Sunteți pe pagina 1din 148

A NEW DERIVATION OF THE PULL-OUT FREQUENCY

FOR SECOND-ORDER PHASE LOCK LOOPS


EMPLOYING TRIANGULAR AND SINUSOIDAL PHASE
DETECTORS
by
ABU-SAYEED A. HUQUE
A DISSERTATION
Submitted in partial fulllment of the requirements
for the degree of Doctor of Philosophy
in
The Department of Electrical and Computer Engineering
to
The School of Graduate Studies
of
The University of Alabama in Huntsville
HUNTSVILLE, ALABAMA
2011
All rights reserved
INFORMATION TO ALL USERS
The quality of this reproduction is dependent on the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.
All rights reserved. This edition of the work is protected against
unauthorized copying under Title 17, United States Code.
ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346
UMI 3492782
Copyright 2012 by ProQuest LLC.
UMI Number: 3492782
ACKNOWLEDGMENTS
All praises be to God who has guided me to this work and facilitated it for me
by providing a strong and supportive supervisory committee.
I would like to express my sincere gratitude to my adviser, Dr. John L. Stensby,
for his immeasurable help and constant guidance from the beginning of the research
until the nal stage of reviewing and editing of this dissertation. His inspirational
role along with his ability to explain diverse and complex materials in clear terms
was the greatest help in this endeavor.
I would like to thank sincerely the other members in my committeeDr. Mervin
C. Budge, Dr. Laurie L. Joiner, Dr. Earl B. Wells and Dr. Shangbing Aifor
providing me with invaluable guidance, support and encouragements. I do appreciate
their time in administering the exams and reviewing this dissertation. Each of them
has imparted a profound learning experience throughout my stay at UAH.
I am grateful to the Department of Electrical and Computer Engineering at
UAH for providing the necessary funding for my entire PhD program. I would like
to recognize Dr. Robert Lindquist, the Chair of the ECE department, who helped
me keep my focus on both the publications as well as the nal stage of dissertation
to ensure the successful completion of the degree.
I also would like to thank Dr. John C. Polkings at Rice University in Houston,
TX for making the source code for pplane8, a Matlab based software package to
generate phase portrait, available to the public for academic research.
vi
I would like to recognize my friend, Dr. Sharif Bhuiyan, who insisted that I
use Latex to write this dissertation and stood by me the entire time with unfailing
support and guidance.
Finally, I extend sincere appreciation to my father, my wife, my sisters, my
brothers and my children for their constant support, encouragements and prayers.
vii
TABLE OF CONTENTS
List of Figures xii
List of Tables xv
List of Symbols xvi
List of Abbreviations xix
Chapter
1 Introduction 1
1.1 Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Error Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Maximum Seeking Followed by Selection Filtering . . . . . . . 3
1.1.3 Nonlinear Operation Followed by Passive Filtering . . . . . . . 4
1.2 History of Phase Lock Loops . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Cycle Slip and Pull-out Frequency . . . . . . . . . . . . . . . . . . . . 6
1.4 Topical Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 Dynamical System Theory 10
2.1 Basic Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.1 Building Blocks of Dynamical Systems . . . . . . . . . . . . . 11
2.1.2 Mathematical Model . . . . . . . . . . . . . . . . . . . . . . . 11
viii
2.1.3 Trajectory or Orbit . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.4 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.5 Equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.6 Singularity and Singular Point . . . . . . . . . . . . . . . . . . 16
2.2 Higher-order Dierential Equation . . . . . . . . . . . . . . . . . . . . 16
2.3 Linear System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.1 Autonomous Homogeneous Linear System . . . . . . . . . . . 20
2.3.2 Planar Linear System . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.2.1 Phase Portrait . . . . . . . . . . . . . . . . . . . . . 24
2.3.2.2 Classication of Equilibrium Points . . . . . . . . . . 24
2.3.2.3 Equilibria Associated with Real Eigenvalue . . . . . 25
2.3.2.4 Equilibria Associated with Complex Eigenvalue . . . 28
2.4 Nonlinear System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.4.1 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . . 32
2.4.2 Equilibria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.4.3 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4.4 Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4.5 Periodic Orbit and Limit Cycle . . . . . . . . . . . . . . . . . 36
2.4.6 Bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.4.7 Planar System . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3 System Model for Second-order Phase Lock Loop 39
3.1 Denition of a Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
ix
3.2 The Need for Models . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3 Block Diagram of a PLL . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4 Model for Component Blocks . . . . . . . . . . . . . . . . . . . . . . 44
3.4.1 Phase Detector . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.4.1.1 Analog Phase Detector . . . . . . . . . . . . . . . . . 45
3.4.1.2 Digital Phase Detector . . . . . . . . . . . . . . . . . 50
3.4.2 Loop Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.4.3 Voltage Controlled Oscillator . . . . . . . . . . . . . . . . . . 54
3.5 System Model of a Second-order PLL . . . . . . . . . . . . . . . . . . 57
4 Second-order Type II Phase Lock Loop 62
4.1 System Model of a Second-order Type II PLL . . . . . . . . . . . . . 62
4.2 Denition of Pull-out Frequency . . . . . . . . . . . . . . . . . . . . . 64
4.3 Derivation of the Formula to Calculate Pull-out Frequency . . . . . . 64
4.3.1 Type II PLL Employing Triangular PD . . . . . . . . . . . . . 66
4.3.1.1 Derivation of an Exact Formula to Calculate Pull-out
Frequency . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3.1.2 Validation of the Exact Formula . . . . . . . . . . . 75
4.3.2 Type II PLL Employing Sinusoidal PD . . . . . . . . . . . . . 78
4.3.2.1 Derivation of a Formula to Approximate Pull-out Fre-
quency . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.3.2.2 Comparison between Derived and Existing Empirical
Formula . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.4 Comparison of Pull-out Frequency between Type II PLLs Employing
Triangular vs. Sinusoidal PD . . . . . . . . . . . . . . . . . . . . . . 90
x
5 Second-order Type I Phase Lock Loop 93
5.1 System Model of a Second-order Type I PLL . . . . . . . . . . . . . . 93
5.1.1 Type I PLL Employing Sinusoidal PD . . . . . . . . . . . . . 94
5.1.1.1 Phase Portrait in Region I . . . . . . . . . . . . . . . 99
5.1.1.2 Phase Portrait in Region II . . . . . . . . . . . . . . 100
5.1.1.3 Phase Portrait in Region III . . . . . . . . . . . . . . 101
5.1.2 Type I PLL Employing Triangular PD . . . . . . . . . . . . . 102
5.1.2.1 Phase Portrait in Region I . . . . . . . . . . . . . . . 107
5.1.2.2 Phase Portrait in Region II . . . . . . . . . . . . . . 108
5.1.2.3 Phase Portrait in Region III . . . . . . . . . . . . . . 109
5.2 Comparison between Type I and Type II Loops Based on the Phase
Portrait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6 Conclusion and Future Work 112
6.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
APPENDIX A: Maple Code to Generate Phase Portrait for a Type II
PLL with Triangular PD 118
APPENDIX B: pplane8 GUI Set up to Generate Phase Portrait 120
APPENDIX C: Matlab Code to Generate Separatrix for Type II PLL122
REFERENCES 125
xi
LIST OF FIGURES
FIGURE PAGE
1.1 Block Diagram of Typical Error Control Feedback System . . . . . . 3
1.2 Binary FSK Modulation scheme . . . . . . . . . . . . . . . . . . . . . 7
1.3 Block Diagram of an FSK Demodulator Implemented with a PLL . . 8
2.1 Phase Portrait for Star Node: (a) Star Sink (b) Star Source . . . . . 26
2.2 Phase Portrait for Node: (a) Unstable Node or Source (b) Stable Node
or Sink . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3 Phase Portrait of Saddle Node for Two Dierent Sets of Eigenvalues . 28
2.4 Phase Portrait of Center Node (a) Counter Clockwise Rotation (b) Clock-
wise Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5 Phase Portrait of Spiral Source (a) Counter Clockwise Rotation (b) Clock-
wise Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.6 Phase Portrait of Spiral Sink (a) Counter Clockwise Rotation (b) Clock-
wise Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1 Key Steps of Mathematical Modeling . . . . . . . . . . . . . . . . . . 40
3.2 Basic Architecture of a Phase Lock Loop System . . . . . . . . . . . 43
3.3 Block Diagram of Phase Detector . . . . . . . . . . . . . . . . . . . . 45
3.4 Block Diagram of Multiplier Type Sinusoidal PD . . . . . . . . . . . 46
3.5 Schematic of Simplied Gilbert Multiplier as Sinusoidal PD
Courtesy: Phase-Locked Loops: Theory and Application; John L. Stensby 47
3.6 Schematic of Diode-ring Doubly-balanced Mixer as a Sinusoidal PD
with Associated Waveforms
Courtesy: Phase-Locked Loops: Theory and Application; John L. Stensby 48
xii
3.7 Triangular Phase Detector Implemented by XOR Multiplier with As-
sociated Truth Table and Waveforms
Courtesy: Phase-Locked Loops: Theory and Application; John L. Stensby 50
3.8 Saw-Tooth (Sequential) Phase Detector Implemented by RS Flip-Flop
with Associated State Diagram and Waveforms
Courtesy: Phase-Locked Loops: Theory and Application; John L. Stensby 51
3.9 Block Diagram of Loop Filter . . . . . . . . . . . . . . . . . . . . . . 52
3.10 Schematic of a First-order Active Filter
Courtesy: Phase-Locked Loops: Theory and Application; John L. Stensby 53
3.11 Schematic of a First-order Passive Filter
Courtesy: Phase-Locked Loops: Theory and Application; John L. Stensby 53
3.12 Block Diagram of Voltage Controlled Oscillator . . . . . . . . . . . . 55
3.13 Astable Multivibrator Type Voltage Controlled Oscillator . . . . . . . 56
3.14 LC Tank Series Resonant Voltage Controlled Crystal Oscillator . . . 57
3.15 Block Diagram of a Second-order PLL . . . . . . . . . . . . . . . . . 60
4.1 Block Diagram of a Second-order Type II PLL . . . . . . . . . . . . . 63
4.2 Block Diagram of a Second-order Type II PLL with Triangular PD . 66
4.3 Triangular Waveform Phase Detector Output Characteristic . . . . . 67
4.4 Typical Phase Portrait of a Second-order Type II PLL with Triangular
PD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.5 Numerical Validation of Derived Formula for Loop Parameter a

= 0.5 76
4.6 Numerical Validation of Derived Formula for Loop Parameter a

= 0.2 77
4.7 Numerical Validation of Derived Formula for Loop Parameter a

= 1.5 77
4.8 Block Diagram of a Second-order Type II PLL with Sinusoidal PD . . 78
4.9 Sinusoidal Waveform Phase Detector Output Characteristic . . . . . . 79
xiii
4.10 Typical Phase Portrait of a Second-order Type II PLL with Sinusoidal
PD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.11 Four-interval Simpsons Rule to Approximate Integrals . . . . . . . . 87
4.12 Separatrices for Sinusoidal and Triangular PD Based Type II PLLs . 91
4.13 Comparison of Pull-out Frequency for Sinusoidal and Triangular PD
Based Type II PLLs . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.1 Block Diagram of a Second-order Type I PLL . . . . . . . . . . . . . 94
5.2 Block Diagram of a Second-order Type I PLL with Sinusoidal PD . . 95
5.3 Region I Phase Portrait of Type I PLL with Sinusoidal PD . . . . . . 100
5.4 Region II Phase Portrait of Type I PLL with Sinusoidal PD . . . . . 101
5.5 Region III Phase Portrait of Type I PLL with Sinusoidal PD . . . . . 102
5.6 Block Diagram of a Second-order Type I PLL with Triangular PD . . 102
5.7 Region I Phase Portrait of Type I PLL with Triangular PD . . . . . . 107
5.8 Region II Phase Portrait of Type I PLL with Triangular PD . . . . . 108
5.9 Region III Phase Portrait of Type I PLL with Triangular PD . . . . . 109
A.1 Maple Code to Generate Phase Portrait of Type II PLL with Triangular
PD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
B.1 pplane8 GUI Set up to Generate Phase Portrait for a PLL with Sinu-
soidal PD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
B.2 pplane8 GUI Set up to Generate Phase Portrait for a PLL with Trian-
gular PD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
C.1 Matlab Code to Generate Separatrix for a PLL with Sinusoidal PD . 123
C.2 Matlab Code to Generate Separatrix for a PLL with Triangular PD . 124
xiv
LIST OF TABLES
TABLE PAGE
4.1 Comparison of Pull-out Frequency from Three Dierent Methods . . 90
A.1 Redenition of System Variables and Loop Parameters in Maple Code 118
B.1 Redenition of System Variables and Loop Parameters in pplane8 . . 120
C.1 Redenition of System Variables and Loop Parameters in Matlab Code 122
xv
LIST OF SYMBOLS
SYMBOL DEFINITION
Phase (error)

Frequency (error)

i
Instantaneous phase of reference input signal

v
Instantaneous phase of VCO output signal

1
Instantaneous relative phase of reference input signal

2
Instantaneous relative phase of VCO output signal

0
Center frequency of VCO
e(t) Error signal
v
ref
(t) Reference input signal
v
V CO
(t) VCO output signal
g() Phase detector output function
g

() Derivative of phase detector output function

Detuning parameter
a Zero of loop lter
b Pole of loop lter
G Closed loop gain
xvi

Gain normalized Detuning parameter


a

Gain normalized Zero of loop lter


b

Gain normalized Pole of loop lter


Damping coecient

n
Damped natural frequency
Bifurcation parameter

c
Critical value of bifurcation parameter
Set of real numbers
C Set of complex numbers
X State vector
A(t) Coecient matrix
R(t) Vector forcing function
Eigenvalue
V Eigenvector

po
Pull-out frequency

h
Hold-in range

p
Pull-in range

2
Half-plane pull-in frequency

po
Gain normalized pull-out frequency
xvii

h
Gain normalized hold-in range

p
Gain normalized pull-in range

2
Gain normalized half-plane pull-in frequency
xviii
LIST OF ABBREVIATIONS
ABBREVIATION DEFINITION
DC Direct Current
DPLL Digital Phase Lock Loop
DSP Digital Signal Processing
FDMA Frequency Division Multiple Access
FSK Frequency Shift Keying
MODEM Modulator Demodulator
PA Power Amplier
PD Phase Detector
PLL Phase Lock Loop
ODE Ordinary Dierential Equation
OFDM Orthogonal Frequency Division Multiplexing
OP-AMP Operational Amplier
SNR Signal to Noise Ratio
TDMA Time Division Multiple Access
VCO Voltage Control Oscillator
VCXO Voltage Control Crystal Oscillator
xix
VLSI Very Large Scale Integration
XOR Exclusive OR
xx
To my parents Nurun Nahar Huque and Abu Obaidul Huque
CHAPTER 1
INTRODUCTION
1.1 Synchronization
When information is transmitted from one location to another, regardless of
the type of processing (analog or digital), the importance of modulating the carrier
is obvious. However, in the case of coherent communications, to be able to retrieve
(reconstruct) the information, rst the receiver has to extract the time base of the
transmitter. The process of this time base extraction is known as synchronization in
the communications literature. The three levels of synchronization commonly found
in most communication systems are briey discussed below.
(I) Carrier synchronization: This type of synchronization is common in both ana-
log and digital communications. Here, a local oscillator in the receiver is locked
to the carrier of the transmitted message (analog or bit stream).
(II) Clock synchronization: This type of synchronization is particularly found in
digital communications. Here, the rising or the falling edge of the receiver
clock is lined up with that of the transmitter clock so that the receiver can
correctly sample the transmitted data stream.
1
(III) Word synchronization: This type of synchronization is applied in digital com-
munications where data is transported in packet or burst mode. Here, the
beginning as well as the end of the valid data packet are identied by the re-
ceiver. Thus, the receiver avoids processing the extra data that is present in the
transmitted bit stream and, instead, processes only the useful data correctly.
The process of synchronization is equally essential for any coherent communi-
cation independent of the medium used, e.g., wire or wireless, cable or ber. However,
the implementation details may vary. It is also true for any transmitter(s)-receiver(s)
connection topology, such as point-to-point, multicast, broadcast, shared, mesh, etc.
For instance, in the case of a shared channel communications, where the dedicated
messages for each user are combined (multiplexed) in dierent time slots (TDMA)
or frequency slots (FDMA), or, a combination of both, the individual receiver has to
perform the synchronization before extracting (decoding) the message intended for
it.
There are at least three ways to implement the synchronizer, and they are
discussed in the following sections.
1.1.1 Error Tracking
Error tracking is a typical feedback control system and is depicted by Fig-
ure 1.1. The reference input contains the phase or timing information of interest.
An error signal e is computed in a functional block commonly known as the phase
detector (PD). This PD can be implemented in analog ways or digitally to produce an
error-dependent output (one of several popular waveforms), which is fed back to ad-
2
Error
Calculator
Output
Adjuster
Ref. Input
e
Figure 1.1: Block Diagram of Typical Error Control Feedback System
just the phase and frequency of a local oscillator to line up with those of the reference
carrier (or clock). In a properly designed loop, the error signal must vanish to achieve
lock. However, a zero error signal is only a necessary but not sucient condition for
phase/frequency lock. This is still one of the most common synchronizers used in
most communications systems, and it is known as a phase lock loop (PLL). In many
implementations, a lowpass lter (active/passive) is added to smoothen the error sig-
nal fed back to the local oscillator. The complexity of model analysis for such loops
increases drastically with the order of the loop. Surprisingly, lower order (rst and
second-order) loop models perform satisfactorily for most applications. The rare use
of higher order (third or fourth-order) models are found in very specic applications.
1.1.2 Maximum Seeking Followed by Selection Filtering
Here, the transmitter sends a predened trainer sequence of pulses that is
known to the receiver. Upon reception, the receiver compares this sequence against a
number of delayed replicas of the same. Then the selection lter chooses the optimum
3
delay that produces the maximum match (correlation) and uses that delay to deter-
mine the time base of the transmitter. The advantage of this approach is that the
optimum delay provides the necessary and the sucient condition for locking, unlike
the previous (PLL) approach. However, in this situation, many error values (equal
to the number of delays chosen) have to be computed as opposed to one in case of
a PLL. This increases the complexity of the circuit by orders of magnitude, which is
why, until the recent advancement of DSP and VLSI technologies, this approach was
primarily of academic interest only.
1.1.3 Nonlinear Operation Followed by Passive Filtering
This technique has been used in many applications until today. The received
(data-carrying) signal is squared (the most commonly applied nonlinear operation)
and then fed through a narrow band passband passive lter centered at the inverse of
the symbol time (1/T). The resultant is almost a sinusoidal signal perturbed by some
random uctuations. After proper analysis, it can be shown that the zero crossings
of the output signal can be used to determine the time base of the transmitter [1].
1.2 History of Phase Lock Loops
The history of phase lock loops (PLL) goes back to the early 1930s and coin-
cides with the introduction of coherent communication. In 1932, a British researcher
named Hann De Bellecize was trying to come up with an alternative receiver to Ed-
win Armstrongs heterodyne receiver. His goal was to minimize the number of tuned
circuits that were needed to build the heterodyne receiver. To this end, he used a local
4
oscillator that was tuned to the signal carrier frequency and multiplied it by the input
to produce the original modulated signal without any phase and/or frequency dis-
tortion. It was then known by dierent names, such as homodyne, direct-conversion
or synchrodyne receiver. Later, an automatic error control circuitry was added to
correct the rapidly drifting frequency of the local oscillator.
In the late 1930s to early 1940s, the widespread use of the homodyne receiver
was found in analog televisions, where vertical and horizontal sweep mechanisms re-
quired that a continuous clock be synchronized with a periodic sync pulse present
in the transmitted signal. In the mid 1960s, its use in various communications sys-
tems skyrocketed when Signetics launched their rst ever complete PLL system in a
monolithic integrated circuit. In the mid 1970s, classic digital PLLs, where the phase
detector was implemented with XOR gates, etc., became available. A few years later,
an all-digital PLL hit the market in which a numerically controlled oscillator replaced
its legacy counterpart. Around same time, laser and optical PLLs started to emerge
and signicantly advanced optical communications.
Today, PLLs are found in every cell phone, modem, router, switch, computer,
television, radio, pager and all telephony devices. In many applications, PLL function-
ality can be implemented entirely in software on sampled data. At high frequencies,
optical PLLs are used to recover clocks from hundreds of Gbps data stream. Thus,
the PLL has turned out to be one of the most prolic feedback systems designed by
engineers.
5
1.3 Cycle Slip and Pull-out Frequency
The phenomenon of cycle slip occurs when the number of cycles present in
the reference input (clock) deviates from the number of cycles found in the VCO
output. It always happens in the power up and reset recovery of a system. However,
it may also happen when a PLL is in a locked condition. The random noise present in
the reference input and/or the VCO output signal can induce enough phase and/or
frequency error to make the PLL lose the lock and miss one or more reference input
(clock) cycles before it locks back to the reference input. Such an event can be
detrimental in many applications, both in digital and in analog communications. In
some applications, this cycle slip may cause the VCO to tune its output in the wrong
direction. In digital packet data synchronization applications, cycle slip causes the
receiver to misread the packet (word) boundaries and, hence, mixes up the received
data stream.
The pull-out frequency of a second-order Type II PLL quanties the maxi-
mum allowable jump in the reference input frequency, that still ensures a pull-in (see
Section 3.3) without a cycle slip. This quantity is extremely critical in frequency
modulation schemes, which include, but are not limited to, Frequency Shift Keying
(FSK), multi-carrier (OFDM), frequency-hopping system, etc., where multiple dis-
crete frequencies are used as carriers. This change of carriers can be treated as an
abrupt jump in the reference input to a locked PLL. Yet, the front-end receiver is
expected to lock on to those frequencies in the same cycle. Therefore, the modulation
bandwidth cannot exceed the pull-out frequency.
6
Figure 1.2: Binary FSK Modulation scheme
In addition, the process of investigating the pull-out frequency includes the
study of a special trajectory on the phase portrait, called the separatrix (see Sec-
tion 2.3.2.1). In a 2 period between two successive saddle nodes (see Section 2.3.2.3),
two separatrices dene the region of convergence that guarantees a zero cycle slip reac-
quisition. This implies that, as long as the phase and the frequency error induced
by the noise (or otherwise) do not cross that region, the PLL will re-achieve the lock
without a cycle slip.
A direct application of the pull-out frequency can be seen in the construction
of the FSK demodulator. Figure 1.2 shows a digital (binary) data stream and the
corresponding analog waveform after FSK modulation. A carrier at frequency f
m
,
commonly known as the Mark frequency, is used to transmit a binary 1 while, a
carrier at frequency f
s
, commonly known as the Space frequency, is used to transmit
a binary 0. Figure 1.3 shows a block diagram of an FSK demodulator built simply
with a PLL. The purpose of the power amplier (PA) before the PLL is to boost
the signal level received by the antenna. The purpose of the comparator after the
7
Figure 1.3: Block Diagram of an FSK Demodulator Implemented with a PLL
PLL, built with an OP-AMP, is to reconstruct the binary waveform by comparing the
PLL output with a threshold. As long as the dierence between f
m
and f
s
is smaller
than the pull-out frequency of the PLL, the demodulator can successfully retrieve the
transmitted binary data.
1.4 Topical Outline
Chapter 2 contains a brief discussion of dynamical system theory. The chapter
starts with the denition of a dynamical system and its dierent attributes. Then
it briey touches upon dierent concepts, including equilibria, stability, trajectory
and separatrix, related to linear and nonlinear systems as well as standard ways to
analyze them. It focuses particularly on second-order systems since this research is
interested in examining such systems. This chapter also touches upon the use of
phase portraits in analyzing second-order (also known as planar) systems.
Chapter 3 starts with the denition of and the need for a mathematical model.
It presents classications of models based on dierent aspects of the system. It then
presents the mathematical model as well as the popular implementations of dierent
8
component blocks of a standard PLL system. Finally, it presents the derivation of
the system equation for a second-order PLL employing a generic phase detector.
Chapter 4 contains the core of this research. It denes the parameter known
as the pull-out frequency in the case of a second-order Type II PLL, which guaran-
tees the re-lock to the reference input without a cycle slip. First, it provides the
derivation of an exact formula to calculate this parameter for a second-order Type II
PLL, employing a triangular waveform PD. Next, it furnishes the derivation of an
approximating formula to calculate the same parameter in the case of a sinusoidal
waveform PD. In both cases, the correctness of the derived formulae were veried
against the numerical results. The end of the chapter presents a comparative study
of the PLLs, employing these two types of PDs based on the pull-out frequency.
Chapter 5 touches on some basic features of a second-order Type I PLL em-
ploying the above-mentioned phase detectors for the sake of completeness. It also
highlights the dierences in system behavior between Type I and Type II PLLs based
on the phase portrait. Lastly, it explains the counterpart of the pull-out frequency for
Type I PLLs, which is known as the pull-in range or the half-plane pull-in frequency
based on the operating region.
Finally, the conclusion of this work and the potential future work is discussed
in Chapter 6.
9
CHAPTER 2
DYNAMICAL SYSTEM THEORY
2.1 Basic Concept
Dynamical system theory is a branch of mathematics that studies the behavior
of systems that vary with time. The goal of the study is to be able to predict
the behavior (outcome) of a particular system at a later time (sometimes in reverse
time) based on information about the system at current (or a specic) time. Though
the theory of dynamical systems concerns one kind of framework that carries out a
methodical study of a system, there may be more than one such framework.
The origin of dynamical systems goes back to the Newtonian mechanics that
began by studying the motion of a particle (body) under the inuence of an exter-
nal force. Despite the discovery of the theory of relativity (also known as Quan-
tum mechanics), Newtonian (also known as classical ) mechanics still remains one
of the prime conceptual cornerstones in physical science. With the successive waves
of new concepts and ideas (many of them have not reached their conclusions), it
still continues to enrich not only the eld of physics but also the the eld of applied
mathematics [2].
10
2.1.1 Building Blocks of Dynamical Systems
In order to study the dynamical behavior of any system, at the very least, we
need to identify the following two variables:
(a) We must determine what constitutes the instantaneous description of the system
under study, often known as state variables. In other words, we need to identify
the necessary and sucient information to characterize that system at a particular
instance of time.
(b) We must determine how those state variables change and interact with each other
over time [2].
2.1.2 Mathematical Model
One of the practical ways to ascertain the nature of any real system is by
conducting specic (methodical) observations or measurements of the system. In
general, one such measurement taken at a particular instance yields a denite number
(real), which will characterize the system at that instance of time. It is obvious that
more than one such measurement at the same instance will increase the precision of
the characterization process. Thus, an ordered set of n-tuple numbers originating from
a nite set of measurements on the system will represent the possible instantaneous
state of the system. Each of the quantities, namely x
1
, x
2
, x
3
, ..., x
n
, determined
by the corresponding measurements is commonly referred to as a state variable in
the dynamical system literature. The state variables may be regarded as n axes of
11
an n-dimensional Euclidian space. Thus the subspace containing all of the possible
n-tuple coordinates that represent the system is called the state space.
Now, the second challenge is to determine the time variation of the system. In
other words, each of the state variables has to be treated as a function of time x
i
(t),
and the description of the system will not be complete until we are able to dene them
all. Unfortunately, we are almost never able to dene them all directly. Instead, the
best we can do is to try to come up with some conditions that must be satised by
these functions and, thereby, extract the functions from those conditions. In fact, the
core of dynamical system theory rests on one such general condition, which states
that the rate of change of a particular state variable x
i
(t) in the vicinity of a given
state [x
1
(t
0
), x
2
(t
0
), x
3
(t
0
), ..., x
n
(t
0
)] only depends on that state and time. This
reduces our complete description of the system to a set of simultaneous rst-order
dierential equations, shown as
x

1
= f
1
(t, x
1
, x
2
, x
3
, ..., x
n
)
x

2
= f
2
(t, x
1
, x
2
, x
3
, ..., x
n
)
.
.
.
x

n
= f
n
(t, x
1
, x
2
, x
3
, ..., x
n
)
_

_
, (2.1)
where x

i
=
dx
i
dt
.
The above description is rather generic due to the presence of the potential
time dependency in each of the real functions f
i
, 1 i n. Systems depicted by this
type of model are known as non-autonomous systems. However for most practical
12
systems, if this time dependency is dropped, the description still remains a valid one
and, thus, signicantly eases the analysis. In other words, a vast majority of the
physical systems can be modeled mathematically by treating the time derivatives
of the state variables in the vicinity of a particular state to be solely dependent on
the state itself. Such systems are termed as autonomous systems and are of prime
concern to this research. Thus, the system model described by Equation 2.1 becomes
the following for an autonomous system:
x

1
= f
1
(x
1
, x
2
, x
3
, ..., x
n
)
x

2
= f
2
(x
1
, x
2
, x
3
, ..., x
n
)
.
.
.
x

n
= f
n
(x
1
, x
2
, x
3
, ..., x
n
)
_

_
, (2.2)
where all f
i
, 1 i n, are dened in and all x

i
, 1 i n, exist in
n
. The above
dynamical equation can be written in a more compact form using the vector notation
as
X

= F(X), (2.3)
.
where X =
_

_
x
1
x
2
.
.
.
x
n
_

_
, X

=
_

_
dx
1
dt
dx
2
dt
.
.
.
dx
n
dt
_

_
and F =
_

_
f
1
(x
1
, x
2
, x
3
, , x
n
)
f
2
(x
1
, x
2
, x
3
, , x
n
)
.
.
.
f
n
(x
1
, x
2
, x
3
, , x
n
)
_

_
.
13
2.1.3 Trajectory or Orbit
The solution of the system will be (t) described by Equation 2.3 on the open
interval I : (t
0
, t) , where t
0
usually denotes the initial time, if and only if the
following two conditions are met:
(a) (t) is dened in the sub space E, also known as the state space, of
n
. In other
words, all of the real-valued component functions of (t), i.e.,
i
(t) for 1 i
n, are dened on I.
(b) The time derivative of (t), i.e., the derivatives of the component functions,

i
(t)
for 1 i n, exits and satises the system equation.
Such (t) is called a trajectory or orbit of the system. With two-dimensional
autonomous systems in
2
, the trajectories do not intersect each other in the phase
plane, due to the guaranteed uniqueness of solution [3].
2.1.4 Stability
Stability of the system is one of the most crucial concepts in dynamical system
analysis. It refers to the systems response to some perturbation; more precisely, the
behavior of the trajectories under perturbation when the time approaches innity, i.e.,
t . In other words, the stability qualies the global properties of unperturbed
trajectories (orbits) under perturbation. Often, this qualication process is not easy
since it requires the extension of the local behavior (information) of the system under
study to the global state space when perturbation is applied. This also involves the
14
existence and uniqueness theorem (see Section 2.4.1). The following provides a brief
description of four types of stability observed in dynamical systems:
(A) A trajectory is called unstable when its neighboring trajectories move further and
further away from it as time progresses. In other words, a small perturbation
will cause a drastic change in the behavior of a particular trajectory over time.
(B) A trajectory is called neutrally or weakly stable if, for a predened degree of
closeness , a measure of closeness = () can always be found such that, all
trajectories initially closer to the unperturbed trajectory than will never move
further away from it than .
(C) A trajectory is called bounded if, for a given nite bound, all trajectories that
are initially suciently close to it never move further away from that bound.
(D) A trajectory is called asymptotically stable if all trajectories that are initially
suciently close to it move arbitrarily close to it as time progresses.
2.1.5 Equilibria
An interesting trajectory for a system (if it exists) is dened by the solution
(t)=[c
1
c
2
c
3
c
n
]
T
, which is a constant vector, since it always satises the right
hand side of Equation 2.2 namely F(X) = 0. Now, if we ip the argument and nd a
constant solution by equating F(X) to zero, this is called the equilibrium point of the
system. In other words, at such a state in the state space, the time derivative of the
state vanishes, i.e., all of the derivatives of the state variables with respect to time
identically become zeros. Thus, if the system initiates at such a state, it stays in the
15
same state forever. There may be systems where no such state (equilibrium) is found.
On the other hand, there may be more than one equilibrium state for some systems.
Lastly, in some systems, that there may exit a unique equilibrium point in the state
space. Frequently, the term critical point is interchangeably used for equilibrium
point. It is important to note here that each equilibrium point itself constitutes a
complete trajectory of the system.
2.1.6 Singularity and Singular Point
The concepts of singularity and singular points are opposite to the concepts
of stability and equilibrium points, respectively. If, in the state space, we nd a state
in which the time derivative of the state becomes innite or indeterminate, such state
is called a singular point. One way to nd such a point or state (if it exists) is by
equating F(X) to innity. Like the equilibrium point, there may be one singular
point, more than one or none for a particular dynamical system.
2.2 Higher-order Dierential Equation
Any dierential equation that involves derivatives that have orders higher than
one is known as a higher-order dierential equation. Here we will only be dealing with
the ordinary dierential equations since they only involve the total derivatives and
not the partial derivatives. Assuming t and x to be the independent and dependent
variables, respectively, the general form of an n
th
- order ordinary dierential equation
can be expressed as
16
x
(n)
= f(t, x, x
(1)
, x
(2)
, ..., x
(n1)
), (2.4)
where the n
th
derivative of x with respect to t is x
(n)
=
d
n
x
dt
n
.
If we substitute x = x
1
, x
(1)
= x
2
, x
(2)
= x
3
, ... , x
(n1)
= x
n
, Equation 2.4
can be written, in the vector form, as a set of simultaneous rst-order equations:
X

=
_

_
x
(1)
x
(2)
.
.
.
x
(n)
_

_
=
_

_
x
(1)
1
x
(1)
2
.
.
.
x
(1)
n
_

_
=
_

_
x

1
x

2
.
.
.
x

n
_

_
=
_

_
x
2
x
3
.
.
.
f(t, x, x
(1)
, x
(2)
, , x
(n1)
_

_
= F(t, X). (2.5)
Thus, the above equation becomes equivalent to Equation 2.1, known as a
non-autonomous system. If we can somehow eliminate the time dependency from the
vector function F, the resultant equation will be equivalent to Equation 2.2, known
as autonomous system. In summary, any higher-order ordinary dierential equation
can be expressed in the form of a standard dynamical system equation. Therefore, it
is important at this point to discuss the general solution and the solutions properties
of a system of simultaneous rst-order dierential equations. We will introduce linear
systems rst to lay ground for nonlinear systems, which will be our research focus
going forward.
17
2.3 Linear System
A rst-order linear system of dierential equation is a special case of Equa-
tion 2.1 and can be written as
x

1
= a
11
(t)x
1
+ a
12
(t)x
2
+ ... + a
1n
x
n
+ r
1
(t)
x

2
= a
21
(t)x
1
+ a
22
(t)x
2
+ ... + a
2n
x
n
+ r
1
(t)
.
.
.
x

n
= a
n1
(t)x
1
+ a
n2
(t)x
2
+ ... + a
nn
x
n
+ r
1
(t)
_

_
. (2.6)
The linear system described by Equation 2.6 can be rewritten in matrix form
as
_

_
x

1
x

2
.
.
.
x

n
_

_
=
_

_
a
11
(t) a
12
(t) a
1n
(t)
a
21
(t) a
22
(t) a
2n
(t)
.
.
.
.
.
.
.
.
.
.
.
.
a
n1
(t) a
n2
(t) a
nn
(t)
_

_
_

_
x
1
x
2
.
.
.
x
n
_

_
+
_

_
r
1
(t)
r
2
(t)
.
.
.
r
n
(t)
_

_
(2.7)
or in a more compact form as
X

= A(t)X+R(t), (2.8)
where X =
_

_
x
1
x
2
.
.
.
x
n
_

_
, A(t) =
_

_
a
11
(t) a
12
(t) a
1n
(t)
a
21
(t) a
22
(t) a
2n
(t)
.
.
.
.
.
.
.
.
.
.
.
.
a
n1
(t) a
n2
(t) a
nn
(t)
_

_
, and R =
_

_
r
1
(t)
r
2
(t)
.
.
.
r
n
(t)
_

_
.
18
A and R are called the coecient matrix and forcing function, respectively,
and they are considered to be continuous if their entries are continuous. Sometimes
R is called the source vector as well. If we set R=0 in Equation 2.8, the equation
becomes
X

= A(t)X (2.9)
and is called a homogeneous linear system; otherwise, it is known as a nonhomoge-
neous linear system.
The existence and uniqueness theorem [3] guarantees that as long as the co-
ecient matrix A and the forcing function (source vector) R are continuous in an
open interval (a,b) and K is an arbitrary constant n-vector for t
0
(a,b), then the
initial value problem
X

= A(t)X+R(t), Y(t
0
) = K (2.10)
has a unique solution.
We are more interested in dealing with a homogeneous linear system, where
R=0. Moreover, to nd a solution to a nonhomogeneous linear system, it is a stan-
dard practice to rst solve the homogeneous counterpart associated with the given
system (sometimes known as a complementary system) and then add it to the partic-
ular solution originating from the nonzero forcing function R of the nonhomogeneous
linear system. One of the most common methods that is used to nd the particular
solution of a nonhomogeneous linear system is called variation of parameters. Thus,
19
for a homogeneous linear system described by Equation 2.9, as long as A(t) is contin-
uous on an interval (a, b), there exists a unique solution at any single time instance in
(a,b). It is obvious that X=0 is always a solution to Equation 2.9. and this is known
as the trivial solution. Any other solution is nontrivial.
One of the basic theorems of homogeneous linear systems is that, if the vector
functions X
1
, X
2
, ..., X
m
dened on an interval (a, b) are solutions to X

= A(t)X, so
is any linear combination of them, i.e., Y = c
1
X
1
+ c
2
X
2
+ ... + c
m
X
m
, where the
arbitrary constants c
i
C for 1 i m [4].
2.3.1 Autonomous Homogeneous Linear System
When A(t) in Equation 2.9 is no longer a function of time, in other words, all
the entries of A are constant, we call it an autonomous homogeneous linear system.
Thus, we can rewrite the constant matrix A(t) as
A =
_

_
a
11
a
12
a
1n
a
21
a
22
a
2n
.
.
.
.
.
.
.
.
.
.
.
.
a
n1
a
n2
a
nn
_

_
, (2.11)
where a
ij
for 1 i n and 1 j n.
A constant matrix Ais always continuous on (, ); therefore, the existence
and uniqueness theorem guarantees the solutions for such systems to be dened for
the entire domain of the independent variable, t.
20
At this point it is worthwhile to refresh some of the linear algebra concepts,
e.g., eigenvector, eigenvalue, characteristic equation, etc., since they will be used in
nding the general solution of the linear system. Whenever a vector is multiplied
by a square matrix, usually the direction as well as the magnitude of the vector are
changed. This process is sometimes known as linear transformation. However there
may exist a vector as a special case, when it is multiplied by a given square matrix
only the magnitude of the vector is aected, but the direction remains unchanged.
This special vector is called an eigenvector of that particular matrix. If the direction
of the vector is ipped after multiplication, the magnitude is considered to be scaled
by a negative number and still ts the denition of an eigenvector. Thus, if V is an
eigenvector of the matrix A then
AV = V, (2.12)
where C and V C
n
. Here the constant is known as the eigenvalue associated
with the eigenvector V. If the eigenvalue = 1, the magnitude of the vector V also
stays unchanged, that is to say, the vector remained unchanged after the multipli-
cation. Likewise, if = 1, the magnitude of the vector stays unchanged but the
direction is ipped. If the eigenvalue > 1, the vector is stretched by that factor.
Likewise, if 0 < < 1, then the vector is shrunk. If < 0, the direction of the vector
is ipped along with the scaling of the magnitude by the absolute value of , and if
= 0, the vector turns into a point. In fact, the eigenvalue can also be a complex
21
constant, and in that case, the associated eigenvector will also be a complex vector,
meaning a vector having a real part and an imaginary part.
By rearranging Equation 2.12, we have
(A I)V = 0, (2.13)
where I is an identity matrix. It is obvious that, if (A I)
1
exists, no nonzero
eigenvector V can be found and, hence, no eigenvalue exists. Conversely, is an
eigenvalue if matrix (A I) is non-invertible. We know from linear algebra that a
matrix is non-invertible if and only if its determinant is zero, which leads us to the
characteristic polynomial det(AI) = 0 whose roots are essentially the eigenvalues.
Eigenvectors are called linearly independent if none of them can be derived from a
linear combination of the others. Otherwise, they are linearly dependent.
Again, one of the basic theorems in linear algebra states that, if the nxn
constant matrix A has n real (not necessarily distinct) eigenvalues
1
,
2
, ... ,
n
with
associated linearly independent eigenvectors V
1
, V
2
, ... ,V
n
, then the functions
X
1
= V
1
e

1
t
, X
2
= V
2
e

2
t
, ... X
n
= V
n
e

n
t
form the fundamental set of solutions for the homogeneous autonomous system X

=
AX. Thus, according to the previously mentioned theorem, the general solution of
22
the system is
X(t) = c
1
V
1
e

1
t
+ c
2
V
2
e

2
t
, ... + c
n
V
n
e

n
t
,
where the arbitrary constants c
i
C for 1 i n [4].
2.3.2 Planar Linear System
Many of the most important physical and engineering systems can be described
quite satisfactorily by second-order dierential equations of the form [x

1
x

2
]
T
=
F(t, x
1
, x
2
, x

1
, x

2
). In Section 2.2, it was shown how this equation can be rewrit-
ten as a set of two simultaneous rst-order dierential equations. If the system is
linear, we can rewrite the system as
_

_
x

1
x

2
_

_
=
_

_
a
11
(t) a
12
(t)
a
21
(t) a
22
(t)
_

_
_

_
x
1
x
2
_

_
+
_

_
r
1
(t)
r
2
(t)
_

_
. (2.14)
If this system is autonomous, we can eliminate the dependency on t from the
elements of the square matrix. Moreover, if this system is homogeneous, we can also
eliminate the forcing function (source vector). The resultant system is known as a
planar linear system with constant coecients and is of the form
_

_
x

1
x

2
_

_
=
_

_
a
11
a
12
a
21
a
22
_

_
_

_
x
1
x
2
_

_
. (2.15)
23
Sometimes the dependent variables of the planar system are renamed to (x,y)
to match the typical two-dimensional rectangular coordinate representation. In fact,
our system of interest is also described by a second-order dierential equation and,
hence, can be analyzed as a planar system. This is why it is important to refresh
some of the concepts related to the planar linear system that will be leveraged against
later on, although our ultimate system will be nonlinear in nature.
2.3.2.1 Phase Portrait
There is more than one way to represent a solution x = (t), y = (t) of a
planar system. One obvious method is to draw separate graphs of (t) and (t). In
many cases, this approach may be useful. However, it does not provide any relation-
ship between the two dependent variables, x and y. The plane whose coordinates
correspond to the values of these two dependent variables, x and y, is called a phase
plane. The curve in this plane described by the parametric equations x = (t) and
y = (t) is known as the trajectory or orbit of a solution. It is common to mark the
orbits with a direction of increasing t (independent variable). A representative col-
lection of orbits (trajectories) drawn on the phase plane is called a phase portrait. A
special orbit in an autonomous planar system is called the separatrix, which separates
orbits with dierent limiting behaviors [5].
2.3.2.2 Classication of Equilibrium Points
In Section 2.3, we have learned that the origin of the Euclidean coordinate sys-
tem is a trivial solution to any linear homogeneous system expressed by Equation 2.9.
24
We have also learned that this is a stationary solution (equilibrium point) composed
of a single point, that is to say, if a solution starts there, it will stay there forever.
The above fact holds true for the planar linear homogeneous system. If the system is
autonomous at the same time, the global behavior of the system can be characterized
based on the property of the constant coecient matrix A in Equation 2.15 around
this equilibrium point at the origin. The following two sections will demonstrate the
global behavior of the planar linear system on the phase plane containing some of the
distinct types of equilibrium points.
2.3.2.3 Equilibria Associated with Real Eigenvalue
When the two eigenvalues (
1
,
2
) of the constant coecient matrix A are
identical, the equilibrium point at the origin is called a star node. If the eigenvalues
are positive it is called an unstable star since any solution that starts at a point
other than the origin (equilibrium point) moves away radially from the origin as time
increases. The phase portrait for such a node is shown in Figure 2.1 (a). Conversely,
if the eigenvalues are negative, it is called a stable star since any solution that starts
at a point other than the origin (equilibrium point) approaches radially towards the
origin as time increases. The phase portrait for such a node is shown in Figure 2.1 (b).
The rate of motion in either case is proportional to the value of the eigenvalue.
When
2
>
1
> 0, there are two trajectories along the eigenvector L
1
asso-
ciated with the weaker eigenvalue
1
, one on each of the half-lines separated by the
origin. Likewise, there are two more trajectories along the eigenvector L
2
associated
with the stronger eigenvalue
2
, one on each of the half-lines separated by the origin.
25
(a) (b)
Figure 2.1: Phase Portrait for Star Node: (a) Star Sink (b) Star Source
For the above-mentioned four trajectories, the motion is away from the origin as time
increases, and they all approach the origin as t . All other trajectories (other
than the origin) move towards the origin and become asymptotically tangent to L
1
as t . Conversely, all other trajectories (other than the origin) move away
from the origin and become asymptotically parallel to L
2
as t . The equilibrium
point at the origin is called the unstable node or source. The phase portrait for such
a system is shown in Figure 2.2 (a), where L
1
is found to be the separatrix. On the
other hand, when 0 >
2
>
1
, there are two trajectories along the eigenvector L
1
associated with the weaker eigenvalue
1
, one on each of the half-lines separated by
the origin. Likewise, there are two more trajectories along the eigenvector L
2
associ-
ated with the stronger eigenvalue
2
, one on each of the half-lines separated by the
origin. The above-mentioned four trajectories approach the origin as t , and
the motion is away from the origin as time decreases. All other trajectories (other
than the origin) move towards the origin and become asymptotically tangent to L
2
as
26
(a) (b)
Figure 2.2: Phase Portrait for Node: (a) Unstable Node or Source (b) Stable Node
or Sink
t . Conversely, all other trajectories (other than the origin) move away from the
origin and become asymptotically parallel to L
1
as t . The equilibrium point
at the origin is called the stable node or sink. The phase portrait for such a system
is shown in Figure 2.2 (b), where L
2
is found to be the separatrix.
When
2
> 0 >
1
, there are two trajectories along the eigenvector L
1
associ-
ated with the negative eigenvalue
1
, one on each of the half-lines separated by the
origin. These trajectories approach the origin as t and move away from the ori-
gin as time decreases. Likewise, there are two more trajectories along the eigenvector
L
2
associated with the positive eigenvalue
2
, one on each of the half-lines separated
by the origin. They move away from the origin as time increases and approach the
origin as t . All other trajectories (other than the origin) move away from the
origin and become asymptotically tangent to L
1
as t . Conversely, all other
trajectories (other than the origin) move away from the origin and become asymp-
27
(a) (b)
Figure 2.3: Phase Portrait of Saddle Node for Two Dierent Sets of Eigenvalues
totically parallel to L
2
as t . Figure 2.3 (a) and (b) are the representative phase
portraits for such systems for two dierent sets of eigenvalues. The equilibrium point
at the origin is called a saddle node, and both L
1
and L
2
are the separatrices in this
case [4].
2.3.2.4 Equilibria Associated with Complex Eigenvalue
As mentioned in Section 2.3.1, the eigenvalues of the coecient matrix A can
be complex, and so is the corresponding eigenvector. Moreover, in this case, they
appear as conjugate pairs. In this section, the geometric property, i.e., the phase
portrait, of such a planar system is reviewed when the constant coecient matrix
A has a complex eigenvalue
1
= + j ( = 0). In this case, the associated
eigenvector p = u +jv is also complex, where the component vectors of the complex
eigenvector, namely u and v, are real vectors. Since no real eigenvector exists for
a complex eigenvalue unlike the previous cases we have discussed, the description
28
of the trajectories, i.e., the construction of the phase portrait in the phase plane, is
not straightforward. However, by introducing a new coordinate system in the x
1
x
2
plane, it is doable. To achieve this, we have to apply a special technique by nding
a p such that its component vectors are orthogonal, in other words u, v = 0. For
most cases, such p may not be readily available. If not, after nding the eigenvector,
we multiply it by a complex constant to obtain a p that satises the orthogonality
condition. Then, we normalize u and v so that U = u/ u and V = v/ v
to nd out the unit vectors for the new coordinate system. The origin of the new
rectangular coordinate system is the same as the origin of the old (x
1
, x
2
) system,
and the coordinates of a point in the new system are the distances in the direction of
U and V.
When = 0, i.e., is purely imaginary, the trajectories of this system are
ellipses that traverse periodically, as shown in Figure 2.4 (a) and (b). Sometimes, this
kind of phase portrait is called circle since a circle is a special case of an ellipse, when
the major and the minor axes become equal. The direction of rotation (clockwise or
counter clockwise) depends on the orientation of U and V. The equilibrium point at
the origin is sometimes called a center.
When > 0, the trajectories of such a system spiral away from the origin as
time increases and approach the origin as t . The equilibrium point at the
origin is sometimes knows as a spiral source. The direction of rotation depends on
the orientation of U and V. Figure 2.5 (a) and (b) shows two representative phase
portraits of this type of system.
29
(a) (b)
Figure 2.4: Phase Portrait of Center Node (a) Counter Clockwise Rotation
(b) Clockwise Rotation
(a) (b)
Figure 2.5: Phase Portrait of Spiral Source (a) Counter Clockwise Rotation
(b) Clockwise Rotation
30
(a) (b)
Figure 2.6: Phase Portrait of Spiral Sink (a) Counter Clockwise Rotation (b) Clock-
wise Rotation
When < 0, the trajectories of such a system spiral toward the origin as
t and spiral away from the origin as time decreases. The equilibrium point at
the origin is known as a spiral sink or focus. The direction of rotation depends on
the orientation of U and V. Like Figure 2.5, Figure 2.6 (a) and (b) shows two repre-
sentative phase portraits of this type of system [4].
2.4 Nonlinear System
When the component functions f
i
, 1 i n, in the vector function F(X) in
Equation 2.3 are not linear in x
1
, x
2
, ..., x
n
, the system is determined to be nonlinear.
We will restrict our discussion to autonomous systems, just like we did for the linear
system. That is, it is assumed that there is no time dependency in the component
functions. It is rare to nd an explicit solution for the nonlinear system, as opposed
to the linear system (autonomous homogeneous in particular), where we always nd
31
the explicit solution for any point on the phase plane. A prime reason for this is that
we do not have enough functions with particular names that can be used to write
down the solution of those systems in explicit form. Moreover, some basic properties,
e.g., the existence and uniqueness of a solution, that are quite obvious for linear
systems, are no longer apparent for nonlinear systems. Some nonlinear systems may
not have a solution for a given initial point. Even if they do, the solution may not
be dened for all t, unlike the linear case. In other words, the solution may blow up
(becomes innite) at some nite t. On the other hand, there may be innitely many
solutions to a system for a given initial point. Another important thing to notice in
the case of nonlinear systems is how they react to a small perturbation applied to
the initial condition or if the corresponding solutions vary continuously. All of these
properties are relatively clear in the case of linear systems. Hence, while dealing with
the nonlinear systems, often we are forced to resort to dierent means. Thus, we
may have to use the combination of analytic, geometric, and topological techniques to
derive the local as well as the global behavior of the solutions [6].
2.4.1 Existence and Uniqueness
As mentioned in the previous section, the existence and uniqueness of the
solutions for nonlinear systems may be neither global nor obvious, which is unlike
what we observe for linear homogenous systems. However, a fundamental theorem in
dierential equations states that, if E is an open subset in R
n
containing X
0
and if the
vector function F is continuous in its rst derivative on E, i.e., F(X) C
1
(E), then
there exists an a > 0 such that the initial value problem (IVP) for Equation 2.3 with
32
the initial value X(t = 0) = X
0
has a unique solution (t) on the interval [a, a].
For a detailed proof, see [7].
2.4.2 Equilibria
As mentioned earlier, the equilibria of a dynamical system are the stationary
points in the state space, where a solution stays forever if it originates there. In
other words, the derivatives of all the state variables with respect to the independent
variable, t, must be identically zero, i.e., X

= 0. Therefore, X
0
is an equilibrium point
for the nonlinear system described by Equation 2.3 if F(X
0
) = 0. This equilibrium is
called hyperbolic if F

(X
0
) = 0; otherwise, it is called nonhyperbolic. In Section 2.3.2.2,
we discussed dierent kinds of equilibria including sink, source, spiral sink, spiral
source, saddle, center and focus, along with their associated phase portraits for linear
homogeneous systems. Their denitions, in terms of limiting behavior, hold true for
nonlinear systems as well though their appearances may look somewhat irregular
than that of their counterparts for linear systems. The nonlinear system described in
Section 2.3.2.2 is said to have a sink, source or saddle at a hyperbolic equilibrium point
X
0
if the linear parts of the component functions f
i
, 1 i n, in the vector function
F evaluated at X
0
had only real negative eigenvalues , or only real positive eigenvalues
or both real positive and real negative eigenvalues, respectively [7]. However, if the
equilibrium point at X
0
is nonhyperbolic, then the above equivalence between the
nonlinear and linear system may not be true, and a dierent approach should be
followed to determine the local behavior of the solutions.
33
2.4.3 Stability
The stability of a nonlinear dynamical system can be looked at from dierent
levels. The highest level is termed as structural stability. The system described by
Equation 2.3 is structurally stable if slight modications to the right hand side of
the equation do not move the equilibrium points of the modied system away from
a close neighborhood of that of the original system or the nature of the equilibrium
points. For instance, a sink, source, or saddle remains a sink, source, or saddle
after the modication. Other features, such as the basin of a sink, stay bounded by
the same separatrice. [3]. Conversely, a system is known as structurally unstable if
the topological character of the trajectories (orbits) of the perturbed system becomes
radically dierent from that of the original system. The next level of extraction is
called the orbital stability. An orbit of a nonlinear system is stable if a suciently
small perturbation does not cause the orbit to move far away from the original orbit.
The orbit is called asymptotically stable if the perturbed orbit approaches the original
orbit as t . The nal level of stability is called the stability of equilibria. If an
equilibrium point is a sink it is stable. If an equilibrium point is a source or saddle
it is unstable. In the case of hyperbolic equilibria, this stability can be determined
relatively easily by observing the eigenvalues that originate from the Jacobian matrix
of the corresponding linear system evaluated at a particular equilibrium point. If the
equilibria are nonhyperbolic, a method developed by Lyapunov is used to determine
their stability.
34
2.4.4 Linearization
The linearization theorem states that, if X
0
is a hyperbolic equilibrium point
of the n-dimensional nonlinear system described by X

= F(X), then the nonlinear


ow is conjugate to the linearized system in a neighborhood of X
0
[6]. Therefore, it is
always a good start to analyze a nonlinear system by determining the equilibria and
describing the local behavior of the system near the equilibria through linearization.
Thus, if X
0
is a hyperbolic equilibrium point of X

= F(X), the local behavior can


be qualitatively determined by the behavior of the linear system X

= AX, where the


matrix A = DF(X
0
). By expanding the dierential operator D, we can rewrite the
matrix as
A =
_

_
f
1
x
1
f
1
x
2
f
1
x
3

f
1
x
n
f
2
x
1
f
2
x
2
f
2
x
3

f
2
x
n
.
.
.
.
.
.
.
.
.
.
.
.
f
n
x
1
f
n
x
2
f
n
x
3

f
n
x
n
_

_
X=X
0
, (2.16)
where f
i
, 1 i n, are the component functions of the vector function F and x
i
,
1 i n, are the state variables that constitute the state vector X.
The matrix Ais called the Jacobian matrix of the system, and AX = DF(X
0
)X
is called the linear part of F at X
0
[7].
35
If we expand the vector function F about the equilibrium point X
0
using a
Taylor series, we have
F(X) = DF(X
0
)(XX
0
) +
1
2
D
2
F(X
0
)(XX
0
)
2
+
1
6
D
3
F(X
0
)(XX
0
)
3
+... (2.17)
Thus, in a close neighborhood around the equilibrium point X
0
, where (XX
0
)
is very small, the higher-order terms can be neglected; this leads us to F(X) =
DF(X
0
)(XX
0
), which is a reasonable approximation of the vector function around
the equilibrium point.
If the eigenvalues of the matrix Ahave only negative real parts, the equilibrium
point of the nonlinear system is a sink. If the eigenvalues of the matrix A have
only positive real parts, the equilibrium point of the nonlinear system is a source.
However, if the eigenvalues of the matrix A have both positive and negative real
parts, the equilibrium point of the nonlinear system is a saddle node. Finally, if all
of the eigenvalues of the matrix A have identically zero real parts, the equilibrium
point is nonhyperbolic, and linearizing the system around such an equilibrium point
will not help in determining its local behavior. Instead, we will have to resort to the
Lyapunov method to determine the behavior of such a system.
2.4.5 Periodic Orbit and Limit Cycle
If (t) is a solution of a system such that (t + T) = (t), it is called a
periodic orbit (trajectory) with a period T. Periodic orbits are also known as closed
orbits since the trajectory returns to the same point in the state space after each
36
period. The limit cycle is a special type of closed orbit for which the neighboring
orbits approach it in the limiting sense. Thus, if the neighboring orbits approach the
limit cycle as t , it is called the stable limit cycle. On the other hand, if the
neighboring orbits approach it as t , it is called the unstable limit cycle.
2.4.6 Bifurcation
Let us consider a nonlinear planar system X

= F(X, ), where is a param-


eter. Usually the phase portrait of the system changes gradually as the parameter
is varied. However for some systems, there may be a drastic change in the phase
portrait when passes through a particular value
c
known as the critical value or
threshold. This phenomenon is called bifurcation. During this remarkable event, a
stable node may turn into a saddle, a node may disappear, a new node may appear, a
limit cycle or separatrix may disappear, a new limit cycle or separatrix may appear or
a trajectory that was in the basin of a sink may change its direction etc. Bifurcation
in a nonlinear system may be analyzed at two levels. It can be analyzed entirely
as the parameter value passes through the threshold by examining the stability of
equilibria, periodic orbit or other invariant set and this is called local bifurcation. In
contrast, the global bifurcation cannot be analyzed by simply examining the stability
of equilibria since it results from a collision of larger invariant sets with each other
and/or with equilibria. Saddle node bifurcation is the most important type of local
bifurcation encountered in the physical systems and involves two equilibria colliding
and annihilating each other. In the case of a discrete time system, this is known as fold
bifurcation. In a local bifurcation, blue skies bifurcation occurs when two equilibria
37
show up suddenly [8]. Bifurcation is a unique phenomenon that is clearly under-
stood in a planar nonlinear system. Unfortunately, there is no clear understanding of
bifurcation for a higher-dimensional system [3, 7, 9].
2.4.7 Planar System
As mentioned in Section 2.3.2, in a planar nonlinear system, there are only two
state variables in the system equation. The concepts of phase plane, phase portrait,
separatrix, etc., discussed earlier in this chapter are still valid for nonlinear systems.
Our system of interest, which will be introduced at the end of Chapter 3, is a planar
nonlinear system. During our analysis, the concepts mentioned in this chapter will
be used.
38
CHAPTER 3
SYSTEM MODEL FOR SECOND-ORDER PHASE LOCK LOOP
3.1 Denition of a Model
In essence, a model is an abstraction of something real. That thing can be an
idea, a condition, an existing system or a potential system. Mathematical modeling,
in particular, is a technique of translating one of the above items from an application
arena into tractable mathematical formulations whose theoretical and/or numerical
analysis provides useful insights, details and guidance. As for an existing or potential
system, the model represents its key characteristics and attributes in a broken-down
component level fashion. Often, models are analyzed in an eort to build a system
or to control, remediate or optimize its performance [10, 11].
The following are three fundamental ingredients in a mathematical model:
(a) VariableThe elements/quanitites that change; sometimes known as state vari-
ables or dependent variables
(b) ParameterThe elements/quanitites that do not change; sometimes known as
exogenous variables or constants
(c) Function The relationship among the variables and the parameters
39
Figure 3.1: Key Steps of Mathematical Modeling
There are also three key steps in formatting a mathematical model, as depicted
in Figure 3.1.
(I) Conceptual abstraction: In this step, a higher-level understanding of the
system on how it works is captured. Many a times, this abstraction is done in
pictorial form, using boxes, arrows and other graphics, and is known as a block
diagram, ow chart, etc.
(II) Dening the details: In this step, the variables and parameters are deter-
mined to describe the system as accurately as possible.
(III) Establishing the interaction: The nal step is to dene the mathematical
relationship betweeen the variables and the parameters.
At times, for a model to be realizable, some inessential features of the system
may have to be ignored. In the end, modeling is essentially a trade-o between
generality, realism and precision [12].
Based on some particular aspect of a system, a model can be divided into
many categories including the following:
40
(i) Deterministic vs. Stochastic
(ii) Static vs. Dynamic
(iii) Linear vs. Nonlinear
(iv) Continuous vs. Discrete
(v) Individual vs. Structured
(vi) Mechanistic vs. Statistical
(vii) Quantitative vs. Qualitative
In this work, the model of interest is continuous time, nonlinear and dynamical.
Mathematical models are not only used in the natural sciences (physics, chem-
istry, biology, etc.) and in the applied sciences (dierent disciplines of engineering)
but also in the area of social sciences (economics, sociology, psychology, etc.).
3.2 The Need for Models
There are many practical reasons to use mathematical models. The following
are some of the important ones [13, 14]:
(1) They allow us to determine the possible behavior of a system without having to
build it, which in many cases, could be huge and costly.
(2) They provide a more complete and thorough understanding of the system under
study.
41
(3) They lead to a detailed break-down of the system under study that facilitates the
analysis.
(4) They sometimes help clarify underlying assumptions in the design process.
(5) They sometimes suggest/identify the crucial data that should be collected while
conducting measurements.
(6) They may suggest systematic improvement/modication to the road map.
(7) They may suggest practical control strategy.
(8) They may generate data that cannot be collected from real-life measurements.
(9) They may predict the outcome under varying conditions by modifying the system
parameters.
3.3 Block Diagram of a PLL
As mentioned in Section 1.1, a phase lock loop (PLL) is a basic form of an
error tracking control system. It tracks the phase of a reference signal (bandpass,
most commonly) and tries to lock to it. Figure 3.2 shows the basic architecture of
a PLL system. The three component blocks in the block diagram are (1) a phase
comparator, (2) a loop lter and (3) a voltage controlled oscillator (VCO). Each of
these component blocks can be realized in many dierent ways.
The external reference input signal is commonly modeled as the sum of a
desired frequency
0
component, undesired frequency components and the additive
noise component. A voltage controlled oscillator with a center frequency at
0
is
42
Figure 3.2: Basic Architecture of a Phase Lock Loop System
chosen to track the phase of the reference input. Suppose
i
(t) and
v
(t) are the
instantaneous phase of the reference input signal and that of the VCO output signal,
respectively. The phase comparator produces a signal that contains a component
quantifying the phase error (t) = (
i
(t)
v
(t)), along with the random noise. This
signal is fed into the loop lter (usually lowpass), and the ltered output is applied
to the VCO as the control input with the hope that it adjusts the phase of the VCO
output signal so that the phase error diminishes monotonically.
If the loop is designed and operated properly, the phase error (absolute value)
becomes very small under the phase-locked condition. The process of achieving this
phase lock from the out-of-lock condition is know as pull-in. This pull-in may or may
not take place for a given PLL. Even if it happens, it may take an unreasonably long
time, that is not practical for the application for which it was intended for. Sometimes
additional circuitry, known as aiding circuitry, is introduced in the architecture to
expedite the acquisition (pull-in) in such situations [1, 1518] .
In the PLL literature, the default architecture shown in Figure 3.2 is called a
short-loop or baseband model, to be specic. There is another version, called long-
43
loop, found in some receivers that deals with an IF signal as reference. In the long-
loop construction, a stage of down conversion to IF followed by bandpass ltering is
followed by another stage of down conversion (to baseband) followed by a lowpass
ltering that is done prior to the VCO feed. Our work will be restricted to the
baseband loop (short-loop) model that operates in a noise-free environment.
3.4 Model for Component Blocks
There are three main component blocks in a basic (short-loop) PLL construc-
tion, as shown in Figure 3.2. The rst one in the block diagram is the phase compara-
tor, also known as phase detector (PD) in the PLL literature. The other two blocks
are called the loop lter and the VCO. Any one of these three blocks can be imple-
mented in various ways using various dierent technologies. Each form has its own
advantages and disadvantages. Commonly, the technical requirements of a given ap-
plication and its associated cost direct the choice. In the following sections, we will
introduce the mathematical models for each of the blocks followed by brief discussions
of some of their popular implementations.
3.4.1 Phase Detector
As mentioned in Section 3.3, the phase comparator, hereafter known as the
phase detector, produces a signal that contains a component quantifying the instan-
taneous phase dierence between the input reference and the VCO output signal.
Thus, a simple model for the phase detector, shown in Figure 3.3, could be expressed
as a function of the instantaneous phase dierence
44
Figure 3.3: Block Diagram of Phase Detector
g(
i

v
) = g(). (3.1)
The construction of the phase detector (PD) can be divided into two categories,
analog and digital. Even within each category, there are several ways to implement the
phase detector. Phase detectors that produce their outputs based on instantaneous
inputs are usually known as analog PDs and, hence, are memoryless. However, PDs
that produce outputs based on an event (e.g., zero crossing, positive or negative edge,
etc.) prior to the current time are known as digital PDs; therefore, memory elements
are required to implement them.
3.4.1.1 Analog Phase Detector
An analog phase detector is primarily a multiplier circuit implemented in
various ways. A basic multiplier block diagram is shown in Figure 3.4. The inputs to
the multiplier are two sinusoids, A
1
sin
1
(t) and A
2
cos
2
(t), where
1
(t) and
2
(t) are
their instantaneous phases and A
1
and A
2
are their amplitudes. Thus, the multiplier
45
Figure 3.4: Block Diagram of Multiplier Type Sinusoidal PD
output is x(t) = A
1
sin
1
A
2
cos
2
= K[sin(
1
+
2
) +sin(
1

2
)], where K is known
as the multiplier gain. After ltering out the double frequency term from the product
(we will see shortly that the loop lter and the VCO that follow the PD both act as
low pass lters), the multiplier output becomes
x(t) = Ksin, (3.2)
where the phase dierence, also known as the phase error, = (
1

2
). Because of
this output characteristic, such phase detectors are commonly known as sinusoidal
PDs, and when two input sinusoids are at quadrature in phase, the output of such a
phase detector goes to zero.
Three of the many popular implementations of multiplier circuits have been
chosen here for brief discussion. One of the most commonly used multipliers is the
four-quadrant doubly-balanced analog multiplier, which is attributed to Gilbert [19].
A simplied schematic for this type of multiplier is shown in Figure 3.5. Here, v
i
and v
o
are two inputs to the multiplier, where v
d
represents the output. I and K are
two constants based on the circuit parameters that regulate the maximum current
46
Figure 3.5: Schematic of Simplied Gilbert Multiplier as Sinusoidal PD
Courtesy: Phase-Locked Loops: Theory and Application; John L. Stensby
and the voltage in the circuit. After analyzing the circuit, the output turns out to be
proportional to the product of the two inputs as
v
d
=
I
K
2
[v
i
v
o
]. (3.3)
This kind of multiplier is very common in many legacy applications. For a detailed
derivation, see [15].
Another commonly found multiplier circuit for sinusoidal PDs is called a diode-
ring doubly-balanced mixer, as shown in Figure 3.6. They can operate at frequencies
well into the microwave region and do not contain any active components. Here,
47
Figure 3.6: Schematic of Diode-ring Doubly-balanced Mixer as a Sinusoidal PD with
Associated Waveforms
Courtesy: Phase-Locked Loops: Theory and Application; John L. Stensby
v
i
and v
o
are two inputs to the multiplier, where v
d
represents the output. As a
result of the presence of the transformers, impedance matching and termination may
become critical design issues. Although this type of mixer can work with a variety
of waveforms, a square wave for v
o
has been chosen for ease of illustration. The nal
output is the average value of v
d
, which is termed as v
o
. For information about the
detailed operation of the circuit, see [2022].
48
The two multipliers that we have discussed so far have sinusoidal output phase
error characteristic. The third kind of multiplier that is discussed here is best suited
for digital (rectangular) waveform and has a triangular output characteristic. The
original inputs to the multiplier can be sinusoidal and, if so, are converted to digital
by passing through a hard limiter. A simple XOR gate can perform the multiplica-
tion operation afterwards. In Figure 3.7, v
i
and v
o
represent two input rectangular
waveforms of the same frequency (period T), and v
d
represents the output of the
XOR gate. It is important to note here that, since the XOR gate acts as a multiplier,
the inputs are expected to be 50% duty cycle waveforms. In most applications, it
is not dicult to maintain this requirement. See [20] for details on the eect of the
duty cycle on such phase detectors operation. It is evident from Figure 3.7 that the
XOR output v
d
is a 50% duty cycle waveform when the two inputs are at quadrature
in phase, having a period that is half (T/2) of that of the inputs. The average of
v
d
, which is denoted as v
d
, in such a condition becomes zero when DC is removed.
This is why the corresponding phase error is dened as zero when the inputs are at
quadrature. The plot of v
d
is a triangular function of the phase error and is linear
over /2 /2 and centered at V/2 instead of zero. The removal of the DC
from the PD output may help or hurt based on the specic apparition.
Any multiplier type of PD output has a strong frequency component at twice
the input frequency. In some applications, the corruption introduced in the VCO
output spectrum by this component may be signicant, if not carefully removed.
On the other hand, the multiplier type of PDs can operate with noisier inputs as
compared to their digital counterparts.
49
Figure 3.7: Triangular Phase Detector Implemented by XOR Multiplier with Asso-
ciated Truth Table and Waveforms
Courtesy: Phase-Locked Loops: Theory and Application; John L. Stensby
3.4.1.2 Digital Phase Detector
Figure 3.8 illustrates the operation of a digital type of phase detector, also
known as sequential PD, implemented with an RS ip-op. The state transitions
take place at the falling edge of the inputs, as shown in the state diagram. Here, v
i
and v
o
represent two input rectangular waveforms of same frequency (period T), and
v
d
represents the output of the RS ip-op. The plot of the output of the PD v
d
,
which is the average of v
o
, is a saw-tooth function of the phase error and is linear
over and centered at V/2. Unlike the XOR multiplier PD, the phase
50
Figure 3.8: Saw-Tooth (Sequential) Phase Detector Implemented by RS Flip-Flop
with Associated State Diagram and Waveforms
Courtesy: Phase-Locked Loops: Theory and Application; John L. Stensby
error is dened as zero when the inputs are complements of each other, instead of at
quadrature.
This type of PD is more susceptible to noise since it is sensitive to missing
or poorly dened transitions of the inputs. Hence, these PDs are more suitable
for a high-SNR situation. Moreover, the presence of a strong undesirable frequency
component at the output of this type of PD poses a greater problem compared to its
multiplier type counterpart since it matches with the frequency of the inputs [15].
Notably, only sinusoidal and triangular waveform PDs have been considered
in this work.
51
Figure 3.9: Block Diagram of Loop Filter
3.4.2 Loop Filter
Mathematically, an nth-order lter, shown in Figure 3.9, can be modeled in
the Laplace domain as
F(s) =
a
m
s
m
+ a
m1
s
m1
+ ... + a
0
b
n
s
n
+ b
n1
s
n1
+ ... + b
0
, m n. (3.4)
In the time domain, the input x(t) and the output e(t) of the lter can be related by
taking the inverse transform of Equation 3.4 as
[b
n
d
n
dt
n
+ b
n1
d
n1
dt
n1
+ ... + b
0
]e(t) = [a
n
d
n
dt
n
+ a
n1
d
n1
dt
n1
+ ... + a
0
]x(t). (3.5)
Both active and passive loop lters are common in the PLLs, though active
lters implemented with OP-AMPs are more common in modern applications. A
rst-order active lter is shown in Figure 3.10, where the DC gain of the lter is
adjusted by choosing appropriate values for R
2
and R
1
. A rst-order passive lter
is shown in Figure 3.11. In most applications, R
1
is chosen to be much larger than
52
Figure 3.10: Schematic of a First-order Active Filter
Courtesy: Phase-Locked Loops: Theory and Application; John L. Stensby
Figure 3.11: Schematic of a First-order Passive Filter
Courtesy: Phase-Locked Loops: Theory and Application; John L. Stensby
R
2
so that
1

2
. Higher-order lters can be built by cascading multiple stages of
rst-order lters or by applying other standard lter design techniques.
However, only a rst-order loop lter will be considered in this work since,
combined with the VCO, it constructs a second-order PLL, which is the topic of this
research.
53
3.4.3 Voltage Controlled Oscillator
The voltage controlled oscillator (VCO) takes in a control voltage e(t) and
produces a sinusoid with instantaneous phase
v
(t). A popular way to mathematically
model such a VCO is by relating these two quantities as
d
v
dt
=
0
+ K
v
e, (3.6)
where constants
0
and K
v
are known as the center or quiescent frequency and gain
parameter of the VCO, respectively. Thus, the frequency of the VCO is proportional
to the control voltage e(t) around the center frequency. Redening the instantaneous
phases of the reference input to the PD and the output of the VCO by subtracting a
phase ramp term, sometimes known as a quiescent phase (
0
t), simplies the analysis.
Thus, the new instantaneous relative phase quantities into the PD and out of the VCO
becomes respectively

1
(t) =
i
(t)
0
t (3.7)
and

2
(t) =
v
(t)
0
t, (3.8)
where
i
(t) is the instantaneous phase of the reference input to the PD, as dened in
Section 3.4.1. With this, the phase error at the output of the PD is also redened as
(t) =
1
(t)
2
(t). (3.9)
54
Figure 3.12: Block Diagram of Voltage Controlled Oscillator
If we substitute Equation 3.8 in Equation 3.6, we have
d
2
dt
= K
v
e, (3.10)
which is represented in the VCO block diagram shown in Figure 3.12. If we take the
Laplace transform of Equation 3.10, assuming
2
(0) = 0, the transfer function of the
VCO becomes
H
vco
(s) =

2
(s)
E(s)
=
K
v
s
. (3.11)
Equation 3.11 implies that the response
2
(t) of the VCO is inversely proportional to
the frequency of the control voltage e(t). Such a rst-order lowpass lter is sometimes
known as an integrator. Thus, the VCO adds an order to the order of the loop lter
(which is usually another lowpass lter) when it comes to the overall order of the
PLL system.
The VCO is an essential component of a PLL. This component can be built
in many dierent ways; the choice depends mostly on the technical specications of
a given application as well as its cost, size, etc. Some of the prime technical features
include center frequency (
0
), phase noise and tunable range.
55

Figure 3.13: Astable Multivibrator Type Voltage Controlled Oscillator
Figure 3.13 shows a simplied schematic of a VCO implemented by an astable
multivibrator with two cross-coupled transistors, Q
1
and Q
1
. Alternating charging
(discharging) of their bases through the cross-coupling capacitors C is controlled by
the two current sources, Q
3
and Q
4
. This kind of VCO oers good linearity of
frequency over tuning voltage but poor spectral purity along with potential startup
problems. For detailed information on the operation of the circuit, see [17, 23].
Another popular way of constructing a VCO is by using a crystal oscillator, as
schematically shown in Figure 3.14. This is commonly known as a voltage controlled
crystal oscillator (VCXO). In this particular conguration, the quartz crystal (XTAL)
is used in its series-resonant mode. The LC tank circuit is appropriately tuned to
resonate at or near the equivalent series-resonant frequency (f
s
) of the crystal. A
loosely coupled link through the inductor L
1
provides positive feedback to facilitate
56
Figure 3.14: LC Tank Series Resonant Voltage Controlled Crystal Oscillator
oscillation. A level above (below) a nominal voltage at the tuning input causes the
resultant reactant of L
1
as well as the tuning diodes to become capacitive (inductive).
This varying reactance combined with the crystal reactance causes the circuit to
oscillate around f
s
. This type of VCO is reasonably stable and smaller in size. They
also oer excellent spectral purity, i.e., very small phase noise. However, their tuning
range is very limited (usually 0.1% of f
s
), and their performance optimization requires
special skill and expertise [15].
3.5 System Model of a Second-order PLL
Now that we have discussed the models for all of the component blocks of a
PLL system in the previous sections, we are ready to construct the model for a full-up
second-order PLL system. As suggested at the end of Section 3.4.2, only a rst-order
57
loop lter (lowpass) is required to build a second-order PLL since the VCO itself also
acts as a rst-order lowpass lter. The transfer function of a rst-order lowpass lter
in the Laplace domain can be written as
F(s) = K
1
s + a
s + b
, (3.12)
where K
1
is the DC gain and a and b are respectively the zero and the pole of
the lter. To ensure the stability of the loop, we also require a > b > 0. By taking
the Laplace transform of Equation 3.12, the time domain equation for the above lter
becomes
[
d
dt
+ b]e(t) = K
1
[
d
dt
+ a]x(t), (3.13)
where x(t) and e(t) are the time domain input and output of the lter, respectively, as
shown in Figure 3.9. Now, if
i
and
v
are the instantaneous phases of the reference
input to the PD and the output of the VCO, respectively, the phase error at the
output of the PD can be expressed as
=
i

v
. (3.14)
Thus, the output of the PD can be written in a generic form (without specifying a
particular waveshape) as
x(t) = K
2
g(), (3.15)
where K
2
is called the PD gain.
58
To simplify the analysis, we dene the relative phases of the reference input
to the PD and the output of the VCO as

1
(t) =
i

0
t = (
i

0
)t =

t (3.16)
and

2
(t) =
v

0
t, (3.17)
respectively, such that,
=
1

2
. (3.18)
Here
i
is the reference input frequency,
0
is the center frequency of the VCO,
0
t
is called the quiescent phase and

is called the detuning parameter of the PLL.


The exact same redenition process was carried out in the beginning of Section 3.4.3;
therefore, all of the equations derived in that section are equally applicable here.
If we insert e(t) from Equation 3.10 and x(t) from Equation 3.15 into Equa-
tion 3.13, we get
[
d
dt
+ b]
1
K
v
d
2
dt
= K
1
[
d
dt
+ a]K
2
g(). (3.19)
Now using Equation 3.18 yields
[
d
dt
+ b][
d
dt

d
1
dt
] = G[
d
dt
+ a]g(), (3.20)
where G K
1
K
2
K
v
is known as the closed loop gain of the system. In other words,
the gains of the individual blocks are lumped together in this single system parameter.
59
Figure 3.15: Block Diagram of a Second-order PLL
It is worth noting that G can be placed anywhere in the loop without aecting the
analysis. Thus, in the block diagram shown in Figure 3.15 for a second-order PLL,
G is arbitrarily included in the VCO block.
By inserting Equation 3.17 into Equation 3.20 and rearranging, we have
d
2

dt
2
+ [b + Gg

()]
d
dt
+ aGg() = b

, (3.21)
where g

()
d
d
g(). It is assumed that the closed loop gain G and the detuning
parameter

, are both positive. Thus, Equation 3.21 represents a second-order


autonomous system (constant coecients with respect to time) that describes the
PLL system of interest [15].
Finally, if the independent variable t (time) is normalized by the closed loop
gain so that = tG, the dependence of G can be removed, and the system equation
can be written as
d
2

d
2
+ [b

+ g

()]
d
d
+ a

g() = b

, (3.22)
where the gain normalized system parameters are
60
a

= a/G,
b

= b/G,
and

/G.
(3.23)
Based on the location of the loop lters pole, a second-order PLL can be
divided into two categories, commonly known as Type I and Type II PLLs. The
next two chapters contain a detailed study of second-order Type II and Type I PLLs,
respectively.
61
CHAPTER 4
SECOND-ORDER TYPE II PHASE LOCK LOOP
4.1 System Model of a Second-order Type II PLL
As suggested at the end of Section 3.5, if the pole of the loop lter is located at
the origin of the s-plane, the system represented by Equation 3.22 is called a second-
order Type II PLL. Thus, the transfer function for the rst-order loop lter with
unity DC gain is expressed, by setting b = 0 and K
1
= 1 in Equation 3.12, as
F(s) =
s+a
s
= 1 +
a
s
,
(4.1)
where a > 0 as dened in Section 3.5. Based on the above transfer function, such
lter can be interpreted as an adder followed by an integrator (pole at the origin).
Therefore, the loop parameter a is commonly known as integrator gain. Since, the
VCO acts as another integrator in the loop, a second-order Type II PLL is also known
as a second-order PLL with perfect integrator.
62
Figure 4.1: Block Diagram of a Second-order Type II PLL
Similarly, by setting b

= b/G = 0 in Equation 3.22, the system equation for a


second-order Type II PLL becomes
d
2

d
2
+ g

()
d
d
+ a

g() = 0, (4.2)
where a

and G are dened in Section 3.5.


Due to its satisfactory performance and relative simplicity, the second-order
Type II is the most widely used PLL model in the design practice.
In this work, we will restrict the analysis of the second-order Type II PLL to
the ones that only employ either a sinusoidal or a triangular PD, because they are
the two most commonly used kinds. Figure 4.1 shows a block diagram of a second-
order Type II PLL. The output of the PD is restricted accordingly to a sinusoid or
a triangular wave with respect to phase error . Moreover, the primary focus of this
work will be to quantify one of the very important parameters of such PLL, known
as pull-out frequency.
63
4.2 Denition of Pull-out Frequency
The pull-out frequency of a second-order Type II PLL is a very important
parameter in many applications. It quanties the ability of a phased-lock PLL to
track an abrupt jump in the input reference frequency. Quantitatively, the pull-out
frequency, denoted as
po
, is the maximum value of input reference frequency step
that can be applied to a phase-locked PLL, yet the loop re-locks to it without slipping
a cycle, i.e. phase error stays within 2 during the pull-in process. This parameter
also gives a qualitative idea about the convergence region, also known as lock-in range,
of the PLL without a cycle-slip.
In the literature, there exists an empirical formula deduced by A. J. Viterbi, to
calculate the pull-out frequency of a second-order Type II PLL containing a sinusoidal
PD [1, 15, 16]. However, there is no formula to calculate the same, for loops based on
a triangular PD. An exact formula to calculate the pull-out frequency of a 2nd-order
Type II PLL that contains a triangular PD is derived in the next section. After that,
an analytical approach is used to derive a formula to approximate the same in case
of sinusoidal PD, that outperforms the empirical formula for the practical range of
the integrator gain a.
4.3 Derivation of the Formula to Calculate Pull-out Frequency
The eort of derivation of the formula to calculate the pull-out frequency,
is dealt with separately for two dierent kinds of phase detectors. First, an exact
formula to calculate the pull-out frequency of a second-order PLL that contains a
64
triangular PD is derived, and then, an approximating formula to calculate the same is
derived for the loop that contains a sinusoidal PD. The process of derivation follows
the standard steps of analyzing a nonlinear planar system as touched upon in the
previous chapter, starting with converting the second-order system equation into a
set of simultaneous rst-order equations, then nding the equilibria, then linearizing
the system around them and nally demonstrating the global behavior of the system
using phase portrait.
Equation 4.2 represents a second-order autonomous nonlinear system. Using
the techniques described in Section 2.2, this system can be expressed in a set of
two simultaneous rst-order dierential equations as
d
d
=

d
= g

()

a

g(),
(4.3)
where a

> 0 is known as gain normalized integrator gain, g() and g

() are the
output of the PD and its derivative respectively, as described in the previous chapter.
Equation 4.3 can also be written in vector form as
_

=
_

_
0 1
0 g

()
_

_
_

_
+
_

_
0
a

g()
_

_
. (4.4)
By removing the independent variable from Equation 4.3, the system can be
written as the rst-order dierential equation
d

d
= g

() a

g()

. (4.5)
65
Figure 4.2: Block Diagram of a Second-order Type II PLL with Triangular PD
4.3.1 Type II PLL Employing Triangular PD
Figure 4.2 shows a block diagram of a second-order Type II PLL containing
a triangular PD, where the PD is implemented with two hard limiters and an XOR
gate.
Here, the sinusoidal reference input (v
ref
) and the VCO output (v
V CO
) are
fed to the hard limiters to convert them to rectangular waveforms feeding into the
XOR gate, which acts as a multiplier. If the reference input and the VCO output are
already in digital format, the need for the hard limiters goes away and the PD can
be realized with the XOR gate alone. Figure 4.3 shows the output characteristic of
the triangular PD.
The analysis is restricted to one period (/2 3/2) of g() (also g

())
without the loss of generality, since the phase portrait repeats after the same period.
Moreover, due to the discontinuity of the rst-order derivative of the PD output g

(),
it simplies the analysis signicantly, if one period is split in two segments as shown
in Figure 4.3. The Segment A, where /2 3/2, is the part of the period in the
66
Figure 4.3: Triangular Waveform Phase Detector Output Characteristic
PD output where the slope of g() is -1 and the Segment B, where /2 /2,
is the part where the slope of g() is +1.
Now, setting g() = ( ) and g

() = 1 in Equation 4.3 yields the


system (linear autonomous non-homogeneous) equations on Segment A as
d
d
=

d
=

+ a

( ).
(4.6)
The equilibria for the above system can be found by equating the right hand
side of Equation 4.6 to zero as described in Section 2.1.5. Thus,

= 0
( ) = 0.
(4.7)
Therefore the sole equilibrium point in Segment A is (,

) = (, 0), since = 0 in
Segment A. After translating the system by shifting the origin to this equilibrium
67
point the, the constant coecient matrix of the corresponding homogeneous system
can be found as
A
A
=
_

_
0 1
a

1
_

_
. (4.8)
If
A
is the eigenvalue of the matrix A
A
, the characteristic equation becomes
det(A
A

A
I) = 0

A
1
a

1
A

= 0

2
A

A
a

= 0,
(4.9)
where I is the identity matrix. If
A1
and
A2
are the roots of the above quadratic
equation, two distinct eigenvalues for the matrix A
A
are found to be

A1
=
1
2
(1

4a

+ 1)

A2
=
1
2
(1 +

4a

+ 1).
(4.10)
Since, the positive normalized integrator gain (a

> 0) implies that


A2
> 0 >

A1
, therefore the sole equilibrium point at (, 0) in Segment A is a saddle node,
as described in Section 2.3.2.3. Again, if V is the eigenvector corresponding to the
eigenvalue
A
such that, AV =
A
V, in case of saddle node as also mentioned in Sec-
tion 2.3.2.3, the separatrix coincides with the eigenvector. Therefore, the separatrix
68
can be found as
_

_
0 1
a

1
_

_
_

_
=
A
_


=
A
.
(4.11)
Thus, by shifting back the origin, the two half-line separatrices associated with

A1
can be expressed as

=
A1
= m

( ) =
1
2
(1

4a

+ 1),
where m


1
2
(1

4a

+ 1).
(4.12)
These two half-line trajectories (separatrices), separated by the saddle node, approach
the saddle node as time approaches innity, since
A1
< 0.
Similarly, by shifting back the origin, the two half-line separatrices associated
with
A2
can be expressed as

=
A2
= m
+
( ) =
1
2
(1 +

4a

+ 1),
where m
+

1
2
(1 +

4a

+ 1).
(4.13)
These two half-line trajectories (separatrices), separated by the saddle node, move
away from the saddle node as time approaches innity, since
A2
> 0.
Now, setting g() = and g

() = 1 in Equation 4.3 yields the system (linear


autonomous homogeneous) equations on Segment B as
d
d
=

d
=

.
(4.14)
69
The equilibria in this segment can be found by equating the right hand side
of Equation 4.14 to zero, as described in Section 2.1.5, that yields

= 0
= 0.
(4.15)
Thus, the sole equilibrium point in Segment B is (,

) = (0, 0), since /2 /2
in Segment B. The constant coecient matrix of this homogeneous system described
by Equation 4.14 is
A
B
=
_

_
0 1
a

1
_

_
. (4.16)
If
B
is the eigenvalue of the matrix A
B
, the characteristic equation becomes
det(A
B

B
I) = 0

B
1
a

1
B

= 0

2
B
+
B
+ a

= 0,
(4.17)
where I is the identity matrix. If
B1
and
B2
are the roots of the above quadratic
equation, two distinct eigenvalues for the matrix A
B
are found to be

B1
=
1
2
(1 j

4a

1)

B2
=
1
2
(1 + j

4a

1).
(4.18)
70
The imaginary part of the complex eigenvalue indicates the rotational feature of the
trajectories and the negative real part indicates that the trajectories approach the
equilibrium point as time approaches innity. Therefore, the sole equilibrium point at
(0, 0) in Segment B is a spiral stable node, commonly known as focus, as described in
Section 2.3.2.4. Further analysis, by decomposing the (real) component vectors of the
complex eigenvector associated with
B
into two orthonormal vectors in the phase
plane, it can be shown that, the direction of rotation of the trajectories is clockwise
around the focus [24].
As mentioned in the beginning of Section 2.4.4, the local behavior of a nonlin-
ear system around a non-hyperbolic equilibrium point is determined by the behavior
of the corresponding linearized system around that equilibrium point. Due to the
presence of (triangular) nonlinearity in g() as well as in g

(), our system of inter-


est described by Equation 4.3 is a nonlinear one. However, the segmented systems
described by Equations 4.6 and 4.14 are individually linear. Therefore, even the global
behavior of the overall system can be constructed by stacking up the behavior of the
segmented systems side by side.
Figure 4.4 is a representative phase portrait, spanned little over a period,
featuring the saddle node, spiral stable node (focus), separatrices etc, drawn with
Maple for a

= 0.5. Here, the foci are the global attracting points for the system,
except for the points residing on the separatrices. In other words, any point on
the phase plane eventually approaches one of the foci as time approaches innity,
except for the points that lie on one of the separatrices. It can be understood from
the phase portrait that, in the upper half plane, the trajectories that start below the
71
Figure 4.4: Typical Phase Portrait of a Second-order Type II PLL with Triangular
PD
separatrix will lock in the same cycle and the trajectories that start over the separatrix
will eventually lock in with one or more cycle slips. Contrast is true for the lower
half plane. It is interesting to notice the symmetry of the phase portrait shown on
Figure 4.4 for such a system. In other words, a trajectory remains a trajectory when
both and

are negated.
4.3.1.1 Derivation of an Exact Formula to Calculate Pull-out Frequency
Since quantitatively speaking, gain normalized pull-out frequency (

po
) is the
frequency (

) axis intercept by the separatrix as shown in Figure 4.4, the separatrix
should be traced back in time to nd that intercept. We have already found out
the four half-line separatrices in Segment A in the previous section. It is clear from
72
Figures 4.5, 4.6 and 4.7 that, the half-line separatrix with slope m

in particular, as
described by Equation 4.12, should be traced back in (reverse) time to nd where
it intersects the segment boundary = /2 (where segments A and B meet). This
intersection point will be used as the initial point for the particular solution (same
separatrix) of the system dened on Segment B.
Now, setting g() = and g

() = 1 in Equation 4.5 for Segment B and


rearranging, we have
d

d
= 1
a

(

/)
= f(

/), (4.19)
If we introduce a new variable u =

/ such that

= u, we have
d

d
= u +
du
d
. (4.20)
Substituting this in Equation 4.19 yields
u +
du
d
= 1
a

u
. (4.21)
After manipulating and rearranging, Equation 4.21 becomes a rst-order separable
ordinary dierential equation as
u
u
2
+ u + a

du =
d

. (4.22)
73
The general solution of Equation 4.22 can be found in a standard integral table
as
1
2
ln|u
2
+ u + a

|
1

4a

1
tan
1
(
2u + 1

4a

1
)
= ln|| + K

,
(4.23)
where K

is the constant of integration [25].


Now, at the boundary of the two segments A and B, it is evident from the
analysis of Segment A, presented in the previous section, that u =

/ = m

and
= /2. By inserting these in Equation 4.23, the constant K

for the particular


solution can be evaluated as
K

= ln|

2
| +
1
2
ln|m
2

+ a

4a

1
tan
1
(
2m

+ 1

4a

1
).
(4.24)
After rearranging and substituting for u =

/ in Equation 4.23, the general
solution becomes
1
2
ln|

2
+

+ a

2
|
1

4a

1
tan
1
(
2

+ 1

4a

1
) = K

, (4.25)
74
Finally setting = 0 in Equation 4.25 produces the desired

intercept (in
logarithmic scale) by that particular solution as
ln|

po
| =
1

4a

1
tan
1
() + K

=

2

4a

1
+ K

K. (4.26)
By raising both sides of Equation 4.26 as exponent of e, the pull-out frequency
turns out to be [26]

po
= e
K
(4.27)
where,
K = ln|

2
| +
1
2
ln|m
2

+ a

4a

1
tan
1
(
2m

+ 1

4a

1
) +

2

4a

1
(4.28)
and
m

=
1
2
(1

4a

+ 1). (4.29)
Note that, the pull-out frequency in Equation 4.27 is normalized by the system
closed loop gain G as shown in Figure 4.4. Therefore, it has to be multiplied by G to
get the absolute value.
4.3.1.2 Validation of the Exact Formula
Sample verications of the derived formula are presented here against numer-
ical results obtained from Matlab simulation, that cover the entire practical range of
the loop parameter a

i.e 0.1 a

2.0. The rst validation as depicted by Figure 4.5,


75
Figure 4.5: Numerical Validation of Derived Formula for Loop Parameter a

= 0.5
is done for a

= 0.5. This particular value of a

translates to = 0.707, a typical value


used for damping coecient in many linear control theory applications including PLL
systems, though in the analysis of this nonlinear model, it may not carry much rel-
evance. In this case, Equation 4.27 yields

po
= 2.65, a value that is in agreement
with the Matlab simulation result. The second validation is done for a

= 0.2, a low
value of the loop parameter and the value of the gain normalized pull-out frequency

po
= 2.15 is a match between the derived formula and the simulation, as depicted by
Figure 4.6. Finally, the last validation is done for a

= 1.5, a relatively higher value


of the loop parameter and

po
= 3.72 is again found consistent result between the
derived formula and the simulation, as depicted by Figure 4.7. The slight dierence
(from the second signicant digit after the decimal) between the value obtained from
the derived formula and the simulation, is due to the nite granularity (step size)
76
Figure 4.6: Numerical Validation of Derived Formula for Loop Parameter a

= 0.2
Figure 4.7: Numerical Validation of Derived Formula for Loop Parameter a

= 1.5
used in the Matlab ode45 function used to calculate the solution data points. It is
important to note here that, varying the closed loop gain G does not aect the vali-
77
Figure 4.8: Block Diagram of a Second-order Type II PLL with Sinusoidal PD
dation process. It also has no eect on the correctness of the result since the system
Equation 3.22 is already normalized by G in the beginning of the analysis.
4.3.2 Type II PLL Employing Sinusoidal PD
Figure 4.8 shows a block diagram of a second-order Type II PLL containing
a sinusoidal PD, where the PD is represented by a multiplier, commonly known as
mixer.
Here, the sinusoid reference input (v
ref
) and the VCO output (v
V CO
) are fed
directly to the multiplier. The output of the PD is a sinusoid of the phase dierence
between the two inputs as depicted by Figure 4.9.
The analysis is restricted to one period (0 2) of g() (also g

())
without the loss of generality, since the phase portrait repeats after the same period.
Now, setting g() = sin and g

() = cos in Equation 4.3 yields the system


(nonlinear autonomous non-homogeneous) equations for a second-order Type II PLL
78
Figure 4.9: Sinusoidal Waveform Phase Detector Output Characteristic
employing a sinusoidal PD as
d
d
=

d
=

cos a

sin .
(4.30)
The equilibria for the above system can be found by equating the right hand
side of Equation 4.30 to zero as described in Section 2.1.5, that yields

= 0
sin = 0
= n,
(4.31)
where (n = 0, 1, 2, 3, ...). Therefore, the two equilibrium points in the period of
interest are (,

) = (0, 0) and (,

) = (, 0).
Now, by linearizing the nonlinear system described by Equation 4.30 around
the rst equilibrium point at (,

) = (0, 0) using Equation 2.16, the constant coe-
79
cient matrix of the corresponding linear homogeneous system can be found as
A
0
=
_

_
0 1
a

1
_

_
. (4.32)
If
0
is the eigenvalue of the matrix A
0
, the characteristic equation becomes
det(A
0

A
I) = 0

0
1
a

1
0

= 0

2
0

0
a

= 0,
(4.33)
where I is the identity matrix. If
01
and
02
are the roots of the above quadratic
equation, the two distinct eigenvalues for the matrix A
0
are found to be

01
=
1
2
(1 j

4a

1)

02
=
1
2
(1 + j

4a

1).
(4.34)
The imaginary part of the complex eigenvalue indicates the rotational feature of
the trajectories and the negative real part indicates that the trajectories approach
the equilibrium point as time approaches innity. Therefore, the equilibrium point at
(0, 0) is a spiral stable node, commonly known as focus, as described in Section 2.3.2.4.
Further analysis, by decomposing the (real) component vectors of the complex eigen-
vector associated with
0
into two orthonormal vectors in the phase plane, it can
80
be shown that the direction of rotation of the trajectories is clockwise around the
focus [24].
Now, by linearizing the nonlinear system described by Equation 4.30 around
the other equilibrium point at (,

) = (, 0) using Equation 2.16, the constant
coecient matrix of the corresponding linear homogeneous system can be found as
A

=
_

_
0 1
a

1
_

_
(4.35)

1
=
1
2
(1

4a

+ 1)

2
=
1
2
(1 +

4a

+ 1).
(4.36)
Since the positive normalized integrator gain (a

> 0) implies that


2
>
0 >
1
, therefore the equilibrium point at (, 0) is a saddle node, as described in
Section 2.3.2.3. Again, if V is the eigenvector corresponding to the eigenvalue

such that, A

V =

V, in case of saddle node as also mentioned in Section 2.3.2.3,


the separatrix coincides with the eigenvector. Therefore, the separatrix can be found
as
_

_
0 1
a

1
_

_
_

_
=

.
(4.37)
Thus, by shifting back the origin, the two half-line separatrices associated with

1
can be expressed as
81

=
1
= m

( ) =
1
2
(1

4a

+ 1),
where m


1
2
(1

4a

+ 1).
(4.38)
These two half-line trajectories (separatrices), separated by the saddle node, approach
the saddle node as time approaches innity, since
1
< 0.
Similarly, by shifting back the origin, the two half-line separatrices associated
with
2
can be expressed as

=
2
= m
+
( ) =
1
2
(1 +

4a

+ 1),
where m
+

1
2
(1 +

4a

+ 1).
(4.39)
These two half-line trajectories (separatrices), separated by the saddle node, move
away from the saddle node as time approaches innity, since
2
> 0.
As mentioned in the beginning of Section 2.4.4, the local behavior of a nonlin-
ear system around a non-hyperbolic equilibrium point is determined by the behavior
of the corresponding linearized system around that equilibrium point. Figure 4.10 is a
representative phase portrait, spanned little over a period, featuring the saddle node,
spiral stable node (focus), separatrices, etc., drawn with Matlab for a

= 0.5. Like
the phase portrait of the PLL with a triangular PD, the foci are the global attracting
points for the system, except for the points residing on the separatrices. In other
words, any point on the phase plane eventually approaches one of the foci as time
approaches innity, except for the points that lie on one of the separatrices. It can
be understood from the phase portrait that, in the upper half plane, the trajectories
that start below the separatrix will lock in the same cycle and the trajectories that
82
Figure 4.10: Typical Phase Portrait of a Second-order Type II PLL with Sinusoidal
PD
start over the separatrix will eventually lock in with one or more cycle slip. Contrast
is true for the lower half plane. The phase portrait demonstrates similar symmetry
as observed in case of the PLL employing a triangular PD.
4.3.2.1 Derivation of a Formula to Approximate Pull-out Frequency
By substituting x = and y =

in Equation 4.30 and removing the indepen-
dent variable from it, we get
dy
dx
= cos x a

sin x
y
. (4.40)
83
The intent here is to nd the y-axis intercept by the separatrix. Therefore
integrating Equation 4.40 in that range produces


0
dy
dx
dx =


0
cos xdx a


0
sin x
y
dx. (4.41)
We know from the previous section that

() = y() = 0. Substituting this in the
above equation yields
y(0) = a


0
sin x
y
dx. (4.42)
Now the integral on the right hand side does not have a closed form solution, therefore
we will have to resort to an approximation technique to calculate it.
Lets expand the function y(x) around the saddle point at x = in an Asymp-
totic series (also known as Poincare expansion) [27] as
y(x) = y() +
y

()
1!
(x ) +
y

()
2!
(x )
2
+
y

()
3!
(x )
3
+ ..., (4.43)
Substituting y() = 0 and redening k
1

y

()
1!
=
1
1!
dy
dx
|
x=
, k
2

y

()
2!
=
1
2!
dy
2
dx
2
|
x=
,
k
3

y

()
3!
=
1
3!
d
3
y
dx
3
|
x=
and so on, Equation 4.43 can be rewritten as
y(x) = k
1
(x ) + k
2
(x )
2
+ k
3
(x )
3
+ .... (4.44)
Dierentiating both sides with respect to x yields
dy
dx
= k
1
+ 2k
2
(x ) + 3k
3
(x )
2
+ 4k
4
(x )
3
+ .... (4.45)
84
Likewise, the two trigonometric functions cos x and sin x can also be expanded re-
spectively around the same point, using Taylor series as
cos x = cos()
sin()
1!
(x )
cos()
2!
(x )
2
+
sin()
3!
(x )
3
+ ..., (4.46)
and
sin x = sin() +
cos()
1!
(x )
sin()
2!
(x )
2

cos()
3!
(x )
3
+ ..., (4.47)
Substituting Equations 4.44, 4.45, 4.46 and 4.47 in Equation 4.40 gives
k
1
+ 2k
2
(x ) + 3k
3
(x )
2
+ 4k
4
(x )
3
+ ....
= [cos()
sin()
1!
(x )
cos()
2!
(x )
2
+
sin()
3!
(x )
3
+ ...]
a

sin() +
cos()
1!
(x )
sin()
2!
(x )
2

cos()
3!
(x )
3
+ ...
k
1
(x ) + k
2
(x )
2
+ k
3
(x )
3
+ ....
(4.48)
After cross multiplication and rearrangement, Equation 4.48 becomes
[k
1
+ 2k
2
(x ) + 3k
3
(x )
2
+ 4k
4
(x )
3
+ ....][k
1
(x ) + k
2
(x )
2
+k
3
(x )
3
+ ....] = [1
1
2
(x )
2
+ ...][k
1
(x ) + k
2
(x )
2
+k
3
(x )
3
+ ....] + a

[(x )
1
6
(x )
3
+ ....].
(4.49)
85
Now, equating the coecients of (x ) from both sides of Equation 4.49
yields
k
2
1
= k
1
+ a

k
2
1
k
1
a

= 0
k
1
=
1
2
(1

4a

+ 1).
(4.50)
From Figure 4.10, it is clear that the separatrix of interest has a negative slope at
the saddle point (,

) (x, y) = (, 0) and per denition of k
1
, it is that slope.
Therefore, the value to keep is
k
1
=
1
2
(1

4a

+ 1). (4.51)
Equating the coecients of (x )
2
from both sides of Equation 4.49 yields
k
1
k
2
+ 2k
1
k
2
= k
2
k
2
(3k
1
1) = 0.
(4.52)
Since negative k
1
implies k
1
= 1/3, therefore k
2
= 0.
Equating the coecients of (x )
3
from both sides of Equation 4.49 yields
k
1
k
3
+ 2k
2
2
+ 3k
1
k
3
= k
3

k
1
2

a

6
. (4.53)
After inserting k
2
= 0 and simplifying the above equation, k
3
can be expressed as
k
3
=
a

+ 3k
1
6(1 4k
1
)
. (4.54)
86
Figure 4.11: Four-interval Simpsons Rule to Approximate Integrals
Thus, the function y(x) can be approximated by the rst three terms in Equa-
tion 4.44 as
y(x) k
1
(x ) +
a

+ 3k
1
6(1 4k
1
)
(x )
3
, (4.55)
where k
1
is given in Equation 4.51.
Now, we are ready to approximate the integral in Equation 4.42 using four-
interval Simpsons rule, that says

b
a
f(x)
h
3
[f(a) + 4f(x
1
) + 2f(x
2
) + 4f(x
3
) + f(b)], (4.56)
where h = (b a)/4, x
1
= a + h, x
2
= a + 2h and x
3
= a + 3h as shown in
Figure 4.11 [28]. In our case f(x) =
sin x
y(x)
, a = 0, b = , x
1
= /4, x
2
= 2/4 and
87
x
3
= 3/4. Substituting these and using Equation 4.55 for y(x) in Equation 4.58, we
get
y(0) a


12
_
sin(0)
y(0)
+
4 sin(

4
)
k
1
(
3
4
) +
a

+3k
1
6(14k
1
)
(
3
4
)
3
+
2 sin(
2
4
)
k
1
(
2
4
) +
a

+3k
1
6(14k
1
)
(
2
4
)
3
+
4 sin(
3
4
)
k
1
(

4
) +
a

+3k
1
6(14k
1
)
(

4
)
3
+
sin()
y()
_
.
(4.57)
Now, the nonzero pull-out frequency implies that

(0) = y(0) = 0. Therefore, the
rst term in Equation 4.57 drops out. The last term
sin()
y()
is in
0
0
form. Applying
LHospital rule, its limiting value at x =

can be evaluated as
cos(

)
y

)
=
1
k
1
. (4.58)
Thus, the pull-out frequency can be expressed as

po
=

(0) = y(0) a


12
[
4 sin(

4
)
k
1
(
3
4
) +
a

+3k
1
6(14k
1
)
(
3
4
)
3
+
2 sin(
2
4
)
k
1
(
2
4
) +
a

+3k
1
6(14k
1
)
(
2
4
)
3
+
4 sin(
3
4
)
k
1
(

4
) +
a

+3k
1
6(14k
1
)
(

4
)
3
+
1
k
1
],
(4.59)
where k
1
is given in Equation 4.51.
Note that, the pull-out frequency in Equation 4.59 is normalized by the system
closed loop gain G as shown in Figure 4.10. Therefore, it has to be multiplied by G
to get the absolute value.
88
4.3.2.2 Comparison between Derived and Existing Empirical Formula
As mentioned in Section 4.2, A. J. Viterbi empirically derived a formula to
calculate the pull-out frequency of a second-order PLL employing a sinusoidal PD,
which is

po
1.8(0.5 +

). (4.60)
However, in the literature, no analytical approach has been found to derive a formula
for

po
. Table 4.1 shows the comparison of the gain normalized pull-out frequency
for the entire practical range of a

, calculated using three dierent methods. The


second column represents the numerically calculated values of the pull-out frequency
using Matlab. The third column represents the value calculated using the empirical
formula described by Equation 4.60. The fourth column lists the percentage error
when the values calculated using the empirical formula are compared with the simu-
lation results. The fth column represents the value calculated by the derived formula
described by Equation 4.59, whereas the last column lists the percentage error when
these values are compared with the simulation results. It is clear from the table that,
the derived formula outperforms the empirical formula for a

0.4 and the percent-


age error when compared against the simulated results is less than 1.5%. For very
low value of the loop parameter a

< 0.4, empirical formula calculates the pull-out


frequency with better accuracy, though such values of a

are rare in practice. In the


PLL design practice, a

= 0.5 is a standard starting value for normalized integrator


gain. This comes from the damping coecient = 0.707, most commonly used value
in the linear control systems, which oers a satisfactory trade o between the over-
89
shoot and the settling time. Although, the linear model of PLL is not adequate, when
the absolute value of the phase error grows beyond 30
0
. Most commonly used values
for the normalized integrator gain stay within 0.51.5, to accommodate larger phase
error.
Table 4.1: Comparison of Pull-out Frequency from Three Dierent Methods
a

Simulation Empirical %err Formula %err


0.1 1.47 1.469 0.054 1.567 6.565
0.2 1.71 1.705 0.294 1.753 3.1
0.3 1.88 1.886 0.314 1.913 1.766
0.4 2.05 2.038 0.566 2.057 0.322
0.5 2.18 2.173 0.330 2.187 0.33
0.6 2.31 2.294 0.68 2.308 0.087
0.7 2.42 2.406 0.579 2.421 0.037
0.8 2.54 2.51 1.181 2.527 0.504
0.9 2.64 2.608 1.227 2.628 0.451
1.0 2.74 2.7 1.46 2.724 0.577
1.1 2.85 2.788 2.179 2.816 1.182
1.2 2.93 2.872 1.986 2.905 0.867
1.3 3.02 2.952 2.241 2.99 1
1.4 3.09 3.03 1.948 3.072 0.583
1.5 3.22 3.105 2.984 3.152 1.509
1.6 3.25 3.177 2.252 3.229 0.649
1.7 3.34 3.247 2.787 3.304 1.081
1.8 3.42 3.315 3.07 3.377 1.257
1.9 3.49 3.381 3.12 3.448 1.198
2.0 3.57 3.446 3.485 3.518 1.468
4.4 Comparison of Pull-out Frequency between Type II PLLs Employing
Triangular vs. Sinusoidal PD
The separatrix, that determines the pull-out frequency in case of a second-
order PLL, is shown in Figure 4.12. The separatrix of interest for a PLL employing
90
Figure 4.12: Separatrices for Sinusoidal and Triangular PD Based Type II PLLs
a sinusoidal PD and one employing a triangular PD, for the loop parameter a

= 0.5,
are drawn on the same set of axis. The area under the separatrix, that represents
the lock-in range for the PLL, is found slightly greater in case of the triangular
PD. Figure 4.13 shows a comparative plot of the gain normalized pull-out frequency
(

po
), for 0.1 a

2. The value of

po
for the PLL that contain a triangular
PD is found around 30% higher than one that contains a sinusoidal PD. Based on
the above observations it can be claimed that, a phase-locked second-order Type II
PLL that uses a triangular PD is more robust in terms of tracking abrupt changes in
reference phase and frequency than one that uses a sinusoidal PD as far as cycle-slip
is concerned. However, when choosing between the two for a particular application,
other constraints and features may have to evaluated as well [29].
91
Figure 4.13: Comparison of Pull-out Frequency for Sinusoidal and Triangular PD
Based Type II PLLs
92
CHAPTER 5
SECOND-ORDER TYPE I PHASE LOCK LOOP
5.1 System Model of a Second-order Type I PLL
As hinted at the end of Section 3.5, if the pole of the loop lter is not located
at the origin of the s-plane i.e. b

= 0, the system represented by Equation 3.22 is


called a second-order Type I PLL. Using the techniques described in Section 2.2, this
system can be expressed in a set of two simultaneous rst-order dierential equations
as
d
d
=

d
= [b

+ g

()]

+ [b

g()],
(5.1)
where a

> b

> 0 are known as loop lters gain normalized zero and pole respectively,
and, g() and g

() are the outputs of the PD and its derivative respectively, as


described in the previous chapters. By removing the independent variable from
Equation 5.1, the system can be written as a rst-order dierential equation
d

d
= [b

+ g

()] +
[b

g()]

. (5.2)
93
Figure 5.1: Block Diagram of a Second-order Type I PLL
Due to the presence of the non-zero pole in the loop lters transfer function,
the analysis of a second-order Type I PLL, becomes much harder than that of the
Type II loops.
Our brief study of the second-order Type I PLL in this chapter will be re-
stricted to the loops, that only employ either a sinusoidal or a triangular PD, because
they are the two most commonly used kinds, and, this is what has been done for the
Type II loops in the previous chapter. Figure 5.1 shows a block diagram of a second-
order Type I PLL. The output of the PD is accordingly restricted to a sinusoid or a
triangular wave with respect to phase error .
5.1.1 Type I PLL Employing Sinusoidal PD
Figure 5.2 shows a block diagram of a second-order Type I PLL containing a
sinusoidal PD, where the PD is represented by a multiplier or mixer.
Here, the sinusoidal reference input (v
ref
) and the VCO output (v
V CO
) are fed
directly to the multiplier. The output of the PD is a sinusoid of the phase dierence
between the two inputs as depicted by Figure 4.9.
94
Figure 5.2: Block Diagram of a Second-order Type I PLL with Sinusoidal PD
The analysis is restricted to one period (0 2) of g() (also g

())
without the loss of generality, since the phase portrait repeats after the same period.
Now, setting g() = sin and g

() = cos in Equation 5.1 yields the system


(nonlinear autonomous non-homogeneous) equations for a second-order Type I PLL
employing a sinusoidal PD as
d
d
=

d
= [b

+ cos ]

+ [b

sin ].
(5.3)
The equilibria for the above system can be found by equating the right hand
side of Equation 5.3 to zero as described in Section 2.1.5, that yields

= 0
sin =
b

= 2n +
0
or (2n 1)
0
,
(5.4)
95
where n = 0, 1, 2, 3, ... and
0
= sin
1
(
b

). Therefore, the two equilibrium


points in the period of interest are (,

) = (
0
, 0) and (,

) = (
0
, 0), where
0
0
< /2.
It is important to note here, that the equilibria are dependent on the loop
parameters a

, b

and

. Moreover, the equilibria exist if and only if |

h
where,

h

a

. Thus, the behavior of the system changes drastically, if any one


or any combination of those parameters keeps varying and cross that limit. This
phenomenon is known as bifurcation, as mentioned in Section 2.4.6. This phenomenon
never occurs in the Type II loops, discussed in the previous chapter.
Now, by linearizing the nonlinear system described by Equation 5.3 around the
rst equilibrium point at (,

) = (
0
, 0) using Equation 2.16, the constant coecient
matrix of the corresponding linear homogeneous system can be found as
A
s1
=
_

_
0 1
a

cos
0
b

+ cos
0
_

_
. (5.5)
If
sf
is the eigenvalue of the matrix A
s1
, after solving the characteristic
equation, it turns out that

sf
=
1
2
[b

+ cos
0
]

1
4
[b

+ cos
0
]
2
a

cos
0
. (5.6)
96
We can rewrite the above two eigenvalues as

sf1
=
1
2
[b

+ cos
0
]

1
4
[b

+ cos
0
]
2
a

cos
0

sf2
=
1
2
[b

+ cos
0
] +

1
4
[b

+ cos
0
]
2
a

cos
0
.
(5.7)
We know that cos(
0
) > 0, since 0
0
< /2. Therefore, if the quantity
under the square root of Equation 5.7 is positive, both the eigenvalues become real and
negative. This implies that the equilibrium point (
0
, 0) is a stable node, given that
it exists. On the other hand, if the quantity under the square root of Equation 5.7 is
negative, both the eigenvalues become complex with negative real parts. This implies
that the equilibrium point (
0
, 0) is a spiral stable node, also known as focus, given
that it exists. Further analysis of the corresponding complex eigenvector can prove
that the direction of the spiral is clockwise [24].
Likewise, by linearizing the nonlinear system described by Equation 5.3 around
the second equilibrium point at (,

) = (
0
, 0) using Equation 2.16, the constant
coecient matrix of the corresponding linear homogeneous system can be found as
A
s2
=
_

_
0 1
a

cos
0
(b

+ cos
0
)
_

_
. (5.8)
If
ss
is the eigenvalue of the matrix A
s2
, after solving the characteristic equa-
tion, it turns out that

ss
=
1
2
[b

cos
0
]

1
4
[b

cos
0
]
2
+ a

cos
0
. (5.9)
97
We can rewrite the above two eigenvalues as

ss1
=
1
2
[b

cos
0
]

cos
0
+
1
4
[b

cos
0
]
2

ss2
=
1
2
[b

cos
0
] +

cos
0
+
1
4
[b

cos
0
]
2
.
(5.10)
Again the equilibria exist if and only if |

h
where,

h

a

. Now,
a

> b

> 0 as dened in Equation 5.1 combined with cos(


0
) > 0, implies that both
the eigenvalues are real, and
ss2
> 0 >
ss1
. Therefore, the equilibrium point at
(
0
, 0) is a saddle node, given that it exists.
Since the existence of the equilibria of a Type I PLL depends on a

, b

and

,
the phase portrait has to be studied separately for dierent ranges of these parameters.
Again, the phase portrait can also be aected by the closed loop gain G of the system,
which has never been the case for Type II loops. However, we will be briey discussing
the behavior of the system for only high gain cases, since the low gain loops are rare
in practice.
As hinted earlier, the bifurcation parameter for a Type I loop can be dened
by any of the loop parameters or any combination thereof. The most common as
well as the practical way to dene it is to use the detuning parameter as

.
Two particular values of this parameter give rise to drastic change in the global
behavior of a second-order Type I loops employing a sinusoidal PD.
The rst one is known as pull-in range (
p
), where a special periodic orbit,
commonly called limit cycle, appears on the phase portrait. This phenomenon is
commonly known as saddle node bifurcation. An approximating formula to calculate
98
the gain normalized pull-in range for a high gain Type I PLL with a sinusoidal PD is
given as [15]

p
=

p
/G

2
a

. (5.11)
A numerical procedure to calculate it accurately can be found in [30].
The second one is known as half-plane pull-in frequency (
2
), where a periodic
limit cycle connects all the saddle nodes on the phase portrait. This phenomenon is
commonly known as separatrix cycle bifurcation, which only applies to high gain
cases. There has neither been an exact nor an approximating formula to calculate
the half-plane pull-in frequency for a Type I loop employing a sinusoidal PD.
5.1.1.1 Phase Portrait in Region I
This region is dened by the range || <

p
. Figure 5.3 is a representative
phase portrait of a second-order Type I PLL employing a sinusoidal PD in this region,
which is drawn for a

= 0.5, b

= 0.1, = 2.5, using a Matlab open source program,


called pplane8 (see appendix B). In this region, the behavior of the Type I PLL
resembles to that of the Type II loops in many respects. The foci are the global
attracting points for the entire phase plane except for the points on the separatrices.
For

< 0, the phase portrait swaps the upper half plane with the lower half plane.
99
Figure 5.3: Region I Phase Portrait of Type I PLL with Sinusoidal PD
5.1.1.2 Phase Portrait in Region II
This region is dened by the range

p
<||<

h
. Figure 5.4 is a representative
phase portrait of a second-order Type I PLL employing a sinusoidal PD in this region,
which is drawn for a

= 0.5, b

= 0.1, = 3.1. In this region, there exist two limit


cycles. The stable limit cycle moves higher and higher up on the phase portrait and
the unstable limit cycle moves closer and closer to the separatrix as increases. All the
trajectories above the stable limit cycle and between the limit cycles asymptotically
approach the stable limit cycle. On the other hand, all the trajectories below the
unstable limit cycle approach one of the foci. For

< 0, the phase portrait swaps


the upper half plane with the lower half plane. Thus, the stable and the unstable
limit cycles will show up in the lower half plane instead of the upper half plane.
100
Figure 5.4: Region II Phase Portrait of Type I PLL with Sinusoidal PD
5.1.1.3 Phase Portrait in Region III
This region is dened by the range || >

h
. Figure 5.5 is a representative
phase portrait of a second-order Type I PLL employing a sinusoidal PD in this region,
which is drawn for a

= 0.5, b

= 0.1, = 6. In this region, the behavior of the Type I


PLL drastically diers from that of the previous two regions. All the equilibrium
points on the phase plane disappear, therefore no pull-in is possible. However, there
exists a stable limit cycle, which asymptotically attracts all the trajectories on the
phase plane. For

< 0, the phase portrait swaps the upper half plane with the
lower half plane. Thus, the stable limit cycle will show up in the lower half plane
instead of the upper half plane.
101
Figure 5.5: Region III Phase Portrait of Type I PLL with Sinusoidal PD
Figure 5.6: Block Diagram of a Second-order Type I PLL with Triangular PD
5.1.2 Type I PLL Employing Triangular PD
Figure 5.6 shows a block diagram of a second-order Type I PLL containing
a triangular PD, where the PD is implemented with two hard limiters and an XOR
gate.
Here, the sinusoid reference input (v
ref
) and the VCO output (v
V CO
) are fed
to the hard limiters to convert them to rectangular waveforms feeding into the XOR
gate, which acts as a multiplier. If the reference input and the VCO output are
102
already in digital format, the need for the hard limiters goes away and the PD can
be realized with the XOR gate alone. Figure 4.3 shows the output characteristic of
the triangular PD.
The analysis is restricted to one period (/2 3/4) of g() (also g

())
without the loss of generality, since the phase portrait repeats after the same period.
Moreover, due to the discontinuity of the rst-order derivative of the PD output g

(),
it simplies the analysis signicantly, if one period is split into two segments as shown
in Figure 4.3. The Segment A, where /2 3/2, is the part of the period in the
PD output where the slope of g() is -1 and the Segment B, where /2 /2,
is the part where the slope of g() is +1 [31].
Now, setting g() = ( ) and g

() = 1 in Equation 5.1 yields the


(linear autonomous non-homogeneous) system equations on Segment A as
d
d
=

d
= [b

1]

+ [b

+ a

( )].
(5.12)
The equilibria for the above system can be found by equating the right hand
side of Equation 5.12 to zero as described in Section 2.1.5, that yields

= 0
=
b

.
(5.13)
Thus, the sole equilibrium point in Segment A is (,

) = (
b

, 0). It is important
to note here that, this equilibrium point is dependent on the loop parameters a

, b

and
103

. Moreover, this equilibrium point becomes indeterminate if, |


b

| >

2
, because
the Segment A is dened only on /2 3/2. Thus, the behavior of the system
changes drastically, if any one or any combination of those parameters keeps varying
and cross a certain limit. This phenomenon is known as bifurcation, as mentioned in
Section 2.4.6.
Thus, the constant coecient matrix of the corresponding linear homogeneous
system described by Equation 5.12 is found as
A
t1
=
_

_
0 1
a

(b

+ 1)
_

_
. (5.14)
If
ts
is the eigenvalue of the matrix A
t1
, after solving the characteristic equa-
tion, it turns out that

ts
=
1
2
[b

+ 1]

1
4
[b

+ 1]
2
+ a

. (5.15)
We can rewrite the above two eigenvalues as

ts1
=
1
2
[b

+ 1]

1
4
[b

+ 1]
2
+ a

ts2
=
1
2
[b

+ 1] +

1
4
[b

+ 1]
2
+ a

.
(5.16)
Since, a

> b

> 0, as dened in Equation 5.1, implies that both the eigenvalues


are real and
ts2
> 0 >
ts1
, therefore, the equilibrium point at (
b

, 0) is a saddle
node.
104
Now, setting g() = and g

() = +1 in Equation 5.1 yields the (linear


autonomous non-homogeneous) system equations on Segment B as
d
d
=

d
= [b

+ 1]

+ [b

].
(5.17)
The equilibria for the above system can be found by equating the right hand
side of Equation 5.17 to zero as described in Section 2.1.5, that yields

= 0
=
b

.
(5.18)
Thus, the sole equilibrium point in Segment B is (,

) = (
b

, 0). It is important to
note here that, this equilibrium point is dependent on the loop parameters a

, b

and

. Moreover, this equilibrium point becomes indeterminate if, |


b

| >

2
, since the
Segment B is dened only on /2 /2. Thus, the behavior of the system
changes drastically, if any one or any combination of those parameters keeps varying
and cross a certain limit. This phenomenon is known as bifurcation, as mentioned in
Section 2.4.6.
Thus, the constant coecient matrix of the corresponding linear homogeneous
system described by Equation 5.17 is found as
A
t2
=
_

_
0 1
a

(b

1)
_

_
. (5.19)
105
If
tf
is the eigenvalue of the matrix A
t2
, after solving the characteristic equa-
tion, it turns out that

tf
=
1
2
[b

+ 1]

1
4
[b

+ 1]
2
a

. (5.20)
We can rewrite the above two eigenvalues as

tf1
=
1
2
[b

+ 1]

1
4
[b

+ 1]
2
a

tf2
=
1
2
[b

+ 1] +

1
4
[b

+ 1]
2
a

.
(5.21)
If the quantity under the square root of Equation 5.21 is positive, both the
eigenvalues become real and negative. This implies that the equilibrium point (
b

, 0)
is a stable node. On the other hand, if the quantity under the square root of Equa-
tion 5.7 is negative, both the eigenvalues become complex with negative real parts.
This implies that the equilibrium point (
b

, 0) is a spiral stable node, also known


as focus. Further analysis of the corresponding complex eigenvector can prove that,
the direction of the spiral is clockwise [24].
Since the existence of the equilibria of a Type I loop depends on a

, b

, and

, the phase portrait has to be studied separately for dierent ranges of these
parameters. Again, the phase portrait can also be aected by the closed loop gain
(G) of the system. However, we will be briey discussing the behavior of the system
for only high gain cases, since the low gain loops are rare in practice.
Unlike, the loops employing a sinusoidal PD, there has neither been an exact
nor an approximating formula to calculate the pull-in range for a Type I loop em-
106
Figure 5.7: Region I Phase Portrait of Type I PLL with Triangular PD
ploying a triangular PD. However, there exists an exact formula deduced by J. L.
Stensby to calculate the half-plane pull-in frequency for a Type I loop employing a
triangular PD [31].
5.1.2.1 Phase Portrait in Region I
This region is dened by the range || <

p
. Figure 5.7 is a representative
phase portrait of a second-order Type I PLL employing a triangular PD in this region,
which is drawn for a

= 0.5, b

= 0.1, = 2.0, using a Matlab open source program


called pplane8 (see appendix B). In this region, the behavior of the Type I PLL
resembles to that of the Type II loops in many respects. The foci are the global
attracting point for the entire phase plane except for the points on the separatrices.
For

< 0, the phase portrait swaps the upper half plane with the lower half plane.
107
Figure 5.8: Region II Phase Portrait of Type I PLL with Triangular PD
5.1.2.2 Phase Portrait in Region II
This region is dened by the range

p
< || <

h
. Figure 5.8 is a represen-
tative phase portrait of a second-order Type I PLL employing a triangular PD in
this region, which is drawn for a

= 0.5, b

= 0.1, = 4.4. In this region, there exist


two limit cycles. The stable limit cycle moves higher and higher up on the phase
portrait and the unstable limit cycle moves closer and closer to the separatrix as
increases. All the trajectories above the stable limit cycle and between the limit
cycles asymptotically approach the stable limit cycle. On the other hand, all the
trajectories below the unstable limit cycle approach one of the foci. For

< 0, the
phase portrait swaps the upper half plane with the lower half plane. Thus, the stable
and the unstable limit cycles will show up in the lower half plane instead of the upper
half plane.
108
Figure 5.9: Region III Phase Portrait of Type I PLL with Triangular PD
5.1.2.3 Phase Portrait in Region III
This region is dened by the range || >

h
. Figure 5.9 is a representative
phase portrait of a second-order Type I PLL employing a sinusoidal PD in this region,
which is drawn for a

= 0.5, b

= 0.1, = 8. In this region, the behavior of the Type I


PLL drastically diers from that of the previous two regions. All the equilibrium
points on the phase plane disappear, therefore no pull-in is possible. However, there
exists a stable limit cycle, which asymptotically attracts all the trajectories on the
phase plane, giving rise to a situation, commonly known as false lock. For

< 0,
the phase portrait swaps the upper half plane with the lower half plane. Thus, the
stable limit cycle will show up in the lower half plane instead of the upper half plane.
109
5.2 Comparison between Type I and Type II Loops Based on the Phase
Portrait
We have presented detail study of a second-order Type II PLL employing
two of the most commonly used phase detectors, namely, sinusoidal and triangular
phase detector, in the previous chapter. In this chapter, we have presented a brief
study of a second-order Type I PLL employing the same phase detectors. It is clear
at this point that the analysis as well as the behavior of the latter is much more
complex than those of the former. We will touch upon a few reasons, based on their
respective phase portraits, in what follows.
The equilibria for Type II loops are independent of any of the loop parameters,
and therefore, they are xed. Whereas, the equilibria for the Type I loops are depen-
dent on the loop parameters, and therefore, they move as any one of the parameters
varies. They even may disappear for certain values of those parameters.
There is a symmetry in the phase portrait for Type II loops, meaning that,
a trajectory remains a trajectory if both the axes are negated. However, in case of
Type I loops, no such symmetry exists.
In case of Type II loops, the entire phase plane constitutes the region of con-
vergence for the foci, except for the separatrices. However, this may be true for
Type I loops up to certain range of the loop parameters values. For another range
of the loop parameters (or some combination thereof), phase portrait may have more
than one cells, exhibiting distinctly dierent behaviors. For another range, the whole
system may fail to produce pull-in all together and so on.
110
In case of Type II loops, there is no drastic change in the phase portrait as the
loop parameters are varied. There is no appearance/disappreacne of equilibrium point
and/or limit cycle. Whereas, Type I loops exhibit two instances of such bifurcation.
In the PLL literature, one is known as saddle node bifurcation and the other one is
known as separatrix cycle bifurcation.
The phase-locked condition of a Type II PLL implies both zero phase and zero
frequency error. However, in case of Type I PLL, it only implies zero frequency error.
For nonzero detuning parameter, phase error is always nonzero.
False lock does not occur in Type II loops. However, in case of Type I loops,
false lock can occur due to potential presence of limit cycle on the phase portrait,
giving rise to beat node.
The close loop gain has almost no aect in the study of Type II loops, it only
acts as a scale factor. Whereas, the high gain Type I loops behave dierently than
the low gain ones, even when the other loop parameters are kept unchanged.
111
CHAPTER 6
CONCLUSION AND FUTURE WORK
6.1 Conclusion
This dissertation provides a detailed study of a second-order Type II PLL
model and a brief study of a second-order Type I PLL model, employing two of the
most widely used phase detectors, namely a sinusoidal and triangular phase detectors.
The rst three chapters provided the necessary background for this study, dealing with
continuous time nonlinear models.
At the end of Chapter 4, we have seen that the analysis as well as the global
behavior of the Type II loops are remarkably simpler than those of their Type I
counterparts. However, the Type I loops do not oer the performance improvement
to the same degree. Therefore, Type II loop models remain the most popular ones
to be used in the PLL design practice. It is worth mentioning here that, in terms of
simplicity of analysis, the linear models outweigh the nonlinear models, though they
are inadequate when the phase error grows more than 30
0
.
As mentioned before, the pull-out frequency is one of the most important
parameters of the Type II PLL, especially, when it comes to cycle slip. It is important
to note that the concept of the pull-out frequency is not valid in the case of the Type I
112
loops. For Type I models, the equilibrium points along with the separatrix are not
xed on the phase plane; rather they move continuously around with the varying
detuning parameter.
A PLL may lose its lock and give rise to cycle slip, when an unintended (in-
tended) circumstance corrupts (changes) the reference input and/or the VCO output
signal. The cycle slip can be extremely detrimental to many applications. Particu-
larly in digital communications, where word (packet) synchronization is necessary, a
miss-count between the reference clock cycle and the VCO clock cycle may destroy
the receive word (packet) alignment. In other cases, the cycle slip may mislead the
VCO to tune its output even in the opposite direction. It is quite common to model
all of the unintended signal interference as AWGN noise although, the intended inter-
ference mostly comes from modulation. It is worth reiterating that all the analyses,
presented in this work, were done on the premise of a noise-free environment, which
does not represent reality. However, instead of including the AWGN noise in the
nonlinear model as a forcing function, the eect of noise can be investigated with the
help of separatrix and pull-out frequency in the case of the Type II loops, that are
studied in this work.
The random noise, radiated or conducted, can corrupt the reference input
and/or the VCO output signal. Although the VCO output is supposed to be very
stable, low frequency power supply noise and/or ground bounce can corrupt its spec-
tral purity if the digital and the analog grounds are not isolated properly on the circuit
board. In any case, if either one of the inputs to the phase detector is corrupted by
the unwanted noise, the loop may lose its lock. The two separatrices on the two half
113
planes in the phase plane, spanned over a 2 period (saddle node to saddle node), can
determine the convergence region of the PLL. As long as the phase and the frequency
error, caused by the random noise, stays within that bound, the loop will regain its
lock without a cycle slip. If the corruption of the input(s) caused by the noise is too
large to induce a phase and/or frequency error to cross the separatrix, one or more
cycles will be missed in the pull-in process. In other words, the region of convergence
provides a measure of the noise-immunity of a Type II PLL.
As for the loss of lock caused by the modulation (intended), the pull-out
frequency becomes the key parameter to guarantee zero cycle slip in relocking. In
the frequency modulation schemes, when a carrier is changed from one frequency (to
which the loop was already locked) to another, it indeed causes an abrupt jump in
the input reference at the zero phase error and zero frequency error condition of a
Type II loop. The pull-out frequency is a direct measure of the maximum allowable
jump that guarantees no cycle slip in the reacquisition of the new frequency. Thus in
a spread spectrum system, as long as the modulation bandwidth is kept within the
pull-out frequency and the loop bandwidth, a Type II PLL can lock to any frequency
in that band and will not lose a cycle in the pull-in process. Likewise, in multi-
carrier (OFDM), frequency hopping, etc., systems, many discrete frequencies from a
specied band are assigned to the users as carriers, in a predened pattern or in a
pseudo random manner and, have to be locked at the receiver front end in the same
cycle. A second-order Type II PLL can adapt to this abrupt (yet frequent) change
in input reference and relock to those frequencies without slipping a cycle, as long as
the frequency band (from which the carriers are chosen) is lower than the pull-out
114
frequency of the PLL. Thus, the pull-out frequency of a Type II PLL can determine
the width of that band, or, the specied band can guide a Type II PLL design eort
to secure a pull-out frequency that can guarantee zero cycle slip in reacquisition.
A direct application of the pull-out frequency can be found in the construction
of the FSK modem (demodulator, to be specic). If the spacing between the Mark
and the Space frequency is kept below the pull-out frequency, an FSK demodulator
can be constructed with only a second-order Type II PLL. In fact, the PLL-based
FSK demodulators have been built in a single chip (XR2211, NJM2211 etc.) for more
than three decades.
In this research, we derived two new formulae to calculate this very important
parameter. We derived an exact formula to calculate the pull-out frequency of a
second-order Type II PLL with a triangular PD and analytically derived a formula to
approximate the same to an acceptable accuracy for a Type II PLL with a sinusoidal
PD. With the formulae derived in this work, this important parameter, which has
an aect on many applications, can be computed easily by using a modern scientic
calculator. The formulae may also be computed in the Matlab command prompt.
Until now, the conventional method to compute the pull-out frequency was by solving
the system equation numerically for a particular trajectory, namely the separatrix,
which is much more involved. In addition, the rst order derivative discontinuity in
the triangular PD output waveform g

() added another level of diculty to this


numerical endeavor.
115
6.2 Future Work
We would like to put similar eort into deriving formulae to calculate the pull-
out frequency for the Type II loops that employ other kinds of phase detectors, such
as saw-tooth or, tanlock waveforms [16].
Another important parameter for a second-order PLL in many applications
is the settling time. Again this can be dened in two ways, just like the cycle slip.
However, we are interested in the settling time not related to the power up or reset
situation, but when the loop loses its lock because of the situations mentioned in the
previous section. As hinted earlier, the pull-out frequency sets the upper limit for the
settling time, which guarantees no cycle slip. Since we have a formula to calculate
the pull-out frequency for Type II loops, in the near future, we intend to calculate or
estimate the associated settling time.
Equally important and equivalent of pull-out frequency in the case of Type I
loops are the pull-in range and half-plane pull-in frequency, depending on the systems
operating region. No formula exists to calculate or approximate the pull-in range of
a second-order Type I PLL employing a triangular PD. Likewise, there is no formula
to calculate or approximate the half-plane pull-in frequency of a loop that employ a
sinusoidal PD. We would like to investigate those open issues as well.
Finally, we would like to extend our work to the discrete time system models,
which can be used directly in the design of a digital phase lock loop (DPLL).
116
APPENDICES
117
APPENDIX A
MAPLE CODE TO GENERATE PHASE PORTRAIT FOR A TYPE II
PLL WITH TRIANGULAR PD
Figure A.1 contains the Maple code that generates the phase portrait for a
second-order Type II PLL employing a triangular PD. Table A.1 shows the redeni-
tion of the system variables and the loop parameters in the Maple code. Non zero
b in the maple code produces the phase portrait for a Type I PLL, whereas b = 0
produces the phase portrait for a Type II PLL.
Table A.1: Redenition of System Variables and Loop Parameters in Maple Code
System equation Maple code
t
x

y
d
d
di(x(t),t)
d

d
di(y(t),t)
a

a
b

c
g() g
g

() gprime
118
Figure A.1: Maple code to generate Phase Portrait of Type II PLL with Triangular
PD
119
APPENDIX B
PPLANE8 GUI SET UP TO GENERATE PHASE PORTRAIT
The pplane8 is a Matlab based phase portrait generator program, created
and supported by Dr. John C. Polkins at Rice University, Houston, TX. He has
gracefully made it available to the public for academic research. Table B.1 shows the
redenition of the system variables and the loop parameters in the Maple code. Non
zero b

produces the phase portrait for a Type I PLL, whereas setting b

= 0 produces
the phase portrait for a Type II PLL.
Table B.1: Redenition of System Variables and Loop Parameters in pplane8
System equation pplane8
x

y
d
d
x

d
y

c
Figure B.1 shows the pplane8 GUI set up for a particular run to generate
phase portrait for a PLL with sinusoidal PD.
Figure B.2 shows the pplane8 GUI set up for a particular run to generate
phase portrait for a PLL with triangular PD.
120
Figure B.1: pplane8 GUI Set up to Generate Phase Portrait for a PLL with Sinu-
soidal PD
Figure B.2: pplane8 GUI Set up to Generate Phase Portrait for a PLL with Trian-
gular PD
121
APPENDIX C
MATLAB CODE TO GENERATE SEPARATRIX FOR TYPE II PLL
Table C.1 shows the redenition of the system variables and the loop param-
eters in the Maple code.
Table C.1: Redenition of System Variables and Loop Parameters in Matlab Code
System equation Matlab code
t
x

y
Figure C.1 contains the Matlab code that produces the separatrix for a Type II
PLL containing a sinusoidal PD. In this particular run, the value for the normalized
integrator gain a

in the system Equation 4.3, is set to 0.5. g3 in the code represents


the system equation.
Figure C.2 contains the Matlab code that generates the sepratrix for a second-
order Type II PLL. First half of the code produces the separatrix for a second-order
Type II PLL containing a triangular PD. It is again split in two parts. The rst part
generates the separatrix for Segment A and the second part does it for Segment B,
as dened in Figure 4.3. g1 and g2 in the code represent the system equation on
122
Figure C.1: Matlab code to generate Separatrix for a PLL with Sinusoidal PD
Segment A and on Segment B, respectively. In this particular run, the value for the
normalized integrator gain a

in the system Equation 4.3, is set to 0.5.


123
Figure C.2: Matlab code to generate Separatrix for a PLL with Triangular PD
124
REFERENCES
[1] Heinrich Meyr and Gerd Ascheid. Synchronization in Digital Communications:
Phase-, Frequency-Locked Loops, and Amplitude Control, volume 1. Wiley-
Interscience, 1990.
[2] Robert Rosen. Dynamical System Theory in Biology, volume 1. Wiley-
Interscience, second edition, 1970.
[3] John H. Hubbard and Beverly H. West. Dierential Equations: A Dynamical
System Approach, page 15. Springer-Verlag, 1995.
[4] William F. Trench. Elementary Dierential Equations with Boundary Value
Problems. Wadsworth Group, 2001.
[5] Bruce P. Conrad. Dierential Equations with Boundary Value Problems: A
Systems Approach. Prentice Hall, 2003.
[6] Morris W. Hirsch, Stephen Smale, and Robert L. Devaney. Dierential Equa-
tions, Dynamical Systems and an Introduction to Chaos. Elsevier Academic
Press, second edition, 2004.
[7] Lawrence Perko. Dierential Equations and Dynamical Systems. Springer-
Verlag, 1991.
[8] Nezavisimyi Moskovskii. Surveys in Modern Mathematics. Campbridge Univer-
sity Press, London,UK, 2005.
[9] Ferdinand Verhulst. Nonlinear Dierential Equations and Dynamical Systems.
Springer-Verlag, 1990.
[10] Norman T. Bailey. The Mathematical Theory of Infectous Diseases. Grin,
London, second edition, 1975.
[11] Akira Okubo and Simon A. Levin. Diusion and Ecological Problems: Modern
Perspective. Springer-Verlag, second edition, 1980.
[12] Richard Levins. The strategy of model building in population biology. American
Scientist, 54:421431, 1966.
[13] Ray Hillborn and Marc Magel. The Ecological Detective: Confronting Models
with Data. Princeton University Press, 1997.
125
[14] Stephen P. Ellner and John Guckenheimer. Dynamic Models in Biology. Prince-
ton University Press, 2006.
[15] John L. Stensby. Phase-Locked Loops: Theory and Applications. CRC Press,
1997.
[16] Floyd M. Gardner. Phaselock Techniques. Wiley-Interscience, second edition,
1979.
[17] Alain Blanchard. Phase-locked Loops: Application to Coherent Receiver Design.
Wiley-Interscience, 1976.
[18] William F. Egan. Phase-Lock Basics. Wiley-Interscience, 1998.
[19] S. Salivahanan and V. S. K. Bhaaskaran. Linear Integrated Circuits. Tata
McGraw-Hill, New Delhi, India, 2008.
[20] Dan H. Wolaver. Phase-locked Loop Circuit Design. Prentice Hall, New York,
1991.
[21] Herbert L. Krauss, Charles W. Bostian, and Fredrick H. Raab. Solid State Radio
Engineering. John Willey and Sons, New York, 1980.
[22] Donald G. Fink and H. Beaty. Standard Handbook for Electrical Engineers.
McGraw-Hill, New York, 15th edition, 2006.
[23] Ronal E. Best. Phase-locked Loops: Design, Simulation and Applications.
McGraw-Hill, fourth edition, 1999.
[24] Aleksandr A. Andronov, Aleksandr A. Vitt, and Semen E. KhaikinMorris. The-
ory of Oscillators. Dover Publications, second edition, December 1987.
[25] D. Zwillinger. CRC Mathematical Tables and Formulae. CRC Press, thirty rst
edition, 2002.
[26] Abu-Sayeed Huque and John L. Stensby. An exact formula for the pull-out
frequency of a 2nd-order Type II phase lock loop. Communications Letters, in
press.
[27] Arthur Erdelyi. Asymptotic Expansions. Dover Publications, 1956.
[28] Francis B. Hildebrand. Introduction to Numerical Analysis. Dover Publications,
second edition, 1974.
[29] Abu-Sayeed Huque and John Stensby. Comparison of pull-out frequencies of
2nd-order Type II PLLs. In Proceedings of IEEE SoutheastCon, 2011.
[30] John L. Stensby. Saddle node bifurcation at a nonhyperbolic limit cycle in a
phase locked loop. J. Franklin Institute, 330,5, 1993.
[31] John L. Stensby. An exact formula for the half-plane pull-in range of a phase
locked loop. J. Franklin Institute, 348,3:671684, 2011.
126

S-ar putea să vă placă și