E-02 1.578E-02 1.539E-02 1.500E-02 1.463E-02 1.426E-02
2.2 1.390E-02 1.355E-02 1.321E-02 1.287E-02 1.255E-02 1.222E-02 1.191E-02 1.160E-02 1.130E-02 1.101E-02
2.3 1.072E-02 1.044E-02 1.017E-02 9.903E-03 9.642E-03 9.387E-03 9.137E-03 8.894E-03 8.656E-03 8.424E-03
2.4 8.198E-03 7.976E-03 7.760E-03 7.549E-03 7.344E-03 7.143E-03 6.947E-03 6.756E-03 6.569E-03 6.387E-03
2.5 6.210E-03 6.037E-03 5.868E-03 5.703E-03 5.543E-03 5.386E-03 5.234E-03 5.085E-03 4.940E-03 4.799E-03
2.6 4.661E-03 4.527E-03 4.397E-03 4.269E-03 4.145E-03 4.025E-03 3.907E-03 3.793E-03 3.681E-03 3.573E-03
2.7 3.467E-03 3.364E-03 3.264E-03 3.167E-03 3.072E-03 2.980E-03 2.890E-03 2.803E-03 2.718E-03 2.635E-03
2.8 2.555E-03 2.477E-03 2.401E-03 2.327E-03 2.256E-03 2.186E-03 2.118E-03 2.052E-03 1.988E-03 1.926E-03
2.9 1.866E-03 1.807E-03 1.750E-03 1.695E-03 1.641E-03 1.589E-03 1.538E-03 1.489E-03 1.441E-03 1.395E-03
3.0 1.350E-03 1.306E-03 1.264E-03 1.223E-03 1.183E-03 1.144E-03 1.107E-03 1.070E-03 1.035E-03 1.001E-03
3.1 9.676E-04 9.354E-04 9.043E-04 8.740E-04 8.447E-04 8.164E-04 7.888E-04 7.622E-04 7.364E-04 7.114E-04
3.2 6.871E-04 6.637E-04 6.410E-04 6.190E-04 5.976E-04 5.770E-04 5.571E-04 5.377E-04 5.190E-04 5.009E-04
3.3 4.834E-04 4.665E-04 4.501E-04 4.342E-04 4.189E-04 4.041E-04 3.897E-04 3.758E-04 3.624E-04 3.495E-04
3.4 3.369E-04 3.248E-04 3.131E-04 3.018E-04 2.909E-04 2.803E-04 2.701E-04 2.602E-04 2.507E-04 2.415E-04
3.5 2.326E-04 2.241E-04 2.158E-04 2.078E-04 2.001E-04 1.926E-04 1.854E-04 1.785E-04 1.718E-04 1.653E-04
3.6 1.591E-04 1.531E-04 1.473E-04 1.417E-04 1.363E-04 1.311E-04 1.261E-04 1.213E-04 1.166E-04 1.121E-04
3.7 1.078E-04 1.036E-04 9.961E-05 9.574E-05 9.201E-05 8.842E-05 8.496E-05 8.162E-05 7.841E-05 7.532E-05
3.8 7.235E-05 6.948E-05 6.673E-05 6.407E-05 6.152E-05 5.906E-05 5.669E-05 5.442E-05 5.223E-05 5.012E-05
3.9 4.810E-05 4.615E-05 4.427E-05 4.247E-05 4.074E-05 3.908E-05 3.747E-05 3.594E-05 3.446E-05 3.304E-05
4.0 3.167E-05 3.036E-05 2.910E-05 2.789E-05 2.673E-05 2.561E-05 2.454E-05 2.351E-05 2.252E-05 2.157E-05
4.1 2.066E-05 1.978E-05 1.894E-05 1.814E-05 1.737E-05 1.662E-05 1.591E-05 1.523E-05 1.458E-05 1.395E-05
4.2 1.335E-05 1.277E-05 1.222E-05 1.168E-05 1.118E-05 1.069E-05 1.022E-05 9.774E-06 9.345E-06 8.934E-06
4.3 8.540E-06 8.163E-06 7.801E-06 7.455E-06 7.124E-06 6.807E-06 6.503E-06 6.212E-06 5.934E-06 5.668E-06
4.4 5.413E-06 5.169E-06 4.935E-06 4.712E-06 4.498E-06 4.294E-06 4.098E-06 3.911E-06 3.732E-06 3.561E-06
4.5 3.398E-06 3.241E-06 3.092E-06 2.949E-06 2.813E-06 2.682E-06 2.558E-06 2.439E-06 2.325E-06 2.216E-06
4.6 2.112E-06 2.013E-06 1.919E-06 1.828E-06 1.742E-06 1.660E-06 1.581E-06 1.506E-06 1.434E-06 1.366E-06
4.7 1.301E-06 1.239E-06 1.179E-06 1.123E-06 1.069E-06 1.017E-06 9.680E-07 9.211E-07 8.765E-07 8.339E-07
4.8 7.933E-07 7.547E-07 7.178E-07 6.827E-07 6.492E-07 6.173E-07 5.869E-07 5.580E-07 5.304E-07 5.042E-07
4.9 4.792E-07 4.554E-07 4.327E-07 4.111E-07 3.906E-07 3.711E-07 3.525E-07 3.348E-07 3.179E-07 3.019E-07
TABLE OF THE Q FUNCTION
Q(x) =
1
2
_

x
e
t
2
/2
dt
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
5.0 2.867E-07 2.722E-07 2.584E-07 2.452E-07 2.328E-07 2.209E-07 2.096E-07 1.989E-07 1.887E-07 1.790E-07
5.1 1.698E-07 1.611E-07 1.528E-07 1.449E-07 1.374E-07 1.302E-07 1.235E-07 1.170E-07 1.109E-07 1.051E-07
5.2 9.964E-08 9.442E-08 8.946E-08 8.476E-08 8.029E-08 7.605E-08 7.203E-08 6.821E-08 6.459E-08 6.116E-08
5.3 5.790E-08 5.481E-08 5.188E-08 4.911E-08 4.647E-08 4.398E-08 4.161E-08 3.937E-08 3.724E-08 3.523E-08
5.4 3.332E-08 3.151E-08 2.980E-08 2.818E-08 2.664E-08 2.518E-08 2.381E-08 2.250E-08 2.127E-08 2.010E-08
5.5 1.899E-08 1.794E-08 1.695E-08 1.601E-08 1.512E-08 1.428E-08 1.349E-08 1.274E-08 1.203E-08 1.135E-08
5.6 1.072E-08 1.012E-08 9.548E-09 9.010E-09 8.503E-09 8.022E-09 7.569E-09 7.140E-09 6.735E-09 6.352E-09
5.7 5.990E-09 5.649E-09 5.326E-09 5.022E-09 4.734E-09 4.462E-09 4.206E-09 3.964E-09 3.735E-09 3.519E-09
5.8 3.316E-09 3.124E-09 2.942E-09 2.771E-09 2.610E-09 2.458E-09 2.314E-09 2.179E-09 2.051E-09 1.931E-09
5.9 1.818E-09 1.711E-09 1.610E-09 1.515E-09 1.425E-09 1.341E-09 1.261E-09 1.186E-09 1.116E-09 1.049E-09
6.0 9.866E-10 9.276E-10 8.721E-10 8.198E-10 7.706E-10 7.242E-10 6.806E-10 6.396E-10 6.009E-10 5.646E-10
6.1 5.303E-10 4.982E-10 4.679E-10 4.394E-10 4.126E-10 3.874E-10 3.637E-10 3.414E-10 3.205E-10 3.008E-10
6.2 2.823E-10 2.649E-10 2.486E-10 2.332E-10 2.188E-10 2.052E-10 1.925E-10 1.805E-10 1.693E-10 1.587E-10
6.3 1.488E-10 1.395E-10 1.308E-10 1.226E-10 1.149E-10 1.077E-10 1.009E-10 9.451E-11 8.854E-11 8.294E-11
6.4 7.769E-11 7.276E-11 6.814E-11 6.380E-11 5.974E-11 5.593E-11 5.235E-11 4.900E-11 4.586E-11 4.292E-11
6.5 4.016E-11 3.758E-11 3.515E-11 3.288E-11 3.076E-11 2.877E-11 2.690E-11 2.516E-11 2.352E-11 2.199E-11
6.6 2.056E-11 1.922E-11 1.796E-11 1.678E-11 1.568E-11 1.465E-11 1.369E-11 1.279E-11 1.195E-11 1.116E-11
6.7 1.042E-11 9.731E-12 9.086E-12 8.483E-12 7.919E-12 7.392E-12 6.900E-12 6.439E-12 6.009E-12 5.607E-12
6.8 5.231E-12 4.880E-12 4.552E-12 4.246E-12 3.960E-12 3.692E-12 3.443E-12 3.210E-12 2.993E-12 2.790E-12
6.9 2.600E-12 2.423E-12 2.258E-12 2.104E-12 1.960E-12 1.826E-12 1.701E-12 1.585E-12 1.476E-12 1.374E-12
7.0 1.280E-12 1.192E-12 1.109E-12 1.033E-12 9.612E-13 8.946E-13 8.325E-13 7.747E-13 7.208E-13 6.706E-13
7.1 6.238E-13 5.802E-13 5.396E-13 5.018E-13 4.667E-13 4.339E-13 4.034E-13 3.750E-13 3.486E-13 3.240E-13
7.2 3.011E-13 2.798E-13 2.599E-13 2.415E-13 2.243E-13 2.084E-13 1.935E-13 1.797E-13 1.669E-13 1.550E-13
7.3 1.439E-13 1.336E-13 1.240E-13 1.151E-13 1.068E-13 9.910E-14 9.196E-14 8.531E-14 7.914E-14 7.341E-14
7.4 6.809E-14 6.315E-14 5.856E-14 5.430E-14 5.034E-14 4.667E-14 4.326E-14 4.010E-14 3.716E-14 3.444E-14
7.5 3.191E-14 2.956E-14 2.739E-14 2.537E-14 2.350E-14 2.176E-14 2.015E-14 1.866E-14 1.728E-14 1.600E-14
7.6 1.481E-14 1.370E-14 1.268E-14 1.174E-14 1.086E-14 1.005E-14 9.297E-15 8.600E-15 7.954E-15 7.357E-15
7.7 6.803E-15 6.291E-15 5.816E-15 5.377E-15 4.971E-15 4.595E-15 4.246E-15 3.924E-15 3.626E-15 3.350E-15
7.8 3.095E-15 2.859E-15 2.641E-15 2.439E-15 2.253E-15 2.080E-15 1.921E-15 1.773E-15 1.637E-15 1.511E-15
7.9 1.395E-15 1.287E-15 1.188E-15 1.096E-15 1.011E-15 9.326E-16 8.602E-16 7.934E-16 7.317E-16 6.747E-16
8.0 6.221E-16 5.735E-16 5.287E-16 4.874E-16 4.492E-16 4.140E-16 3.815E-16 3.515E-16 3.238E-16 2.983E-16
8.1 2.748E-16 2.531E-16 2.331E-16 2.146E-16 1.976E-16 1.820E-16 1.675E-16 1.542E-16 1.419E-16 1.306E-16
8.2 1.202E-16 1.106E-16 1.018E-16 9.361E-17 8.611E-17 7.920E-17 7.284E-17 6.698E-17 6.159E-17 5.662E-17
8.3 5.206E-17 4.785E-17 4.398E-17 4.042E-17 3.715E-17 3.413E-17 3.136E-17 2.881E-17 2.646E-17 2.431E-17
8.4 2.232E-17 2.050E-17 1.882E-17 1.728E-17 1.587E-17 1.457E-17 1.337E-17 1.227E-17 1.126E-17 1.033E-17
8.5 9.480E-18 8.697E-18 7.978E-18 7.317E-18 6.711E-18 6.154E-18 5.643E-18 5.174E-18 4.744E-18 4.348E-18
8.6 3.986E-18 3.653E-18 3.348E-18 3.068E-18 2.811E-18 2.575E-18 2.359E-18 2.161E-18 1.979E-18 1.812E-18
8.7 1.659E-18 1.519E-18 1.391E-18 1.273E-18 1.166E-18 1.067E-18 9.763E-19 8.933E-19 8.174E-19 7.478E-19
8.8 6.841E-19 6.257E-19 5.723E-19 5.234E-19 4.786E-19 4.376E-19 4.001E-19 3.657E-19 3.343E-19 3.055E-19
8.9 2.792E-19 2.552E-19 2.331E-19 2.130E-19 1.946E-19 1.777E-19 1.623E-19 1.483E-19 1.354E-19 1.236E-19
9.0 1.129E-19 1.030E-19 9.404E-20 8.584E-20 7.834E-20 7.148E-20 6.523E-20 5.951E-20 5.429E-20 4.952E-20
9.1 4.517E-20 4.119E-20 3.756E-20 3.425E-20 3.123E-20 2.847E-20 2.595E-20 2.365E-20 2.155E-20 1.964E-20
9.2 1.790E-20 1.631E-20 1.486E-20 1.353E-20 1.232E-20 1.122E-20 1.022E-20 9.307E-21 8.474E-21 7.714E-21
9.3 7.022E-21 6.392E-21 5.817E-21 5.294E-21 4.817E-21 4.382E-21 3.987E-21 3.627E-21 3.299E-21 3.000E-21
9.4 2.728E-21 2.481E-21 2.255E-21 2.050E-21 1.864E-21 1.694E-21 1.540E-21 1.399E-21 1.271E-21 1.155E-21
9.5 1.049E-21 9.533E-22 8.659E-22 7.864E-22 7.142E-22 6.485E-22 5.888E-22 5.345E-22 4.852E-22 4.404E-22
9.6 3.997E-22 3.627E-22 3.292E-22 2.986E-22 2.709E-22 2.458E-22 2.229E-22 2.022E-22 1.834E-22 1.663E-22
9.7 1.507E-22 1.367E-22 1.239E-22 1.123E-22 1.018E-22 9.223E-23 8.358E-23 7.573E-23 6.861E-23 6.215E-23
9.8 5.629E-23 5.098E-23 4.617E-23 4.181E-23 3.786E-23 3.427E-23 3.102E-23 2.808E-23 2.542E-23 2.300E-23
9.9 2.081E-23 1.883E-23 1.704E-23 1.541E-23 1.394E-23 1.261E-23 1.140E-23 1.031E-23 9.323E-24 8.429E-24
Appendix C
PROOFS
C.1 Purpose
Students are not required to understand the material in this Appendix. It has been
provided for the benet of those who are curious about the proofs of some of the results
which were stated without proof in the notes.
C.2 Bounds on the Autocorrelation Function
We suppose that we have real random processes X(t) and Y (t) which are not necessarily
stationary. Since the square of a real function is nonnegative, it follows that
0 E
X(t
1
)
_
E{(X(t
1
)
2
}

Y (t
2
)
_
E{(Y (t
2
)
2
}
(C.1)
On expanding, we can obtain the result
|R
xy
(t
1
, t
2
)|
_
E{X(t
1
)
2
}
_
E{Y (t
2
)
2
}. (C.2)
Similarly
|R
xx
(t
1
, t
2
)|
_
E{X(t
1
)
2
}
_
E{X(t
2
)
2
}. (C.3)
For stationary random processes these inequalities become
|R
xy
()|
_
R
xx
(0)
_
R
yy
(0) (C.4)
and
|R
xx
()| R
xx
(0) (C.5)
It is the second of these results which was asserted in equation 6.4.
201
202 APPENDIX C. PROOFS
C.3 The Wiener Khinchine Theorem
The Wiener Khinchine theorem states that the autocorrelation function and the power
spectral density are Fourier transform pairs. The theorem automatically raises the ques-
tion of what is the power spectral density.
We have a clear, empirically-based belief that the concept of a power spectral density
should exist. The situation of a random process emitting a signal for all time, and that
signal being passed through a band pass lter, and the power in the output signal being
observed on some average basis, and the lter being tuned to dierent frequencies, suggests
an empirical context in whch we can give the concept of power spectral density a meaning.
Although we recognise that power spectral density should have meaning, we cannot
base our concept of power spectral density on the Fourier transform X(f) of a time wave
form X(t). This is for the two dierent but compelling reasons stated below.
We recall that Fourier transforms exist for functions of nite energy, and the time
waveform of many ranndom processes may not have nite energy.
Even in the situation where X(t) represents the voltage across a 1 resistor so
that X(t)
2
represents the instantaneous power, the quadratic function X(f)X
(f),
which we might be tempted to adopt for the power spectral density, has the wrong
units: it has an extra dimension of time.
To overcome these diculties, we introduce for each sample function X(t) of the
process the variable X
T
(t) which has the same value as X(t) in the interval T < t < T,
and is zero outside that range, and we denote its Fourier transform by X
T
(f), so that
X
T
(f) =
_
T
T
X
T
(t)e
j2ft
dt =
_
T
T
X(t)e
j2ft
dt (C.6)
By the Rayleigh energy theorem, the energy E(T) contained in the signal may be
written in the two forms below.
E(T) =
_
T
T
X
2
T
(t)dt =
_

|X
T
(f)|
2
df (C.7)
We note that the integration over time is from T to T, while the integration over
frequency is from to , but involves the Fourier transform of the time-truncated
sample function.
We now dene the power spectral density S
XX
(f) as
S
XX
(f) =
lim
t
E[|X
T
(f)|
2
]
2T
(C.8)
Next we substitute for X
T
(f) and its conjugate their Fourier transforms, and obtain
S
XX
(f) =
lim
t
E
_
1
2T
_
T
T
X(t
1
)e
j2ft
1
dt
1
_
T
T
X(t
2
)e
j2ft
2
dt
2
_
(C.9)
=
lim
t
1
2T
_
T
T
_
T
T
E[X(t
1
)X(t
2
)e
j2f(t
1
t
2
)
dt
2
dt
1
(C.10)
C.3. THE WIENER KHINCHINE THEOREM 203
When t
1
and t
2
lie in the range from T to T, the expectation in the integrand of
the second equation is identied as the autocorrelation function of X(t), so this equation
becomes
S
XX
(f) =
lim
t
1
2T
_
T
T
_
T
T
R
XX
(t
1
, t
2
)e
j2f(t
1
t
2
)
dt
2
dt
1
(C.11)
We now have a relation between the power spectral density and the autocorrelation
function. What we wish to show relates to the inverse Fourier transform of the power
spectral density. The inverse transform, to the domain, is
_

S
XX
(f)e
j2f
df (C.12)
=
_

lim
t
1
2T
_
T
T
_
T
T
R
XX
(t
1
, t
2
)e
j2f(t
1
t
2
)
dt
2
dt
1
e
j2f
df (C.13)
=
lim
t
1
2T
_
T
T
_
T
T
R
XX
(t
1
, t
2
)
_

e
j2f(+t
1
t
2
)
dfdt
2
dt
1
(C.14)
We can recognise the inner integral as the inverse Fourier transform of a delta function,
so
_

S
XX
(f)e
j2f
df =
lim
t
1
2T
_
T
T
_
T
T
R
XX
(t
1
, t
2
)( +t
1
t
2
)dt
2
dt
1
(C.15)
The properties of the delta function allow the immediate evaluation of the integral
over t
2
, so
_

S
XX
(f)e
j2f
df =
lim
t
1
2T
_
T
T
_
T
T
R
XX
(t
1
, +t
1
)dt
1
(C.16)
This expression is still subject to the condition that the times are within the range T
to T. In the limit when T this restriction has little meaning. What we see on the
right hand side of the above equation is the time average of the processs autocorrelation
function.
In the important case when the process is stationary, the time average is just the
autocorrelation function R
XX
(), so we recognise that S
XX
(f) and R
XX
() are Fourier
transform pairs.
S
XX
(f) =
_

R
XX
()e
j2f
d (C.17)
R
XX
() =
_

S
XX
()e
j2f
df (C.18)
These relations are called, in celebration of the work of N. Wiener and A. I Khinchine,
the Wiener-Khinchine relations. They form the basic link between the description of
random processes in the time domain by means of autocorrelation functions, and their
description in the frequency domain by means of their power spectrum.
From the above we can see that a knowledge of the power spectrum of a random process
allows complete recovery of the autocorrelation function when the process is stationary.
When the process is non-stationary, only the time average of the autocorrelation function
is recoverable.
C.4 Analysis of the Bessel-Thompson Filter
Assume we have H(s
n
) = K/G(s
n
) where G(s
n
) is a polynomial of degree n in s
n
.
H(s
n
) =
K
G(s
n
)
G(s
n
) =
n
i=0
g
i
s
i
n
(C.19)
This will be an all pole lter.
H(j) = A() e
j()
(C.20)
() = Im{ln H(j)} (C.21)
= Im{ln G(j)} (C.22)
d()
d
= Im
_
jG
(j)
G(j)
_
(C.23)
where the dash denotes the derivative with respect to s and Im is the imaginary part.
The derivative of () with respect to is called the normalised group delay
g
(j).
Writing this in terms of s
n
= j and using G
(j) = G(s
n
) gives
g
(s
n
) =
1
2
_
G
(s
n
)
G(s
n
)
+
G
(s
n
)
G(s
n
)
_
(C.24)
=
N(s
n
)
2G(s
n
)G(s
n
)
(C.25)
where N(s
n
) = G(s
n
)G
(s
n
) + G(s
n
)G
(s
n
) is an even polynomial in s
n
of degree
2n 2.
Since () is an odd function of , its derivative must be an even function, and hence
g
(s
n
) is an even function of s
n
.
Suppose that the desired value of
g
(s
n
) is 1. Then we can write
g
(s
n
) = 1 +
P(s
n
)
G(s
n
)G(s
n
)
(C.26)
where P(s
n
) will be an even polynomial of degree 2n with the coecient of s
2n
n
equal
to ()
n+1
g
2
n
.
To have as many derivatives as possible of
g
(s
n
) at s
n
= 0 to be zero requires that
P(s
n
) = ()
n+1
g
2
n
s
2n
.
Equating the two expressions for
g
(s
n
) gives
G
(s
n
)G(s
n
) +G
(s
n
)G(s
n
) = 2G(s
n
)G(s n) + 2()
n+1
g
2
n
s
2n
n
(C.27)
Multiplying both sides by s
2n
n
and forming the derivative gives (after some algebra)
Ev{ [s
n
G(s
n
) 2(s
n
+n)G
(s
n
) + 2nG(s
n
)] G(s
n
) } = 0 (C.28)
where Ev means Even Part. (The even part of any function f(s) is (f(s) +
f(s))/2.) This means that the expression in { } must be either an odd polynomial
C.4. ANALYSIS OF THE BESSEL-THOMPSON FILTER 205
(no even powers of s
n
) or identically zero. An odd polynomial must have at least one root
at s
n
= 0 and the remainder in plus and minus pairs (i.e. if s
no
is a zero then so is s
no
).
Since G(s
n
) has n zeros in the right half s
n
plane, there must be n zeros in the left
half s
n
plane plus a zero at s
n
= 0 which can only come from the term in [ ]. Since this
polynomial is of degree n, this is not possible, so the term in [ ] is identically zero.
s
n
G
(s
n
) 2(s
n
+n)G
(s
n
) + 2nG(s
n
) = 0. (C.29)
The solution of this equation leads to polynomials called the Bessel polynomials. The
solutions (with g
n
= 1) are shown below, and the transfer function of the lter is
H(s
n
) =
g
0
G(s
n
)
. (C.30)
The coecients can be shown to satisfy a recurrence relation
g
i+1
= g
i
2(n i)
(2n i)(i + 1)
. (C.31)
This gives
G
0
(s
n
) = 1 (C.32)
G
1
(s
n
) = s
n
+ 1 (C.33)
G
2
(s
n
) = s
2
n
+ 3s
n
+ 3 (C.34)
G
3
(s
n
) = s
3
n
+ 6s
2
n
+ 15s
n
+ 15 (C.35)
G
4
(s
n
) = s
4
n
+ 10s
3
n
+ 45s
2
n
+ 105s
n
+ 105 (C.36)
G
5
(s
n
) = s
5
n
+ 15s
4
n
+ 105s
3
n
+ 420s
2
n
+ 945s
n
+ 945 (C.37)
G
n
(s
n
) = (2n 1)G
n1
(s
n
) +s
2
n
G
n2
(s
n
) (C.38)
The zeros of these polynomials, which correspond to the poles of H(s
n
), are given in
the table below.
Order Poles of H(s
n
)
1 1.00000
2 1.50000 j0.86603
3 2.32219
1.83891 j1.75438
4 2.89621 j0.86723
2.10379 j2.65742
5 3.64674
3.35196 j1.74266
2.32467 j3.57102
C.4. ANALYSIS OF THE BESSEL-THOMPSON FILTER 207
Appendix D
ADVICE ON STUDYING FOR
THE EXAMINATION
Attend all lectures.
Make notes during a lecture when no notes, or only brief notes, are supplied.
Remember that the process of note-taking engages more of your attention than does
passive listening.
Do all homework exercises.
Attempt to solve all tutorial problems.
Attend all tutorials to validate your answers.
Work through any previous examinations available.
Be aware that learning takes place when you yourself perform the set exercises, or
you yourself attempt to solve problems.
Be aware that you learn very little from reading someone elses solutions to problems
that you have not yourself attempted.
So reading a solution before really trying hard to solve a problem is very unwise.
Remember the Solemn Promise About Examinations made to all students by the
lecturer. If you have not heard it, please ask.
Be aware that for the year 2003 signicant changes have been made in the lecture
notes and the notation contained within them from that used in related courses
delivered in previous years, and such changes are reected in the examination ques-
tions.
It follows that, even if you have studied a related course before, you should obtain
and study the the lecture notes for this course.
209
210 APPENDIX D. ADVICE ON STUDYING FOR THE EXAMINATION
211
212 APPENDIX D. ADVICE ON STUDYING FOR THE EXAMINATION
Appendix E
NOTES ON STUDENT ERRORS
This Appendix contains notes on student errors which have been noted in the marking of
assignments and examinations for related courses delivered in the past. It is hoped that
it will serve to minimise the occurrence of such errors in the future.
E.1 Common Errors in Quiz 1
Almost always, and amazingly, a zero value was given for the value at time t = 0 of
a sinusoid represented by real, albeit negative, phasor.
There was often a failure to supply the units of the answer. Many marks were lost
in this way.
Sometimes a colon was incorrrectly supplied in place of a decimal point.
Sometimes an instantaneous value was given the units of V r.m.s. This betrays a
serious misunderstanding of the concept of r.m.s.
Sometimes a negative value was given for an r.m.s. quantity. This betrays another
serious misunderstanding; they should always be positive.
Sometimes peak volts was given as the units of an instantaneous value. This betrays
a misunderstanding of the concept of an instantaneous value.
The units of power were sometimes quoted as ohm or joule; the units of current were
sometimes given as volt; the units of magnetic eld were sometimes given as henry;
the units of inductance were sometimes given as farad; and the units of impedance
were sometimes give as Z. There is scope for improvement here.
A negative value was sometimes given as available source power. It is hard to
imagine how this can be.
Sometimes a plural was denoted by an apostrophe s. This is a misuse of the English
language.
213
214 APPENDIX E. NOTES ON STUDENT ERRORS
E.1. COMMON ERRORS IN QUIZ 1 215
216 APPENDIX E. NOTES ON STUDENT ERRORS
Appendix F
STANDARD INTERNATIONAL
TERMINOLOGY AND UNITS
F.1 Objective
This Appendix was originally written as part of a course in elementary electrodynamics.
It is provided here because the SI units for important common quantities seem to be still
not known by all students.
It was the original objective of this Appendix to make available to students up to date
information on those parts of the Standard International system of units which should be
used for study in the area of electodynamics.
F.2 Informal discussion
Students are hopefully familiar with the usage of Standard International units in the
subjects of mechanics and dynamics, although perhps not with recent changes in the way
the fundamental quantities of mass, length and time are dened. Although it might appear
that those fundamental quantities should be denable in a way which is independent of
electrodynamic theory, the modern approach to those denitions requires for its proper
undersanding a knowlege of the fundamenal principles of electodynamics, and in particular
of the properties of electromagnetic waves and the relation of their velocity to fundamenal
physical constants.
During the course we have encountered the fundamental physical constants
0
which
appeared in the law of force between electric charges and
0
which appeared in the law
of force between current carrying wires. What we did not have the opportunity to do,
because of the limited scope of the course, was to establish that the values of these
constants are not indepedent, but form part of a basic relation which links themselves and
the value of the velocity of light.
In the further development of the subject of electrodynamics it may be shown that
the magnetic permeability of free space, the dielectric permittivity of free space and the
velocity of light in free space are related by the equation
c
2
0
= 1 (F.1)
In the past,
0
was always given the dened value 4 10
7
H/m in the process
217
218 APPENDIX F. STANDARD INTERNATIONAL TERMINOLOGY AND UNITS
of setting up the Standard International system of units. On the assumption that the
units of mass, length and time were already established, and hence the unit of force
was established, the action of giving
0
this dened value in eect established the unit
of current. The velocy of light was then an experimentally determined value, and the
dielectric permittiviy of free space became a calulated value based in part on a dened
value and in part on an experimentally determined value, the calculation being based
upon equation F.1 above.
In the modern approach, the unit of time (the second) is dened in terms of the
frequency of oscillation of an atomic clock based on a particular form of radiation. Instead,
however, of dening, as was done in the past, the unit of distance (the metre) as the
distance between two marks on the internationl prototype metre maintained in a controlled
environment, the metre has been re-dened in terms of the distance traveled by light in a
particular time. Such a dention requires that the velocity of light be known. It becomes
known by giving it the dened value of 299,979,245.8 m/s. What has happened is that
the length of the meter has been adjusted, (hopefully not by much), so that the velocity
of light now has exactly this value.
Among the electrodynamic constants, the value of
0
is still given the dened value
4 10
7
H/m. Equation F.1 is regarded as being rmly established by fundamental
physical law as providing a relation between
0
,
0
and the velocity of light c. As both c
and
0
have now received dened values, so now also has
0
. Its value is exactly
0
=
1
4 10
7
(299, 979, 245.8)
2
(F.2)
Its value is quoted to a convenient number of decimal places later in this Appendix.
In the modern system, the unit of mass (the kg) is still dened as the mass of the
international prototype kilogram maintained in a controlled environment, and from which
other standard kilogram masses can be copied and distributed.
As the units of mass, length and time are now established, then the unit of force is
establised through the hopefully well known equations of dynamics.
The unit of current is then established through the equation describing the force
between current carrying conductors, and reproduced for convenience below.
Force per unit
length of one wire
=

0
I
1
I
2
2s
(F.3)
and hence the unit of charge (the coulomb) is established as the charge which a current
of one ampere would deliver in one second.
Although for economy of expression and to celebrate the work of eminent scientists,
particular names have been given to the remainder of the electrodynamc quantities dis-
cussed in this course, they can all be reduced to alternative terms involving the fundamen-
tal units of mass, length, time and current which are dened, among others, as belonging
in the set of base units listed in Table F.1 below, and hence the meaning of all such
quantites is clearly established.
F.3 The Base Units
We provide in Table F.1 the formal denitions of the base units of the SI system.
F.3. THE BASE UNITS 219
BASE UNITS
QUANTITY NAME SYMBOL DEFINITION OF UNIT
length metre m
The metre is the length of the path travelled
by light in vacuum during a time interval of
1/299 792 458 of a second. (What this does
is give the velocity of light a dened value of
299 979 245.8 m/s.)
mass kilogram kg
The kilogram is the unit of mass; it is equal
to the mass of the international protype of
the kilogram.
time second s
The second is the duration of 9 192 631 770
periods of the radiation corresponding to the
transition between the two hyperne levels of
the ground state of the caesium-133 atom.
electric current
ampere A
The ampere is that constant current which,
if maintained in two straight parallel conduc-
tors of innite length, of negligible circular
cross-section, and placed one metre apart in
a vacuum, would produce between these con-
ductors a force equal to 2 10
7
newton per
metre of length.
thermodynamic
temperature
kelvin K
The kelvin unit of thermodynamic tempera-
ture, is the fraction 1/273.16 of the thermo-
dynamic temperature of the triple point of
water.
amount of sub-
stance
mole mol
The mole is the amount of substance which
contains as many elementary entities as there
are atoms in 0.012 kilograms of carbon-12.
When the mole is used, the elementary en-
tities must be specied and may be atoms,
molecules, ions, electrons, other particles, or
specied groups of such particles.
luminous inten-
sity
candela cd
The candela is the luminous intensity, in a
given direction, of a source that emits mono-
chromatic radiation of frequency 540 10
12
hertz and that has a radiant intensity in that
direction of 1/683 watt per steradian.
Table F.1: The base units of the SI system.
F.4 Abbreviations
We provide in Table F.2 a summary of the names and standard abbreviations for com-
monly used electrodynamic variables.
Full name Abbreviation
ampere A
coulomb C
farad F
henry H
joule J
metre m
newton N
ohm
siemen S
tesla T
volt V
watt W
weber Wb
Table F.2: Full and abbreviated forms of SI units used for electrodynamic variables.
F.5 The Fundamental Vectors
1. The names, usual symbols, and standard international units for the four vectors
used to describe electromagnetic leds in a general medium are
E electric eld intensity Vm
1
H magnetic eld intensity Am
1
D electric ux density Cm
2
B magnetic ux density Wbm
2
The units of magnetic ux density B have the alternative name of Tesla, for which
the abbreviation is T.
2. The names, usual symbols, and standard international units for the vectors used
to describe the state of a dielectric medium and the state of a magnetic medium
respectively are
P polarisation Cm
2
M magnetisation Am
1
3. The fully general relations between the above six vectors are
(i) D =
0
E +P by denition
(ii) B =
0
(H+M) by denition
F.6. LIST OF STANDARD SYMBOLS 221
4. The values of the magnetic permeability and the dielectric permittivity of free space
are in standard international units
(i)
0
4 10
7
Hm
1
by denition
(ii)
0
8.854 10
12
Fm
1
approximately
In the most recent denition of the standard international system of units, both of
the above values have become, in eect, as explained above, dened values.
F.6 List of Standard Symbols
We provide in Table F.3 a summary of the names and SI units used for the variables
employed in these notes.
VARIABLE SYMBOL SI UNITS
Area A m
2
Capacitance C F
Charge Q C
Conductance G S
Conductivity S/m
Current I A
Dielectric permittivity F/m
Dielectric susceptibility
e
None
Distance l, r m
Electric dipole moment p Cm
Electric eld intensity E V/m
Electric ux C
Electric ux density D Cm
2
Energy W J
Force F N
Inductance L H
Linear charge density q
l
Cm
1
Magnetic eld intensity H A/m
Magnetic ux Wb
Magnetic ux density B Wbm
2
or T
Magnetic moment m Am
2
Magnetic permeability H/m
Magnetic susceptibility
m
None
Magnetisation M A/m
Mobility m
2
Vs
Polarisation P Cm
2
Potential V V
Power P W
Resistance R
Surface charge density q
s
Cm
2
Surface current density J
s
Am
2
Torque T Nm
Velocity v m/s
Volume v m
3
Volume current density J
v
Am
2
Volume charge density q
v
or
v
Cm
3
Work W J
Table F.3: Names symbols and units for electrodynamic variables.
F.6. LIST OF STANDARD SYMBOLS 223
Appendix G
STOP PRESS
This appendix contains late breaking news and additional material which may be useful
in understanding the course.
G.1 Thevenins and Nortons Theorems
G.1.1 Motivation
Thevenins and Nortons theorems have been continuously useful to the lecturer through-
out his professional career in providing a simple understanding complex networks. It is
believed that a careful enunciation, in compact language, of the theorems and the con-
ditions under which they apply can capture all of the important aspects of the theorems
including their range of applicability. Knowing all of these aspects is essential for their
correct application. Such a compact and compete enunciation is provided in the next two
sections.
G.1.2 Thevenins theorem
The external circuit behaviour of a network of linear circuit elements at a pair of terminals
A,B can be modelled as that produced by an equivalent circuit consisting of an ideal
voltage generator V in series with an impedance Z.
The voltage V of the ideal voltage generator is the voltage appearing at the terminals
A,B when the are terminated in an open circuit.
The impedance Z is the impedance seen looking back into the network at terminals
A,B when all of the internal sources are reduced to zero.
G.1.3 Nortons theorem
The external circuit behaviour of a network of linear circuit elements at a pair of terminals
A,B can be modelled as that produced by an equivalent circuit consisting of an ideal
current generator I in parallel with an impedance Z.
The current I of the ideal current generator is the current owing externally between
the terminals A,B when the are terminated in a short circuit.
The impedance Z is the impedance seen looking back into the network at terminals
A,B when all of the internal sources are reduced to zero.
225
226 APPENDIX G. STOP PRESS
G.1.4 Relations between the circuits
Note that the same impedance Z occurs in both circuits. Note also that since the circuits
are both equivalent to the same original circuit they must be, as far as external circuit
behaviour is concerned, be equivalent to one another, and hence
V = ZI (G.1)
G.1.5 Commentary
The theorems have been deliberately worded so that they may apply, with an appropriate
interpretation of notation, to two cases, viz.
a network of d.c. sources and resistors; or
a network of sinusoidal steady state a.c. sources, resistors, inductors and capacitors.
G.1.6 Cautions
Note that other forms of time varying signals are excluded. Note also the requirement
that the circuit elements be linear in their voltage current relation.
Note that an ideal voltage generator has only a Thevenin circuit, and an ideal current
generator has only a Norton circuit. (Why?)
Note that the polarity of the sources is important, although it is not mentioned ex-
plicitly in the enunciation. This is because stating the polarity of a voltage or current
source is an essential part of its denition.
G.1. THEVENINS AND NORTONS THEOREMS 227
228 APPENDIX G. STOP PRESS
Appendix H
SOME NOTES ON PHASOR
ANALYSIS
H.1 Objective
The objective of these notes is to clarify the basic concepts of phasor analysis of AC
circuits in the sinusoidal steady state.
H.2 A Simple Circuit
Our vehicle for this instruction will be the analysis of the simple circuit shown in the
Figure H.1.
L
i(t)
L
v (t)
+
v(t)
+
_
_
R
R
v (t)
+
_
Figure H.1: A simple circuit for analysis.
In analysing this circuit, we will assume the sinusoidal current with peak value I
m
and
phase and .
i(t) = I
m
cos(t + ). (H.1)
H.3 Analysis by Dierential Equations
For the basic circuit elements R, L, and C, we can write dierential equations relating
the terminal voltage and the element current, using reference directions for voltages and
currents as shown in Figure H.2.
229
230 APPENDIX H. SOME NOTES ON PHASOR ANALYSIS
R
i(t)
v(t)
+
_
v(t) = R i(t)
L
i(t)
v(t)
+
_
di(t)
dt
v(t) = L
i(t)
v(t)
+
_
C
dv(t)
dt
i(t) = C
Figure H.2: Sign conventions for circuit variables.
For a resistor R
v(t) = Ri(t) = RI
m
cos(t + ). (H.2)
For an inductor of self inductance L,
v(t) = L
di(t)
dt
= LI
m
sin(t + ). (H.3)
For a capacitor of capacitance C,
v(t) =
1
C
_
i(t)dt =
1
C
I
m
sin(t + ). (H.4)
Thus for the circuit of Figure H.1 we have
v(t) = Ri(t) +L
di(t)
dt
(H.5)
= RI
m
cos(t + ) LI
m
sin(t + ) (H.6)
= I
m
_
R
2
+ (L)
2
sin(t + + ) (H.7)
where
sin =
L
_
R
2
+ (L)
2
and cos =
R
_
R
2
+ (L)
2
. (H.8)
H.4 Disadvantageous Features
Most people nd this analysis burdensome, particularly in the last step where a sine wave
and cosine wave of excitation must be combined.
We will show in the below sections that analysis using complex phasors provides the
correct results with less trouble.
Although we achieve this result by introducing complex numbers to represent the
sinusoidal variables, all the voltages and currents within our circuit remain real numbers.
H.5. REPRESENTATION OF SINUSOIDS 231
H.5 Representation of Sinusoids
It is well known that the circular functions sin(t) and cos(t) bear simple relations to
the complex exponential e
jt
. It is also well known that the complex exponential e
jt
has a particulary simple time derivative i.e. je
jt
, in which the functional form of the
variable does not change; it is merely multiplied by a constant.
This prompts us to investigate whether complex exponentials can be used to provide
a simpler version of AC circuit analysis than that just illustrated.
We rst introdce the concept of phasor to represent a sine wave (or more correctly a
cosine wave) excitation.
In doing this, we rst observe that, in a situation in which all variables are sinusoidal
and at a common angular frequency , the sinusoidal current i(t) = I
m
cos(t + ) is
completely determined by its magnitude I
m
and its phase .
We may therefore represent that magnitude and phase information by a complex
number of magnitude I
m
and argument , i.e. the complex number I
m
e
j
as shown in the
Argand diagram on Figure H.3 at the point P
1
.
imag
real
f
wt
Instantaneous value at time t = 0
Instantaneous value at time t
Rotation
Phasor (fixed) Rotating arm
O
I
m
I
m
P
1
2
P
Figure H.3: Phasors, rotating arms and projections.
We notice that the point P
1
does not move. Moreover, if we take the projection of the
point P
1
on the horizontal axis, we obtain the value I
m
cos , which is the value of the
sinusoidal excitation at the time t = 0.
If we add an additional time-varying phase angle t to the phase angle of the phasor
leading to the point P
1
, we obtaine the rotating arm in which the point P
2
describes a
circle centred on the origin and of radius I
m
. At any time, the projection of the point P
2
on
the horizontal axis gives the instantaneous value of the sinusoidal quantity I
m
cos(t +)
at time t. We notice that the horizontal projection is just the real part of the complex
number represented by the rotating arm.
H.6 Analysis Leading to Algebraic Equations
Let us now investigate the way in which taking our variables as the real parts of complex
exponentials may simplify the work involved in circuit analysis when sinusoidal signals
are present. We begin with
i(t) = {I
m
e
j(t+)
}. (H.9)
Then
di(t)
dt
=
d{I
m
e
j(t+)
}
dt
(H.10)
=
dI
m
cos(t + )
dt
(H.11)
= sin(t + ) (H.12)
= {je
j(t+)
}. (H.13)
So we see that we may achieve the object of dierentiation, which is initially to be
done externally to the operation of taking the real part, by performing the dierentiation
rst on the complex function, and then the taking the real part of the result. This is
actually a quite general result which applies to complex functions of a real variable when
we dierentiat with respect to that real variable. What is, however, interesting for us
here is rstly that the dierentiation operation has the simple eect of multiplying the
complex function by j, and secondly that there is no other change in the functional form
of the complex function.
H.7 Application to Circuit Analysis
So in our analysis of the simple circuit given in Figure H.1 with the sinusoidal excitation
i(t) = {I
m
e
j(t+)
} (H.14)
we may derive the input voltage by the steps
v(t) = {RI
m
e
j(t+)
+jLe
j(t+)
} (H.15)
= {(R +jL)I
m
e
j(t+)
}. (H.16)
H.8 The Physical and Mathematical Systems
These steps lead to the following interpretation, in which it is useful to distinguish between
a physical system and a mathematical system.
H.8. THE PHYSICAL AND MATHEMATICAL SYSTEMS 233
H.8.1 The physical system
In the physical system, we have an excitation current and the resulting voltage which are
both real and sinusoidal, but are expressed as the real part of complex exponentials, as
below.
i(t) = {I
m
e
j(t+
} ; v(t) = {(R +jL)I
m
e
j(t+)
}. (H.17)
It is very evident that the voltage and current are real, as they must be.
H.8.2 The mathematical system
In the mathematical system we might treat our input current as the complex function
I
m
e
j(t+)
(H.18)
and treat resistors, inductors, and capacitors as having complex impedances of R,
jL and 1/(jC) resepctively, and calculate voltages, (and in more complicated circuits
maybe other currents), by the same rules of circuit theory as we have been used to in DC
circuits, but where the resistances are replaced by the complex impedances dened in the
equations above. To nd the answer to the physical problem, we take of the real parts of
the complex functions. Thus
i(t) = {I
m
e
j(t+)
} (H.19)
and
v(t) = {V e
(jt)
} (H.20)
= {(R +jL)I
m
e
j
e
jt
}. (H.21)
H.8.3 Deletion of the time function
We notice that the time function e
jt
is a common factor of all of the above operations.
Thus we can delete it from from the function I
m
e
j(t+)
at the beginning of the calculation,
to obtain the phasor I
m
e
j
. Then we operate on that phasor by our complex impedances,
(in our case we multiply by our complex impedance (R+jL), to obtain our input phasor
V = (R + jL)I
m
e
j
. We then re-attach the common factor e
jt
, and take the real part
of the result, to get the real solution to the physical problem.
v(t) = {V e
jt
} (H.22)
= {(R +jL)I
m
e
j
e
jt
}. (H.23)
H.8. THE PHYSICAL AND MATHEMATICAL SYSTEMS 235
Appendix I
MODULATION SYSTEMS
I.1 Acknowledgment
This appendix contains notes assembled from the Text Book by A. Bruce Carlson Com-
munication systems, McGraw Hill, 1975. For the benet of students who wish to extend
their studies through the use of that text book, the notation of that book is adopted here.
I.2 Introduction and Objective
These notes provide an introduction to and analysis of the performance of some mod-
ulation methods used in communication systems. The material provides a justication
for the use of modulation, give examples of circuits by means of which modulation and
subsequent detection can be achieved, and a compares the performance of communication
systems using various forms of continuous wave modulation, i.e. systems in which the
un-modulated carrier is a sinusoid.
I.3 Modulation Systems
I.3.1 Denition of modulation
Modulation may be dened as the systematic alteration of one waveform, called the
carrier, according to the characteristics of another waveform, which we call the modulating
signal or the message. Normally the modulating signal or message is denoted by x(t),
which is a dimensionless variable, and is subject to |x(t)| < 1. The message is assumed
to be contained within a bandwidth W.
Sometimes the modulating signal is the single tone dened by
x(t) = A
m
cos 2f
m
t with A
m
< 1 and f
m
< W. (I.1)
For such a tone, the double sided Fourier spectrum is given by
X(f) =
A
m
2
[(f +f
m
) + (f f
m
)] (I.2)
and is illustrated in Figure I.1.
237
238 APPENDIX I. MODULATION SYSTEMS
Fourier spectrum for a single tone message.
For this figure you could consult Figure 7.3 of
Peter Cookes book
Figure I.1: Fourier spectrum for a single tone message.
I.3.2 Classes of modulation
We will recognise in our study the classes of linear modulation, in which the envelope of
the carrier bears a simple and linear relation to the message, and so-called exponential
modulation in which the phase or the frequency of the carrier is varied in response to the
message.
I.4 Bandpass Signals and Systems
A bandpass signal is one whose spectrum is concentrated in the vicinity of the carrier
frequency. It appears as a sinusoid at the carrier frequency with slowly changing amplitude
and phase, as shown in the equation below.
v(t) = R(t)[cos(
c
t + (t))] with R(t) 0 (I.3)
Another way of writing this signal is in terms of slowly varying in-phase and quadrature
components as shown below.
v(t) = v
i
(t) cos
c
t v
q
(t) sin
c
t (I.4)
This last description has the advantage is that it is more readily transformed into the
frequency domain than is the amplitude and phase description.
I.5 Linear modulation systems
We study rst the class of linear modulation systems in which the envelope of the signal is
linearly related to the message. The simplest of these is amplitude modulation as described
below.
I.5.1 Amplitude Modulation
In amplitude modulation, the modulated carrier wave takes the form
x
c
(t) = A
c
(t)[1 +mx(t)] cos
c
t (I.5)
The envelope of the signal is therefore 1+mx(t). The envelope, relative to the carrier,
follows the message signal x(t).
I.5. LINEAR MODULATION SYSTEMS 239
Normally m 1 to avoid envelope distortion and phase reversals in the carrier.
The term linear modulation comes from the fact that an easily recognisable aspect of
the signal, namely the envelope, is a linear function of the message. An illustration of an
amplitude modulated signal is given in Figure ??.
Illustration of simple amplitude modulated signal.
Peter Cookes book
Figure I.2: Illustration of a simple amplitude modulated signal.
For amplitude modulation with a single tone, the double sided Fourier spectrum is
given by
X(f) =
A
m
2
[(f +f
m
) + (f f
m
)] (I.6)
and is illustrated in Figure I.3.
Fourier spectrum with amplitude modulation by a single tone.
Peter Cookes book
Figure I.3: Fourier spectrum with amplitude modulation by a single tone.
It is clear that part of the single tone spectrum, and indeed the general amplitude
modulation spectrum, on the negative frequency axis is completely predictable from the
part on the positive frequency axis.
The transmission bandwidth of the general amplitude modulated signal B
T
= 2W,
i.e. twice the bandwidth of the message signal. Amplitude modulation is sometimes said
to be wasteful of bandwidth in that the message could be sent at baseband (i.e. without
modulation) at half the transmission bandwidth.
The average power of an amplitude modulated signal is given by
S
T
= E{x
2
c
(t)} = (1 +m
2
x
2
)
A
2
c
2
(I.7)
from which we see that at least half the transmitted power resides in the carrier. It can
be said that this part of the carrier power is wasted as it is independent of the message,
and thus it carries no information.
I.5.2 AM circuits
A range of circuit arrangements can be used to accomplish amplitude modulation. One of
them, shown in the Figure I.4, uses a textitbalanced mixer to form a product between
a scaled version mx(t) of the message signal and the carrier wave A
c
cos
c
t, the product
then being added to the carrier wave to produce the modulated carrier wave x
c
(t).
x
c
(t) mx(t)
X
~
A cos t
c c
w
+
Figure I.4: Amplitude modulation using a balanced mixer.
Other ways of producing amplitude modulation exist. Some of them naturally produce
a modulation signal plus carrier without the carrier needing to be added.
The simplest way to recover the message from an amplitude modulated signal is to
employ the envelope detector shown Figure I.5.
+ +
_ _
v
in
v
out
R
1
R
2
C
1
C
2
Figure I.5: Circuit for envelope detection.
Although other forms of demodulation circuits, which deal with a wider range of
modulation methods, are available, this circuit is often preferred for simple amplitude
modulation because of its simplicity and its lack of dependence of knowledge of frequency
and phase of the original carrier signal.
I.5.3 Double sideband suppressed carrier modulation
In amplitude modulation the carrier signal is not considered to carry information, and, as
has been said before,, is said to cause the transmission of unnecessary power. Eliminating
the carrier term and setting of the modulation index to unity produces double sideband
suppressed carrier modulation as described in the equation below.
x
c
(t) = A
c
x(t) cos
c
t (I.8)
The spectrum bandwidth of double sideband suppressed carrier modulation is still
2W.
There is the question of whether the envelope is still of the same shape as the message,
so that envelope detection may be employed. The simple example of tone modulation
shown in Figure I.6 shows that it is not, so simple detection of the waveform of the
envelope which is possible in simple amplitude modulation will no longer recover the
Illustration of a double sideband suppressed carrier modulated signal.
Peter Cookes book
Figure I.6: Illustration of a double sideband suppressed carrier modulated signal.
message. Instead, there is a need for special detectors e.g. the synchronous detector of
Figure I.10, which is discussed later.
I.5.4 Single sideband modulation
We have already eliminated some wasted power from amplitude modulation by eliminat-
ing the carrier signal. Because both sidebands carry the same information, we can also
eliminate some wasted bandwidth by eliminating either the upper or lower sideband of
the signal. Such sideband suppression can be accomplished by means of a single sideband
lter as shown in Figure I.7.
x(t)
~
cos t w
c
DSB
SSB
Balanced
mixer
Sideband
filter
H(f)
Figure I.7: Single sideband modulation using sideband lter.
An alternative method of generation of a single sideband signal is shown in Figure I.8.
This circuit can produce either the upper sideband or the lower sideband, depending
on whether we add or subtract the outputs of the two balanced modulators. There is
again the question of whether the envelope bears a simple relation to the message, so that
envelope detection may be employed. We will not prove the result here, but it is may be
shown that the envelope of the double sideband suppressed carrier signal is given by
R
SSB
(t) =
A
c
2
_
_
x(t)
2
+

x(t)
2
_
(I.9)
where

x(t) is the Hilbert transform of x(t). As the Hilbert transform in general
produces considerable change in shape of a signal, it is not surprising that the envelope
of a single sideband suppressed carrier signal is no longer similar to that of the original
message.
Bearing in mind that changing the phase of all components of the message signal by
90 degrees at baseband is equivalent to taking its Hilbert transform, we can see that the
0.5x(t)
modulator
Balanced
modulator
Balanced
+
90
o
90
o
~
+
+_
SSB
A cos t
c
w
0.5x(t)A cos t
c
w
0.5x(t)A sin t
c
w
>
Figure I.8: Phase shift single sideband modulator.
single sideband generation process may be interpreted as adding two double sideband
signals having quadrature carriers, modulated respectively by the original message signal
and by its Hilbert transform.
I.5.5 Vestigial sideband modulation
Practical single sideband systems have a poor low frequency response but good bandwidth
eciency. In addition, it is dicult to achieve the sharp cut-o required to completely
suppress one of the two sidebands.
The double sideband systems have good low frequency response but the message band-
width is twice that of single sideband systems.
A compromise system which achieves some of the benets of each of these two systems
is the vestigial sideband (VSB) modulation system in which the sideband lter is as shown
in Figure I.9.
If vestigial sideband ltering is applied to double sideband suppressed carrier modula-
tion, we produce pure vestigial sideband modulation in which a carrier is not present. If
vestigial sideband ltering is applied to simple amplitude modulation, we produce vesti-
gial sideband plus carrier (VSB+C) modulation in which a carrier is present. Sometimes
additional carrier amplitude is added for reasons that it may permit simpler, although
approximate, detection by an envelope detector.
A illustration of vestigial sideband generation and procesing is provided in Figure I.9
The vestigial sideband modulation system, with envelope detection, does produce
appreciable signal distortion, but will transmit signals down to DC, and will transmit
high frequency signals with good bandwidth eciency. This system is normally applied
in contexts, such as television picture transmission, is where the distortion is not readily
noticed by the uses of the system, and good picture sharpness (associated with good high
frequency response) is valued.
(a)
(b)
(c)
f
X(f)
O
2f
c
-2f
c
f
X(f)
O
f
c
-f
c
f +W
c
-f -W
c
f
X(f)
O
W
W
-W
-W
Figure I.9: Illustration of vestigial sideband ltering.
I.5.6 Detection systems
Linear modulation can always be detected by the product detector of Figure I.10. However
the Figure does require the availability of a local oscillator at the same frequency and phase
as the carrier frequency, which may not have been transmitted with the signal, or may
have been transmitted only with low amplitude.
x
c
(t)
~
A cos t
LO
w
c
y (t)
D
LPF
B=W
x
Figure I.10: Synchronous detection of a modulated signal.
Thus the carrier may need to be regenerated in some way which we will not discuss
in detail here. This fact makes the product detector more complicated than the envelope
detector shown in Figure I.5.
It is interesting to examine what happens in the detection process in the case of
vestigial sideband modulation. Figure I.9 has already provided an illustration.
This diagram pertains to pure vestigial sideband modulation and synchronous detec-
tion, the latter as shown in Figure I.10. As the correct operation of the detector requires
a local oscillator which is synchronous and in phase with the transmitted carrier, and that
carrier is absent in the case of pure vestigial sideband modulation, the system of vestigial
sideband modulation with carrier is more often used. In that case it may be shown that
with an appropriate level of carrier, and tolerance of some distortion in the detection
process, an envelope detector, with its advantage of simplicity, may be used.
I.6 Receivers
In a receiver for modulated signal we need
Provision of frequency selectivity so that we receive the desired signal among many
that can enter the inputs to the receiver.
Amplication so that the received signal is at a suitable level for recovery of the
message. A suitable level is required because the recovery process often exploits
the nonlinearity of some component, and the appropriate nonlinearity is usually
available only over a suitable range of signal levels.
Usually, further amplication so that the output signal is increased from a level at
which the demodulation takes place, to the level desired at the destination.
The most common form of receiver is the superheterodyne receiver, described in the
next section. The principal benet of the superheterodyne receiver is that it provides the
appropriate degree of signal selectivity for us to be able to eciently manage a crowded
communication spectrum.
I.6.1 AM superheterodyne receiver
A block diagram of a superheterodyne receiver for amplitude modulation is shown in
Figure I.11.
Block diagram of a superheterodyne receiver.
Peter Cookes book
Figure I.11: Block diagram of a superheterodyne receiver.
The way in which the frequency is the selectivity of the superheterodyne receiver is
provided is illustrated in Figure I.12.
I.7 Angle modulation
We now turn to consider angle modulation, sometimes called exponential modulation,
because of the connection between sinusoids and exponential functions. In angle modu-
lation we may vary the phase or the frequency of the transmitted signal in accord with
the message.
I.7. ANGLE MODULATION 245
f
V
0
(f)
f
IF
B
T
O
2f
IF
f
V
0
(f)
f
LO
f -f
LO IF
f +f
LO IF
B
T
B
T
O
f
V
0
(f)
f -f
LO IF
B
T
O
f
V
0
(f)
f -f
LO IF
O
(a)
(b)
(c)
(d)
Figure I.12: Passbands in a superheterodyne receiver.
I.7.1 Phase modulation
In phase modulation with a message x(t) the modulated carrier is
x
c
(t) = A
c
cos[t +
x(t)] (I.10)
where
is called the phase-deviation constant, and is the maximum phase deviation

produced byx(t), the message being still subject to the message restriction |x(t)| 1.
In the above expression it is clear that the instantaneous phase of the modulated carrier
relative to the unmodulated carrier is
x(t).
A method of producing phase modulation is shown in Figure I.13.
x(t)
+
_
Balanced
mixer
f
D
90
o
NBPM
cos t w
c
sin t w
c
~
Figure I.13: Narrow band phase modulator using balanced mixer.
I.7.2 Frequency modulation
In frequency modulation it is the instantaneous frequency f of the modulating signal
which is varied in accord with the message, that is
f = f
x(t) (I.11)
where f
is called the frequency deviation constant. Since frequency is the time

derivative of phase, the modulated carrier in frequency modulation may be expressed as
x
c
(t) = A
c
cos[
c
t + 2f
_
t
x()d (I.12)
This expression indicates the connection between frequency modulation and phase
modulation and that they are both cases of angle modulation.
So that the integral does not diverge, it is assumed that the message signal has no
d.c. component. However, a message with a d.c. component can be made to make sense
in the frequency modulation context, provided we do not attempt a description of it in
phase modulation terms.
The appearance of the message term inside the argument of the cosine function pro-
duces quite profound dierences between angle modulation and amplitude modulation.
The most signicant of these lies in the spectrum of the modulated carrier. In amplitude
modulation, when the message contains a number of tones, the sideband produced by the
modulation process is the sum of the sidebands provided by each of the tones separately.
In angle modulation, this superposition behaviour does not occur. The sideband structure
of the modulated carrier which is angle modulated by a multitone message is much more
complex than the simple superposition would give. It is for this reason that our analysis
below will turn for a time to the study of single tone messages.
I.7.3 Spectral analysis
To provide a suitable basis for analysis, to be pursued later, of a frequency modulation
spectrum, we will consider the case of a phase modulated signal in which the message is
x(t) = A
m
cos
m
t (I.13)
rather than a cosine wave. So the modulated carrier becomes
x
c
(t) = A
c
cos[
c
t +
_
2f
m
_
A
m
sin
m
t] (I.14)
= A
c
cos(
c
t + sin
m
t) (I.15)
where is called the modulation index for phase modulation. It is the maximum phase
deviation for the tone of amplitude A
m
.
It may be shown that the Fourier spectrum of this signal is
x
c
(t) = A
c
n=
J
n
() cos(
c
+n
m
t) (I.16)
where J
n
() is of the Bessel function of the rst kind of order n and argument . The
Bessel functions up to order seven are illustrated in Figure I.14.
The Bessel functions of negative order may be derived from those for positive order
by the relation
J
n
() = (1)
n
J
n
() (I.17)
An illustration of the Fourier spectrum of this modulated signal is shown in Figure I.15.
We note in this gure that in accord with equation above the even order sidebands
have the same signs above and below the carrier, while the odd order sidebands have
opposite signs above and below the carrier.
An illustration of the magnitude of the spectrum such as shown in Figure I.16. This
gure does not show how the phases of the upper and lower sidebands are dierently
related for even and odd order sidebands, as was shown in Figure I.15.
In the case of a frequency modulated signal of the single tone and message x(t) =
A
m
cos[
m
t we have by performing the integration in equation ?? the carrier signal
x
c
(t) = A
c
cos
_
c
t +
2f
m
sin
m
t
_
. (I.18)
This is of the same form as was studied for phase modulation if we set
=
2f
A
m
m
=
A
m
f
f
m
(I.19)
~
cos t w
c
0 1 2 3 4 5 6 7 8 9 10
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Figure I.14: Bessel functions of various orders.
f
f
c
f f
c m
+ f 3f
c m
+
f 2f
c m
+
-J ( )
1
b
J ( )
1
b
J ( )
2
b
-J ( )
3
b
J ( )
3
b
J ( )
0
b
J ( )
2
b
X(f)
O
Figure I.15: Amplitude spectrum of an angle modulated signal.
f
f
c
f f
c m
+ f f
c m
- f 3f
c m
+ f 3f
c m
-
f 2f
c m
+ f -2f
c m
J ( )
1
b J ( )
1
b
J ( )
2
b
J ( )
3
b J ( )
3
b
J ( )
0
b
J ( )
2
b
|X|(f)
O
Figure I.16: Magnitude spectrum of an angle modulated signal.
The Fourier spectrum of the FM signal is again
x
c
(t) = A
c
n=
J
n
() cos(
c
+n
m
t) (I.20)
which is of the same form as was obtained for phase modulation with modulation
index .
Note that the two forms of modulation share the property that is the maximum
phase deviation for the tone of amplitude A
m
, but may be contrasted by the fact that in
phase modulation
= A
m
(I.21)
and is independent of the modulating frequency f
m
, while in a frequency modulation
=
2f
A
m
m
(I.22)
And depends on modulating frequency, and increases without limit as f
m
0.
I.7.4 Bandwidth requirements
The above analysis of angle modulation spectra indicates that for transmission with com-
plete delity, of even the single tone, an innite number of sidebands occupying an innite
bandwidth is required.
However, as a study of Figure I.16 will show, for each modulation index, sidebands of
beyond a certain number are of negligible amplitude. The higher the modulation index,
the more sidebands are required, but the above analysis shows that in the case of frequency
modulation, the combination of high sidebands order and amplitude is achieved only at
the lower modulation frequencies, so the actual bandwidth required for transmission with
reasonable delity is not extreme.
1
2
5
10
20
50
0.1 0.2
0.5 1 2 5 10 20 50
Deviation ratio D
Average of values for = 0.01 and = 0.1 e e
N
u
m
b
e
r
M
o
f
s
i
g
n
i
f
i
c
a
n
t
s
i
d
e
b
a
n
d
s
Figure I.17: Number of signicant sidebands in a frequency modulated signal.
This matter is illustrated in more detail in Figure I.17 wherein the number of signicant
sidebands required for transmission with reasonable delity is plotted as a function of the
deviation ratio
=
f
f
m
(I.23)
which is also equal to the modulation index which applies at the maximum signal
amplitudeA
m
= 1 and modulation frequency f
m
for a particular maximum frequency
deviationf
.
If we make the approximation from this curve that
M() + 2 (I.24)
We may derive the required transmission bandwidth B
T
for transmitting both upper
and lower sidebands is
B
T
= 2(f
+W). (I.25)
I.7.5 Detectors of angle modulation
A number of circuits have been devised for the recovery of the message from a frequency
modulated signal. One of the simplest to comprehend is shown in Figure I.18.
x (t)
c
+
_
Kx(t)
+
_
f >f
0 c
f <f
0 c
Figure I.18: A balanced discriminator circuit.
Its operation depends upon the fact that the output from the rectied output of a tuned
circuit, operating on the shoulder of the resonance curve, depends upon the frequency.
Two tuned circuits are used, one tuned so the unmodulated carrier is below the resonance,
and the other tuned so that the unmodulated carrier is above the resonance, as illustrated
in Figure I.18.
Also as shown in Figure I.18, the output signal is taken as the dierence between
the two rectied output is and exhibits good linearity. It also has the ability to follow
frequency changes quickly.
Other frequency modulation discriminator circuits are of the Foster-Seely discrimina-
tor, the ratio detector, and the phase locked loop oscillator, but these are not explored
here.
I.8 Performance of modulation systems
We now proceeded to make an analysis of the extent to which noise that enters the
transmission path of the communications system can appear at the destination.
I.8. PERFORMANCE OF MODULATION SYSTEMS 251
f
V
0
(f)
O
Top side
Bottom side
Sum
Figure I.19: Derivation of balanced FM discriminator transfer function.
I.8.1 A general communication system
The system we consider is shown in Figure I.20.
x(t) x (t)
c
v(t) y (t)
D
y(t)
x (t)
c
K
R
H (f)
R
BPF
Det
LPF
W
L
Channel
+
S
T
S
R
S
T
L
=
Trans
Noise and interference
Receiver
Figure I.20: A general CW communication system.
Although much of it has been dened before, for conveneince we summarise below the
notation used in Figure I.20.
x(t) The message signal.
S
T
The transmitted signal power.
L The channel power loss factor: 1/L is the channel power gain. It is not in dB.
W The bandwidth ofthe message signal.
x
c
(t) The modulated signal sent to the channel.
B
T
The transmisson bandwidth of the channel, assumed disortionless.
S
R
The channel output power after the loss L.
K
R
The channel amplitude gain associated with the power loss L. Thus K
R
= 1/
L
.
A
c
The unmodulated carrier amplitude. The carrier amplitude to the detector input
is thus K
R
A
c
.
v(t) The modulated singal coming out of the channel. This will be K
R
x
c
(t).
H
R
(f) The response of the band pass pre-detection lter in the receiver.
R
v
(t),
v
(t), v
i
(t), and v
q
(t), These are all descriptions of the signal v(t) presented
to the detector, either described in amplitude and phase terms, or in-phase (relative
to the unmodulated carrier) and quadrature component terms.

In phase modulation, phase deviation constant. It is the peak phase deviation,

occurring when x(t) = 1.
f
In frequency modulation, the frequency deviation constant. It is the peak fre-

quency deviation, occurring when x(t) = 1.
The frequency modulation deviation ratio
f
fm
.
I.8.2 Illustration of modulation eects
A comparison of the eects of various system of modulation carrier wave, and of the
introduction of noise in the communications path, is shown in Figure I.21. As the gure
is complex, it is best studied with the aid of the commentary below.
The gure shows an un-modulated carrier drawn, arbitrarily, upright. It is left to
the reader to imagine what happens to the carrier as it is amplitude modulated with
presumably 100 percent modulation. In such modulation, the vector OP
c
varies in length
from zero size to double the size shown here.
Next in the gure we may pay attention to the eect of noise introduced either in phase
with or in quadrature with the carrier as shown by the vectors P
c
P
1
and P
c
P
2
respectively.
In an amplitude detector the in phase noise will be present at the output while the
quadrature noise will have practically no eect. The relative proportions of output noise
to output signal will depend upon the ratio of noise power to the unmodulated carrier
power.
Next in the gure we should give attention to the angle modulated carrier of which
the instantaneous position in the modulation cycle is shown by the vector OP
3
. We note
that this is for the same carrier power as for the amplitude modulation case described
above. When we come to consider the eect of noise in an angle modulated system we
see it is the quadrature noise component represented by the vector P
c
P
2
that governs the
noise phase angle
n
.
To see the eect of this noise on the output signal-to-noise ratio products of angle
modulation system we must compare the distance P
c
P
2
with the distance swept out by
the pointP
3
during a modulation cycle.
For the case of phase modulation this distance can be up to half the circumference of
the circle shown, as for phase modulation of phase angle must stay within the range to
to avoid ambiguity in decoding. But this distance is considerably more than the radius
of the circle, so phase modulation is potentially able to produce signal-to-noise ratios at
the destination substantially better than those produced by amplitude modulation. Of
course if we mistakenly designed a phase modulation system so that the maximum of
phase angle deviation is only small, the noise performance of phase modulation system is
inferior to that of a well-designed amplitude modulation system.
O
f
n
f
s Unmodulated
carrier
Angle modulated
carrier
Carrier with
in-phase noise
Carrier with
quadrature
noise
P
1
P
2
Practical excursion of P
with phase modulation
2
Practical excursion of P
with frequency modulation
3
Unit circle
P
3
P
c
Figure I.21: Signal and noise in various modulation systems.
Now we consider the possible excursion of the point P
3
in a frequency modulation
system. The phase deviation is no longer constrained to cover the range to , but can
wind multiple times around the circle, roughly as shown by the outer spiral curve. It can
be seen that without any increase in the carrier power, but just by choice of a suitably
large frequency deviation, with consequently large phase deviation, the detected signal-
to-noise ratio can be made much greater than is possible with an amplitude modulation
system. This is particularly so at the lower frequency is where the phase deviation can
become very large.
This diagram indicates quite well how phase modulation and frequency modulation
may have, (depending on modulation index), noise performance both inferior to all su-
perior to amplitude modulation systems, but does not show the extent to which noise
generated over the transmission band manifests itself at the post detection output. These
matters are considered in the next two sections.
I.8.3 PM post detection noise spectrum
For the evaluation of the noise spectrum at the output of the phase modulation detector
a number of assumptions are made. One is that the band pass noise can be thought of
as a summation of random sinusoids distributed equally over 30 frequency band f
c

B
T
/2. Another is that, as already has been explained, only the quadrature component
contributes to phase noise. A third is that the signal-to-noise ratio is reasonably high.
With these assumptions, it can be shown that the power spectral density G
PM
(f) of the
phase modulation detector output signal is given by
G
PM
(f) =

2S
R
Rect
_
f
B
T
_
(I.26)
where

2
is the power spectral density (one side of a double sided spectrum] at the
input to the band pass lter H
R
(f), i.e. at the input to the receiver. This spectrum is
illustrated in Figure I.22.
f
B
T
2
B
T
2
_
_
W W
G (f)
zPM
O
h
2S
R
Figure I.22: PM post detection noise spectrum.
This diagram is of considerable interest because it shows us that although a transmis-
sion bandwidth B
T
considerably in excess of the message bandwidth W was needed to
transmit the signal without distortion, not all of the noise in that transmission bandwidth
falls within the recovered signal bandwidth W. Thus from a noise point of view, we are
not paying an increased penalty for the additional transmission bandwidth.
I.8.4 FM post detection noise spectrum
We observe that for frequency modulation the signal recovered from the detector is a
frequency, which is is the derivative of the phase. It can then be shown that the power
spectral density G
FM
(f) of the frequency modulation detector output signal is attainable
from the above derived expression for the phase modulation detector by multiplication by
|f|
2
, and is therefore given by
G
FM
(f) =
f
2
2S
R
Rect
_
f
B
T
_
(I.27)
where the rectangular function Rect and are as dened above. This spectrum is
illustrated in Figure I.23.
f
B
T
2
B
T
2
_
_
W W
G (f)
zFM
O
hf
2
2S
R
Figure I.23: FM post detection noise spectrum.
This gure is of even greater interest in that it rstly reinforces the notation that
not all of the output detector noise will fall within the message detection bandwidth W,
and also shows that the very low frequency noise components are suppressed by the factor
|f|
2
. This observation is entirely in accord with the earlier observation that with frequency
modulation, at low frequencies, the phase deviation becomes very large, and the phase
jitter introduced by quadrature noise becomes relatively unimportant.
I.8.5 Pre-emphasis and de-emphasis in frequency modulation
We have just observed that frequency modulation systems deal well with low frequency
noise, and less well with high frequency noise. It is also true that in broadcasting or pro-
gram material does not normally contain high frequency components of large amplitude.
In addition, the high frequency components do not produce the large phase deviations of
the low frequency components. For all of these reasons, it is expedient to apply, before
frequency modulation, a controlled amount of enhancement of the amplitude of high fre-
quency signals, and after frequency the modulation a corresponding attenuation of those
high frequency components. This process is known as pre-emphasis and de-emphasis,
and further improves signal-to-noise ratio. It is assumed that the pre-emphasis does not
signicantly increase the transmission bandwidth.
I.8.6 Parameters for broadcast FM
In broadcast FM systems for the following the parameters are used.
Deviation f
75 kHz.
Message bandwidth W 15 kHz.
In consequence, the deviation ratio is 5.
Time constant of the emphasis lter 75 s.
I.9 Destination signal to noise ratios
At good signal-to-noise ratios, the calculation of the output signal-to-noise ratio from a
frequency modulation detection system and its post-detection lter is a straightforward
matter from the results previously obtained, but we will not present the details here.
The results depend upon the variance of the message signal, the deviation ratio, and the
signal-to-noise ratio at the receiver input, and are presented in a later section.
I.9.1 FM threshold eect and mutilation
However, when the assumption of high signal-to-noise ratio is no longer valid, some of the
formulae employed in the preceding calculation break down, and the high signal-to-noise
ratio is no longer obtained. We have already seen in our qualitative discussion in relation
to gure ??, that angle modulation systems can perform a with inappropriate parameters
less well than amplitude modulation systems. The results to be presented are to a degree
a reection of this fact.
But more seriously, unless the signal-to-noise ratio is high, frequency and phase mod-
ulation decoding can produce very serious signal distortion which is known as mutilation.
The poor output signal-to-noise ratio performance at low input signal-to-noise ratios are
represented in in Figure I.24 which emerges from a detailed analysis. But this gure does
not tell the whole story about the utility of frequency modulation systems at all failures
of signal-to-noise ratio. It is a fact that below the knee of the curve, and in the region
where the signal-to-noise ratio at the destination still appears to be good, the mutilation
eect has rendered the signal unusable.
I.9.2 FM threshold extension (advanced topic to be omitted)
The limitations just discussed above provided serious constraint on the design of minimum
power frequency modulation systems. There is a technique, shown in Figure I.25, of using
I.10. COMPARISON OF CW MODULATION SYSTEMS 257
g
O
dB
(S/N) dB
D
Baseband
D=2 D=5
Figure I.24: FM noise performance as a function of gamma.
a frequency compressing feedback loop in the receiver, that will extend the range of useful
operation of FM system to lower signal levels. The use of a phase lock loop receiver can
produce a similar improvement
v(t) y (t)
D
v (t)
R
H (f)
IF
H (f)
RF
detector
LPF X
Frequency
VCO
Receiver
input
Figure I.25: FMFB receiver block diagram.
I.10 Comparison of CW modulation systems
In the table below we compare the performance of eight signal transmission systems, in
respect of, in column order: transmission bandwidth to signal bandwidth ratio, destination
signal to noise ratio to channel signal to noise ratio, normally used channel signal to noise
ratio, d.c. coupling, circuit complexity, type of detection, and common application.
Most of the entries in the table are readily derivable from theory presented in these
notes.
Type B
T
/W (S/N)
D
/
th
DC Cmplxty Comment Application
BB 1 1 N Small No mod Short link
AM 2
m
2
x
2
1+m
2
x
2
20 N Small Env det Bcst Radio
DSB 2 1 Y Large Sync det Analog data
SSB 1 1 N Medium Sync det PP voice
VSB 1+ 1 Y Large Sync det Dig data
VSB+C 1+
m
2
x
2
1+m
2
x
2
20 Y Medium Env det TV
PM 2M(
)
2

x
2
10B
T
/W Y Medium Phase det Dig data
FM 2M() 3
2
x
2
10B
T
/W Y Medium Freq det FM Radio
Table I.1: Comparison of Continuous Wave Modulation Systems.
I.10. COMPARISON OF CW MODULATION SYSTEMS 259

Communications S&S

Caricato da

Informazioni sul documento

Descrizione originale:

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Communications S&S

Caricato da

Copyright:

Formati disponibili

LEVEL 3 3015 COMMUNICATIONS, SIGNALS

AND SYSTEMS 2003

(f), so the magnitude

x() d X(f)/j2f + (1/2)X(0) (f)

X(f) df = area of X(f) (6.31)

x(t) dt = area of x(t) (6.32)

(x) (x) dx = (0) (6.43)

x() y(t ) d = x(t) y(t) (6.57)

h(t, ) x() d (6.72)

x() h(t ) d = h(t) x(t) (6.73)

sin 2 = 2 sin cos

(f) sgn(t) rect(t/T)

f(x nP) X(0) = area of x(t)

(f) if x(t) is real

is called the phase-deviation constant, and is the maximum phase deviation

is called the frequency deviation constant. Since frequency is the time

In phase modulation, phase deviation constant. It is the peak phase deviation,

In frequency modulation, the frequency deviation constant. It is the peak fre-

Potrebbero piacerti anche

Communications S&amp;S

Caricato da

Informazioni sul documento

Descrizione originale:

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Communications S&amp;S

Caricato da

Copyright:

Formati disponibili

LEVEL 3 3015 COMMUNICATIONS, SIGNALS

AND SYSTEMS 2003

(f), so the magnitude

x() d X(f)/j2f + (1/2)X(0) (f)

X(f) df = area of X(f) (6.31)

x(t) dt = area of x(t) (6.32)

(x) (x) dx = (0) (6.43)

x() y(t ) d = x(t) y(t) (6.57)

h(t, ) x() d (6.72)

x() h(t ) d = h(t) x(t) (6.73)

sin 2 = 2 sin cos

(f) sgn(t) rect(t/T)

f(x nP) X(0) = area of x(t)

(f) if x(t) is real

is called the phase-deviation constant, and is the maximum phase deviation

is called the frequency deviation constant. Since frequency is the time

In phase modulation, phase deviation constant. It is the peak phase deviation,

In frequency modulation, the frequency deviation constant. It is the peak fre-

Potrebbero piacerti anche

Communications S&S

Communications S&S