Vigo University

Department of Electronic Technology

Dissertation submitted for the degree of

European Doctor of
Philosophy in Electrical Engineering

Contributions to Grid-Synchronization
Techniques for Power Electronic Converters

Author: Francisco Daniel Freijedo Fernández

Director: Jesús Doval Gandoy

Vigo, 9 June 2009

Ph.D. Committee Members:

Carlos Martínez Peñalver

Carlos Couto
Enrique Acha
Emilio Bueno
Pedro Roncero
Grzegorz Benysek
Marco Liserre
Paolo Mattavelli
Enrique Romero Cadaval
Enrique Barreiro

9 June 2009
University of Vigo
Vigo, Spain
Vita

Francisco Daniel Freijedo Fernández was born in Ponteareas (Pontevedra), Spain, in

December 1978. He received the M.Sc. in Physics in 2002 from the University of Santiago
de Compostela.

Since 2003 to 2005 he worked as software developer. From February 2005 he is part
time assistant lecturer of the University of Vigo. From June 2008 to October 2008 he
had a research stay at the Power Engineering Group of the University of Glasgow. He
had a research stay at the University of Alcalá in February 2009. His research interests
include digital control of Power Electronics Converters, specially for Power Quality and
renewable energy applications.

He has authored or coauthored about 20 technical papers in the field of power elec-
tronics, 8 of which are IEEE Transactions grade. He is member of the IEEE since 2007
and also of the IEEE Industrial Electronics and IEEE Power Electronics Societies.
Contributions to Grid-Synchronization Techniques for
Power Electronic Converters
Francisco Daniel Freijedo Fernández

(Abstract)
Real time grid-synchronization techniques are studied in this dissertation. The syn-
chronization block is a crucial part of the controller in grid-connected power converters.
The phase-angle of the voltage/current vector fundamental component at the point of
common coupling (PCC) should be tracked online in order to control the energy transfer
between the power converter and the ac mains. Synchronization algorithms have been
evolving since the first analog zero-cross detectors to current high performance digital
implementations.

An in-depth review of the state-of-art in grid synchronization is contributed in the

Chapter 2 of this PhD dissertation. In some ways, it could be said that the grid syn-
chronization state-of-art has been evolving to avoid malfunctions due to power quality
phenomena. The Ainsworth proposal of using a voltage controlled oscillator (VCO) in-
side the control loop of a High Voltage Direct Current (HVDC) successfully dealt with
the novel, at that time, harmonic instability problem. Subsequently, analog phase locked
loops (PLL) were proposed to be used as measurement blocks, which provide frequency
adaptation in motor drives. Since then, controllers using analog and digital PLLs have
been proposed in all kind of applications such as controlled rectifiers, Flexible ac Transmis-
sion Systems (FACTS) and power line conditioners. Nowadays, the fields of Distributed
Power Generation Systems (DPGS), power line conditioners and traction applications
have an important demand of optimized synchronization algorithms. Even though the
main part of chapter 2 is only dedicated to synchronization with fundamental component
(and positive-sequence), the issue of harmonic and negative sequences extraction is also
dealt with. The advanced synchronization/extraction methods based on synchronous ref-
erence frames (SRFs) and moving average filters (MAFs), which are major contributions
of this PhD dissertation, are introduced in this chapter.

Digital PLLs are the most employed synchronization algorithm, mainly due to their
simple implementation, good frequency adaptation and acceptable filtering versus tran-
sient response trade-off. Chapter 3 is devoted to digital PLLs design. In a general way,
PLLs with a low bandwidth (low-gain PLLs) are required when handling with distorted
voltages. It is analytically demonstrated that low-gain PLLs have more trade-offs than
high-gain PLLs (e.g. PLLs for communications): it is not possible to optimize the settling
time for a phase-jump without getting slower the PLL response to frequency variations.
Existing tuning methods do not take into account low-gain features, which may result in
non-optimum designs. An intuitive tuning methodology based on inspection of frequency-
domain diagrams is contributed in this chapter. Contrary to the other existing tuning
methods, it takes into account low-gain dynamics. It is assured an optimized performance
in the presence of any kind of disturbances in the grid. In the second part of the chapter,
the DCO based on a RC oscillator is presented: the digital model of a sinusoidal oscil-
lator is implemented, instead of explicit trigonometric functions. This solution reduces
the needed digital resources without reducing the performance, which could be specially
useful for DSP-based control of power converters.

SRF synchronization algorithms based on moving average filters (SRF-MAF), which

have been introduced in chapter 2, provide a quite fast transient response and very good
harmonic cancellation. However, its main drawback with respect of PLLs is its lack of
frequency adaptation working open loop. This problem is also observed in other schemes
based on stochastic filtering. A novel synchronization technique based on filtering in the
SRF provided in chapter 4. The main novelty with respect to previous works is the fact
that it is frequency adaptive working open loop. This is achieved by cascading two kinds
of finite impulse response (FIR) filters: a moving average FIR filter providing an excellent
harmonic cancellation and a predictive filter compensating for the negative phase (delay)
introduced by the first one. The resulting schemes are named Predictive SRF-MAFs.
It is proved that predictive SRF-MAFs have a very high performance under distorted
conditions, specially when compared with PLLs.

Chapter 5 provides a novel approach in the implementation of Proportional+Resonant

(PR) current controllers, specially suitable for active power filters (APFs) compensating
for high order harmonics. The difference equation to implement each resonant controller
has been obtained from the impulse invariant discretization method. An explicit estima-
tion of the grid-frequency deviation, which is obtained from a PLL, is employed to update
online the PR current controller. An excellent frequency adaptation is achieved with this
technique. Technical details of the whole discrete-time controller are provided. A 1 kVA
rated lab prototype has been implemented and tested. Experimental tests show how the
system performs both in steady-state and under different transients. Perfect tracking in
steady-state for a wide range of grid-frequencies is achieved. Regarding transient response,
different tests were programmed such as load transients (change in the load), sudden and
large changes in the ac mains frequency (frequency steps) and strong source voltage faults
(voltage sags with phase-jump). Experimental results prove how the system is robust and
fast responding to all of these possible real operation faults. Even though this chapter is
focused in APF applications, the proposed PR controller is suitable for a vast range of
industrial applications such as power line conditioners (including APFs), DPGS, FACTS,
controlled rectifiers, motor drives, etc.

Chapter 6 is devoted to review all the contributions of this PhD thesis. A brief outline
of the future work is also provided.
Acknowledgments

First of all, I would like to thank my director, Dr. Jesús Doval, for his patience,
guidance and support. He introduced me to the interesting field of power electronics, and
more specifically in the issue of grid-synchronization.

Secondly, I would like to express my appreciation to my colleagues of the Electronic

Technology Department of the University of Vigo. Specially, I would like to thank Dr.
Oscar López for the technical feedback, his continuous support and his help with LATEX.

I would like to thank Professor Enrique Acha for giving me the opportunity to be
at the University of Glasgow, and for his invaluable help during the development of this
dissertation.

I would also like to thank University of Vigo for the grant I was given to stay in
Glasgow.

I wish to thank Dr Emilio Bueno for the good moments at IECON 2008, and for
having invited me to the University of Alcala.

I would like to thank the PhD Committee Members for their comments which help to
improve the overall quality of this dissertation.

I wish to thank all the colleagues and students for the good moments we shared.
Especially, I would like to mention Enrique Ortega, Pablo, Jano, Jacobo, Renato and
Alejandro.

I would like to thank Carmenza by her English classes and for the good moments we
shared with Miguel.

Finally, I would like to dedicate this dissertation to my family, specially to my parents

for having encouraged me to study and to Silvinha for her unconditional love.
 Á miña familia,
Silvinha, Ma Victoria, Bernardo, Marivi, Carmiña e Amelia
Contents

Contents i

List of Figures ix

List of Tables xvii

List of Abbreviations and Acronyms xix

Nomenclature xxiii

1 Introduction 1

1.1 Objectives and Motivation of this Work . . . . . . . . . . . . . . . . . . . . 1

1.2 Background on Electric Power Definitions . . . . . . . . . . . . . . . . . . 1

1.2.1 Power Definitions under Sinusoidal Conditions . . . . . . . . . . . . 2

1.2.1.1 Electrical Variables with Phasor Notation . . . . . . . . . 4

1.2.2 Power Definitions under Non-Sinusoidal Conditions . . . . . . . . . 6

1.2.2.1 Power Definitions in the Frequency Domain by Budeanu . 7

1.2.2.1.1 Apparent Power S: . . . . . . . . . . . . . . . . . 7

1.2.2.1.2 Active Power P : . . . . . . . . . . . . . . . . . . 7

1.2.2.1.3 Reactive Power Q: . . . . . . . . . . . . . . . . . 8

i
ii

1.2.2.1.4 Distortion Power D . . . . . . . . . . . . . . . . . 8

1.2.2.1.5 Power Factor λ . . . . . . . . . . . . . . . . . . . 8

1.2.2.1.6 Displacement Factor cos (φ) . . . . . . . . . . . . 8

1.2.2.1.7 Distortion Factor cos(γ) . . . . . . . . . . . . . . 8

1.2.2.2 Power Definition in the Time Domain by Fryze . . . . . . 9

1.2.2.2.1 Active Power Pw . . . . . . . . . . . . . . . . . . 9

1.2.2.2.2 Apparent Power Ps . . . . . . . . . . . . . . . . . 9

1.2.2.2.3 Active Power Factor λ . . . . . . . . . . . . . . . 9

1.2.2.2.4 Reactive Power . . . . . . . . . . . . . . . . . . . 9

1.2.2.2.5 Reactive Power Factor λq . . . . . . . . . . . . . 9

1.2.2.2.6 Active Voltage vw and Active Current iw . . . . . 9

1.2.2.2.7 Reactive Voltage vq and Reactive Current iq . . 10

1.2.3 Power Definitions in Three-Phase Systems . . . . . . . . . . . . . . 10

1.2.3.1 Power in Balanced Three-phase Systems . . . . . . . . . . 10

1.2.3.2 Analysis under Unbalanced Conditions . . . . . . . . . . . 11

1.2.3.2.1 ”Per-phase” Calculation: . . . . . . . . . . . . . . 11

1.2.3.2.2 ”Aggregate rms Value” Calculation: . . . . . . . 12

1.2.3.3 Power Definitions and Alternatives in Power Electronic

Converters Control . . . . . . . . . . . . . . . . . . . . . . 12

1.3 Background on Symmetrical Components . . . . . . . . . . . . . . . . . . . 12

1.3.1 Alternatives in Implementation . . . . . . . . . . . . . . . . . . . . 16

1.4 Background on Power Quality . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.4.1 Transients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1.4.1.1 Impulsive Transient . . . . . . . . . . . . . . . . . . . . . 17

 iii

1.4.1.2 Oscillatory Transients . . . . . . . . . . . . . . . . . . . . 17

1.4.2 Short-duration Variations . . . . . . . . . . . . . . . . . . . . . . . 17

1.4.2.1 Interruptions . . . . . . . . . . . . . . . . . . . . . . . . . 18

1.4.2.2 Sags (Dips) . . . . . . . . . . . . . . . . . . . . . . . . . . 18

1.4.2.3 Swells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

1.4.3 Long Duration Variations . . . . . . . . . . . . . . . . . . . . . . . 18

1.4.3.1 Over-Voltage . . . . . . . . . . . . . . . . . . . . . . . . . 18

1.4.3.2 Under-Voltage . . . . . . . . . . . . . . . . . . . . . . . . 19

1.4.3.3 Sustained Interruptions . . . . . . . . . . . . . . . . . . . 19

1.4.4 Voltage Imbalance . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

1.4.5 Waveform Distortion . . . . . . . . . . . . . . . . . . . . . . . . . . 19

1.4.5.1 Dc Offset . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

1.4.5.2 Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1.4.5.3 Inter-Harmonics . . . . . . . . . . . . . . . . . . . . . . . 20

1.4.5.4 Notching . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1.4.5.5 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.4.6 Voltage Fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.4.7 Power Frequency Variations . . . . . . . . . . . . . . . . . . . . . . 21

1.5 Power Electronic Converters and Applications . . . . . . . . . . . . . . . . 22

1.5.1 Thyristor Converters . . . . . . . . . . . . . . . . . . . . . . . . . . 22

1.5.1.1 Three-phase Full Rectifier . . . . . . . . . . . . . . . . . . 22

1.5.1.2 Applications of Thyristor Converters . . . . . . . . . . . . 25

1.5.1.2.1 High Voltage dc Transmission . . . . . . . . . . . 25

iv

1.5.1.2.2 Flexible ac Transmission Systems . . . . . . . . . 27

1.5.1.2.3 Ac Motor Drives . . . . . . . . . . . . . . . . . . 27

1.5.2 Pulse-Width-Modulated Converters . . . . . . . . . . . . . . . . . . 28

1.5.2.1 Overview of Modulation Techniques . . . . . . . . . . . . 28

1.5.2.1.1 Sinusoidal PWM . . . . . . . . . . . . . . . . . . 29

1.5.2.2 Voltage Source Converters . . . . . . . . . . . . . . . . . . 30

1.5.2.3 Applications of PWM Converters . . . . . . . . . . . . . . 34

1.5.2.3.1 Ac Motor Drives . . . . . . . . . . . . . . . . . . 34

1.5.2.3.2 Uninterruptible ac power supplies . . . . . . . . . 34

1.5.2.3.3 Power Line Conditioners . . . . . . . . . . . . . . 35

1.5.2.3.4 Flexible ac Transmission Systems . . . . . . . . . 35

1.6 Control of Power Electronic Converters and Synchronization . . . . . . . . 36

1.7 Structure of the Document . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

1.7.1 Chapter 2: State-of-the-Art in Grid Synchronization . . . . . . . . . 41

1.7.2 Chapter 3: Dynamics Study of Low-Gain PLLs . . . . . . . . . . . 41

1.7.3 Chapter 4: Predictive based SRF-MAF Synchronization Algorithms 42

1.7.4 Chapter 5: Frequency Adaptive PR Current Controller for APFs . . 42

1.7.5 Chapter 6: Conclusions and Future Work . . . . . . . . . . . . . . . 43

2 State-of-the-Art in Grid Synchronization 45

2.1 Methods based on Zero-Cross Detection . . . . . . . . . . . . . . . . . . . 46

2.1.1 The First High-Voltage Direct Current Systems . . . . . . . . . . . 48

2.1.1.1 Constant-α Method . . . . . . . . . . . . . . . . . . . . . 48

2.1.1.2 Inverse-Cosine Method . . . . . . . . . . . . . . . . . . . . 49

 v

2.1.2 Drawbacks of Zero-cross Detectors . . . . . . . . . . . . . . . . . . 49

2.1.2.1 Improvements in Zero-cross Detectors . . . . . . . . . . . 50

2.1.2.2 Harmonic Instability of Zero-cross based HVDCs . . . . . 50

2.2 Phase-Locked Oscillator Control Systems . . . . . . . . . . . . . . . . . . . 52

2.2.1 PLO Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

2.2.1.1 Design Example . . . . . . . . . . . . . . . . . . . . . . . 56

2.2.2 Other Phase-Locked Oscillator Schemes . . . . . . . . . . . . . . . . 62

2.2.2.1 PLO for Voltage Control . . . . . . . . . . . . . . . . . . . 62

2.2.2.2 PLO for Speed Control of dc Motors . . . . . . . . . . . . 62

2.3 Phase Locked Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

2.3.1 Charge-Pump PLLs . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

2.3.2 Digital SRF-PLL . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

2.3.2.1 SRF-PLL Operation . . . . . . . . . . . . . . . . . . . . . 69

2.3.2.2 Pre-filters for the SRF-PLL . . . . . . . . . . . . . . . . . 70

2.3.2.2.1 All-pass Filters based Sequence Detector . . . . . 71

2.3.2.2.2 Delayed Signal Cancellation in the αβ Frame . . 73

2.3.2.2.3 Generalized Integrators in PLLs . . . . . . . . . . 74

2.3.2.2.4 Extended and Generalized Delayed Signal Can-

cellation . . . . . . . . . . . . . . . . . . . . . . . 75

2.3.3 Digital Single-phase PLL . . . . . . . . . . . . . . . . . . . . . . . . 76

2.3.4 Single-phase PLLs Vs SRF-PLLs . . . . . . . . . . . . . . . . . . . 78

2.3.5 Amplitude Compensation in Digital PLLs . . . . . . . . . . . . . . 78

2.3.6 Limit Cycles of Digital PLLs . . . . . . . . . . . . . . . . . . . . . . 79

2.4 Digital Alternatives to PLLs . . . . . . . . . . . . . . . . . . . . . . . . . . 82

vi

2.4.1 Filtering in the αβ Frame . . . . . . . . . . . . . . . . . . . . . . . 82

2.4.1.1 Transformation Angle Detector (LP-TAD) . . . . . . . . . 82

2.4.1.2 Normalised Positive-sequence Synchronous Frame (NPSF) 83

2.4.2 Filtering in Synchronous Reference Frames . . . . . . . . . . . . . . 85

2.4.2.1 Single-phase SRF Synchronization Algorithms . . . . . . . 85

2.4.2.2 MAF as LPF for SRF based Schemes . . . . . . . . . . . . 87

2.4.2.3 Equivalent Implementations . . . . . . . . . . . . . . . . . 92

2.4.2.4 Three-phase SRF Synchronization Algorithm . . . . . . . 93

2.4.2.5 SRF-MAFs for Harmonics and Negative-sequences Extrac-

tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

2.4.3 Stochastic Filtering based Synchronization Schemes . . . . . . . . . 96

2.4.3.1 Kalman Filtering . . . . . . . . . . . . . . . . . . . . . . . 96

2.4.3.1.1 Single-phase Example . . . . . . . . . . . . . . . 96

2.4.3.2 Weighted Least-Squares Estimation (WLSE) algorithms . 99

2.4.3.2.1 Three-phase WLSE Algorithm . . . . . . . . . . . 99

2.5 Frequency Adaptation of Synchronization Algorithms . . . . . . . . . . . . 104

2.5.1 Frequency Adaptation in WLSE Schemes . . . . . . . . . . . . . . . 106

2.5.2 Frequency Adaptive MAFs . . . . . . . . . . . . . . . . . . . . . . . 107

2.5.3 Adaptative Filtering for Frequency Adaptation/Estimation . . . . . 109

2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

3 Dynamics Study of Low-Gain PLLs 115

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

3.2 Frequency Domain Based Tuning . . . . . . . . . . . . . . . . . . . . . . . 116

 vii

3.2.1 Stability Margins and Cut-off Frequency . . . . . . . . . . . . . . . 117

3.2.2 Steady-state Distortion and Bandwidth . . . . . . . . . . . . . . . . 118

3.2.3 Grid Events and Transient Responses . . . . . . . . . . . . . . . . . 119

3.2.3.1 Underdamped Case: ζ < 1 . . . . . . . . . . . . . . . . . . 121

3.2.3.2 Overdamped Case: ζ > 1 . . . . . . . . . . . . . . . . . . 121

3.2.4 Limitations of Existing Tuning Approaches for Low-gain PLLs . . . 123

3.2.5 Design Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

3.2.5.1 High Bandwidth SRF-PLL (HB-PLL) . . . . . . . . . . . 124

3.2.5.2 Unbalance/Notch SRF-PLL (UN-PLL) . . . . . . . . . . . 125

3.2.5.3 SRF-PLL with Moving Average Filter (MA-PLL) . . . . . 126

3.2.6 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . 127

3.3 RC Model of Digitally Controlled Oscillator . . . . . . . . . . . . . . . . . 129

3.3.1 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . 131

3.3.1.1 Simulation Results . . . . . . . . . . . . . . . . . . . . . . 131

3.3.1.2 Single-phase PLL Implementation . . . . . . . . . . . . . . 131

3.3.1.3 SRF-PLL . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

4 Predictive based SRF-MAF Synchronization Algorithms 135

4.1 Calculation of Predictive Filters . . . . . . . . . . . . . . . . . . . . . . . . 137

4.2 Design Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

4.2.1 Design Example 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

4.2.2 Design Example 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

4.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

viii

4.4 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

4.4.1 Brief Comparative . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

4.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

5 Frequency Adaptive PR Current Controller for APFs 153

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

5.2 Active Power Filter Prototype . . . . . . . . . . . . . . . . . . . . . . . . . 156

5.3 Phase Locked Loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

5.3.1 PLL Tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

5.4 PR Current Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

5.4.1 Resonant Controllers Implementation . . . . . . . . . . . . . . . . . 161

5.4.2 Frequency Adaptive Implementation of Resonant Blocks . . . . . . 162

5.4.2.1 Effect of the Frequency Correction in Dynamics . . . . . . 163

5.4.3 Design and Tuning of C(z) . . . . . . . . . . . . . . . . . . . . . . . 165

5.5 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

5.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172

6 Conclusion and Future Work 173

6.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

6.2 Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

6.3 Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

A Matlab Scripts 175

References 183
List of Figures

1.1 Simple electrical system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Waveforms associated to power concepts in a linear single-phase circuit. . . 5

1.3 Voltage/current and Power phasors for Fig. 1.2 example. . . . . . . . . . . 6

1.4 Example of decomposition of an unbalanced three-phase set of fundamental

voltages/currents. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.5 vabc in the SRF established by θ. . . . . . . . . . . . . . . . . . . . . . . . 15

1.6 6-pulse thyristor based rectifier. . . . . . . . . . . . . . . . . . . . . . . . . 23

1.7 Working points of the thyristor full wave rectifier. . . . . . . . . . . . . . . 25

1.8 Steady-state voltages of the thyristor three-phase full rectifier. . . . . . . . 26

1.9 Simple HVDC diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.10 ac Motor controlled with Thyristor-based inverter and rectifier. . . . . . . . 27

1.11 One leg of a switch mode inverter. . . . . . . . . . . . . . . . . . . . . . . . 29

1.12 Set of S+ firing signal with sinusoidal PWM. . . . . . . . . . . . . . . . . . 30

1.13 va output with sinusoidal PWM. . . . . . . . . . . . . . . . . . . . . . . . . 31

1.14 Fourier Components of va . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

1.15 VSC connected to an ac source. . . . . . . . . . . . . . . . . . . . . . . . . 32

1.16 Per-phase representation of VSC connected to an ac source. . . . . . . . . 32

1.17 Complex power diagram of grid-connected VSC. . . . . . . . . . . . . . . . 33

ix
x

1.18 P − Q diagram. Operation points of grid-connected VSCs. . . . . . . . . . 34

1.19 Control of induction motor by PWM-VSI with diode rectifier . . . . . . . . 34

1.20 Simplified dc-link voltage control loop of VSC. . . . . . . . . . . . . . . . . 37

1.21 VSC operation in steady-state. Effect of phase-error. . . . . . . . . . . . . 38

1.22 VSC operation in steady-state. Effect of ripple in θ̂1 . . . . . . . . . . . . . 39

1.23 VSC operation during a transient. Effect of the re-tracking speed. . . . . . 40

2.1 Zero-cross detector example. . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2.2 OA based zero-cross detector example. . . . . . . . . . . . . . . . . . . . . 47

2.3 Simple HVDC diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

2.4 Constant-α vs Inverse-Cosine HVDCs. . . . . . . . . . . . . . . . . . . . . 49

2.5 Main drawbacks of zero-cross detectors. . . . . . . . . . . . . . . . . . . . . 51

2.6 Weak Grid. Short circuit ratio Xr /Xs ≈ 3 − 6. . . . . . . . . . . . . . . . . 52

2.7 Scheme of the PLO controller for the HVDC rectifier proposed in [1]. . . . 53

2.8 Equivalent block diagram of PLO circuit and control. . . . . . . . . . . . . 55

2.9 Equivalent block diagram of PLO circuit and controller for firing maximum
delay. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

2.10 PLO equivalent model at different α values. . . . . . . . . . . . . . . . . . 57

2.11 Start-up simulation of PLO for different i∗dc (dotted), idc is the blue line. . . 58

2.12 Steady-state system voltages for Figs. 2.12. . . . . . . . . . . . . . . . . . . 59

2.13 PLO tested under a very unbalanced system of voltages. . . . . . . . . . . 59

2.14 PLO tested in the presence of an unbalanced and polluted with harmonics
system of voltages. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

2.15 PLO transient test: sudden change in zL . . . . . . . . . . . . . . . . . . . . 61

2.16 PLO tested under a sudden change in the input frequency at 10 s. . . . . . 61

 xi

2.17 Control scheme of a dc motor in a single-phase grid. . . . . . . . . . . . . . 63

2.18 Basic conceptual models of PLLs. . . . . . . . . . . . . . . . . . . . . . . . 64

2.19 Block diagram of an analog CP-PLL. . . . . . . . . . . . . . . . . . . . . . 65

2.20 Key diagrams for the EXOR-Charge Pump Phase Detector . . . . . . . . . 66

2.21 Time domain simulation results from the CP-PLL model at f = 49 Hz. . . 67

2.22 SRF-PLL block diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

2.23 SRF-PLL with Fortescue operator based pre-filter. . . . . . . . . . . . . . . 71

2.24 Matlab Simulink model and all-pass filter features. . . . . . . . . . . . . . . 72

2.25 Positive-sequence estimation with all-pass filters (f1 = 50 Hz). . . . . . . . 72

2.26 SRF-PLL with pre-filters in the αβ frame. . . . . . . . . . . . . . . . . . . 73

2.27 Comparative of the dynamics of A(z) and D(z) (fs = 10 kHz, so n = 50 in D(z)). 74

2.28 Hd (s) and Hq (s) dynamic response. . . . . . . . . . . . . . . . . . . . . . . 76

2.29 Single-phase PLL with multiplier as PD. . . . . . . . . . . . . . . . . . . . 77

2.30 ALL-PLL compensation block. . . . . . . . . . . . . . . . . . . . . . . . . . 79

2.31 Key figures of the ALL-PLL. . . . . . . . . . . . . . . . . . . . . . . . . . . 80

2.32 Phase-plane portraits of a SRF-PLL implementation. The curves are ob-

tained through different initial conditions (color legends indicate the initial
θ̂1 ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

2.33 LP-TAD scheme. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

2.34 Significant simulation results for LP-TAD: fault and unbalance conditions. 84

2.35 v1 in the single-phase SRF. . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

2.36 Single-phase SRF synchronization algorithm: ωsrf = ω1 to decouple v1 . . . 86

2.37 Time and frequency responses of H(z)1 implemented at fs = 10 kHz and

n = 100. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

2.38 Simulation results for SRF-MAF1 working open loop (ωsrf = ω1 = ω1n ). . . 91
xii

2.39 DCT based implementation figures. . . . . . . . . . . . . . . . . . . . . . . 92

+
2.40 Three-phase SRF synchronization algorithm; ωsrf = ω1 to decouple vabc 1
. . 93

2.41 Simulation results for SRF-MAF3 working in open loop (ωsrf = ω1 = ω1n ). 95

2.42 Single-phase Kalman based synchronization algorithm. . . . . . . . . . . . 98

2.43 WLSE experimental results for different λs. . . . . . . . . . . . . . . . . . 102

2.44 WLSE experimental results. Effect of adding extra sequences to H. . . . . 103

2.45 Frequency adaptive MAFs. . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

2.46 Adaptive filter scheme of SRF-MAF1 with frequency adaptation/estimation.110

2.47 Adaptive SRF-MAF1 Vs open loop SRF-MAF1 during the start-up and
steady-state. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

2.48 Frequency estimation with Adaptive SRF-MAF3. . . . . . . . . . . . . . . 113

3.1 Studied linear model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

3.2 Frequency response of different filters for canceling harmonics. . . . . . . . 119

3.3 Comparative between underdamped and overdamped cases. . . . . . . . . . 123

3.4 Frequency response of H(z) for the HB-PLL (L(s)HB−P LL was discretized
using the ’zoh’ method). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

3.5 Frequency response of H(z) for the UN-PLL (L(s)U N −P LL was discretized
using the ’zoh’ method). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

3.6 Frequency response of H(z) for the MA-PLL (PI filter was discretized using
the ’zoh’ method). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

3.7 Experimental results for different SRF-PLLs: Ch1 (black) is the Va input in
p.u./V, Ch4 is the instantaneous phase-angle measurement (100 mV/rad),
Ch2 is the error signal vq in p.u./V (1 p.u. = π/2 deg of phase error), Ch3
is ∆ωo (10 mV/(rad/s)). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

3.8 RC Oscillator features. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

3.9 P1+ generation from the oscillator signals. . . . . . . . . . . . . . . . . . . . 131

 xiii

3.10 Simulation results for single-phase PLL (ω1 = 2π50.3 rad/s). . . . . . . . . 132

3.11 Experimental results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

4.1 Filtering blocks of the design examples. . . . . . . . . . . . . . . . . . . . . 139

4.2 Design example 1: time and frequency responses of H1 (z) · H2 (z). . . . . . 140
′
4.3 Design example 2: Frequency and Step responses of H1 (z) · H1 (z) · H2 (z). . 142

4.4 S1: Test to show predictive filters action. . . . . . . . . . . . . . . . . . . . 143

−
4.5 S1 response to a big frequency step at 0.2 s. Unbalanced (va1 max
= 0.1 ·
+
va1max ) input wave. Frequency step (up), phase error (center) and fre-
quency error (down). Zero steady state error and transient duration of
0.01 s are achieved. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

4.6 S1 tested under a distorted set of input waves rotating at 49.5 Hz. At 1 s,
a 1 p.u. to 0.2 p.u. sag with +45 deg phase jump has been programmed. . . 145

4.7 S1: steady state error for a balanced set of inputs oscillating at 48 Hz. . . . 146

− +
4.8 S1: Steady state error for an unbalanced (va1 max
= 0.1 · va1 max
) set of inputs
oscillating at 51 Hz. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

4.9 S1: Transient response for a sag with −45 deg phase jump. . . . . . . . . . 148

4.10 S1: steady-state phase measurement at 48 Hz. Ch1 is θ1 , Ch2 is va , Ch3 is

vb , Ch4 is vc (phase in ≈ 2 rad, voltages in p.u.). The measurement does
not have ripple. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

+
4.11 Comparative between S1 and S2 in terms of noise rejection. Ch1 is va1 max
,
Ch2 is va , Ch3 is vb , Ch4 is vc (voltages in p.u., ω1 = 2π · 49.75 rad/s). . . 150

+
4.12 S1 Vs S2 in terms of transient response. Ch1 is va1 max
, Ch2 is va , Ch3 is
vb , Ch4 is vc (voltages in p.u., ω1 = 2π · 49 rad/s). . . . . . . . . . . . . . . 150

5.1 APF prototype and controller. . . . . . . . . . . . . . . . . . . . . . . . . . 157

5.2 Experimental set-up features. . . . . . . . . . . . . . . . . . . . . . . . . . 158

5.3 Phase locked loop. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

5.4 Open loop PLL Bode diagram. . . . . . . . . . . . . . . . . . . . . . . . . 160

xiv

5.5 Two integrators based implementations, as proposed in [2]. . . . . . . . . . 161

5.6 Comparative of the error induced by different implementations (fs = 10 kHz).163

5.7 Frequency adaptive resonant block implementation. . . . . . . . . . . . . . 163

5.8 Accuracy of the resonant controllers (fs = 10 kHz). . . . . . . . . . . . . . 164

5.9 Frequency response of C(z) · P (z) around 150 Hz. When ∆ω̂1 > 0 the peak
”moves” to higher frequencies and vice versa. . . . . . . . . . . . . . . . . . 164

5.10 Current control block diagram. . . . . . . . . . . . . . . . . . . . . . . . . 165

5.11 PR block (C(z)). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

5.12 Bode diagram of C(z) · P (z). . . . . . . . . . . . . . . . . . . . . . . . . . 167

5.13 Steady-state currents and vP CC for different input frequencies. Ch1 is iL ,

Ch2 is iF , Ch3 is iS and Ch4 is vP CC . . . . . . . . . . . . . . . . . . . . . . 168

5.14 Transient response when there is a load change. Ch1 is iL , Ch2 is iF , Ch3
is iS and Ch4 is vP CC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

5.15 Transient response when there is a frequency step in f1 from 48 Hz to 52 Hz.

Ch1 is iL , Ch2 is iF , Ch3 is iS and Ch4 is vP CC . . . . . . . . . . . . . . . . 169

5.16 Transient response when there is a frequency step in f1 : from 52 Hz to

48 Hz. Ch1 is iL , Ch2 is ∆ω̂1 (scale at 2π∆3 rad/div), Ch3 is iS and Ch4
is vP CC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

5.17 Transient response when there is a frequency step in f1 from 48 Hz to

52 Hz. Effect of ripple of ≈ ±1 Hz in ∆ω̂1 . Ch1 is iL , Ch2 is ∆ωˆ1 (scale at
2π∆3 rad/div), Ch3 is iS and Ch4 is vP CC . . . . . . . . . . . . . . . . . . . 170

5.18 Steady state figures when error in estimation of ∆ω̂1 is considered. ∆ω̂1 =
0 rad/s and ∆ω1 = −4π rad/s (f1 = 48 Hz). Ch1 is iL , Ch2 is ∆ωˆ1 (scale
at 2π∆3 rad/div), Ch3 is iS and Ch4 is vP CC . . . . . . . . . . . . . . . . . 171

5.19 Transient response when there is a voltage sag (1 p.u. → 0.8 p.u.) with
phase-angle jump of +45 deg in vP CC (vS ). Ch1 is iL , Ch2 is ∆ωˆ1 (scale at
2π∆3 rad/div), Ch3 is iS and Ch4 is vP CC . . . . . . . . . . . . . . . . . . . 171

A.1 PLO system Matlab script. Main file. . . . . . . . . . . . . . . . . . . . . . 176

A.2 PLO system Matlab script. Function file (page 1 of 2). . . . . . . . . . . . 177
 xv

A.3 PLO system Matlab script. Function file (page 2 of 2). . . . . . . . . . . . 178

A.4 Matlab script to depict SRF-PLL trajectories in the Phase-plane. . . . . . 179

A.5 Matlab script of the Kalman Filter Single-phase synchronization example. . 180

A.6 Matlab script of the WLSE syncrhonization example. . . . . . . . . . . . . 181

A.7 Matlab script of the single-phase PLL with RC-Oscillator based DCO. . . . 182
xvi
List of Tables

1.1 States of diode full rectifier depending of θ1 . . . . . . . . . . . . . . . . . . 24

1.2 State of controlled rectifier depending of θ1 and α. . . . . . . . . . . . . . . 24

1.3 VSC based Rectifier Values. . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.1 HVDC values. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

2.2 Controller Values. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

2.3 Controller Values. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

2.4 Look-up table: control of n depending on ω̂1 . . . . . . . . . . . . . . . . . 107

2.5 Adaptive algorithm values. . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

3.1 Significant parameters of HB-PLL . . . . . . . . . . . . . . . . . . . . . . . 124

3.2 Significant parameters of UN-PLL . . . . . . . . . . . . . . . . . . . . . . . 125

3.3 Significant parameters of MA-PLL . . . . . . . . . . . . . . . . . . . . . . 126

4.1 Design Example 1 Values. . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

4.2 Design Example 2 Values. . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

4.3 Brief comparative among significant systems with good unbalance rejection 148

5.1 Significant values of the power circuit. . . . . . . . . . . . . . . . . . . . . 156

5.2 Values of PLL parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

xvii
xviii

5.3 Values of parameters of the PR controller. . . . . . . . . . . . . . . . . . . 166

List of Abbreviations and Acronyms

ac Alternate current.

ADC Analog-to-digital converter.

ADALINE Adaptive Linear Combiner.

APF Active Power Filter.

CP Charge-Pump (PLL).

dc Direct current.

DCO Digitally Controlled Oscillator.

DCT Discrete Cosine Transform.

DFT Discrete Fourier Transform.

DPGS Distributed Power Generation System.

DSP Digital Signal Processor.

DSRF Double SRF. Applied to a PLL (DSRF-PLL).

DST Discrete Sine Transform.

EXOR Exclusive OR.

EKF-TAD Enhanced Kalman Filter-TAD (Synchronization Algorithm).

FACTS Flexible ac Transmission Systems.

FIR Finite Impulse Response (filter).

FPGA field-programmable gate array.

GI Generalized integrator.

GTO gate turn-off thyristor.

HVDC High-Voltage Direct Current.

xix
xx

IC Integrated Circuit.

IEEE Institute of Electrical and Electronics Engineers.

IIR Infinite Impulse Response (filter).

IGBT insulated-gate bipolar transistor.

IGCT integrated gate-commutated thyristor.

LF Loop filter (PLL).

LPF Low pass filter.

LP-TAD Low pass-TAD (Synchronization Algorithm).

MAF Moving average Filter.

MSRF Multiple Synchronous Reference Frame.

NPSF Normalised Positive-sequence Synchronous Frame.

OA Operational Amplifier.

PCC Point of common coupling.

PD Phase detector (PLL).

PI Proportional Integrator (filter).

PLL Phase locked loop.

PLO Phase locked oscillator (control system).

PM Phase margin.

PQ Power Quality.

PR Proportional+Resonant (current controller).

PWM Pulse Width Modulation.

RDFT Recursive DFT.

SRF Synchronous Reference Frame.

SSSC Static Synchronous Series Compensator.

SRF-MAF Synchronization algorithm based on SRF and MAFs.

STATCOM static compensator.

SVC static VAr compensator.

SVF-TAD Space Vector Filter-TAD (Synchronization Algorithm).

 xxi

TAD Transformation angle detector.

THD Total Harmonic Distortion.

UPFC unified power-flow controller.

VCO Voltage Controlled Oscillator (IC).

VSC voltage-source controller.

VSI voltage-source inverter.

WLSE Weighted Least Squares Estimation (Synchronization Algorithms).

xxii
Nomenclature

α Firing angle in thyristor based converters.

B Angle transformation from an αβ rotating frame to a dq frame.

C Clarke transformation from an abc rotating frame to an equivalent αβ frame.

∆α Small perturbation in α.

d Duty cycle of PWM converters..

ǫ Optimization parameter of predictive filters.

f Frequency in Hz..

f1n Nominal fundamental frequency (e.g. 50 Hz in Europe).

fs Sampling frequency of a discrete model.

φ Relative phase-angle between i(t) and v(t).

ϕ Offset phase-angle.

i Single-phase current.

i Vector of currents in a multiphase system or reference frame.

j Imaginary Unit.

m Modulating/control signal of PWM converters.

n Set the order of a filter.

xxiii
xxiv

N1 Set the order of a MAF (chapter 4).

N2 Set the order of a predictive filter (chapter 4).

P Active Power.

P Park transformation from an abc rotating frame to a dq static frame.

Q Reactive Power.

R Resistance.

S Apparent Power.

S Fortescue matrix.

t Time, in seconds.

T = 1/f . Time cycle.

∆t Very short instant, in seconds.

θ Instantaneous rotating phase-angle.

v Single-phase voltage.

v Vector of voltages in a multiphase system or reference frame.

ω = 2πf . Angular frequency or pulsation.

z Impedance.

Subscript

1 Fundamental component.

abc Voltage/current vector in the abc frame.

a Phase a in the abc frame.

b Phase b in the abc frame.

 xxv

c Phase c in the abc frame.

co Relative to controlled oscillator (voltage).

cp Relative to charge-pump (voltage).

αβ Voltage/current vector in the αβ frame.

α Component α in the αβ frame.

β Component β in the αβ frame.

d Relative to d component in a srf .

dc Voltages/currents in a DC bus.

e Relative to the error between real and estimated value.

i Input value.

j Relative to jittering of synchronization algorithms (phase).

ll Line to line (voltage).

LP F Low pass filter.

max Maximum. ”Peak value” referred to voltages/currents.

min Minimum.

n Nominal value, e.g. f1n = 50 Hz nominal fundamental frequency.

o Output value.

q Relative to q component in a srf .

rms Root-mean square.

s Relative to sampling (f and T ).

s Relative to settling time.

srf Relative to SRF.

xxvi

Superscript

+ Positive sequence.

− Negative sequence.

0 Zero sequence.

T Transpose of a matrix.

(−1) Inverse of a matrix.

∗ Reference value.

Accents

x̄ Average value of x.

x̂ Estimated value of x.

Ẋ x vector of voltages/currents in phasor notation.

Chapter 1

Introduction

1.1 Objectives and Motivation of this Work

This work approaches an in-depth study of synchronization algorithms employed in

the control of grid-connected power electronics converters. The goal of this research is to
supply an added knowledge to all the researchers and engineers working in this field.

Due to the high amount of recent proposals and applications, an in-depth review of the
state-of-art in grid synchronization seems to be necessary. Active fields of research such
as Flexible ac Transmission Systems (FACTs), Distributed Power Generation Systems
(DPGS), Power Conditioning and Traction Systems claim for an ever-increasing perfor-
mance of synchronization algorithms. This works provides a study of existing schemes and
implementation techniques and also provides novel high performance algorithms. Special
emphasis is devoted to take advantage of digital implementation.

It is also a goal of this work to show how a good knowledge in the synchronization field
permits to improve other blocks employed in the controllers of power electronic converters.
A novel discrete-time implementation of the Proportional+Resonant (PR) controllers is
contributed in chapter 5. The added value of this implementation is the manner how each
resonant controller is made frequency adaptive using a phase locked loop (PLL). This
provides robustness in the presence of grid frequency deviations.

1.2 Background on Electric Power Definitions

Some concepts should be defined before the analysis of power systems. These concepts
and definitions of electric power for sinusoidal ac systems are well stated and accepted
nowadays when considering ideal conditions. However, under distorted conditions, several

1
2 Chapter 1 Introduction

Ideal AC
Source Linear Load
v i

ω = 2 ⋅π ⋅ f

Figure 1.1: Simple electrical system.

and different power definitions and theories are still in use [3].

1.2.1 Power Definitions under Sinusoidal Conditions

The simple single-phase electrical systems depicted in Fig. 1.1 is considered. The ideal
ac source generates a sinusoidal voltage rotating at the the angular frequency, also named
frequency pulsation ω. The frequency of oscillation in [ Hz] is obtained as f = ω/(2π).
The ac source is connected to a linear load. The instantaneous voltage and current can
be analytically represented by

√
v(t) = 2 · vrms · sin (ωt + ϕv ), (1.1)

where ϕv is a constant offset phase-angle, and

√
i(t) = 2 · vrms · sin (ωt + ϕi ), (1.2)

where ϕi is also a constant offset phase-angle. The relative angle between v(t) and i(t)
(φ) is given by

φ = ϕv − ϕi . (1.3)

φ is set by the load reactance: φ > 0 (the current lags the voltage) for inductive loads,
and φ < 0 (the current leads the voltage) for capacitive loads. The value of φ in Fig. 1.1
is limited in the range [−90 deg ≤ φ ≤ 90 deg]. The instantaneous active power p(t) is
1.2 Background on Electric Power Definitions 3

given by the product of the instantaneous voltage and current:

p(t) = v(t) · i(t) = P [1 − cos (2ω1 t)] − Q sin (2ω1 t), (1.4)
where

P = vrms · irms · cos (φ) (1.5)

is named active power, for which the measurement unit is the Watt [ W].

Q = vrms · irms · sin (φ) (1.6)

is named reactive power, which is measured in volt-ampere reactive [ VAr] .

From (1.4) to (1.6), p(t) is not constant, since it has an oscillating component rotating
at 2ω. P represents an unidirectional power flow from the ac source to the load. Therefore,
it could be said that P gives a measurement of the averaged energy supplied in any time
interval from the ac source to the load.

The term associated to Q is an oscillating at 2ω component and its average value

is equal to zero. Q is commonly associated as the oscillating power, since the total
energy transfered to the load over an entire cycle is zero. For this reason, Q is commonly
associated to a parasitic or non desired effect.

The apparent power S is defined as:

S = vrms · irms . (1.7)

The unit of S is the Volt-Ampere [ VA] .

The power factor (PF) is defined as cos (φ), which is obtained as:

P
cos (φ) = . (1.8)
S

The physical meaning of S is clear from (1.7) and (1.8). S represents the maximum
reachable active power P at unity power factor.

A PF of one, or “unity power factor”, is the goal of any electric utility company, since,
if the power factor is less than one, they have to supply more current to the user for a
given amount of power consumption. In so doing, they incur more line losses. They also
must have larger capacity equipment in place than would be otherwise necessary. As a
result, an industrial facility will be charged a penalty if its PF is much different from 1.
4 Chapter 1 Introduction

The scheme of Fig. 1.1 has been simulated in the time domain through PSPICE in
order to supply the waveforms shown in Fig. 1.2. The linear load is form of a resistor in
series with an inductance with a XL /R = 0.6/0.8 ratio, which results in φ ≈ 36 deg (XL
is the inductance impedance at ω). The values of the components have been arranged so
rms values of v(t) and i(t) are 1 p.u..

The above figures show both v(t) and i(t) and the time delay ≈ 2 ms associated to φ.
The figure below shows p(t) and its average value P . As expected, P = V I cos(φ) = 0.8 p.u..
In some points of time, p(t) is negative, which corresponds with an instantaneous power
flow from the load to the source. However, the area highlighted with ’A’, corresponding
with a positive power flow from the source to the load, is higher than the area highlighted
with ’B’, corresponding with the flow from the load to the source. By inspection of (1.4)
it is clear that a higher weight of the inductance (XL ) in the load results in a higher ’B’
area, which means a higher Q. If there was no resistive component φ would be equal to
90 deg, and hence, area ’B’ and area ’A’ would be the same; Q would be maximum and
P = 0. The sign of Q depends on the kind of load: inductive load (φ > 0) results in a
positive Q and a capacitive load (φ < 0) results in a negative Q.

1.2.1.1 Electrical Variables with Phasor Notation

A powerful tool in the analysis of power systems is the use of the so-called phasor nota-
tion instead of analytical expressions in the time domain. The phasor notation approach
is detailed as follows.

Any sinusoidal time function (f (t)) rotating at an angular frequency ω can be repre-
sented as the real part of a complex number:

f (t) = A · cos (ωt + φ) = ℜ{A · ei(ωt+φ) } = ℜ{A · ejωt · ejφ }. (1.9)

Assuming steady-state and known frequency, f (t) can be represented by the complex
number Ḟ :

Ḟ = A · ejφ = A∠φ. (1.10)

A voltage phasor v̇ and a current phasor i̇ can be represented by complex numbers

both in polar and Cartesian notation:

V̇ = vmax ∠ϕv = vmax cos (ϕv ) + j · vmax sin (ϕv )

(1.11)
I˙ = imax ∠ϕi = imax cos (ϕi ) + j · imax sin (ϕi ).
1.2 Background on Electric Power Definitions 5

1.41

0.5

-0.5

-1

-1.5
0.04 0.045 0.05 0.055 0.06 0.062 0.065 0.07 0.075
0.08

1.5

0.8

0.5

0.045 0.05 0.055 0.065 0.07 0.075

0.04 0.06 0.08

Figure 1.2: Waveforms associated to power concepts in a linear single-phase circuit.

6 Chapter 1 Introduction

ℑ ωt ℑ

ℜ
S ω
V& jQ

φ = 36º
φ = 36º

I&
P ℜ
(a) Voltage and current phasor representation for (b) Graphical representation of the
Fig. 1.2. complex power for Fig. 1.2.

Figure 1.3: Voltage/current and Power phasors for Fig. 1.2 example.

Usually, the phase-angle of one variable, (e.g. ωt + ϕv ) is chosen to be zero, so the

phasor diagram could be represented as function of a relative angle (e.g. φ).

Fig. 1.3a shows the phasor notation for the example depicted in Fig. 1.2. As in the
time domain representation, the current lags the voltage, since there is an inductive load.

The phasor notation can be extended to the power definitions. The complex apparent
power S is defined as the product of the voltage and current phasors:

S = V̇ · I˙∗ = vmax · imax cos (φ) + jvmax · imax sin (φ) = P + jQ. (1.12)

S can be depicted in the complex plane as function of P , Q and PF. Fig. 1.3b shows
this complex power representation for the Fig. 1.2 example. The sign of the reactive
power is positive because of the inductive behavior of the load. A negative Q would refer
to a capacitive reactive part of the load.

1.2.2 Power Definitions under Non-Sinusoidal Conditions

When the system voltages and currents contain components of non fundamental fre-
quency, it is said that there is distortion. The origin of this distortion can be in the origin
or in the consumer (non-linear loads).

The assessment of power definitions under distorted conditions is not a trivial task
at all. Two different theories have been provided: the power definitions based on the
frequency domain, proposed by Budeanu [4], and the power definitions in the time domain
by Fryze [5].
1.2 Background on Electric Power Definitions 7

There is so much discussion concerning which one of that strategies is more suitable
for the control of power converters. For example, Czarnecki proved that some of the
Budeanu definitions do not have physical meaning [6]. Moreover, there are quite a lot
difficulties with frequency domain power definitions applied to three-phase and four-wire
systems [3].

1.2.2.1 Power Definitions in the Frequency Domain by Budeanu

A set of power definitions set by Budeanu [4] is still very useful for the analysis of
power systems in the frequency domain. These definitions are valid for generic currents
and voltages in steady-state. However, they are not suitable to analyze transients.

A periodic ac single-phase wave, as voltages and currents, can be decomposed in a

Fourier series. Then, the corresponding phasor for each harmonic can be calculated. The
Budeanu definitions are:

1.2.2.1.1 Apparent Power S:

S = vrms · irms , (1.13)

where vrms and irms are voltage and current rms values These rms values can be calculated
as function of each hth (h = 1, 2, ...) order harmonic component:

s v
u ∞
1ZT 2 u1 X
vrms = v (t)dt = t v2
T 0 T h=1 rmsh
s u
v (1.14)
∞
1ZT 2 u1 X
irms = i (t)dt = t i2
T 0 T h=1 rmsh

where vrmsh and irmsh are the rms values of each harmonic component, and T the funda-
mental component period.

1.2.2.1.2 Active Power P :

∞
X ∞
X
P = Ph = vrmsh · irmsh cos (φh ), (1.15)
h=1 h=1

where the displacement angle of each pair of hth order harmonic voltage and current
components is represented as φh .
8 Chapter 1 Introduction

1.2.2.1.3 Reactive Power Q:

∞
X ∞
X
Q= Qh = vrmsh · irmsh sin (φh ) (1.16)
h=1 h=1

This Q definition tried to quantify the amount of power that does not realize work
in steady-state. However, this per-component approach does not take into account other
interactions such as cross products between voltages and currents at different frequencies.
For this reason, Budeanu also defined the distortion power D to quantify the loss of power
quality due to harmonic distortion:

1.2.2.1.4 Distortion Power D

q
D= S 2 − P 2 − Q2 . (1.17)

The physical meaning of P is clear, since it represents the average value of the instan-
taneous power p(t), that is, the average ratio of energy transfered between two systems.
However, both Q and D are just mathematical expressions without a clear physical mean-
ing. Another drawback of Budeanu’s approach is its poor applicability in practical cases
of power quality assessment [3]. Other definitions by Budeanu are:

1.2.2.1.5 Power Factor λ

P
λ= . (1.18)
S

1.2.2.1.6 Displacement Factor cos (φ)

P
cos (φ) = √ . (1.19)
P2 + Q2

1.2.2.1.7 Distortion Factor cos(γ)

√
P 2 + Q2
cos (γ) = . (1.20)
S

The following relation is valid

P
λ= = cos (φ) · cos (γ). (1.21)
S
1.2 Background on Electric Power Definitions 9

1.2.2.2 Power Definition in the Time Domain by Fryze

The main definition for power components in the time domain have been presented
by Fryze:

1.2.2.2.1 Active Power Pw

1ZT 1ZT
Pw = p(t) = v(t) · i(t) = vrms · iw = vw · irms . (1.22)
T 0 T 0

where vrms and irms are voltage and current rms values, and vw and iw are the active
voltage and active current defined below.

1.2.2.2.2 Apparent Power Ps

Ps = vrms · irms . (1.23)

1.2.2.2.3 Active Power Factor λ

Pw
λ= . (1.24)
Ps

1.2.2.2.4 Reactive Power

q
Pq = Ps2 − Pw2 = vq · irms = vrms · iq . (1.25)

where vq and iq are the reactive voltage and current defined below.

1.2.2.2.5 Reactive Power Factor λq

√
λq = 1 − λ2 . (1.26)

1.2.2.2.6 Active Voltage vw and Active Current iw

vw = λ · vrms
(1.27)
iw = λ · irms .
10 Chapter 1 Introduction

1.2.2.2.7 Reactive Voltage vq and Reactive Current iq

vq = λq · vrms
(1.28)
iq = λq · irms .

1.2.3 Power Definitions in Three-Phase Systems

The extension of the power definitions from a single-phase to a three-phase systems

is not a trivial task. Any three-phase system could have some properties not present in
single-phase systems:

• The presence of a fourth wire: if a three-phase system is grounded in more than one
point there is an additional path for current circulation. In other systems, a fourth
wire connected to the neutral is presented.

• Balance/Unbalance among phases: this feature refers to a set of three-phase voltages

(or currents). If their amplitudes are the same and the displacement angles between
the phase is 2π/3 the system is balanced. If not, the system is unbalanced.

1.2.3.1 Power in Balanced Three-phase Systems

Considering a balanced system of voltages and currents:

   √ 
va 2 · vrms · sin(ωt + ϕv )
   √ 2π

2 · vrms · sin(ωt + ϕv −
 
=

 vb   3
) , (1.29)
   √ 2π

vc 2 · vrms · sin(ωt + ϕv + 3
)

and

   √ 
ia 2 · irms · sin (ωt + ϕi )




 √ i 2π

 ib  =
 2 · irms · sin (ωt + ϕ − 3
)

. (1.30)
   √ 2π

ic 2 · irms · sin (ωt + ϕi + 3
)

For a three-phase system the instantaneous active power p3φ (t) describing the energy
flow per time unit transfered between two systems is given by:

p3φ (t) = va (t)ia (t) + vb (t)ib (t) + vc (t)ic (t) = pa (t) + pb (t) + pc (t). (1.31)
1.2 Background on Electric Power Definitions 11

Substituting (1.29) and (1.30) in (1.29): (1.29) and (1.30) in (1.29):

p3φ (t) = 3 · vrms · irms cos ϕv − ϕi = 3P. (1.32)

Equation (1.32) shows the different behavior with respect to single-phase systems:
p3φ (t) is constant and does not have a second harmonic oscillating term. This constant
component defines the active power (P3φ ), which is three times the single-phase active
power.

The three-phase apparent power (S3φ ) is:

S3φ (t) = 3S = 3 · varms · iarms , (1.33)

so S3φ is also three times the single-phase apparent power.

The definition of three-phase reactive power (Q3φ ) is:

Q3φ (t) = 3 · varms · iarms sin(ϕv1 − ϕi ) = 3Q. (1.34)

However, Q3φ does not have the same physical meaning that the reactive power has
in single-phase systems. In fact, a balanced three-phase system feeding a three-phase
balanced load does not cause power oscillations [3].

1.2.3.2 Analysis under Unbalanced Conditions

The traditional concepts of apparent power and reactive power are not suitable for
unbalanced three-phase systems. As example, in an unbalanced system the fact that
the line currents are proportional to system voltages does not assure maximum average
(active) power transfer between systems [3].

Based on rms values of voltages and currents, two different definitions of apparent
power have been proposed:

1.2.3.2.1 ”Per-phase” Calculation:

X X
S3φ = Sk = Vk Ik , k = (a, b, c). (1.35)
k k
12 Chapter 1 Introduction

1.2.3.2.2 ”Aggregate rms Value” Calculation:

sX sX
SP = Vk2 Ik2 , k = (a, b, c). (1.36)
k k

It has been proved that under unbalanced and distorted conditions SP ≤ S3φ . SP
have been used by some authors as ”the maximum reachable active power at unity power
factor” [7].

However, the physical meaning of these definitions is not universally accepted, and
some authors state that they are only mathematical expressions. For example, Akagi
et al only consider the instantaneous active power as universal concept in three-phase
systems [3].

1.2.3.3 Power Definitions and Alternatives in Power Electronic Converters

Control

Power definitions are employed in the control of power converters, specially in power
line conditioners [3, 8–10], since Akagi et al presented their ”instantaneous active power
theory” [11]. However, dealing with power magnitudes in real-time has some drawbacks,
specially in three-phase and four-wires systems [3, 12].

As an alternative to instantaneous reactive power theory, Bhattacharya et al pro-

posed the synchronous reference frame (SRF) theory, in which specific voltage/currents
components/sequences to compensate are extracted by digital signal processing [12–16].
Equivalent SRF approaches for single-phase systems were proposed in the literature [17–
19]. Nowadays, with the suitability of powerful digital devices and the advance in the
state-of-art in digital signal processing, working directly with voltages/currents is a very
widespread option.

1.3 Background on Symmetrical Components

One fundamental theory in the analysis of multi-phase system is the symmetrical

components decomposition discovered by Fortescue [20]. The theory of symmetrical com-
ponents allows to handle with sequences instead of independently per-phase.

From direct application to a three-phase system of the brilliant analysis by Fortescue,

any unbalanced three-phase system of voltages/currents (V̇a , V̇b , V̇c ) can be represented by
other three phasors: a positive-sequence phasor V̇+ , a negative-sequence phasor V̇− and a
1.3 Background on Symmetrical Components 13

ω ω ω ω
vbmax vb + vb -

φab vabc 0
120º va + 120º va -
φbc = 120º + 120º +
vamax
φca 120º 120º
vc + vc -
vcmax
va max ≠vb max≠ vcmax Positive Sequence Negative Sequence Zero Sequence

φab ≠ φbc ≠ φca va + + +

max =vb max= vc max va max =vb max= vc-max
- - va 0max =vb 0max= vc0max
(a) Phasor diagram of an unbalanced system.
Three-Phase Unbalanced system Positive-sequence Negative-sequence Zero-sequence
1 1 0.15 0.08

Amplitude (p.u.)
Amplitude (p.u.)

0.06
Amplitude (p.u.)

0.1

Amplitude (p.u.)
0.5 0.5 0.04
0.05
0.02

0
= 0
+ -0.05
0
+ 0

-0.02

-0.5 -0.5 -0.04

Va Va+ -0.1
Va- Va0
Vb Vb+ Vb- -0.06 Vb0
Vc Vc+ Vc- Vc0
-1 -1 -0.08
0.45 0.46 0.47 0.48 0.49 0.5 0.45 0.46 0.47 0.48 0.49 0.5 0.45 0.46 0.47 0.48 0.49 0.5 0.45 0.46 0.47 0.48 0.49 0.5
Time (s) Time (s) Time (s) Time (s)

(b) Time domain decomposition.

Figure 1.4: Example of decomposition of an unbalanced three-phase set of fundamental

voltages/currents.

zero-sequence phasor V̇0 [21]. The Fortescue transformation by means of the S matrix is:

     
V̇ 0 1 1 1 V̇a
  1    
V̇ +  a2 
     
= · 1 a ·  V̇  , (1.37)
  b 

  3 
V̇ − 1 a2 a V̇c
| {z }
S

where a is a complex number also called Fortescue operator:

√
j(2π/3) −1 3
a = 1∠120 deg = e = +j . (1.38)
2 2

Fig. 1.4a shows an example of decomposition of an unbalanced three-system of volt-

ages (vabc = [va , vb , vc ]) in a balanced system rotating at (ω) which is the positive-
sequence (V̇ + ) plus a balanced system rotating at (−ω) which is the negative-sequence
(V̇ − ) plus a zero-sequence (V̇ 0 ). Fig. 1.4 shows the equivalent time domain vectors:
+
vabc = [va+ , vb+ , vc+ ], vabc
− 0
= [va− , vb− , vc− ] and vabc = [va0 , vb0 , vc0 ]. The superscripts “+”, “−”
and “0” refer to positive, negative and zero sequences, respectively. The superscript “1”
refer to the fundamental component: e.g. ω1 is the fundamental frequency.
14 Chapter 1 Introduction

The real time implementation of S is approached in some works; the imaginary unit
j is implemented by means of digital filters with a 90 deg delay at ω1 [22]. However, this
approach have some limitations, as shown in chapter 2.

Clarke and Park transformations [23, 24], also based on the symmetrical components
theory, offer a much higher versatility, specially for real-time implementations such as
power converter controllers, monitoring systems, etc.

The Clarke transform C converts a three-phase system vabc to an equivalent two-phase

system vαβ :

 
    va
vα 2  1 −0.5 −0.5   


  = √ √ ·  vb , (1.39)
vβ 3 0 3
−2 3  
2
| {z } | {z } vc
vαβ C | {z }
vabc

where vα and vβ are two orthogonal waves rotating at the same frequency that vabc , which
have information of both positive and negative sequences; zero-sequences do not appear
in vαβ [21]. The Clarke transformation is employed in several controllers such as the ones
implementing the Space Vector Modulation (SVM) technique [16, 25]. It is also widely
employed in grid synchronization.

Real-time implementation of Park transformations allow to decouple positive from

negative sequences, since they are “moved” to different frequencies, including dc. In fact,
the possibility of dealing with oscillating waves as dc values is the most interesting feature
of the Park transformations. The most widely employed Park transformation is the P+
matrix to decouple the positive-sequence. Considering an input vector of voltages as
+
(1.29), but with only positive-sequence, vabc rotating at ω, a vector of two dc components
+
(vdq ) is obtained as follows:

 
    va+
vd+ 2  sin(θ) sin(θ − 2π
3
) sin(θ + 2π
3
)  
 vb+
· 
  = , (1.40)
vq+ 3 cos(θ) cos(θ − 2π ) cos(θ + 2π )  
3 3
| {z } | {z } vc+
v+
dq
P+ | {z }
vabc
R π
where θ = −π ωt+ϕ, ϕ being an offset constant. When referring to (1.40) implementation,
it is usually said that vabc is transformed to the synchronous reference frame (SRF) stated
by θ. Fig. 1.5 shows the graphical representation of vabc in the SRF established by θ.

In (1.40) negative-sequence rotating at ω is not considered to recall the fact that

positive-sequence components are ”moved” to dc. However, it should be known that
P+ · vabc
−
results in a vector whose components rotate at 2ω [26]. On the other hand, any
other sequence (positive or negative), rotating at ω ′ not equal to ω, results in components
1.3 Background on Symmetrical Components 15

q ω

vq+
v dq

vd+ d

Figure 1.5: vabc in the SRF established by θ.

of frequency ω − ω ′ 1 and ω + ω ′ [27]. For this reason, Park transformations used to

be implemented together with low pass filters or Proportional Integrator (PI) controllers
[12–16].

The Park transformation for negative-sequence (P− ) is also implemented, specially in

power line conditioners dealing with unbalance [26].

 
2π 2π
2 sin (θ) sin (θ + 3
) sin (θ − 3
)
P− = , (1.41)
3 cos (θ) cos (θ + 2π
) cos (θ − 2π
)
3 3

π R
where θ = −π ωt + ϕ. Through P− the negative-sequence rotating at ω are trans-
formed to dc components and other sequences are “moved” to different frequencies. When
representing sequences in the SRF of P− , the sense of rotation is clock-wise.

0
Regarding to zero-sequences of any frequency, it should be noticed that P− · vabc =0
0
and P · vabc = 0. Therefore, zero-sequences are neglected when working with SRF based
−

approaches.

In this section, P+ and P− have been defined for a generic frequency (ω). However,
it should be noticed that electric power applications usually deal with signals containing
sequences (positive and/or negative) of more than one frequency: a fundamental com-
ponent rotating at fundamental frequency and its harmonics components. Starting from
this point, the subscript “1” is also added both to P+ and P− to refer to the fundamental
component and subscripts ”h”, where h = 2, 3, 4, 5, ... to harmonic components. E.g. P+ 1
refers to the Park transformation for the fundamental positive-frequency.

1
Negative frequencies implies a change in the kind of sequence (e.g. from negative to positive).
16 Chapter 1 Introduction

1.3.1 Alternatives in Implementation

It is common to find some of these transformations expressed in a different

q way or
+
with different constants, as the so-called constant power P which uses 3/2 instead of
2/3 [25]. In this work, the constants have been chosen so the amplitude of the vectors
involved is kept constant, since amplitudes could be also estimated by synchronization
algorithms.

On the other hand, the Park transformation can be considered as the combination of
the Clarke transformation and the so-called angle transformations B+ and B− [21]:

 
cos (θ) sin (θ)
B+ =  , (1.42)
− sin (θ) cos (θ)

 
cos(θ) − sin(θ)
B− =  . (1.43)
sin(θ) cos(θ)

So

 
2π 2π
2 cos(θ) cos(θ − ) cos(θ + )
P+ = B+ · C = 3 3  (1.44)
3 − sin(θ) − sin(θ − 2π ) − sin(θ + 2π )
3 3

and

 
2 cos(θ) cos(θ + 2π
3
) cos(θ − 2π
3
)
P− = B− · C =  . (1.45)
3 − sin(θ) − sin(θ + 2π ) − sin(θ − 2π )
3 3

Apparently, Park transformations (1.40) and (1.41) could seem different from (1.44)
and (1.45). However, they are totally equivalent, the difference is in how to express the
vabc vector, either as cosine waves [24, 28] or as sine waves [29, 30]. Anyway, both
approaches lead to the same final results and equivalent intermediate analysis.

1.4 Background on Power Quality

As stated in [31], the term “power quality” refers to a wide variety of electromagnetic
phenomena that characterize the voltage and current at a given time and location on
1.4 Background on Power Quality 17

the power system. This section summarizes the terminology to describe the most com-
mon events. location on the power system. This section summarizes the terminology to
describe the most common events.

1.4.1 Transients

It refers to very short duration undesirable events in the voltages, currents or power in
an electric circuit. Depending on the wave-shape of the transient, there are two categories
of transients:

1.4.1.1 Impulsive Transient

It is a sudden, non-power frequency change in the steady-state condition of volt-

age/current, that is unidirectional polarity. It is characterized by the rise and decay
times. The most common cause of impulsive transients is lighting. Due to the high fre-
quencies involved, impulsive transients are damped quickly by resistive circuit components
and are not conduced far from their source.

1.4.1.2 Oscillatory Transients

An oscillatory transient is a voltage/current whose instantaneous values change polar-

ity rapidly. It is described by its predominant frequency, duration and magnitude.

Power electronics devices produce oscillatory voltage transients as a result of commu-

tation and RLC snubbers. They are in the high kHz range and last only a few cycles of
that frequency.

Back to back capacitor energization gives rise to current oscillatory transients in the
tens of kHz.

Low frequency oscillatory transients (< 5 kHz) are present both in sub-transmission
and distribution systems mainly due to capacitor bank and transformers energization.

1.4.2 Short-duration Variations

They last more than transients: between 0.5 cycles of the fundamental frequency
(50 or 60 Hz) and 1 minute. They are mainly caused by fault conditions, energization
of large loads and intermittent loose connections in power wiring. These faults could
18 Chapter 1 Introduction

result in voltage rises (swells), voltage drops (sags) and even a complete loss of voltage
(interruptions).

1.4.2.1 Interruptions

It occurs when the supply voltage/current decreases to below 0.1 p.u.. The causes of
interruptions are power system faults, equipment failures and control malfunctions.

1.4.2.2 Sags (Dips)

As said, the term sag refers to a voltage drop. They are characterized by their mag-
nitude and duration. The terminology recommended by [31] to refer to voltage sags is ”a
sag to 20 %”, which means that the line voltage is reduced to a 20%.

The main causes of sags are power system faults, switching of heavy loads or starting
of large motors. A very detailed work about kind of sags and causes is shown in [32].

1.4.2.3 Swells

It is an increase on the rms voltage or current at the nominal frequency. They are
also associated with system fault conditions but they are much less common than sags.
They are characterized by their magnitude and duration. Typical magnitudes are in the
range 1.1 p.u. and 1.8 p.u..

1.4.3 Long Duration Variations

Long duration variations are rms deviations lasting for more than 1 minute. They
can be either over-voltages and under-voltages, depending on the cause of the variations.
They are caused by load variations on the system and switching operations. They are
characterized by plots voltage versus time.

1.4.3.1 Over-Voltage

Over-voltages may be result of load switching (e.g. switching off a large load) or
variations in the reactive compensations of a power system (e.g. switching on a capacitor
bank). Poor system voltage regulation capabilities or controls result in over-voltages.
1.4 Background on Power Quality 19

1.4.3.2 Under-Voltage

Their causes are the opposite to over-voltage ones: e.g. a big load switching on or a
capacitor bank switching off. Overloaded circuits may lead to under-voltage, too.

1.4.3.3 Sustained Interruptions

A decrease to zero of the supply voltage for a period of time longer than 1 minute
is a sustained interruptions. They are often permanent in nature and require manual
intervention to restoration.

1.4.4 Voltage Imbalance

Voltage imbalance or unbalance is defined as the ratio of the negative or zero sequence
component to the positive-sequence component. They main causes of these zero and neg-
ative sequences are unbalanced loads causing which results in a flow of zero and negative
sequence currents.

Imbalance is estimated (expressed as %) using the expression:

maximum deviation f rom average voltage

voltage imbalance = · 100%. (1.46)
average voltage

The main source of smooth voltage imbalance (less than 2%) is the single-phase loads
(unbalanced by definition) in three-phase circuits. Severe voltage imbalances (greater
than 5%) can result from single-phasing conditions.

1.4.5 Waveform Distortion

Waveform distortion is steady-state deviation from the ideal sinusoidal wave of funda-
mental frequency. It is well characterized by the spectral content. They are five primary
types of waveform distortion:

1.4.5.1 Dc Offset

It is the presence of a dc voltage or current in an ac power system. Geomagnetic

disturbances and effects of half-wave rectification may lead to dc offset. Direct current in
20 Chapter 1 Introduction

ac systems has some damaging effects such as an increase in the transformer saturation,
additional stressing of insulation and others.

1.4.5.2 Harmonics

Harmonics are sinusoidal voltages or currents having frequencies that are integer mul-
tiples of the fundamental (50 or 60 Hz). Harmonics combine with the fundamental voltage
or current, and produce distortion.

The cause of harmonics is the non-linearity of devices and loads which produce current
harmonics. These current harmonic also produces voltage harmonics because of the volt-
age drops in the power system impedances. Harmonic distortion is a growing concern for
many customers and for the overall power system due to increasing application of power
electronics equipment.

Harmonic distortion levels can be characterized by the complete harmonic spectrum

(magnitudes and phases). It is also common to use to use the total harmonic distortion
(THD) [33]:

s
sum of all squares of amplitude of all wave harmonics
T HD = · 100% =
square of the amplitude of the f undamental component
qP
∞
(1.47)
h=2 Vh
· 100%.
V1

1.4.5.3 Inter-Harmonics

Inter-harmonics are voltage or current components with a frequency that is a non-

integral multiple of the fundamental frequency. They can appear as discrete frequencies
or as a wide-band spectrum. The main sources of inter-harmonics are static frequency
converters, cyclo-converters, induction motors and arcing devices. Their effects are still
not well studied, but it is known that they produce visual flicker in display devices and
also affect to power line carrier signaling. Inter-harmonics are harmonic components at
non multiple

1.4.5.4 Notching

It is a periodic voltage disturbance caused by the normal operation of power elec-

tronics devices when there is commutation. Since notching occurs continuously it can
be characterized through the frequency spectrum; however, typical notching frequencies
1.4 Background on Power Quality 21

are high and therefore difficult to measure accurately with normal equipment. As shown
in the next chapter, notching can cause severe faults in power electronics converters for
which the control algorithm uses zero-cross detection circuits.

1.4.5.5 Noise

Noise is unwanted electrical signals with broadband spectral content lower than 200 kHz
superimposed over the power system voltages or currents. It is caused by power electronics
devices, control circuits, arcing equipment, non-linear loads and switching power supplies.
Noise phenomena is also very related with grounding. Basically, noise is any unwanted
distortion that cannot be classified as harmonics or transient.

1.4.6 Voltage Fluctuations

Voltage fluctuations are systematic variations of the voltage envelope for which the
voltage range does not exceed the range 0.95 − 1.05 p.u..

Any load with continuous and significant current variations, specially in the reactive
components, may cause voltage fluctuations. Arc furnaces are the most common cause of
voltage variations on the transmission and distribution systems. Voltage fluctuations are
defined by their rms magnitude expressed as a percent of the fundamental.

1.4.7 Power Frequency Variations

The power system frequency is directly related to the rotational speed of the generators
on the power system. At any instant, the frequency depends on the balance between
the load and the generation. When this dynamic balance changes, small changes in
frequency occur. The size of the frequency variations 2 and its duration depends on the
load characteristics and the response of the generation system to load changes.

Frequency deviations outside accepted limits for normal steady-state operation can
be caused by faults on the bulk power transmission system, a disconnection of a large
load or a disconnection of a large source of generation. Frequency variation affecting to
equipment also appear in systems powered by isolated generators.

2
The term power frequency deviations is also employed.
22 Chapter 1 Introduction

1.5 Power Electronic Converters and Applications

This section dealts with power electronic converters connected to an ac mains 3 . The
main devices, architectures, and applications are summarized in this section. A special
emphasis is put in the importance of synchronization and extraction algorithms in their
controllers.

1.5.1 Thyristor Converters

The thyristor is one of the most important types of power semiconductor devices. It
is suitable for circuits where high currents and voltages are involved. It is a four-layer
semiconductor device with three pn-juntions and three terminals: anode, cathode and
gate.

When the anode voltage is positive with respect to the cathode voltage the thyristor is
said to be in off-state condition. It can be turned on by applying a positive voltage between
gate and cathode. When the thyristor is turned on, it goes on conducting while the anode
to cathode current is above the so-called holding current; there is no influence of the gate
signal during the on-state. When the anode to cathode voltage is negative, the thyristor
behaves like a reverse biased diode: there is not current through them (independently of
the gate signal).

Even though the thyristor characteristics show that it is a semi-controlled device, it

can be assumed as ideal switch for many applications such as the thyristor rectifiers.

A rectifier is a power circuit which convert ac input power to dc output power. A

thyristor rectifier uses thyristor as switches. With respect to diode rectifiers, a thyristor
rectifier allows to control the output voltage, that is the power flow from the ac side to
the dc side, by controlling the firing instants of the thyristors gate.

1.5.1.1 Three-phase Full Rectifier

Fig. 1.6 shows a 6-pulse thyristor full rectifier with linear load (zL ). The average
current through the transmission cable (īdc ) directly depends on the average voltage across
the dc side terminals (v̄dc ). The active power in the dc bus is Pdc = v̄dc · īdc .

A balanced set of input voltages can be expressed as function of the line-to-line voltage

3
Even though dc-dc conversion is a very important branch of power electronics, it is approached neither
in this section nor in this Ph.D. dissertation.
1.5 Power Electronic Converters and Applications 23

idc
S1 S3 S5
va Ls

vb Ls
vdc
zL
vc Ls

S2 S4 S6

Figure 1.6: 6-pulse thyristor based rectifier.

vllmax as:

vll vll
va = √max sin(ω1 t) = √max sin(θ1 )
3 3
vllmax 2π
vb = √ sin(θ1 − ) (1.48)
3 3
vll 2π
vc = √max sin(θ1 + ),
3 3
R
where θ1 = ω1 dt is the phase-angle oscillating at ω1 (ϕ1 = 0).

In normal operation, one upper and one lower thyristor of different branches are con-
ducing. Hence, The total number of combinations, or states, is 6. The notation used to
show in which state is the rectifier is Sij , where i refers to the upper branch and j the
lower branch. For example, the state Sab refers to thyristor T 1 and T 4 conducing.

The switch between states is set by the firing of the thyristor. The normal operation
sequence is Scb → Sab → Sac → Sbc → Sba → Sca → Scb → Sab → ... and so on.

From (1.48), va is the highest of the three in the range [30, 150] deg, vb in the range
[150, 270] deg and vc in the range [270, 390 (30)] deg. Moreover, va is the lowest of the
three in the range [210, 330] deg, vb in the range [330, 450 (90)] deg and vc in the range
[90, 210] deg.

If diodes are used instead of thyristor, the switching time are not controlled: each
diode starts conducting when its anode to cathode voltage is positive. Table 1.1 shows
the state dependence with respect to the line phase-angle.

However, if thyristors are used, their conduction could be delayed by choosing the
desired firing angle, named α; the α angle can be set in the range 0 deg < α < 180 deg,
24 Chapter 1 Introduction

Table 1.1
States of diode full rectifier depending of θ1 .

Range of θ1 State
[0, 30] deg Scb
[30, 90] deg Sab
[90, 150] deg Sac
[150, 210] deg Sbc
[210, 270] deg Sba
[270, 330] deg Sca
[330, 360] deg Scb

Table 1.2
State of controlled rectifier depending of θ1 and α.

Range of θ1 State
[0 + α, 30 + α] deg Scb
[30 + α, 90 + α] deg Sab
[90 + α, 150 + α] deg Sac
[150 + α, 210 + α] deg Sbc
[210 + α, 270 + α] deg Sba
[270 + α, 330 + α] deg Sca
[330 + α, 360 + α] deg Scb

that is when the thyristors anode to cathode voltage is positive. The switching times as
function of α are the shown in table 1.2.

v̄dc can be expressed as function of α and vllmax :

1 Zα Z α+pi/6
3
v̄dc = [ vc − vb + va − vb + ....] = · vllmax · cos (α). (1.49)
T 0 α π

As expected, the maximum value of v̄dc corresponds with diode operation, that is
α = 0 deg. The average current īdc depends on v̄dc and the resistive component of zf
(zf (ω = 0)):

v̄dc
īdc = . (1.50)
zf (ω = 0)

The operation of the rectifier is not linear: vdc has ripple components over v̄dc . This
1.5 Power Electronic Converters and Applications 25

α=0 deg.
vdc

α=90 deg.

idc

α=180 deg.

Figure 1.7: Working points of the thyristor full wave rectifier.

ripple is composed of even harmonic components and also affects to īdc . Therefore, this
rectifier is producing reactive power for any α.

Fig. 1.8 shows the rectifier working in different points. The firing instant are the
points where the switching state changes are highlighted as discontinuous lines. It is clear
the non-linear behavior of this converter and its inherent reactive power consumption for
any zf . The ripple is higher as lower is the module of v̄dc .

1.5.1.2 Applications of Thyristor Converters

Nowadays, with the suitability of better controllable switching devices, the use of
thyristors is in three-phase and high-power applications.

1.5.1.2.1 High Voltage dc Transmission Electrical plants generate power in the

form of ac voltages and currents. This power is transmitted to the load centers on three-
phase ac transmission lines. However, for a given quantity of power transmitted, specially
over long distances, dc transmission is more efficient [34].

Fig. 1.9 shows a simple scheme of a HVDC. In order to minimize the losses in the
transmission line due to parasitic reactances the origin ac voltages are transformed into
high dc voltages. A power converter working as rectifier to convert the origin ac signal
to dc is employed. Another power converter is placed in the destination point to convert
the dc voltage to ac (inverter).
26 Chapter 1 Introduction

(a) Three-phase rectifier working at α = 0 deg (Diode mode).

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

Scb Sab Sac Sbc Sba Sca Scb

-0.8
1.955 1.965 1.97 1.975

(b) Three-phase rectifier working at α = 30 deg.

0.6

0.4

0.2

-0.2

-0.4

-0.6

Sca
Sbc Sba Scb Sab Sac Sbc Sba
-1
1.95 1.955 1.965 1.97 1.975

(c) Three-phase =120

rectifier
deg. working at α = 120 deg.

Figure 1.8: Steady-state voltages of the thyristor three-phase full rectifier.

1.5 Power Electronic Converters and Applications 27

idc
Origin ac System Destination ac System

Rectifier Inverter

Figure 1.9: Simple HVDC diagram.

Ldc
ac dc
ac
Motor

Rectifier Inverter

Figure 1.10: ac Motor controlled with Thyristor-based inverter and rectifier.

θ

1.5.1.2.2 Flexible ac Transmission Systems A Flexible Alternating Current Trans- α

mission System (FACTS) is a power electronic based system and other static equipment
that provide control of one or more ac transmission system parameters to enhance con-
trollability and increase power transfer capability [35]. Thyristor based devices are a very
important part of the FACTS family.

Static VAR Compensators (SVC) provide fast-acting reactive power compensation on

high-voltage electricity transmission networks: Thyristor Controlled Reactors (TCR), and
Thyristor Switched Capacitors (TSC) are shunt-connected devices controlled with thyris-
tor which absorb/provide reactive power. SVCs are used both on bulk power transmission
circuits to regulate voltage and contribute to steady-state stability; they also are useful
when placed near high and rapidly varying loads, such as arc furnaces.

There are also series devices such as Thyristor Controlled Series Capacitors (TCSC),
Thyristor Controlled Series Reactances (TCSR), etc citeHingorani2000.

1.5.1.2.3 Ac Motor Drives Induction motors with squirrel cage rotors are very suit-
able for industry because of their low cost and robustness. When operated from the line
voltage as input, they work at constant speed and torque. However, by means of power
electronics converters, it is possible to control speed, torque and even position.

Fig. 1.10 shows a scheme used for the speed control of an ac induction motor using two
thyristor converter, one as rectifier and another one as inverter. This scheme is employed
in very large power rating applications, where a very fast dynamic response is not needed.
28 Chapter 1 Introduction

The rectifier converter controls the dc voltage, and the inverter controls the ac voltage
in the input of the ac motor. Varying this ac input voltage, the point of operation of the
motor (speed and torque) is controlled.

1.5.2 Pulse-Width-Modulated Converters

Pulse Width Modulation (PWM) is the more widespread technique to set both the on
and off firing signals of power converter switches.

The use of thyristors in PWM converters requires a forced commutation scheme. How-
ever, there are some other devices, which allow a better on-off switching control and are
suitable from lower to medium-high power rated applications: Insulated Gate Bipolar
Transistors (IGBTs), Metal Oxide Semiconductor Field Effect Transistors (MOSFETs),
Bipolar Junction Transistors (BJTs), etc. The main advantage of these devices with re-
spect to Thyristor is the fact that they are bidirectional in current. However, they only
block voltage in one way (unidirectional in voltage).

In general, PWM converters are suitable to work in four quadrant operation under the
power flow point of view. Owing to this flexibility, PWM power converters are employed
in vast range of applications.

1.5.2.1 Overview of Modulation Techniques

A brief explanation of PWM operation is shown in this section. Lets suppose the
one-leg switch inverter of Fig 1.11. Ideal switches are considered instead of real power
electronic devices (S+ and S− ). Each switch has a control signal, which set if the switch
is in conducting (on state, boolean 1) or not (off state, boolean 0). Only one switch per
phase can be in on state, else the dc link would be short-circuited. The instantaneous
value of the voltage in the point “a” (va ) depends on which switch is on: vdc /2 or −vdc /2.
The rate of commutation for the switches sets the switching frequency (fs ). The maximum
value of fs is limited by the real devices (e.g. around 20 kHz for IGBTs). The average
value of va , averaged each commutation cycle, defined as v̄a will depend on the time each
switch in on. The duty cycle (d) is defined as the ration of time per commutation cycle
the switch is on per total commutation cycle. So:

vdc −vdc
v̄a = d + (1 − d) . (1.51)
2 2

Therefore setting d for each commutation cycle accordingly, the output voltage in “a”
can be controlled in real time. Equation 1.51 can also be expressed by the PWM control
1.5 Power Electronic Converters and Applications 29

+ vdc + S+
2
vdc -
a
vdc +
S- +
2 - van
- -
n
Figure 1.11: One leg of a switch mode inverter.

or modulating signal (m) (−1 < m < 1):

vdc
v̄a = m . (1.52)
2

1.5.2.1.1 Sinusoidal PWM Sinusoidal PWM is one of the most employed technique
to generate the control signals of the power switches. An example applied to the branch
of Fig. 1.11 is provided as follows.

The firing signals of the switches are obtained by comparing a triangular wave (vtri )
of amplitude 1 and switching frequency (fs ) with a control signal m. The switching rules
are:

vdc
m > vtri S+ ON, S− OF F ⇒ va =
2 (1.53)
−vdc
m < vtri S+ OF F, S− ON ⇒ va = .
2

Fig. 1.12 shows the firing signal of S+ obtained with these switching rules (S− is
the opposite to S+ ). m rotates at 50 Hz, a frequency much lower than vtri frequency,
fs = 10 kHz.

Using these switching rules, the v̄a , averaged each switching cycle, is:

vdc
v̄a = m . (1.54)
2
30 Chapter 1 Introduction

0.8

0.6 vtri m
0.4

0.2

-0.2

-0.4

-0.6

-0.8

1 1.1 1.2 1.3 1.4 1.5

+

Figure 1.12: Set of S+ firing signal with sinusoidal PWM.

Fig. 1.13 shows the output of instantaneous and filtered va .

Fig. 1.14 shows Fourier coefficients of va . It is shown how, even though the control
signal has only a 50 Hz component, va has much more spectrum components. There are
relative high components around the switching frequency (and its harmonics), and, in
general, a continuous non-zero spectrum.

There are several modulation techniques for power electronic converters. E.g. SVM
algorithms, which result in different switching sequences and spectrum contents.

1.5.2.2 Voltage Source Converters

Voltage-source converters (VSC) with PWM modulation are, by far, the most popular
converters for ac power supplies up to a few megawatts [36]. Fig. 1.15 shows the topology
of a three-phase VSC; ei (i = a, b, c) is the per-phase output voltage of the VSC. The
average value of ei (ēi ), averaged each commutation cycle (fs ), can be set through the
PWM or SVM techniques [16, 25].

Considering only the fundamental component of each ei , the power flow between the
ac mains and the dc side of the VSC, both active and reactive, can be controlled. This is
called four quadrant operation.
1.5 Power Electronic Converters and Applications 31

0.8

0.6

0.4
m
0.2

-0.2

-0.4
vtri
-0.6

-0.8

0.08 0.082 0.084 0.086 0.088 0.09 0.092 0.094 0.096 0.098 0.1

vdc/2
a

-vdc/2
0.082 0.084 0.086 0.088 0.09 0.092 0.094 0.096 0.098

Figure 1.13: va output with sinusoidal PWM.

Figure 1.14: Fourier Components of va .

32 Chapter 1 Introduction

id
S1 S3 S5
va La ea

vb Lb eb vd Cd

vc Lc ec

S2 S4 S6

Figure 1.15: VSC connected to an ac source.

vi ii Li ei

Figure 1.16: Per-phase representation of VSC connected to an ac source.

Fig. 1.16 is the per-phase representation of the VSC, where i means a, b or c phase, and
vi refers only to the supply fundamental component. The system is considered rotating
at ω1 .

From Fig. 1.16 results clear that the current through Li (ii ) depends on vi and ei . In
phasor notation this can be expressed as:

V̇i − Ėi
I˙i = = imax ∠φ1 , (1.55)
jω1 Li
where V̇i = vmax ∠0 and Ėi = emax ∠β1 ; β1 being the phase-angle offset between vi and ei
4
.

From power definitions:

S = 3 · vrms · irms
P = 3 · vrms · irms · cos(φ1 ) (1.56)
Q = 3 · vrms · irms · sin(φ1 )
4
It should be noticed that the angle references can be set with respect to V̇i .
1.5 Power Electronic Converters and Applications 33

ℑ=Q
Quadrant II Quadrant I
VSC supplies active power to the grid. VSC absorbs active power from the grid.
VSC absorbs reactive power from the grid. VSC absorbs reactive power from the grid.

ℜ=P
Region III Region III
VSC supplies active power to the grid. VSC absorbs active power from the grid.
VSC supplies reactive power to the grid. VSC supplies reactive power from the grid.

Figure 1.17: Complex power diagram of grid-connected VSC.

where rms values are easily obtained from the peak values as:

vmax
vrms = √
2
imax (1.57)
irms = √ .
2

Therefore, by assessing emax and β1 accordingly, the energy transfer between the ac
side and dc side of the VSC can be controlled. Fig. 1.17 shows the complex plane to
represent the four quadrant operation of the VSC connected to the ac grid.

These complex power can also be represented using P and Q as axis (P − Q diagram).
Fig. 1.18 shows the different operation points of the VSC connected to the ac mains. It
is clear that the VSC is a very flexible device when connected to the grid: from the point
of view of the fundamental components, it is possible to absorb and supply active power,
but also positive and negative reactive power. It is typical to refer to the VSC as Voltage
Source Inverter (VSI) when it is working suppling active power to the ac grid, that is
dc-ac conversion.

VSCs can handle components rotating at higher frequencies than ω1 , e.g. harmonics.
The maximum frequency that a VSC is able to control in steady-state is related to the
controller bandwidth. Generally, it is assumed that the VSC bandwidth is the tenth (or
eleventh) part of the PWM frequency. For example, an IGBT based VSC commutating
at 10 kHz can handle with harmonics around the 20 one (1 kHz) [37, 38]. VSCs can be
connected in series, between the ac mains and another electrical system.
34 Chapter 1 Introduction

V&i Q
E& i
Q

β1
I&Lsi
β1 V&Lsi
P P
I&Lsi φ1 V&Lsi
φ1 V&i
E& i

E& i V&Lsi Q I&Lsi Q

P
P
E& i
φ1
I&Lsi V&
φ1
i
V&i V& Lsi

E& i V&Lsi I&Lsi

β1
Q Q
V&i
φ1
P P
I&Lsi
φ1 V&i β1 V&Lsi
E& i
Q
E& i
Q
I&Lsi V&i
I&Lsi β1 V&Lsi
P P
β1
φ1 V&i
V&Lsi
E& i

Figure 1.18: P − Q diagram. Operation points of grid-connected VSCs.

ac dc
ac
Motor

Rectifier Inverter

Figure 1.19: Control of induction motor by PWM-VSI with diode rectifier

θ
1.5.2.3 Applications of PWM Converters
α
PWM converters are employed in a vast range of applications.

1.5.2.3.1 Ac Motor Drives As said, VSC and other PWM converter topologies used
as inverter can control both frequency and magnitude of their outputs. On the other hand,
the speed of rotation of ac motors depends on the frequency of the stator voltage and the
torque on the magnitude. Therefore, it is possible to control both torque and speed of
an induction motor by means of, for example, a PWM converter as depicted in Fig. 1.19.
The ac line voltage is rectified by means of an uncontrolled diode rectifier and the inverter
feeds the stator with a voltage which controls its torque and rotation speed.

1.5.2.3.2 Uninterruptible ac power supplies An uninterruptible power supply

(UPS) is a device which maintains a continuous supply of electric power to connected
1.5 Power Electronic Converters and Applications 35

equipment by supplying power from a separate source when utility power is not available.
A UPS can be used to provide uninterrupted power to equipment for 1–20 minutes until
a generator can be turned on or utility power is restored.

1.5.2.3.3 Power Line Conditioners A power line conditioner is a device intended

to improve the “quality” of the power that is delivered to electrical equipment. Power line
conditioners based on PWM converters are employed to prevent power line disturbances
from disrupting the operation of critical loads (medical equipment, industrial processes,
...).

Important power line conditioners are:

• Reactive power compensator [3, 11].

• Power Factor Correction: a PWM converter can be placed in parallel with the rest
of loads supplying a fundamental current so the power factor from the ac source
remains around 1 [39, 40].

• Active Power Filters (APF) for harmonic compensation: the PWM converter can
be placed in series or in parallel with the protected loads, to compensate for voltage
or current harmonics [8, 41–45].

• Dynamic Voltage Restorer (DVR): the PWM converter is placed in series with a
sensible loads so it is prevented from voltage sags [46–50].

• UPSs can be considered power line conditioners since they provide protection against
all common power problems [51–53].

1.5.2.3.4 Flexible ac Transmission Systems Due to the ever-increasing power

rating of PWM converters they were proposed as FACTs devices [35]. The most significant
applications are

• Static Synchronous Compensators (STATCOM). A STATCOM operates as a shunt-

connected static var compensator whose capacitive or inductive output current can
be controlled.

• Static Synchronous Series Compensators (SSSC). A SSSC operates without any

external electric energy source as a series compensator whose output voltage is in
quadrature with, and controllable independently of, the line current for the purpose
of increasing or decreasing the overall reactive voltage drop across the line and
thereby controlling the transmitted electric power. The SSSC may include energy
storage to enhance the dynamic behavior of the power system.
36 Chapter 1 Introduction

• Universal Power Flow Controllers (UPFC). An UPFC is a combination of a STAT-

COM and a SSSC which are coupled via a common dc-link, to allow bi-directional
flow of active power between the series output terminals of the SSSC and the shunt
output terminals of the STATCOM, and are controlled to provide concurrent real
and reactive series line compensation without any external energy source.

1.6 Control of Power Electronic Converters and Syn-

chronization

Even though analog control circuits are still being proposed and used, the clear trend
is the implementation of the control algorithms in digital devices. Current devices such as
Digital Signal Processors (DSP), Field Programmable Gate Arrays (FPGA) and Micro-
controllers allow the digital implementation of controllers for power electronics converters
[54–56]. Digital implementation offers quite a lot of advantages with respect to analog
circuits, above all its flexibility in the design, implementation and maintenance.

It said, the P-Q operation point of a grid-connected VSC depends on φ1 . In general,

the controllers of power electronic converters implement synchronization algorithms to
assess exchanges of electrical energy. Measurements at the point of common coupling
(PCC) are made in order to estimate voltage/current absolute phase-angles (θ1v and/or
θ1i ). In this work, the estimated values have a “hat”: e.g. the estimated value of θ1v is
expressed as θ̂1v . Estimated values are employed to assess φ1 and hence, the P-Q operation
point.

Fig. 1.20 shows a simplified scheme of dc-link voltage (vdc ) feedback controller using a
PI filter and SRFs obtained through the Park transformation (P1+ ). The phase-angle for
P1+ generation is obtained from a synchronization block (θ̂1v ). Assuming perfect tracking
(θ̂1v = θ1v ) the grid current (iabc ) is in-phase with the grid voltage (vabc ), so there is only
active power flow from the PCC to the dc-link.

Dc-link voltage controllers are implemented in a wide range of applications, where

capacitors are employed as energy storage device; e.g.:

• Power Line Conditioners [16, 41–44].

• Controlled Rectifiers with a high power factor [16, 26, 57–59].
• Motor Drives [60].
• FACTs devices [35].

Using the values of table 1.3, the dc controller has been simulated in order to show
the role of the synchronization algorithm.
1.6 Control of Power Electronic Converters and Synchronization 37

PCC iabc Lr
Current ac dc
vabc Sensors Side Side
Cdc vdc ZL

VSC
Firing
iabc Signals Voltage
Sensor
ac Voltage iq1+ iq1+*=0
Sensors
SRF
P1+ + Current
id1 Regulator id1+*
- vdc
θ1
PI
)v +
vabc
Synch. vdc*
Digital Controller

Figure 1.20: Simplified dc-link voltage control loop of VSC.

Table 1.3
VSC based Rectifier Values.

vmax 1 p.u.
vdc
∗
1.5 p.u.
Cdc 3 p.u. at 50 Hz
ZL (resistive) 1 p.u.
Voltage Regulator ≈ 50 Hz
Bandwidth
Current Regulator ∞
Bandwidth
38 Chapter 1 Introduction

0.5

-0.5

-1
5.01 5.02 5.03 5.04 5.06 5.07 5.08 5.09 5.1

0.5

-0.5

-1

-1.5
5.01 5.02 5.03 5.04 5.06 5.07 5.08 5.09 5.1

Figure 1.21: VSC operation in steady-state. Effect of phase-error.

Fig. 1.21 shows steady-state voltages/currents of the power system for perfect tracking
(dotted) and when there is a constant steady-state phase error (θ1e = θ1v − θ̂1v ) of ≈
14 deg (solid). The effect of non-zero phase-error is a displacement power factor cos(φ1 )
reduction, both for single-phase and three-phase systems. A cos(φ1 ) 6= 1 results in Q 6= 0,
so higher rated currents are needed to maintain the bus voltage at the reference value
(vdc
∗
). Therefore, average steady state phase error (θ̄1e ) 5 should be canceled to get an
optimized real-time controller.

Fig. 1.22 shows steady-state voltages/currents when there is ripple of 2th harmonic
over the phase estimation. As shown, ripple in the measurement infers also ripple in iabc
and vdc 6 . Therefore, noise in phase estimation reduces the whole performance of the
5
Even though in PWM/SVM the average values are averaged each cycle of fs , in synchronization the
averaging is related to f1 .
6
The frequency of vdc ripple is 4ω1 , not 2ω1 .
1.6 Control of Power Electronic Converters and Synchronization 39

1.501
1.5008

1.5006

1.5004

1.5002

1.5
1.4998

1.4996

1.4994
1
1.4992
1 1
1.499
5.01 5.02 5.03 5.04 5.06 5.07 5.08 5.09 5.1

Figure 1.22: VSC operation in steady-state. Effect of ripple in θ̂1 .

dc-link controller.

A comparative between a relatively fast and a slow synchronization block is provided

in Fig. 1.23. As expected, the system with a faster estimation recovers first the steady-
state, and the vdc
∗
maximum deviation from its nominal or reference value (vdc∗
) is minor.
This second point may be critical in some systems, since power electronic converters
experiment malfunctions when the dc-link voltage is lower than a threshold value.

From this significant examples, it results clear the important role of synchronization
in the control of grid-connected power electronic converters. The performance of the
synchronization block is critical in very fast responding applications as well as for high
power rated systems.
40 Chapter 1 Introduction

0.8
a
b
0.6

0.4

0.2

-0.2

-0.4

-0.6
c
-0.8

-1
13.98 13.985 13.99 13.995 14.005 14.01 14.015 14.025 14.03

50

1
1
30

20

10

-10

-20
13.9 14.1 14.2 14.3 14.4 14.6 14.7 14.8

1.6

1.55

1.45

1.4

1.35
dc
1.3
dc

13.9 14.1 14.2 14.3 14.4 14.6 14.7 14.8

Figure 1.23: VSC operation during a transient. Effect of the re-tracking speed.
1.7 Structure of the Document 41

1.7 Structure of the Document

This work is organized as follows.

1.7.1 Chapter 2: State-of-the-Art in Grid Synchronization

This chapter provides an in-depth study of the state-of-art in grid synchronization.

The most significant proposals in the field are analyzed, making special emphasis in pros
and cons of each one. The study starts with the most simple strategies such as the zero-
cross detectors and their big limitation dealing with power quality phenomena. Next, the
proposal of using a VCO inside the controller of a power converter (PLO system) proposed
for the first HVDC systems is deeply analyzed. The PLO based HVDC successfully
handle with the harmonic instability problem. Subsequently, the study of different PLLs
is provided, making special emphasis in the PLLs implemented in digital devices such
as the SRF-PLL. In the following section, some interesting alternatives to PLLs, such
as the use of FIR and stochastic filters in SRFs, are analyzed. Finally, the frequency
adaptability is studied for each synchronization technique. Besides this deep State-of-art
study, some sections of this chapter are novel contributions of this dissertation. More
specifically, the schemes based on SRFs and Moving Average Filters (MAFs), as well as
frequency adaptation techniques.

1.7.2 Chapter 3: Dynamics Study of Low-Gain PLLs

Chapter 3 is devoted to digital PLLs design. In a general way, PLLs with a low
bandwidth (low-gain PLLs) are required when handling with distorted voltages. It is
analytically demonstrated that low-gain PLLs have more trade-offs than high-gain PLLs
(e.g. PLLs for communications): it is not possible to optimize the settling time for a
phase-jump without getting slower the PLL response to frequency variations. Existing
tuning methods do not take into account low-gain features, which may result in non-
optimum designs. An intuitive tuning methodology based on inspection of frequency-
domain diagrams is contributed in this chapter. Contrary to the other existing tuning
methods, it takes into account low-gain dynamics. It is assured an optimized performance
in the presence of any kind of disturbances in the grid. In the second part of the chapter,
the DCO based on a RC oscillator is presented: the digital model of a sinusoidal oscillator
is implemented, instead of explicit trigonometric functions. This solution reduces the
needed digital resources without reducing the performance, which could be specially useful
for DSP-based control of power converters.
42 Chapter 1 Introduction

1.7.3 Chapter 4: Predictive based SRF-MAF Synchronization

Algorithms

Chapter 4 presents a novel open-loop synchronization system, designed from SRF-

MAF schemes. Accurate measurements of phase, frequency and amplitude are carried
out in real-time.

Previous works establish that the fundamental positive-sequence vector of a set of

utility voltage/current vectors can be decoupled using Park transformations and low pass
filters. However, the filtering process introduces delays that impair the system perfor-
mance. More specifically, when there is grid frequency deviations, a non zero average
steady-state phase error appears in the measurements.

To overcome such limitations, a suitable combination of predictive and moving av-

erage FIR filters is proposed to achieve a robust synchronization system for all input
frequencies. MAFs are linear phase FIR filters which have a constant time delay at low
frequencies. A characteristic which is exploited to good effect to design a predictive filter
which compensates such time delays, enabling zero steady-state phase errors for shifted
input frequencies.

In summary, the main attributes of the new system are its good frequency adaptation,
good filtering/transient response trade-off and the fact that its dynamics is independent of
the input vector amplitude. Comprehensive experimental results validate the theoretical
approach and the high performance of the proposed synchronization algorithm.

1.7.4 Chapter 5: Frequency Adaptive PR Current Controller

for APFs

Chapter 5 shows experimental results of an Active Power Filter implementation. A

feedforward control scheme has been implemented, where the main parts are the SRF-
MAFs extraction algorithms working together a proportional-resonant (PR) based current
regulator which assures perfect tracking at the selected harmonics. The resonant blocks
are tuned using frequency domain techniques. In order to get frequency adaptation, that
is, perfect tracking for any input frequency, the difference equation implementing each
resonant block depends explicitly on the frequency deviation. This frequency deviation
is estimated from a PLL. An in-depth study of the optimized implementation and prac-
tical considerations about the resonant controllers are contributed. Experimental results
obtained from a 1 kVA rated laboratory prototype prove the accuracy of the theoretical
approaches and the feasibility for industrial applications such as power line conditioners
and distributed power generation.
1.7 Structure of the Document 43

1.7.5 Chapter 6: Conclusions and Future Work

The main contributions of this work are reviewed in the last chapter. An outline of
programmed future works is also provided.
44 Chapter 1 Introduction
Chapter 2

State-of-the-Art in Grid
Synchronization

Abstract —

The control of the power flow between a grid-connected power converter and the ac-mains
requires an online tracking of the fundamental component voltage (or current) phase-angle.

The very first synchronization schemes were based on an open loop estimation of the phase-
angle from the observation of zero-crosses. However, the presence of power quality phenomena,
specially in weak-grids, led to malfunctions in zero-cross based controllers. The introduction
of voltage controlled oscillators (VCOs) resulted in more robust controllers such as the Phase
Locked Oscillator (PLO) system and the Charge-Pump Phase Locked Loops (CP-PLL).

With the suitability of discrete devices such as micro-controllers the number of synchroniza-
tion algorithms and filtering techniques has grown drastically, in parallel with the appearance of
high performance applications (high bandwidth) and new technical requirements in fields such as
renewable energy applications, traction systems and power line conditioners. Digital Phase locked
Loops (PLL) and algorithms implementing stochastic and/or FIR filters are clear examples of
the high performance offered by digital implementation.

However, the research in the synchronization field cannot be considered complete at all. The
appearance of new applications and/or technical requirements of existing ones would need the
development of specific solutions. A review in synchronization strategies for power converters is
contributed in this chapter.

Even though the main part of this chapter is devoted to synchronization with fundamen-
tal component vectors, the issues of harmonics and negative-sequences extraction are also ap-
proached, since to synchronize with those components is required in several applications.

It should be noticed that, even though this chapter may seem to be presented in a temporal-line
manner, many of the approaches have coexisted since their proposal to nowadays.

45
46 Chapter 2 State-of-the-Art in Grid Synchronization

Input D1 D3 R3 Output

Dgen Dgen 1k
VAC1 V C1 V
1000u

0 Q1
R2
VAC2 0
1k
D2 QgenN

R1
Dgen
VAMPL = 21.2V 100k
FREQ = 50Hz

0
(a) Simple analog zero-cross detector circuit.
25
V(output)
20

15

10
Voltage (V)

-5

-10
V(input)
-15

-20

-25
0 0.01 0.02 0.03 0.04 0.05 0.06
Time (s)

(b) Key waves for a sinusoidal input wave.

Figure 2.1: Zero-cross detector example.

2.1 Methods based on Zero-Cross Detection

A zero-cross detector is a circuit which detects the transition of a signal waveform

from negative and positive, and vice versa, providing a narrow pulse that coincides with
the zero voltage condition.

When using with ac networks, zero-cross detectors obtain the information of the phase-
angle: 0 deg and 180 deg coincide with the pulses at the outputs.

In Fig. 2.1 an example of analog zero-cross detector connected to the utility grid is
shown. The input wave is supposed to be obtained from the secondary of a 220/15 V and
50 Hz transformer. The transistor Q1 is polarized in the active zone except during the
zero crosses. Under such a situation Q1 is in cut-off. As there is no current through Q1,
the voltage in the output is a bit lower than the peak value of the input voltage.

Fig. 2.1b shows the result of a time domain simulation, obtained with PSPICE, when
the input wave is ideal. As expected, the output has pulses during the zeros crosses.
2.1 Methods based on Zero-Cross Detection 47

Input U1
+ Output
VAC1 V R1
OUT
1k
V R2
- OPAMP 1k
0
VAMPL = 21.2V 0
FREQ = 50Hz
0
(a) Simple analog zero-cross detector circuit.

25

20

15

10
Voltage (V)

5
V(output)
0

-5

-10

-15

-20
V(input)
-25
0 0.01 0.02 0.03 0.04 0.05 0.06
Time (s)

(b) Key waves for a sinusoidal input wave.

Figure 2.2: OA based zero-cross detector example.

Moreover, as the input wave frequency is just 50 Hz, the phase between zero crosses can
be extrapolated by means a multi-vibrator delay circuit or a digital counter [61].

The zero-cross detector circuit works per-phase. In three-phase systems it is common

to use one zero-cross circuit per-phase. A simpler solution is to use the phase-angle
extrapolated from one phase to set the zero crosses of the other phases, assuming balanced
conditions. The circuit of Fig. 2.1a can be easily adapted to obtain separate signals for
the zero-crosses at 0 deg and 180 deg; another transistor is needed.

The operational amplifier (OA) has been widely used for zero-cross detection. Fig. 2.2a
shows a simple circuit of zero-cross detection using an OA. Fig. 2.2b shows the result of
the time domain PSPICE simulation. When the input signal is sinusoidal, a square wave
in-phase with the input signal is obtained.
48 Chapter 2 State-of-the-Art in Grid Synchronization

idc
Origin ac System Destination ac System

Rectifier Inverter

Figure 2.3: Simple HVDC diagram.

2.1.1 The First High-Voltage Direct Current Systems

The first implementation of power converters handling high power flows can be placed
in time with the high-voltage direct current systems (HVDC). HVDCs were initially used
to transmit large amounts of power between two ac systems over long distances. The
first commercial HVDC was developed between 1950 and 1954 in Sweden connecting the
Swedish island of Gotland with mainland. At that time, silicon devices were not suitable
for such a high power, so power converters were based on mercury-arc valves [62]. The
firing of these mercury-arc valves were equivalent to thyristor firing.

Fig. 2.3 shows a simple scheme of a HVDC. In order to minimize the losses in the
transmission line due to parasitic reactances, the origin ac voltages are transformed into
high dc voltages. A power converter is used working as rectifier to convert the origin ac
signal to dc. Another power converter is placed in the destination point to convert the dc
voltage to ac (inverter).

The control of a HVDC system is quite complex and requires many controllers includ-
ing steady-state operation and protection against transients; this subsection is focused in
the steady-state operation.

The most widely method employed to control the average 1 power flow from the origin
ac side to the dc bus is the so-called constant current (CC). The average value of the
rectifier current (īdc ) is controlled through the firing angle α.

2.1.1.1 Constant-α Method

These first rectifier controllers were based on the so-called constant-α and inverse-
cosine methods. Fig. 2.4a shows a simplified scheme of the constant-α method.

The control scheme is a circuit which implements a zero-cross detector and a variable
1
averaged each cycle of the fundamental f1 .
2.1 Methods based on Zero-Cross Detection 49

Mercury-Arc based 6-pulse Mercury-Arc based 6-pulse

Power Electronics Rectifier Power Electronics Rectifier

idc idc
AC DC v AC DC v
dc dc
Side Side Side Side

Per-phase Per-phase
AC Voltage AC Voltage
Sensors Sensors

Per-valve Per-valve
Zero-cross Controlled phase-lag Per-valve Firing
variable Firing Zero-cross
detection network circuits
circuits delay circuits detection
and
idc* circuits idc* comparator circuits
(external ref) (external ref) Controller circuit
Controller circuit

(a) Constant-α rectifier control scheme. (b) Inverse-cosine rectifier control scheme.

Figure 2.4: Constant-α vs Inverse-Cosine HVDCs.

delay circuit per valve. The zero-cross detection circuits should detect separately 0 deg
and 180 deg zero crosses: 0 deg cross for upper valves and 180 deg for the lower ones. The
α angle is controlled externally (angle reference) which controls the delay of the variable
delay circuits.

2.1.1.2 Inverse-Cosine Method

The basic principle is that the delay angle can be obtained from the crossing point of
an associated “cosine-wave” with a control reference voltage at a comparator. The cosine-
wave is obtained from a phase-lag network so the output of that network has a peak at
the delay angle (α) equal to zero [63]. The firing pulse is generated at the point at which
the cosine-wave is equal to the control voltage (vr ), which is directly proportional to ī∗dc .
emax being the peak of the cosine-wave, the firing condition is set by the equation:

cos(α) = vr /emax . (2.1)

Fig. 2.4b shows a simplified scheme of the constant-α method. The main advantage
of this method is the fact that v̄dc varies linearly with vr .

2.1.2 Drawbacks of Zero-cross Detectors

Even though this circuit obtains good results for a lot of applications, the growing up
concern about power quality problems have led to consider other high performance solu-
tions. The main problems of zero-cross detectors are well summarized in [64]: harmonics
and noise.
50 Chapter 2 State-of-the-Art in Grid Synchronization

Fig. 2.5a shows the results of the time domain SPICE simulation when the input
wave has a 20% of 5th harmonic over the fundamental. As said, this harmonic changes
the zero-cross detection points, so there is non zero average phase error in steady-state.
With regard to the control of power converters, this error could easily lead to instability
and malfunction [65].

Fig. 2.5b shows the results of the time domain SPICE simulation when the input wave
contains notches. There are misleading zero-cross detections due to the notching, which
result in very high phase error.

Another drawback appears when considering input wave transients. Zero-cross de-
tectors only estimate the phase-angle one or two times per cycle, so when a phase-angle
jump occurs is not detected until the next zero-cross. Moreover, zero-cross detectors do
not provide frequency estimation or adaptation, which could lead to malfunctions in the
presence of frequency deviations.

2.1.2.1 Improvements in Zero-cross Detectors

In order to improve the performance in the zero-cross detection, some simple but
practical actions have been proposed, e.g.:

• In three-phase circuits a common practice has been to filter the b − phase and
extrapolate the a − phase from the filtered signal [66].

• In order to deal with frequency variation in applications of motor control, some

enhanced zero-cross detector circuits have been contributed [67, 68].

• With the suitability of digital devices implementing discrete operation, the possi-
bility of digital filtering resulted in a vast range of filtering alternatives. e.g. in [69]
the fundamental component is reconstructed from a notching wave, assuming that
the short-circuit parameters are known; in [70] an 11th order FIR filter is proposed
to cancel out harmonics.

However, digital implementation makes suitable the implementation of algorithms with

a much higher performance than these based on zero-cross detectors. The suitability of
implementing trigonometric functions, multipliers, IIR and FIR filters, etc, gives much
more versatility than analog integrated circuits (ICs), as shown below.

2.1.2.2 Harmonic Instability of Zero-cross based HVDCs

Nowadays, it is well known that zero-cross detectors implementation in systems con-

taining harmonics could lead to malfunctions. It was in a work published by J. D.
2.1 Methods based on Zero-Cross Detection 51

30

Detected
20
Zero cross
Real
10 Zero cross
Voltage (V)

-10

Fundamental
-20 of V(input)

V(input)
-30
0 0.01 0.02 0.03 0.04 0.05 0.06
Time (s)
(a) Key figures under harmonic contaminated input.

25
Good
20 zero-cross
detection
15

10
Misleading
Voltage (V)

5 zero-cross
detection
0

-5

-10

-15

-20 V(input)

-25
0 0.01 0.02 0.03 0.04 0.05 0.06
Time (s)
(b) Key waves for a input wave with notches.

Figure 2.5: Main drawbacks of zero-cross detectors.

52 Chapter 2 State-of-the-Art in Grid Synchronization

Xs Xr

AC DC
Side Side
AC
Source
Thyristor or Mercury-arc valves
Controlled Power Electronics Converter

Figure 2.6: Weak Grid. Short circuit ratio Xr /Xs ≈ 3 − 6.

Ainsworth in 1967 [65], when a problem called “harmonic instability” was reported 2 .

Harmonic instability occurs when the rectifier was connected to a weak system, a
system for which the short-circuit ratio (Xr /Xs ) is low: e.g. in the range 3 to 6 [71].

The harmonic instability is a problem that is characterized by magnification of non-

characteristic harmonics (6= 5, 7, 11, 13) in steady-state. This is due to the fact that any
distortion in the ac side such as unbalance and harmonics affects to the firing instants.
This leads to a situation where the firing instants are not equidistant in steady-state
and to the generation of non-characteristic harmonics. The problem of generation of
non-characteristic harmonics could be even aggravated if some of that harmonics are in
parallel resonance with the circuit [34, 65].

In order to overcome the problem of harmonic instability, passive filters were proposed,
but it was an insufficient alternative [34]. The real milestone in the control of the HVDC,
and indeed in the control of power electronic converters was the Ainsworth proposal of [1]:
the firing instants are independent from voltage zero-crosses through a voltage controlled
oscillator (VCO) inside the control loop. This proposal is deeply analyzed in section 2.2.

2.2 Phase-Locked Oscillator Control Systems

A surprising novel scheme for the control of power converters was contributed by J.
D. Ainsworth in [1]: the phase-locked oscillator (PLO) control system. The work of
Ainsworth is related with the need of a robust controller for HVDC systems connected to
weak ac grids, a problem he had posed in his previous paper [65].

The first PLO was a control system for the HVDC rectifier. Under ideal operation
conditions and during steady-state, this converter produces 5th , 7th , 11th , 13th , etc har-
monics in the ac side and 6th , 12th , etc in the dc side. Moreover, other harmonics may
2
’Harmonic instability’ does not mean instability in a feedback control system.
2.2 Phase-Locked Oscillator Control Systems 53

Xs Xr Ld idc

AC DC
vdc zL
Side Side
Current
Sensor

Firing C1
Signals
C2 R1
a
- idc*
-
ω̂1
VCO +
vco +

Figure 2.7: Scheme of the PLO controller for the HVDC rectifier proposed in [1].

also appear if unbalance and timing errors in the control systems are considered. The
presence of voltage harmonics in the ac side is higher as higher is the impedance between
the power converter and the generation (weak system) [65, 71]. The problem of those
converters working in a weak grid is the dependence of the control from the measure-
ment of the ac voltages, which could result in harmonic instability. Ainsworth proposed
a closed-loop controller containing a voltage controlled oscillator (VCO) in order to reject
the harmonics effect.

Fig. 2.7 depicts a PLO simplified scheme. The inductance current (idc ) is controlled in
a closed loop; i∗dc being its reference value. An analog filter is placed after the comparator;
the output of this filter is the input voltage of the VCO (vco ). The VCO generates
output pulses at a repetition frequency six times the system voltages/currents fundamental
frequency; the frequency of the VCO output signal implicitly represents the estimated
fundamental frequency (ω̂1 ). The output pulses, generated through a ring counter, are
employed to fire the thyristors/valves. When there is a positive difference in the output of
the comparator, the vco voltage tends to rise and, therefore, ω̂1 increases, so α is reduced.
For negative difference, the behavior is the opposite one.

2.2.1 PLO Linearization

Even though Fig. 2.7 represents a quite simplified model of the PLO system, it has
a very non-linear behavior. This subsection describes a linearization of the PLO system.
The linearization of the PLO was made by Ekstrom and Liss in [72], and, even though
Ainsworth was skeptical with this procedure (see discussion of [72]), nowadays it is clear
that linearization is a powerful tool to describe non-linear systems.

The average value of the voltage in the dc link (v̄dc ) is function of the line to line peak
54 Chapter 2 State-of-the-Art in Grid Synchronization

voltage (vllmax ) and the firing angle α:

3
v̄dc = · vllmax · cos(α) = vdcmax · cos(α), (2.2)
π
3
where vdcmax = π
· vllmax .

Lets suppose a perturbation in the firing angle ∆α << 1, a small change in vdc (∆vdc )
is supposed. By introducing ∆α and ∆vdc in (2.2)

∆vdc = −vdcmax · sin(α) · sin(∆α), (2.3)

which can be approximated to

∆vdc = −vdcmax · sin(α) · ∆α. (2.4)

∆α is function of the difference between ω1 and ω̂1 , during a time ∆t:

∆α = −(ω̂1 − ω1 ) · ∆t = −∆ω1 · ∆t, (2.5)

and, in the Laplace domain,

∆ω1
∆α = − . (2.6)
s

Therefore,

∆ω1
∆v̄dc = vdcmax · sin(α) · , (2.7)
s
which goes on being a non linear equivalence.

The relation between vdc and idc depends on the inductance Ld and the load impedance
(zL ):

vdc
idc = . (2.8)
Ld s + zL

Linking (2.4) and (2.8),

∆ω1
vdcmax · sin(α) · s
idc = . (2.9)
Ld s + zL
2.2 Phase-Locked Oscillator Control Systems 55

∆ω̂1 vd max ⋅ sin(α )

CONTROLLER PLANT
C2 ⋅ R1 ⋅ s + 1
⋅e
K vco − g ( t ) t
*
idc 1
s C1 ⋅ C2 ⋅ R1 ⋅ s + (C1 + C2 ) Ld ⋅ s + z L
+-
s
idc
Analog filter VCO and Firing Circuit Rectifier circuit

Figure 2.8: Equivalent block diagram of PLO circuit and control.

∆ω1 has a linear relation with the input of the VCO:

∆ω1 = Kvco · vco , (2.10)

so

vdcmax · sin(α) · Kvcos·vco

idc = . (2.11)
Ld s + zL

Ainsworth designed a controller for the plant set by (2.11). This equation is non linear;
the plant depends on the firing angle through sin(α). Another source of non linearity is
due to the fact that the controller only acts at 60 deg intervals, that is, there is a time
dependent delay function, which should be considered in the design of a controller.

The controller proposed by Ainsworth was:

C2 · R1 · s + 1 Kc
C(s) = · (2.12)
C2 · C1 · R1 · s + (C2 + C1 ) s

Kc being a constant. Fig. 2.8 shows the control block diagram of the PLO.

In order to study the dynamics of the system, this model should be evaluated in all
the range of possible values of α. In [1] the minimum α was 5 deg. It is clear that, the
lower α, the lower the gain of the loop.

As pointed by Ainsworth the circuit in steady state is always in the range [0, 180] deg.
This implies in practice that ω̂1 ≈ ω1 so

1
g(t) ≤ . (2.13)
6 · f1n

Therefore, the firing delay can be modeled taken into account the worst case: the
delay is maximum ( 6·f11n ) (Fig. 2.9).
56 Chapter 2 State-of-the-Art in Grid Synchronization

∆ω̂1 vd max ⋅ sin(α )

CONTROLLER PLANT
C2 ⋅ R1 ⋅ s + 1 −
⋅e
* 1
idc Kc
s C1 ⋅ C2 ⋅ R1 ⋅ s + (C1 + C2 ) Ld ⋅ s + z L
+-
K vco 6 f1n
t

s
idc
Analog filter VCO and Firing Circuit Rectifier circuit

Figure 2.9: Equivalent block diagram of PLO circuit and controller for firing maximum
delay.

Table 2.1
HVDC values.

vdcmax 1 p.u.
zL (Resistive) 1 p.u.
Ld 1/(10 · π) p.u.

The tuning of the PLO controller presents several difficulties. In sum:

• The plant changes with sin(α).

• It has two poles at the origin. The phase versus frequency response is −180 deg at
dc.

• The introduced delay tends to lag the phase, reducing the phase margin (PM).

• There are another two poles: one at −(C1 + C2 )/C1 · C2 · R1 rad/s due to the con-
troller and another at −zL /Ld rad/s due to the plant.

• There is only a zero due to the controller in −1/(C2 · R1 ) rad/s.

In order to avoid a negative phase margin for very low frequencies and, therefore, to
have an unstable system, the controller should have its zero very near to the origin. The
bandwidth of the system is limited by the second pole of the controller.

2.2.1.1 Design Example

A more in-depth study of a model would require the knowledge of the practical sys-
tem values employed by Ainsworth. In this section, an example of the PLO system is
contributed in order to show its performance 3 .

The most important data of this table is the relation between zL and the equivalent
inductance Ld , because this relation sets a pole in the plant. The controller parameters
of the example are shown in table 2.2.
2.2 Phase-Locked Oscillator Control Systems 57

Table 2.2
Controller Values.

R1 100 kΩ
C1 100 pF
C2 100 µF
α range [5, 90] deg
Kc 10−5
Kvco 100 rad/V

300 300
Total
Total
200 200

Controller
Magnitude (dB)

Amplitude (dB)

100 100
Controller
Plant

0 0
Delay Delay

-100 -100
Plant

-200 -3 -2 -1 0 1 2 -200 -3 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10 10 10

0 0
Delay Delay Controller
-50 Controller -50
Plant
-100 -100
Phase (deg)

Phase (deg)

Total Plant
-150 -150 Total
-180 -180
-200 -200

-250
-250

-300 -3
10 10
-2 -1
10 10
0
10
1 2
10 -300 -3 -2 -1 0 1 2
10 10 10 10 10 10
Frequency (Hz)
Frequency (Hz)

(a) Equivalent linearized model for α = 5 deg. (b) Equivalent linearized model for α = 90 deg.
P M = 66.5 deg, at 1.35 Hz. P M = 19.92 deg, at 8.25 Hz.

Figure 2.10: PLO equivalent model at different α values.

58 Chapter 2 State-of-the-Art in Grid Synchronization

0.75

0.65
0.94

0.93 0.6

0.55
0.92

0.91

0.45

0.89 1 1.5 2 2.5 3 3.5 4 4.5

0.5 1.5 2 2.5 3 3.5 4 4.5

(a) PLO Start-up. Step in i∗dc from 0.9 p.u. to (b) PLO Start-up. Step in i∗dc from 0.55 p.u. to
0.95 p.u.. 0.5 p.u..

Figure 2.11: Start-up simulation of PLO for different i∗dc (dotted), idc is the blue line.

As shown in Figs. 2.10 it is possible to tune a controller stable for all the α range. As
pointed by Ekstrom in [72], for higher α values the system is difficult to stabilize. In fact,
the PM for α = 90 deg is very small. But, on the other hand, as pointed by Ainsworth
in the discussion of [72], it is also true that for lower α values the tuning becomes more
concerning since the bandwidth becomes very small and the system could easily go out
from normal operation.

From Figs. 2.10 an important feature of the PLO is observed. The magnitude gain
decreases very fast with the frequency, so it is expected a good current harmonic cancel-
lation.

The simplified scheme of Fig. 2.7 with the parameters and component values of
table 2.2 has been simulated through the PLO Matlab scripts of appendix (from A.1
to A.3).

Fig. 2.11 shows the start-up for a the PLO working in different points (α). After the
initial transient the current reference is perfectly tracked. Fig. 2.11a shows the start-
up when α is 18.2 deg. In such a situation, the output current in steady-state should
be 0.95 p.u.. Fig. 2.11b shows the start-up when α is 60 deg, so the output current in
steady-state should be 0.5 p.u.. A comparative between both figures shows how for lower
α the overshoot is lower, since the PM is higher. Higher α results in a shorter rise time
but with a higher overshoot, since the PM is shorter. These results are expected from
Figs. 2.10.

Figs. 2.12 show the system voltages when steady-state is reached for both cases shown
in Fig. 2.11. As zL has been chosen to be 1 p.u. pure resistive, v̄dc should be equal to īdc .

Figs. 2.13 show the results of a simulation when the voltages system is unbalanced
both in amplitudes and angles. Specifically the phase to neutral voltages of the test have
3
Component values and tests made in this section may differ from first PLO tests.
2.2 Phase-Locked Oscillator Control Systems 59

1 1
0.95
Vdc Vdc
0.8 0.8
Vdc Vdc
0.6 0.6
0.5
0.4 0.4

0.2 0.2

0 0

Vb Vc
-0.2 -0.2 Va Vb
Va Vc

-0.4 -0.4

-0.6 -0.6
5.95 5.955 5.96 5.965 5.97 5.975 5.98 5.985 5.99 5.995 6 5.95 5.955 5.96 5.965 5.97 5.975 5.98 5.985 5.99 5.995 6

(a) Steady state voltages for Fig. 2.11a. (b) Steady state voltages for Fig. 2.11b.

Figure 2.12: Steady-state system voltages for Figs. 2.12.

1
Vdc Vdc
0.8
0.78
0.7
0.6
0.76

0.4
0.74
0.2

0.72
0
Vb Vc
-0.2
Va

0.68 -0.4

0.66 -0.6
1 1.5 2 2.5 3 3.5 4 4.5 5.95 5.955 5.96 5.965 5.97 5.975 5.98 5.985 5.99 5.995 6

(a) PLO Start-up for an unbalanced set of voltages. (b) Steady state voltages for Fig. 2.12b.
Step in i∗dc (dotted) from 0.8 p.u. to 0.7 p.u., idc is
the blue line.

Figure 2.13: PLO tested under a very unbalanced system of voltages.

the expression

va = 1 · sin(ω1n t)
vb = 0.8 · sin(ω1n t − (2 ∗ pi/3 + 0.1)) (2.14)
vc = 0.9 · sin(ω1n t + (2 ∗ pi/3 + 0.1)).

Perfect tracking under unbalanced conditions is observed; the PLO rejects voltage
unbalance effects.

Figs. 2.14 show the results of a simulation when the voltages system is unbalanced
both in amplitudes and angles and also very distorted by harmonics. Specifically, the
phase to neutral voltages of the test have been:
60 Chapter 2 State-of-the-Art in Grid Synchronization

1.2
0.84

Vdc
0.82
Vdc
0.8

0.6

0.78

0.4

0.76
0.2

0.74

0.72
Vb
-0.2 Va
Vc
-0.4

0.68 -0.6

-0.8
1 1.5 2 2.5 3 3.5 4 4.5 5.955 5.96 5.965 5.97 5.975 5.98 5.985 5.99 5.995

(a) PLO Start-up for an unbalanced set of voltages. (b) Steady state voltages for Fig. 2.14a.
Step in i∗dc (dotted) from 0.8 p.u. to 0.7 p.u., idc is
the blue line.

Figure 2.14: PLO tested in the presence of an unbalanced and polluted with harmonics
system of voltages.

va = 1 · sin(ω1n t) + 0.1 sin(5ω1n t) + 0.1 sin(7ω1n t)

vb = 0.8 · sin(ω1n t − (2 ∗ pi/3 + 0.1)) + 0.1 sin(5ω1n t + 0.1) + 0.1 sin(7ω1n t − 1)
vc = 0.9 · sin(ω1n t + (2 ∗ pi/3 + 0.1)) + 0.1 sin(5ω1n t + 0.9) + 0.1 sin(7ω1n t − 0.7).
(2.15)

Perfect tracking under unbalance and harmonics is observed. In this way, the problem
of harmonic instability in weak grids was overcome with the PLO.

Figs. 2.15 show the results of a simulation when the voltages system is very distorted
and there is a load transient: at 2.5 s a change from 1 p.u. to 1.1 p.u. has been programmed.
The PLO system retracks perfectly after the transient, so it could be said that the PLO
is robust in the presence of load transients.

Figs. 2.16 show the results of a simulation when the input frequency has a step change
and how the VCO frequency estimation (fˆ1 ) follows its reference f1 . Therefore, another
important feature of the PLO is its good frequency adaptation.

In sum, the PLO does not only fix the harmonic adaptation problem. The proposal
of using a VCO circuit inside the controller of a power converter also resulted in a very
effective tool to overcome other problems such as unbalance, frequency adaptation and
noise cancellation.
2.2 Phase-Locked Oscillator Control Systems 61

0.84
1.2
Vdc
0.82
1

Vdc

0.78
0.6

0.76
0.4

0.74
0.2

0.72
0
Vb
-0.2
Va
Vc
0.68
-0.4

0.66
-0.6

0.64
1 1.5 2 3.5 4 4.5 -0.8
5.955 5.96 5.965 5.97 5.975 5.98 5.985 5.99 5.995

(a) Ainsworth model Start-up and transient at (b) Steady state voltages for Fig. 2.15a.
2.5 s for a distorted set of voltages. The reference
current is 0.7 p.u., idc is the blue line.

Figure 2.15: PLO transient test: sudden change in zL .

0.85
50.7

50.6

50.4
0.75

50.3

50.2

50.1

2 4 6 8 12 14 16 18 2 4 6 8 12 14 16 18

(a) PLO current response. i∗dc (dotted) is 0.7 p.u., (b) Grid (f1 ) VS VCO (fˆ1 ) Frequencies for Fig.
idc is the blue line. 2.16a.

Figure 2.16: PLO tested under a sudden change in the input frequency at 10 s.
62 Chapter 2 State-of-the-Art in Grid Synchronization

2.2.2 Other Phase-Locked Oscillator Schemes

The PLO system was also proposed with other schemes. Some of the most significant
ones are detailed below.

2.2.2.1 PLO for Voltage Control

An alternative to the Ainsworth controller has been posed in [73]. In this scheme a
thyristor based controller is controlled in a closed loop with a VCO inside. The difference
of this scheme with respect to the Ainsworth one is the fact that the controlled variable
was the dc link voltage, so in practice, the second pole of the plant disappeared. This
allowed to Skjellnes et al to develop a closed loop with a bandwidth of ≈ 100 Hz, so
the transient response is highly enhanced with respect to the Ainsworth PLO. The main
limitation of this scheme is due to the fact that most of the practical applications require
a current control loop.

2.2.2.2 PLO for Speed Control of dc Motors

The PLO scheme has been adapted to the speed control of dc motors [63, 74, 75]. A
speed encoder and a phase comparator circuit are implemented in order to obtain the error
signal which controls the VCO frequency and, hence, the firing of the thyristor converter.

Based on this scheme, a frequency adaptive system with an acceptable transient re-
sponse (around one cycle of ω1n ) is achieved in [74].

2.3 Phase Locked Loops

As shown in the previous section, the Ainsworth proposal of using a VCO inside the
contro loop of a power converter provides harmonics/unbalance rejection and frequency
adaptation. However, the PLO system is very difficult to tune, and extra controllers
monitoring correct operation are needed [1].

An alternative use of the VCO is in phase locked loops (PLLs) tracking the funda-
mental phase-angle, as proposed in [76, 77]. PLLs had been used in other fields, specially
communications, since the 40’s [78, 79]. The works of [76, 77] extended the applicability
of PLLs to the control of power electronic converters.

The main novelty of the PLL proposal is the fact that the VCO is not a part of the
main control algorithm; the VCO is employed inside a measurement circuit (estimation).
2.3 Phase Locked Loops 63

vd
ac

dc Motor
ac Voltage Sensor

θˆ1
ω̂1
va PLL Comparator PWM

α ref
Circuit
Main
vd* Control
Controller Circuit

Figure 2.17: Control scheme of a dc motor in a single-phase grid.

Fig. 2.17 shows the scheme proposed in [76] for the control of a thyristor controlled
rectifier in a single-phase motor drive application.

Even though the application is simpler than the HVDC rectifier of [1], it is clear that
the use of the PLL as a measurement circuit “liberates” the main controller of two poles
at the origin, which tend to make the system unstable.

The schemes of [76, 77] are very simple because the main control reference (vdc
∗
) is set
open loop (Fig. 2.17). However, as shown below, the kind of PLLs employed in [76, 77]
has more limitations considering power quality phenomena than PLO systems. Even so,
it is clear that the proposal of using PLLs inside the controllers is, without any doubt, a
milestone in the field of synchronization and power electronics control in general.

Fig. 2.18a shows a simplified PLL scheme. A PLL is a non-linear circuit which
synchronizes its output signal (vo ) with a reference or input signal (vi ) in frequency as
well as in phase.

The
Rπ
fundamental frequency and phase-angle of vi are ω1 and θ1 ; they are related by
θ1 = −π ω1 t + ϕ1 where ϕ1 is an offset dc constant. In the same way, the output signal
vo parameters are the estimated values of vi : ω̂1 , θ̂1 and ϕ̂1 . In steady-state it is expected
that estimated and real values are the same.

The PLL scheme is composed by the three basic functional blocks: the phase detector
(PD), the loop filter (LF) and the voltage controlled oscillator (VCO). The role of each
block could be summarized as follows:

• The PD is a circuit for which the average value of its output, averaged each cycle
64 Chapter 2 State-of-the-Art in Grid Synchronization

θ1 = ω1t + ϕ1 θ1e ∆ω̂1

ω1n
PD VCO (DCO)
vi ve
Phase Loop Loop + +
+-
Detector Filter Filter

θˆ1 = ωˆ1t + ϕˆ1 ω̂1

∫π
π
vo
Voltage
vco
Controlled

ωˆ1 = ω1n + K vco vco −

Oscillator

(a) Block diagram of a PLL. (b) Linear model of a PLL.

Figure 2.18: Basic conceptual models of PLLs.

ω +θ
of the fundamental component of the input, is zero only when the input signals are
in-phase, that is, they are synchronized in phase and frequency.

• The LF supplies a filtered control signal (vco ) to the VCO.

• The VCO generates a signal of frequency ω̂1 from its central frequency, which, in
grid applications, should be adapted to ω1n and v̄c . The estimated frequency is
ω̂1 = ω1n + Kvco · v̄c , where Kvco is the VCO constant. Later, in digital PLLs, the
term Digitally Controlled Oscillator (DCO) is employed instead of VCO (integrated
circuit).

PLLs should be linearized in order to study its dynamics and tuning the LF. This
process is done assuming that the PLL is near locked state, that is, ω1 = ω̂1 and θi ≈ θ̂1
[78, 79]. Under such a situation, the equivalent system is the linear PLL (LPLL) of Fig.
2.18b.

The LPLL is a typical closed loop servo system. The instantaneous phase-angle error
at the fundamental component is θ1e = θ1 − θ̂1 . The LF path provides a frequency
correction around the nominal (central) frequency (∆ω̂1 ). When designing a PLL, its
most important features are:

1. Stability: the LPLL model should have high enough stability margins in order to
assure the stability of the PLL.

2. The “type” of the PLL: it is set by the number of origin poles of the equivalent
LPLL. A type 2 PLL (two origin poles) is necessary to assure θ̄1e = 0 when frequency
deviations are considered [78, 79].

3. The LPLL bandwidth sets

• the transients response. The higher the bandwidth, the faster the PLL retracks
the phase-angle after a transient.
2.3 Phase Locked Loops 65

if( (V(%IN1)>0.9 & V(%IN2)<0.1) |

U7
(V(%IN2)>0.9 & V(%IN1)<0.1),-5,5) U5
+
VAC2 Raux V OUT 1 Vcp R1 Vc
1k 3 IN OUT
Raux2 2
V
- 1k
OPAMP 100k V C1 V
Phase Detector & Charge Pump 10u VCO_sqr
0
VAMPL = 21.2V 0 0
FREQ = 49
0

Figure 2.19: Block diagram of an analog CP-PLL.

• the harmonic/noise filtering. The presence of noise and harmonics in the inputs
affects the PLL measurements. The term “phase jittering” is employed by
Best and Gardner to refer to the effect of noise in the phase estimation: phase
jittering (θ1ej ) is the error in the value of θ1e during zero crosses due to noise 4 .
2 2
In a general way, the average value of θ1ej (θ̄1ej ) is inversely proportional to
the PLL bandwidth [78, 79].

2.3.1 Charge-Pump PLLs

As said, the schemes implemented in [76, 77] are pioneer in power electronics, and
gave rise to much more development in the PLL technique during the end of 70’s and
early 80’s [80, 81].

PLLs proposed in [76, 77, 81] are Charge-Pump PLLs (CP-PLLs) [79, 82]. CP-PLLs
have a sequential-logic PD working together with a charge pump. The purpose of the
charge pump is to convert the logic states of the PD into analog signals suitable for
controlling the VCO. The main advantage of the charge pump is the chance of using
analog filters.

Fig. 2.19 shows the PSPICE model of a simple CP-PLL using an EXOR gate as PD.
The main characteristic of the CP-PLL using an EXOR gate is that, in tracking state,
the output leads the input signal 90 deg (cosine-wave).

The parts of this model are explained as follows:

• A comparator with zero is placed in order to generate a digital signal for which the
logic value changes occur synchronized with the ac grid zero crosses.

• The input and feedback signals from the VCO are the inputs of an EXOR gate.
The output of the EXOR gate controls a Charge Pump. The output of the EXOR-
Charge Pump system is an analog wave (vcp ) for which the average value, averaged
each cycle of the fundamental, depends on the relative phase between the two digital
66 Chapter 2 State-of-the-Art in Grid Synchronization

Input 2
1 (Feedback)
(AC Comp.)
Input 1
Input 1
AC Comp
AC Comp
Input 2 Input 2
Feedback Feedback

EXOR inputs
EXOR inputs

0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Time (s) Time (s)

Averaged

Charge Pump Output (V)

Charge-Pump Output (V)

Averaged

Instantaneous

Instantaneous

0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Time (s) Time (s)

′
(a) Key diagrams for θ1e = π/2 rad (θ1e = 0 rad). (b) Key diagrams for θ1e = π/4 rad
′
(θ1e = −pi/4 rad).

Figure 2.20: Key diagrams for the EXOR-Charge Pump Phase Detector

input signals. Through defining θ1e

′
= θ1e − pi/2:

2vcpmax ′ ′
v̄cp = · θ1e = Kd · θ1e . (2.16)
π
Fig. 2.20 shows the outputs of the EXOR-Charge-Pump PD.
• Considering v̄cp ≈ vco , from eq. (2.16) a negative θ1e
′
results in a negative vco , so the
ˆ
VCO frequency (f1 ) decreases. As the feedback signal has a lower frequency than
the input one, θ1e
′
tends to decrease, and therefore, the system tends to retrack the
phase. For positive θ1e′
the behavior is opposite.
• The role of the LF is to cancel the ripple in the output signal of the PD. As shown
in Fig. 2.20, the output of the PD has a fundamental component (v̄cp ) but also
harmonics, which affect to the VCO dynamics. For this reason, the LF should be a
low pass filter (LPF). In the model a first order LPF with cut-off frequency equal
to 100 rad/s has been implemented.

Fig. 2.21 shows time domain simulation results for the CP-PLL of Fig. 2.19. The
values of table 2.3 were employed. The system tracks the input phase and frequency in
around 3 cycles of the fundamental (f1n = 50 Hz).

The PD detector of this simulation is of EXOR type. However, the so called phase-
frequency detectors (PFDs) providing a feedback signal in-phase with the input (no 90 deg
2.3 Phase Locked Loops 67

25
20
15
10
AC input (V)

5
0

-5
-10

-15
-20

-25
0 0.05 0.1 0.15
Time (s)
Input 2 (VCO)
Input 2 (VCO)
1

0.8
EXOR inputs

0.6

0.4

0.2

0
0 Input 1 (AC) 0.05 0.1 Input 1 (AC) 0.15
Time (s)
VCO input, averaged value (Averaged each 20.4 ms)
5
VCO input
Voltage (V)

0 -0.1V

-5
0 0.05 0.1 0.15
VCO input Time (s)
Charge-Pump output

Figure 2.21: Time domain simulation results from the CP-PLL model at f = 49 Hz.
68 Chapter 2 State-of-the-Art in Grid Synchronization

Table 2.3
Controller Values.

Kd 3.18 rad/V
R 100 Ω
C 10 µF
Kvco 10 V/Hz

shift) have been the most widely employed in power converters [75].

On the other hand, even though some current works implement CP-PLL [83], the use
of these kind of solutions are being abandoned by digital implementations. The limitations
of CP-PLL are:

• Most of the classical VCOs such as the ICs 8038, 4044, 4046, and Siemens TCA
780-785, etc are becoming “obsolete components” [84]. Digital implementation is
definitively a current trend.

• The PD is based on zero-cross detection and therefore, could introduce phase error.

• In three-phase systems, as they work per phase, they cannot detect the fundamental
positive-sequence.

2.3.2 Digital SRF-PLL

Fig. 2.22 shows the Synchronous Reference Frame PLL (SRF-PLL or dq-PLL). The
main feature of the SRF-PLL is that it estimates the phase-angle of the fundamental
+ +
positive-sequence (vabc 1
): therefore, the estimated phase-angle (θ̂1 ) corresponds with va1
+ +
phase-angle. If vabc waves are considered sine-waves, P1 of (1.40) is implemented, but
most of the authors usually present the SRF-PLL with the Park transformation obtained
+
as in (1.44), which implies that vabc components are considered cosine-waves. Anyway,
both approaches are totally equivalent.

The SRF-PLL structure was proposed by Gole et al in some simulation models of

rectifiers and Static var compensators [85, 86]. On the other hand, the SRF-PLL real
time implementation in a digital device was presented by Kaura and Blasko in [28], even
though a similar implementation had been previously proposed by Blasko et al in [87].

SRF-PLLs are very suitable to work together with other SRF controllers: the phase-
+
angle of va1 sets the origin of the positive-sequence SRF, so that a SRF-PLL assures an
accurate power flow control. This fact was crucial in the success of the SRF-PLL, since
SRF-based current controllers have been employed in all kind of high performance power
converters applications, such as:
2.3 Phase Locked Loops 69

vd+1 + f (hω1 )
PD
va

vq+1 + f (hω1 )
P1+ ∆ω̂1
vb
vc LOOP
FILTER
+ + ω1n

θˆ1 = ωˆ1t + ϕˆ1 ω̂1

∫π
π

−
DCO

Figure 2.22: SRF-PLL block diagram.

• Motor drives [60, 88–90].

• Controlled rectifiers [26, 58, 59, 91, 92].

• Power Line Conditioners [8, 27, 41–43, 93].

• DPGSs and Grid Monitoring [94, 95].

• UPS [51, 53, 96, 97].

The field of application of the SRF-PLL is very wide and, hence, its optimization has
been approached in several works [22, 28, 46, 94, 98–114]. The feasibility of powerful
digital devices and novel implementation techniques have led to more and more enhanced
implementations.

2.3.2.1 SRF-PLL Operation

+
Considering vabc a set of phase to neutral “sine-waves”, the Park transformation (2.17)
+
acts as phase detector and the quadrature component (vq1 ) as phase error signal (θ1e ), so
+
that vq1 ≈ θ1e .

 
  va
vd1+  
  = P+ 
1 ·

v , (2.17)
vq1+  b 
| {z } vc
v+
dq
| {z }
vabc

where

 
2  sin(θ̂1 ) sin(θ̂1 − 2π ) sin(θ̂1 + 2π )
P+
1 = 3 3 . (2.18)
3 cos(θ̂1 ) cos(θ̂1 − 2π ) cos(θ̂1 + 2π )
3 3
70 Chapter 2 State-of-the-Art in Grid Synchronization

The estimated phase-angle θ̂1 is employed to generate P+ 1 . The SRF-PLL is imple-

mented in a digital device; a digital controlled oscillator (DCO) substitutes the VCO
circuit. The DCO is an integrator and a feed-forward constant with the nominal fre-
quency so the central frequency of the PLL is ω1n . The LF path provides a correction
value (∆ω̂1 ) around ω1n .

For the SRF-PLL linearization, the system is considered around the tracking point,
that is, ω1 = ω̂1 and θ1 ≈ θ̂1 , and also balanced inputs [28]. Under such a situation

   
+
vd1+ va1max cos(θ1 − θ̂1 ) ≈ va1 max
  = +
 (2.19)
vq1+ va1max sin(θi − θ̂1 ) ≈ va1 max
· (θi − θ̂1 ) = va1max · θ1e
| {z }
v+
dq1

+ + + +
where va1 max
(= vb1 max
= vc1 max
) should be rearranged so that va1 max
= 1 p.u.. It is
+
expected that v̄q1+ , which represents θ̄1e , to be zero in steady-state and, therefore, v̄d1 =
+
va1max . If amplitude measurement is not employed, only the q component of the P+ 1 is
generated (less needed resources).

+
It should be noticed that, the zero-sequence of vabc is eliminated in the PD. In fact,
several three-phase and three-wire systems could not have a physical neutral connection.
This is not a problem, but an advantage: phase to phase voltages can be employed in
order to reduce the number of voltage sensors (from 3 to 2) and components of P+1 matrix.
Eq. (2.17) can be rewritten as

 
  va − vc
vd1+  
  = P+ 
1 · vb − vc

, (2.20)
vq1+  
| {z } vc − vc = 0
v+
dq
| {z }
Line to line voltages
being the discrete implementation optimized accordingly.

2.3.2.2 Pre-filters for the SRF-PLL

From PLL theory, the level of distortion in the input signal should be taken into
account to tune PLLs, which is approached in detail for the SRF-PLL in the next chapter.
However, it should be highlighted here that a significant bandwidth reduction may lead
to a very slow transient response.

In order to improve the whole dynamics of the SRF-PLL some authors have proposed
the use of “pre-filters”, which act in a rotating reference frame (vabc or vαβ ). In this
manner, the input of the SRF-PLL is cleaner and therefore it can be tuned with a higher
bandwidth. Some interesting proposals are summarized below:
2.3 Phase Locked Loops 71

vd+1 + f (hω1 )
PD
va va+1 + f (hω1 )

S+ vb+1 + f (hω1 )
P1+ vq+1 + f (hω1 )
∆ω̂1
vb
vc v + f (hω1 )
+
c1
LOOP + + ω1n
FILTER

θˆ1 = ωˆ1t + ϕˆ1 ω̂1

∫
π

DCO −π

Figure 2.23: SRF-PLL with Fortescue operator based pre-filter.

2.3.2.2.1 All-pass Filters based Sequence Detector A sequence detector to pre-

filter the negative-sequence component and feed the SRF-PLL only with positive-sequence
was proposed by Lee et al in [22].

Considering only components at fundamental frequency, the positive-sequence can be

expressed by means of the Fortescue matrix (S+ ):

     
+
va1 1 a a2 va1

 +

 1
 2  
  
vb1 =  a 1 a  · v  , (2.21)
  b1 

  3
+
vc1 a a2 1 vc1
| {z }
S+
√ 2π
3
where a = −1
2
+j 2
or a = ej 3 .

The imaginary term j corresponds with a delay of 90 deg in a real signal, so digital
implementation of (2.21) is possible by means of all-pass filters delaying 90 deg the com-
ponents rotating at ω1 (phase at f1 ). Another important feature of all-pass filters is the
unitary gain for all frequencies.

Fig. 2.24 shows the Matlab/Simulink model of a sequence detector using an all-pass
filter (A(z)) of the form 5 :

−0.967z + 1
A(z) = , (2.22)
z − 0.967
in the Z-domain, implemented at fs = 10 kHz. The frequency and transient response of
A(z) are also shown. A phase of exactly 90 deg at ω1n is achieved.

Fig. 2.25 shows simulation results obtained from the model of Fig. 2.24: the input
signal is unbalanced but the output has only the (estimated) positive-sequence.

This technique is also suitable when there are harmonics in the inputs. As shown
below, some harmonic sequences are canceled with this implementation. However, the
5
It seems that there is not any specific detail of the original A(z) employed in [22].
72 Chapter 2 State-of-the-Art in Grid Synchronization

-0.967 z+1 1/3*(u(1)-0.5*u(3)-0.5*sqrt(3)*u(4)-0.5*u(5)+0.5*sqrt(3)*u(6))

z-0.967
Va Vap
All -pass

-0.967 z+1 1/3*(-0.5*u(1)+0.5*sqrt(3)*u(2)+u(3)-0.5*u(5)-0.5*sqrt(3)*u(6))

z-0.967 Vbp Scope 1

Vb All pass

1/3*(-0.5*u(1)-0.5*sqrt(3)*u(2)-0.5*u(3)+0.5*sqrt(3)*u(4)+u(5))
-0.967 z+1
Vcp
z-0.967
Vc All pass

Frequency Response Step Response

5 1.5
Magnitude (dB)

1
0
Amplitude (p.u.) 0.5

-5
0
-75
Phase (deg)

-80
-0.5
-85

-90
-95 -1

-100

-105 -1.5
40 42 44 46 48
50 52 54 56 58 60 0.018
0.02
0.022 0.024 0.026 0.028
0.03
0.032 0.034

Frequency (Hz) Time (s)

Figure 2.24: Matlab Simulink model and all-pass filter features.

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1

0.01 0.02 0.03 0.04 0.05

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1

0.01 0.02 0.03 0.04 0.05

Figure 2.25: Positive-sequence estimation with all-pass filters (f1 = 50 Hz).

2.3 Phase Locked Loops 73

vα+ vd+1 + f (hω1 )

PD
va vα

Q+ B+ vq+1 + f (hω1 )
∆ω̂1
vb

ω1n
vβ+
C vβ
vc LOOP
+ +
FILTER

θˆ1 = ωˆ1t + ϕˆ1 ω̂1

∫π
π

DCO −

Figure 2.26: SRF-PLL with pre-filters in the αβ frame.

SRF-PLL should filter the non canceled sequences.

2.3.2.2.2 Delayed Signal Cancellation in the αβ Frame Fig. 2.26 shows a

method for negative-sequence cancellation in the αβ frame [115, 116] applied to the SRF-
PLL [102, 117]. It is quite remarkable the fact that working in the αβ frame reduces the
number of filters from 3 to 2.

The input signal vabc is expressed in the αβ frame by means of the Clarke Transform:

 
    va
vα 2  1 −0.5 −0.5   


  = √ √ ·  vb  . (2.23)
vβ 3 0 3
−2 3  
2
| {z } | {z } vc
vαβ C | {z }
vabc

The fundamental positive-sequence is obtained through the Q+ matrix as

     
vα+ 1 1 −q   vα 
 =  · , (2.24)
vβ+ 2 q 1 vβ
| {z }
Q+

−π
where q = ej 2 .

+
Finally the angle transformation (B+ ) 6
is employed to get vdq

     
+
vd1 cos(θ̂1 ) sin(θ̂1 ) vα+

+
 = · . (2.25)
vq1 − sin(θ̂1 ) cos(θ̂1 ) vβ+
| {z }
B+

6
B+ · C = P + .
74 Chapter 2 State-of-the-Art in Grid Synchronization

Frequency Response Step Response

Phase (deg) Magnitude (dB)
5 1.5

D(z)

Amplitude (p.u.)
D(z) A(z) 1
0
0.5
A(z)

-5 0
-84
-86
D(z) -0.5
-88

-90 A(z)
-92 -1

-94

-96 47 48 52 53
-1.5
0.018 0.022 0.026 0.028 0.032 0.034
49 50 51 0.02 0.024 0.03
Frequency (Hz) Time (s)

Figure 2.27: Comparative of the dynamics of A(z) and D(z)

(fs = 10 kHz, so n = 50 in D(z)).

As shown in Fig. 2.26, the pre-filtering is at the real time generation of Q+ . A 90 deg
delay at f1n is applied to get the implementation of (2.24). This delay is obtained as a
FIR filter of the form

1
D(z) = (2.26)
zn
n being the order of the filter, which is so that D(z) has 90 deg of delay at f1n . D(z)
has unitary gain for all the frequencies and linear phase, so it can also be considered an
all-pass filter.

Fig. 2.27 shows a comparison between D(z) and A(z). It is expected a faster transient
response of DSC method, but a worse frequency adaptation.

2.3.2.2.3 Generalized Integrators in PLLs Generalized Integrators (GIs) have

been employed in the control of power converters because they have a very high gain
at a desired harmonic frequencies; e.g. their implementation in Proportional+Resonant
(PR) current controllers [44, 88, 118–122]. The use of damped GIs have been proposed
by Rodriguez et al as SRF-PLL pre-filters in the αβ frame [123].

The generic form of a GI is:

N um(s)
G(s) = . (2.27)
s2 + kω ′ s + ω ′ 2

′ ′
From (2.27), G(s) has a maximum at ω (s2 + ω = 0). When k 6= 0 the GI is said
to be damped. The numerator of G(s) (N um(s)) is set so that the frequency response,
′
specially at ω , is the expected. Two GI filters were proposed in [123]: a direct GI (Hd (s))
2.3 Phase Locked Loops 75

and a quadrature (Hq (s))

kω1n s
Hd (s) = 2
, (2.28)
s2 + kω1n s + ω1n
with a 0 deg phase at f1n = 50 Hz, and

2
kω1n
Hq (s) = 2
, (2.29)
s2 + kω1n s + ω1n
with a 90 deg phase at f 1n = 50 Hz.

So Q+ is implemented in real time (even though expressed in the S-domain) as:

 
Hd (s) −Hq (s)
Q+ =  . (2.30)
Hq (s) Hd (s)

Fig. 2.28a shows the frequency response of both filters. At f1 = 50 Hz both filters
have unitary gain; in terms of phase, Hd (s) = 0 deg and Hq (s) = 90 deg at 50 Hz; Hq (s)
has a better harmonic filtering response since it decays 40 dB/dec. The settling times (ts )
of both filters last around one fundamental cycle, what lags the transient considerably.

2.3.2.2.4 Extended and Generalized Delayed Signal Cancellation As shown,

the real time generation of the Fortescue operator (S+ ) is obtained from delayed signals:
the complex operator i is implemented by delaying 90 deg the fundamental component of
the input signal.

However, when digital delays of the form (2.26) are implemented, the equivalent phase
at harmonic frequencies is different. Indeed, these are “linear-phase” filters. Therefore, at
some harmonic frequencies and sequences the implementation of S+ results in harmonic
cancellation [124].

On the other hand, the real time generation of S+ with delayed signals is not unique.
Taking advantage of this feature, different combinations of the fundamental component
positive-sequence reconstruction can be linked in order to cancel some harmonic sequences
in the αβ frame [124].

The proposal of [124] has been generalized by Bradaschia et al in [111]. Through

mathematical operations in the αβ frame, they prove that any sequence (positive or
negative) at any harmonic of the fundamental frequency (odd and even) can be canceled
choosing the correct S implementation and/or link of some of them.
76 Chapter 2 State-of-the-Art in Grid Synchronization

10

Magnitude (dB)
0
-10

-20

-30

-40

-50
90
Phase (deg)

-90

-180
0 1 2 3
10 10 50 10 10
Frequency (Hz)
√
(a) Hd (s) and Hq (s) frequency response (k = 2).
500

300

200

100

-100
0 0.005 0.01 0.015 0.025 0.03

√
(b) Hd (s) and Hq (s) impulse (transient) response (k = 2).

Figure 2.28: Hd (s) and Hq (s) dynamic response.

2.3.3 Digital Single-phase PLL

The suitability of digital implementation allows to implement PDs with a higher per-
formance than the CP-PLLs ones, which are based on zero-cross detection. From PLL
theory [78, 79], the “linear multiplier” PD seems to be the most intuitive option [112, 125–
128]. Fig. 2.29 shows a simple single-phase PLL scheme based on linear multiplier as PD.

For the linearization, the input signal of equation (2.31) is considered:

v = v1max · sin(ω1 t + ϕ1 ) + f (3ωi , 5ωi , 7ωi , ...) p.u.. (2.31)

| {z }
θ1
2.3 Phase Locked Loops 77

θ1e ω1n
PD
va ( p.u.)
X LOOP FILTER + +

cos(θˆ1 ) θˆ1
DCO

∫π ω̂1
π

cos(u)
−

sin(θˆ1 ) sin(u) θˆ1

Figure 2.29: Single-phase PLL with multiplier as PD.

The wave at the PD output is

v1max
v× cos(θ̂1 ) = · (sin(θ1 − θ̂1 ) + sin(θ1 + θ̂1 )) + f (2ωi , 4ωi , 6ωi , ...). (2.32)
| {z } 2
F eedback wave

Again, assuming that the PLL is locked in steady-state ω1 = ω̂1 and ϕ1 ≈ ϕ̂1 [78, 79],

v1max
v × cos(θ̂1 ) ≈ sin(ϕ1 − ϕ̂1 ) +
| 2 {z }
≈ϕ1 −ϕ̂1 =θ1e
v1 (2.33)
+ max sin(2ω1 t + 2ϕ1 ) + f (2ω1 , 4ω1 , 6ω1 , ...) .
| 2
| {z }
{z } Other harmonics
Generated second harmonic

Equation (2.33) shows that in steady-state the wave has a small dc signal with the
phase error information (θ1e ), a high second harmonic and other even harmonic compo-
nents. Assuming that all the harmonic components are canceled in the LF, the linearized
model is also valid for single-phase PLLs.

Other alternatives for single-phase PLL implementation have been proposed, such as:

• The adaptation of the SRF-PLL to single-phase systems [129–133].

• The implementation of a more complex PD such as the one of the “Enhanced PLL”
(EPLL) of [134], or the one based on the Coullon Oscillator [135]7 .

However, all single-phase PLLs for grid applications need of a drastic bandwidth reduc-
tion due to the PD non-linearities, more specifically the internal generation of harmonics
7
Even though the title could be misleading, [135] proposes a single-phase PLL
78 Chapter 2 State-of-the-Art in Grid Synchronization

at 2ω1 . Therefore, grid-connected single-phase PLLs should not be considered “fast”, so

that they are “low-gain” PLLs. In fact, their typical settling times are not lower than 1
cycle of ω1n [112, 126, 127, 133–136].

2.3.4 Single-phase PLLs Vs SRF-PLLs

+
As said, in a three-phase system the SRF-PLL estimates the phase-angle of va1 . On
the other hand, a single-phase PLL connected in the a-phase tracks the phase-angle of va1 .
Therefore, it seems that SRF-PLLs are more reliable when SRF controllers are employed.
That is, the use of single-phase PLLs in the presence of negative/zero sequences may
result in θ̄1e 6= 0 when tracking the positive-sequence, which may lead to a non efficient
energy transfer control. Some works have proposed the use of single-phase PLLs instead
of SRF-PLLs, since a SRF-PLL cannot detect individual phases [126, 135]. Even though
this assessment could be right, SRF controllers do not seem the most suitable to work
together with single-phase PLLs. This point should have been clarified by those authors.
It could be said that single-phase PLLs are more suitable for single-phase circuits and
three-phase circuits with per-phase controllers.

Other authors suggest a per-phase estimation of va1 , vb1 and vc1 phase-angles and
+
amplitudes in order to calculate the phase-angle of va1 [134]. Even though this is feasible,
there is no apparent reason to use individual single-phase PLLs, since the phase-detector
of the SRF-PLL (P+ +
1 ) totally decouples the positive-sequence: the dc value of vdq1 only
+
depends on vabc1 .

2.3.5 Amplitude Compensation in Digital PLLs

When considering amplitude variations such as voltage sags and swells, the drawback
of amplitude dependence in PLLs appears: from (2.19) and (2.33), it is clear that the
dynamics of digital PLLs depends on the amplitude of the tracked signal. In fact, the
amplitude is considered a proportional gain, which is normalized to 1 p.u. in order to
make easier the tuning process.

The amplitude dependence may lead to instability when the PLL phase margin is too
low. However, in practice, the most reported problem is due to the slow dynamics in the
presence of voltage sags [126, 127, 137].

This amplitude dependence can be avoided in part by compensating the PLL input
with an amplitude estimation. In single-phase PLLs, some works propose an extra loop
tracking the amplitude, which could be named amplitude locked loop (ALL) [126, 127,
136]. Similar approaches have been proposed for the SRF-PLL [111, 137].

Fig. 2.30 shows the combined ALL-PLL structure proposed in [127]. As a general
2.3 Phase Locked Loops 79

v v vˆ 1 max θˆ1
sin(ωˆ1t + ϕˆ1 )
Amplitude Phase
Locked v / vˆ 1 max Locked
loop. Loop.

(a) ALL-PLL global structure.

sin(ωˆ1t + ϕˆ1 )
u2 +- PI

v 1
X
+
+

vˆ 1 max
1/u
v / vˆ 1 max
X
(b) ALL block.

Figure 2.30: ALL-PLL compensation block.

rule, the ALL closed loop bandwidth should be lower than the PLL one, because of the
fact that the amplitude measurement also depends on the phase-angle.

Fig. 2.31 shows experimental results from the ALL-PLL structure. When there is a
voltage sag, the dynamics is faster and the amplitude can be estimated in real-time.

However, the range of improvement achieved with this technique could be considered
quite limited. In sum:

• In the single-phase case, the dependence between loops limits the response. The
study of the combined dynamics is very difficult [127].

• Also in the single-phase case, the bandwidth of the PLL should still consider the
PLL second harmonic [126, 127].

• If a division block is implemented, division by zero or low values could result in

instability.

2.3.6 Limit Cycles of Digital PLLs

Digital PLLs are non-linear systems. Their trajectories can be depicted in the so-called
phase plane portraits [138, 139]. The frequency error ω1 − ω̂1 is represented in the Y axis
and the phase error θ1 − θ̂1 in the X axis. During normal operation, it is expected from a
80 Chapter 2 State-of-the-Art in Grid Synchronization

1.1

0.9

Amplitude (p.u.)
0.8

0.7

0.6

0.5

0.4
0.1 0.15 0.2 0.25 0.3 0.35
Time (s)
(a) ALL-PLL response to a 0.5 sag with (b) v̂1max obtained with the ALL (v gray
phase-angle jump of +45 deg. (v gray dashed, v̂1 dashed, v̂1 black solid).
black solid).

Figure 2.31: Key figures of the ALL-PLL.

PLL to converge to the limit cycle (0, 0). Other solutions of the difference equation defined
by the PLL are expected to be unstable limit cycles. If not, a misleading estimation could
lead to malfunctions of the power electronic converter.

Different parameters of the PLL set its phase-plane portrait: the loop filter, the feed-
forward constant, the word length and other non-linear operations. It should be noticed
that the effect of these parameters cannot be assessed through the PLL linearized model.

Some digital implementations could lead to non desired stable limit cycles (misleading
steady-state estimations). In order to avoid this, the PLL region of operation can be
delimited explicitly.

A study of the phase-plane trajectories of a SRF-PLL is contributed in this section.

The influence of some parameters is shown. Fig. 2.32 shows phase-plane portraits ob-
tained for a SRF-PLL implementation. The Matlab script employed to depict these por-
traits is in appendix A.4. The different curves are obtained considering different initial
conditions of θ̂1 (color legends). Initially this implementation does not take into account
negative values of ω̂1 , that is, the sign of ω̂1 is obviated in the integration step to get θ̂1 .
This action can be considered attractive under the resources point of view, especially in
a fixed point implementation where the word length is very limited. Fig. 2.32a and 2.32b
show the phase-plane portrait with and without feedforward constant. Some important
results are obtained from these figures:

1. A stable limit cycle appears at (2ω1 , −π) when the sign of ω̂1 is obviated. The
outcome of such a wrong estimation in a real time controller could be unpredictable.
2.3 Phase Locked Loops 81

1000 1000

π-0.1 π-0.1
π/2 π/2
π/4 π/4
0 0
π/2 π/2
-π+0.6 -π+0.6
-π+0.1 -π+0.1

-500 -500
-4 -3 -2 -1 1 2 3 4 -4 -3 -2 -1 1 2 3 4

(a) Phase-plane portrait of a SRF-PLL obviating (b) Phase-plane portrait of a SRF-PLL obviating
ω̂1 sign. Feedforward constant implemented (ω1n = ω̂1 sign. No feedforward constant (ω1n = 0).
2π · 50 rad/s).
1000 1000

π-0.1 π-0.1
π/2 π/2
π/4 π/4
0 0
π/2 π/2
-π+0.6 -π+0.6
-π+0.1 -π+0.1

-500 -500
-4 -3 -2 -1 1 2 3 4 -4 -3 -2 -1 1 2 3 4

(c) Phase-plane portrait of a SRF-PLL with ω̂1 sign (d) Phase-plane portrait of a SRF-PLL with ω̂1 sign
information (ω1n = 2π · 50 rad/s). information. The LF path is limited by the word
length (ω1n = 2π · 50 rad/s).

Figure 2.32: Phase-plane portraits of a SRF-PLL implementation. The curves are ob-
tained through different initial conditions (color legends indicate the initial θ̂1 ).

2. Thanks to the feedforward constant there are more trajectories converging to the
good limit cycle (0, 0), but the malfunction remains to be quite likely.

Fig. 2.32c shows the phase-plane portrait considering ω̂1 sign. In such a case, (2ω1 , −π)
is an unstable limit cycle and the SRF-PLL converges to the good estimation. Fig. 2.32d
shows the effect of the word length consideration. The trajectories in the phase-plane
portrait are limited in a smaller region.

Returning to the theoretical approach, a high presence of harmonics may result in a

PLL tracking a harmonic phase. Another real possibility is a non converging PLL. E.g.
a single-phase PLL based on a multiplier may lead to a solution where ω̂1 = 0 (θ̄1e = 0)
and, therefore, the X axis is θ1 .

It could be said that, in general, the region of operation of the PLL should be limited
in order to avoid non desired stable limit cycles and diverging trajectories. In practice,
the best manner to limit the region of operation is by placing the feedforward constant
and limiting the word length [112].
82 Chapter 2 State-of-the-Art in Grid Synchronization

2.4 Digital Alternatives to PLLs

The main drawback of PLLs is their transient response speed, specially in the presence
of distorted conditions [91, 134, 140]. On the other hand, digital implementation allows
to implement very high performance synchronization algorithms without a DCO inside a
closed-loop (a digital PLL).

Several algorithms have been proposed as an alternative to PLLs. The most significant
approaches are analyzed in this section. It should be highlighted that some of these
proposals are novel contributions of this thesis.

2.4.1 Filtering in the αβ Frame

Some algorithms based on the Clarke transformation are analyzed below.

2.4.1.1 Transformation Angle Detector (LP-TAD)

An open loop system, named low-pass transformation angle detector (LP-TAD), able
to synchronize with the positive-sequence of the fundamental vector of a three-phase
voltages/currents system was presented in [141]. This system was proposed as a digital
alternative to CP-PLLs for VSIs based on vector control, where the performance of zero-
cross detectors was very poor in the presence of input harmonics.

Fig. 2.33 shows the scheme of LP-TAD. A three-phase system of voltages system
(va , vb , vc ) is transformed to its equivalent Clarke variables vα and vβ :

 
    va
vα 2  1 −0.5 −0.5   


  = √ √ ·  vb  . (2.34)
vβ 3 0 3
−2 3  
2
| {z } | {z } vc
vαβ C | {z }
~vabc

Assuming that there is not negative-sequence at fundamental frequency, the phase-

angle of the positive-sequence is estimated from v̂α1 and v̂β1 , obtained by filtering har-
monics from ~vαβ , through:

θ̂1 = atan2(v̂α1 , v̂β1 ) ∈ [−π, π], (2.35)

where atan2 is the four-quadrant inverse tangent [142].
2.4 Digital Alternatives to PLLs 83

vα vˆα 1
x vˆα 1
norm

÷ θˆ1
va LPF

vˆα 1 + vˆβ 1
÷ vˆ
vb
C 2 2

β1
norm
(atan2)

vc vβ
LPF x
vˆβ 1

Figure 2.33: LP-TAD scheme.

As said, LPFs are placed in order to filter higher harmonics; they should be tuned
taking into account the trade-off between filtering and transient response. Filters delay
ω
should be compensated at the outputs, which is not a trivial issue considering frequency
ω
∫
variations [105]. π

−π
Fig. 2.34 shows some significant simulation results of the LP-TAD. For these simula-
tions an order 1 Butterworth filter with cut-off frequency 15 Hz has been employed.

As shown, the LP-TAD has a good behavior under balanced conditions. However, it
has a very poor performance in the presence of negative-sequence in the inputs. This lack
of sequence decoupling ability is due to fact that the Clarke variables include information
of both positive and negative sequences [21]. The transient response depends only on the
employed LPFs.

The LP-TAD proves that open loop schemes and inverse trigonometric functions
(atan2) could be an alternative to PLLs if a good decoupling and filtering strategy is
carried out.

Also in [141], the dynamic response of the LP-TAD is enhanced by using recur-
sive/stochastic filters, giving rise to the Space Vector Filter-TAD (SVF-TAD) and En-
hanced Kalman Filter-TAD (EKF-TAD). Stochastic filtering for synchronization is ana-
lyzed below.

2.4.1.2 Normalised Positive-sequence Synchronous Frame (NPSF)

In [143, 144] a scheme to decouple the positive-sequence and then synchronize in the
αβ frame is shown. S+ is implemented using GIs to get 90 deg delay at nominal frequency.
Once decoupled the positive-sequence, open loop estimation is accurate. However, like in
the GI based PLLs, the transient response is not its strongest point.
84 Chapter 2 State-of-the-Art in Grid Synchronization

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1
0.47 0.48 0.49 0.51 0.52 0.53 0.54

(a) Input waves.

0.8

norm
0.6

0.4

0.2
norm

-0.2

-0.4 norm

-0.6

-0.8

-1
0.47 0.48 0.49 0.51 0.52 0.53 0.54

(b) Normalized eα and vβ .

-1

-2

-3
0.47 0.48 0.49 0.51 0.52 0.53 0.54

(c) Phase-angle measurement.

Figure 2.34: Significant simulation results for LP-TAD: fault and unbalance conditions.
2.4 Digital Alternatives to PLLs 85

2.4.2 Filtering in Synchronous Reference Frames

Digital devices allow to handle electrical variables in SRFs. The main advantage of
the SRF approach is the suitability of controlling oscillating waves as dc signals. This is a
big advantage under the designer point of view, since low order infinite impulse response
(IIR) filters (Chebishew, Butterworth), moving average FIR filters (MAF), Proportional
Integrator (PI) controllers, lag/lead controllers, etc, perform much better with dc signals
[12, 44, 58, 145].

It is shown in this section how to implement SRF schemes in order to obtain high per-
formance synchronization algorithms. Firstly, algorithms to synchronize with the funda-
mental components are analyzed, both for single-phase and three-phase systems. Special
emphasis is put on the proposal of using MAFs together with SRF schemes, which is a
major contribution of this thesis [146–148]. Later, the issues of harmonics and negative-
sequences extraction based on SRFs are approached [13, 19, 148–151].

2.4.2.1 Single-phase SRF Synchronization Algorithms

It is considered that the input wave (v) is a single-phase wave with multiple harmonics

v = v1max sin(ω1 t + ϕ1 ) +v3max sin(3ω1 t + ϕ3 ) + v5max sin(5ω1 t + ϕ5 )+

| {z }
v1 (2.36)
+ v7max sin(7ω1 t + ϕ7 ) + ...

v1 is the fundamental component and rotates at ω1 ; from Fourier theory, it can be

expressed as the sum of two orthogonal waves also rotating at ω1 . These two waves
rotating at a frequency ωsrf = ω1 , named d (direct) and q (quadrature), set a single-phase
SRF.

v1 = vd1 · sin(ω1 t + ϕsrf ) +vq1 · cos(ω1 t + ϕsrf ) .

| {z } | {z } (2.37)
d q

It is important to note that (2.37) is valid for any ϕsrf ; Fig. 2.35 depicts v1 in the
SRF.

Fig. 2.36 shows a generic SRF scheme suitable for reconstruction of v1 . The goal of
the SRF algorithm is to measure v1d and v1q , which are dc values, in order to reconstruct
v1 and estimate its phase-angle (θ1 ). The frequency of the implemented SRF (ωsrf ) sets
86 Chapter 2 State-of-the-Art in Grid Synchronization

ω1
q

v q1
v1

vd1 d

Figure 2.35: v1 in the single-phase SRF.

ω1

∫
)
π
ω srf
vd 1
sin(u) X 2·H(z)LPF X
−π

θ srf = ∫ ωsrf t + ϕ srf f (dc,2ω1 ,4ω1 ,...) f (dc)

d v v̂1
Σ
cos(u) X 2·H(z)LPF ) X
Open Loop Synchronous q vq1
Reference Frame generation

Figure 2.36: Single-phase SRF synchronization algorithm: ωsrf = ω1 to decouple v1 .

the component to reconstruct; as said ωsrf = ω1 is set in order to track v1 . The input
wave (v) is multiplied by the d and q, yielding the following results:

1
v × sin(θsrf ) = [v1 cos(ϕ1 − ϕsrf ) + v1 cos(θsrf + θ1 ) + v3 cos(θ3 − θsrf ) +
2| {z } | {z }
vd1 rotating at 2ω1
(2.38)
v3 cos(θ3 − θsrf ) + v5 cos(θ5 − θsrf ) +...],
| {z }
rotating at 4ω1

1
v × cos(θsrf ) = [v1 sin(ϕ1 − ϕsrf ) + v1 sin(θsrf + θ1 ) + v3 sin(θ3 − θsrf ) +
2| {z } | {z }
vq1 rotating at 2ω1
(2.39)
v3 sin(θ3 − θsrf ) + v5 sin(θ5 − θsrf ) +...].
| {z }
rotating at 4ω1

From (2.38) and (2.39), it is expected that vd1 and vq1 , which are dc coefficients, can
2.4 Digital Alternatives to PLLs 87

be obtained with no ripple by means of LPFs. After the filtering stage, v1 is reconstructed
using (2.37). The estimated phase-angle of v1 (θ̂1 ) is:

θ̂1 = θsrf + atan2(v̂q1 , v̂d1 ), (2.40)

| {z }
β (relative angle)

where β is a relative phase-angle in the SRF, which is function of ϕsrf . v1max is estimated
by

q
2 2
v̂1max = v̂d1 + v̂q1 . (2.41)

From Fig. 2.36, it is clear that the dynamics of this system depends only on the LPFs.

2.4.2.2 MAF as LPF for SRF based Schemes

The most important feature of a LPF is its amplitude versus frequency response. In
this particular case, unitary gain for low frequency values and high attenuation at higher
frequencies, specially for harmonics, are wanted. A filter with this frequency response
is relatively easy to design and implement in current digital devices. However, another
important feature of the LPF is its step response. The step response sets the system
response to a transient. In filtering there is a trade-off between the cancellation pattern
and the step-response [152].

MAFs have been implemented in electric power applications because of their excellent
behavior canceling harmonics [153–157]. In this work, it is shown how MAFs also offer a
very optimized performance for grid synchronization [146–148].

The difference equation which defines how the input signal (x) is related with the
output signal (y) of a FIR filter is:

n−1
X
y(k) = b0 · xk + b1 · xk−1 + .... + bn−1 · xk−n−1 = bi · xk−i , (2.42)
i=0

where k represents the current number of sample, bi are the filter coefficients and n − 1
is the filter order, so n must be an integer number. The window length, or time length,
(tw ) of a FIR filter is set by the number of samples, and the sampling time (1/fs ):

1
tw = n · . (2.43)
fs

An important feature of some FIR filters is the linear-phase [158]. A FIR filter is
88 Chapter 2 State-of-the-Art in Grid Synchronization

linear-phase if and only if its coefficients are symmetrical around the center coefficient; in
such a condition the delay through the filter (td ) is function of n:

n 1
td = · . (2.44)
2 fs

As shown in chapter 4, this linear-phase feature is key in the optimization of SRF

based synchronization algorithms.

A particular case of linear-phase FIR filters are MAFs, which implement the average
function over the last n samples. The filter coefficients of the MAFs are found via the
following equation:

1
b(i)i=0,..,n−1 = . (2.45)
n

Eq. (2.42) can be expressed in the Z domain as

1  
Y (z) = H1 (z) · X(z) = · X(z) + X(z)z −1 + .... + X(z)z −(n−1) =
n
1 n−1
X
! (2.46)
· z −i · X(z),
n i=0

H1 (z) being the Z domain transfer function relating the input and output of the filter.
From (2.46) and through the basic properties of the Z transform, a more suitable for
implementation form of H1 (z) is obtained:

1 1 − z −n
H1 (z) = . (2.47)
n 1 − z −1

The magnitude versus frequency response of H1 (z) is mathematically described by the

Fourier transform of a rectangular pulse of amplitude 1/n and duration n samples in the
time domain:

 
sin( πffsn )
M ag [H1 (f )] = M ag  , (2.48)
n · sin( πf
fs
)
where f is defined in the range [0, fs /2] [159].

In grid applications, a cancellation of the even harmonics of f1 = f1n = 50 Hz is

expected (100, 200, ... Hz). From (2.48), the n value canceling all the even harmonics is
2.4 Digital Alternatives to PLLs 89

derived:

fs
n= , (2.49)
2 · f1n
n being an integer number, since it sets the order of a digital filter; hence, fs should be
an integer multiple of 2 · f1n .

Fig. 2.37 shows the features of a MAF with fs = 10 kHz and n = 100 [146].Fig 2.37a
shows that the even harmonics of f1n are canceled and the gain is also very low around
these frequencies. The linear-phase feature can also be noticed. Figs. 2.37c and 2.37d
show the impulse and step responses set by (2.43), in this case 0.01 s.

Figs. 2.37e and 2.37f show a comparative among different kinds of discrete filters ex-
tracted from [148] 8 . As shown, the MAF has a very good cancellation pattern considering
grid applications (odd harmonics) and very good transient response.

Fig. 2.38 shows the very good performance of the SRF and MAF based algorithm,
which is named SRF-MAF1.

These simulations have been carried out considering ωsrf = ω1 = ω1n . In a practical
environment frequency deviations should be taken into account, as shown in section 2.5.

8
Butt(2,10 Hz) means: second order filter of type Butterworth, with cut-off frequency equal to 10 Hz.
90 Chapter 2 State-of-the-Art in Grid Synchronization

1.0002
1

Magnitude (abs)
Magnitude (abs)

0.8
1
0.9998
0.6

0.4 0.9996

0.2 0.9994

0 0.9992
0 0

Phase (deg)
Phase (deg)

-45 -1

-90 -2

-135 -3

-180 -4
0 50 100 150 200 250 300 350 400 450 500 550 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Frequency (Hz) Frequency (Hz)
(a) Frequency response (in Hz). (b) Fig. 2.37a zoomed around dc (in Hz).

0.01
0.009
1
0.008
Amplitude
Amplitude

0.007 0.8

0.006
0.6
0.005

0.004
0.4
0.003

0.002
0.2

0.001

0 00
0 0.002 0.004 0.006 0.008
0.01 0.012
0.014 0.002 0.004 0.006 0.008 0.01 0.012 0.014
Time (sec) Time (sec)
(c) Impulse response (tw = 0.01 s). (d) Step response (tw = 0.01 s).

1 Butt(2,30Hz)

0.9
Amplitude (p.u.)

1 STEP
0.8

0.7
Butt(2,30Hz) 0.8
Butt(2,10Hz)
0.6
Gain

MAF
0.5 Butt(1,10Hz) 0.6
Butt(1,30Hz)
0.4
MAF 0.4
0.3

0.2
0.2
0.1
Butt(2,10Hz)
0 0
0 50 100 150 200 250 300 350 400 0.15 0.2 0.25 0.3 0.35 0.4
Frequency (Hz) Time (s)
(e) Comparative with other filters (Frequency Re- (f) Comparative with other filters (Step/transient
sponse.) response.)

Figure 2.37: Time and frequency responses of H(z)1 implemented at fs = 10 kHz and
n = 100.
2.4 Digital Alternatives to PLLs 91

1
0.8

0.6

0.4

1
0.2

0
-0.2

-0.4

-0.6

-0.8

-1

0 0.01
0.02 0.03 0.04 0.05
0.06

1 1max
0.8 d1
0.6

0.4

0.2

0
-0.2 q1

-0.4

-0.6

-0.8

-1

0 0.01
0.02 0.03 0.04 0.05 0.06

1 1
1

-1

-2

-3

0 0.01
0.02 0.03 0.04 0.05 0.06

Figure 2.38: Simulation results for SRF-MAF1 working open loop (ωsrf = ω1 = ω1n ).
92 Chapter 2 State-of-the-Art in Grid Synchronization

FIR generation 10

Taps 0

Magnitude (dB)
(off-line) X n=100
a N −1
-10

z −1
n=200
-20
Impulse
. -30
.
. . -40
v1
Σ
.
SRF-MAF1 . -50

X
180
a2 z −1
90
a1

Phase (deg)
X

a0 z −1 0

X
-90

v -180
0 100 150 200 250 300 350 400 450 500
HDCT(z) (on-line) 50
Frequency (Hz)

(a) Technique to obtain FIR taps from the impulse (b) SRF-MAF1 with n = 100 and n = 200 (=RDFT
response. of [158]); ωsrf = ω1n .

Figure 2.39: DCT based implementation figures.

2.4.2.3 Equivalent Implementations

It should be commented that, even though the theoretical approach proposed by Mc-
Grath et al in [158], based on a Recursive Discrete Transfer Function (RDFT) and SRF-
MAF1 are different, their implemented schemes could be considered equivalent (Fig. 2.36
vs Fig. 2 of [158]). It can be checked analytically that, even though both schemes seem
to be non-linear, they could be also implemented as linear filters based on the Discrete
Cosine Transform (DCT). In this manner:

2 n−1
X
HDCT (z) = cos(ω1 · k · Ts )z −k . (2.50)
n k=0

HDCT (z) taps can be also obtained through the impulse response of SRF-MAF1 (or
RDFT-based) as depicted in Fig. 2.39a. Fig. 2.39b shows the frequency response of the
DCT filters obtained from SRF-MAF1 with n = 100 and n = 200, which is equivalent
to the RDFT based scheme of [158]. As expected, the double window length provides a
better filtering at the cost of doubling the transient time.

It is a common practice to neglect even harmonics in system voltages/currents, which

is accurate for the most of practical electric circuits [27, 97, 160–165]. Therefore, it could
be said that SRF-MAF with n = 100 seems to be a more efficient alternative. Indeed,
it takes advantage of the potential DCT half-cycle window, which is the main advantage
of the DCT over the DFT [166]. Another advantage of the DCT (half cycle) over DFT
appears when frequency deviations are considered [166]: as can be observed in Fig. 2.39b,
when the input wave is not rotating at nominal frequency DFT (or DCT with n = 200)
phase error doubles the DCT one.
2.4 Digital Alternatives to PLLs 93

f (dc, ω2 , ω4 ,...)
va vˆd+1 vˆa+1
vˆq+1 P + (-1) vˆb+1
H(z)LPF
vb
vc P1+ H(z)LPF 1 vˆc+1
θ srf = ωsrf t + ϕ srf
∫
π
ω srf
−π

ω1
Open Loop Synchronous
Reference Frame Generation

+
Figure 2.40: Three-phase SRF synchronization algorithm; ωsrf = ω1 to decouple vabc 1
.

On the other hand, several works in other signal processing fields (image, speech,
communications,...) proved that the IIR equivalent implementation to a DCT are GIs,
as the ones of section 2.3.2.2 [167–170]. The main advantage of such IIR over FIR filters
appears mainly when considering very large scale integration (VLSI) implementations;
the transient response does not seem an important issue in those works. However, for
electric power applications, the performance of FIR based implementations results in a
much better filtering versus transient response trade-off, as shown in this dissertation.

2.4.2.4 Three-phase SRF Synchronization Algorithm

A three-phase synchronization in the SRF can be obtained by the Park transformation

matrix P+ for the positive-sequence with a random phase-angle; ωsrf = ω1 should be
chosen in order to decouple the fundamental positive-sequence from vabc , so

 
2  sin(ω1 t + ϕsrf ) sin(ω1 t + ϕsrf − 2π ) sin(ω1 t + ϕsrf + 2π )
P+
1 = 3 3  (2.51)
3 cos(ω1 t + ϕsrf ) cos(ω1 t + ϕsrf − ) cos(ω1 t + ϕsrf + )
2π 2π
3 3

is implemented.

Fig. 2.40 shows a suitable scheme for three-phase systems. LPFs are employed to
+ +
cancel ripple from the dc coefficients vd1 and vq1 . The inverse of P+ is used to estimate
+
vabc . When using MAFs, the name SRF-MAF3 refers to the algorithm depicted in Fig.
2.40.
94 Chapter 2 State-of-the-Art in Grid Synchronization

+
The phase-angle and amplitude of vabc are estimated by

+ +
θ̂1 = (ω1n t + ϕsrf ) + atan2(v̂q1 , v̂d1 ), (2.52)
| {z } | {z }
θsrf β (relative angle)

and

q
+ +2 + 2
v̂a1 max
= v̂d1 + v̂q1 . (2.53)

The phase-angle estimation can also me made by using the angle transform B+(−1) to
+
get vα1 and vβ+ so the offset angle β is not needed for θ1 assessment [147].

+ +
θ̂1 = atan2(v̂α1 , v̂β1 ) ∈ [−π, π]. (2.54)

Fig. 2.41 shows simulation results of SRF-MAF3 and n = 100. The good performance,
both in steady and transient states is proved when ideal conditions are considered.

2.4.2.5 SRF-MAFs for Harmonics and Negative-sequences Extraction

Previous approaches showed how to decouple the fundamental component of single-

phase signals and the fundamental positive-sequence of three-phase systems. In the same
way, higher harmonic components can be decoupled using the expression ωsrf = hω1 ,
where h is the order of the selected harmonic. Negative-sequences can be decoupled by
means of P− and its inverse. [12–15, 19, 148–151].

These approaches are specially suitable for power line conditioners. SRF-MAFs have
been successfully implemented as extraction algorithms in [148–151].

SRF-MAFs for selective harmonic extraction have been implemented in the APF con-
troller analyzed in chapter 5. Even though the main part of this chapter is devoted to the
PR current controller block, experimental results prove the excellent performance offered
by the SRF-MAF based extraction, both in steady-state and in the presence of input
transients.
2.4 Digital Alternatives to PLLs 95

1.5

Vc
Va Vb
1
Amplitude (p.u.)

0.5

-0.5

-1

-1.5
0 0.01 0.02 0.03
0.04 0.05 0.06 0.07 0.08

Time (s)
1 ^ +
Va1
0.8
^
V +
b1
0.6 ^ +
Vc1
Amplitude (p.u.)

0.4

0.2

0
-0.2

-0.4

-0.6

-0.8

-1

0 0.01 0.02 0.03

0.04 0.05 0.06 0.07 0.08

Time (s)
3

^
θ1
Phase-angle (rad)

-1

-2

-3

0 0.01 0.02 0.03

0.04 0.05 0.06 0.07 0.08

Time (s)

Figure 2.41: Simulation results for SRF-MAF3 working in open loop (ωsrf = ω1 = ω1n ).
96 Chapter 2 State-of-the-Art in Grid Synchronization

2.4.3 Stochastic Filtering based Synchronization Schemes

Digital devices allow to implement online recursive filters such as Kalman and adaptive
filters [171, 172]. The problem of synchronization with a polluted wave can be posed as
the linear-quadratic-Gaussian control problem [171] or as the Wiener problem of state
estimation [172]. Quite significant proposals in synchronization and harmonic extraction
based on stochastic filtering are present in technical literature [48, 140, 141, 173–181].

The most interesting feature of stochastic filtering based algorithms is their potential
fast transient response, even in the presence of some distorted conditions. For this reason,
they were mainly proposed to work in applications where the transient response is crucial
such as Uninterruptible Power Suppliers (UPSs), Dynamic Voltage Restorers (DVR),
traction systems, relaying protecting and harmonic identification [48, 140, 173–178, 182].

A global sight of synchronization algorithms based on stochastic filtering is detailed

as follows, both for single-phase and three-phase systems.

2.4.3.1 Kalman Filtering

The Kalman filter is an efficient recursive filter that estimates the state of a dynamic
system from a series of noisy measurements. Kalman filters in grid applications have
been employed as extraction algorithm in [183] (off-line) and [145, 173, 181, 182, 184]
(real time). Relating to the scheme of Fig. 2.33, the EKF-TAD of [141] obtains v̂α and
v̂β in a recursive manner, rejecting noise and harmonics in vα and vβ .

It should be highlighted here that the EKF-TAD does not decouple negative-sequences,
while the Kalman based SRF schemes of [145] do it. In general, it cannot be asserted
that stochastic filtering schemes proposed novel architectures. It seems more suitable
to say that they provided a novel approach to digital filtering. In fact, MAFs 9 and
IIRs are also recursive filters. Therefore, it could be said that the manner in which
the synchronization or extraction problem is posed is much more important than the
stochastic filter coefficients.

2.4.3.1.1 Single-phase Example A very simple single-phase example to show Kalman

filtering features is provided in this section. The discrete Kalman model with stationary
reference for identification of the fundamental component (v1 , rotating at ω1 ) of a single-
phase wave (v) can be written as:

x(k + 1) = M · x(k) + W(k) (State equation), (2.55)

9
Even though MAFs are FIR filters, usually they are implemented as IIR structures using (2.47).
2.4 Digital Alternatives to PLLs 97

M being

 
cos( ω̂fs1 ) sin( ω̂fs1 )
M= , (2.56)
− sin( ω̂fs1 ) cos( ω̂fs1 )

and

v1 (k) = H · x(k) + V(k) (M easurement equation), (2.57)

H being

H = [0 1]. (2.58)

W (k) and V (k) are the stochastic processes of the system noise and measurement
noise, respectively. System covariance matrices for W (k) and V (k) are assumed as
E[W (k) W (k)T ] = q and E[W (k) W (k)T ] = r. The error covariance matrix (p) is
ˆ T ]. The second coefficient of
updated in each step; its initial value is p(0) = E[X̂(k) X(k)
x̂(k) is v1 estimation.

The sequential recursive computation steps for v1 estimation are:

1. Compute of Innovation Matrix (I):

I = v(k) − H · x̂(k). (2.59)

2. Compute of Covariance of Innovation (s):

s = H · p(k) · HT + r. (2.60)

3. Compute of Gain Matrix (K):

K = M · p(k) · HT · s(−1) . (2.61)

4. State estimation:
x̂(k + 1) = M · x̂(k) + K · I. (2.62)

5. Update of the error covariance matrix p:

p(k + 1) = M · p(k) · MT + q − M · p(k) · HT · s(−1) · H · p(k) · MT (2.63)

98 Chapter 2 State-of-the-Art in Grid Synchronization

0.8
0.8
1
0.6
0.6

0.4 0.4

0.2 0.2

-0.2
-0.2

-0.4
-0.4
-0.6
-0.6
-0.8

-0.8
-1

-1
0.01 0.02 0.03 0.04 0.06 0.07 0.08 0.09 0.01 0.02 0.03 0.04 0.06 0.07 0.08 0.09

2 2

1
1 1

-1 -1

-2 -2

-3 -3
0.01 0.02 0.03 0.04 0.06 0.07 0.08 0.09 0.01 0.02 0.03 0.04 0.06 0.07 0.08 0.09

(a) Test for an input with noise and 0.5 p.u. sag (b) Test for an input with harmonics.
with −45 deg phase jump.

Figure 2.42: Single-phase Kalman based synchronization algorithm.

The filtering pattern and dynamic response of the Kalman process mainly depend on
the initial values of the covariances. Of course, there is a trade-off between filtering and
transient response. The phase-angle of v1 for each k instant is obtained by

θ1 = atan2(x̂1 , x̂2 ) ∈ [−π, π]. (2.64)

Figs. 2.42a and 2.42b show some significant results for the Kalman filtering synchro-
nization method obtained through the Matlab Script detailed in appendix A.5. These
simulations have been carried out considering ω̂1 = ω1 = ω1n , which is revised in section
2.5.

These results show that the Kalman filter approach is suitable for grid synchroniza-
tion, since it can offer an acceptable filtering versus transient response, specially when
comparing with single-phase PLLs.

The dynamics this Kalman based synchronization algorithm can be set in a different
manner by means of the intial parameters which set the system gain K (see appendix
A.5). Furthermore, by introducing higher harmonic components in the M matrix, the
filtering response can be enhanced at the cost of increasing computational burdens [183].
2.4 Digital Alternatives to PLLs 99

2.4.3.2 Weighted Least-Squares Estimation (WLSE) algorithms

Song et al proposed the implementation of a recursive least squares (RLS) adaptive

filter, named Weighted Least-Squares Estimation (WLSE), adapted to the synchronization
problem [140, 176–178]. The WLSE algorithms approach the problem of synchronization
in an elegant way, so the dynamics of the system mainly depends on the so-called forgetting
factor λ.

Firstly, the problem was posed for three-phase systems with negative-sequence [140,
176]: usually, after a voltage sag, there is a phase-jump and unbalance (negative-sequence)
in the grid voltage [32]; therefore, the energy transfer between the ac mains and a grid-
connected power converter would not be well controlled while phase-angle error is big, so
the retracking should be as fast as possible. On the other hand, unbalance generates a
second harmonic ripple in the positive SRF [26]. Song et al discarded the use of notch
filters because of their “sluggish” transient response; they proposed the WLSE algorithm
instead, which indeed offers a surprising almost instantaneous retracking even under un-
balanced conditions. Song et al proved the WLSE performance in a DVR. Of course, this
approach is suitable for any other approach where SRF controllers are employed.

It is also remarkable the single-phase version of the WLSE algorithm, which also
achieves almost instantaneous retracking [177, 178]. The single-phase WLSE algorithm
was proposed for a traction system in [178]: when a locomotive changes its voltage source,
the new phase-angle should be retracked as soon as possible in order to avoid malfunctions.

2.4.3.2.1 Three-phase WLSE Algorithm In [140, 176], the problem of phase-angle

estimation of a three-phase input signal (vabc ) is posed relating the SRF with the αβ frame
as follows.

Considering only the fundamental component of vabc , it can be expressed in the αβ

frame

T
vαβ = C · vabc , (2.65)

where

 
2 1 − 12 − 21 
C=  √ √ . (2.66)
3 0 3
− 3
2 2

vαβ = [vα vβ ] has information of positive and negative sequences: when there is
unbalance (negative-sequence) vαmax 6= vβmax . In order to decouple negative from positive-
sequence of the fundamental component the angle transformations are employed: the
100 Chapter 2 State-of-the-Art in Grid Synchronization

+(−1) +(−1)
matrix H is composed of the link of B1 and B1 , which rotate at ω̂1 and have a
random offset phase. That is:

   
 cos(ω̂1 · t(k)) sin(ω̂1 · t(k))
 
cos(ω̂1 · t(k)) − sin(ω̂1 · t(k)) 


H=  − sin(ω̂1 · t(k)) cos(ω̂1 · t(k)) sin(ω̂1 · t(k)) cos(ω̂1 · t(k)) 
 , (2.67)
| {z } | {z }
+(−1) −(−1)
B1 B1

so

+−T
vαβ = H · vdq1 , (2.68)
where

h i
+− + +
vdq1 = vd1 vq1 −
vd1 −
vq1 , (2.69)
+ + − −
vd1 , vq1 , vd1 and vq1 being the components of vabc1 in the positive and negative SRFs.

+−
The sequential recursive computation steps for the WLSE estimation of vdq1 , and
hence the phase-angle from vabc1 are detailed below. The dynamic response depends on
the forgetting factor λ and initial error covariance (p0 ):

1. Compute of vαβ with eq. (2.65).

2. Update H with the current t(k).

3. Compute of r Matrix:
r = I + H · p · HT . (2.70)

4. Compute of K gain:
K = p(k) · H · r(−1) . (2.71)

5. Update the error covariance matrix p:

p(k) = λ−1 · p(k − 1) · (1 − K · H). (2.72)

+−
6. Update the estimation of vdq1 :
+− +− +−
v̂dq1 (k) = v̂dq1 (k − 1) + K · (vαβ − H · v̂dq1 (k − 1)). (2.73)

The phase-angle of the positive-sequence is obtained using vd+ , vq+ . Owing to the good
decoupling between positive and negative sequences there is not an undesired second
harmonic in estimated values.
2.4 Digital Alternatives to PLLs 101

Fig. 2.43 shows experimental results of the WLSE algorithm. The Matlab script of
appendix A.6 has been employed. It can be noticed that the transient response is very
good and it does not have ripple under unbalanced conditions. The transient response
could be even improved by the covariance reseting technique proposed in [140, 176].

Fig. 2.44 shows experimental results of adding extra sequences to H when dealing
with harmonics in the input: the simple implementation identifies the 5th harmonic as
noise and the λ should be reduced in order to filter it. The second option is to add positive
and negative sequences of the 5th harmonic to H: positive and negative sequences of 5th
harmonics are decoupled and the λ is kept. This second option could be more resource-
consuming, but in terms of performance seems to be an interesting one. Of course, in
very polluted practical systems more harmonic sequences in H could be implemented.

Until now, the effect of frequency deviations in stochastic filters has not been con-
sidered. However, stochastic filters are very sensitive to them [140, 181]. The frequency
adaptation of stochastic filters, specially the WLSE ones, is revised in the next section.
102 Chapter 2 State-of-the-Art in Grid Synchronization

1 1
Vb Va Vα
Vβ

Amplitude (p.u.)
Amplitude (p.u.)

0 0

Vc
-1 -1
0 0.01 0.02 0.03 0.04
0.05 0.06 0.07 0.08 0.09
0.1 0 0.01 0.02 0.03 0.04
0.05
0.06 0.07 0.08 0.09
0.1
Time (s) Time (s)
1 1.5
^
Vd1+
^
1 Vd1+

^
Amplitude (p.u.)

^ Vq1+
Vq1+
Amplitude (p.u.)

0
^ ^
Vq1-
Vd1-
0 ^
^ Vq1-
Vd1-
-0.2 0.01 0.02 0.03 0.04 0.06 0.07 0.08 0.09
-1 0.01 0.02 0.03 0.04 0.06 0.07 0.08 0.09
0 0.05 0.1 0 0.05 0.1
Time (s) Time (s)
3 3
Phase-angle (rad)

2
Phase-angle (rad)

1
^
θ1 ^
θ1
0 0
-1

-2

-3 -3
0 0.01 0.02 0.03 0.04
0.05 0.06 0.07 0.08 0.09
0.1 0 0.01 0.02 0.03 0.04
0.05 0.06 0.07 0.08 0.09
0.1
Time (s) Time (s)

Figure 2.43: WLSE experimental results for different λs.

2.4 Digital Alternatives to PLLs 103

1 1
Va Vα
Vb

Amplitude (p.u.)
Amplitude (p.u.)

Vβ

0 0

Vc
-1 -1
0 0.01 0.02 0.03 0.05
0.04 0.06 0.07 0.08 0.1
0.09
0 0.01 0.02 0.03 0.04
0.05 0.06 0.07 0.08 0.09
0.1
Time (s) Time (s)
1.5 2
1.5
^
Vd1+ ^
1 Vd1+
1
Amplitude (p.u.)
Amplitude (p.u.)

^
Vq1+ 0.5
0.5 ^
Vq1+
0
0 -0.5

-1
^ ^
Vq1-
-0.5
+(-1) -(-1) Vd1- +(-1) -(-1) +(-1) -(-1)
-1.5
1 1 1 1 5 5
-1 -2 0.01 0.02 0.03 0.04
0 0.01 0.02 0.03 0.04
0.05 0.06 0.07 0.08 0.09
0.1 0 0.05 0.06 0.07 0.08 0.09
0.1
Time (s) Time (s)
3 3
+(-1) -(-1) +(-1) -(-1) +(-1) -(-1)
2
1 1 1 1 5 5
2
Phase-angle (rad)

Phase-angle (rad)

1 1
^
θ1 ^
θ1
0 0

-1 -1

-2 -2

-3 -3
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0.1 0 0.01 0.02 0.03 0.04
0.05 0.06 0.07 0.08 0.09
0.1
Time (s) Time (s)

Figure 2.44: WLSE experimental results. Effect of adding extra sequences to H.

104 Chapter 2 State-of-the-Art in Grid Synchronization

2.5 Frequency Adaptation of Synchronization Algo-

rithms

The study of the linearized model of PLLs provides an accurate assessment of dynam-
ics. However, studies of the dynamics provided for other schemes are not as deep and
accurate as the provided by PLL theory.

Section 2.3 revised the criteria followed to assess the performance of PLLs. Following
the same criteria, any synchronization algorithm should have:

1. high stability margins,

2. a fast transient response,

3. zero steady-state average phase error (θ̄1e = 0), and

2
4. low phase jittering (low θ̄1ej ),

in order to be considered a high performance one.

As shown in section 2.4, many of the alternative synchronization algorithms have a

better performance on the whole than PLLs when working around nominal frequency
(ω1 ≈ ω1n ). However, as also pointed in section 2.4, this assertion should be reconsidered
in practical operation. Faults on the bulk power transmission system, a disconnection
of a large load or of a large source of generation cause frequency deviations (from ω1n ).
These frequency shifts are larger in equipment powered by isolated generators.

Therefore, when assessing the performance of a synchronization algorithm, it should

be taken into account how it performs in the presence of frequency deviations and, in
some cases, detects this change.

International standards define ranges of normal operation for frequency deviations.

• The European Standard UNE-EN 50160, which sets f1n = 50 Hz, establishes limits
in maximum allowable frequency deviations. For interconnected systems the mean
value of f1 , estimated each 10s should not exceed ±1% ([49.5, 50.5] Hz) during the
99.5% of a week, but exceptional deviations between +4% and −6% are permited.
For isolated systems, normal operation of ±2% and exceptional deviations of ±15%
are specified [185].

• The IEC-61000 Standard sets the normal limit operation in ±2% for interconnected
system and extends these limits for isolated systems [144].
2.5 Frequency Adaptation of Synchronization Algorithms 105

• Regarding to distribution generation, and more specifically to the new grid codes
for wind turbines, normal operation of wind turbines allows frequency deviations of
around ±3 Hz [94].

When ω1 6= ω1n , it should be noticed that:

• Digital PLLs using a lag/lead filter, as proposed in [46, 100, 105], would not have
θ̄1e = 0 (type I PLLs) [78, 79].

• Filters employed in the αβ frame as SRF-PLL pre-filters could lead to θ̄1e 6= 0 in

steady-state when ω1 6= ω1n : some of these filters (e.g. D(z)) do not have the
expected frequency response at f1 6= f1n (e.g. phase response of D(z) in Fig. 2.27).

• SRF based schemes, including the stochastic filtering ones, do not have θ̄1e = 0
when ωsrf 6= ω1 because of the filtering delay. This phase error is:

θ1e = (ω1 − ωsrf ) · td (f1 ), (2.74)

where td (f ) is function of the input frequency (phase response), except for linear
phase filters [147, 158]. In the case of stochastic filtering, td is lower as faster is
the system (e.g. in the WLSE algorithms, the lower λ the faster the system, so the
lower td ).

• FIR filters such as numeric delays (e.g. D(z)) and MAFs do not achieve complete
harmonic cancellation, which causes phase jittering. This assertion could be ex-
tended to stochastic filters.

Of course, the issue of frequency adaptation has been posed by most of the authors
proposing novel synchronization algorithms. In sum:

• The use of PLLs to get frequency adaptation has been proposed to work with TAD
schemes [141] and SRF schemes [148, 186]. These hybrid schemes enhance their
whole performance using the frequency estimation from PLLs to adapt internal
signal and/or filter coefficients.
E.g. in [148] ωsrf is estimated by a single-phase PLL. Then, Fig. 2.36 is implemented
with ωsrf = h·ω̂1 for selective harmonic extraction of different harmonic components.
The implementation of an extra frequency loop, instead of a PLL, to estimate ω1
and update ωsrf has also been proposed [140, 143, 144, 151, 176, 177].

• The update of the sampling rate of the digital device was proposed in the RDFT
algorithm of [158] and also in the single-phase PLL of [128]. Zero steady-state phase
error and low jittering is achieved. However, technical problems could appear, since
the frequency response of discrete filters is changing continuously [158].
106 Chapter 2 State-of-the-Art in Grid Synchronization

An alternative phase-offset correction technique is provided in [158]. In the case of

a PLL, a closed loop system, this limitation seems to be very important and the
PLL should have a very low bandwidth, and hence, a slow transient time, as shown
in Fig. 14 of [128].

• GI discrete-time implementation can be updated using ω̂1 as proposed in [119] for

PR current controllers. This has been applied in synchronization algorithms to avoid
phase error and/or jittering [123, 143, 144]. The chapter 5 of this thesis contributes
a novel approach concerning frequency adaptation of PR current controllers, which
can be also applied to synchronization.

• Svensson and Bongiorno have studied the frequency adaptation of DSC filters. The
choice of the device sampling rate is very dependant on the grid frequency [187]. In
[188] a very similar technique to the phase-offset correction of [158] is provided for
DSC implementation.

• Similarly to the case of DSC filters implementation, the choice of the sampling
frequency when implementing the pre-filters of [111] is not a trivial task. In a general
way, it could be concluded that the most difficult issue in the implementation of
pre-filters is to assure that the input wave is delayed an integer number of samples
[188].

2.5.1 Frequency Adaptation in WLSE Schemes

Dealing again with the WLSE synchronization algorithm, Song et al proposed a PI

based controller to correct the estimated frequency. The input of that PI was the phase
offset error between two consecutive samples. It should be noticed that the frequency
adaptation of the WLSE results in a very slow dynamics (≈ 2 s) when compared with
phase-jump and amplitude-change transients (≤ 1 ms) [140, 176, 177]. This problem has
also been reported in [108].

The main reason of this slow response is the very big forgetting factor (λ = 0.9999)
employed in the implementation. From Figs. 2.43, a large λ provides very good filtering
but slow dynamics. Song et al ”fixed” this problem proposing the so-called “covariance
resetting technique”, which resets the covariances to initial values when a grid fault is
detected. In this manner, it is achieved an excellent transient and steady-state (filtering)
responses. However, this approach has a drawback considering frequency deviations:
even though the WLSE in the form proposed by Song et al has a fast transient response
(when fault condition is correctly detected), in steady-state the WLSE is tuned with slow
dynamics. Frequency deviations do not reset the covariances so the system dynamics
remains very slow.

Therefore, it could be said that the “covariance resetting” technique does not seem a
good choice when considering considerable frequency deviations.
2.5 Frequency Adaptation of Synchronization Algorithms 107

Table 2.4
Look-up table: control of n depending on ω̂1

Optimized for n ω̂1 ⇒ n + 1 ω̂1 ⇒ n − 1

ω1 = 2π48.5 rad/s 103 2π48.8 rad/s -
ω1 = 2π49 rad/s 102 2π49.3 rad/s 2π48.7 rad/s
ω1 = 2π49.5 rad/s 101 2π49.8 rad/s 2π49.2 rad/s
ω1 = 2π50 rad/s 100 2π50.3 rad/s 2π49.7 rad/s
ω1 = 2π50.5 rad/s 99 2π50.8 rad/s 2π50.2 rad/s
ω1 = 2π51 rad/s 98 2π51.3 rad/s 2π50.7 rad/s
ω1 = 2π51.5 rad/s 97 - 2π51.2 rad/s

2.5.2 Frequency Adaptive MAFs

This technique has been proposed in [148]. The frequency estimation from a PLL (ω̂1 )
is employed to adjust online the order of the MAFs. Fig. 2.45 shows how the frequency
response for different n. Table 2.4 show the rule of change of the n value: upper and
lower thresholds are different from the center value in order to provide some “hysteresis”
behavior, avoiding continuous changes due to ripple in ω̂1 .
108 Chapter 2 State-of-the-Art in Grid Synchronization

Constant2

1 K Ts
In z-1
50.8-50.7 Product Discrete-Time
Integrator
1
50.2-50.3 Out
2 4 -98
w Z
2*2
frequency Integer Delay
estimation 49.8-49.7 Add
Multiport
Switch -1
Z
49
49.3-49.2 Integer Delay1
Constant1

49.5 -1
Constant3 Z
Integer Delay2
50

Constant4
-1 Multiport
Z Switch1
50.5
Integer Delay3
Constant5

51 -1
Constant6 Z
Integer Delay4

(a) Simulink implementation of MAFs with frequency adaptation.

1
n=98
0.9 n=99
0.05 n=100
0.8 n=101
0.04 (zoom) n=102
0.7
0.03
Gain

0.6
0.02
Gain

0.5
0.01

0.4 0
97 98 99 100 101 102 103
Frequency (Hz)
0.3

0.2

0.1

0
0 50 100 150 200 250 300
Frequency (Hz)
(b) Frequency response of MAFs as function of n (fs = 10 kHz).

Figure 2.45: Frequency adaptive MAFs.

2.5 Frequency Adaptation of Synchronization Algorithms 109

2.5.3 Adaptative Filtering for Frequency Adaptation/Estimation

Adaptive and stochastic filters have been also proposed for frequency estimation, e.g.
[189, 190]. This section analyzes the frequency adaptation technique proposed in [151] to
enhance the performance of SRF-MAFs.

Fig. 2.46 shows the frequency update algorithm proposed for SRF-MAF1. Fig. 2.46a
shows SRF-MAF1 with frequency adaptation. Frequency adaptive MAFs are employed.
The frequency of the SRF is updated online, so ωsrf = ω̂1 . This assures that vq1 and vd1
are dc components in steady-state, and therefore, there is not filtering delay, so θ̄1e = 0
is achieved. Fig. 2.46b represents the adatptive SRF-MAF1 (Fig. 2.46) depicted as an
adaptive filter.

Fig. 2.46c shows the flowchart for frequency estimation. This adaptive filter has been
designed taking into account the singular features of grid-applications. The main parts
of the proposed frequency estimation are detailed below:

• The error signal (e), or cost function, is e = atan2(vd1 , vq1 ) variation with respect
to the previous sample (numerical derivative), as in [140, 177].

• The abs(e) < ǫ condition is placed in order not to estimate the frequency during
transients. Hence, it is assured that the transient response is the same as in open
loop operation.

• The update of ω̂1 is proportional to a gain factor (λ) and e2 ; the e2 factor causes the
system to adapt rapidly when there is a high frequency shift and smoothly when
the shift is small.

• The value of e is saturated in a range [emin , emax ] avoiding very big changes between
consecutive ω̂1 .

• The value of ω̂1 is saturated in the range [ω̂1min , ω̂1max ], since practical frequency
deviations are within a reduced range.

• The optimum λ mainly depends on the input signal features, specially uncorrelated
white noise. The optimum λ depends on the main algorithm, i.e. SRF-MAF1
optimum λ is not optimum for SRF-MAF3.

The approach made for single-phase systems is suitable for the three-phase approach
+ +
using the fundamental positive-sequence set of values (v̂d1 ,v̂q1 ) to calculate e.

Table 2.5 shows the algorithm values used in the implementation. Different gains (λ)
have been used in order to show the noise dependence of this parameter.

Fig. 2.47 shows simulation results for the frequency adaptive SRF-MAF1 implementa-
tion using λ = 5 ∗ 104 s/rad. The test shows the response of the system during the initial
110 Chapter 2 State-of-the-Art in Grid Synchronization

θ srf = ∫ ωsrf t + ϕ srf

)

∫
π vd 1
sin(u) X 2H(z)MAF
−π
f (dc, ω2 , ω4 ,...)
Frequency
d v
ω̂1
) Estimation
vq1 Algorithm
z-1 X
ωsrf (k ) = ω1 (k − 1)
cos(u) 2H(z)MAF
)
q

(a) SRF-MAF1 with frequency adaptation/estimation.

Initialize: x(0)=0; k=0

New sample: k=k+1

Estimate: vd1(k) and vq1(k)

x(k ) = atan2(vˆq1 (k ), vˆd 1 (k ))

Calculate error: e=x(k)-x(k-1)

abs(e)<ε no

yes

Limit e: e min≤ e ≤ emax

atan2(vˆq1 (k ), vˆd 1 (k ))
v
ωˆ1 (k ) = ωˆ1 (k − 1) + λ ⋅ sig (e) ⋅ e 2
update
SRF-MAF1 +
(Adaptive Fiter)
z-1 -

ω̂1
Saturate ω1 estimation

ωˆ1 (k ) min < ωˆ1 (k ) <= ωˆ1 (k ) max

e(k)
Frequency
Estimation
(b) Fig. 2.46a depicted as an adaptive filter. (c) Flowchart of the adaptive filter
based frequency estimation.

Figure 2.46: Adaptive filter scheme of SRF-MAF1 with frequency adaptation/estimation.

2.5 Frequency Adaptation of Synchronization Algorithms 111

Table 2.5
Adaptive algorithm values.

ǫ 10−3 rad/s
emin −0.7 ∗ 10−3 rad/s
emax 0.7 ∗ 10−3 rad/s
ω̂1max 2π51.5 rad/s
ω̂1min 2π48.5 rad/s
λ 1 ∗ 105 s/rad
λ′ 1 ∗ 104 s/rad
λ′′ 5 ∗ 104 s/rad

transient (start-up) of an input signal with ω1 = 49 Hz. Fig. 2.47a shows the fundamental
component reconstruction. As expected, the transient response lasts half a cycle of f1 .
Fig. 2.47b shows the phase error during and after the transient. The same result is shown
for the scheme without frequency adaptation in Fig. 2.47c. These results prove the im-
provement in steady-state, keeping the transient response features. Fig. 2.47d shows the
frequency adaptation; as said, during the initial transient ω̂1 is not updated (abs(e) > ǫ
condition); the update is made only in steady-state.

Regarding to SRF-MAFs, due to selectivity of MAFs, they cancel well harmonics but
not specially well other components. The effect of uncorrelated white noise is studied
below.

The optimum value of λ, under different noisy conditions, is studied for the SRF-
MAF3. This tests show the speed of the algorithm to adapting itself to the input frequency
during the start-up transient. Fig. 2.48a shows a set distorted (unbalance and harmonics)
input waves. The input frequency is 50.8 Hz. Fig. 2.48b shows how the system with higher
λ (λ = 1 ∗ 105 s/rad) is adapted faster and with good accuracy. Fig. 2.48c shows the same
results when white noise of BW 10 kHz and PSD of 4 × 10−8 p.u. is added to Fig. 2.48a
waves. In this case ten times lower λ (λ = 1 ∗ 104 s/rad) is more suitable, since it has a
much lower level of frequency jittering. It is clear the influence of noise in the election of
the λ value, so the variables acquisition is key.
112 Chapter 2 State-of-the-Art in Grid Synchronization

Input Frequency= 51 Hz.

1 0
0.8 -10
0.6 ZOOM
-20
0.4

Phase error (deg)

-30
Amplitude (p.u.)

Phase error (deg)

0.2
-40
0
-50
-0.2 ZERO AVERAGE STEADY-
-60
STATE PHASE ERROR
-0.4
-70
-0.6
-80
-0.8
-90
-1
0 -100
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Time (s) Time (s)

(a) Input wave and reconstructed (f1 = 51 Hz). (b) Phase-error with frequency adaptation.

51.2
0

-10 51

-20 ZOOM
System Frequency (Hz) 50.8
-30
Phase error (deg)

Phase error (deg)

-40 50.6

-50
50.4
-60
NON ZERO AVERAGE STEADY-
-70 STATE PHASE ERROR 50.2

-80
50
-90

-100 49.8
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Time (s) Time (s)

(c) Phase-error without frequency adaptation (≈ (d) Frequency estimation.

+2 deg average steady-state phase error).

Figure 2.47: Adaptive SRF-MAF1 Vs open loop SRF-MAF1 during the start-up and
steady-state.

2.6 Conclusion

This chapter provides an in-depth review of the state-of-the art in grid synchronization.
It is clear that the simplest open loop systems based on zero-cross detection are not
suitable for distorted environments. The first milestone in grid synchronization was the
PLO proposed by Ainsworth. The PLO system is almost immune to harmonics and
frequency deviations in weak grids. The drawback of the PLO is its complex dynamics (α
dependence). Later, the use of CP-PLLs as measurement blocks was proposed in motor
drives applications, mainly to handle with big frequency deviations. The use of CP-PLLs
resulted in simpler controllers. However, this first CP-PLLs have important limitations
due to their PD are based on zero-cross detection. With the suitability of discrete devices,
digital PLLs were proposed. They perform much better than analog CP-PLLs. It should
be specially highlighted the SRF-PLL proposal. In fact, SRF-PLL implementation is being
investigated nowadays; several optimization techniques have been proposed since its first
appearances. Digital single-phase PLLs are also an interesting alternative to CP-PLL in
single-phase systems. On the other hand, by taking advantage of digital implementation
some interesting schemes have been proposed as alternative to PLLs, e.g. SRF-MAFs
and WLSE. However, PLLs provide a big advantage with respect to other systems: its
2.6 Conclusion 113

1.5 51.5

51
0.5
Amplitude (p.u.)

Frequency (Hz)
0 50.5

-0.5

50
-1

-1.5 49.5
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time (s) Time (s)

(a) Unbalanced and harmonic contaminated set of (b) Frequency estimation (green
input waves (blue is va , green is vb and red is vc ). with λ = 1 ∗ 104 s/rad, red with λ = 1 ∗ 105 s/rad).

51.5

51
Frequency (Hz)

50.5

50 Input Frequency
Freq. Meas with K2
Freq. Meas with K1

49.5
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time (s)

(c) Frequency estimation for a noisy input (green (d) dSpace implementation of Adaptive SRF-
with λ = 1 ∗ 104 s/rad, red with λ = 1 ∗ 105 s/rad) MAF3. Frequency estimation using λ = 1 ∗
104 s/rad.

Figure 2.48: Frequency estimation with Adaptive SRF-MAF3.

very good frequency adaptation. Some frequency adaptation techniques for alternative
algorithms are reviewed.
114 Chapter 2 State-of-the-Art in Grid Synchronization
Chapter 3

Dynamics Study of Low-Gain PLLs

Abstract —

This chapter presents a novel approach in the tuning of phase locked loops (PLLs) for power
electronic converters. PLLs are implemented inside a higher level controller to estimate the grid
voltage phase-angle and then control the energy transfer between the power converter and the ac
mains. The tuning of the PLL is not a trivial task, specially when considering power quality
phenomena.

In a general way, PLLs with a low bandwidth (low-gain PLLs) are required when handling
with distorted voltages. It is analytically demonstrated that low-gain PLLs have more trade-offs
than high-gain PLLs (e.g. PLLs for communications): it is not possible to optimize the settling
time for a phase-jump without getting slower the PLL response to frequency variations. Existing
tuning methods do not take into account low-gain features, which may result in non-optimum
designs.

The proposing PLL tuning methodology is based on inspection of frequency-domain diagrams

and, contrary to the other existing tuning methods, takes into account ’low-gain’ dynamics. It is
assured an optimized performance in the presence of any kind of disturbances in the grid.

From practical point of view, the proposed tuning procedure is very intuitive for controller
designers. Some significant design examples and experimental results, obtained from a discrete
implementation (dSpace platform), are provided in order to validate the theoretical approaches.

In the second part of the chapter, the DCO based on a RC oscillator is presented: the digi-
tal model of a sinusoidal oscillator is implemented, instead of explicit trigonometric functions.
This solution reduces the needed digital resources without reducing the performance, which could
be specially useful for DSP-based control of power converters. Experimental results are shown
proving the good performance of the proposal.

115
116 Chapter 3 Dynamics Study of Low-Gain PLLs

3.1 Introduction

As said in chapter 2, distortion in the ac mains should be taken into account when
tuning a PLL: there is a trade-off between filtering and transient response. With the
suitability of an ever increasing potential of discrete devices, several techniques to improve
this trade-off have been proposed, such as the placement of extra filter inside the loop
[106, 109, 112, 126] or before the phase detector (pre-filters) [111, 117, 123], and the
implementation of complex feedback structures [108, 126, 127]. However, it can be checked
that the implementation of filtering techniques always leads to a drastic reduction in the
whole bandwidth and therefore in the transient response speed. More specifically, when
optimization techniques affect to the closed-loop dynamics (filters inside and/or feedback),
the PLL should be tuned with a low bandwidth accordingly, giving rise to a low-gain PLL.

This chapter analyzes in depth the dynamics of low-gain PLLs through the frequency
response of their equivalent linearized model. It is analytically proved that low-gain PLLs
do not respond equally to phase or frequency changes. More specifically, it is not possible
to control the overshoot in the step response (phase-jump) without affecting very much
the dynamics when a frequency variation occurs. This fact is not taken into account in
previous works, which could easily result in a non-optimal tuning.

Taking into account ’low-gain’ features, a tuning method for grid-connected PLLs is
provided. This method is based on inspection of Bode and pole/zero diagrams, so it could
be said that it is very intuitive and useful for the controller designer. Some significant
design examples of SRF-PLL tuning are provided and tested in real time. These tests
contemplate both steady-state distortion and grid transients (faults). The dSpace DS1103
platform, using the discrete solver, and a programmable ac source have been employed
for the tests. Experimental results prove all the theoretical approaches.

In the second part of the chapter, the DCO based on a RC oscillator is presented. The
digital model of a sinusoidal oscillator is implemented, instead of explicit trigonometric
functions. This reduces the needed digital resources without reducing the performance,
which could be specially useful for DSP-based control of power converters.

3.2 Frequency Domain Based Tuning

Fig. 3.1 shows the PLL linear model, which is suitable for studying the dynamics of
PLL for which the linearization approach is accurate, such as SRF-PLLs and multiplied
based single-phase PLLs. The linear model open loop transfer function in the Laplace
3.2 Frequency Domain Based Tuning 117

PD
θ1e ω̂1
ω1t + θ1
+- L(s )

ωˆ1t + θˆ1
DCO
1
H (s ) s

Figure 3.1: Studied linear model.

domain is given by the equation:

1
H(s) = L(s) · . (3.1)
| {z } s
|{z}
Loop F ilter
DCO

3.2.1 Stability Margins and Cut-off Frequency

The stability condition of the linear model through the open loop frequency response
is: |H(s)| < 1 when ∠H(s) ≥ −180 deg. Two quantities are directly related with this
stability criterion: the phase margin (PM) and the gain margin (GM). The crossover
frequency (ωc ) is referred to the frequency at which |H(s)| is 1 (0 dB). In this point the
PM is measured. PM is very useful to specify control system performance because it is
related with the damping ratio of the system (ζ) [191]:

PM
ζ∼
= f or P M < 70 deg . (3.2)
100

The information of PLL settling time (ts ) can be obtained from the H(s) frequency
response, since ωc is approximately equal to the closed loop bandwidth (ω3dB ) and natural
(ωn ) frequencies: ts to within 2% of the final value of a second order closed loop system
can be estimated with

4 4
ts = ≈ (3.3)
ζωn ζωc
[192]. Therefore, by inspection of the Bode diagram of H(s) the PLL transient response
can be estimated accurately. From (3.3) a high ωc results in a fast system. However, a
high ωc leads to a bad steady-state filtering: the choice of ωc is the main trade-off in the
tuning [28, 99]. ts can also be optimized increasing ζ [117].
118 Chapter 3 Dynamics Study of Low-Gain PLLs

Stability margins and ωc can be estimated by different frequency domain methods

such as Nyquist, Nichols and Bode diagrams. The use of Bode diagrams is employed in
this work, since they provide information of both transient and steady-state performance.

3.2.2 Steady-state Distortion and Bandwidth

The needed information for phase-angle tracking is the dc component after the phase
detector. Any other harmonic content in the error signal causes jittering in the estimation
and should be canceled [78]. For the single-phase PLL it is clear that the second harmonic
is an annoying problem even though the input signal is totally clean. The problem of the
second harmonic appears in the SRF-PLL when there is fundamental negative-sequence
in the set of input voltages [99]. The presence of higher harmonics in the inputs could
also result in a loss of performance [28, 99].

The very first action to cancel the presence of non dc components in the error signal was
the reduction of the PLL bandwidth (ωc ), so the closed loop has low gain at non desired
frequencies [28, 99]. The suitability of ever-increasing performance digital devices allows
the implementation of more complex discrete filters and techniques. Some significant
proposals are

• the implementation of different kind of filters inside the loop filter to cancel for
specific harmonics and unbalance effect [106, 109, 112, 126].

• the use of filters before the phase detector (pre-filters) to cancel for specific har-
monics and unbalance effect [111, 117, 123]. It should be noticed that the use of
pre-filters does not affect to PLL tuning, but reduces the whole bandwidth, since
they cause a lag when there is an input transient.

• the implementation of complex feedback structures to cancel for second harmonic

(single-phase PLLs) and unbalance effect (SRF-PLL) [108, 126, 127].

Fig. 3.2 explains by itself the pros and cons of introducing extra filters inside the
loop (L(s)). Indeed, the magnitude versus frequency response allows to cancel for specific
components. However, each filter introduces phase inside the loop, which results in a
reduction of stability margins. Therefore, in order to have an acceptable PM, when extra
filters are placed inside the loop, the bandwidth (ωc ) should be reduced. It could be said
that, in general, the better the filter cancellation pattern of a specific filter, the worse
its phase response. Moreover, it should be also noticed that the effect of adding extra
feedback signals results in a drastic bandwidth reduction [108].
3.2 Frequency Domain Based Tuning 119

Magnitude (abs)
0.8

0.6 Notch (Q=10)

0.4
Notch (Q=1)
Moving Average
0.2

0
90
Phase (deg)

-90

-180 50 100 150 200 250 300 350 400

Frequency (Hz)

Figure 3.2: Frequency response of different filters for canceling harmonics.

3.2.3 Grid Events and Transient Responses

Typical power system electromagnetic phenomena related with transients in the input
of the PLL are voltage variation (sag and swells), phase-angle shifts or phase jumps and
frequency deviations [31].

The bandwidth is set by the gain of the loop filter, but also, by the amplitude of the
input signal. This is indeed a limitation of PLLs: the dynamics depends on the input
amplitude. A practical problem appears when a voltage sag is in the inputs. Usually,
voltage sags have a phase-jump associated [32]. Under such a situation, a low-gain PLL
is even slower than usual, and therefore the re-tracking time could be unacceptable. To
overcome this situation the PD input can be normalized using an amplitude estimation,
but it adds complexity to the PLL as well as to the dynamics assessment [111, 117, 126,
127].

Another limitation of low-gain PLLs, which is analytically proved below, is the fact
that the settling times in the presence of phase-jumps or frequency deviations could be
very different. More specifically, the step response of Fig. 3.1 model is not accurate for
frequency steps when H(s) overdamped. This introduces a new trade-off in the design of
the PLL, since, it is not possible to optimize ts for phase jumps without getting worse the
response to frequency deviations.

The error transfer function relating the phase error θe to the input phase θ1 is:

1
He (s) = . (3.4)
H(s) + 1
120 Chapter 3 Dynamics Study of Low-Gain PLLs

The Laplace transform of the phase error when a phase step or phase jump (∆θ) is
applied at time t = 0 is [78]:

∆θ
Θ∆θ
e (s) = He (s) · . (3.5)
s

Through the final value theorem, a PLL has zero average steady state phase error
after a phase jump if:

θe∆θ (∞) = s→∞

lim s · Θ∆θ
e (s) = 0. (3.6)

In the same way, the Laplace transform of the phase error when a frequency step (∆ω)
is applied at time t = 0 is:

∆ω
Θ∆ω
e (s) = He (s) · . (3.7)
s2
and

lim s · Θ∆ω
θe∆ω (∞) = s→∞ e (s) = 0 (3.8)
assures zero steady-state phase error after a frequency step.

The condition imposed by (3.8) requires of two origin poles in H(s) giving rise to a
type 2 PLL [79]. Applied to the PLLs of Fig. 2.18 if there is no origin poles in L(s)
this PLL does not have zero average steady-state phase error when the input signal is
not rotating at the frequency set by the feedforward constant of the VCO (ωvco ). Typical
digital PLLs implements PIs as loop filter resulting in type 2 PLL [99]. The use of a
lag-lead filter (lag compensator) instead of a PI filter, as proposed e.g. in [46], results in
type 1 PLL which does not assure tracking under frequency deviations.

Eqs. (3.6) and (3.8) do not provide information about how long the transients last.
In order to study the transient responses of the PLL, the inverse Laplace transform is
applied to eqs. (3.6) and (3.8). This analysis is performed assuming that L(s) is a PI
filter (type 2 PLL):

Ki
L(s) = Kp + , (3.9)
s
where Kp and Ki are the proportional and integral constants, respectively. This approach
is valid even though extra discrete filters are present inside the loop, since, from stability
conditions, their implementation require a ’low gain’ and therefore, the dominant roots
are the PI filter ones.
3.2 Frequency Domain Based Tuning 121

Eqs. (3.5) and (3.7), expressed also as function of ζ and ωn , are:

∆θs ∆θs
Θ∆θ
e (s) = = 2 , (3.10)
s2 + Kp s + Ki s + 2ζωn s + ωn2
and

∆ω ∆ω
Θ∆ω
e (s) = = 2 . (3.11)
s2 + K p s + Ki s + 2ζωn s + ωn2

The expression of θe∆θ (t) and θe∆ω (t) in the time domain (being t = 0 the moment of
the transient) depends on the roots of the denominator (p1 and p2 ):

q
p1,2 = −ωn (ζ ± ζ 2 − 1). (3.12)

3.2.3.1 Underdamped Case: ζ < 1

∆θ q
θe∆θ (t) = √ · e−ζωn t
· sin(ω 2
n 1 − ζ t + φ) (3.13)
1 − ζ2
√
where φ = tan−1 ( 1 − ζ/ − ζ), and

∆ω q
θe∆ω (t) = √ · e−ζωn t
· sin(ωn 1 − ζ 2 t). (3.14)
ωn 1 − ζ 2

3.2.3.2 Overdamped Case: ζ > 1

∆θ q √
−ζωn (1− ζ 2 −1)t
θe∆θ (t) = √ 2
[(1 − ζ − 1) · e −
2 ζ2 − 1 | {z }
slow pole
q √
ζ 2 −1)t
− (1 + ζ 2 − 1) · e−ζωn (1+ ]=
| {z } (3.15)
f ast pole
∆θ q q
√ 2 [(1 − ζ 2 − 1) · ep2 t − (1 + ζ 2 − 1) · ep1 t ]
2 ζ −1 | {z } | {z }
slow pole f ast pole
122 Chapter 3 Dynamics Study of Low-Gain PLLs

and

∆ω √
ζ 2 −1)t
θe∆ω (t) = √ 2 [e| −ζωn (1− }−
2ωn ζ − 1 {z
slow pole
√ (3.16)
−ζωn (1+ ζ 2 −1)t ∆ω
− e| {z }] = 2ω √ζ 2 − 1 [ |{z} ep1 t ]
ep2 t − |{z}
n
f ast pole slow pole f ast pole

From eqs. (3.13) and (3.14), it is clear that for the underdamped case the transient
response can be improved both increasing Ki (ωn ) or Kp (ζ). However, for overdamped
situation the situation is not clear at all. Eqs. (3.15) and (3.16) have two terms, one
decaying with ep1 t (fast pole) and another with ep2 t (slow pole).

In the case of θe∆θ (t) (phase-jump) the term associated with the slow pole has associated
a smaller coefficient. If the system is tuned so |p1| >> |p2| the transient response only
depends on the fast pole, so this transient response can be improved increasing ζ. A step
response with almost no overshoot can be achieved. It should be noticed that an increase
in ζ is also equivalent to an increasing in the PM. The problem of this approach arises
when θe∆ω (t) (frequency deviation) is considered (eq. (3.16)). The terms associated to the
slow pole has the same magnitude than the other one. If p2 is very close to the origin,
the transient defined by θe∆ω (t) lasts very much.

Fig. 3.3a shows a diagram of poles/zeros for underdamped and overdamped situations.
As larger is the distance of the dominant poles to the imaginary axis, the faster is the
transient response. Moreover, the higher the imaginary part of a pole, the higher the
overshoot. Therefore, for the case of Θ∆ e (s), it is clear that the transient response to a
θ

phase-jump is better for the overdamped case because the dominant pole is p1 , since p2
is almost canceled by the origin pole (Fig. 3.3b). However, in the case of Θ∆ e (s) the
ω

absence of zero in the overdamped tuning leads to a situation where p2 is the dominant
root, and therefore, the underdamped case has a better frequency step transient (Fig.
3.3c). Therefore, it seems that a tuning around critical damping (Kp2 ≈ 4Ki , so that
ζ ≈ 1) results in the better trade-off for transient responses, even though the overshoot
in θe∆θ (t) is not canceled.

Finally, it should be noticed that a similar approach can be made in the Z-domain
obtaining the equivalent results. A very overdamped system in the Z-domain has its
dominant poles very close to the unity circle [117]. As PLLs are implemented in a discrete
device, an analysis/tuning in the Z-domain seems to be more suitable. However, the
analysis/tuning could be made in the Laplace domain if the sampling frequency (fs ) is
high enough when compared with the dominant poles frequency; the poles/zeros place in
the Laplace domain is much more intuitive to predict the time constants of the system.
3.2 Frequency Domain Based Tuning 123

(a) Poles-zeros diagram for Θe (s)∆θ and Θe (s)∆ω . Blue poles refer to ζ < 1
and green poles to ζ > 1. The origin zero (red) is only present in Θe (s)∆θ .

(b) θe∆θ (t) comparative. (c) θe∆ω (t) comparative.

Figure 3.3: Comparative between underdamped and overdamped cases.

3.2.4 Limitations of Existing Tuning Approaches for Low-gain

PLLs

First of all, it should be noticed that the Evans root-locus diagram does not provide an
accurate tuning of ’low-gain’ PLLs: a type II PLL has a zero close to the origin (dominant
root). Its effect is negligible when the dominant poles are high, but not when they are
also close to the origin. In practice, Evans root locus does not provide a reliable overshoot
information from its ζ for low-gain type II PLLs. It should be noticed that, in control
theory, the Evan root-locus is provided to analyze second order systems without zeros
[191].

In [28, 99, 117] analytical methods to predict the bandwidth and overshoot of PI
based type II PLLs are provided. It can be easily checked that low overshoot tuning
results in very overdamped systems which lead to big transient response in the presence of
frequency steps. Moreover, if extra filters (notch, MAF, DSC) are placed, those methods,
by themselves, are not accurate (P M and ωc change).
124 Chapter 3 Dynamics Study of Low-Gain PLLs

When comparing with previous tuning methods, two important new outcomes can be
extracted from this chapter:

• PM and ωc information from the Bode diagram provides a very reliable information
to tune a PLL when extra harmonics/noise filters are placed inside the loop (low-
gain PLLs).

• The PI filter coefficients tuning should not give rise to very overdamped system,
which damages the transient response in the presence of frequency deviations, which
had been previously reported as an optimum tuning [117].

3.2.5 Design Examples

Some significant examples are shown in order to provide a comparison among tuning
strategies. The sampling frequency (fs ) is 10 kHz and the nominal frequency 50 Hz.

3.2.5.1 High Bandwidth SRF-PLL (HB-PLL)

A SRF-PLL with a very high Bandwidth is designed:

Ki
L(s)HB−P LL = (Kp + ) (3.17)
| {z s }
P I f ilter

Table 3.1
Significant parameters of HB-PLL

ωc ( rad/s) P M (deg) Kp Ki p1 ( rad/s) p2 ( rad/s)

3 7 3
4.56 · 10 49 4000 10 (−2 + i2.45) · 10 p∗1

From Fig. 3.4 and table 3.1, it is expected a PLL with a very fast transient response
both for phase jumps and frequency deviations. Some overshoot is expected. There is
not harmonic/unbalance cancellation.
3.2 Frequency Domain Based Tuning 125

80

60

40

20

-20
-120

-150

-210
2 3 4 5
10 10 10 10

Figure 3.4: Frequency response of H(z) for the HB-PLL (L(s)HB−P LL was discretized
using the ’zoh’ method).

3.2.5.2 Unbalance/Notch SRF-PLL (UN-PLL)

A SRF-PLL for canceling the second harmonic generated by the negative sequence
using a notch filter is designed:

Ki
L(s)U N −P LL = (Kp + ) ·N (s) (3.18)
| {z s }
P I f ilter

where N (s) is a notch filter at 100 Hz with Q = 1.25.

Table 3.2
Significant parameters of UN-PLL

ωc ( rad/s) P M (deg) Kp Ki p1 ( rad/s) p2 ( rad/s)

145 (23 Hz) 70 141.44 5000 −71.9 −69.6

From Fig. 3.5 and table 3.2 it is expected a PLL with an acceptable transient response
both for phase and frequency jumps with some overshoot, since the system is near critical
damping (p1 ≈ p2 ). It has very good unbalance cancellation and an acceptable harmonic
filtering.
126 Chapter 3 Dynamics Study of Low-Gain PLLs

100

50

-50

-100

-150
0

-45

-90

-135

-180

-225
-1 0 1 2 3 4
10 10 10 10 10 10

Figure 3.5: Frequency response of H(z) for the UN-PLL (L(s)U N −P LL was discretized
using the ’zoh’ method).

3.2.5.3 SRF-PLL with Moving Average Filter (MA-PLL)

A SRF-PLL is designed to have a very good harmonic/unbalance cancellation in

steady-state. L(s) is:

Ki
L(s)M A−P LL = (Kp + ) ·M A(s) (3.19)
| {z s }
P I f ilter

where M A(s) is the transfer function of a moving average filter tuned to cancel even
harmonics; it is better expressed in the Z-domain (fs = 10 kHz):

1 1 − z −100
M A(z) = (3.20)
100 1 − z −1

Table 3.3
Significant parameters of MA-PLL

ωc ( rad/s) P M (deg) Kp Ki p1 ( rad/s) p2 ( rad/s)

69 (11 Hz) 69 70 70 −68.98 −1.02

From Fig. 3.6 and table 3.3 it is expected a SRF-PLL with an excellent harmonic/unbalance
cancellation, an acceptable phase-jump transient response (decaying with ep1 t ) with no
overshoot (high PM), but with a bad frequency variation transient (decaying with ep2 t ).
3.2 Frequency Domain Based Tuning 127

-100

-200

-300

-90

-135

-225

-270

-315
0 1 2 3 4
10 10 10 10 10

Figure 3.6: Frequency response of H(z) for the MA-PLL (PI filter was discretized using
the ’zoh’ method).

3.2.6 Experimental Results

The SRF-PLLs of section 3.2.5 have been implemented in a digital device (dSpace
DS1103) using the discrete solver at fs = 10 kHz. The execution times of all SRF-PLLs
were lower than 10 µs. The three-phase input voltages have been generated with a three-
phase programmable ac source (three Chroma 61501 modules). In order to obtain the test
signals the three-phase ac source has been used as an arbitrary power amplifier connected
to a three-phase arbitrary waveform generator based on a DSP card. The input signals
have been acquired through the dSpace I/O interface by means of LEM LV25-P voltage
transducers and adapted to p.u. units. Fig. 3.7 shows the key oscillograms, obtained by
three different tests.

Figs. 3.7a, 3.7b and 3.7c show experimental results when the SRF-PLLs were tested
using a clean signal (no harmonics/noise) having a phase-jump of +45 deg; the input
frequency was 49.5 Hz. Steady state and phase-jump transient responses are shown. From
these results it is clear that the HB-PLL is the best option when there is not distortion,
since it has a very fast re-tracking (see Ch2: vq ≈ phase-error) and good steady-state
phase-angle estimation (see Ch4). As expected from their bandwidth, the other two
systems are slower.

In the frequency step tests (Figs. 3.7d, 3.7e and 3.7f) sudden frequency changes in
the input waves (alternating between 49 Hz and 51 Hz) were programmed in order to
show the response of each system in the presence of frequency deviations. As expected,
the frequency step and phase jump transient times are almost equivalent for not very
overdamped systems (HB-PLL and UN-PLL). For MA-PLL (very overdamped PLL) the
optimization in the phase transient settling time (low overshoot) resulted in a very slow
response in the presence of frequency deviations (see Ch2: vq ≈ phase-error).

In the distorted signal test (Figs. 3.7g, 3.7h and 3.7i), the pre-programed input waves
128 Chapter 3 Dynamics Study of Low-Gain PLLs

(a) Clean signal Test (b) Clean signal Test (c) Clean signal Test
(HB-PLL). (UN-PLL). (MA-PLL).

(d) Frequency Step Test (e) Frequency Step Test (f) Frequency Step Test
(HB-PLL). (UN-PLL). (MA-PLL).

(g) Distorted signal Test (h) Distorted signal Test (i) Distorted signal Test
(HB-PLL). (UN-PLL). (MA-PLL).

Figure 3.7: Experimental results for different SRF-PLLs: Ch1 (black) is the Va input
in p.u./V, Ch4 is the instantaneous phase-angle measurement (100 mV/rad), Ch2 is the
error signal vq in p.u./V (1 p.u. = π/2 deg of phase error), Ch3 is ∆ωo (10 mV/(rad/s)).
3.3 RC Model of Digitally Controlled Oscillator 129

have unbalance (≈ 10% of negative sequence ) and high ammount of harmonics (10%
of 5th , 5% of 5th and 5% of 11th ); the input frequency was 50.5 Hz and a phase-jump
of +45 deg was pre-programmed. The HB-PLL responds very fast to the phase-jump
transient. However, it does not filter either harmonics or unbalance effect because its
high bandwidth, so that the quality of its phase-angle estimation is very poor due to the
jittering (see Ch4 of Fig. 3.7g). Excellent steady-state results are obtained with MA-PLL
thanks to the moving average filter, which cancels all the even harmonics caused by the
presence of odd harmonics and unbalance in the inputs (see Ch3 of Fig. 3.7i: ∆ωo ).
However, its phase-jump transient response is slow, since it has a low bandwidth. UN-
PLL has a very good unbalance cancelation and acceptable harmonic filtering (see Ch3
of Fig. 3.7h: ∆ωo ), and an average phase-jump transient response.

From these results, it could be stated that a previous knowledge of the grid level
distortion is very recommended when tuning a PLL. It should be noticed that, very
overdamped (e.g. MA-PLL) and very high bandwidth (e.g. HB-PLL) PLLs present some
important drawbacks when dealing with power quality phenomena, which may lead to
a poor energy exchange control (between the grid and the power electronic converter).
Therefore, an ‘average’ system (no overdamped with a medium bandwidth), such as UN-
PLL, could be a good choice for most of the applications.

3.3 RC Model of Digitally Controlled Oscillator

This section provides a novel technique of trigonometric functions and SRF matrices
implementation, which has been presented in [112]. With regard to the PLL structure,
these blocks are part of a DCO.

The implementation of trigonometric functions is not a trivial task, specially in low-

cost fixed point DSP; technical reports provide some techniques such as the use of look-up
tables, linear interpolation methods and IIR filters [193–195].

This work proposes an efficient IIR based algorithm based on the model of a RC
electronic oscillator; its block diagram is depicted in Fig. 3.8a. Two unitary orthogonal
waves are obtained. This implementation reminds to GI based on two integrators [119],
but without input.

Fig.3.8b shows the closed loop poles of the digital oscillator for ω̂1 = ω1n ; of course, in
real time operation, the oscillation frequency ω̂1 is updated online. Following Barkhausen
criteria, this system oscillates at ω̂1 : from control theory, this system tends to instability
since its poles are on the imaginary axis; however, as the integrators are saturated, the
signal amplitude can be controlled [196]. Also from oscillators theory, the output of one
of the integrators must be non-zero in order to start the oscillation. e.g. sin(θ̂1 ) = 0,
cos(θ̂1 ) = 1 can be chosen to quickly achieve the oscillation steady-state.
130 Chapter 3 Dynamics Study of Low-Gain PLLs

cos(ωot + θ o )

ωo ∫
Pole-Zero Map
0.99 400

sin(ωot + θ o )
Pole : 0 + 314i
300
− 0.99
200

Imaginary Axis
∫
100

ωo
0.99
0
-
− 0.99
-100

ωo t + θ o
Reset

∫
-200
π

ωo
-300 Pole : 0 - 314i

−π
-400
-1 -0.5 0 0.5 1
Real Axis
(a) Block diagram of the digital oscillator. (b) Poles/Zeros map of the oscillator (in
rad/s, continuous model).

Figure 3.8: RC Oscillator features.

The estimated phase (θ̂1 ) is calculated through the numeric integration of ω̂1 between
[−π, π]. When sin(θ̂1 ) crosses zero in the falling edge θ̂1 is reset to −π. Another option is
to use the range [0, 2π] and reset on the rising edge.

As seen, this diagram is very low resource-consuming. A small drawback of the oscil-
lator is that the generated waves sin(θ̂n ) and cos(θ̂1 ) are not pure sinusoidal waveforms,
due to the non-linear behavior of the saturation in the integrators. However, the total
harmonic distortion (THD) of these waves is negligible (0.70% when implemented at a
sampling frequency (fs ) of 10 kHz).The limits of the integrator should be chosen accord-
ingly in order to assure that the amplitude of the fundamental component is 1. E.g. the
integrators are saturated at ±0.99, not at ±1.

As said, two orthogonal waves are the outputs of the oscillator. With regard to single-
phase the in-phase and feedback signals are directly obtained. With regard to SRF-PLLs,
the assessment of Park transformations is immediate by:

√
2π 3
sin (θ̂1 ± ) = −0.5 · sin (θ̂1 ) ∓ · cos(θ̂1 ). (3.21)
3 2

Fig. 3.9 shows the P1+ matrix calculation from the oscillator signals (Fig. 3.8a) and
(3.21) identities.
3.3 RC Model of Digitally Controlled Oscillator 131

2/3 sin(2π/3)
cos(θ^1)

-2/3

2/3 cos(2π/3)
2/3 sin(2π/3)
sin(θ^1)

-2/3

2/3 cos(2π/3)

Figure 3.9: P1+ generation from the oscillator signals.

3.3.1 Experimental Results

3.3.1.1 Simulation Results

Fig. A.7 of the Appendix shows a Matlab script which simulates the single-phase PLL
start-up tracking. A notch filter is placed to cancel the second harmonic. This script
also proves the simplicity and significance of the oscillator implementation: trigonometric
functions are not explicitly present. It is important to mention that the output of the
LF should be saturated (∆ωmax/min = ±128 rad/s). This is implicit in a DSP fixed point
implementation.

3.3.1.2 Single-phase PLL Implementation

The single-phase PLL has been implemented in the Technosoft MSK2407 board con-
taining a TMS320LF2407 fixed point DSP of Texas Instruments; the word length of this
DSP is 16 bits. In the script of Fig. A.7 of appendix it is shown the representation for
each variable in Q format. It is important to comment that there is a trade-off in the
choice of the ’ylf’ pipeline format. A smaller pipeline (e.g. Q9) gives rise to a system
slower since it saturates easily. However, a bigger pipeline (e.g. Q7) would give rise to a
noticeable lose of performance due to the truncation after the LF multipliers. Q8 resulted
to be the optimum choice.

Fig. 3.11a shows the steady state phase measurement of the single-phase PLL when
a clean sinusoidal of 51 Hz is at the input. Fig. 3.11b shows the phase tracking when
the real grid signal is at the input. Fig. 3.11c shows the excellent transient response of
the PLL: after a strong fault such as a voltage sag from 1 p.u. to 0.5 p.u. and +45 deg
phase-jump the PLL re-tracks in less than a cycle. These results confirm the expected
132 Chapter 3 Dynamics Study of Low-Gain PLLs

Amplitude (p.u.), Phase-angle (rad/s)

2

-1

-2
Input signal
Mysin
-3 Phase-angle
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Time (s)

(a) Single-phase PLL start-up simulation (ideal input signal).

3
Amplitude (p.u.), Phase-angle (rad/s)

-1

-2 Input signal
Mysin
-3 Phase-angle

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Time (s)

(b) Single-phase PLL start-up simulation (highly harmonic con-

taminated input signal).

Figure 3.10: Simulation results for single-phase PLL (ω1 = 2π50.3 rad/s).

performance of the system.

3.3.1.3 SRF-PLL

The SRF-PLL have been implemented in the floating point PowerPC inside the dSpace
DS1103 prototyping platform. This platform contains a floating point micro PowerPC
and an I/O interface.

Fig. 3.11d shows the steady-state phase measurement for a highly unbalanced (both in
amplitude and phase) voltage system at the inputs of SRF-PLL. As expected, the second
harmonic is canceled. Fig. 3.11e shows the good tracking of the SRF-PLL when there is
superimposed white noise of 10 kHz BW and a power spectral density (PSD) of 10−6 p.u.
to the previous test wave. Fig. 3.11f shows the excellent performance of the SRF-PLL
tracking an unbalanced voltage system with notches.
3.3 RC Model of Digitally Controlled Oscillator 133

Zoom

(a) Phase measurement for sinusoidal input with (b) Phase measurement for real grid voltage.
input frequency of 51 Hz.

Trigger

(c) PLL response under a strong fault. (d) SRF-PLL steady state phase measurement
for a set of unbalanced voltages as input.
(fin = 50.5 Hz)

(e) SRF-PLL steady state phase measurement (f) SRF-PLL steady state phase measurement
of a highly noisy signal (fIN = 49.5 Hz). of a signal with deep notches (fIN = 50.8 Hz).

Figure 3.11: Experimental results.

134 Chapter 3 Dynamics Study of Low-Gain PLLs

3.4 Conclusions

This chapter presents an in-depth study of the dynamics of digital PLLs for grid-
connected power electronic converters.

Section 3.2 reviews some PLL theory important concepts and proposes a tuning ap-
proach based on inspection of Bode and pole/zero diagrams. From a practical point of
view it could be said that the proposed tuning method is very intuitive. Moreover, two
new significant outcomes are contributed in this chapter:

1. PM and ωc information from the Bode diagram provides a very reliable information
to tune a PLL when extra harmonics/noise filters are placed inside the loop (low-gain
PLLs). This information is much more reliable that the obtained from analytical
methods having into account only the PI filter roots.

2. It is analytically proved that low-gain PLLs have a trade-off between the responses
to phase-jumps and frequency deviations. A very overdamped PLL has an opti-
mized phase-jump response since the overshoot is minimized, but it has a very slow
response in the presence of frequency deviations. Therefore, the PI filter coeffi-
cients should not give rise to a very overdamped system, which had been previously
reported as an optimum tuning.

Some significant design examples of SRF-PLL tuning are provided in order to check in
real time the theoretical approaches. Experimental results prove the accuracy and validity
of the analysis.

The second part of the chapter shows the good performance of the RC-oscillator based
DCO. The low ratio grid over sampling frequency allows to implement an RC-oscillator
algorithm to work as DCO. In this way, a simple and accurate method of trigonometric
functions implementation is achieved. This method is specially suitable for low-cost fixed
point DSPs. Experimental results prove that, in practice, the proposed DCO performs as
an ideal oscillator, both for single-phase and three-phase PLLs.
Chapter 4

Predictive based SRF-MAF

Synchronization Algorithms

Abstract —

Chapter 4 presents a novel open-loop synchronization system, designed from SRF-MAF

schemes. Accurate measurements of phase, frequency and amplitude are carried out in real-
time.

Previous works establish that the fundamental positive-sequence vector of a set of utility volt-
age/current vectors can be decoupled using Park’s transformation and low pass filters. However,
the filtering process introduces delays that impair the system performance. More specifically,
when the input signal frequency is shifted above the nominal, a non zero average steady-state
phase error appears in the measurements (estimations).

To overcome such limitations, a suitable combination of predictive and moving average FIR
filters is proposed to achieve a robust synchronization system for all input frequencies inside
typical normal operation ranges. MAFs are linear phase FIR filters which have a constant time
delay at low frequencies. A characteristic which is exploited to good effect to design a predictive
FIR filter which compensates such time delays, enabling zero steady-state phase errors for shifted
input frequencies.

In summary, the main attributes of the new system are its good frequency adaptation, good
filtering/transient response trade-off, the fact that its dynamics is independent of the input vector
amplitude and no trade-off between frequency deviation and phase-jump transient responses.

Comprehensive experimental results validate the theoretical approach and the high perfor-
mance of the proposed synchronization algorithm.

135
136 Chapter 4 Predictive based SRF-MAF Synchronization Algorithms

As shown, synchronization is an issue of paramount importance in the control of

grid-connected power electronics equipment. Most control algorithms use some form of
systems voltage/current information, such as the phase, amplitude and frequency of the
fundamental component. It is expected that, the speedier and the more accurate these
measurements are, the better the generation of reference signals will be, and therefore the
more efficient the control actions.

Arguably, phase locked loops (PLLs) are the most widespread synchronization algo-
rithms employed in grid-connected converters, as shown in previous chapters.

PLLs have been employed successfully in systems where an accurate phase measure-
ment is required in high and medium voltage grid connected converters [64, 83, 197, 198],
and in systems with self adapting frequency variations characteristics [92, 145, 199–201].
However, the dynamics of PLLs presents some drawbacks, as analyzed in chapter 3: low-
gain features and input amplitude dependent dynamics.

As reviewed in Chapter 2, some alternative algorithms perform better than PLLs

under certain conditions, particularly when the input frequency is close to the nominal.
However, as the input frequency shifts, their performance tend to worsen. Frequency
control algorithms have been added to open loop schemes in order to make them frequency
adaptive. However, this is at the expense of increasing the complexity of their dynamics
and implementation by a considerable margin. So, in practical implementations, the PLL
continues to be the most employed synchronization algorithm, mainly due to its simplicity,
frequency adaptation features and versatility.

SRF-MAF based synchronization algorithms, have amplitude independent dynamics

and good filtering/transient-response. SRF-MAFs take advantage of both frequency and
transient (step) responses of MAFs. However, similarly to other open loop systems, they
lack frequency of adaptation features if a closed loop frequency loop is not added, which
adds complexity to the scheme.

In order to achieve frequency adaptation, predictive filters for compensating the delay
through the MAFs are proposed in this chapter: MAFs have linear-phase which implies
that the time of delay is constant and known for low frequencies [152]. The oscillation
frequency of Park variables is set by the input frequency deviation from the nominal. The
time of delay through the MAFs is always much smaller than the oscillation period of the
Park variables. Therefore, the trajectory in the time domain of the Park variables can be
approximated to a straight line, where only the time interval between the measured value
and the actual value are considered [202–208]. Using this approach, the transfer function
of a predictive filter which predicts the current value of Park variables is obtained.

In summary, the use of moving average and predictive filters enables a purely open loop
system with frequency adaptation, amplitude independent dynamics and good filtering
versus transient-response trade-off. The dynamics of this open loop system is mainly
stated by the frequency and transient responses at the filtering stage. Moreover, more
than one moving average filter can be used in order to improve the cancellation pattern,
4.1 Calculation of Predictive Filters 137

but this at the expense of making the transient response slower.

Two design examples are presented. Their high performance is shown in the experi-
mental results section. Simulation and real time implementation results are provided and
amply discussed.

4.1 Calculation of Predictive Filters

As shown in section 2.5, the most problematic drawback of SRF-MAF3 and SRF-
MAF1 (Fig. 2.40 and 2.36) is that ω1 is unknown. Focusing in the design of SRF-MAF3,
+ +
when ω1 6= ω1n , vd1 and vq1 are low frequency components rotating at ωd = ω1 − ω̂1n . In
+ +
such a situation vd1 and vq1 are obtained with a time delay (td ) set by the phase versus
frequency response of the low pass filter employed. If it is not compensated, this time
delay would cause a constant average steady-state phase error (θ1e ) proportional to td and
ωd :

θ1e = θ1 − θ̂1 = ωd · td . (4.1)

The only way to know td in open loop, without knowing ωd is using a linear-phase FIR
filters. Linear-phase refers to the condition where the phase response of the filter is a linear
function of frequency, excluding phase wraps at ±180 deg. This results in td through the
filter being the same at all frequencies [152].As shown in section 2.4.2.2, MAF filters have
linear-phase. in the Z-domain a MAF is 1 :

1 1 − z −N1
H1 (z) = . (4.2)
N1 1 − z −1

The time delay through this filter is

N1 1
td = · . (4.3)
2 fs

A predictive filter compensating for MAF lag can be calculated as follows. From (2.44)
td is calculated off-line. While td << ω2πd , it is correct to say that in any td interval both
+ +
vd1 and vq1 trajectories in the time domain fit very well to a straight line. Therefore, it is
possible to predict future samples from current samples into this trajectory [202–208].

y2 (k + 1) being the current sample of a straight line trajectory, it can be expressed as:

1
y2 (k + 1) = y2 (k) + m · (4.4)
fs
1
In this chapter, it is used N1 instead of n.
138 Chapter 4 Predictive based SRF-MAF Synchronization Algorithms

m being the slope. In the same way:

1
y2 (k + 2) = y2 (k + 1) + m · . (4.5)
fs

From eqs. (4.4) and (4.5):

y2 (k + 2) = 2 · y2 (k + 1) − y2 (k). (4.6)
N2 being the number of samples of delay through a moving average FIR filter, it is set by
td (eq. 4.3) and fs :

N1
N2 = td · fs = . (4.7)
2

Knowing N2 from the moving average FIR filter parameters, through eq. (4.6) the
future N2 sample can be predicted from the current and past samples as follows:

y2 (k + N2 ) = 2 · y2 (k + N2 − 1) − y2 (k + N2 − 2) =
= 3 · y2 (k + N2 − 2) − 2 · y2 (k + N2 − 3) = (4.8)
= (N2 + 1) · y2 (k) − N2 · y2 (k − 1).

Eq. (4.8) can be expressed as the Z domain transfer function:

H2 (z) = (N2 + 1) − N2 z −1 . (4.9)

The predictive filter defined by eq. (4.9) compensates the delay through the moving
average FIR filter: for low frequencies, the phase versus frequency response of H2 (z) has
a linear positive phase which cancels the negative linear phase of H1 (z).

By inspecting the phase versus frequency response of the whole filtering stage (H1 (z) ·
H2 (z)) , H2 (z) can be optimized accordingly in order not to have almost any phase delay
without introducing amplitude error. H2 (z) is redefined as:

(N2 + 1) − (N2 − ǫ)z −1

H2 (z) = , (4.10)
1+ǫ
where ǫ is an optimization parameter (ǫ << 1).
4.2 Design Examples 139

vˆd+1
vˆd+1
Vabc H1(z) H1(z)’ H2(z)
H1(z) H2(z)
vˆq+1
Vabc
vˆq+1
[P1+ ]
[P1+ ]
H1(z) H1(z)’ H2(z)
H1(z) H2(z)

Filtering Block Filtering Block

(a) Filtering stage of S1. (b) Filtering stage of S2.

Figure 4.1: Filtering blocks of the design examples.

4.2 Design Examples

Following the theoretical approach of the previous section, two design examples are
contributed.

4.2.1 Design Example 1

Fig. 4.1a shows the filtering block which employs two pairs of linked moving average
and predictive filters. The values of the parameters are summarized in table 4.1.

Table 4.1
Design Example 1 Values.

fs N1 N2 ǫ
10 kHz 100 50 0.0095

The whole system implementing this filtering block is named S1. Fig. (4.2a) shows
H1 (z)·H2 (z) frequency response. The cancellation pattern is optimized for even harmonics
after P+1 caused by odd harmonics in the input signal vabc [146]. Fig. (4.2b) proves
the feasibility of the predictive filter approach: the phase delay is almost zero for low
frequencies. This error is even minimized by the optimization of the ǫ value (Fig. (4.2c)).
Figs. 4.2d and 4.2e show that the transient response is kept in half a fundamental cycle.

As said, in this approach only odd harmonics in vabc are considered since non-linear
electrical loads causing high even harmonic disturbances (in vabc ) are not usual and their
use must be precluded in ac networks [163–165]. To deal with the presence of even
harmonics in vabc , N1 = 200 and N2 = 100 could be considered.
140 Chapter 4 Predictive based SRF-MAF Synchronization Algorithms

1.5 1.002
Magnitude (abs)

Magnitude (abs)
1.0015
1
1.001

1.0005
0.5

0 0.9995
90 0.04
Phase (deg)

Phase (deg)
45 0.03

0 0.02

-45 0.01

-90
0
-135 -0.01
0 50
100 150 200 250 300 350 400 450 500
550 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Frequency (Hz) Frequency (Hz)
(a) Frequency response (in Hz). (b) Fig. 4.2a zoomed around low frequency with
ǫ = 0.

1.0115
0.5 1 sample
Magnitude (abs)

1.011 0.4

0.3
1.0105
0.2
Amplitude

1.01 0.1
100 samples of 0.01
1.0095 0
0.001 -0.1
Phase (deg)

0 -0.2
-0.001
-0.3
-0.002
-0.4
-0.003
-0.5 1 sample
-0.004
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014
Frequency (Hz) Time (sec)
(c) Fig. 4.2a zoomed around low frequency with (d) Impulse response (tw = 0.01 s). .
ǫ = 0.0095.

1.8

1.6

1.4
Amplitude

1.2

0.8

0.6

0.40 0.002 0.004 0.006 0.008 0.01 0.012 0.014

Time (sec)
(e) Step response (tw = 0.01 s).

Figure 4.2: Design example 1: time and frequency responses of H1 (z) · H2 (z).
4.3 Simulation Results 141

4.2.2 Design Example 2

Even though S1 is a very good practical solution dealing with unbalance and harmon-
ics, the presence of other factors such as interharmonics, noise, notching (also noise), dc
offset ([132]) could degrade the performance of the measurements. In fact, the limitations
of the proposed system are mainly set by its cancellation pattern. In order to improve the
′
cancellation pattern two moving average filters H1 (z) and H1 (z) can be linked, as pro-
posed in this section (Fig. 4.1b). The predictive filter should be recalculated according
to eq. (4.7), since the time of delay through the two filters is doubled. The whole system
implementing this filtering block is named S2, and its values are summarized in table 4.2:

Table 4.2
Design Example 2 Values.
′ ′
fs N1 N1 (Eq. (4.2) for H1 ) N2 ǫ
10 kHz 100 100 100 0.009

Fig. 4.3 shows the frequency and step responses of design example 2. The cancellation
pattern is improved at cost of doubling the transient time. As in other systems such as
PLLs, there is a trade-off between filtering and transient response.

4.3 Simulation Results

The S1 system proposed in the previous section has been simulated in order to test
its performance. Simulation results have been obtained through Matlab/Simulink, using
the fixed step discrete time solver at fs = 10 kHz.

The most important feature of time domain simulation with respect to real time
implementation is the possibility of obtaining curves of phase error, frequency error and
amplitude error in the time domain. These errors are defined as the difference between
the actual and estimated values.

The amplitudes of the input waves are in p.u. units, so they can represent both
voltages or currents. The magnitude displayed to indicate the amount of unbalance in
−
vabc is the negative sequence vector magnitude (va1 max
).

− +
Fig. 4.4 shows the phase error for an unbalanced (va1 max
= 0.1 · va1 max
) input wave
rotating at 51 Hz (ω1 = 2π51 rad/s). A −45 deg phase jump has been programmed to
show both steady-state and transient responses. The result is also displayed for the system
without predictive filters. As expected, the system without predictive filters has the non-
zero average steady-state phase error set by eq. (4.1). With predictive filter, steady-state
zero error is achieved. As also expected, the transient lasts 0.01 s.
142 Chapter 4 Predictive based SRF-MAF Synchronization Algorithms

2 1.6
Magnitude (abs)

1.5 1.4

1 1.2

Amplitude
0.5 1

0 0.8
360
Phase (deg)

0 0.6
-360
-720
0.4
-1080
-1440 0.2
-1800

-2160 00
0 50
100 150 200 250 300 350 400 450 500 550 0.005 0.01 0.015 0.02 0.025
Frequency (Hz) Time (sec)
(a) Frequency response (in Hz). (b) Step response (tw = 0.02 s).
Phase (deg) Magnitude (abs)

1.01
1.008

1.006

1.004

1.002

1
0.998
0.01

0
-0.01

-0.02

-0.03

-0.04
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

(c) Fig. 4.3a zoomed around low frequency with

ǫ = 0.009.
′
Figure 4.3: Design example 2: Frequency and Step responses of H1 (z) · H1 (z) · H2 (z).
4.3 Simulation Results 143

Va
1

Vc
Amplitude (p.u.)

Vb

-1

0.99 1 1.01 1.02 1.03 1.04 1.05

Time (s)
− +
(a) Phase jump transient in unbalanced (va1 max
= 0.1 · va1 max
) set of input waves
rotating at ω1 = 2π · 51 rad/s.

20 10

0
10
Average steady-state phase
Phase error (deg)

Phase error (deg)

-10 error equal to 1.8 deg.

0

Average steady-state -20

-10 phase error equal to 0
-30

-20 -40

-30 -50
0.99 1 1.01 1.02 1.03 1.04 1.05 0.99 1 1.01 1.02 1.03 1.04 1.05
Time (s) Time (s)

(b) Phase error: transient and steady-state (c) Phase error: transient and steady-state
with predictive filters (H2 (z)). Zero average without predictive filters (H2 (z)). There is no
phase error is achieved. zero average phase error.

Figure 4.4: S1: Test to show predictive filters action.

144 Chapter 4 Predictive based SRF-MAF Synchronization Algorithms

Fig. 4.5 shows the phase error when the input wave of the previous test has a big
frequency step in 0.2 s: from 48 Hz to 52 Hz. Before and after the transient of duration
0.01 s, the average phase and frequency error are zero. As shown, frequency deviation
and phase jump responses are equivalent, on the contrary to overdamped PLLs.
Input frequency (Hz)

53

52
51
50
49

48
47
0.17 0.18 0.19 0.2 0.21 0.22 0.23
Time (s)
2
Phase error (deg)

1.5 Average steady- Average steady-

state phase error state phase error
1
equal to 0 equal to 0
0.5

0
-0.5
0.17 0.18 0.19 0.2 0.21 0.22 0.23
Time (s)
Frequency error (Hz)

4
Average steady- Average steady-
state phase error state phase error
2 equal to 0 equal to 0

-2
0.17 0.18 0.19 0.2 0.21 0.22 0.23
Time (s)

− +
Figure 4.5: S1 response to a big frequency step at 0.2 s. Unbalanced (va1 max
= 0.1 · va1 max
) input wave. Frequency step (up), phase error (center) and frequency error (down). Zero
steady state error and transient duration of 0.01 s are achieved.

− +
Fig. 4.6 shows the results for an unbalanced (va1 max
= 0.1 · va1 max
) and harmonic
th th th
contaminated (≈ 10% 5 , 7% 7 , 4% 11 ) input wave rotating at 49.5 Hz. In 1 s a
deep magnitude change (sag) from 1 p.u. to 0.2 p.u. with a +45 deg phase jump has been
programmed. Fig. 4.6b shows how steady-state phase error is achieved; the transient
+ + + +
lasts 0.01 s. Fig. 4.6c shows v̂a1 max
, amplitude error, v̂d1 and v̂q1 ; as expected v̂d1 and
+
v̂q1 rotate at 0.5 Hz. The average amplitude error is non zero only during the transient
+ +
(0.01 s). Fig. 4.6d shows the Clarke variables (v̂α1 and v̂β1 ) employed in the calculation
of θ1 .

4.4 Experimental Results

The proposed algorithm has been implemented in the dSpace platform DS1103 at
(fs = 10 kHz) and tested in the laboratory. The execution time of S1 and S2 was 11.2 µs
and 15.8 µs respectively. The three phase voltage system has been generated with a three
phase programmable AC source (three Chroma 61501 modules). In order to obtain the
4.4 Experimental Results 145

40
1.5

Va Vb Vc
20
1

0
0.5
-20

0
-40

-0.5
-60

-1
-80

-1.5
0.95 0.96 0.97 0.98 0.99 1 1.01 1.02 1.03 1.04 1.05 -100
0.95 0.96 0.97 0.98 0.99 1 1.01 1.02 1.03 1.04 1.05

(a) S1: set of distorted input waves in time domain. (b) Phase error: the transient lasts 0.01 s and zero
− +
Unbalance (va1 max
= 0.1 · va1 max
) and harmonics average steady-state error.
th th th
(≈ 10%5 , 7%7 , 4%11 ).

1.2
1
1 |Vp|
0.8
0.8
0.6
Vq
0.6
0.4
0.4 0.2
0.2 0

0 -0.2
-0.4
-0.2
Vd -0.6
-0.4
-0.8
-0.6
-1
-0.8
0.95 0.96 0.97 0.98 0.99 1 1.01 1.02 1.03 1.04 1.05 0.95 0.96 0.97 0.98 0.99 1 1.01 1.02 1.03 1.04 1.05

+ + + + +
(c) v̂a1 max
, v̂d1 and v̂q1 measurements. The tran- (d) vα1 and vβ1 estimations. They are not dis-
sient lasts 0.01 s and zero average steady-state er- torted.
ror.

Figure 4.6: S1 tested under a distorted set of input waves rotating at 49.5 Hz. At 1 s, a
1 p.u. to 0.2 p.u. sag with +45 deg phase jump has been programmed.
146 Chapter 4 Predictive based SRF-MAF Synchronization Algorithms

Vb Vc Va, Vα Va

Vα

Va–Vα
Va–Vα

(a) Balanced set of voltages oscillating at 48 Hz and (b) Zoom of Fig. 4.7a around the zero cross. The
+
Vα measurement. M1 is va −v̂α1 which tends to zero. average error is slightly positive due to the sampling
process.

Figure 4.7: S1: steady state error for a balanced set of inputs oscillating at 48 Hz.

test signals containing sags, unbalance, harmonics and notching signals the three phase AC
source has been used as an arbitrary power amplifier connected to a three phase arbitrary
waveform generator based on a DSP card. The input signals have been acquired through
the dSpace I/O interface by means of LEM LV25-P voltage transducers. As in previous
section, p.u. units have been employed.

As the system is an open loop system, no feedback error signals can be presented.
+
However, considering only the fundamental component, va = v̂α1 for balanced conditions
0 − 0
and for unbalalanced systems when va = −va , that is vabc and vabc
−
cancel each other
+
for the a phase. Therefore, testing the systems under such conditions va − v̂α1 represents
+
the error; that is va − v̂α1 = 0 implies zero error in phase, frequency and amplitude. The
subtraction function of the oscilloscope has been used to obtain this error (channel M 1).

Fig. 4.7 shows the error signal of S1 in steady-state when the set of inputs is balanced,
and oscillates at 48 Hz. As shown, the error is negligible. In the zoomed figure, it could
be noticed the small positive delay between the input (va ) and the output or processed
+
signal (v̂α1 ). This error is due to the sampling process (execution and conversion times).

Fig. 4.8 shows the error signal of S1 in steady-state for an unbalanced set of inputs
+
oscillating at 51 Hz. As explained before, va− = −va0 and va1 −
max
= 0.1 · va1 max
in order
to test the error signal. These figures also prove the absence of steady state error of the
algorithm. Fig. 4.8b shows again the small delay in the visualization due to the sampling
process.

Fig. 4.9 shows the error signal of S1 before, during and after a transient. The pro-
+
grammed transient is a sag with post-fault component [32]. During the fault va1 max
=
0 −
0.8 p.u. and va = −va with va1max = 0.05 p.u., and a −45 deg phase jump with respect to
−
4.4 Experimental Results 147

Vb Va
Va, Vα
Vc
Vα

Va–Vα
Va–Vα

(a) Steady state measurements for a balanced set of (b) Zoom of Fig. 4.8a around the zero cross. The
+
voltages oscillating at 51 Hz. M1 depicts va − v̂α1 average error is slightly positive due to the sampling
as error signal tends to zero in steady-state. process.
− +
Figure 4.8: S1: Steady state error for an unbalanced (va1 max
= 0.1 · va1 max
) set of inputs
oscillating at 51 Hz.

+
the balanced wave of va1 max
= 1 p.u..

Once proved the theoretical approach in terms of zero steady state error and transient
response, other interesting figures are shown. Because of the features of the input waves,
the error signal is not available for these tests.

Fig. 4.10 shows how the measurement of θ1 through S1 for a set of unbalanced
+
−
(va1 max
= 0.09 · va1 max
) and harmonic contaminated (≈ 9% 5th , 5% 7th , 3% 11th ) in-
put waves rotating at 48 Hz is rippleless.

Figs. 4.11) and 4.12 compare S1 with S2 in terms of the trade-off between filtering
and transient response.

Fig. 4.11 shows the test made for a balanced input wave containing notches, which,
+
under the frequency domain point of view, is noise. Fig. 4.11a shows v̂α1 measured with
+
S1 contains notches, even though smoother than in input waves. Fig. 4.11b shows as v̂α1
measured with S2 is clean. These results are expected from Figs. 4.2a and 4.3a.

Fig. 4.12 shows how S2 improves the frequency rejection of S1 at cost of incrementing
the resources (execution time) and transient response. This test has been realized with
− +
an unbalanced (va1 max
= 0.09 · va1 max
) input wave which presents a very deep sag: from
1 p.u. to 0.1 p.u. with −45 deg phase jump. S1 is faster than S2 as expected from Figs.
4.2e and 4.3b. Figs. 4.12 emphasize the fact that the system gain is independent from
the input amplitude.
148 Chapter 4 Predictive based SRF-MAF Synchronization Algorithms

Vb
Vc

Va Va

Vα
Vα

Va–Vα Va–Vα

(a) S1: Transients response to a sag with −45 deg (b) S1: Transients response to a sag with −45 deg
+ +
phase jump. M1 depicts va − v̂α1 as error signal phase jump. M1 depicts va − v̂α1 as error signal
which is zero in steady-state. As expected, the tran- which is zero in steady-state. As expected, the tran-
sients lasts 0.01 s. Time scale at 5 ms. sients lasts 0.01 s. Time scale at 2 ms.

Figure 4.9: S1: Transient response for a sag with −45 deg phase jump.

4.4.1 Brief Comparative

Table 4.3 shows a brief comparative with other significant proposals which successfully
deal with unbalance [108, 140, 144] 2 .

Table 4.3
Brief comparative among significant systems with good unbalance rejection

System Settling time Execution Device Frequency Amplitude

(Ts ) Time Loop Dependence
S1 0.01 s 11.2 µs dSpace 1103 NO NO
S2 0.02 s 15.8 µs dSpace 1103 NO NO
DSRF-PLL ≈ 0.036 s –(< 50 µs) dSpace 1103 – YES
NPSF > 0.02 s 6.5 µs TMS320F2812 (DSP) YES YES
WLSE < 0.005 s 10 µs TMS320C31 (DSP) YES NO

The settling time (ts ) is defined as the time required for the system to settle within a
certain percentage of the steady-state value [192].

As proved, for S1 and S2 ts is accurately set by the step response of the FIR filtering
stage (eq. (2.43)): 0.01 s and 0.02 s respectively. In relation to device implementation,
it is clear that the moving average filters are a part resource consuming as shown in the
2
The comparative is made with respect to data provided in those works. DSRF-PLL is the Double
SRF-PLL provided in [108]
4.4 Experimental Results 149

Figure 4.10: S1: steady-state phase measurement at 48 Hz. Ch1 is θ1 , Ch2 is va , Ch3 is
vb , Ch4 is vc (phase in ≈ 2 rad, voltages in p.u.). The measurement does not have ripple.

increment of execution time from S1 to S2. However, this and other implementations
[128, 157, 158] prove that current digital devices are powerful enough to implement them
in real time. With regard to amplitude dependence, due to the linearity of all the employed
digital filters the system is amplitude independent.

The settling time ts (to within 2% of the final value) of the DSRF-PLL has been
estimated using the data (ωc = 2π25 rad/s, ξ = 0.707 ) of the equivalent linearized
systems studied in [108] through the formula [192]:

4
ts = . (4.11)
ξ · ωc

The accuracy of this estimation can be checked in Figs. 11 and 12 of [108]. The
execution time of the DSRF-PLL is not shown in [108], but, it is smaller than the imple-
mentation sampling time (50 µs).

The non-linearity introduced by the phase detector in PLLs, and therefore also in the
DSRF-PLL has the drawback of amplitude dependence, even though this can be reduced
through a normalization stage [126, 127, 137].

ts of NPSF presents a high vary depending on the kind of transient, as shown in Figs
12 of [144], which is a non linear feature. This time is always higher than 0.02 s.

The WLSE algorithm , tested when the input frequency is tracked (Figs. 6 and 7 of
[140]) presents a very quick response and good harmonic/noise rejection. However, its
frequency adaptation algorithm is very slow when compared with the NPSF one. In fact,
the frequency adaptation seems to be the main lack of the WLSE algorithm [108].
150 Chapter 4 Predictive based SRF-MAF Synchronization Algorithms

Va Va

Va Va

+ +
(a) Measurement with S1. v̂α1 also contains (b) Measurement with S2. v̂α1 is clean.
notches.
+
Figure 4.11: Comparative between S1 and S2 in terms of noise rejection. Ch1 is va1 max
,
Ch2 is va , Ch3 is vb , Ch4 is vc (voltages in p.u., ω1 = 2π · 49.75 rad/s).

Va Va

Va Va

(a) Transient response of S1. The steady-state is (b) Transient response of S2. The steady-state is
reached in 0.01 s. reached in 0.02 s. The settling time is doubled.
+
Figure 4.12: S1 Vs S2 in terms of transient response. Ch1 is va1 max
, Ch2 is va , Ch3 is vb ,
Ch4 is vc (voltages in p.u., ω1 = 2π · 49 rad/s).
4.5 Conclusions 151

In general, it could be said that adaptive filtering based algorithms such as Kalman
based and WLSE have a trade-off between transient response and frequency adaptation,
as proved in [145]. The WLSE approach of [140] is in the extreme of very fast transient
response but with a very slow frequency adaptation.

From table 4.3, it is clear that S1 and S2 present very good dynamics and are suitable
for real time implementation with a high performance.

4.5 Conclusions

This chapter shows how the SRF-MAF systems can be made frequency adaptive by
means of predictive filters. The methodology developed to obtain FIR predictive filters is
contributed. Some interesting features of the proposal should be highlighted.

• It has a very good filtering/transient response trade-off. It is specially advantageous

in the presence of unbalances and harmonic distortion.

• Since it is based on the implementation of linear filters its implementation is simple

and not very much resource-consuming. No explicit trigonometric functions are re-
quired if the Park’s transformations are implemented by means of the RC oscillator.

• Its dynamics are independent from the amplitude of the input waves, and therefore
the system is highly stable. This advantage is specially advantageous for phase-angle
tracking of variable rms waves such as the current in a SSSC [209].

• On the contrary to PLLs, specially overdamped tuned ones, there is not trade-off
between frequency and phase responses.

• It is frequency adaptive without resorting to an extra frequency loop. In It yields to

zero steady-state average error for all the measurements with independence of the
input signal frequency.

These features, bode well for the applicability of the algorithm in the control of grid
connected power converters, grid monitoring in distributed power generation systems, and
other applications where a fast and accurate on-line measurement of phase, frequency and
amplitude are mandatory.

Even though they were not developed or commented in the chapter, some interesting
points of this proposal should be also mentioned:

• The application of the algorithm in single-phase systems of working per-phase is

immediate, resulting in a predictive SRF-MAF1 system [209].
152 Chapter 4 Predictive based SRF-MAF Synchronization Algorithms

• The theoretical approach to achieve the predictive filters is based on predictive FIR
filters. However, IIR all-pass filters with a more advantageous amplitude response
(unitary gain) and positive linear-phase could be developed. In return, this solution
has a longer transient response.
Chapter 5

Frequency Adaptive PR Current

Controller for APFs

Abstract —

This chapter provides a novel approach in the implementation of Proportional+Resonant

current controllers, specially suitable for active power filters compensating for high order har-
monics. The difference equation to implement each resonant controller has been obtained from
the impulse invariant discretization method; an explicit estimation of the grid-frequency devia-
tion, which is obtained from a PLL, is employed to update online the current controller. The
method to update the difference equation, which is a proposal of this thesis, is based on a Tay-
lor approximation which avoids explicit trigonometric functions. Technical details of the whole
discrete-time controller are provided in the chapter.

A 1 kVA rated lab prototype has been implemented and tested. Experimental tests show
how the system performs both in steady-state and under different transients. Perfect tracking
in steady-state for a wide range of grid-frequencies is achieved. Regarding transient response,
different tests were programmed such as load transients (change in the load), sudden and large
changes in the ac mains frequency (frequency steps) and strong source voltage faults (voltage
sags with phase-jump). Experimental results prove how the system is robust and fast responding
to all of these possible real operation faults.

153
154 Chapter 5 Frequency Adaptive PR Current Controller for APFs

5.1 Introduction

The issue of current control for power electronic converters connected to the grid is a
field of paramount interest at present [210–217]. More specifically, schemes including res-
onant controllers to achieve perfect tracking/rejection in steady-state of periodic currents
have been proposed for all kind of power converter applications, such as active power fil-
ters (APFs) [119, 120, 210], distributed power generation systems (DPGSs) [2, 217–219],
rectifiers [213, 220] and motor drives [36, 88, 221].

Among other resonant-based controllers, Proportional+Resonant (PR) current regu-

lators are very interesting thanks to their potential high bandwidth [88, 120]. PR current
regulators can achieve perfect tracking for high order harmonics and also a fast transient
response. Another characteristic of PR current controllers is that they are suitable for
single-phase and three-phase applications: three-phase schemes, implemented in the αβ
frame, provide perfect tracking both for positive and negative sequences [88, 120].

When comparing with other structures, PR current regulators have some interesting
features, specially when active filtering operation is considered; in sum:

• Hysteresis current controllers are simple and achieve perfect tracking but, in general,
they have poor switching characteristics (variable switching frequency, maximum on-
time). Linear regulators, as PR schemes, overcome these limitations by separating
the modulation and regulation functions [222].

• Basic schemes based on linear regulators such as simple proportional and dead-beat
schemes do not achieve zero average steady-state error. In a general way, it can
be said that these kinds of controllers perform worse as higher is the harmonic
component to compensate [3]. PR current regulators seem to be a much better
alternative for APFs.

• It has been demonstrated that PR current controllers can perform as the ones based
on Synchronous Reference Frame (SRF) [88]. Under the discrete implementation
point of view, PR controllers are an interesting alternative since Park’s transforma-
tions are not needed [88, 120].

• Lately, repetitive regulators based on the discrete cosine transform (DCT) have been
proposed as an alternative to PR controllers [219, 223]. There are more similarities
than differences when both schemes are compared [219]. However, the PR structure
has been chosen in this work due to the fact that its frequency adaptation is more
approachable, since less digital filter coefficients (taps) are involved.

Focusing the problem in PR current regulators for APFs, their open loop transfer
function should have resonant peaks at selected frequencies, as well as at the fundamental
one [119, 120]. The resonance at fundamental frequency provides perfect tracking at
5.1 Introduction 155

fundamental currents (i.e. for dc bus control) as well as total rejection to the fundamental
component of the grid voltage (disturbance) [119].

A practical problem of resonant filters, and hence in PR controllers, arises with grid-
frequency deviations owing to their high selectivity; if not compensated, frequency varia-
tions cause that resonant peaks do not coincide with input wave frequencies and, therefore,
perfect tracking/rejection is not assured [88]. This problem may be specially critical in
isolated systems where ranges of operation within a range of ≈ ±2 Hz, or higher, are
taken into account [224]. Indeed, the higher the frequency deviation, the lower the gain;
so, in practice, for so high non-compensated frequency deviations, a PR controller works
as a simple proportional one [88].

With regard to frequency adaptation of resonant controllers, it should be reminded

that there is no benefit to be gained in using damped resonant implementations [88].
Different works have proposed the use of phase locked loops (PLLs) to adapt repetitive and
resonant controllers (resonant peaks) to the grid frequency [2, 119, 218, 223]. A variable
sample rate repetitive controller, where the sampling frequency is driven by the PLL,
has been proposed in [223]. This solution could be approached for PR implementation.
However, the drawback of this option is that, depending on the discrete device, a variable
sample rate implementation could not be a trivial task. Another interesting proposal is
to decompose each resonant controller in two integrators and use the resonant frequency
as an input. The difference equation is updated online using the frequency estimation of
a PLL [2, 119, 218]. The problem of this approach is the fact that the “two integrators”
based implementation causes a significant displacement of the resonant poles, as proved
below. On the other hand, the maximum level of ripple from the PLL estimation was not
assessed in previous works.

In this chapter the discrete-time implementation of single-phase APF based on a novel

frequency adaptive PR current controller is approached. The APF performs a selective
harmonic cancellation of 3rd , 5th and 7th harmonics. The difference equation of each
resonant controller is obtained from the impulse invariant discretization method, which
provides an accurate placement of the resonant peaks. These difference equations are
updated online accordingly by means of the PLL estimation of the grid-frequency devi-
ation. Experimental results show how the proposed APF achieves perfect tracking for a
big range of input frequencies as well as its robustness under different kinds of distorted
conditions. The maximum allowable level of ripple from the PLL frequency estimation
has been assessed experimentally.
156 Chapter 5 Frequency Adaptive PR Current Controller for APFs

Table 5.1
Significant values of the power circuit.

Parameter Value
vdc
∗
220 V
vSrms 110 V
C 1 mF
ZF LF + R F
LF 5 mH
RF 0.5 Ω
LS 50 µH
fsw = fs 10 kHz
tcomp 12 µs

5.2 Active Power Filter Prototype

Fig. 5.1 depicts the scheme of the lab prototype and its discrete controller. Table 5.1
shows the significant values of the hardware.

The power supply is composed of two programmable single-phase ac sources “Chroma

61501” working in parallel. This configuration supplies a maximum power of 1 kVA. The
APF is an IGBTs based voltage source converter (VSC) connected to the point of common
coupling (PCC) through the filter impedance (ZF ). The whole load is composed of two
non-linear loads also connected to the PCC: an uncontrolled rectifier and an ac regulator.
A manual switch has been placed before the ac regulator in order to cause load transients.

The discrete controller has been implemented in the dSpace DS1103 platform, which
implements Matlab/Simulink models in real-time. The discrete solver has been chosen
so all the controller blocks are digital filters in the Z-domain (neither “ode” solvers nor
“continuous” blocks). The sampling frequency (fs ) and IGBTs switching rate (fsw ) have
been set to 10 kHz; the sampling period Ts is defined as 1/fs . System currents and voltages
have been sensed using LEM-LA55P and LEM-LV26V sensors, respectively. These sensed
signals are acquired through the I/O interface of the dSpace target. The average time of
operation tcomp was around 12 µs. Fig. 5.2 shows a picture of the built prototype as well
as a capture of a ControlDesk layout in real-time operation.

The goals of the digital control are: to compensate for the selected load harmonic
currents and to keep constant the dc-link voltage (vdc ). The proposed controller works as
follows.

• The instantaneous load current iL is detected.

5.2 Active Power Filter Prototype 157

AC REGULATOR
MANUAL
SWITCH
vS
LS iS PCC iL
vPCC
VSC
UNCONTROLLED
vF ZF iF RECTIFIER
vDC C

PWM
m
vPCC iL
P+RESONANT
∆ω
ˆ1 ω̂1 HARMONIC
CURRENT
REGULATOR
PLL EXTRACTION
vDC sin(θˆ 1 )
-
PI X
+ iF
- +
i1*
vDC *
+ +
iF* iLh*
DIGITAL CONTROLLER (dSpace DS1103)

Figure 5.1: APF prototype and controller.

• The reference harmonic currents to compensate (i∗F h ) are extracted from iL by means
of digital signal processing. A selective extraction of 3rd , 5th and 7th components is
performed using SRF-MAFs [148] (i∗F h = îL3 + îL5 + îL7 ).

• The reference current i∗F 1 , calculated to maintain vdc , is obtained through a PI

controller and the in-phase signal from the PLL [18]. The subscript “1” emphasizes
the fact that only the fundamental component is controlled to set the active power
flow from the PCC to the VSC dc-link.

• The total reference of current for the APF (i∗F ) is calculated as i∗F h + i∗F 1 .

• The PR current regulator assures that (i∗F − iF ) is zero in steady-state (for the
fundamental, 3rd , 5th and 7th harmonic components).

The PLL and PR schemes are detailed in subsequent sections.

158 Chapter 5 Frequency Adaptive PR Current Controller for APFs

(a) Detail of dSpace ControlDesk layout.

VSC
I/O Interf
Non-linear
Load circuits
Box with
sensors (PCC)

PC with
dSPace
Non-linear
DS1104 Load circuits

AC Sources
connected
in parallel

(b) Photo of the lab prototype.

Figure 5.2: Experimental set-up features.

5.3 Phase Locked Loop 159

2π ·3
∆ω̂1
− 2π ·3
MAF2

vPCC ( p.u.) θ1e

X PI MAF1 + + ω1n

cos(θˆ1 ) θˆ1 ω̂1

∫
π 2π ·53

2π ·47
cos(u)
−π

sin(θˆ )
1 sin(u)

Figure 5.3: Phase locked loop.

5.3 Phase Locked Loop

A single-phase PLL has been implemented to track online the phase-angle (θ1 ) and
frequency (ω1 ) of vP CC fundamental component. Fig. 5.3 shows the PLL block diagram.
It uses a multiplier as phase detector. The loop filter is composed of a proportional
integrator (PI) controller and a moving average filter (MAF1), for which the equivalent
transfer functions are

c1 − c0 · z −1
C(z)P I = (5.1)
1 − z −1
and

1 1 − z −n
C(z)M AF 1 = · , (5.2)
n 1 − z −1
respectively. The PI controller assures that θ1 and ω1 are tracked with zero average
steady-state error, even though ω1 6= ω1n [78, 112–114]. A second MAF (MAF2) with
the same transfer function as C(z)M AF 1 can be placed before ∆ω̂1 estimation in order to
control the ∆ω̂1 ripple.

The PLL outputs are employed with the following goals:

1. A normalized in-phase signal (sin(θ̂1 )) is multiplied by the output of the PI controller

to obtain i∗F 1 [18].
2. The frequency estimation is employed to generate the multiple SRF waves of the se-
lective extraction algorithm. The frequency value is saturated to avoid malfunctions
during vP CC transients [148].
3. Usually, digital PLLs are designed to have a feedforward frequency (ω1n ), so the
closed-loop path provides a correction (∆ω̂1 ) around it (ω̂1 = ω1n + ∆ω̂1 ). This
160 Chapter 5 Frequency Adaptive PR Current Controller for APFs

Table 5.2
Values of PLL parameters.

Parameter Value
c1 2
c0 1.995
n (MAF1) 100
n (MAF2) 100
100

50
Magnitude (dB)

-50

-100

-150

-200

-90
Phase (deg)

-135

PM=31 deg
-180

-225

-270
-1

1 10 28 100 1000 10000

Frequency (Hz)

Figure 5.4: Open loop PLL Bode diagram.

work extends this approach to resonant controllers design: the difference equation
of each one is function of h · ω1n and h · ∆ω̂1 (see section 5.4).

5.3.1 PLL Tuning

PLLs have a trade-off between transient response and filtering: the higher the band-
width the faster the transient response but the worse the harmonic cancellation. Specially
critical is the internal second harmonic generation in multiplier based single-phase PLLs.
If not canceled, this second harmonic causes ripple in ∆ω̂1 , ω̂1 and θ̂1 , which could affect
to the whole controller performance [112–114].

Fig. 5.4 shows the open loop frequency response of the linear PLL. The values of the
coefficients are shown in table 5.2. Stability and bandwidth optimization has been sought
during the tuning process [112–114]. The phase margin (PM) could be considered a bit
low for a controller design, but, in the case of PLL tuning, a very big increase in vSrms is
not expected, so stability is assured with this PM [112–114].
5.4 PR Current Controller 161

Forward
Euler
iF*-iF Ts z − 1 output
1 − z -1
+
-

ω̂ 12
Ts
1 − z -1 X
Backward
Euler

Figure 5.5: Two integrators based implementations, as proposed in [2].

From Fig. 5.4, when ω1 = ω1n , thanks to MAF1 all the harmonics, including the
second one, are canceled. However, when ω1 6= ω1n perfect harmonic cancellation is not
achieved.

It is out of the scope in this chapter to show how to optimize more the PLL loop filter
in order to reduce jittering. Instead of it, harmonic ripple effect in PR current controllers
is studied. In this work, control over ripple has been achieved by placing/removing MAF2.

5.4 PR Current Controller

5.4.1 Resonant Controllers Implementation

Resonant filters are undamped GIs which can be implemented as two linked integra-
tors [119]. Frequency adaptation is provided by means of PLL estimation [2, 119, 218].
Regarding the discretization method, it has been suggested to discretize the direct integra-
tor using Forward Euler method while the feedback one is discretized using the Backward
Euler method, as shown in Fig. 5.5 [2]. The resulting transfer function in the Z-domain
is:

z −1 − z −2
Rtwo (z) = Ts . (5.3)
1 + z −1 (ω12 Ts2 − 2) + z −2

It can be checked that to discretize the two integrators separately results in a displace-
ment of the resonant peaks from the expected values. However, the linear dependence of
the different equation on ω12 eases the frequency adaptation.

To use the Z-transform of the cosine wave was proposed in [120]. This means, in
practice, to discretize a second order generalized integrator with the impulse invariant
162 Chapter 5 Frequency Adaptive PR Current Controller for APFs

method, giving rise to the following resonant controllers:

1 − z −1 cos(ω1 Ts )
Rii (z) = Ts · . (5.4)
1 − 2z −1 cos(ω1 Ts ) + z −2

This method results much more accurate, as shown in Fig. 5.6. Its drawback ap-
pears when frequency adaptation is desired, since trigonometric functions are involved. A
method to avoid explicit cosine wave calculations is provided in section 5.4.2.

The error induced by the two integrators based implementation may be acceptable for
some applications handling with very low order harmonics and fundamental. However,
the loss of performance is very significant for higher harmonics. Rii (z) seems to be a
better choice, specially for APFs.

A PR controller (C(z)) of the form of (5.5) has been selected for implementation. Ts
is a constant term so it can be included in each Kh .

X ′ 1 − z −1 · cos(h · ω1 · Ts )
C(z) = KP + KRh · Ts · =
h=1,3,5,7 1 − 2 · z −1 · cos(h · ω1 · Ts ) + z −2
(5.5)
X 1 − z −1 · cos(h · ω1 · Ts )
KP + KRh · .
h=1,3,5,7 1 − 2 · z −1 · cos(h · ω1 · Ts ) + z −2

5.4.2 Frequency Adaptive Implementation of Resonant Blocks

From (5.5), it seems immediate to update the terms cos(h · ω1 · Ts ) using ω̂1 from
the PLL [225]. The main drawback of this approach is that it requires computation of
trigonometric functions online. The explicit implementation of the cosine function can be
avoided, using the proposed Taylor approximation detailed as follows.

In a similar manner that the digital PLL of Fig. 5.3 works around the equilibrium
point established by ω1n , the PR current regulator of (5.6) has been derived from (5.5):
assuming that h · ∆ω̂1 · Ts ≈ 0, the “equilibrium point strategy” of PLL is also applied to
resonant controllers.

X 1 − z −1 (cos(h · ω1n · Ts ) − h∆ω̂1 Ts sin(h · ω1n · Ts ))

C(z) = KP + KRh · .
h=1,3,5,7 1 − 2 · z −1 (cos(h · ω1n · Ts ) − h∆ω̂1 Ts sin(h · ω1n · Ts )) + z −2
(5.6)

Fig. 5.7 shows the implementation block for each resonant controller. Block diagram
5.4 PR Current Controller 163

Two integrators based implementation

Impulse Invariant based implementation

Hz

100 200 300 400 500 600

Hz

Figure 5.6: Comparative of the error induced by different implementations (fs = 10 kHz).

iF*-iF
KRh +
+ + To adder
- Z-1 ah ++ -

Z-1 bh X

∆ω
Precalculated constants:
ah = cos(h·w1n·Ts) ^ h·Ts
1
bh = sin(h·w1n·Ts) (From PLL)

Figure 5.7: Frequency adaptive resonant block implementation.

algebra has been applied to optimize the implementation set by (5.6), so the number of
multiplications by non-fixed coefficients (∆ω̂1 ) has been reduced to one.

Fig. 5.8 shows the accuracy of the approximation for each resonant block. The higher
the harmonic order, the worse the approximation, since a shift of ±1 Hz at fundamental
results in a ±h · 1 Hz shift at the harmonic h. At any rate, this error can be considered
low enough, even for significant frequency deviations, as proved in section 5.5.

5.4.2.1 Effect of the Frequency Correction in Dynamics

As explained, when ∆ω̂1 6= 0 the resonant peak is moved to the estimated frequency;
i.e. Fig. 5.9 shows the effect of the frequency correction around the 3rd harmonic resonant
peak. It could be said that when the peak is “well centered” the PR is working as an ideal
resonant controller (infinite gain), but in other case the performance tends to approach
164 Chapter 5 Frequency Adaptive PR Current Controller for APFs

0.35

Fundamental
3rd harmonic
5th harmonic

Hz
0.25 7th harmonic

0.15

0.05

-1.5 -0.5 0.5 1.5

1 Hz

Figure 5.8: Accuracy of the resonant controllers (fs = 10 kHz).

100
90

∆ω1<0
Magnitude (dB)

80
∆ω1>0

70

60
50

40
30
Proportional Resonant Proportional
controller controller controller
20
45
Phase (deg)

0
-45

-90

-135

-180
149.8 149.9 149.98 150 150.02 150.1 150.2
Frequency (Hz)

Figure 5.9: Frequency response of C(z) · P (z) around 150 Hz. When ∆ω̂1 > 0 the peak
”moves” to higher frequencies and vice versa.

the one of a proportional controller. The exact place to depict the resonant/predictive
borderline in that figure could be considered a subjective issue. The idea is to show how
a practical very high gain is sought with frequency adaptation.

∆ω̂1 , as obtained from a PLL, can be written as

¯ 1 + ∆ω̂
∆ω̂1 = ∆ω̂ ˜1, (5.7)
where ∆ω̂¯ 1 and ∆ω̂
˜ 1 are the average value and an oscillating term associated with jittering,
respectively.

From Fig. 5.9, the best performance occurs when ∆ω̂ ¯ 1 is tracking perfectly, that is,
˜ 1 = 0 (no jittering). The first condition is
the resonant peak is well centered, and ∆ω̂
easily performed by any PLL, but the second one is more complicated, specially under
5.4 PR Current Controller 165

vPCC

iF
−1
iF* m -
+- C ( z) z + G (z )
iF Plant Model≡ P(z)

Figure 5.10: Current control block diagram.

distorted conditions [112–114]. Analytical expressions of the maximum level of jittering

seem to be very complex to obtain. Alternatively, the quantification of the maximum
level of jittering has been assessed by experimental tests.

5.4.3 Design and Tuning of C(z)

Fig. 5.10 depicts the current control system including the PR controller C(z) and the
model of the plant P (z). P (z) has been modeled including the computational delay (z −1 )
and the PWM converter operation. The PWM reference (m) is kept constant over each
sampling interval, so the power converter can be assumed to be a zero order hold (ZOH)
circuit. Therefore, the filter inductance discrete-time model (G(z)) should be obtained
through the ZOH method, as follows [3, 226]:

1 ZOH 1 1 − e−Rf ·Ts /Lf

G(s) = → G(z) = · . (5.8)
sLF + RF Rf z − e−Rf ·Ts /Lf

The effect of the ZOH circuit is noticeable at high frequencies, since an extra phase
lag is introduced in the current control loop, and it should be taken into account when
tuning C(z).

As said in the introduction, resonant controllers provide perfect tracking (and distur-
bance rejection) of currents for which frequency C(z) · P (z) has infinite gain.

Fig. 5.11 shows C(z) implementation. Four resonant blocks have been implemented
in the APF controller: one at fundamental component to provide perfect tracking of i∗F 1
(vdc control) and vP CC fundamental rejection, and the other three ones to provide perfect
tracking of i∗F 3 , i∗F 5 and i∗F 7 . The term 1/vdc
∗
is included so Kp and Kh tuning is made
independent from vdc (plant).

The controller C(z) has been tuned by inspection of C(z) · P (z) transfer function
[120, 219]; ω1 = ω1n has been considered. Fig. 5.12 shows C(z) · P (z) frequency response
and the parameters are specified in Table 5.3. From Fig. 5.12, the implemented PR
controller has the following features:
166 Chapter 5 Frequency Adaptive PR Current Controller for APFs

iF*-iF
KP +

Res.1st +
m
Res.3rd +

Res.5th +

∆ω
Res.7th +
^1
(From PLL)

Figure 5.11: PR block (C(z)).

Table 5.3
Values of parameters of the PR controller.

Parameter Value
KP 25
KR1 0.1
KR3 0.1
KR5 0.1
KR7 0.1

• High enough phase margin (P M = 46 deg) so that stability is assured.

• Perfect tracking and disturbance rejection at resonant peaks are expected.

• A fast transient response is expected: the open loop cut-off frequency and hence the
system bandwidth are around 800 Hz.
5.5 Experimental Results 167

100

80

Magnitude (dB)
60

40

20

-20
180
135
Phase (deg)

90
45
0
-45
-90
-135
PM=46 deg.
-180
10 100 800 1000
Frequency (Hz)

Figure 5.12: Bode diagram of C(z) · P (z).

5.5 Experimental Results

Different tests have been carried out to show how the APF performs both in steady-
state and under transients.

Fig. 5.13 shows experimental results obtained from steady-state operation when an
“input frequency sweep” has been carried out. Oscilloscope figures show results at nomi-
nal (50 Hz) and very shifted (48 Hz) frequencies; system currents and vP CC are displayed.
It should be noticed that iL spectrum varies slightly with the frequency due to load reac-
tances. Fig. 5.13g depicts a table with all the results of the “input frequency sweep”. These
results prove that the 3rd , 5th and 7th harmonics of iL are not present in iS . Therefore,
perfect tracking of the APF harmonic reference is achieved. The fundamental component
of iS is slightly higher than iL1 , since vdc should be maintained.

Fig. 5.14 shows experimental results when there is a load transient. Initially only the
uncontrolled rectifier is connected to the PCC and then the ac regulator starts up. As
expected from the PR controller bandwidth (800 Hz) the system has a very fast transient
response. It should be noticed that the extraction algorithm settling time is ≈ 0.01 s and
also affects to dynamics [148].

Fig. 5.15 illustrates the APF response when there is a sudden step in f1 . Fig. 5.16
depicts results from a similar test but showing ∆ω̂1 from the PLL in Ch2. In both cases,
the PR current regulator and, therefore, the whole system presents a very good frequency
adaptation. The frequency step has been detected inside the dSpace model and connected
168 Chapter 5 Frequency Adaptive PR Current Controller for APFs

(a) Steady-state currents and voltage at the (b) Fourier spectra of iL in (c) Fourier spectra of iS in
PCC when f1 = 50 Hz. Fig. 5.13a (Ch1). Fig. 5.13a (Ch3).

(d) Steady-state currents and voltage at the (e) Fourier spectra of iL in (f) Fourier spectra of iS in
PCC when f1 = 48 Hz. Fig. 5.13d (Ch1). Fig. 5.13d (Ch3).

48 Hz 49 Hz 50 Hz 51 Hz 52 Hz
h iL iS iL iS iL iS iL iS iL iS
1 3.02 A 3.2 A 2.96 A 3.13 A 2.87 A 3A 2.79 A 2.94 A 2.71 A 2.87 A
3 886 mA 30 mA 874 mA 10 mA 856 mA 4 mA 826 mA 14 mA 738 mA 25 mA
5 223 mA 8 mA 206 mA 5 mA 199 mA 4 mA 190 mA 6 mA 186 mA 9 mA
7 176 mA 13 mA 153 mA 9 mA 134.6 mA 3 mA 115 mA 6 mA 94 mA 11 mA
(g) Values of harmonic current of the load and the source for different frequencies.

Figure 5.13: Steady-state currents and vP CC for different input frequencies. Ch1 is iL ,
Ch2 is iF , Ch3 is iS and Ch4 is vP CC .
5.5 Experimental Results 169

Figure 5.14: Transient response when there is a load change. Ch1 is iL , Ch2 is iF , Ch3 is
iS and Ch4 is vP CC .

Figure 5.15: Transient response when there is a frequency step in f1 from 48 Hz to 52 Hz.
Ch1 is iL , Ch2 is iF , Ch3 is iS and Ch4 is vP CC .

to the oscilloscope external trigger. Again, it is noticeable the change in iL spectrum due
to the frequency dependence of load reactances.

Figs. 5.15 and 5.16 have been obtained by implementing the PLL with MAF2. As
expected, ∆ω̂1 is ripple-less (two MAFs). Fig. 5.17 results have been obtained repeating
the previous test, but eliminating MAF2 from the PLL. In this test, a second-harmonic
oscillating term ∆ω̂˜ 1 oscillates ≈ ±1 Hz around ∆ω̂ ¯ 1 . This result is very interesting,
and even quite surprising, since the practical high performance of the PR controller is
kept. Even though, posterior tests made by the authors with a higher level of jittering
(±3 Hz) resulted in instability, it is clear that this implementation is very robust under
the presence of harmonics/noise ripple in the frequency estimation.

Another test has been made considering ∆ω̂ ˜ 1 = 0, that is, a PR without frequency
adaptation. Fig. 5.18 shows results obtained when the input frequency is 48 Hz. The
same results have been obtained testing with a current controller C ′ (z) = KP . Therefore,
the PR controller without frequency adaptation behaves equal that a simple proportional
regulator.
170 Chapter 5 Frequency Adaptive PR Current Controller for APFs

Figure 5.16: Transient response when there is a frequency step in f1 : from 52 Hz to 48 Hz.
Ch1 is iL , Ch2 is ∆ω̂1 (scale at 2π∆3 rad/div), Ch3 is iS and Ch4 is vP CC .

Figure 5.17: Transient response when there is a frequency step in f1 from 48 Hz to 52 Hz.
Effect of ripple of ≈ ±1 Hz in ∆ω̂1 . Ch1 is iL , Ch2 is ∆ωˆ1 (scale at 2π∆3 rad/div), Ch3
is iS and Ch4 is vP CC .

Finally, the prototype has been tested when there is a voltage sag (from 1 p.u. to
0.8 p.u.) with phase-angle jump (+45 deg) in vP CC (programmed in vS ). The test has
been made at 49 Hz (∆ω̂1 = −2π rad/s). When the fault occurs the PLL frequency
increases, but ∆ω̂1 to the PR is limited to 2π∆3 rad/s. As shown in Fig. 5.19, this
protection is crucial and steady-state is achieved in ≈ 1 − 2 cycles. It should be noticed
that the fault also affects to the extraction algorithm [148].
5.5 Experimental Results 171

(a) Steady-state currents and voltage at the (b) Fourier spectra of iL (c) Fourier spectra of iS
PCC when f1 = 48 Hz. (Ch1 of Fig. 5.18a). (Ch3 of Fig. 5.18a).

h 48 Hz 49 Hz 50 Hz 51 Hz 52 Hz
1 3.14 A 3.08 A 3.02 A 2.90 A 2.85 A
3 349 mA 282 mA 4 mA 297 mA 320 mA
5 71 mA 60 mA 4 mA 73 mA 81 mA
7 56 mA 38 mA 3 mA 29 mA 32 mA
(d) Values of iS for ∆ω̂1 = 0 (No frequency adaptation).

Figure 5.18: Steady state figures when error in estimation of ∆ω̂1 is considered.
∆ω̂1 = 0 rad/s and ∆ω1 = −4π rad/s (f1 = 48 Hz). Ch1 is iL , Ch2 is ∆ωˆ1 (scale at
2π∆3 rad/div), Ch3 is iS and Ch4 is vP CC .

Figure 5.19: Transient response when there is a voltage sag (1 p.u. → 0.8 p.u.) with phase-
angle jump of +45 deg in vP CC (vS ). Ch1 is iL , Ch2 is ∆ωˆ1 (scale at 2π∆3 rad/div), Ch3
is iS and Ch4 is vP CC .
172 Chapter 5 Frequency Adaptive PR Current Controller for APFs

5.6 Conclusions

A novel approach in the implementation of PR current regulators is contributed in

this chapter. More specifically, the issue of frequency adaptation using a PLL is deeply
analyzed. From this study, some practical conclusions of significant importance can be
extracted. In sum:

1. Resonant controllers implementations based on the impulse invariant discretization

are more reliable for APFs than the ones based on two discrete integrators.

2. The impulse invariant resonant controllers can be made frequency adaptive by means
of a PLL estimation of the grid-frequency deviation. The proposed implementation
has good performance and no needed of explicit trigonometric functions.

3. The most critical parameter of the PLL is its accuracy detecting the average value of
the fundamental frequency. Jittering in ∆ω̂1 should be reduced as much as possible,
even though high robustness to this parameter is provided.

4. The tested resonant controllers achieve perfect tracking even for the worst case (7th
harmonic and ±2 Hz deviation), so it is also expected a good performance in APF
applications compensating for higher order harmonics.

5. The proposed PR controller has a very high bandwidth and, therefore, the system
has a very fast transient response. The whole controller is robust under all kind of
transients such as load changes, frequency variations, and even input voltage sags.
It should be remarked that some PLL outputs (estimations) have been saturated to
avoid malfunctions.
Chapter 6

Conclusion and Future Work

6.1 Conclusion

This dissertation addresses the issue of grid-synchronization for power electronics con-
verters. Its main contributions and conclusions are summarized below.

• A thorough review of the State-of-the-Art is contributed. The problem of frequency

adaptation is presented as the most important practical issue in the implementation
of discrete algorithms.

• Methods based on SRF-MAF both for grid-synchronization as well as for harmon-

ics/sequences extraction are provided. It is also proved that there is a close relation
between these implementations and other based on DFT and DCT.

• An in-depth analysis of the dynamics of low-gain PLLs. It is proved that, opposite as

considered in some works, overdamped tuning should be avoided when considering
frequency deviations.

• A novel methodology based on RC oscillator model to optimize the discrete imple-

mentation is contributed.

• An open-loop frequency adaptive SRF-MAF algorithm providing an accurate and

fast grid phase-angle estimation is contributed.

• A novel approach on the implementation of PR current controllers is contributed.

The provided APF implementation proves the high performance of the proposed
implementation as well as the robustness and frequency adaptability of the solution.

173
174 Chapter 6 Conclusion and Future Work

6.2 Publications

Research work included in the dissertation has given rise, at the moment, to three
journal papers [114, 147, 148] and eight conference papers [112, 113, 127, 146, 149–
151, 209]. A summary of the contributions of each journal paper is detailed as follows:

1. The basics of SRF-MAF synchronization/extraction algorithms, presented in chap-

ter 2, are detailed in [148]. This paper focuses more in extraction for active power
filters than in synchronization itself.

2. The tuning methodology for PLLs proposed in chapter 3 is provided in [114].

3. The synchronization method based on SRF-MAFs and predictive filters, presented

in chapter 4, is provided in [147].

6.3 Future Research

There are several interesting topics suggested for further research. Some of them are
described in the following:

• As shown in chapter 5, frequency adaptation of PR controllers is not a trivial task.

An immediate future task is to compare more deeply the provided technique based
on Taylor approximations with other alternatives presented in technical literature.

• A more in-depth research of the applicability of the different synchronization algo-

rithms (proposed and studied) in distributed power generation applications in order
to fulfill with existing normative (Grid codes).

• Rotor position and speed estimation of synchronous machines operating at any

speed.

• Implementation of the most significant proposals (Predictive SRF-MAF, Adaptive

PR current controllers,...) in industrial devices such as FPGAs and DSPs.

• It is also expected to carry out solutions for industry based on the contributions of
this thesis.
Appendix A

Matlab Scripts

175
176 Chapter A Matlab Scripts

1 %% Matlab Script for executing the Ainsworth model

2 %% Main File
3 %% Francisco D. Freijedo, Glasgow, August 2008
4
5 clear all
6 % Step of the simulation.
7 % Input Frequency
8 % Final time of the simulation
9 % Time of a transient
10 % Time vector
11 %R eq
12 % L eq
13
14 % Initial value of current
15 % initalize output current reference
16 % initial condition for VCO phase
17 %inital value of grid phase
18 % initalize frequency correction of VCO
19 % initalize error vector
20 % initalize output current vector
21 % initalize output current vector average value
22 % initizalize Vd output vector
23 % initialize output frequency correction vector
24 while
25 % time for each simulation loop
26 if taux>ttran
27 % transient changes
28
29 end
30 %function call
31 [il,Vd,Vln,phaseVCO,wcorr,e,phaseGrid]=calc_icl(tsim,Ts,w1,...
32
33 taux=taux+tsim % show current simulation time
34 % update output vector
35 % save output current value for next function call
36 % save output current average value
37
38
39 end
40 % Visualize values
41 % PLOT CURRENTS
42
43
44
45
46
47 %PLOT STEADY-STATE VOLTAGES
48 plot((0:length(Vln)-1)*Ts+tfinal-0.5,Vln,...
49 (0:length(Vd)-1)*Ts+tfinal-0.5,Vd,[Ts+tfinal-0.5 Ts+tfinal]...
50
51 xlabel('Time (s)','FontSize'
52 ylabel('Current (p.u.)','FontSize'
53
54
55
56
57
58

Figure A.1: PLO system Matlab script. Main file.

 177

1 %% Matlab Script for executing the Ainsworth model

2 %% Function file
3 %% Francisco D. Freijedo, Glasgow, August 2008
4
5 function [il,Vd,Vln,phaseVCO,wcorr,e,phaseGrid]=calc_icl(tsim,Ts,w1,R,L,phia,phib,
phic,iin,phaseVCOi,wcorri,iref,ei,phaseGridi)
6 % 30 deg phase offset
7
8 % Initialize current values from previous function calls
9 % Grid phase
10 % VCO phase
11 % VCO correction around the nominal frequency (2*pi*50 rad/s)
12 % Set VCO frequency
13 % current error signal
14 % current value of the system current
15
16
17 % The Ainsworth problem is stated as a state state.
18
19 % .
20 % x=Ax+Bu
21 % y=x
22 % where x=y is the current through L and R and
23 % u is the set line to neutral voltages vector.
24
25 % A in the continuous space state
26 % Discretize of A
27
28 % Definition of the switching functions:
29 % Different swtiching functions are defined since the circuit
30 % depends on the VCO phase. That is, B is changing with VCO phase.
31
32 % Continuos swtiching function
33 % Discretized switching function
34
35
36 % ss en con
37
38
39
40
41
42
43
44
45 % Initial value of Bd
46
47 for
48
49 % Calculation of instantaneous phase to neutral voltages
50 Va=sin(offset+phaseGrid+phia)/sqrt(3)+0.1*sin(5*phaseGrid)+0.1*sin

51 Vb=1*sin(offset+phaseGrid+phib)/sqrt(3)+0.1*sin(5*phaseGrid+0.1)+0.1*sin

52 Vc=1*sin(offset+phaseGrid+phic)/sqrt(3)+0.1*sin(5*phaseGrid+0.9)+0.1*sin

53
54 % Error signal vector update
55
56

Figure A.2: PLO system Matlab script. Function file (page 1 of 2).
178 Chapter A Matlab Scripts

57 % The output of the filter is obtained through the discretized

58 % transfer function using the tustin method
59 tmp=1.3333*wcorr(2)-0.33329*wcorr(1)+33.334*e(3)+0.0033332*e(2)-33.331*e

60
61
62 % % VCO frequency update
63
64
65 % The phase must be in the range [0,2*pi]
66 while phaseGrid>=2*pi
67
68 end
69
70 while phaseVCO>=2*pi
71
72 end
73
74 while phaseVCO<0
75
76 end
77
78 % Set of Bcd as function of VCO phase (switching functions)
79 if phaseVCO<=pi/3
80
81
82 end
83 if phaseVCO>pi/3 && phaseVCO<=2*pi/3
84
85
86 end
87 if phaseVCO>2*pi/3 && phaseVCO<=pi
88
89
90 end
91 if phaseVCO>pi && phaseVCO<=4*pi/3
92
93
94 end
95 if phaseVCO>4*pi/3 && phaseVCO<=5*pi/3
96
97
98 end
99 if phaseVCO>5*pi/3
100
101
102 end
103
104 % Save output current vector
105
106 % Save Vd voltage
107 % Save line to neutral voltages
108 % Grid phase update
109 % VCO phase update
110 end

Figure A.3: PLO system Matlab script. Function file (page 2 of 2).
 179

1 %% Matlab Script to get SRF-PLL trajectories in the Phase-Plane

2 % Francisco Freijedo
3 Ts=1e-4; Tfinal=0.1; t=0:Ts:Tfinal;
4 wn=2*pi*50; % Feedforward constant at nominal frequency (or =0).
5 ypd=0; ylfP=0; ylfIv=[0 0 0 0 0 0 0]; theta=0; thetadif=0; difw=[];
6 diftheta=[];
7 figure; hold on;
8 colors=['k' 'y' 'b' 'm' 'g' 'r' 'c']';
9 phiuv=[pi-0.1 pi/2 pi/4 0 -pi/2 -pi+0.6 -pi+0.1];
10 limitI=[1000 1000 1000 1000 1000 1000 1000]; %Limit the word lenght
11 ylf=ylfIv;
12 for u=1:size(phiuv')
13 % Create INPUT WAVE for SIMULATION
14 wu=2*pi*51; % Input wave frequency
15 phiu=phiuv(u);
16 thetain=rem((wu*t+phiu)+pi,2*pi)'-pi;
17 ua=sin(wu*t+phiu); % Input wave definition
18 ub=sin(wu*t+phiu-2*pi/3); % Input wave definition
19 uc=sin(wu*t+phiu+2*pi/3); % Input wave definition
20 % SIMULATE THE PROCESS
21 ylf=ylfIv(u); cont=2; difw=wu-(wn+ylf); diftheta=[phiuv(u)];
22 for n = 1:Tfinal/Ts % Number ot iterations
23 ypd(n+1)=ua(n)*cos(theta(n))+ub(n)*cos(theta(n)-2*pi/3)+...
24 uc(n)*cos(theta(n)+2*pi/3); % Phase Detector (Q15)
25 ylf(n+1)=(1*ylf(n)+250*ypd(n+1)*1-248*ypd(n)*1);
26 ylf(n+1)=max([ylf(n+1) -1*limitI(u)]); % Limit LF
27 ylf(n+1)=min([ylf(n+1) 1*limitI(u)]); %(according to a pipeline)
28 wo=wn+ylf(n+1);
29 theta(n+1)=theta(n)+abs(wo)*Ts; % Effect of discard frequency sign
30 % theta(n+1)=theta(n)+wo*Ts; % Good operation
31 if theta(n+1)>=pi % Reset phase-angles
32 theta(n+1)=theta(n+1)-2*pi;
33 end
34 if theta(n+1)<=-pi
35 theta(n+1)=theta(n+1)+2*pi;
36 end
37 thetadif(n+1)=thetain(n)-theta(n);
38 if thetadif(n+1)>pi
39 thetadif(n+1)=thetadif(n+1)-2*pi;
40 end
41 if thetadif(n+1)<-pi
42 thetadif(n+1)=thetadif(n+1)+2*pi;
43 end
44 if abs(thetadif(n+1)-thetadif(n)) < pi
45 difw(cont)=wu-wo;
46 diftheta(cont)=thetadif(n+1);
47 cont=cont+1;
48 else
49 plot(diftheta,difw,colors(u),'LineWidth',2); %plot phase plane
50 cont=1; difw=[]; diftheta=[];
51 end
52 end
53 plot(diftheta,difw,colors(u),'LineWidth',2); %plot phase plane
54 difw=[];
55 diftheta=[];
56 end

Figure A.4: Matlab script to depict SRF-PLL trajectories in the Phase-plane.

180 Chapter A Matlab Scripts

1 clear all;
2 t=1e-4:1e-4:0.1;
3
4 w1n=2*pi*50; % Kalman filter frequency = Nominal frequency.
5 w1=2*pi*50; % Fundamental frequency of the input wave.
6
7 % Sinusoidal input
8 % v=sin(w0i*t-pi/4);
9
10 % Non sinusoidal input (harmonics)
11 v=sin(w1*t-pi/4)+0.11*sin(3*w1*t)+0.08*sin(5*w1*t);
12
13 % Voltage Sag with pi/4 phase jump
14 % v=-sin(w0i*t(1:500));
15 % v=[v -0.5*sin(w0i*t(501:1000)-pi/4)];
16
17 %add white noise to the input
18 % noise=[];
19 % for i=1:1:1e3
20 % noise=[noise;randn*0.05];
21 % end
22 % v=v+noise';
23
24 xhat=[0;0];
25 M=[cos(w1n*1e-4) sin(w1n*1e-4);-sin(w1n*1e-4) cos(w1n*1e-4)];
26 H=[1 0];
27 p=100;
28 P=[p 0; 0 p]; % Set P(0).
29 q=0.001;
30 Q=[q 0; 0 q]; % States covariance Matrix. High q ==> high K.
31 R=5; % Measurement covariance. High R ==> low K.
32 x=[];
33 phase=[];
34 for i=1:1:1e3
35 % Innovation
36 Inn = v(i) - H * xhat;
37 % Covariance of Innovation
38 s = H * P * H' + R;
39 % Gain matrix.
40 K = M * P * H' * inv(s);
41 % State estimate
42 xhat = M * xhat + K * Inn;
43 % Covariance of prediction error
44 P = M * P * M' + Q - M * P * H' * inv(s) * H * P * M';
45 %Save values
46 x=[x xhat];
47 phase=[phase atan2(xhat(1),xhat(2))];
48 end
49 plot(t,v,t,x);
50 figure;
51 plot(t,phase);

Figure A.5: Matlab script of the Kalman Filter Single-phase synchronization example.
 181

1 %% Matlab Script for testing the WLSE synchronization algorithm (3-phase)

2 %% Francisco D. Freijedo, Vigo, December 2008.
3
4 clear all;
5 t=1e-4:1e-4:0.1;
6 w1n=2*pi*50; % Nominal frequency ==> System frequency
7 w1=2*pi*50; % Input wave frequency.
8 % A transient is programmed in the input wave. The distorted signal has
9 % unbalance.
10 Va=1*cos(w1*t(1:499)-pi/6);
11 Vb=1*cos(w1*t(1:499)-2*pi/3-pi/6);
12 Vc=1*cos(w1*t(1:499)+2*pi/3-pi/6);
13 Va=[Va 1*(sin(w1*t(500:1e3)))];
14 Vb=[Vb 0.6*sin(w1*t(500:1e3)-2*pi/3+0.4)];
15 Vc=[Vc 0.9*sin(w1*t(500:1e3)+2*pi/3+0.3)];
16 % The error covariance is initialized.
17 P=300*[1 0 0 0;0 1 0 0;0 0 1 0; 0 0 0 1];
18 lamb=0.95; % Forgetting factor
19 lambinv=1/lamb;
20 ang1vec=[];
21 x=[0 0 0 0]'; % x is v_{dq}^{+-}
22 xvec=[];
23 angvec=[];
24 vab=[];
25 for i=1:1:1e3
26 Vabc=[Va(i) Vb(i) Vc(i)]; % Input wave
27 Talfabeta=(2/3)*[1 -1/2 -1/2;0 -sqrt(3)/2 sqrt(3)/2];
28 Valbe=Talfabeta*Vabc'; % input wave in alfa beta frame
29 vab=[vab Valbe]; % Save alfa beta frame values
30 H=[cos(w1n*t(i)) sin(w1n*t(i)) cos(w1n*t(i)) -sin(w1n*t(i));...
31 -sin(w1n*t(i)) cos(w1n*t(i)) sin(w1n*t(i)) cos(w1n*t(i)) ];
32 r=[1 0; 0 1]+H*P*H'; % compute of R matrix
33 rinv=inv(r);
34 k=P*H'*rinv; % Compute of K gain.
35 P=lambinv*P-lambinv*k*H*P; % Update of error covariance
36 xhat=x+k*[Valbe-H*x]; % Output voltage update
37 xvec=[xvec xhat]; % Save values of SRF coefficients
38 % Compute of phase angle of fundamental positive sequence vector
39 angvec=[angvec atan2(xhat(1)*cos(w1n*t(i))+xhat(2)*+sin(w1n*t(i)) ...
40 ,(-xhat(1)*sin(w1n*t(i))+xhat(2)*cos(w1n*t(i))))];
41 x=xhat; % for next sample...
42 end
43 plot(t,Va,t,Vb,t,Vc);
44 figure;
45 plot(t,vab);
46 figure;
47 plot(t,xvec);
48 figure;
49 plot(t,angvec);
50

Figure A.6: Matlab script of the WLSE syncrhonization example.

182 Chapter A Matlab Scripts

1 clear all;
2 Ts=1e-4; % Sampling time (= 1/fs).
3 Tfinal=0.8; % Time for the simulation.
4 t=0:Ts:Tfinal; % time vector.
5 wn=2*pi*50; % Nominal frequency.
6 Mysin=[0;0]; % Initialize Mysin.
7 Mycos=[0;0.99]; % Initialize Mycos.
8 ypd=[0;0]; % Initialize PD output.
9 ylf=[0;0]; % Initialize LF output.
10 ynotch=[0;0]; % Initialize notch filter output.
11 theta=[0;0]; % Initialize phase angle
12 % Create INPUT WAVE for SIMULATION
13 wu=2*pi*50.3; % Input wave frequency
14 u=sin(wu*t+pi); % Input wave definition
15 % SIMULATE THE PROCESS
16 for n = 2:Tfinal/Ts % Number ot iterations
17 ypd(n+1)=u(n)*Mycos(n); % Phase Detector (Q15)
18 % Notch Filter (Q15)
19 %ynotch(n+1)=ypd(n+1);
20 ynotch(n+1)=1.94*ynotch(n)-0.944*ynotch(n-1)+...
21 0.972*ypd(n+1)-1.94*ypd(n)+0.972*ypd(n-1);
22 % Loop Filter Q(8)
23 ylf(n+1)=(1*ylf(n)+250*ynotch(n+1)-247.8*ynotch(n));
24 % Limit LF according to its Q8 size pipeline
25 ylf(n+1)=max([ylf(n+1) -128]);
26 ylf(n+1)=min([ylf(n+1) 128]);
27 % Update Output frequency and compensate 1/2 (Q6)
28 wo=wn+2*ylf(n+1);
29 % Integration process (digital oscillation)
30 Mysin(n+1)=Mysin(n)+wo*Ts*(Mycos(n)); %(Q15)
31 Mycos(n+1)=Mycos(n)-wo*Ts*(Mysin(n)); %(Q15)
32 % Limit the oscillator integrators
33 Mysin(n+1)=max([Mysin(n+1) -0.99]);
34 Mysin(n+1)=min([Mysin(n+1) 0.99]);
35 Mycos(n+1)=max([Mycos(n+1) -0.99]);
36 Mycos(n+1)=min([Mycos(n+1) 0.99]);
37 % Update the output phase (Q12)
38 theta(n+1)=theta(n)+wo*Ts;
39 % Output phase reset condition
40 if Mysin(n)>=0 && Mysin(n+1)<=0
41 theta(n+1)=-pi;
42 end
43 end
44 plot(t,u,t,Mysin,t,theta); % Plot experimental results

Figure A.7: Matlab script of the single-phase PLL with RC-Oscillator based DCO.
References

[1] J. D. Ainsworth, “The phase-locked oscillator-a new control system for controlled
static convertors,” IEEE Transactions on Power Apparatus and Systems, vol. 87,
no. 3, pp. 859–865, Mar. 1968.

[2] R. Teodorescu, F. Blaabjerg, U. Borup, and M. Liserre, “A new control struc-

ture for grid-connected LCL PV inverters with zero steady-state error and selective
harmonic compensation,” in Applied Power Electronics Conference and Exposition,
2004. APEC ’04. Nineteenth Annual IEEE, vol. 1, 2004, pp. 580–586.

[3] H. Akagi, E. H. Watanabe, and M. Aredes, Instantaneous Power Theory and Ap-
plications to Power Conditioning, M. E. El-Hawari, Ed. Wiley-IEEE Press, 2007.

[4] C. Budeanu, “Puissances reactives et fictives,” in Instytut Romain de l’Energie.

Bucharest, Romania., vol. -, 1927, pp. –.

[5] S. Fryze, “Wirk-, blind- und scheinleistung in elektrischen stromkreisen mit nicht-
sinusförmigem verlauf von strom und spannung,” ETZ, Arch Electrotech, vol. 53,
pp. 596–599, 625–627, 1932.

[6] L. S. Czarnecki, “What is wrong with the budeanu. concept of distorted power and
why it should b e. abandoned ?” IEEE Trans. on Instr. Meas., vol. IM-36, pp.
834–837, 1987.

[7] A. E. Emanuel, “Apparent and reactive powers in three-phase systems: in search of

a physical meaning and a better resolution,” ETEP, Eur. Trans. Elect. Power Eng,
vol. 3, pp. 7–14, 1993.

[8] J. Afonso, C. Couto, and J. Martins, “Active filters with control based on the p-q.
theory,” IEEE Industrial Electronics Society Newsletter, vol. 47, pp. 5–11, 2000.

[9] J. G. Pinto, R. Pregitzer, L. Monteiro, C. Couto, and J. Afonso, “A combined series

active filter and passive filters for harmonics, unbalances and flicker compensation,”
in Proceedings of Power Engineering, Energy and Electrical Drives, 2007.

[10] M. I. M. Montero, E. R. Cadaval, and F. B. Gonzalez, “Comparison of control

strategies for shunt active power filters in three-phase four-wire systems,” IEEE
Transactions on Power Electronics, vol. 22, no. 1, pp. 229–236, Jan. 2007.

183
184 REFERENCES

[11] H. Akagi, Y. Kanazawa, and A. Nabae, “Instantaneous reactive power compensators

comprising switching devices without energy storage components,” IEEE Transac-
tions on Industry Applications, vol. 20, no. 3, pp. 625–630, May 1984.

[12] S. Bhattacharya, D. M. Divan, and B. B. Banerjee, “Synchronous frame harmonic

isolator using active series filter,” in Proceedings of EPE Conference, 1991., Seattle,
WA, USA, Jun. 1991, pp. 779–786.

[13] ——, “Control and reduction of terminal voltage total harmonic distortion(THD)
in a hybrid series active and parallel passive filter system,” in Power Electronics
Specialists Conference, 1993. PESC ’93 Record., 24th Annual IEEE, Seattle, WA,
USA, Jun. 1993, pp. 779–786.

[14] S. Bhattacharya and D. Divan, “Synchronous frame based controller implementation

for a hybrid series active filter system,” in Industry Applications Conference, 1995.
Thirtieth IAS Annual Meeting, IAS ’95., Conference Record of the 1995 IEEE,
vol. 3, Orlando, FL, Oct. 1995, pp. 2531–2540.

[15] ——, “Active filter solutions for utility interface of industrial loads,” in Power Elec-
tronics, Drives and Energy Systems for Industrial Growth, 1996., Proceedings of the
1996 International Conference on, vol. 2, New Delhi, India, Jan. 1996, pp. 1078–
1084.

[16] J. Doval, “Aportaciones a la mejora de la calidad de la red eléctrica. utilización de

DSPs en convertidores electrónicos de potencia en fuente de tensión (in spanish).”
Ph.D. dissertation, University of Vigo, 1999.

[17] J. L. Tepper, J. Dixon, G. Venegas, and L. Moran, “A simple frequency-independent

method for calculating the reactive and harmonic current in a non-linear load,”
IEEE Transactions on Industrial Electronics., vol. 43, pp. 647–653, 1996.

[18] J. Doval-Gandoy, A. Nogueiras, C. M. Penalver, and A. Lago., “Shunt active power

filter with harmonic current control strategy.” in Proceedings of PESC., 1998.

[19] P. C. Tan, P. C. Loh, and D. G. Holmes, “High-performance harmonic extraction

algorithm for a 25 kv traction power quality conditioner,” in Electric Power Appli-
cations, IEE Proceedings -, vol. 151, no. 5, Sep. 2004, pp. 505–512.

[20] C. Fortescue, “Method of symmetrical coordinates applied to the solution of

polyphase netwotks.” Trans. AIEE., vol. 37, pp. 1027–1140, 1918.

[21] G. C. Paap, “Symmetrical components in the time domain and their application
topower network calculations,” IEEE Transactions on Power Systems, vol. 15, no. 2,
pp. 522–528, May 2000.

[22] S.-J. Lee, J.-K. Kang, and S.-K. Sul, “A new phase detecting method for power con-
version systemsconsidering distorted conditions in power system,” in Industry Ap-
plications Conference, 1999. Thirty-Fourth IAS Annual Meeting. Conference Record
of the 1999 IEEE, vol. 4, Phoenix, AZ, USA, 1999, pp. 2167–2172.
REFERENCES 185

[23] E. Clarke, Circuit Analysis of AC Power Systems, ., Ed. Wiley, 1950.

[24] R. Park, “Two reaction theory of synchronous machines,” AIEE Transactions,

vol. 48, pp. 716–730, 1929.

[25] O. Lopez, “Space vector pulse-width modulation for multilevel multiphase voltage-
source converters,” Ph.D. dissertation, Department of Electronic Technology, Uni-
versity of Vigo, 2009.

[26] H.-S. Song and K. Nam, “Dual current control scheme for PWM converter under
unbalancedinput voltage conditions,” IEEE Transactions on Industrial Electronics,
vol. 46, no. 5, pp. 953–959, Oct. 1999.

[27] I. Etxeberria-Otadui, A. Lopez De Heredia, H. Gaztanaga, S. Bacha, and M. Reyero,

“A single synchronous frame hybrid (SSFH) multifrequency controller for power
active filters,” IEEE Transactions on Industrial Electronics, vol. 53, no. 5, pp. 1640–
1648, Oct. 2006.

[28] V. Kaura and V. Blasko, “Operation of a phase locked loop system under distorted
utilityconditions,” IEEE Transactions on Industry Applications, vol. 33, no. 1, pp.
58–63, Jan./Feb. 1997.

[29] SimPowerSystem Toolbox for Matlab/Simulink. [Online]. Available: http:

//www.mathworks.es/

[30] P. Pillay and R. Krishnan, “Control characteristics and speed controller design for a
highperformance permanent magnet synchronous motor drive,” IEEE Transactions
on Power Electronics, vol. 5, no. 2, pp. 151–159, Apr. 1990.

[31] “IEEE Standard 1159-1995. IEEE Recommeded Practice for Monitoring Electric
Power Quality.”

[32] M. H. J. Bollen., Understanding power quality problems. Voltage sags and interru-
pions., R. J. Herrick., Ed. IEEE Press Editorial Board., 2000.

[33] “Ieee recommended practices and requirements for harmonic control in electrical
power systems. ieee std 519-1992.”

[34] K. R. Padiyar, HVDC Power Transmission Systems, ., Ed. New Age International,
1990.

[35] N. G. Hingorani and L. Gyugyi, Understanding FACTS. Concepts and technology

of Flexible AC Transmission Systems, R. J. Herrick, Ed. IEEE, 2000.

[36] A. Garcia-Cerrada, P. Roncero-Sanchez, P. Garcia-Gonzalez, and V. Feliu-Batlle,

“Detailed analysis of closed-loop control of output-voltage harmonics in voltage-
source inverters,” in Electric Power Applications, IEE Proceedings -, vol. 151, no. 6,
Nov. 2004, pp. 734–743.
186 REFERENCES

[37] S. Hiti, D. Boroyevich, and C. Cuadros, “Small-signal modeling and control of three-
phase PWM converters,” in Industry Applications Society Annual Meeting, 1994.,
Conference Record of the 1994 IEEE, Denver, CO, USA, Oct. 1994, pp. 1143–1150.

[38] J. Doval-Gandoy and C. M. Penalver, “Dynamic and steady state analysis of a three
phase buck rectifier,” IEEE Transactions on Power Electronics, vol. 15, no. 6, pp.
953–959, Nov. 2000.

[39] A. I. Maswood and F. Liu, “A unity-power-factor converter using the synchronous-

reference-frame-based hysteresis current control,” IEEE Transactions on Industry
Applications, vol. 43, no. 2, pp. 593–599, March 2007.

[40] B. Wang, G. Venkataramanan, and A. Bendre, “Unity power factor control for
three-phase three-level rectifiers without current sensors,” IEEE Transactions on
Industry Applications, vol. 43, no. 5, pp. 1341–1348, Sep./Oct. 2007.

[41] P. Verdelho and G. D. Marques, “An active power filter and unbalanced current
compensator,” IEEE Transactions on Industrial Electronics, vol. 44, no. 3, pp. 321–
328, Jun. 1997.

[42] ——, “Four-wire current-regulated PWM voltage converter,” IEEE Transactions on

Industrial Electronics, vol. 45, no. 5, pp. 761–770, Oct. 1998.

[43] V. Soares, P. Verdelho, and G. D. Marques, “An instantaneous active and reactive
current component method foractive filters,” IEEE Transactions on Power Elec-
tronics, vol. 15, no. 4, pp. 660–669, Jul. 2000.

[44] D. N. Zmood, “A systematic development of improved linear regulators for sinu-

soidal power converters,” Ph.D. dissertation, Department of Electrical and Com-
puter Systems Engineering, Monash University, Victoria (Australia), 2002.

[45] A. Garcia-Cerrada, O. Pinzon-Ardila, V. Feliu-Batlle, P. Roncero-Sanchez, and

P. Garcia-Gonzalez, “Application of a repetitive controller for a three-phase active
power filter,” IEEE Transactions on Power Electronics, vol. 22, no. 1, pp. 237–246,
Jan. 2007.

[46] C. Zhan, V. K. Ramachandaramurthy, A. Arulampalam, C. Fitzer, S. Kromlidis,

M. Bames, and N. Jenkins, “Dynamic voltage restorer based on voltage-space-vector
PWM control,” IEEE Transactions on Industry Applications, vol. 37, no. 6, pp.
1855–1863, Nov./Dec. 2001.

[47] M. J. Newman, D. G. Holmes, J. G. Nielsen, and F. Blaabjerg, “A dynamic

voltage restorer (DVR) with selective harmonic compensation at medium voltage
level,” IEEE Transactions on Industry Applications, vol. 41, no. 6, pp. 1744–1753,
Nov./Dec. 2005.

[48] M. I. Marei, E. F. El-Saadany, and M. M. A. Salama, “A new approach to control

DVR based on symmetrical components estimation,” IEEE Transactions on Power
Delivery, vol. 22, no. 4, pp. 2017–2024, Oct. 2007.
REFERENCES 187

[49] H. Gaztanaga, I. Etxeberria-Otadui, S. Bacha, and D. Roye, “Fixed-speed wind farm

operation improvement by using DVR devices,” in Industrial Electronics, 2007. ISIE
2007. IEEE International Symposium on, Vigo„ Jun. 2007, pp. 2679–2684.

[50] P. Roncero-Sanchez, E. Acha, J. E. Ortega-Calderon, V. Feliu, and A. Garcia-

Cerrada, “A versatile control scheme for a dynamic voltage restorer for power-quality
improvement,” IEEE Transactions on Power Delivery, vol. 24, no. 1, pp. 277–284,
Jan. 2009.

[51] P. Mattavelli, “Synchronous-frame harmonic control for high-performance AC power

supplies,” IEEE Transactions on Industry Applications, vol. 37, no. 3, pp. 864–872,
May/Jun. 2001.

[52] P. C. Loh, M. J. Newman, D. N. Zmood, and D. G. Holmes, “A comparative analysis

of multiloop voltage regulation strategies for single and three-phase UPS systems,”
IEEE Transactions on Power Electronics, vol. 18, no. 5, pp. 1176–1185, Sep. 2003.

[53] J. M. Guerrero, L. Hang, and J. Uceda, “Control of distributed uninterruptible

power supply systems,” IEEE Transactions on Industrial Electronics, vol. 55, no. 8,
pp. 2845–2859, Aug. 2008.

[54] S. K. Mitra, Digital Signal Processing. A Computer-Based Approach, S. W. Director,

Ed. Mc Graw Hill, 1998.

[55] Rodriguez-Andina, J. J., M. J. Moure, and M. D. Valdes, “Features, design tools,

and application domains of FPGAs,” IEEE Transactions on Industrial Electronics,
vol. 54, no. 4, pp. 1810–1823, Aug. 2007.

[56] E. Monmasson and M. N. Cirstea, “Fpga design methodology for industrial control
systems—a review,” IEEE Transactions on Industrial Electronics, vol. 54, no. 4, pp.
1824–1842, Aug. 2007.

[57] N. R. Zargari and G. Joos, “Performance investigation of a current-controlled

voltage-regulatedPWM rectifier in rotating and stationary frames,” in Industrial
Electronics, Control, and Instrumentation, 1993. Proceedings of the IECON ’93.,
International Conference on, Maui, HI, USA, Nov. 1993, pp. 1193–1198.

[58] ——, “Performance investigation of a current-controlled voltage-regulated PWM

rectifier in rotating and stationary frames,” IEEE Transactions on Industrial Elec-
tronics, vol. 42, pp. 396–401, Nov. 1995.

[59] P. Verdelho and G. D. Marques, “DC voltage control and stability analysis of
PWM-voltage-typereversible rectifiers,” IEEE Transactions on Industrial Electron-
ics, vol. 45, no. 2, pp. 263–273, Apr. 1998.

[60] G. D. Marques and P. Verdelho, “A simple slip-power recovery system with a DC

voltage intermediatecircuit and reduced harmonics on the mains,” IEEE Transac-
tions on Industrial Electronics, vol. 47, no. 1, pp. 123–132, Feb. 2000.
188 REFERENCES

[61] J. Millman and H. Taub, Pulse, Digital and Switching Waveforms, M. G. Hill, Ed.
Mc Graw Hill International, 1965.

[62] “Abb website. hvdc section.” [Online]. Available: http://www.abb.com/hvdc

[63] P. C. Sen, Power electronics, ., Ed. Tata Mc Graw Hill, 1987.

[64] F. Blaabjerg, R. Teodorescu, M. Liserre, and A. V. Timbus, “Overview of control

and grid synchronization for distributed power generation systems,” IEEE Trans-
actions on Industrial Electronics, vol. 53, no. 5, pp. 1398–1409, Oct. 2006.

[65] J. D. Ainsworth, “Harmonic instability between controlled static convertors and a.c.
networks,” Proceedings of the IEE, vol. 114, pp. 949–957, 1967.

[66] B. Singh, V. Verma, and J. Solanki, “Neural network-based selective compensation

of current quality problems in distribution system,” IEEE Transactions on Indus-
trial Electronics, vol. 54, no. 1, pp. 53–60, Feb. 2007.

[67] R. Arockiasamy and S. Doraipandy, “A novel trigger scheme for thyristor operating
under variable-frequency anode supply,” IEEE Transactions on Industrial Electron-
ics and Control Instrumentation, vol. 22, no. 1, pp. 83–85, Feb. 1975.

[68] S. Murugesan and C. K. Rao, “Simple adaptive analog and digital trigger circuits
for thyristors working under wide range of supply frequencies,” IEEE Transactions
on Industrial Electronics and Control Instrumentation, vol. 24, no. 1, pp. 46–49,
Feb. 1977.

[69] R. Weidenbrug, F. P. Dawson, and R. Bonert, “New synchronization method for

thyristor power converters to weakAC-systems,” IEEE Transactions on Industrial
Electronics, vol. 40, no. 5, pp. 505–511, Oct. 1993.

[70] O. Vainio and S. J. Ovaska, “Noise reduction in zero crossing detection by predictive
digitalfiltering,” IEEE Transactions on Industrial Electronics, vol. 42, no. 1, pp. 58–
62, Feb. 1995.

[71] J. Galloway, “Harmonic instability in phase controlled rectifiers,” in Petroleum and

Chemical Industry Conference, 1999. Industry Applications Society 46th Annual,
San Diego, CA, USA, 1999, pp. 171–175.

[72] A. Ekstrom and G. Liss, “A refined HVDC control system,” IEEE Transactions on
Power Apparatus and Systems, vol. 89, no. 5, pp. 723–732, May 1970.

[73] A. Skjellnes, B. Hanssen, and T. Arnulf, “Phase-locked loop control of thyristor

convertors,” Proceedings of the IEE, vol. 123, pp. 999–1001, 1976.

[74] R. Simard and V. Rajagopalan, “Economical equidistant pulse firing scheme for
thyristorized DC drives,” IEEE Transactions on Industrial Electronics and Control
Instrumentation, vol. 22, no. 3, pp. 425–429, Aug. 1975.

[75] G.-C. Hsieh and J. C. Hung, “Phase-locked loop techniques. a survey,” IEEE Trans-
actions on Industrial Electronics, vol. 43, no. 6, pp. 609–615, Dec. 1996.
REFERENCES 189

[76] H. Le-Huy, “A digitally controlled thyristor trigger circuit,” Proceedings of the

IEEE, vol. 66, no. 1, pp. 89–91, Jan. 1978.

[77] B. K. Bose and K. J. Jentzen, “Digital speed control of a DC motor with phase-
locked loop regulation,” IEEE Transactions on Industrial Electronics and Control
Instrumentation, vol. 25, no. 1, pp. 10–13, Feb. 1978.

[78] R. E. Best, Phase Locked Loops. Design, Simulation and Applications. 4th Edition.
McGraw-Hill, 1999.

[79] F. M. Gardner, PhaseLock Techniques. John Wiley and Sons, 2004.

[80] S. C. Gupta, K. Venkatesan, and K. Eapen, “A generalized firing angle controller

using phase-locked loop for thyristor control,” IEEE Transactions on Industrial
Electronics and Control Instrumentation, vol. 28, no. 1, pp. 46–49, Feb. 1981.

[81] F. J. Bourbeau, “LSI based three-phase thyristor firing circuit,” IEEE Transactions
on Industry Applications, vol. 19, no. 4, pp. 571–578, Jul. 1983.

[82] F. M. Gardner, “Charge-pump phase-lock loops,” IEEE Transactions on Commu-

nications, vol. COM-28, pp. 1849–1953, 1980.

[83] C. N. m. Ho, H. S. H. Chung, and K. T. K. Au, “Design and implementation of

a fast dynamic control scheme for capacitor-supported dynamic voltage restorers,”
IEEE Transactions on Power Electronics, vol. 23, no. 1, pp. 237–251, Jan. 2008.

[84] “Electronic components on-line shop Digi-Key.” [Online]. Available: www.digikey.

com/

[85] A. Gole, V. K. Sood, and L. Mootoosamy, “Validation and analysis of a grid control
system using d-q-z transformation for static compensator systems,” in Proc. Can.
Conf. Elect. Comput. Eng., 1989, pp. 745–748.

[86] A. M. Gole and V. K. Sood, “A static compensator model for use with electro-
magnetic transients simulation programs,” IEEE Transactions on Power Delivery,
vol. 5, no. 3, pp. 1398–1407, Jul. 1990.

[87] V. Blasko, J. C. Moreira, and T. A. Lipo, “A new field oriented controller utilizing
spatial positionmeasurement of rotor end ring current,” in Power Electronics Spe-
cialists Conference, 1989. PESC ’89 Record., 20th Annual IEEE, Milwaukee, WI,
USA, Jun. 1989, pp. 295–300.

[88] D. N. Zmood and D. G. Holmes, “Stationary frame current regulation of PWM

inverters with zero steady-state error,” IEEE Transactions on Power Electronics,
vol. 18, no. 3, pp. 814–822, May 2003.

[89] T. Ahmed, K. Nishida, and M. Nakaoka, “Advanced control of PWM converter with
variable-speed induction generator,” IEEE Transactions on Industry Applications,
vol. 42, no. 4, pp. 934–945, Jul./Aug. 2006.
190 REFERENCES

[90] T. Ahmed, K. Nishida, and M. Nakaoka., “A novel stand-alone induction gener-

ator system for AC and DC power applications,” IEEE Transactions on Industry
Applications, vol. 43, no. 6, pp. 1465–1474, Nov./Dec. 2007.

[91] I. Etxeberria-Otadui, U. Viscarret, M. Caballero, A. Rufer, and S. Bacha, “New

optimized PWM VSC control structures and strategies under unbalanced voltage
transients,” IEEE Transactions on Industrial Electronics, vol. 54, no. 5, pp. 2902–
2914, Oct. 2007.

[92] P. Xiao, K. A. Corzine, and G. K. Venayagamoorthy, “Multiple reference frame-

based control of three-phase PWM boost rectifiers under unbalanced and distorted
input conditions,” IEEE Transactions on Power Electronics, vol. 23, no. 4, pp.
2006–2017, Jul. 2008.

[93] P. Mattavelli, “A closed-loop selective harmonic compensation for active filters,”

IEEE Transactions on Industry Applications, vol. 37, no. 1, pp. 81–89, Jan./Feb.
2001.

[94] A. Timbus, “Grid monitoring and advanced control of distributed power generation
systems,” Ph.D. dissertation, Aalborg University, Institute of Energy Technology,
2007.

[95] P. . Rodriguez, A. V. Timbus, R. . Teodorescu, M. . Liserre, and F. . Blaabjerg,

“Flexible active power control of distributed power generation systems during grid
faults,” IEEE Transactions on Industrial Electronics, vol. 54, no. 5, pp. 2583–2592,
Oct. 2007.

[96] G. Escobar, P. Mattavelli, A. Stankovic, A. Valdez, and J. Leyva-Ramos, “An adap-

tive control for UPS to compensate unbalance and harmonic distortion using a com-
bined capacitor/load current sensing,” IEEE Transactions on Industrial Electronics,
vol. 54, no. 2, pp. 839–847, April 2007.

[97] G. Escobar, J. Leyva-Ramos, P. Mattavelli, and A. Valdez, “Repetitive-based con-

troller for a UPS inverter to compensate unbalance and harmonic distortion,” IEEE
Transactions on Industrial Electronics, vol. 54, no. 1, pp. 504–510, Feb. 2007.

[98] S. K. Chung, “Phase-locked loop for grid-connected three-phase power conversion

systems,” in Electric Power Applications, IEE Proceedings -, vol. 147, no. 3, May
2000, pp. 213–219.

[99] S.-K. Chung, “A phase tracking system for three phase utility interface inverters,”
IEEE Transactions on Power Electronics, vol. 15, no. 3, pp. 431–438, May 2000.

[100] C. Zhan, C. Fitzer, V. Ramachandaramurthy, A. Arulampalam, M. Barnes, and

N. Jenkins, “Software phase-locked loop applied to dynamic voltage restorer
(DVR),” in Power Engineering Society Winter Meeting, 2001. IEEE, vol. 3, 28
Jan.-1 Feb. 2001, pp. 1033–1038vol.3.
REFERENCES 191

[101] P. Rodriguez, L. Sainz, and J. Bergas, “Synchronous double reference frame PLL
applied to a unified power quality conditioner,” in Harmonics and Quality of Power,
2002. 10th International Conference on, vol. 2, Oct. 2002, pp. 614–619.

[102] H. Awad, J. Svensson, and M. J. Bollen, “Phase-locked loop for static series com-
pensator,” in Proceedings of EPE Conference, 2003.

[103] ——, “Tuning software phase-locked loop for series-connected converters,” IEEE
Transactions on Power Delivery, vol. 20, no. 1, pp. 300–308, Jan. 2005.

[104] M. C. Benhabib and S. Saadate, “A new robust experimentally validated phase

locked loop for power electronic control,” EPE Journal, vol. 15, pp. 36–48, 2005.

[105] A. Timbus, M. Liserre, R. Teodorescu, and F. Blaabjerg, “Synchronization methods

for three phase distributed power generation systems. an overview and evaluation,”
in Power Electronics Specialists Conference, 2005. PESC ’05. IEEE 36th, Recife„
2005, pp. 2474–2481.

[106] A. V. Timbus, R. Teodorescu, F. Blaabjerg, M. Liserre, and P. Rodriguez, “PLL

algorithm for power generation systems robust to grid voltage faults,” in Power
Electronics Specialists Conference, 2006. PESC ’06. 37th IEEE, Jeju, Jun. 2006,
pp. 1–7.

[107] L. G. B. Barbosa Rolim, D. R. Rodrigues da CostaJr., and M. Aredes, “Analysis

and software implementation of a robust synchronizing PLL circuit based on the pq
theory,” IEEE Transactions on Industrial Electronics, vol. 53, no. 6, pp. 1919–1926,
Dec. 2006.

[108] P. Rodriguez, J. Pou, J. Bergas, J. I. Candela, R. P. Burgos, and D. Boroyevich,

“Decoupled double synchronous reference frame PLL for power converters control,”
IEEE Transactions on Power Electronics, vol. 22, no. 2, pp. 584–592, Mar. 2007.

[109] T. Ostrem, W. Sulkowski, L. Norum, and W. Yadong, “Phase-locked loop with

adaptive signal cancellation for three-phase network side voltage source inverter,”
in Power Electronics and Applications, 2007 European Conference on, Aalborg„
Sep. 2007, pp. 1–10.

[110] A. M. Salamah, S. J. Finney, and B. W. Williams, “Three-phase phase-lock loop

for distorted utilities,” IET Electric Power Applications, vol. 1, no. 6, pp. 937–945,
Nov. 2007.

[111] F. Bradaschia, J. P. Arruda, H. Souza, G. M. S. Azevedo, F. A. S. Neves, and

M. C. Cavalcanti, “A method for extracting the fundamental frequency positive-
sequence voltage vector based on simple mathematical transformations,” in Power
Electronics Specialists Conference, 2008. PESC 2008. IEEE, Rhodes„ Jun. 2008,
pp. 1115–1121.

[112] F. D. Freijedo, J. Doval-Gandoy, O. Lopez, and C. Jacobo, “Robust phase locked

loops optimized for dsp implementation in power quality applications,” in 34th
192 REFERENCES

Annual Conference on IEEE Industrial Electronics, IECON 2008, Orlando, USA,

Nov. 2008.

[113] F. D. Freijedo, J. Doval-Gandoy, O. Lopez, and E. Acha, “Tunning of phase locked

loops for power converters under distorted utility conditions,” in Proceedings of
IEEE Applied Power Electronics Conference, Washinton DC, USA, Nov. 2009.

[114] ——, “Tuning of phase locked loops for power converters under distorted utility
conditions,” To be published in IEEE Transactions on Industry Applications, vol. .,
pp. 1–9, Nov. 2009.

[115] G. Saccomando and J. Svensson, “Transient operation of grid-connected voltage

source converterunder unbalanced voltage conditions,” in Industry Applications
Conference, 2001. Thirty-Sixth IAS Annual Meeting. Conference Record of the 2001
IEEE, vol. 4, Chicago, IL, USA, Sep./Oct. 2001, pp. 2419–2424.

[116] H. Awad and J. Svensson, “Compensation of unbalanced voltage dips using vector-
controlled static series compensator with LC-filter,” in Industry Applications Con-
ference, 2002. 37th IAS Annual Meeting. Conference Record of the, vol. 2, 2002, pp.
904–910.

[117] H. Awad, J. Svensson, and M. J. Bollen, “Tuning software phase-locked loop for
series-connected converters,” IEEE Transactions on Power Delivery, vol. 20, no. 1,
pp. 300–308, Jan. 2005.

[118] X. Yuan, J. Allmeling, W. Merk, and H. Stemmler, “Stationary frame generalized

integrators for current control ofactive power filters with zero steady state error for
current harmonicsof concern under unbalanced and distorted operation conditions,”
in Industry Applications Conference, 2000. Conference Record of the 2000 IEEE,
vol. 4, Rome, Italy, Oct. 2000, pp. 2143–2150.

[119] S. Fukuda and T. Yoda, “A novel current-tracking method for active filters based
on asinusoidal internal model [for PWM invertors],” IEEE Transactions on Industry
Applications, vol. 37, no. 3, pp. 888–895, May/Jun. 2001.

[120] X. Yuan, W. Merk, H. Stemmler, and J. Allmeling, “Stationary-frame generalized

integrators for current control of active power filters with zero steady-state error for
current harmonics of concern under unbalanced and distorted operating conditions,”
IEEE Transactions on Industry Applications, vol. 38, no. 2, pp. 523–532, Mar./Apr.
2002.

[121] S. Fukuda and R. Imamura, “Application of a sinusoidal internal model to current

control of three phase utility-interface-converters,” in Power Electronics Specialist
Conference, 2003. PESC ’03. 2003 IEEE 34th Annual, vol. 3, Jun. 2003, pp. 1301–
1306.

[122] ——, “Application of a sinusoidal internal model to current control of three-phase

utility-interface converters,” IEEE Transactions on Industrial Electronics, vol. 52,
no. 2, pp. 420–426, Apr. 2005.
REFERENCES 193

[123] P. Rodriguez, R. Teodorescu, I. Candela, A. V. Timbus, M. Liserre, and F. Blaab-

jerg, “New positive-sequence voltage detector for grid synchronization of power con-
verters under faulty grid conditions,” in Power Electronics Specialists Conference,
2006. PESC ’06. 37th IEEE, Jun. 2006, pp. 1–7.

[124] R. Cutri and L. Matraska Jr, “A fast instantaneous method for sequence extraction,”
in Ninth Brazilian Conference on Power Electronics, 2007.

[125] G. H. Jung, G. C. Cho, S. W. Hong, and G. H. Cho, “DSP based control of high
power static VAr compensator using novelvector product phase locked loop,” in
Power Electronics Specialists Conference, 1996. PESC ’96 Record., 27th Annual
IEEE, vol. 1, Baveno, Italy, Jun. 1996, pp. 238–243.

[126] D. Jovcic, “Phase locked loop system for FACTS,” IEEE Transactions on Power
Systems, vol. 18, no. 3, pp. 1116–1124, Aug. 2003.

[127] F. D. Freijedo, J. Doval-Gandoy, O. Lopez, D. Pineiro, C. M. Penalver, and A. A.

Nogueiras, “Real-time implementation of a SPLL for FACTS,” in 32nd Annual
Conference on IEEE Industrial Electronics, IECON 2006, Paris, France, Nov. 2006,
pp. 2390–2395.

[128] M. A. Perez, J. R. Espinoza, L. A. Moran, M. A. Torres, and E. A. Araya, “A robust

phase-locked loop algorithm to synchronize static-power converters with polluted
AC systems,” IEEE Transactions on Industrial Electronics, vol. 55, no. 5, pp. 2185–
2192, May 2008.

[129] L. N. Arruda, S. M. Silva, and B. J. C. Filho, “PLL structures for utility con-
nected systems,” in Industry Applications Conference, 2001. Thirty-Sixth IAS An-
nual Meeting. Conference Record of the 2001 IEEE, vol. 4, Chicago, IL, USA,
Sep./Oct. 2001, pp. 2655–2660.

[130] S. M. Silva, B. M. Lopes, B. J. C. Filho, R. P. Campana, and W. C. Bosven-

tura, “Performance evaluation of PLL algorithms for single-phase grid-connected
systems,” in Industry Applications Conference, 2004. 39th IAS Annual Meeting.
Conference Record of the 2004 IEEE, vol. 4, Oct. 2004, pp. 2259–2263.

[131] M. Ciobotaru, R. Teodorescu, and F. Blaabjerg, “A new single-phase PLL struc-

ture based on second order generalized integrator,” in Power Electronics Specialists
Conference, 2006. PESC ’06. 37th IEEE, Jun. 2006, pp. 1–6.

[132] M. Ciobotaru, R. Teodorescu, and V. G. Agelidis, “Offset rejection for PLL based
synchronization in grid-connected converters,” in Applied Power Electronics Con-
ference and Exposition, 2008. APEC 2008. Twenty-Third Annual IEEE, Austin,
TX„ Feb. 2008, pp. 1611–1617.

[133] S. Shinnaka, “A robust single-phase PLL system with stable and fast tracking,”
IEEE Transactions on Industry Applications, vol. 44, no. 2, pp. 624–633, Mar./
2008.
194 REFERENCES

[134] M. R. Iravani and M. Karimi-Ghartemani, “Online estimation of steady state and

instantaneous symmetrical components,” in Generation, Transmission and Distri-
bution, IEE Proceedings-, vol. 150, no. 5, Sep. 2003, pp. 616–622.
[135] C. H. da Silva, R. R. Pereira, L. E. B. da Silva, G. Lambert-Torres, and V. F.
Silva, “DSP implementation of three-phase PLL using modified synchronous ref-
erence frame,” in Industrial Electronics Society, 2007. IECON 2007. 33rd Annual
Conference of the IEEE, Taipei„ Nov. 2007, pp. 1697–1701.
[136] M. Karimi-Ghartemani, H. Karimi, and M. R. Iravani, “A magnitude/phase-locked
loop system based on estimation of frequency and in-phase/quadrature-phase am-
plitudes,” IEEE Transactions on Industrial Electronics, vol. 51, no. 2, pp. 511–517,
Apr. 2004.
[137] S. Muyulema, E. J. Bueno, F. J. Rodriguez, S. Cobreces, and D. Diaz, “Response
of the grid converters synchronization using p.u. magnitude in the control loop,” in
Industrial Electronics, 2007. ISIE 2007. IEEE International Symposium on, Jun.
2007, pp. 186–191.
[138] A. viterbi, Principles of Coherent Communications. McGraw-Hill Inc.,US, 1967.
[139] J.-J. E. Slotine and W. Li, Applied Nonlinear control, ., Ed. Prentice Hall, 1991.
[140] H. S. Song and K. Nam, “Instantaneous phase-angle estimation algorithm under
unbalancedvoltage-sag conditions,” in Generation, Transmission and Distribution,
IEE Proceedings-, vol. 147, no. 6, Nov. 2000, pp. 409–415.
[141] J. Svensson, “Synchronisation methods for grid-connected voltage sourceconvert-
ers,” in Generation, Transmission and Distribution, IEE Proceedings-, vol. 148,
no. 3, May 2001, pp. 229–235.
[142] “Matlab help.” [Online]. Available: http://www.mathworks.com/access/helpdesk/
help/helpdesk.html
[143] R. F. de Camargo, A. T. Pereira, and H. Pinheiro, “New synchronization method
for three-phase three-wire PWM converters under unbalance and harmonics in the
grid voltages,” in Power Electronics Specialists Conference, 2005. PESC ’05. IEEE
36th, Recife„ 2005, pp. 506–512.
[144] R. F. de Camargo and H. Pinheiro, “Synchronisation method for three-phase PWM
converters under unbalanced and distorted grid,” in Electric Power Applications,
IEE Proceedings -, vol. 153, no. 5, Sep. 2006, pp. 763–772.
[145] V. M. Moreno, M. Liserre, A. Pigazo, and A. Dell’Aquila, “A comparative analy-
sis of real-time algorithms for power signal decomposition in multiple synchronous
reference frames,” IEEE Transactions on Power Electronics, vol. 22, no. 4, pp.
1280–1289, Jul. 2007.
[146] F. Freijedo, J. Doval-Gandoy, O. Lopez, and C. Martinez-Penalver, “New algorithm
for grid synchronization based on fourier series,” in EPE Conference., Aalborg,
Denmark., Sep. 2007.
REFERENCES 195

[147] F. D. Freijedo, J. Doval-Gandoy, O. Lopez, and E. Acha, “A generic open-loop algo-

rithm for three-phase grid voltage/current synchronization with particular reference
to phase, frequency, and amplitude estimation,” IEEE Transactions on Power Elec-
tronics, vol. 24, no. 1, pp. 94–107, Jan. 2009.

[148] F. D. Freijedo, J. Doval, O. Lopez, and C. Martinez-Penalver, “A signal processing

adaptive algorithm for selective current harmonic cancellation in active power fil-
ters,” IEEE Transactions on Industrial Electronics, vol. forthcoming papers, p. 9,
2009.

[149] F. D. Freijedo, J. Doval-Gandoy, O. Lopez, C. Martinez-Penalver, A. Nogueiras,

and E. Acha, “Novel harmonic identification algorithm based on fourier correlation
and moving average filtering,” in IEEE Power Engineering Society General Meeting,
Anaheim, California, USA, Oct. 2007.

[150] F. D. Freijedo, J. Doval-Gandoy, O. Lopez, and J. Cabaleiro, “Harmonic identifica-

tion methods based on moving average filters for active power filters,” in Industry
Applications Society Annual Meeting, 2008. IAS ’08. IEEE, Edmonton, Alberta,
Canada„ Oct. 2008.

[151] F. D. Freijedo, J. Doval-Gandoy, O. Lopez, and C. Jacobo, “Reference generation

techniques for active power line conditioners,” in 34th Annual Conference on IEEE
Industrial Electronics, IECON 2008, Orlando, USA, Nov. 2008.

[152] S. K. Mitra, Digital Signal Processing. A Computer-Based Approach, S. W. Director,

Ed. Mc Graw Hill, 1998.

[153] R. W. Menzies and G. B. Mazur, “Advances in the determination of control param-

eters for static compensators,” IEEE Power Engineering Review, vol. 9, no. 10, pp.
44–44, Oct. 1989.

[154] A. Nakata, A. Ueda, and A. Torii, “A method of current detection for an active
power filter applyingmoving average to pq-theory,” in Power Electronics Specialists
Conference, 1998. PESC 98 Record. 29th Annual IEEE, vol. 1, Fukuoka, Japan,
May 1998, pp. 242–247.

[155] Power Systems Blockset, User Manual, Mathworks, 2000.

[156] M. Albu and G. Heydt, “On the use of RMS values in power quality assessment,”
IEEE Transactions on Power Delivery, vol. 18, no. 4, pp. 1586–1587, Oct. 2003.

[157] H. Awad, H. Nelsen, F. Blaabjerg, and M. J. Newman, “Operation of static se-

ries compensator under distorted utility conditions,” IEEE Transactions on Power
Systems, vol. 20, no. 1, pp. 448–457, Feb. 2005.

[158] B. P. McGrath, D. G. Holmes, and J. J. H. Galloway, “Power converter line syn-

chronization using a discrete fourier transform (DFT) based on a variable sample
rate,” IEEE Transactions on Power Electronics, vol. 20, no. 4, pp. 877–884, Jul.
2005.
196 REFERENCES

[159] S. W. Smith, The Scientist and Engineer’s Guide to Digital Signal Processing., .,
Ed. California Technical Publishing, 1998.

[160] E. Acha and M. Madrigal, Power Systems Harmonics, J. Willey, Ed. Jonh Willey
& Sons, Ltd, 2001.

[161] R. Grino, R. Cardoner, R. Costa-Castello, and E. Fossas, “Digital repetitive control

of a three-phase four-wire shunt active filter,” IEEE Transactions on Industrial
Electronics, vol. 54, no. 3, pp. 1495–1503, June 2007.

[162] L. Asiminoaei, F. Blaabjerg, and S. Hansen, “Evaluation of harmonic detection

methods for active power filter applications.” in Proceedings of APEC, 2005.

[163] R. Costa-Castello, R. Grino, and E. Fossas, “Odd-harmonic digital repetitive control

of a single-phase current active filter,” IEEE Transactions on Power Electronics,
vol. 19, no. 4, pp. 1060–1068, Jul. 2004.

[164] ——, “Reply to "concerning ’odd-harmonic digital repetitive control of a single-

phase current active filter’",” IEEE Transactions on Power Electronics, vol. 21,
no. 4, pp. 1159–1160, Jul. 2006.

[165] K. Zhou, K.-S. Low, D. Wang, F.-L. Luo, B. Zhang, and Y. Wang, “Zero-phase odd-
harmonic repetitive controller for a single-phase PWM inverter,” IEEE Transactions
on Power Electronics, vol. 21, no. 1, pp. 193–201, Jan. 2006.

[166] M. D. Kusljevic, “A simple method for design of adaptive filters for sinusoidal
signals,” IEEE Transactions on Instrumentation and Measurement, vol. 57, no. 10,
pp. 2242–2249, Oct. 2008.

[167] L.-P. Chau and W.-C. Siu, “Direct formulation for the realization of discrete cosine
transformusing recursive structure,” IEEE Transactions on Circuits and Systems
II: Analog and Digital Signal Processing, vol. 42, no. 1, pp. 50–52, Jan. 1995.

[168] J.-F. Yang and C.-P. Fan, “Recursive discrete cosine transforms with selectablefixed-
coefficient filters,” IEEE Transactions on Circuits and Systems II: Analog and Dig-
ital Signal Processing, vol. 46, no. 2, pp. 211–216, Feb. 1999.

[169] ——, “Compact recursive structures for discrete cosine transform,” IEEE Trans-
actions on Circuits and Systems II: Analog and Digital Signal Processing, vol. 47,
no. 4, pp. 314–321, Apr. 2000.

[170] C.-H. Chen, B.-D. Liu, J.-F. Yang, and J.-L. Wang, “Efficient recursive structures
for forward and inverse discrete cosine transform,” IEEE Transactions on Signal
Processing, vol. 52, no. 9, pp. 2665–2669, Sep. 2004.

[171] R. E. Kalman, “A new approach to linear filtering and prediction problems,” Trans-
actions on ASME, Journal of Basic Engineering, vol. 82, pp. 33–45, 1960.

[172] B. Widrow and M. E. Hoff, “Adaptive switching circuits,” in IRE WESCON Con-
vention Records, 1960.
REFERENCES 197

[173] H. Ma and A. A. Girgis, “Identification and tracking of harmonic sources in a power

systemusing a kalman filter,” IEEE Transactions on Power Delivery, vol. 11, no. 3,
pp. 1659–1665, Jul. 1996.

[174] P. K. Dash, D. P. Swain, H. P. Khincha, and A. C. Liew, “Digital protective relay-

ing using an adaptive neural network,” in Energy Management and Power Delivery,
1995. Proceedings of EMPD ’95., 1995 International Conference on, vol. 2, , Singa-
pore, Nov. 1995, pp. 684–689.

[175] P. K. Dash, D. P. Swain, A. C. Liew, and S. Rahman, “An adaptive linear com-
biner for on-line tracking of power systemharmonics,” IEEE Transactions on Power
Systems, vol. 11, no. 4, pp. 1730–1735, Nov. 1996.

[176] H.-S. Song, H.-G. Park, and K. Nam, “An instantaneous phase angle detection
algorithm under unbalancedline voltage condition,” in Power Electronics Specialists
Conference, 1999. PESC 99. 30th Annual IEEE, vol. 1, Charleston, SC, USA, Aug.
1999, pp. 533–537.

[177] H.-S. Song, K. Nam, and P. Mutschler, “Very fast phase angle estimation algorithm
for a single-phase system having sudden phase angle jumps,” in Industry Applica-
tions Conference, 2002. 37th IAS Annual Meeting. Conference Record of the, vol. 2,
2002, pp. 925–931.

[178] H.-S. Song, R. Keil, P. Mutschler, J. van der Weem, and K. Nam, “Advanced con-
trol scheme for a single-phase PWM rectifier in traction applications,” in Industry
Applications Conference, 2003. 38th IAS Annual Meeting. Conference Record of the,
vol. 3, Oct. 2003, pp. 1558–1565.

[179] M. I. Marei, E. F. El-Saadany, and M. M. A. Salama, “A processing unit for symmet-

rical components and harmonics estimation based on a new adaptive linear combiner
structure,” IEEE Transactions on Power Delivery, vol. 19, no. 3, pp. 1245–1252,
Jul. 2004.

[180] B. Han and B. Bae, “Novel phase-locked loop using adaptive linear combiner,” IEEE
Transactions on Power Delivery, vol. 21, no. 1, pp. 513–514, Jan. 2006.

[181] R. Cardoso, R. F. de Camargo, H. Pinheiro, and H. A. Grundling, “Kalman filter

based synchronisation methods,” Generation, Transmission & Distribution, IET,
vol. 2, no. 4, pp. 542–555, Jul. 2008.

[182] J. F. Petit, “Control de filtros activos de potencia para la mitigacion de armonicos

y mejora del factor de potencia en sistemas desequilibrados.” Ph.D. dissertation,
Departamento de Ingenieria Electrica, Universidad Carlos III de Madrid, 2007.

[183] A. A. Girgis, W. B. Chang, and E. B. Makram, “A digital recursive measurement

scheme for online tracking of powersystem harmonics,” IEEE Transactions on Power
Delivery, vol. 6, no. 3, pp. 1153–1160, Jul. 1991.
198 REFERENCES

[184] R. Alcaraz, E. J. Bueno, S. Cobreces, F. J. Rodriguez, F. Espinosa, and

S. Muyulema, “Power system voltage harmonic identification using kalman filter,”
in EPE-PEMC, 2006.

[185] “UNE-EN50160 Standard.”

[186] R. Alcaraz, E. J. Bueno, S. Cobreces, F. J. Rodriguez, C. Giron, and M. Liserre,

“Comparison of voltage harmonic identification methods for single-phase systems,”
in Proceedings of IECON, 2006.

[187] M. Bongiorno, J. Svensson, and A. Sannino, “Effect of sampling frequency and

harmonics on delay-based phase-sequence estimation method,” IEEE Transactions
on Power Delivery, vol. 23, no. 3, pp. 1664–1672, Jul. 2008.

[188] J. Svensson, M. Bongiorno, and A. Sannino, “Practical implementation of delayed

signal cancellation method for phase-sequence separation,” IEEE Transactions on
Power Delivery, vol. 22, no. 1, pp. 18–26, Jan. 2007.

[189] A. K. Pradhan, A. Routray, and A. Basak, “Power system frequency estimation

using least mean square technique,” IEEE Transactions on Power Delivery, vol. 20,
no. 3, pp. 1812–1816, Jul. 2005.

[190] C.-H. Huang, C.-H. Lee, K.-J. Shih, and Y.-J. Wang, “Frequency estimation of
distorted power system signals using a robust algorithm,” IEEE Transactions on
Power Delivery, vol. 23, no. 1, pp. 41–51, Jan. 2008.

[191] G. F. Franklin, J. D. Powell, and A. Emami-Naeini, Feedback Control of Dynamic

Systems. Pearson Education, 2002.

[192] R. C. Dorf and R. H. Bishop, Modern Control Systems, Ninth Edition, ., Ed. Pren-
tice Hall, 2000.

[193] ., “Digital sine-wave generation using the tms320c2xx. spra179,” Texas Instruments,
Tech. Rep., 1998.

[194] D. Garcia, “Precision digital sine-wave generation with the tms32010. application
report spra007.” Texas Instruments., Tech. Rep., 1997.

[195] Y.-T. Cheng, “Tms320c62x algorithm: Sine wave generation. spra708.” Texas In-
struments, Tech. Rep., 2000.

[196] G. Gonzalez, Foundations of Oscillator Circuit Design, A. H. Publishers, Ed.

Artech House Publishers, 2007.

[197] Y. W. Li, F. Blaabjerg, D. M. Vilathgamuwa, and P. C. Loh, “Design and com-

parison of high performance stationary-frame controllers for DVR implementation,”
IEEE Transactions on Power Electronics, vol. 22, no. 2, pp. 602–612, Mar. 2007.

[198] H. Akagi, S. Inoue, and T. Yoshii, “Control and performance of a transformerless

cascade PWM STATCOM with star configuration,” IEEE Transactions on Industry
Applications, vol. 43, no. 4, pp. 1041–1049, Jul./Aug. 2007.
REFERENCES 199

[199] C. Lascu, L. Asiminoaei, I. Boldea, and F. Blaabjerg, “High performance current

controller for selective harmonic compensation in active power filters,” IEEE Trans-
actions on Power Electronics, vol. 22, no. 5, pp. 1826–1835, Sep. 2007.

[200] M. Liserre, R. Teodorescu, and F. Blaabjerg, “Multiple harmonics control for three-
phase grid converter systems with the use of PI-RES current controller in a rotating
frame,” IEEE Transactions on Power Electronics, vol. 21, no. 3, pp. 836–841, May
2006.

[201] C. A. Busada, H. G. Chiacchiarini, and J. C. Balda, “Synthesis of sinusoidal wave-

form references synchronized with periodic signals,” IEEE Transactions on Power
Electronics, vol. 23, no. 2, pp. 581–590, Mar. 2008.

[202] R. Wu, S. B. Dewan, and G. R. Slemon, “A PWM AC-to-DC converter with fixed
switching frequency,” IEEE Transactions on Industry Applications, vol. 26, no. 5,
pp. 880–885, Sep./Oct. 1990.

[203] T. G. Habetler, “A space vector-based rectifier regulator for AC/DC/AC convert-

ers,” IEEE Transactions on Power Electronics, vol. 8, no. 1, pp. 30–36, Jan. 1993.

[204] D. Holmes and D. Matin, “Implementation of a direct digital predictive current

controller for single and three phase voltage source inverters.” in Proceedings of
IAS., 1996.

[205] Y. A. R. I. Mohamed and E. F. El-Saadany, “Robust high bandwidth discrete-time

predictive current control with predictive internal model—a unified approach for
voltage-source PWM converters,” IEEE Transactions on Power Electronics, vol. 23,
no. 1, pp. 126–136, Jan. 2008.

[206] P. Cortes, J. Rodriguez, D. E. Quevedo, and C. Silva, “Predictive current con-

trol strategy with imposed load current spectrum,” IEEE Transactions on Power
Electronics, vol. 23, no. 2, pp. 612–618, Mar. 2008.

[207] S.-J. Jeong and S.-H. Song, “Improvement of predictive current control performance
using online parameter estimation in phase controlled rectifier,” IEEE Transactions
on Power Electronics, vol. 22, no. 5, pp. 1820–1825, Sep. 2007.

[208] G. Abad, M. . Rodriguez, and J. Poza, “Two-level VSC based predictive direct
torque control of the doubly fed induction machine with reduced torque and flux
ripples at low constant switching frequency,” IEEE Transactions on Power Elec-
tronics, vol. 23, no. 3, pp. 1050–1061, May 2008.

[209] P. Fernandez-Comesaña, F. D. Freijedo, J. Doval-Gandoy, and J. Malvar., “Con-

trol algorithm for a SSSC,” in Proceedings of Compatibilty and Power Eletronics
Conference, Badajoz, Spain, Nov. 2009.

[210] C. Lascu, L. Asiminoaei, I. Boldea, and F. Blaabjerg, “Frequency response analysis

of current controllers for selective harmonic compensation in active power filters,”
IEEE Transactions on Industrial Electronics, vol. 56, no. 2, pp. 337–347, Feb. 2009.
200 REFERENCES

[211] M. S. Carmeli, F. Castelli-Dezza, and G. Superti-Furga, “Generalized decoupling

method for current-controlled multiswitching systems,” IEEE Transactions on In-
dustrial Electronics, vol. 56, no. 2, pp. 348–359, Feb. 2009.

[212] A. Gensior, H. Sira-Ramirez, J. Rudolph, and H. Guldner, “On some nonlinear

current controllers for three-phase boost rectifiers,” IEEE Transactions on Industrial
Electronics, vol. 56, no. 2, pp. 360–370, Feb. 2009.

[213] M. H. Bierhoff and F. W. Fuchs, “Active damping for three-phase PWM recti-
fiers with high-order line-side filters,” IEEE Transactions on Industrial Electronics,
vol. 56, no. 2, pp. 371–379, Feb. 2009.

[214] J. Dannehl, C. Wessels, and F. W. Fuchs, “Limitations of voltage-oriented PI current

control of grid-connected PWM rectifiers with LCL filters,” IEEE Transactions on
Industrial Electronics, vol. 56, no. 2, pp. 380–388, Feb. 2009.

[215] P. Lezana, J. Rodriguez, M. A. Perez, and J. Espinoza, “Input current harmon-

ics in a regenerative multicell inverter with single-phase PWM rectifiers,” IEEE
Transactions on Industrial Electronics, vol. 56, no. 2, pp. 408–417, Feb. 2009.

[216] A. Luo, Z. Shuai, W. Zhu, and Z. J. Shen, “Combined system for harmonic sup-
pression and reactive power compensation,” IEEE Transactions on Industrial Elec-
tronics, vol. 56, no. 2, pp. 418–428, Feb. 2009.

[217] J. Hu, Y. He, L. Xu, and B. W. Williams, “Improved control of DFIG systems
during network unbalance using PI–r current regulators,” IEEE Transactions on
Industrial Electronics, vol. 56, no. 2, pp. 439–451, Feb. 2009.

[218] T. R., Blaabjerg, and L. M., “Proportional-resonant controllers: A new breed of

controllers suitable for grid-connected voltage-source converters,” in Proceedings
of The 9th International Conference on Optimization of Electrical and Electronic
Equipments, OPTIM, vol. 1, 2004.

[219] R. A. Mastromauro, M. Liserre, and A. Dell’Aquila, “Study of the effects of inductor

nonlinear behavior on the performance of current controllers for single-phase PV
grid converters,” IEEE Transactions on Industrial Electronics, vol. 55, no. 5, pp.
2043–2052, May 2008.

[220] P. Lezana, C. A. Silva, J. Rodriguez, and M. A. Perez, “Zero-steady-state-error

input-current controller for regenerative multilevel converters based on single-phase
cells,” IEEE Transactions on Industrial Electronics, vol. 54, no. 2, pp. 733–740,
Apr. 2007.

[221] D. N. Zmood, D. G. Holmes, and G. H. Bode, “Frequency-domain analysis of

three-phase linear current regulators,” IEEE Transactions on Industry Applications,
vol. 37, no. 2, pp. 601–610, Mar./Apr. 2001.

[222] M. P. Kazmierkowski and L. Malesani, “Current control techniques for three-phase

voltage-source PWM converters: a survey,” IEEE Transactions on Industrial Elec-
tronics, vol. 45, no. 5, pp. 691–703, Oct. 1998.
REFERENCES 201

[223] P. Mattavelli and F. P. Marafao, “Repetitive-based control for selective harmonic

compensation in active power filters,” IEEE Transactions on Industrial Electronics,
vol. 51, no. 5, pp. 1018–1024, Oct. 2004.

[224] M. Bollen and I. Gu, Signal Processing of Power Quality Distrurbances, M. E. El-
Hawary, Ed. John Wiley and Sons, 2006.

[225] X. Guillaud, P. Degobert, and R. Teodorescu, “Use of resonant controller for grid-
connected converters in case of large frequency fluctuations,” in European Confer-
ence on Power Electronics and Applications, 2007, Aalborg, Sep. 2007, pp. 1–8.

[226] M. Sedighy, S. Dewan, and F. Dawson, “A robust digital current control method for
active power filters,” IEEE Transactions on Industry Applications, vol. 36, no. 4,
pp. 1158–1164, July-Aug. 2000.