This material is only for your personal use.

You are NOT allowed

to deliver the material to third parties in electronic form.

The pages covered in the course will be given in the MyCourses portal.
Feedback on this material is welcome!

Control of Voltage-Source Converters

and Variable-Speed Drives

Lennart Harnefors Marko Hinkkanen

Oskar Wallmark Alejandro Gómez Yepes

 1

Preface
2

.
Contents

1 Introduction 9
1.1 Electric Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.1 Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.2 Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.1.3 Transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.1.4 Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.5 Consumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Electrical Drives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.1 Benefits of Electrical Drives . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2.2 Motor Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2.3 Variable-Speed Drives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3 Power Electronic Converters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.1 DC–DC Converters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.2 AC–DC Converters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.3 DC–AC Converters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.4 AC–AC Converters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.4 Challenges in Control of Converters and Drives . . . . . . . . . . . . . . . . . . . . 20

2 Systems and Circuits in Power Electronics and Drives 23

2.1 Single-Phase AC Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.1.1 Sinusoids and Phasors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.1.2 Active and Reactive Powers . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.1.3 Complex Impedances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.4 Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2 Linear Continuous-Time Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.1 Definition of Linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.2 Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.3 Transfer Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.4 Relation Between Poles and System Dynamics . . . . . . . . . . . . . . . . 31
2.2.5 Determination of Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2.6 Frequency Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.7 Rise Time and Bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.2.8 Damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.9 Differential Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.10 State-Space Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3 Nonlinear Continuous-Time Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3.1 Global and Local Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.3.2 Analysis of Local Stability by Linearization . . . . . . . . . . . . . . . . . . 40
2.4 Linear Discrete-Time Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4 CONTENTS

2.4.1 Sampling Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

2.4.2 Sampling Theorem and Prefiltering . . . . . . . . . . . . . . . . . . . . . . . 43
2.4.3 Z Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.4.4 Difference Equations and Transfer Functions . . . . . . . . . . . . . . . . . 45
2.4.5 Relation Between Poles and System Dynamics . . . . . . . . . . . . . . . . 46
2.4.6 Shift Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.4.7 Euler Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.5 Three-Phase Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.5.1 Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.5.2 Instantaneous Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.5.3 Wye (Y) and Delta (∆) Connections . . . . . . . . . . . . . . . . . . . . . . 53
2.5.4 Multifrequency Waveforms With Balanced Harmonic Content . . . . . . . . 55
2.6 Space Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.6.1 Two-Phase Representation With Complex Space Vectors . . . . . . . . . . . 56
2.6.2 Vector Representation of Positive- and Negative-Sequence Components . . 57
2.6.3 Relations Between Complex Space Vectors and Rotating Phasors . . . . . . 58
2.6.4 Real Space Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.6.5 Complex Gains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.6.6 Coordinate Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.6.7 dq Transformation of Multifrequency Waveforms . . . . . . . . . . . . . . . 63
2.6.8 Coordinate Transformations of Real Space Vectors . . . . . . . . . . . . . . 64
2.6.9 Instantaneous Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.6.10 Power of Multifrequency Waveforms . . . . . . . . . . . . . . . . . . . . . . 66
2.6.11 Selection of the Space-Vector Scaling Constant . . . . . . . . . . . . . . . . 67
2.7 Symmetrical Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.7.1 Complex System Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.7.2 Complex Transfer Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.7.3 Coordinate Transformations of Symmetrical Systems . . . . . . . . . . . . . 73
2.7.4 Discrete-Time Symmetrical Systems . . . . . . . . . . . . . . . . . . . . . . 75
2.8 Normalized (Per-Unit) Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.8.1 Single-Phase and DC Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.8.2 Three-Phase Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

3 Voltage-Source Converters 81
3.1 Buck Converters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.1.1 Switch-Mode Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.1.2 Load-Current Ripple . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.1.3 Hysteresis Current Control . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.2 Single-Phase Voltage-Source Converters . . . . . . . . . . . . . . . . . . . . . . . . 85
3.2.1 Pulsewidth Modulation: Bipolar Switching . . . . . . . . . . . . . . . . . . 85
3.2.2 Suboscillation Method: Bipolar Switching . . . . . . . . . . . . . . . . . . . 87
3.2.3 Pulsewidth Modulation: Unipolar Switching . . . . . . . . . . . . . . . . . . 89
3.2.4 Frequency Spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.2.5 Digital Implementation Using Synchronous Sampling . . . . . . . . . . . . . 90
3.3 Three-Phase Voltage-Source Converters . . . . . . . . . . . . . . . . . . . . . . . . 93
3.3.1 Suboscillation Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.3.2 Symmetrical Suboscillation Method . . . . . . . . . . . . . . . . . . . . . . 97
CONTENTS 5

3.3.3 Prevention of Overmodulation . . . . . . . . . . . . . . . . . . . . . . . . . 98

3.3.4 Digital Implementation Using Synchronous Sampling . . . . . . . . . . . . . 100
3.3.5 Switching Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

4 DC Motor Drives 105

4.1 Mechanics for Rotational Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.2 DC Machine Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.2.1 Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.2.2 Electrical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.2.3 Dynamic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.2.4 Magnetization by Field Winding . . . . . . . . . . . . . . . . . . . . . . . . 112
4.2.5 Rating and Base Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.2.6 Normalization of the DC Machine Model . . . . . . . . . . . . . . . . . . . 114
4.3 Open-Loop Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.4 Current Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.4.1 One-Degree-of-Freedom Controller Design . . . . . . . . . . . . . . . . . . . 120
4.4.2 Load-Disturbance Rejection . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.4.3 Feedforward of the Back EMF . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.4.4 Two-Degrees-of-Freedom Controller Design . . . . . . . . . . . . . . . . . . 123
4.4.5 Voltage Saturation and Antiwindup . . . . . . . . . . . . . . . . . . . . . . 125
4.4.6 Digital Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.4.7 Bandwidth Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
4.5 Field Weakening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.6 Speed Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.6.1 Speed Estimation for Sensorless Control . . . . . . . . . . . . . . . . . . . . 133

5 AC Motor Drives 135

5.1 AC Machine Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.1.1 Synchronous Machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.1.2 Induction Machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.2 Dynamic Model for the Induction Motor . . . . . . . . . . . . . . . . . . . . . . . . 142
5.2.1 Inverse-Γ Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
5.2.2 Transformation to Synchronous Coordinates . . . . . . . . . . . . . . . . . . 148
5.2.3 Perfect Field Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
5.2.4 Torque Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
5.2.5 Steady-State Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.2.6 Field Weakening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.2.7 Motor Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5.2.8 Nominal Flux and Torque . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5.2.9 Steady-State Relations for Line-Connected IMs . . . . . . . . . . . . . . . . 153
5.2.10 Transient Impedance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.2.11 Starting Dynamics of Line-Connected IMs . . . . . . . . . . . . . . . . . . . 158
5.2.12 Volts-per-Hertz Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
5.3 Dynamic Model for the Permanent-Magnet Synchronous Motor . . . . . . . . . . . 162
5.3.1 Transformation to Synchronous Coordinates . . . . . . . . . . . . . . . . . . 163
5.3.2 Motor Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
5.3.3 Torque Production and Control Principle . . . . . . . . . . . . . . . . . . . 164
6 CONTENTS

5.3.4 Rotor Flux and Base Frequency . . . . . . . . . . . . . . . . . . . . . . . . . 165

5.3.5 Field Weakening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.3.6 Transient Impedance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
5.3.7 Power Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
5.3.8 Salient PMSMs and Reluctance SMs . . . . . . . . . . . . . . . . . . . . . . 167
5.4 Unified IM/PMSM Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
5.5 Mechanical Dynamics and Speed Control . . . . . . . . . . . . . . . . . . . . . . . 171
5.5.1 Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
5.6 Determination of Motor Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5.6.1 Induction Motors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5.6.2 Permanent-Magnet Synchronous Motors . . . . . . . . . . . . . . . . . . . . 176
5.6.3 Parameter Variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

6 Vector Control Fundamentals 179

6.1 Principle of Vector Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.1.1 Transformation to Synchronous Coordinates . . . . . . . . . . . . . . . . . . 180
6.2 Introduction to Vector Current Control . . . . . . . . . . . . . . . . . . . . . . . . 181
6.3 Review of Methods for Three-Phase Current Control . . . . . . . . . . . . . . . . . 182
6.3.1 Hysteresis Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
6.3.2 Stationary-Frame PI Control . . . . . . . . . . . . . . . . . . . . . . . . . . 183
6.3.3 Synchronous-Frame PI Control . . . . . . . . . . . . . . . . . . . . . . . . . 184
6.4 Design of Synchronous-Frame Current Controllers . . . . . . . . . . . . . . . . . . 185
6.4.1 Bandwidth Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
6.4.2 PWM, Time-Delay Compensation, and Antiwindup . . . . . . . . . . . . . 187
6.5 Synchronization Control Using a Phase-Locked Loop . . . . . . . . . . . . . . . . . 189
6.5.1 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192

7 Vector Control of Grid-Connected VSCs 195

8 Vector Control of PMSM Drives 197

8.1 Position-Sensored Control Using Resolvers . . . . . . . . . . . . . . . . . . . . . . . 197
8.1.1 Excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
8.1.2 Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
8.2 Position-Sensorless Control Using Back-EMF Information . . . . . . . . . . . . . . 200
8.2.1 PLL for Sensorless PMSM Drives . . . . . . . . . . . . . . . . . . . . . . . . 201
8.2.2 Analysis and Parameter Selection . . . . . . . . . . . . . . . . . . . . . . . . 202
8.3 Position-Sensorless Control Using Saliency and Signal Injection . . . . . . . . . . . 206
8.3.1 Signal Injection and Position Extraction . . . . . . . . . . . . . . . . . . . . 207
8.4 Vector Current Control of Salient PMSMs . . . . . . . . . . . . . . . . . . . . . . . 209
8.5 Field-Weakening Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
8.5.1 Gain Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212

9 Vector Control of IM Drives 215

9.1 Speed-Sensored Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
9.1.1 Direct Field Orientation Using the Current Model . . . . . . . . . . . . . . 216
9.1.2 Properties of the Current Model . . . . . . . . . . . . . . . . . . . . . . . . 216
9.1.3 Indirect Field Orientation Using the Current Model . . . . . . . . . . . . . 216
9.1.4 Analysis of the Current Model . . . . . . . . . . . . . . . . . . . . . . . . . 218
CONTENTS 7

9.1.5 Improved Flux Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220

9.1.6 Modified Current Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
9.1.7 Analysis and Parameter Selection of the Modified Current Model . . . . . . 223
9.1.8 Parameter Sensitivity of the Modified Current Model . . . . . . . . . . . . . 225
9.2 Speed-Sensorless Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
9.2.1 Direct Field Orientation Using the Voltage Model With Modifications . . . 227
9.2.2 Indirect Field Orientation Using the Statically Compensated Voltage Model 229
9.2.3 Properties of the Statically Compensated Voltage Model . . . . . . . . . . . 231
9.2.4 Instability Phenomena for Regeneration-Mode Operation . . . . . . . . . . 233
9.2.5 Complete Stabilization of the Statically Compensated Voltage Model . . . . 234
9.3 Field-Weakening Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
9.3.1 Gain Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
9.3.2 Maximum Torque . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
9.4 Direct Torque Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
9.4.1 Control Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
9.4.2 Control System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
9.4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
9.4.4 Stator-Flux Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

A Lyapunov Stability Theory 253

B CORDIC Algorithm 257

8 CONTENTS

.
CHAPTER 1

Introduction

The purpose of this chapter is to give a brief overview of the field of electric power engineering,
with particular focus on power electronic converters and drives. A general introduction to electric
energy is first made. We then proceed with an introductory presentation of drives and the types of
machines that are commonly used, followed by an overview of converters. The chapter is finished
by summing up some of the various challenges that might face an engineer who is about to enter
the field of control of converters and drives, and where in the book to find the information needed
to solve various problems.

1.1 Electric Energy

The usage of energy in the world has been continuously increasing for the last century. In the
Western world the energy usage has levelled off due to improved efficiency, whereas it still increases
in countries which are under rapid development, notably China and India. Electric energy is not
a primary energy source, but rather a means for transporting energy. Yet, it is the energy form
that best fulfills requirements on flexibility and efficiency. Its main drawback is that storage of
large amounts of electric energy is difficult. Pumped hydro storage, i.e., water that is pumped
to reservoirs at higher elevation during periods of low load and released through turbines during
periods of high load, is today the most effective method. Batteries may become more relied upon in
the future, but they are still bulky and costly. Therefore, since electrical energy storage is difficult,
costly, and also gives losses, it is desired that generation should match demand to the greatest
possible extent.

1.1.1 Sources
A great percentage of all electric energy is generated from the combustion of fossil fuels—oil,
coal, and natural gas—in thermal generating stations. Steam is raised in boilers and passed to
steam turbines. Thermal generating stations typically have a rated capacity from 100 MW up to
a few gigawatts, where the larger stations are equipped with ten or more turbine-generator sets.
Due to the threat of global warming and, on the longer horizon, the (abundant but yet) limited
supply of fuel, many countries have presently agreed to reduce the utilization of fossil fuel over the
decades to follow.
In many countries, a significant contribution is obtained from nuclear reactors, where steam
is produced from the heat generated by fission of uranium-235 atoms. Nuclear generating stations
typically have a rated capacity of 1 GW and up. Nuclear fusion may eventually replace fission—
thereby solving the problems of limited fuel supply and safe storage of used fuel—but due to the
difficulties of handling the extremely high temperatures and pressures required for a controlled and
encapsulated fusion reaction, it is difficult to predict exactly when this will occur.
10 Chapter 1. Introduction

It is generally agreed that a reduction in the usage of fossil fuel must be met by further improved
energy efficiency and renewable energy souces. These include the following.
Hydroelectric (water) energy, possibly the oldest form of energy conversion, was utilized
long before the term “renewable energy source” was even coined. Larger hydroelectric generating
stations have a rated capacity which matches that of a thermal generating station. The Itaipu
dam in South America was long the world’s largest hydroelectric installation, with a capacity of
14 GW shared by 20 generating units. Currently, the largest installation is the Three Gorges dam
in China, with a capacity of 22.5 GW.
Biofuels. Humans have since long generated heat by burning wood, dung, and other fuels
obtained from decaying biomass. In modern generating stations, such fuel, particularly wood chip,
peat, vegetable oil, ethanol, and methane, can be burned to produce steam to drive turbines.
Geothermal energy. In some regions, for example in Iceland, Italy, and New Zealand, hot
springs or lava streams come close enough to the earth’s surface to be tapped. Such energy sources
have been used since the early 1900s.
Wind energy obtained from generators mounted horizontally on towers has become eco-
nomically viable over the last couple of decades. This is because of the increasing capacity of
wind-turbine generators, where the largest today are rated 10 MW. In Denmark, Germany, and
Spain, wind energy already accounts for a significant share of the total generation. Wind energy
is obviously intermittent to its nature and must be balanced by other sources.
Solar energy can either be converted to heat for steam production, by concentrating incident
rays to a receiver using tracking heliostats, or be tapped by photovoltaics, i.e., directly converted
to electricity. The latter generation method is currently growing quickly, particularly in central
and southern Europe. The obvious drawback is that extraction is possible during at most 12 hours
each day, on average. It is also intermittent because of clouds and fog. Solar energy therefore too
must be balanced by other sources and by energy storage.
Tidal energy generation is feasible along coastlines with large tidal differences. It is cyclic to
its nature, but a lot more predictable than wind. Tidal generating stations have been operational
since the 1960s, but they are so far few.

1.1.2 Generation
The majority of all electric energy is created via rotating mechanical motion, where the prime
mover is a hydro, steam, or wind turbine. Electrical machines are converters between mechanic
and electric energy; in that direction in generator mode and in the opposite direction in motor
mode.
The great majority of all electric energy is generated using synchronous machines (synchronous
generators). The main exception so far is conversion of wind energy, where induction machines
are often used. Both machine types are quite efficient: most synchronous generators have an
efficiency of above 90%. Photovoltaic technology is one of the few exceptions where electric energy
is generated without intermediate mechanical motion.

1.1.3 Transmission
Generators and consumers of electric energy often have significant geographic separation. For
example, hydroelectric generating stations are often located in rivers flowing through sparsely
populated regions. Transmission of electric energy over long distances requires high voltage, several
hundred kilovolts, in order to give reasonably low currents and low resistive losses as a result. AC,
unlike dc, transmission allows the usage of transformers for changing voltage levels.Three-phase ac
1.1. Electric Energy 11

transmission, i.e., with three parallel lines, has several benefits over single-phase ac transmission.
Therefore, already in the early 1900s it was decided to use three-phase ac for bulk transmission of
electric energy.
Long-distance three-phase transmission is mostly made by overhead power lines, with rms line-
to-line1 voltages ranging from 100 kV up to (and in a few cases exceeding) 1000 kV. At these voltage
levels close to 100% efficiency is obtained, even over very long distances. Since most generators are
designed for a line-to-line voltage between 11 kV and 25 kV, step-up transformers are employed to
connect generators to power lines.
For bulk transmission over long distances, high-voltage direct-current (HVDC) transmissions
are attractive. The additional expense of power electronic converters for ac–dc conversion is offset
by lower cost of the overhead transmission lines (or underground cables), as only two wires are
required for a bipolar system as opposite to three wires in an ac line.

1.1.4 Distribution

Distribution networks supply electricity to the consumers, typically at line-to-line voltages between
2 kV and 35 kV. An ac transmission grid serves as bulk supply. It is tapped using step-down
transformers located in one or several substation(s).
Larger consumers such as industries tap the distribution network directly. Most consumers,
however, private homes included, draw electricity via step-down transformers providing a line-to-
line voltage of a few hundred volts, for example 400 V in Europe and 190 V in North America.
Larger loads—such as stoves, water heaters, and some washing machines—are connected to three-
phase outlets, whereas the majority—lamps, computers, stereos, TV sets, etc.—draw power from
single-phase outlets, which are connected between one of the three phases and neutral, i.e., a phase
voltage. Hence, the nominal single-phase rms voltage is 230 V and 115 V, respectively, in Europe
and North America. Single-phase loads are as evenly distributed between the three phases as
possible in order to reduce the imbalance in the total current drawn by a distribution grid.

1.1.5 Consumption

The main end users of electric energy are industrial, residential (homes), and commercial (offices,
shops, schools, hospitals, etc.). Industrial consumptions accounts for approximately 40%. The
worldwide annual growth rate of electric-energy consumption has since the early 1900s remained
constant at roughly 7%.
Even as the total energy consumption may level off or decrease in the future due to increase
efficiency, it is likely that usage of electric energy will continue to grow at a fairly high rate. Where
today the end user employs fossil fuel directly, in the future electric energy may in some cases be
used instead. For example, transportation today depends to more than 90% on fossil fuel (refined
oil and, to a lesser extent, natural gas); the main exception is electrified railways (including subways
and light-rail lines). Transportation should gradually shift to reliance upon electric energy. It is
likely that traffic-intensive railways which today use diesel haulage (notably in North America and
Australia) eventually will be electrified. With improved battery technology and possibly electrified
highways, we are also going to see abundant usage of electric and hybrid road vehicles. Knowledge
of electric energy technology and electrical drives therefore should become all the more important
to engineers in various disciplines.
1 The
√
line-to-line voltage is 3 times larger than the line-to-neutral voltage (which is also called the phase voltage).
12 Chapter 1. Introduction

1.2 Electrical Drives

Since electric energy is mostly obtained using mechanical motion, it is not surprising that the
opposite also is true, to a significant extent. Electrical motors are electrical machines which always
or mostly operate in motor mode. An electrical drive consists of a motor, its mechanical load,
and—if such is used—the motor control system, which may include a power electronic converter.
Between 50% and 60% of all electric energy produced, or about 20% of the total energy pro-
duced, is consumed in electrical drives. The majority of these are in the low-power (fractional
kW) and medium-power (1–100 kW) ranges. Examples of application are electric drills, washing
machines, fans and air conditioners, clocks and watches, CD and DVD players, and cameras and
projectors. A modern automobile is equipped with up to 50 low-power electrical motors. In the
higher power ranges we find industrial drives rated up to 100 MW. Typical industrial applications
are pumps, fans, saws, winches, grinders, presses, hoists, and rollers.
Applications in electric traction include electric and diesel-electric locomotives, light-rail ve-
hicles, and subway trains. As mentioned above, a growth area in the future will be electric and
hybrid road vehicles.
The motor may be the most important component of a drive, but it is not the only one.
As Figure 1.1 shows, the motor is often fed from a power electronic converter, which in turn is
connected to a control unit (today often a digital signal processor, DSP). Therefore, to design and
construct an electrical drive, knowledge in the following areas of engineering is needed:

• Mechanics. To model and understand the characteristics and dynamics of the mechanical
load.

• Electrical machines. To understand the characteristics and dynamics of the motor, and
to select the proper motor “off the shelf” for the application in case. Also, to obtain motor
parameters needed for various control algorithms.

• Power electronics. To understand the operating principles and limitations of power elec-
tronic converters, and to select (or design) the converter best suited for the application in
case.

• Sensors and measurement (or estimation). To find a cost-effective solution to the

problem of measuring voltages, currents, fluxes, and rotor speed and/or position. Also, when
a sensor cannot be used or is deemed too expensive, replace it by an estimation algorithm
while understanding how this affects the performance of the drive.

• Signal processing and control. To apply filters when needed, and to design control
algorithms for the power electronic converter that fulfill given demands on dynamic response
and accuracy, etc., given limited knowledge of machine parameters and unavailability of
certain measurements (i.e., to take into account that estimates are used instead).

• Computer engineering. To select and/or develop suitable hardware and software on which
the signal processing, control, and measurement/estimation algorithms can be implemented.

As can be seen, electrical drives are a highly multi-disciplinary topic. As the title of this book
suggests, the focus is signal processing and control, although all the above areas are covered at
least briefly.
1.2. Electrical Drives 13

Electric
power

.
Reference Power Mechanical
Control electronic Motor load
converter
Voltage Current Speed/position
measurement measurement measurement
Feedback

Figure 1.1. Block diagram of an electrical drive. Thick lines indicate flow of power; thin lines signals.

1.2.1 Benefits of Electrical Drives

Compared to other drive types, such as combustion engines and hydraulic or pneumatic drives,
electrical drives have some very attractive properties:
• They are available in the entire power range, from less than 1 W to 100 MW.
• There is a wide range of speed and torque available. In rolling mills, torques in excess of 106
Nm are required, whereas in watches and medical applications, for example, the torque is
sometimes less than 10−6 Nm.
• Electrical drives can be adapted to almost any operating condition. Whereas some drives
require forced cooling, other can be completely encapsulated. Electrical drives can operate
in liquids or be exposed to radioactive or potentially explosive environments without harm
or risk.
• Refueling is not needed and service requirements are modest. There is no exhaust and the
acoustic noise level is low.
• The efficiency is high (typically 75% to 95% for operation at rated load) and the low-load
losses are minimal. The short-term overload capability is high.
• Finally, electrical drives can be designed to operate in both directions of rotation and in both
directions of power flow. For example, an electric vehicle can run forward and backward
at full speed, and can in both directions regenerate energy when braking. The latter is
particularly attractive as compared to a combustion engine.

1.2.2 Motor Types

Electrical motors can be classified in two main categories.
• DC motors (DCMs) were the motors first invented for practical use. As is further discussed
in Chapter 4, dc motors use a directly magnetized stator, whereas a mechanical commutator
distributes current in the rotor, so that the desired torque is produced. The commutator
utilizes sliding carbon brushes, which wear down and have to be replaced. A dc motor
therefore requires significantly more maintenance than an ac motor (see below).
On the other hand, the commutator makes the dc motor controllable without power electronic
converters. The dc motor was therefore up until the 1980s the only motor type available for
variable-speed drives. Classic applications are electric traction, rolling mills, and servos.
14 Chapter 1. Introduction

• AC motors (ACMs) can be regarded as dc motors turned inside out: the rotor is magnetized
directly or indirectly, and the desired torque is produced by applying a rotating (ac) stator
current of proper magnitude, frequency, and phase. Except in constant-speed drives, this
requires power electronic converters which are more advanced than for dc drives. So, whereas
ac drives are advantageous due to less maintenance and higher efficiency, it was not until
the late 1980s that the development of power semiconductors (and microprocessors for their
control) finally reached levels that allowed performance comparable to dc drives. From the
1990s and on, ac drives have increasingly been replacing dc drives, particularly in industrial
and traction applications.
There are mainly three sub-categories of ac motors.

1. Induction motors (IMs) use a so-called cage rotor, in which current is induced. Torque
is produced due to the interaction of the stator current and the flux produced by the
rotor current. As is further described in Section 5.2, this allows the IM to start by direct
line connection. Called direct line start, this is very convenient, but the drive can be
used only in constant-speed or near-constant-speed operation. In such applications the
IM has been used since the late 1800s, but is today commonplace also in variable-speed
drives.
Due to the induction of rotor current the IM rotates with a slip, i.e., slightly slower than
the applied stator current when operating in motor mode.
2. Synchronous motors (SMs) use a directly magnetized rotor, either by permanent
magnets (a so-called permanent-magnet SMs—PMSMs), or by a field winding in the
rotor (so-called electrically excited SMs). “Synchronous” refers to the rotation, which
is of the same frequency as the flux resulting from application of stator current. For
this reason, an SM must be supplied from a power electronic converter to allow variable
speed, but many constant-speed SMs allow direct line start. SMs have traditionally
been used in high-power, constant-speed applications, but are today increasingly used
also in variable-speed drives. Electrically excited SMs are typically utilized for high-
power applications, whereas PMSMs mainly are used in high-performance variable-speed
drives (such as servos) of low and medium power ratings. The dynamics and control
of an electrically excited SM are in many respects similar to those of IMs and PMSMs,
but there are nevertheless several special issues. Therefore, we have in this book elected
to focus on IMs and PMSMs, which also are much more commonly used as motors.
3. Reluctance motors (RMs) in many ways resemble synchronous motors, but rely on
reluctance rather than rotor magnetization for production of torque. As their con-
struction is very simple and rugged—no magnets or rotor winding is used—reluctance
motors are very attractive in low-power, variable-speed applications. Supply from a
power electronic converter is required. The synchronous reluctance motor—or reluc-
tance synchronous motor—is treated as a special case of the PMSM in Section 5.3. The
more popular switched reluctance motor differs so much from other ac motor types,
however, that it is not considered further in this book. For more information, see e.g.,
[60].

Figure 1.2 summarizes how the different motor types are characteristically used in the power range
from fractional kW up to 100 MW.
1.3. Power Electronic Converters 15

DCM
IM
. EESM
PMSM
RM
<< 1 kW 1 kW 10 kW 100 kW 1 MW 10 MW 100 MW

Figure 1.2. Power ranges for different motor types.

1.2.3 Variable-Speed Drives

As its title suggests, this book focuses to a great extent on variable-speed drives. The number of
applications where variable speed is required—or at least preferred—is increasing, due to higher
demands on precision and efficiency. As an example, whereas in bygone days it was considered
perfectly fine that a washing machine always rotated with the same speed, it is nowadays taken for
granted that different speeds should be used depending on the fabric being washed and the stage
in the washing program.
Many applications, such as robotics, rolling mills, and paper machines, require accurate torque,
speed, and position control, often with high dynamic requirements (short response times). In
other applications, accuracy and dynamic performance matter less, but speed control is still highly
important, as this implies significant energy savings. This is particularly true for pumps and fans.
Controlling the flow of liquid or gas by adjusting the pump or fan speed, rather than having the
motor operate at constant speed and using a throttle, means that the energy efficiency can improve
tremendously.
The usual approach when teaching electrical drives is to start with constant-speed drives. Due
to their increasing importance, we here consider variable-speed drives already from the beginning.
Constant-speed drives then follow as special cases.

1.3 Power Electronic Converters

Variable-speed ac drives cannot be (fully) controlled without power electronic converters. Whereas
variable-speed drives are one important application of power electronics converters, it is by no
means the only one. Power electronic converters come in four fundamental variants, see Figure
1.3. The flow of power in this figure is generally—but not always—from left to right, i.e., the input
port is on the left-hand side and the output port is on the right-hand side. Converters which do
not allow and do allow power flow in both directions are, respectively, called unidirectional and
bidirectional. The same terms apply with regards also to current flow.

1.3.1 DC–DC Converters

Converters of this category are mostly unidirectional, and act as transformers for direct voltage
and current. Unlike a transformer, though, an output voltage which is variable within a wide range
can be obtained. To lower the voltage, step-down or buck converters are used, whereas step-up
or boost converters increase the voltage. The hybrid buck–boost converter allows output voltage
variable between zero and (theoretically) infinity.
16 Chapter 1. Introduction

(a) (b)

(c) .
(d)

Figure 1.3. Power electronic converters. (a) dc–dc. (b) ac–dc. (c) dc–ac. (d) ac–ac.

DC–dc converters operate in switched mode: the output voltage is a train of pulses. Such
converters are said to be self-commutated. This enables fast control, with response times in
the millisecond range. The switching device is in most cases a transistor. Inductances and/or
capacitances are needed to smoothen the output voltage and give a (virtually) ripple-free output
current.
DC–dc converters are used in various applications, many of them low-power, including voltage
supplies for dc circuits which require precise voltage control. Low-power dc–dc converters can be
fully integrated, for example to permit variable voltage levels in different blocks of a CMOS circuit.
In higher power ranges dc–dc converters are used for example to power variable-speed dc motor
drives.

1.3.2 AC–DC Converters

Conversion from ac to dc is called rectification. There are two fundamental, but related, converter
types in this category.

Diode Rectifiers
The diode rectifier is the most common converter, being used in most voltage supplies for low-power
dc circuits. The “diamond-shaped” single-phase rectifier depicted in Figure 1.4(a) may be familiar
to those readers who have studied the circuit diagram of, e.g., an audio amplifier. As shown in this
figure, at the output of the rectifier, a capacitor is utilized in order to give a smooth direct voltage.
The load is modeled as a direct-voltage source, but is in many cases partly or purely resistive.

Thyristor Converters
Diode rectifiers do not allow the direct voltage to be varied. This is remedied by replacing the diodes
in Figure 1.4(a) by thyristors. Whereas a diode has a simple P-N junction, a thyristor has four
layers, where the middle P layer is connected to a gate. This prevents the device from conducting
when the anode–cathode voltage turns positive, until triggered by a short pulse at the gate. It is
said that the thyristor’s turn-on instant is delayable. As shown in Figure 1.4(b), both an inductor
and a capacitor may be used to smoothen the output direct voltage, but in many applications only
one of the two components is installed. Thyristor converters allow negative output voltage; the
1.3. Power Electronic Converters 17

(a) (b)

(d)

AC source DC source

(c)

Figure 1.4. (a) Single-phase diode rectifier with a capacitor smoothing the direct voltage. (b) Three-phase
thyristor converter with an inductor and/or a capacitor for smoothing the direct voltage. (c) Antiparallel-
connected thyristor bridges, allowing bidirectional current flow. (d) Three-phase voltage-source converter.

direction of power flow is then reversed. Hence, thyristor converters are bidirectional with regards
to power flow, and are said to operate in rectifier mode when the output voltage is positive and
inverter mode otherwise (“invert” refers to inverted operation as compared to a rectifier).
One important application for thyristor converters is to feed variable-speed dc drives. Neglect-
ing resistive losses, a dc motor—as will be seen in Chapter 4—can be modeled as a series connection
of an inductance and a voltage source (the back emf). Figure 1.4(b) with the capacitor removed
illustrates the circuit diagram. Other important applications are HVDC transmissions and static
var compensators (SVCs). The latter allows control of the reactive power at a certain node in a
power system, thereby stabilizing the voltage at that node. There is no transfer of active power
between the ac and dc sides.
The direction of current in a thyristor converter cannot be reversed, since thyristors—like
diodes—conduct only in one direction. To allow negative direct voltage and current, two antiparallel-
connected thyristor bridges, as illustrated in Figure 1.4(c), must be used.
We finally note that, since thyristor converters do not employ switch-mode operation, their
control is relatively slow, and they generate sizeable low-order current harmonics on the ac side.
Thyristor converters as well as diode rectifiers are said to be line commutated.

1.3.3 DC–AC Converters

Converters of this category are referred to as inverters. Of course, a birectional ac–dc converter
is also a bidirectional dc–ac converter, so a thyristor converter which operates in inverter mode
may be referred to as an inverter. The term inverter is, however, often reserved for inherent dc–ac
converters, which are able to actually create an alternating voltage of variable frequency. Similarly
18 Chapter 1. Introduction

to dc–dc converters, this requires switch-mode operation, which enables fast control with response
times in the millisecond range.
A voltage-source inverter (VSI) uses a stiff direct voltage—kept up by one or several capacitor(s)—
as source, which is “inverted” into ac. As switching devices, gate-turn-off (GTO) thyristors were
initially used, but today transistors dominate. The circuit symbol for a general on/off-switching
device is a that of a diode with two gate symbols. Figure 1.4(d) shows the circuit diagram of a
three-phase VSI. The antiparallel-connected diodes enable bidirectional current (and power) flow.
To highlight that a VSI may operate in rectifier mode just as well as in inverter mode, the term
voltage-source converter (VSC) will be used henceforth.

1.3.4 AC–AC Converters

First, we note that direct ac–ac conversion is rarely used. It is often beneficial to convert from ac
to ac via an intermediate dc bus, which is often referred to as a dc link if it interconnects only two
converters. Conversion from ac to ac has three important applications.

• Variable-speed ac drives: In many ac drive applications, a multi-terminal dc bus is not

available. Power is drawn directly from a three-phase line using an ac–ac converter. The line
side of the converter has constant frequency and the drive side variable frequency.

• HVDC transmissions: Two asynchronous ac grids—alternatively two distant nodes within

one grid—may be joined together by an HVDC transmission, allowing controlled transfer of
power. HVDC transmissions are also used to connect a distant source or load to an ac grid;
examples include offshore wind farms and oil or gas platforms.

• Frequency conversion: Two ac systems with different fundamental frequencies can be

coupled together by an ac–ac converter. A typical application of this is supply to electrified
railways. For example, in many European countries, the three-phase grid has fundamental
frequency 50 Hz, whereas the railways are electrified with 16 2/3-Hz single-phase system.

Indirect AC–AC Conversion

In ac drives, the converters on each terminal of the dc link are often placed adjacent to each other.
This is known as a back-to-back converter system. The converters in most HVDC transmissions
are located in different stations, which are separated by a dc cable or an overhead line. However,
the cable or line can be accurately modeled as an inductive or capacitive dc link (see below) as
seen from the converters, so the converters are electrically back-to-back also in this case.
There are three dc-link variants, see Figure 1.5. Selection is made depending on the converter
types used.

(a) Voltage-stiff (capacitive) dc link: This type of dc link is used with VSCs, which allows
full bidirectional power and current flow. If unidirectional power flow is sufficient, a diode
rectifier may be used as grid-side converter. Back-to-back VSCs with a voltage-stiff dc link
are also increasingly used in HVDC transmissions.

(b) Current-stiff (inductive) dc link: When thyristor converters are employed at both ter-
minals, a current-stiff dc link is used. Line-commutated, i.e., thyristor-based, HVDC trans-
missions have this configuration.
1.3. Power Electronic Converters 19

(c) Resonant (inductive–capacitive) dc link: Although this configuration is sometimes used

when the left-port converter is a diode rectifier or a thyristor converter, cf. Figure 1.4(b),
we here specifically mean the situation where the dc-link components are selected to give
a very high LC resonant frequency (often above 100 kHz). Forcing the dc-link voltage
to oscillate with the resonant frequency, together with a VSC this allows switchings to be
made when the dc-link voltage is zero (and possibly also the current), which reduces losses
and electromagnetic interference (EMI). Since extra components and relatively complicated
control are required, resonant-link converters have so far been applied fairly sparsely. The
interested reader is referred to the literature devoted to the topic [47].

(a) (b)

(c)

Figure 1.5. Indirect ac–ac conversion with three dc-link variants. (a) Voltage-stiff. (b) Current-stiff. (c)
Resonant.

Direct AC–AC Conversion

There are two converter types which are able to convert directly from ac to ac without an interme-
diate dc link. Both are used in ac drive applications, but sparsely. The interested reader is again
referred to the literature specializing in the topic, e.g., [75].

• Matrix converters interconnect two three-phase systems as illustrated in Figure 1.6. Each
interconnection—marked by a dot in Figure 1.6—requires two switching devices to allow
bidirectional current flow. That is, 18 switching devices are required for a three-phase matrix
converter. This, along with other issues such as complicated control, has so far prevented
widespread usage of matrix converters.

• Cycloconverters for three-phase ac–ac conversion consist of six thyristor bridges: two
antiparallel-connected bridges—cf. Figure 1.4(c)—per phase, which enables bidirectional cur-
rent flow. Cycloconverters inherit the drawbacks of thyristor rectifiers: slow control and large
low-order current harmonics. They are predominantly used in high-power, variable-speed ac
drives, particularly SM drives, where these drawbacks are acceptable.
20 Chapter 1. Introduction

Figure 1.6. Three-phase matrix converter.

1.4 Challenges in Control of Converters and Drives

Converter and drives control is the focus of the book; what problems are there? In one sense,
VSC control is straightforward, because similar solutions can be applied to solving various control
problems, due to similarity between the various loads and systems connected to the VSC. In
fact, many control problems boil down to controlling the current i through a load consisting of a
resistance R, an inductance L, and a voltage source E (the motor’s back emf) connected in series,
by manipulating the converter output voltage v. This is illustrated in Figure 1.7.

R L
i

v E

Figure 1.7. Model of a VSC with voltage v connected to an alternating or direct voltage source. The
task is to properly control the current i.

Whereas VSC control is quite simple if the load is a dc motor, things immediately get more
complicated for ac.
• It is easily verified that standard proportional-plus-integral (PI) controllers are not partic-
ularly useful for control of ac quantities. How can better controllers be designed? (This is
answered in Chapter 6.)
• The VSC operates in switched mode. How can the output signal of a linear controller (for
example, an improved PI controller) be converted to suitable switching signals? [Answer:
pulsewidth modulation (PWM), which is covered in Chapter 3.]
• In addition to correct voltage and frequency, the phase angle of the VSC’s output voltage must
also be correct, in order to obtain proper active and reactive powers. Thus, the switchings
must be synchronized in some way with the back emf. How is that accomplished? [By a
phase-locked loop (PLL), as shown in Chapter 6.]
• What if the back emf cannot be measured? How do we synchronize the converter in that
case? (By developing further the PLL. This is often very challenging, however, and applica-
tion-specific adaptations must be made. See Chapters 6, 8, and 9.)
1.4. Challenges in Control of Converters and Drives 21

We have so far discussed just control of a VSC. There are in many cases several cascaded or intercon-
nected control loops, particularly in a back-to-back configuration where the load is a variable-speed
ac drive fed via a dc link from an ac grid, as shown in Figure 1.8.

DC−link voltage
DC−link
voltage reference
controller Flux
controller
Grid Motor

Grid−side Motor−side
current current Speed Speed
controller controller
controller reference

Synch. Synch.
controller controller
(PLL) (PLL)

Figure 1.8. Loops in a back-to-back VSC control system.

There is a
1. speed control loop, by which the desired motor speed is obtained. The output of the speed
controller is the reference torque (or power) of the motor, which is regulated by the
2. motor-side current control loop. If increased current to the motor is demanded, more
power is drawn from the dc link, so a
3. dc-link voltage control loop is required to keep the dc-link voltage at the desired level.
The output of the dc-link voltage controller is the power that should be drawn from the
source, which is the reference for the
4. grid-side current control loop, which is the mirror image of the motor-side current control
loop. In addition, there may be a
5. flux control loop for the ac motor, since the flux must be decreased to allow high speeds
(so-called field weakening). Finally, it should not be forgotten that the
6. synchronization mentioned above is also a control loop. Two synchronization loops are
required, one for each current control loop.
Thus, there are up to seven control loops, where loops 1 and 2 as well as loops 3 and 4, respectively,
operate in cascade: the two current control loops are “slaves” to the speed and dc-link voltage
control loops, respectively. What are the challenges facing the engineer whose task is to design the
entire control system?
22 Chapter 1. Introduction

1. Speed control loop. To achieve performance that is robust to changes in the load torque
and in the system’s mechanical parameters (Section 4.6).
2. Motor-side current control loop. To permit control of amplitude and phase angle inde-
pendently of each other, and to achieve robustness to changes in the back emf (Section 4.4
and Chapter 6).
3. DC-link voltage control loop. To achieve robustness to changes in the power drawn from
the load-side converter.
4. Grid-side current control loop. To permit control of amplitude and phase angle indepen-
dently of each other, and to achieve robustness to variations in the grid voltage, particularly
voltage sags (Chapter 6).
5. Flux control loop. To achieve stability and insensitivity to various motor parameters
(Sections 8.5 and 9.3).
6. Synchronization loops. To achieve stability and insensitivity to various parameters. This
is particularly critical for variable-speed drives at low speeds (Chapters 6–9).
From this summary it is apparent that many of the challenges involve controller design that makes
the control loop less sensitive to disturbances and changes in the system parameters. This is
normally solved loop for loop. However, since there are as many as seven control loop, it is
apparent that they may interact with each other in ways that are not desirable.
So, we see that control of VSCs and variable-speed drives can indeed be quite challenging. Al-
though many problems have been solved—as will be seen by studying further this book—technology
advances and new applications will certainly pose new challenges for researchers and development
engineers for the foreseeable future.
CHAPTER 2

Systems and Circuits in Power

Electronics and Drives

This chapter introduces general theory that will be relied upon in subsequent chapters. Many
of the topics—especially those in the beginning of the chapter—should be well known already,
so the presentation is often kept brief. Starting with single-phase ac systems, we move on to
linear dynamic systems, covering the Laplace transform, transfer functions, and state-space models.
Nonlinear systems are also covered briefly; this is perhaps a new topic to many readers. Sampling
and discrete-time systems and covered next. We then discuss three-phase ac systems and the
convenient analysis tool of space vectors, along with so-called symmetrical dynamic systems, which
can be modeled with complex transfer functions. The chapter is finished by the principle of
normalization using per-unit quantities.

2.1 Single-Phase AC Systems

2.1.1 Sinusoids and Phasors
From Fourier analysis it is known that all signals—periodic as well as transient—can be decomposed
in a (usually infinite) number of sinusoids. The sinusoid is chosen as base waveform because it
is the only signal that can pass a linear system without waveform distortion: its amplitude and
phase are generally altered, but the output signal remains a sinusoid of the same frequency as the
input signal. A sinusoidal signal, e.g., a voltage, is determined by its peak value (or amplitude) V! ,
angular frequency ω1 , and phase angle ϕv
v(t) = V! cos(ω1 t + ϕv ). (2.1)
The subscript 1 of ω1 denotes the fundamental frequency, of which harmonics—see Paragraph
2.1.4—are multiples. The rotating phasor associated with (2.1) is, following the so-called “jω
method,” given by
v̄(t) = V ej(ω1 t+ϕv ) (2.2)
√ √
where V = V! / 2 is the rms value, so v(t) = 2 Re{v̄(t)}. Similarly, a sinusoidal current can be
expressed as
√
i(t) = I! cos(ω1 t + ϕi ) = 2 Re{ī(t)}, ī(t) = Iej(ω1 t+ϕi ) (2.3)
√
where I = I/! 2. A certain voltage or current is often taken as the phase reference: its phase angle
is by definition set to zero, e.g.,
v(t) = V! cos ω1 t. (2.4)
24 Chapter 2. Systems and Circuits in Power Electronics and Drives

We further note that the sine function of course can be picked as base function instead of the cosine
function, e.g., v(t) = V! sin ω1 t. When calculating the actual voltage or current,
√ the imaginary part
of the phasor should then be taken instead of the real part, e.g., v(t) = 2 Im{v̄(t)}, as illustrated
in Figure 2.1.

Im v (t ) / 2
v (t )
ω 1t + ϕ
Re t

√
Figure 2.1. Rotating phasor v̄(t) projected on the imaginary axis yields v(t)/ 2 when sine is used as
base function.

Remark 2.1 Often, fixed phasors are used instead of rotating phasors. A fixed phasor can be
considered as a snapshot of a rotating phasor at t = 0, e.g.,
v̄(t) = V ej(ω1 t+ϕb ) ⇒ V̄ = v̄(0) = V ejϕv . (2.5)

2.1.2 Active and Reactive Powers

The instantaneous power developed in a circuit with input voltage v(t) and input current i(t) is
given by P (t) = v(t)i(t). For (2.1) and (2.3) we obtain
P (t) = v(t)i(t) = V! I! cos(ω1 t + ϕv ) cos(ω1 t + ϕi )
= V I[cos ϕ + cos(2ω1 t + ϕv + ϕi )] (2.6)
where ϕ = ϕv − ϕi . The power oscillates with twice the fundamental frequency. Lamps in
Europe thus blink with 100 Hz and the torque (which is proportional to the power according to
the relation torque=power/angular speed) of a line-connected single-phase ac motor or a series
dc motor pulsates with 100 Hz. The mean value of the instantaneous power is called the (mean)
active power, and is given by
P = <P (t)> = V I cos ϕ (2.7)
where cos ϕ is called the power factor. Maximum power, P = V I, is developed when ϕ = 0, i.e.,
cos ϕ = 1 (unity power factor), zero power is developed when ϕ = ±90◦ , and negative power (i.e.,
power is drawn from the circuit) is developed when 90◦ < |ϕ| ≤ 180◦ .
Phasors allow power calculations where the oscillating terms vanish averaging. The complex
power (the modulus of which is called the apparent power) is introduced as
S̄ = v̄(t)ī∗ (t) (2.8)
where the superscript ∗ stands for complex conjugate. From this quantity, the active power can
be obtained by taking the real part
P = Re{S̄} = V I cos ϕ. (2.9)
2.1. Single-Phase AC Systems 25

Taking instead the imaginary part, we obtain the reactive power

Q = Im{S̄} = V I sin ϕ. (2.10)

When maximum active power is developed, i.e., for ϕ = 0, the reactive power is zero. It is desirable
that loads should draw zero reactive power, i.e., have unity power factor, cos ϕ = 1, since then
the drawn current is fully utilized for active power. To distinguish apparent and reactive power
from active power, the former are given the units volt–amps [VA] and volt–amps (reactive) [VAr],
respectively, whereas active power is measured in watts [W].

2.1.3 Complex Impedances

Phasors allow straightforward calculations of voltage, current, and power using complex impedances.
Since the derivative of a phasor corresponds to multiplication by jω1 , for example
dv̄(t)
v̄(t) = V ej(ω1 t+ϕv ) ⇒ = jω1 V ej(ω1 t+ϕv ) = jω1 v̄(t) (2.11)
dt
the voltage–current relations for an inductor and a capacitor
di(t) dv(t)
v(t) = L i(t) = C (2.12)
dt dt
for phasors become
dī(t) dv̄(t)
v̄(t) = L = jω1 Lī(t) ī(t) = C = jω1 C v̄(t) (2.13)
dt dt
respectively. The complex impedances for an inductor and a capacitor are, thus,
1
ZL (jω1 ) = jω1 L = jXL ZC (jω1 ) = = −jXC (2.14)
jω1 C
respectively, where XL = ω1 L and XC = 1/(ω1 C) are the reactances.
Example 2.1 Let us calculate the current through and the mean active and reactive powers
developed√in an RL and an RC series circuit, respectively, given that the voltage across the circuit
is v(t) = 2V cos ω1 t. The RL circuit has the complex impedance ZRL (jω1 ) = R + jω1 L; thus we
get
v̄(t) V ejω1 t V
ī(t) = = = " ej[ω1 t−arctan(ω1 L/R)]
ZRL (jω1 ) R + jω1 L R + (ω1 L)2
2

and √
2V
i(t) = " cos[ω1 t − arctan(ω1 L/R)].
R2 + (ω1 L)2
It can be observed that when R = 0, the current is displaced − arctan ∞ = −90◦ from the voltage.
Furthermore,
V2 (R + jω1 L)V 2
S̄ = v̄(t)ī∗ (t) = =
R − jω1 L R2 + (ω1 L)2
giving
RV 2 ω1 LV 2
P = Re{S̄} = Q = Im{S̄} = .
R2 + (ω1 L)2 R2 + (ω1 L)2
26 Chapter 2. Systems and Circuits in Power Electronics and Drives

It can be noted that P = 0 when R = 0, which agrees with the finding that no active power is
developed when the phase angle between the voltage and the current is ±90◦ . It should also be
noted that Q ≥ 0. An inductive circuit is said to consume reactive power.
For the RC circuit, we just need to replace ω1 L by −1/(ω1 C), getting
√
v̄(t) V ejω1 t 2ω1 CV
ī(t) = = =" ej[ω1 t+arctan(1/ω1 RC)]
ZRC (jω1 ) R − j/ω1 C 1 + (ω1 RC)2

and √
2ω1 CV
i(t) = " cos[ω1 t + arctan(1/ω1 RC)].
1 + (ω1 RC)2
In this case, when R = 0, the current is displaced arctan ∞ = +90◦ from the voltage. We now
calculate the complex power:

V2 (R − j/ω1 C)V 2 ω1 C(ω1 RC − j)V 2

S̄ = v̄(t)ī∗ (t) = = 2 =
R + j/ω1 C R + 1/(ω1 C)2 1 + (ω1 RC)2

giving
(ω1 C)2 RV 2 ω1 CV 2
P = Re{S̄} = Q = Im{S̄} = − .
1 + (ω1 RC)2 1 + (ω1 RC)2
Also in this case, P = 0 when R = 0, whereas Q ≤ 0 for ω1 > 0. A capacitive circuit is said to
produce reactive power

PROBLEM 2.1
How should R be chosen in the RL and RC circuits, respectively, in order to maximize the
active power?

2.1.4 Harmonics
Suppose that v(t) is a periodic waveform of zero mean with period T1 = 2π/ω1 . This waveform
can be expanded into the Fourier series
∞
# ∞
#
v(t) = ah cos hω1 t + bh sin hω1 t = V!h cos(hω1 t + ϕh ) (2.15)
h=1 h=1

where
$ T1 /2 $ T1 /2
ω1 ω1
ah = v(t) cos hω1 t dt bh = v(t) sin hω1 t dt
π −T1 /2 π −T1 /2
bh
V!h2 = a2h + b2h tan ϕh = − . (2.16)
ah

In (2.15), h = 1 gives the fundamental component, V!1 cos(ω1 t + ϕ1 ), whereas the components that
are frequency multiples of the fundamental, h = 2, 3, . . ., are called harmonics. For the harmonics,
direct calculations using (2.16) reveal the following results:
2.1. Single-Phase AC Systems 27

1) When v(t) is half-wave symmetric, i.e., the negative half period is the mirror image of the
positive half period, all even-order harmonics vanish. This is the case for most waveforms in
electrical engineering. The Fourier series then reduces to

v(t) = V!1 cos(ω1 t + ϕ1 ) + V!3 cos(3ω1 t + ϕ3 ) + V!5 cos(5ω1 t + ϕ5 ) + · · ·

2) For many—although not all—waveforms, the amplitude V!h decreases monotonically with h,
and at least as fast as 1/h. That is, the amplitude of the 5th harmonic is 3/5 or less of that
of the 3rd harmonic, etc. For example, an odd squarewave of zero mean and amplitude 1 has
ah = 0 for all h and
⎧
⎨ 4 , h = 1, 3, 5, . . .
bh = πh
⎩ 0, h = 2, 4, 6, . . .

PROBLEM 2.2
Show that!

Harmonics in electric circuits and systems appear partly due to nonlinear components, primarily
semiconductors and—as we shall see in Chapter 5—nonlinear phenomena in electrical motors.
Whereas, as mentioned, a linear circuit excited by a sinusoidal voltage draws a sinusoidal current
of the same frequency as the voltage, a nonlinear circuit distorts the waveform. The nonlinear
circuit is said to add harmonics. Furthermore, as will be seen in Chapter 3, power electronic
converters create switched waveforms that are far from sinusoidal.
It is often realistic to assume that, at a certain node in a circuit, either the voltage or the
current can be approximated as purely sinusoidal, whereas the other quantity contains harmonics.
Suppose that the voltage is sinusoidal and is taken as phase reference
√
v(t) = 2V cos ω1 t (2.17)

whereas the current contains harmonics as

∞
# √
i(t) = 2Ih cos(hω1 t + ϕh ). (2.18)
h=1,3,5,...

Since 2 cos ω1 t cos(hω1 t+ϕ) = cos[(h−1)ω1 t+ϕ]+cos[(h+1)ω1 t+ϕ], we find that the instantaneous
power is given by

P (t) = v(t)i(t) = V I1 [cos ϕ1 + cos(2ω1 t + ϕ1 )]

+ V I3 [cos(2ω1 t + ϕ3 ) + cos(4ω1 t + ϕ3 )] + · · · . (2.19)

Only the fundamental frequency component produces power of nonzero mean; the harmonics merely
produce power pulsations of 2, 4, 6, etc. times the fundamental frequency. This is one reason why
the sinusoid is the desired waveform in electric power engineering: harmonics just add distortion
that does not contribute to the efficiency of the system. For example, when a power electronic
converter is connected to the grid, passive filters—at the very least a series inductance—must be
used to suppress harmonics from the current drawn from or injected into the grid.
28 Chapter 2. Systems and Circuits in Power Electronics and Drives

2.2 Linear Continuous-Time Systems

In the previous section, we considered steady-state ac excitation of RLC circuits. Such circuits are
dynamic systems. When subjected to a change in some variable, e.g., the input voltage, a transient
results before a steady state is again attained. In this section, general linear dynamic systems (but
RLC circuits in particular) will be studied, using the Laplace transform as mathematical tool.
A dynamic system is often described in terms of its input and output signals. For a general
dynamic system they are often denoted as u(t) and y(t), respectively; see Figure 2.2. The input
signal is available for manipulation and the output signal describes the system’s response to that
manipulation. In the context of electric circuits, the input and output signals are voltages and/or
currents (possibly magnetic flux linkages).

u(t) y (t)
S {.}

Figure 2.2. System y(t) = S{u(t)}.

2.2.1 Definition of Linearity

A linear system S is such that the principle of superposition holds. That is, if the input signal is
the sum of two or more terms, for example

u(t) = a1 u1 (t) + a2 u2 (t) (2.20)

then the output signal y(t) = S{u(t)} can be expressed as

y(t) = a1 S{u1 (t)} + a2 S{u2 (t)}. (2.21)

If ideal components are assumed (i.e., parasitic phenomena such as magnetic saturation are ne-
glected), all RLC circuits are linear systems.

PROBLEM 2.3
Dynamic systems are modeled using differential equations. Which ones of the following are
models for linear systems?

dy(t) dy(t)
a) + ay(t) = b sin u(t) b) + ay(t) = bu(t)
dt dt
dy(t) dy(t)
c) + ay(t) = btu(t) d) + ay 3 (t) = bu(t)
dt dt
( )2
dy(t) dy(t)
e) + ay(t) = bu(t) f) + ay(t) = bu(t)y(t)
dt dt

2.2.2 Laplace Transform

Phasors and complex impedances allow only calculation of the response to steady-state ac exci-
tation. To calculate also the transient response, e.g., resulting from a switch being closed at a
2.2. Linear Continuous-Time Systems 29

Table 2.1. Laplace transform pairs.

Signal (for t ≥ 0) Laplace transform

−at 1
e (Exponential)
s+a
δ(t) (Impulse) 1
1
1 (Step)
s
1
t (Ramp)
s2
s
cos ωt (Cosinusoid)
s2 + ω 2
ω
sin ωt (Sinusoid)
s + ω2
2
$ t
U (s)
u(τ ) dτ (Integration)
0 s
−sTd
u(t − Td ) (Time delay) e U (s)
$ ∞
y(t) = h(τ )u(t − τ ) dτ (Convolution) Y (s) = H(s)U (s).
0

certain time (which is often taken as t = 0), the Laplace transform can instead be used. Having
the definition
$ ∞
Y (s) = L{y(t)} = y(t)e−st dt (2.22)
0

the Laplace transform has the important property that a time derivative is transformed to multi-
plication by s
* +
dy(t)
L = sY (s) − y(0−) (2.23)
dt

where often y(0−) = 0 can be assumed (the circuit or system is assumed not to be excited for
t < 0). We have the correspondence
d
s↔ . (2.24)
dt
So, when calculating the transient response of an electric circuit, the following Laplace impedances
should be used instead of the corresponding complex impedances:
1
ZL (s) = sL ZC (s) = . (2.25)
sC
Some important Laplace transform pairs are shown in Table 2.1.

Example 2.2 We wish to calculate the response of the current i(t) through an LR √ series circuit
to a direct voltage v(t) = V (i.e., the step response) and an alternating voltage v(t) = 2V cos ω1 t,
respectively. Both voltages are applied at t = 0, and i(t) = 0 can be assumed for t < 0.
30 Chapter 2. Systems and Circuits in Power Electronics and Drives

The Laplace transform of a constant signal v(t) = V applied at t = 0 (i.e., a step of amplitude
V ) is V (s) = V /s. The transform of the current is obtained by dividing this by the Laplace
impedance of the RL circuit
, -
V (s) V V V 1 1
I(s) = = = = − .
sL + R s(sL + R) sL(s + R/L) R s s + R/L
Rewriting the expression in partial fractions, as above, makes identification of the terms in Table
2.1 easy. We obtain
V
i(t) = L−1 {I(s)} = (1 − e−Rt/L ), t ≥ 0
R
−1
where “L ” denotes the inverse Laplace transform.
When calculating the response to the ac excitation, a lot of work would result were we to take the
Laplace transform of v(t) directly. Instead, we introduce the rotating phasor v̄(t) = V ejω1 t , t ≥ 0,
whose Laplace transform is V̄ (s) = V /(s − jω1 ), giving

¯ V V
I(s) = =
(s − jω1 )(sL + R) L(s − jω1 )(s + R/L)
, -
V 1 1
= −
R + jω1 L s − jω1 s + R/L
and
¯ V
ī(t) = L−1 {I(s)} = (ejω1 t − e−Rt/L ), t ≥ 0.
R + jω1 L
√
Taking the real part and multiplying by 2 then yields
√ √
2V 2V R
i(t) = " cos[ω1 t − arctan(ω1 L/R)] − 2 e−Rt/L , t ≥ 0.
2
R + (ω1 L) 2 R + (ω1 L)2

The second term is the transient part of the response, which decays to zero with the time constant
L/R, leaving the steady-state part (first term), which verifies Example 2.1.
♣

PROBLEM 2.4
Find expressions for the current through an RC series
√ circuit when the voltage across the
circuit is given by a) v(t) = V, t ≥ 0 and b) v(t) = 2V sin ω1 t, t ≥ 0, respectively.

2.2.3 Transfer Functions

The transfer function H(s) is the quotient between the Laplace transforms of the output and input
signals:
L{output signal} Y (s)
H(s) = = . (2.26)
L{input signal} U (s)
Example 2.3 The transfer function of an RL series circuit with the voltage v(t) across the cir-
cuit taken as input signal and the current i(t) through the circuit as output signal is simply the
admittance of the circuit:
I(s) 1
H(s) = = .
V (s) sL + R
2.2. Linear Continuous-Time Systems 31

PROBLEM 2.5
Find the transfer function of an RLC series circuit with voltage input and current output.

PROBLEM 2.6
Consider an RC series circuit where the input signal is the voltage across both components
and the output signal is the voltage across the capacitor.
a) Find the transfer function for the circuit.
b) Find the step response for a step of amplitude V : u(t) = V, t ≥ 0.
c) Find the response to a short-duration pulse u(t) of amplitude V and duration T . Hint:
u(t) ≈ AT δ(t).

2.2.4 Relation Between Poles and System Dynamics

It is generally desired that a dynamic system should be stable. The stability property of a system
can be linked to the system’s impulse response h(t), i.e., the response y(t) to a short-duration pulse
with pulse area 1: u(t) = δ(t). Because L{δ(t)} = 1, we have

y(t) = L−1 {H(s)} = h(t). (2.27)

The system is
• asymptotically stable if h(t) → 0 as t → ∞;
• marginally stable (or conditionally stable) if h(t) is limited as t → ∞;
• unstable if |h(t)| → ∞ as t → ∞.
The system’s stability is associated with the locations of the poles of the transfer function. Ex-
pressing the transfer function as the quotient

B(s) bN sN + bN −1 sN −1 + · · · + b0
H(s) = = N (2.28)
A(s) s + aN −1 sN −1 + · · · + a0
where N is called the order of the system, the poles pi , i = 1, 2, . . . , N are the roots to the
denominator polynomial A(s), whereas the zeros zi , i = 1, 2, . . . , N are the roots to the numerator
polynomial B(s). This allows the factorization

(s − z1 )(s − z2 ) · · · (s − zN )
H(s) = bN . (2.29)
(s − p1 )(s − p2 ) · · · (s − pN )

If all poles are of multiplicity 1 [an example where this is not true is A(s) = (s + 1)2 ], then H(s)
can be expressed in partial fractions as
k1 k2 kN
H(s) = k0 + + + ···+ (2.30)
s − p1 s − p2 s − pN
giving the impulse response

h(t) = k0 δ(t) + k1 ep1 t + k2 ep2 t + · · · + kN epN t , t ≥ 0. (2.31)

32 Chapter 2. Systems and Circuits in Power Electronics and Drives

The poles may be complex, i.e., pi = σi + jωi , i = 1, 2, . . . The term of the impulse response
associated with the ith pole can thus be expressed as
ki epi t = ki e(σi +jωi )t = ki eσi t (cos ωi t + j sin ωi t), t ≥ 0. (2.32)
From (2.32) the following conclusions can be drawn regarding stability. The system is
• asymptotically stable if (and only if) Re{pi } = σi < 0, i = 1, 2, . . . , N , i.e., all poles are
located in the left half of the s plane (LHP);
• marginally stable if there is at least one pole located on the imaginary axis: Re{pi } =
σi = 0, but none in the right half of the s plane (RHP), and
• unstable if there is at least one pole located in the RHP: Re{pi } = σi > 0.
The result also holds for systems with poles of higher multiplicity than 1 with one exception: for
the system to be marginally stable, any poles on the imaginary axis must be of multiplicity 1.
Thus, the system H(s) = 1/s is marginally stable, whereas H(s) = 1/s2 is unstable. In addition
to stability, (2.32) reveals the following properties of the system dynamics.
• The distance from the pole to the imaginary axis, i.e., the absolute value of the real part, |σi |,
determines the exponential divergence rate for RHP poles (σi > 0) or the exponential
convergence rate for LHP poles (σi < 0) associated with the ith pole. A pole further into
the LHP gives faster exponential convergence than a pole close to the imaginary axis.
• The distance from the pole to the imaginary axis, i.e., the absolute value of the imaginary
part, |ωi |, determines the angular frequency of oscillation associated with the ith pole.
A pole further away from the real axis gives a higher oscillation frequency than a pole closer
to the real axis. A pole on the negative real axis is associated with an exponential convergence
which does not oscillate.
These properties are summarized in Figure 2.3.
Remark 2.2 If all coefficients of a transfer function are real, then the poles and zeros must appear
in conjugated pair. For example, if there is a pole at s = σ1 + jω1 there must also be a pole at
s = σ1 − jω1 . All imaginary parts in (2.32) then add up to zero, making h(t) real.

2.2.5 Determination of Stability

Solving the poles of a transfer function is often impractical, because the transfer function often
does not have numerical coefficients, but analytic coefficients that are functions of the system
parameters, such as RLC circuit parameters. The following rules are helpful for determination of
the stability of a system:
• First- and second-order systems are asymptotically stable if (and only if) all denominator
coefficients are positive (a1 > 0 and a0 > 0). H(s) = 1/(s2 + s + 1) is thus asymptotically
stable, but not H(s) = 1/(s2 − s + 1).
• For higher-order systems it is necessary but not sufficient for asymptotic stability that all
denominator coefficients are positive: ai > 0, i = 0, 1, . . . , N − 1. H(s) = 1/(s3 − s2 + s + 1)
is thus not asymptotically stable, whereas nothing conclusive can be said about H(s) =
1/(s3 + s2 + s + 1) without further checking, e.g., using the Routh–Hurwitz test.
2.2. Linear Continuous-Time Systems 33

s
jω

increased frequency

instability
σ
increased

increased frequency
convergence rate

Figure 2.3. Pole locations in the s plane and the associated exponential responses.

• Passive RLC circuits are asymptotically stable for any system order. (In the unrealistic case
when there is no resistance at all, the circuit is marginally stable.)

2.2.6 Frequency Response

The transform rule for the convolution shown in Table 2.1 reveals the connection between the
Laplace transform and the jω method. If the input signal to a system with transfer function H(s)
is taken as a complex phasor u(t) = U ! ejωt (whose angular frequency ω is arbitrary), then the
output signal is given by
$ ∞ ($ ∞ )
y(t) = !
h(τ )U e jω(t−τ )
dτ = !
h(τ )U e −jωτ
dτ U! jωt
0 0
. e/0 1 = H(jω)u(t). (2.33)
. /0 1 u(t)
H(jω)
We find that the frequency response for a certain system is evaluated simply by setting s = jω in
the transfer function. The special case s = ω = 0 yields the static gain H(0), i.e., the gain for a
constant input signal. This follows also from the final value theorem, which says that, provided
that y(t) converges as t → ∞, then
y(∞) = lim sY (s). (2.34)
s→0

If Y (s) = H(s)U (s) and the input signal is a unit step, i.e., u(t) = 1, t ≥ 0 ⇒ U (s) = 1/s, then
y(t) converges provided that H(s) is asymptotically stable. In that case, (2.34) can be applied to
34 Chapter 2. Systems and Circuits in Power Electronics and Drives

calculate the final value

H(s)
y(∞) = lim sY (s) = lim s = H(0). (2.35)
s→0 s→0 s

2.2.7 Rise Time and Bandwidth

The rise time of a system is defined as the time tr required for the output signal to go from 10% to
90% of the final value, given that the input signal is a step, see Figure 2.4(a). For the first-order
system

b k 1 b
H(s) = = , T = , k= (2.36)
s+a sT + 1 a a
where T is the known as the time constant. We find that the time constant is the inverse distance
from the pole to the imaginary axis. The rise time can be expressed as

ln 9
tr = T ln 9 = (2.37)
a
where ln 9 ≈ 2.2.

PROBLEM 2.7
Show that!

PROBLEM 2.8
Show that if a step is applied to (2.36) at t = 0, then the step response has reached 63% of
the final value at t = T .

The bandwidth of a system of low-pass characteristic [i.e., |H(j0)| = H(0) > 0 and |H(j∞)| = 0]
is defined as the angular frequency ωB for which

|H(jωB )| 1
=√ (2.38)
H(0) 2

as shown in Figure 2.4(b). The bandwidth of (2.36) is a.

PROBLEM 2.9
Show that!

Thus, (2.37) can also be expressed as

ln 9
rise time =
bandwidth
which holds exactly for (2.37) but approximately also for higher-order systems.

PROBLEM 2.10
Determine approximately the rise time of the system H(s) = 1/(s2 + 2s + 2).
2.2. Linear Continuous-Time Systems 35

(a) (b)
1

1
0.8

0.8
0.6

|H(jω)|
y(t)

0.6
0.4
0.4

0.2
0.2
tr ωB
0 0
0 2 4 6 0 1 2 3 4 5
t ω

Figure 2.4. (a) Step response y(t) with rise time tr . (b) Magnitude function |H(jω)| with bandwidth
ωB .

2.2.8 Damping
Second-order systems are often parametrized in the following way:

kω02
H(s) = (2.39)
s2 + 2ζω0 s + ω02

where ω0 (which is positive) is called the angular eigenfrequency of the undamped system and ζ
the relative damping. If 0 ≤ ζ < 1, the poles of the system are given by
2 " 3
s = −ζ ± j 1 − ζ 2 ω0 . (2.40)

When ζ is varied between 0 and 1, the poles move on a half circle with radius ω0 in the left half
plane. For ζ > 1, the pole move along the negative real axis, one toward the origin and one
toward −∞. Step responses for a few different values of ζ are depicted in Figure 2.5(a) and their
corresponding poles are depicted in Figure 2.5(b).

The following observations can be made.

• For ζ = 0, the step response oscillates with constant amplitude. The angular frequency of
the oscillation is ω0 (which explains the name of this parameter). The poles are then located
on the imaginary axis, at s = ±jω0 .

• The overshoot, which is the maximal value of y(t) with the final value (here, 1) subtracted,
increases as ζ decreases, peaking at 100% for ζ = 0. Large overshoot corresponds to poles
closer to the imaginary axis.

• For ζ = 0.7, there is only a small√ overshoot (about 5%). For this value√of the relative
damping, or more precisely, ζ = 1/ 2, the poles are located at (−1 ± j)ω0 / 2, i.e., with an
angle of 45◦ relative both the real and imaginary axes.
36 Chapter 2. Systems and Circuits in Power Electronics and Drives

(a) (b)
2
1
0 0
0.3
0.7
1.5 0.5
0.3

Im{s}
y(t)

1 0.7 0
1.2 1 1.2
1

0.5 1.2 −0.5

0.7
0.3
0
−1
0
0 2 4 6 8 10 −2 −1.5 −1 −0.5 0
t (s) Re{s}

Figure 2.5. (a) Step responses and (b) poles of (2.39) for k = ω0 = 1 and five different values ζ =
{0, 0.3, 0.7, 1, 1.2} as indicated.

• For ζ = 1 there is no overshoot. The system is said to be critically damped for ζ = 1. Both
poles are located at s = −ω0 .
• For ζ = 1.2, a slower response is obtained than for ζ = 1. This is because the poles then are
located on the real axis, but one has moved closer to the origin as compared to ζ = 1. This
pole dominates and slows down the response. If ζ > 1 the system is called overdamped.
√
If small overshoots are desired, ζ should not be made smaller than 1/ 2. This means that the
poles should not be located closer to the imaginary axis than with an angle of 45◦ , i.e., to the left
of the solid lines in Figure 2.5(b). This yields the recommendation
1
√ ≤ ζ ≤ 1. (2.41)
2
PROBLEM 2.11
How would letting ω0 = 2 alter the step responses in Figure 2.5?

2.2.9 Differential Operator

Sometimes it is convenient to use the differential operator p = d/dt. In many cases it can be
used interchangeably with the Laplace variable s. A transfer function H(s) is replaced by the
corresponding transfer operator H(p), giving the correspondence
Y (s) = H(s)U (s) ↔ y(t) = H(p)u(t). (2.42)
Whereas H(s) multiplies with the Laplace transformed input signal, H(p) operates on the actual
input signal, i.e., the function of time u(t). For this reason we can never divide by u(t) in the
second equality in (2.42), as we can in the first equality: Y (s)/U (s) = H(s).
Remark 2.3 The Laplace variable and the differential operator shall never be confused with the
Laplace operator, which is the partial-derivative operator
∂2 ∂2 ∂2
∆ =∇·∇ = 2
+ 2+ 2
∂x ∂y ∂z
2.2. Linear Continuous-Time Systems 37

used in vector analysis.

2.2.10 State-Space Models

The state-space model for a linear system is given by

ẋ(t) = Ax(t) + Bu(t) (2.43)

y(t) = Cx(t) + Du(t) (2.44)

where x(t) is the state vector, whose components x1 (t), x2 (t), . . . , xN (t) are called the state vari-
ables. Whereas the transfer function only gives an input–output description of a system, a state-
space model also gives an internal description, because the state variables can be chosen as physical
variables. A, B, and C are matrices, whereas D is a scalar (as long as the input and output signals
are scalars). For electric circuits, the state variables are normally chosen as

• inductor currents, and

• capacitor voltages.

By taking the Laplace transform and eliminating the state vector

sX(s) = AX(s) + BU (s) ⇒ (sI − A)X(s) = BU (s) ⇒

X(s) = (sI − A)−1 BU (s) ⇒ Y (s) = [C(sI − A)−1 B + D]U (s)

the transfer function corresponding to (2.43)–(2.44) is found to be

Y (s) C adj (sI − A)B

H(s) = = C(sI − A)−1 B + D = +D (2.45)
U (s) det(sI − A)

where adj (·) is the adjoint matrix. The poles of the transfer function are given by the roots to
the characteristic polynomial det(sI − A). The roots of the characteristic polynomial are the
eigenvalues of A. Thus, since det(sI − A) is the denominator polynomial of (2.45), the poles of the
system are the same as the eigenvalues of A. For 2 × 2 matrices, the determinant and the adjoint
matrix are calculated as follows:
( ) ( ) ( )
a b a b d −b
det = ad − bc, adj = . (2.46)
c d c d −c a

Example 2.4 The state-space model for an RLC series circuit, where the total voltage is the
input signal u(t) and the current is the output signal y(t), can be found as follows. Let x1 (t) be
the current through the inductor (which equals the output signal) and let x2 (t) be the voltage
across the capacitor. Kirchhoff’s second law then says that

dx1 (t)
Rx1 (t) + L + x2 (t) = u(t)
dt
whereas the derivative of the capacitor voltage times C equals the current

dx2 (t)
C = x1 (t).
dt
38 Chapter 2. Systems and Circuits in Power Electronics and Drives

Rewriting these in state-space component form yields

R 1 1
ẋ1 (t) = − x1 (t) − x1 (t) + u(t)
L L L
1
ẋ2 (t) = x1 (t)
C
y(t) = x1 (t)

from which the state matrices are easily identified

⎡ ⎤ ⎡ ⎤
R 1 1
− − : ;
⎢ L L ⎥
A=⎣ ⎦ B=⎣ L ⎦ C= 1 0
1
0 0
C
whereas D = 0.

PROBLEM 2.12
A state-space model is not unique; there exist an infinite number of different, but from
the “outside” equivalent, state-space models for a certain system.

a) Find the state matrices for the system in Example 2.4 if instead x1 (t) is taken as the
capacitor voltage and x2 (t) as the inductor current.
b) Show that the same transfer function H(s) is obtained for both choices of the state
variables.
c) Suppose that a new state vector w(t) = P x(t) (where P is an invertible transformation
matrix) is introduced in a general state-space system (2.43)–(2.44). Show that the
transfer function (2.45) is invariant of P .

PROBLEM 2.13
Figure 2.6 shows the circuit diagram of a so-called LCL filter.

a) Find a state-space model for the system.

b) Find the transfer function H(s).
c) For L1 = L2 = L, √ select C such that maximum input-to-output transfer of power is
achieved for u(t) = 2U cos ω1 t. Also find an expression for the maximum mean power
developed in the resistor R. It can be assumed that ω1 L ≫ R. Hint: |H(jω1 )|2 .
d) Find an expression for the input impedance and characterize its properties when C is
selected according to c). It can again be assumed that ω1 L ≫ R.

2.3 Nonlinear Continuous-Time Systems

Although many dynamic systems can be modeled using linear models (transfer functions and state-
space models), most are in reality nonlinear, more or less heavily so. The linear model is valid
2.3. Nonlinear Continuous-Time Systems 39

L1 L2

u C R y

Figure 2.6. LCL filter.

only about a certain operating point; often, the linear condition (2.21) holds only when u(t) is
sufficiently small.
When power electronic converters and variable-speed drives are studied, nonlinear system mod-
els will inevitably be encountered. State-space models can be generalized to cover also nonlinear
systems:
ẋ(t) = f (x(t), u(t)) (2.47)
y(t) = g(x(t)). (2.48)
Equations (2.43)–(2.44) for D = 0 represent a special case of (2.47)–(2.48) with f (x(t), u(t))
= Ax(t) + Bu(t) and g(x(t)) = Cx(t). Naturally, it is much more difficult to analyze the stability
of (2.47)–(2.48) than of the linear system (2.43)–(2.44). Although it represents only a special case,
it is often necessary to restrict analysis to a constant input signal, as otherwise the analysis tends
to become extremely complicated. A constant input signal allows u(t) to be dropped from the
state-space equations, obtaining what is known as an autonomous system:

ẋ(t) = f (x(t)). (2.49)

Quite often, the system is of order two, i.e., with x(t) = [x1 (t), x2 (t)]T , or can be approximated as
such. We may then express the system in component form as
ẋ1 (t) = f1 (x1 (t), x2 (t)) (2.50)
ẋ2 (t) = f2 (x1 (t), x2 (t)). (2.51)
The questions of importance are: is (2.49) stable, and if so, to which value does x(t) converge when
initialized at a certain point x(0) in the state space? Naturally, if x(t) converges, then ẋ(t) = 0,
so the points to which x(t) may possibly converge—which are known as equilibrium points—are
obtained by setting ẋ(t) = 0 in (2.49) and solving for x(t). An equilibrium point x⋆ thus satisfies
the relation

f (x⋆ ) = 0. (2.52)

A linear autonomous system ẋ(t) = Ax(t) has only one equilibrium point, the origin x⋆ = 0, but
a nonlinear system may have several. Consider for example the following system:
ẋ1 (t) = −x1 (t) + x21 (t) (2.53)
ẋ2 (t) = x21 (t) − x2 (t). (2.54)
There are two equilibrium points: x⋆(1) = [0, 0]T and x⋆(2) = [1, 1]T .
40 Chapter 2. Systems and Circuits in Power Electronics and Drives

2.3.1 Global and Local Stability

Some nonlinear systems have the excellent property that they are globally stable. This means that,
regardless of the starting point x(0) in the state space, x(t) converges to the same equilibrium
point x⋆ . Global stability may be difficult to prove, though. So-called Lyapunov functions (or
Lyapunov’s second—or direct—method) are very powerful for this purpose, but are often hard to
find; see Appendix A.
Local stability is easier to prove, as will be shown momentarily. Local stability implies that, if
the starting point x(0) is close enough to the equilibrium point x⋆ , then x(t) will converge to x⋆ .
As will be seen in the subsequent chapters, in control of converters and drives, the starting point
is—with only a few exceptions—close to the equilibrium. Therefore, local stability is often good
enough a property.

2.3.2 Analysis of Local Stability by Linearization

If x(t) ≈ x⋆ we have f (x(t)) ≈ 0. Hence, f (x(t)) can be approximated as a first-order Taylor
series expansion
f (x(t)) ≈ f (x⋆ ) +f ′ (x⋆ )[x(t) − x⋆ ]. (2.55)
. /0 1
0
It therefore appears logical that close to an equilibrium point, (2.49) can be approximated as a
linear system
<˙ (t) = A<
x x(t) A = f ′ (x⋆ ) (2.56)
where x <(t) = x(t) − x⋆ . Linearization is the most commonly used method for stability analysis
of nonlinear systems and is also called Lyapunov’s first (or indirect) method. Stability of the
linearized system (2.56) can be analyzed simply by calculating the eigenvalues of A. If we restrict
the analysis to second-order systems—which is the case in the applications encountered in this
book—this matrix is given by
⎡ ⎤
∂f1 (x1 , x2 ) ∂f1 (x1 , x2 )
⎢ ∂x1 ∂x2 ⎥
A = f ′ (x⋆ ) = ⎢ ⎥
⎣ ∂f2 (x1 , x2 ) ∂f2 (x1 , x2 ) ⎦ . (2.57)

∂x1 ∂x2 x=x⋆

There are two eigenvalues, which we denote as λ1 and λ2 , i.e., the characteristic polynomial is
given by
(s − λ1 )(s − λ2 ) = s2 −(λ1 + λ2 ) s + λ1 λ2 . (2.58)
. /0 1 . /0 1
a1 a0
The type of equilibrium point is determined by the signs of the real parts of the eigenvalues.
1. Stable equilibrium point—sink: Re{λ1 } < 0, Re{λ2 } < 0 (a1 > 0, a0 > 0). Any
trajectory starting in the vicinity of x⋆ will converge to x⋆ . The system behaves as an
asymptotically stable linear system about x⋆ .
2. Unstable equilibrium point—source: Re{λ1 } > 0, Re{λ2 } > 0 (a1 < 0, a0 > 0). Any
trajectory starting in the vicinity of x⋆ will diverge from x⋆ . The system behaves as an
unstable linear system about x⋆ .
2.3. Nonlinear Continuous-Time Systems 41

3. Unstable equilibrium point—saddle point: λ1 > 0, λ2 < 0 (a0 < 0). (The eigenvalues
are in this case always real.) Trajectories close to a saddle point tend first to approach the
saddle point and then move away from it. Hence, most trajectories starting in the vicinity
of x⋆ will diverge from x⋆ . Theoretically, there are also initial values x(0) for which x(t)
will converge to x⋆ . These are called the stable separatrices or stable manifolds, and are
important, as they determine the stable regions about each sink [48].
4. Center: Re{λ1 } = Re{λ2 } = 0 (a1 = 0). Nothing conclusive about stability can be said.
Centers are not considered here; for more information, see [48].

Example 2.5 Suppose that an RLC circuit is connected to the input of a dc–dc converter as
shown in Figure 2.7. The input voltage E is constant. If the converter can be assumed static and
lossless, the following input–output power balance must hold:

v1 (t)i1 (t) = v2 (t)i2 (t).

If v2 (t) and i2 (t) are controlled such that the output power is constant: v2 (t)i2 (t) = P , the input
current will be given by
P
i1 (t) = .
v1 (t)
Unlike RLC circuits, the dc–dc converter has a nonlinear input voltage–current relation.

L R i i1 i2

E C v1 v2

Figure 2.7. RLC circuit connected to a dc–dc converter.

Taking i(t) and v1 (t) as state variables, the following nonlinear state-space model is obtained:
di(t) R 1 E
= − i(t) − v1 (t) + (2.59)
dt L L L
dv1 (t) 1 1 P
= i(t) − . (2.60)
dt C C v1 (t)
. /0 1
i1 (t)
The equilibrium points (EPs) are found by solving for i(t) on the right-hand side of (2.59) and
substituting in (2.60), giving the following equation:
= > = ?
2 ⋆(1,2) E E2 E 4P R
v1 − Ev1 + RP = 0 ⇒ v1 = ± − PR = 1± 1−
2 4 2 E2
⎧
( , -) ⎪ PR
E 2P R ⎨ E− ≈ E (EP1)
≈ 1± 1− = E
⎩ PR
2 E 2 ⎪
(EP2)
E
42 Chapter 2. Systems and Circuits in Power Electronics and Drives

the approximations if R can be assumed small. Linearization of (2.59)–(2.60) involves only one
term: i1 (t), since all other terms are linear. Equation (2.57) can either be applied directly, or we
can proceed as follows. Introduce v<1 (t) = v1 (t) − v1⋆ . Then
( )
P P P 1 P v<1 (t)
i1 (t) = = ⋆ = ⋆ ≈ ⋆ 1− ⋆
v1 (t) v1 + v<1 (t) v1 (t)/v1⋆
v1 1 + < v1 v1
P P
= ⋆ − ⋆ 2 v<1 (t)
v1 (v1 )
where the approximation results from the MacLaurin series expansion 1/(1 − x) = 1 + x + x2 + · · · ,
which is truncated after two terms. For P > 0 the dc–dc converter thus acts as a negative
conductance −P/(v1⋆ )2 . The linearized system is obtained as
d<i(t) R 1
= − <i(t) − v<1 (t)
dt L L
d<
v1 (t) 1 1 P
= <i(t) + v<1 (t)
dt C C (v1⋆ )2
giving
⎡R 1 ⎤
− −
⎢ L L ⎥
A=⎣ 1 P ⎦
C C(v1⋆ )2
and the characteristic polynomial
( ) ( )
R P 1 PR
det(sI − A) = s2 + − s + 1 − .
L C(v1⋆ )2 LC (v1⋆ )2
For the two equilibrium points we obtain, respectively
, - , -
2 R P 1 PR
EP1: s + − s+ 1− 2
L CE 2 LC E
, -
R P 1
≈ s2 + − s+
L CE 2 LC
( ) ( )
2 R P 1 PR
EP2: s + − s+ 1−
L C(P R/E)2 LC (P R/E)2
2 2
E E
≈ s2 − 2
s− .
P CR P LRC
Thus, EP2 is a saddle point (unstable), whereas EP1 is a sink (asymptotically stable) if
R P PL
− >0⇒R>
L CE 2 CE 2
and otherwise a source (unstable). A sufficiently large resistance R is required to compensate for
the negative conductance of the converter for P > 0. In Figure 2.8, a simulation for E = 100 V,
P = 500 W, L = 10 mH, and C = 1000 µF is shown, where R = 1 Ω for 0 ≤ t < 0.5 s. The initial
values are v1 (0) = 100 V and i(0) = 0 A. It is seen that v1 (t) converges to a steady-state value
after an initial damped oscillation. The criterion for stability is R > P L/CE 2 = 0.5 Ω. At t = 0.5
s, R is decreased to 0.45 Ω. As can be seen, the equilibrium point turns unstable; an oscillation
with increasing amplitude commences.
2.4. Linear Discrete-Time Systems 43

120

100

80
v1(t) (V)
60

40

20

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t (s)

Figure 2.8. Simulation of an RLC circuit connected to the input of a dc–dc converter that draws constant
power. At t = 0.5 s, resistance R is decreased.

PROBLEM 2.14
Characterize the equilibrium points of the nonlinear system (2.53)–(2.54).

PROBLEM 2.15
Suppose that in the T filter in Figure 2.6, L1 = L whereas L2 = 0, and R is replaced by a
nonlinear resistor with the following voltage–current characteristic:

vR (t) = ki3R (t).

a) Find a nonlinear state-space model for the system.

b) Find and characterize the equilibrium points when the input signal is constant: u(t) =
U . What can be said about the system’s stability properties?

2.4 Linear Discrete-Time Systems

2.4.1 Sampling Process
Virtually all converter control systems are today implemented digitally, i.e., they process discrete-
time samples of continuous-time signals. The sampling process can be illustrated as a switch which
closes periodically with the sampling interval Ts . Each time t = kTs , where k (an integer) is called
the sampling instant, that the switch closes a snapshot of the continuous-time signal y(t) is taken.
This is illustrated in Figure 2.9(a). A discrete-time signal

yk = y(kTs ) (2.61)

results, whose values are called samples. See Figure 2.9(b).

2.4.2 Sampling Theorem and Prefiltering

It is well known that a signal to be sampled should not contain components at frequencies above half
the sampling frequency, fs /2, where fs = 1/Ts , which is also called the Nyquist frequency. In the
44 Chapter 2. Systems and Circuits in Power Electronics and Drives

y (t)
y
y0 y1 2 y3
y4 y
5
y (t) yk = y (kTs)
t=kTs
t
0 Ts 2Ts 3Ts 4Ts 5Ts
(a) (b)

Figure 2.9. Sampling process illustrated (a) as a block and (b) in the time domain.

sampling process, such components will be aliased down into the frequency range [0, fs /2], possibly
causing significant distortion. In situations where there are known to be significant components
above the Nyquist frequency, prefiltering of the measured signal y(t) should be applied before the
signal is sampled. This can be made using a continuous-time low-pass filter (LPF), as shown in
Figure 2.10(a). An alternative is oversampling, i.e., the measured signal is sampled L times as often
as the desired sampling rate, such that all frequency components that are within the bandwidth
of the measurement sensor are captured without aliasing. A discrete-time low-pass filter is then
applied. The output signal of the low-pass filter is then decimated, meaning that every Lth sample
is picked out, giving the desired sampling rate. Often, oversampling, discrete-time prefiltering, and
decimation are made using an FPGA. If needed, a combination of the two methods can be used,
where both continuous- and discrete-time prefilters are applied. In this case, the continuous-time
prefilter can have low order.

y (t) yk y (t) yk
LPFc t=kTs t=kTs/L LPFd L

(a) (b)

Figure 2.10. (a) Continuous-time prefiltering. (b) Oversampling, discrete-time prefiltering, and decima-
tion.

2.4.3 Z Transform
The discrete-time correspondence to the Laplace transform is the Z transform, which is defined as
∞
#
Y (z) = Z{yk } = yk z −k . (2.62)
k=0

One important property is the transform of a one-sample delay

∞
# ∞
# ∞
#
Z{yk−1 } = yk−1 z −k = {k = n + 1} = yn z −(n+1) = y−1 + z −1 yn z −n . (2.63)
k=0 n=−1 n=0
. /0 1
Y (z)
2.4. Linear Discrete-Time Systems 45

Table 2.2. Z transform pairs.

Signal (for k ≥ 0) Z transform

k z
a (Exponential)
z−a
δk (Unit pulse) 1
z
1 (Step)
z−1
z
k (Ramp)
(z − 1)2
yk+1 (Forward shift) z[Y (z) + y0 ]
k
#
yk = hn uk−n (Convolution) Y (z) = H(z)U (z).
n=0

Assuming that y−1 = 0, we thus obtain

Z{yk−1 } = z −1 Y (z). (2.64)

In words, a one-sample delay is transformed to a multiplication by z −1 . Suppose that yk is obtained

by sampling of y(t), as illustrated in Figure 2.9. A one-sample delay obviously corresponds to the
time delay y(t − Ts ), whose Laplace transform is given by

L{y(t − Ts )} = e−sTs Y (s). (2.65)

This shows intuitively that the correspondence between the two transform variables is z −1 = e−sTs ,
i.e,

z = esTs . (2.66)

In complex analysis this is known as a conformal mapping. Some important Z transform pairs are
given in Table 2.2.

2.4.4 Difference Equations and Transfer Functions

The discrete-time correspondence to differential equations are difference equations, which have the
form

yk + aN −1 yk−1 + aN −2 yk−2 + · · · + a0 yk−N = bN uk + bN −1 uk−1 + bN −2 uk−2 + · · · + b0 yk−N (2.67)

where N is the system order. Unlike a differential equation, a difference equation is its own
numerical solution. The output at a certain sampling instant k is computed recursively as the sum
of weighted input and output samples, back to the sampling instant k − N , as

yk = bN uk + bN −1 uk−1 + bN −2 uk−2 + · · ·+ b0 yk−N − aN −1 yk−1 − aN −2 yk−2 − · · ·− a0 yk−N (2.68)

By finding a discrete-time approximation—also called discretization—of a continuous-time system,

this system can be simulated using the difference equation of the discretization. In a similar
46 Chapter 2. Systems and Circuits in Power Electronics and Drives

fashion, filters and controllers can be designed as continuous-time systems and then be discretized
for discrete-time implementation. To determine whether a certain discretization will be successful
or not, some analysis is required, particularly stability theory for discrete-time systems. To this
end, let us apply the Z transform to (2.67). The following transfer function is obtained:
Y (z) bN + bN −1 z −1 + bN −2 z −2 + · · · + b0 z −N
H(z) = =
U (z) 1 + aN −1 z −1 + aN −2 z −2 + · · · + a0 z −N
bN z N + bN −1 z N −1 + bN −2 z N −2 + · · · + b0
= . (2.69)
z N + aN −1 z N −1 + aN −2 z N −2 + · · · + a0

2.4.5 Relation Between Poles and System Dynamics

The Z transform allows the dynamics of a discrete-time system to be assessed using the poles of the
transfer function in a similar fashion as continuous-time systems. The denominator and numerator
of a general transfer function (2.69) can be factorized respectively in poles and zeros as

(z − z1 )(z − z2 ) · · · (z − zN )
H(z) = b0 . (2.70)
(z − p1 )(z − p2 ) · · · (z − pN )
If all poles are of multiplicity 1, then H(z) can be expressed in partial fractions as
k1 z k2 z kN z
H(s) = k0 + + + ···+ (2.71)
z − p1 z − p2 z − pN
giving the unit-pulse response

hk = k0 δk + k1 pk1 + k2 pk2 + · · · + kN pkN , k ≥ 0. (2.72)

The poles may be complex; for discrete-time systems it is convenient to use polar form as pi =
ri ejΩi , i = 1, 2, . . . The term of the unit-pulse response associated with the ith pole can thus be
expressed as

(ri ejΩi )k = rik (cos Ωi k + j sin Ωi k), k ≥ 0. (2.73)

From (2.73) the following conclusions can be drawn regarding system stability. The system is
• asymptotically stable if (and only if) |pi | = ri < 1, i = 1, 2, . . . , N , i.e., all poles are
located within the unit circle in the z plane;
• marginally stable if there is at least one pole located on the unit circle: |pi | = ri = 1, but
none outside, and
• unstable if there is at least one pole located outside the unit circle: |pi | = ri > 1.
The result also holds for systems with poles of higher multiplicity than 1 with one exception: for
the system to be marginally stable, any poles on the unit circle must be of multiplicity 1. In
addition to stability, (2.32) reveals the following properties of the system dynamics.
• The modulus of the pole, i.e., distance from the pole to the origin, |pi | = ri , determines the
exponential divergence rate if |pi | > 1 and the exponential convergence rate |pi | < 1
associated with the ith pole. A pole closer to the origin gives faster exponential convergence
than a pole close to the imaginary axis.
2.4. Linear Discrete-Time Systems 47

• The angle of the pole relative the positive real axis, Ωi , determines the normalized fre-
quency of oscillation associated with the ith pole. A pole with a larger angle gives a
higher oscillation frequency than a pole with a smaller angle. A pole on the positive real axis
within the unit circle, i.e., 0 ≤ pi < 1 is associated with an exponential response that does
not oscillate.
These properties can be derived from the corresponding properties of continuous-time systems by
applying the mapping (2.66) between the s and z planes. They can also be deduced by sampling
the exponential response e(σ+jω)t , which gives

e(σ+jω)kTs = (eσTs )k ejωkTs (2.74)

showing that

r = eσTs Ω = ωTs . (2.75)

Now, it is obvious that asymptotic stability in the s plane, i.e., σ < 0, corresponds directly to
asymptotic stability in the s plane, i.e., r < 1.
Discrete-time systems have two unique properties as compared to continuous-time systems.
• The fastest convergence rate of a discrete-time system is one sample. This corresponds to a
unique point in the z plane: the origin. This fact can be deduced by letting s → −∞, which
via (2.66) gives z → 0. A continuous-time system cannot have an infinitely fast convergence
rate, though, and therefore no corresponding point exists in the s plane.
• The highest oscillation frequency results from a pole on the negative real axis, i.e., Ω = π.
Equation (2.75) shows that this corresponds to the angular Nyquist frequency ω = π/Ts .
This property is a direct consequence of the sampling theorem.
The found properties are summarized in Figure 2.11.

2.4.6 Shift Operator

Sometimes it is convenient to apply the shift operator q, which corresponds to z in the same fashion
as the differential operator p corresponds to s

qyk = qk+1 q −1 yk = yk−1 . (2.76)

2.4.7 Euler Discretization

It is often convenient to design filters, controllers, etc. in the s domain as if they were continuous-
time systems. To allow discrete-time implementation, discretization of the continuous-time model
is needed. Unfortunately, (2.66) cannot be applied directly for this purpose, because it would
imply the substitution s → (ln z)/Ts , resulting in a transfer function which is irrational in z. Any
discretization will therefore be an approximation and will consequently give altered properties as
compared to the underlying continuous-time system.
We shall here consider the Euler discretization method, where a first-order MacLaurin expansion
is made of the right-hand side of (2.66)

z = esTs ≈ 1 + sTs . (2.77)

48 Chapter 2. Systems and Circuits in Power Electronics and Drives

oscillation at the Nyquist frequency

increased
convergence
rate
increased
frequency

instability

maximum convergence rate

Figure 2.11. Pole locations in the z plane and the associated exponential responses.

Solving for s yields the substitution rule

z−1
s→ (2.78)
Ts
which in the time domain corresponds to
dy(t) yk+1 − yk
= u(t) → = uk . (2.79)
dt Ts
This is a a trapezoidal—or forward difference—derivative approximation. An Euler discretized
integrator is thus obtained as

yk+1 = yk + Ts uk (2.80)

which, by substituting k → k − 1, also can be expressed as

yk = yk−1 + Ts uk−1 (2.81)

and whose block diagram is shown in Figure 2.12(a).

It may be noted in Figure 2.12(a) that a delayed integrator is obtained, because of the one-
sample delay z −1 in the signal path from the input signal to the output signal. This delay makes
the Euler method easy to apply, but caution must be exercised. We know that the unit circle is
the boundary of the stability region in the z plane. Letting z = ejωTs in (2.78) thus gives the
2.4. Linear Discrete-Time Systems 49

u Ts y
z−1
1/T s

(a)

(b)

Figure 2.12. (a) Block diagram of an Euler discretized integrator. (b) Stability region in the s plane for
Euler discretization.

mapping of the z-plane unit circle onto the s plane for Euler discretization. A circle which is
centered at s = −1/Ts and with radius s = −1/Ts is obtained, as shown by the circle in Figure
2.12(b). This means that s-plane poles outside of this circle will be mapped to the outside of the
the z-plane unit circle. That is, even when the continuous-time system is stable, i.e., all poles are
in the LHP, the Euler method may render an unstable discretization. In addition, the frequency
response of an Euler discretized system will deviate from that of the underlying continuous-time
system, particularly for higher frequencies. It is said that the frequency response becomes warped.
The warping can be quantified by setting s = jω in (2.66), giving z = ejωTs . Applying this to
(2.78) yields

ejωTs − 1
jω → . (2.82)
Ts

A MacLaurin expansion of the right-hand side of (2.82) yields

ejωTs − 1 1 + jωTs + (jωTs )2 /2! + · · · − 1

= ≈ jω (2.83)
Ts Ts
where the approximation holds for small ω, but not for larger. Particularly, at the Nyquist fre-
quency we have

ej(ωs /2)Ts − 1 ej(π/Ts )Ts − 1 2

= =− (2.84)
Ts Ts Ts

which is completely different from the right-hand side of (2.82), i.e., jωs /2 = jπ/Ts .
The Euler method is well suited for discretization of well-damped continuous-time systems, as
long as Ts is sufficiently small. Poorly damped systems on the other hand, i.e., with poles very
close to, or on, the imaginary axis in the s plane, will be rendered unstable by the Euler method.
In the latter case there are two options.
50 Chapter 2. Systems and Circuits in Power Electronics and Drives

1) Apply another discretization method, e.g., the prewarped Tustin method [81]
z−1 ωh
s → KT , KT = . (2.85)
z+1 tan( ωh2Ts )
This method maps the LHP to the inside of the z-plane unit circle, thus preserving the
stability properties. It also gives an accurate mapping of the angular frequency ωh , which
can be verified by setting z = esTs and, in turn, s = jωh in (2.85). Unfortunately, Tustin
discretization easily results in quite complicated equations.
2) Apply the Euler method to a well-damped s-domain system and then make a transformation
in the z domain. This is often a more straightforward option, which will be exemplified in
Paragraph 2.7.4.

2.5 Three-Phase Systems

A three-phase system is formed by three quantities, oscillating with the angular frequency ω1 and
separated in phase angle by (ideally) 120◦ (2π/3 radians). Three-phase systems are preferable to
single-phase systems mainly for two reasons (which hold under idealized conditions).
1) The sum of the instantaneous values is zero.
2) The instantaneous three-phase power is constant.
As a consequence thereof, three-phase grids today handle the majority of all transmission and dis-
tribution of electric power, while three-phase machines and three-phase power electronic converters
are used for power conversion.
The three phases are usually denoted as a, b, and c (sometimes as r, s, and t). The peak values
of the three components are, if not exactly equal, then at least approximately equal. Phases b and
c (ideally) lag 120◦ and 240◦, respectively, behind phase a. A three-phase voltage system can be
written as
va (t) = V!a cos(ω1 t + ϕa )
vb (t) = V!b cos(ω1 t − 2π/3 + ϕb ) (2.86)
vc (t) = V!c cos(ω1 t − 4π/3 + ϕc )
where va (t), vb (t), and vc (t) are called the phase voltages. Hence, during normal operation (i.e.,
except when a larger disturbance occurs), V!a ≈ V!b ≈ V!c and ϕa ≈ ϕb ≈ ϕc . This is illustrated by
the phasor diagram in Figure 2.13. The phasors rotate counterclockwise with angular frequency
ω1 (in Figure 2.13 they are frozen at ω1 t + ϕa = 0).

2.5.1 Sequences
As mentioned, the ideal situation is a three-phase system with equal peak values in all three
phases, and exactly 120◦ phase distribution. This is called a balanced system. Taking va as phase
reference, we have
va (t) = V! cos(ω1 t)
vb (t) = V! cos(ω1 t − 2π/3) (2.87)
vc (t) = V! cos(ω1 t − 4π/3).
2.5. Three-Phase Systems 51

vc

120 va
120
120

vb ω1

Figure 2.13. Phasor diagram of a three-phase system.

When phase b lags 120◦ behind phase a, as here, we have a positive-sequence component. If the
ordering of phases is modified, e.g., phase b is taken as the phase lagging 240◦ behind phase a
va (t) = V! cos(ω1 t)
vb (t) = V! cos(ω1 t − 4π/3) (2.88)
vc (t) = V! cos(ω1 t − 2π/3)
then a negative-sequence component is obtained. Positive and negative sequences can be visualized
as rotating counterclockwise and clockwise, respectively. This is very relevant for ac motors: if
two of the phases feeding an ac motor are interchanged, the rotation of the motor is reversed.1
Oscillograms of positive- and negative-sequence components are depicted in Figure 2.14. Ideally, a
three-phase system should be pure positive sequence. Small negative-sequence voltage and current
components appear during normal operation due to unbalanced impedances (this will be exem-
plified later), whereas large transient negative-sequence components appear during unbalanced
faults, such as line-to-ground or line-to-line flashovers, for example caused by lightning strikes on
overhead power lines. Since faults normally are cleared within a few periods of the fundamental,
unbalances caused by large negative-sequence components typically have duration only of a few
hundred milliseconds.
One interesting property of any three-phase system which is a positive sequence, a negative
sequence, or a sum thereof is that the instantaneous sum of the components is zero:

va (t) + vb (t) + vc (t) = 0 ∀t. (2.89)

In situations when (2.89) does not hold, the nonzero mean value
va (t) + vb (t) + vc (t)
v0 (t) = (2.90)
3
is called the zero-sequence component. This represents a common-mode unbalance among the
three phases. If a neutral Y-point (star-point) connection is not used, i.e., the dashed line in
Figure 2.15(a) is not present, then it is guaranteed that ia (t) + ib (t) + ic (t) = 0, even if there
should be a zero-sequence voltage component. That is, {va′ (t), vb′ (t), vc′ (t)} where

va′ (t) = va (t) − v0 (t) vb′ (t) = vb (t) − v0 (t) vc′ (t) = vc (t) − v0 (t) (2.91)
1 In German, three-phase current is very appropriately called Drehstrom—turning current.
52 Chapter 2. Systems and Circuits in Power Electronics and Drives

1
va vb vc

0.5

Positive sequence
0

−0.5

−1
0 π 2π 3π
ω1t

1
v v v
a c b
Negative sequence

0.5

−0.5

−1
0 π 2π 3π
ω1t

Figure 2.14. Positive-sequence and negative-sequence components.

or, in matrix/vector form,

⎡ ⎤ ⎡ ⎤⎡ ⎤
va′ (t) 2 −1 −1 va (t)
⎣ vb′ (t) ⎦ = 1 ⎣ −1 2 −1 ⎦ ⎣ vb (t) ⎦ (2.92)
vc′ (t) 3 −1 −1 2 vc (t)

for which va′ (t) + vb′ (t) + vc′ (t) = 0, give the same phase currents as {va (t), vb (t), vc (t)}. A zero-
sequence component may be freely added to or subtracted from the phase voltages without affecting
the currents, as long as the Y point is left ungrounded. (A neutral wire is normally not needed in
a three-phase system.)
The results are summarized below.

Result 2.1 The unbalanced three-phase system

⎡ ⎤ ⎡ ⎤
va (t) V!a cos(ω1 t + ϕa )
⎣ vb (t) ⎦ = ⎢ ⎥
⎣ V!b cos(ω1 t − 2π/3 + ϕb ) ⎦ (2.93)
vc (t) V!c cos(ω1 t − 4π/3 + ϕc )

can be uniquely decomposed in a positive-sequence component, a negative-sequence component,

2.5. Three-Phase Systems 53

and a zero-sequence component as

⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
va (t) V!+1 cos(ω1 t + ϕ+1 ) V!−1 cos(ω1 t + ϕ−1 ) v0 (t)
⎣ vb (t) ⎦ = ⎢ ⎥ ⎢ ⎥
⎣ V!+1 cos(ω1 t − 2π/3 + ϕ+1 ) ⎦ + ⎣ V!−1 cos(ω1 t − 4π/3 + ϕ−1 ) ⎦ + ⎣ v0 (t) ⎦
vc (t) V!+1 cos(ω1 t − 4π/3 + ϕ+1 ) V!−1 cos(ω1 t − 2π/3 + ϕ−1 ) v0 (t)
(2.94)
where
va (t) + vb (t) + vc (t)
v0 (t) = V!0 cos(ω1 t + ϕ0 ) = . (2.95)
3

2.5.2 Instantaneous Power

As stated initially, one of the benefits of three-phase electric power conversion is that the instan-
taneous power is ideally constant. To show this, let us consider a balanced impedance, i.e., three
identical impedances Z connected to a positive-sequence voltage as given by (2.87); see Figure
2.15(a). This gives a positive-sequence current

ia (t) = I! cos(ω1 t − ϕ)
ib (t) = I! cos(ω1 t − 2π/3 − ϕ) (2.96)
ic (t) = I! cos(ω1 t − 4π/3 − ϕ)

where I! = V! /|Z(jω1 )| and ϕ = arg Z(jω1 ). Through straightforward but tedious trigonometrical
manipulations it is found that the total instantanoues active power P (t) developed in the three
impedances is given by

3V! I!
P (t) = va (t)ia (t) + vb (t)ib (t) + vc (t)ic (t) = cos ϕ
2
= 3V I cos ϕ (2.97)
√ √
where, as in the single-phase case, rms values are introduced as V = V! / 2 and I = I/ ! 2. In
contrast, in a single-phase only the instantaneous power pulsates with the angular frequency 2ω1
about the mean value V I cos ϕ.
PROBLEM 2.16
A relevant question at this is point is whether two-phase systems, such as

vα (t) = V! cos ω1 t, vβ (t) = V! sin ω1 t

would be a feasible alternative to three-phase systems. Simple trigonometry shows that also
in this case, the instantaneous active power is constant. Why are three-phase systems yet
preferable to two-phase systems? Give one reason!

2.5.3 Wye (Y) and Delta (∆) Connections

A balanced three-phase load with per-phase impedance Z can be connected either in a Y (or star)
or in a ∆, as illustrated in Figures 2.15(a) and (b), respectively. Whereas in the Y connection
54 Chapter 2. Systems and Circuits in Power Electronics and Drives

there is a phase voltage across each impedance, the ∆ connection has a line-to-line voltage. From
the geometry in Figure 2.15(c) it is found that the line-to-line voltages are given by
√
vab (t) = va (t) − vb (t) = 3V! cos(ω1 t + π/6)
√
vbc (t) = vb (t) − vc (t) = 3V! cos(ω1 t − π/2) (2.98)
√
vca (t) = vc (t) − va (t) = 3V! cos(ω1 t − 7π/6)
√
still with va (t) as phase reference. The peak value of each line-to-line voltage is 3 times larger
than the peak value of each phase voltage. Thus, the total power developed in a three-phase load
is increased three times when the connection is changed from a Y to a ∆: P∆ = 3PY .

ia Z ia
va
Z vca
vab
vc
ib Z ib 120 30
vb va
vca

Z vbc
vb
in = 0

vbc

ic Z Z vab
vc ic

.
(a): Y connection (b): ∆ connection (c): Phasor diagram

Figure 2.15. Balanced three-phase loads.

It is unimportant from the standpoint of dynamics and control whether a load is connected in
a Y or in a ∆. A balanced ∆-connected load can be treated as if it were connected in a Y, but
with all impedances reduced to 1/3 of the actual values. This is called an equivalent Y.

PROBLEM 2.17
In a Y-connected three-phase load with a 230-V, 50-Hz phase voltage, the total active power
is measured as 4.5 kW whereas the phase current is 9.2 A.

a) What is the inductance per phase if the load is resistive–inductive?

b) What is the capacitance per phase if the load is resistive–capacitive?

PROBLEM 2.18
A load whose resistance per phase is 8 Ω is ∆-connected to a 230-V (rms phase voltage),
50-Hz grid. The load phase voltage is phase shifted 30◦ relative the load phase current.

a) Determine the active and reactive powers as well as the phase current.
b) Is the load inductive or capacitive? Determine the inductance or capacitance.
c) What are the total active power and the line current if the load is instead Y-connected?

PROBLEM 2.19
In a three-phase load with 3 Ω/phase connected to a 400-V (line-to-line voltage), 50-Hz grid,
the total active power is measured as 145 kW and the line current as 440 A.

a) Is the load connected in a Y or in a ∆?

2.5. Three-Phase Systems 55

b) What is the power factor and the inductance per phase of the load?

PROBLEM 2.20
A balanced ∆-connected three-phase impedance is connected as load to a balanced three-
phase grid, whose phase-a voltage is selected as phase reference:
√
va (t) = 230 2 cos ω1 t V

where ω1 = 100π rad/s. In phase b, the following current is measured:

√
ib (t) = 11 2 cos(ω1 t − 55◦ ) A.

Suggest a circuit diagram (with component values) of the impedance per phase of the ∆-
connected load.

2.5.4 Multifrequency Waveforms With Balanced Harmonic Content

Major sources of harmonic pollution in three-phase grids are diode rectifiers and thyristor convert-
ers. They act as balanced but nonlinear loads, which draw balanced but nonsinusoidal currents.
That is, the waveforms in the three phases are identical, only time shifted corresponding to 120◦
of the fundamental. Electrical machines tend to act in a similar way, though they usually draw
less distorted currents. The dominant harmonics thus form a balanced, multifrequency three-phase
system, which can be expressed as the Fourier series expansion (cf. Paragraph 2.1.4)
⎡ ⎤ ⎡ ⎤
va (t) # ⎢ V!h cos[hω1 t + ϕh ]
⎣ vb (t) ⎦ = ⎥
⎣ V!h cos[h(ω1 t − 2π/3) + ϕh ] ⎦ . (2.99)
vc (t) h=1,3,5,... V!h cos[h(ω1 t − 4π/3) + ϕh ]

The 3rd, 5th, and 7th harmonics (with normalized amplitudes V!h = 1 and zero phase angles
ϕh = 0) are shown in Figure 2.16.
As can be seen, the 3rd harmonic has the same phase angle in all three phases, and is thus zero
sequence. It is also seen that the 5th harmonic is negative sequence (cf. Figure 2.14) and the 7th
harmonic is positive sequence. The obvious extension of this is

h= 1 3 5 7 9 11 13 15 · · ·
seq. + 0 − + 0 − + 0 ···

All harmonics that are multiples of 3—so-called triplen harmonics—are zero-sequence components.
Therefore, even if there are triplen voltage harmonics there will be no triplen current harmonics
as long as the Y point ungrounded, cf. Fig. 2.15. In a three-phase multifrequency system with
balanced harmonics but where there may be a fundamental negative-sequence component, the
signed frequency-component orders are as follows:

··· −17 −11 −5 −1 +1 +7 +13 +19 · · · (2.100)

or, more compactly written, ±1 and ±6n + 1, n = 1, 2, . . .

56 Chapter 2. Systems and Circuits in Power Electronics and Drives

phase a
0

−1
0 0.5π π

1
phase b

−1
0 0.5π π

1
phase c

−1
0 0.5π π
ω1t

Figure 2.16. Three-phase harmonics: 3rd (solid), 5th (dashed), and 7th (dash-dotted).

2.6 Space Vectors

2.6.1 Two-Phase Representation With Complex Space Vectors
Previously it was concluded that zero-sequence components often can be disregarded, because if
the Y point is ungrounded a zero-sequence current component is not drawn. The system of three-
phase voltages {va (t), vb (t), vc (t)} is, for practical purposes, equal to a system {va′ (t), vb′ (t), vc′ (t)},
see (2.92), for which
va′ (t) + vb′ (t) + vc′ (t) = 0 ∀t.
This removes one degree of freedom since one of the components—vc′ (t), for example—always can
be expressed in the other two, as vc′ (t) = −[va′ (t) + vb′ (t)]. As a result, it is possible to describe a
three-phase system as an equivalent two-phase system that has two perpendicular axes, denoted
with the subscripts α and β. These axes can be considered as the real and imaginary axes in a
complex plane, allowing a complex representation. The three-phase-to-two-phase transformation—
which is also known as the Clarke transformation [20]—results in the complex space vector
2
vs (t) = vα (t) + jvβ (t) = K[va (t) + ej2π/3 vb (t) + ej4π/3 vc (t)] (2.101)
3
where K is known as the space-vector scaling constant. (We shall discuss the selection of K in
Paragraph 2.6.11.) Figure 2.17 illustrates the construction of a space vector. The contributions
of the phase voltages are equipped with their respective directions (i.e., the directions defined by
2.6. Space Vectors 57

the unit vectors 1, ej2π/3 , and ej4π/3 ), then added together, and, finally, scaled with the constant
2K/3.

β
j2π/3
e
vc e j4π/3
3 vs
vs 2K
va α
j2π/3
vb e

e j4π/3
Figure 2.17. Construction of a voltage space vector.

“Space” in space vector originates from their first usage: to describe two-dimensional spatial
flux distributions in ac machines. “Vector” is used rather than phasor, partly to make a distinction
to phasors and partly because the complex notation of (2.101) sometimes needs to be replaced by
the corresponding real notation in standard two-dimensional vector form (see further Paragraph
2.6.4)
( ) ( )
s vα (t) Re{vs (t)}
v (t) = = . (2.102)
vβ (t) Im{vs (t)}

The superscript s denotes that the space vector is expressed in stationary coordinates. It is also
said that the space vector is referred to the stationary (αβ) frame. When referred to the αβ frame,
the space vector rotates with the same angular frequency as the three-phase quantities which it
represents, i.e., normally the fundamental angular frequency ω1 .

2.6.2 Vector Representation of Positive- and Negative-Sequence Components

Let study the three-phase-to-two-phase transformations of fundamental-frequency positive- and

negative-sequence components. For the positive-sequence component
⎡ ⎤ ⎡ ⎤
va (t) V!+1 cos(ω1 t + ϕ+1 )
⎣ vb (t) ⎦ = ⎢
⎣
⎥
V!+1 cos(ω1 t − 2π/3 + ϕ+1 ) ⎦ (2.103)
vc (t) V!+1 cos(ω1 t − 4π/3 + ϕ+1 )

by direct substitution in (2.101) it is found that the corresponding space vector is given by

vs (t) = K V!+1 ej(ω1 t+ϕ+1 ) . (2.104)

58 Chapter 2. Systems and Circuits in Power Electronics and Drives

It can be observed that the space vector rotates counterclockwise in the αβ plane and that it is
aligned with the phase-a quantity va (t). Transforming the negative-sequence component
⎡ ⎤ ⎡ ⎤
va (t) V!−1 cos(ω1 t + ϕ−1 )
⎣ vb (t) ⎦ = ⎢ ⎥
⎣ V!−1 cos(ω1 t − 4π/3 + ϕ−1 ) ⎦ (2.105)
vc (t) V!−1 cos(ω1 t − 2π/3 + ϕ−1 )

we obtain

vs (t) = K V!−1 ej(−ω1 t+ϕ−1 ) . (2.106)

Also this space vector is aligned with the phase-a quantity, but it rotates clockwise in the αβ
plane. Extension of this principle to harmonics is straightforward. As the 5th harmonic is negative
sequence, the 7th is positive sequence and so on, the space vector of a multifrequency waveform
with balanced harmonics, but possibly including a fundamental negative-sequence component, can
be expressed as
A B
vs (t) = K V!+1 ej(ω1 t+ϕ+1 ) + V!−1 ej(−ω1 t+ϕ−1 ) + V!−5 ej(−5ω1 t+ϕ−5 ) + V!+7 ej(7ω1 t+ϕ+7 ) + · · · .
(2.107)

2.6.3 Relations Between Complex Space Vectors and Rotating Phasors

Complex space vectors and rotating phasors are obviously related, but they are not the same.
Their main similarities and differences are as follows.
• Similarity. Space vector can be regarded as rotating phasors generalized to three-phase
systems. The space vector
vs (t) = K V! ej(ω1 t+ϕ)
corresponds to the three-phase voltage
⎡ ⎤ ⎡ ⎤
va (t) V! cos(ω1 t + ϕ)
⎣ vb (t) ⎦ = ⎢ ⎥
⎣ V! cos(ω1 t − 2π/3 + ϕ) ⎦
vc (t) V! cos(ω1 t − 4π/3 + ϕ)

just as the rotating phasor

V!
v̄(t) = √ ej(ω1 t+ϕ)
2
corresponds to the single-phase voltage (with cosine as base function)

v(t) = V! cos(ω1 t + ϕ).

• Differences.
1) Each phasor normally represents just one frequency component, whereas a complex space
vector can represent several, as shown in (2.107).
2) Whereas the phasor is only a mathematical tool for steady-state analysis of ac circuits,
a complex space vector can be regarded as a complex-valued signal. Although voltages and
2.6. Space Vectors 59

currents etc. are three-phase quantities, after measurement using suitable sensors followed
by sampling, the signals can easily be transformed to two-phase quantities by code in a
microprocessor.
3) For phasors and single-phase ac systems, the concept of negative frequency is not physically
relevant. For example, replacing ω1 in v(t) = sin ω1 t by −ω1 yields v(t) = − sin ω1 t =
sin(ω1 t + π), i.e., just a phase shift. For space vectors negative frequency is very relevant:
positive and negative frequencies correspond to positive and negative sequences, respectively,
cf. (2.107).

Example 2.6 In the three-phase system (2.86) the peak values are

V!a = 591 V V!b = 673 V V!c = 465 V

and the phase angles are

ϕa = 11.7◦ ϕb = −5.7◦ ϕc = −6.5◦ .

We wish to calculate the corresponding space vector. The system is unbalanced, i.e., it contains
both positive- and negative-sequence components. Let us first check that there is no zero-sequence
component. We have

591 cos(ω1 t + 0.204) + 673 cos(ω1 t − 2π/3 − 0.0995)

+465 cos(ω1 t − 4π/3 − 0.113) = 0.58 cos(ω1 t) + 0.25 sin(ω1 t)

so the zero-sequence component is negligible. Using Euler’s relations, we obtain the following space
vector:

vs (t)
(
2 ej(ω1 t+0.204) + e−j(ω1 t+0.204) ej(ω1 t−2π/3−0.0995) + e−j(ω1 t−2π/3−0.0995)
= K 591 + ej2π/3 673
3 2 2
j(ω1 t−4π/3−0.113) −j(ω1 t−4π/3−0.113)
)
e +e
+ej4π/3 465 V (2.108)
2

which through some effort can be simplified to

vs (t) = K[570ejω1 t + 120e−j(ω1 t+1.5) ] V.

PROBLEM 2.21
At a certain time instant, an rms-value-scaled space vector is given as

vs = 130 + j75 V.

A zero-sequence component exists, whose instantaneous value is 9 V. Determine the instan-

taneous values of the phase voltages.
60 Chapter 2. Systems and Circuits in Power Electronics and Drives

2.6.4 Real Space Vectors

A complex space vector, e.g., vs (t) = vα (t) + jvβ (t), is a complex variable. As such, it has a scalar
notation, though with real and imaginary parts. This allows straightforward addition, multiplica-
tion, conjugation etc. using well-known complex arithmetics. Calculation involving complex space
vectors is just somewhat more difficult to perform than calculation involving real scalar variables.
For this reason, complex space vectors are very convenient and should be used whenever possible,
which turns out to be in the majority of situations.
Sometimes, though, the real equivalent to a complex space vector has to be used. This is
usually the case for digital implementation of control algorithms, because programming languages
suitable for fast real-time applications, such as converter and drives control, do not always support
complex variables. Another situation is when there are unbalanced three-phase impedances.
A real space vector is a standard vector with two elements, which is related to its corresponding
complex space vector as shown in (2.102). By evaluating the real and imaginary parts of the three-
phase-to-two-phase transformation (2.101), it is found that the real space vector is obtained by
applying a transformation matrix T32 as
⎡ ⎤
2 1 1 ⎡ ⎤
( ) − − va (t)
vα (t) ⎢ 3 3 3 ⎥
v s (t) = =K⎢⎣
⎥ ⎣ vb (t) ⎦ . (2.109)
vβ (t) 1 1 ⎦
0 √ −√ vc (t)
3 3
. /0 1
T32

For the inverse transformation, i.e., when going from a two-phase representation to a three-phase
representation, the natural choice of transformation matrix T23 is
⎡ ⎤
1 0
⎡ ⎤ ⎢ √ ⎥(
va′ (t) ⎢ 1 3 ⎥ )
⎣ vb (t) ⎦ =
′ 1 ⎢ − ⎥ vα′ (t)
⎢ 2 2 ⎥ (2.110)
′ K⎢ √ ⎥ vβ′ (t)
vc (t) ⎣ ⎦
1 3
− −
. 2/0 2 1
T23

since then T32 T23 = I (where I here is the 2×2 identity matrix), and [vα (t), vβ (t)]T = [vα′ (t), vβ′ (t)]T
if (2.110) is substituted in (2.109). If, on the other hand, (2.109) is substituted in (2.110), we get
⎡ ⎤ ⎡ ⎤⎡ ⎤
va′ (t) 2 −1 −1 va (t)
⎣ vb′ (t) ⎦ = 1 ⎣ −1 2 −1 ⎦ ⎣ vb (t) ⎦ . (2.111)
vc′ (t) 3 −1 −1 2 vc (t)
. /0 1
T23 T32

This relation is identical to (2.92), i.e., the result obtained if the zero-sequence component v0 =
[va (t) + vb (t) + vc (t)]/3 is subtracted from all phase quantities. This shows that, when transform-
ing from a three-phase representation to a two-phase representation and back to a three-phase
representation, all that is lost is the zero-sequence component, should there be any.
2.6. Space Vectors 61

2.6.5 Complex Gains

As will be evident in the chapters to follow, when modeling three-phase circuits or machines as
well as when designing controllers for three-phase VSCs, space vectors are frequently multiplied
by complex gains. Consider the multiplication of the input space vector us = uα + juβ 2 by the
complex gain c = a + jb, forming the output space vector ys = yα + jyβ , as shown in Figure
2.18(a). The multiplication can be resolved in its components, i.e., the real and imaginary parts,
as

ys = cus = (a + jb)(uα + juβ ) = auα − buβ +j (auβ + buα ) . (2.112)

. /0 1 . /0 1
yα yβ

A component-wise block diagram illustrating this multiplication is shown in Figure 2.18(b).

uα a yα
b
s c s
u y b

uβ a yβ
(a)

(b)

Figure 2.18. Block diagrams showing multiplication by the complex gain c = a + jb. (a) Complex
space-vector schematic. (b) Component-wise schematic.

It can be observed that component-wise realization of a complex gain involves four multiplica-
tions and two additions. It can also be noted that the real part of the complex gain gives a direct
mapping: the d component of the input vector maps to the d component of the output vector with
gain a, and similarly for the q component. On the other hand, the imaginary part of the complex
gain gives a cross coupling: the d component of the input vector maps to the q component of the
output vector with gain b, whereas the q component of the input vector maps to the d component
of the output vector with gain −b. We thus find that the real correspondence to (2.112) can be
expressed using a matrix gain C as
( ) ( )( )
yα a −b uα
= . (2.113)
yβ b a uβ
. /0 1 . /0 1 . /0 1
ys C us

A 2 × 2 matrix gain which can be expressed in this fashion, i.e., the diagonal elements are equal,
whereas the antidiagonal elements are equal but with opposite signs, always has a corresponding
complex gain. We shall call this property that the gain is symmetrical, see further Section 2.7.
The special cases of a real gain (b = 0) and an imaginary gain (a = 0) are illustrated in Figure
2.19. The number of multiplications here reduce to two and the additions vanish.

2 For the sake of simplicity, from here on—with a few exceptions—we drop the time argument “(t).”
62 Chapter 2. Systems and Circuits in Power Electronics and Drives

uα a yα uα a =0 yα
b =0 b
b =0 b

uβ a yβ uβ a =0 yβ

(a) (b)

Figure 2.19. Block diagrams showing component-wise schematics of the special cases (a) real gain and
(b) imaginary gain.

Remark 2.4 The correspondence between the complex gain c and the matrix gain C becomes
yet more clear by the introduction of the matrix equivalent to j
( )
0 −1
J= . (2.114)
1 0

J used as a matrix gain yields the mapping

( )( ) ( )
0 −1 uα −uβ
= (2.115)
1 0 uβ uα
. /0 1 . /0 1
J us
which is a direct correspondence to the mapping obtained by the corresponding complex gain j

jus = j(uα + juβ ) = −uβ + juα . (2.116)

Using (2.114) allows us to express C in (2.113) as

( ) ( ) ( )
1 0 0 −1 a −b
C = aI + bJ = a +b = (2.117)
0 1 1 0 b a

showing clearly the correspondence to c = a + jb.

2.6.6 Coordinate Transformations

Consider a fundamental positive-sequence space vector:

vs = K V! ej(ω1 t+ϕ) . (2.118)

The transformation

v = e−jω1 t vs = K V! ejϕ (2.119)

removes the rotation of the vector, making it similar to a fixed phasor, cf. (2.5). This is called
the dq transformation (which is also known as the Park transformation) and can be regarded
as observing the space vector from a coordinate system rotating with the angular fundamental
frequency ω1 . This coordinate system is called synchronous coordinates, and it is said that space
2.6. Space Vectors 63

vector v is referred to the synchronous (dq) frame. See Figure 2.20. We shall denote a space vector
in synchronous coordinates without a superscript, and its components with the subscripts d and
q:

v = vd + jvq . (2.120)

Since the rotation of the space vector is removed—provided that vs contains no other frequency
components than the fundamental positive-sequence component—its components vd and vq will be
constant in the steady state. This fact makes the dq transformation not useful just for analysis,
e.g., of electrical machines, but also for implementation of controllers for three-phase VSCs. This
is because it is easier to design controllers for quantities that are constant—rather than constantly
oscillating—in the steady state. PI controllers, with some additions, can generally be relied upon.

β
q vs
vβ
d
vq vd
θ1 α
vα

Figure 2.20. Projections of the vector vs on the αβ and dq frames.

Equation (2.118) implies that ω1 is constant, but this is rarely the case, particularly not for a
variable-speed drive. To allow a varying fundamental frequency, the general definitions of the dq
transformation and its inverse, the αβ transformation, are given by

v = e−jθ1 vs (dq transformation) (2.121)

s jθ1
v = e v (αβ transformation) (2.122)

where
$
θ1 = ω1 dt. (2.123)

2.6.7 dq Transformation of Multifrequency Waveforms

Applying the dq transformation to the multifrequency space vector (2.107) gives
A B
v = K V!+1 ejϕ+1 + V!−1 ej(−2ω1 t+ϕ−1 ) + V!−5 ej(−6ω1 t+ϕ−5 ) + V!+7 ej(6ω1 t+ϕ+7 ) + · · · . (2.124)

The 5th and 7th harmonics are both transformed to 6th harmonics in synchronous coordinates
(which are negative and positive sequence, respectively). Similarly, the 11th and 13th harmonics
64 Chapter 2. Systems and Circuits in Power Electronics and Drives

are both transformed to 12th harmonics. In the dq frame, balanced harmonics thus have the
angular frequencies 6nω1 , n = 1, 2, . . . The fundamental-frequency negative-sequence component
is transformed to −2. Thus, the orders as given with sign are transformed as
αβ ··· −11 −5 −1 +1 +7 +13 · · ·
↓ ↓ ↓ ↓ ↓ ↓ ↓ (2.125)
dq ··· −12 −6 −2 0 +6 +12 · · ·

2.6.8 Coordinate Transformations of Real Space Vectors

The dq transformation (2.121) can be resolved in the components of the αβ- and dq-frame space
vectors, as well as in the real and imaginary parts of the transformation factor e−jθ1 , as
v = e−jθ1 vs = (cos θ1 − j sin θ1 )(vα + jvβ )
= vα cos θ1 + vβ sin θ1 +j (−vα sin θ1 + vβ cos θ1 ) . (2.126)
. /0 1 . /0 1
vd vq
The underbraced component identities can be expressed with the help of a dq transformation
matrix Tdq , giving the following equivalence to (2.126) for the corresponding real space vectors:
( ) ( )( )
vd cos θ1 sin θ1 vα
= . (2.127)
vq − sin θ1 cos θ1 vβ
. /0 1 . /0 1 . /0 1
v Tdq vs

Since αβ transformation involves multiplication by the factor ejθ1 , the αβ transformation matrix
Tαβ is obtained by substituting θ1 → −θ1 in Tdq , giving
( ) ( )( )
vα cos θ1 − sin θ1 vd
= . (2.128)
vβ sin θ1 cos θ1 vq
. /0 1 . /0 1 . /0 1
vs Tαβ v
−1 T
Note that Tαβ = Tdq = Tdq ; Tdq and Tαβ are orthogonal matrices. A block diagram for the αβ
transformation is shown in Figure 2.21. Since the αβ transformation ejθ1 factor is a complex gain,
Figure 2.21 obviously is a special case of Figure 2.18(b) with a = cos θ1 and b = sin θ1 .

vd vα

vq vβ

cos θ1 sin θ1
.

Figure 2.21. Block diagram for the αβ transformation.

2.6. Space Vectors 65

PROBLEM 2.22
How should Figure 2.21 be modified in order to obtain the dq transformation?

PROBLEM 2.23
Using the matrix correspondence J to the imaginary unit j, as given by (2.114), should
logically allow the correspondence

vs = ejθ1 v ⇔ v s = eJ θ1 v.

That is, eJ θ1 = Tαβ . Show that! Hint: L{eAt } = (sI − A)−1 .

2.6.9 Instantaneous Power

Equation (2.8) states the well-known relation between the phasors for voltage and current and
the complex power in a single-phase system. Based upon this, it may be conjectured that the
∗
complex power in a three-phase system is proportional to vs (is ) . Taking respectively the real
and imaginary parts should then give the active and reactive powers. Proving this conjecture only
requires some algebraic exercise. From the definition (2.101) of a complex space vector, we have
, -2 2 32 3∗
s s ∗ 2K
v (i ) = va + ej2π/3 vb + ej4π/3 vc ia + ej2π/3 ib + ej4π/3 ic . (2.129)
3

Observing that ej4π/3 = e−j2π/3 and assuming that there is no zero-sequence current component,
i.e., ia + ib + ic = 0, we obtain
⎡

, -2 ⎢
⎢
> √ ?
2K ⎢ 1 3
vs (is )∗ = ⎢va ia + vb ib + vc ic + − + j (va ic + vb ia + vc ib )
3 ⎢ 2 2
⎣ . /0 1
ej2π/3
⎤
> √ ? ⎥
1 3 ⎥
⎥
+ − −j (va ib + vb ic + vc ia )⎥
2 2 ⎥
. /0 1 ⎦
−j2π/3
e
⎡ ⎛ ⎞
, -2
2K ⎢ 1⎜ ⎟
= ⎣va ia + vb ib + vc ic − ⎝va (ib + ic ) +vb (ia + ic ) +vc (ia + ib )⎠
3 2 . /0 1 . /0 1 . /0 1
−ia −ib −ic
√ I
3
+ j (va (ic − ib ) + vb (ia − ic ) + vc (ib − ia ))
2
( )
2K 2 1
= va ia + vb ib + vc ic + j √ (va (ic − ib ) + vb (ia − ic ) + vc (ib − ia )) . (2.130)
3 3
The instantaneous active power developed in a three-phase circuit is the sum of the instantaneous
active powers per phase, i.e., va ia +vb ib +vc ic . This shows that (2.130) should be scaled by 3/(2K 2)
66 Chapter 2. Systems and Circuits in Power Electronics and Drives

to obtain the complex power for all three phases, i.e.,

3 s s ∗ 3
S= v (i ) = vi. (2.131)
2K 2 2K 2
[Due to the complex conjugation of the current space vector, the expression is reference-frame
invariant. Thus, as shown in (2.131), the relation for the corresponding dq-frame space vectors is
formally identical to that for the αβ-frame space vectors.] So, the instantaneous active power is
given by
3 3
P = Re{S} = Re {vs (is )∗ } = Re {vi∗ } = va ia + vb ib + vc ic (2.132)
2K 2 2K 2
whereas the instantaneous reactive power is given by
3 3
Q = Im{S} = Im {vs (is )∗ } = Im {vi∗ }
2K 2 2K 2
1
= √ [va (ic − ib ) + vb (ia − ic ) + vc (ib − ia )] . (2.133)
3
It should be stressed that (2.131)–(2.133) give the instantaneous powers, unlike the corresponding
relations for phasors representing single-phase systems, from which just the mean powers are
obtained. In the ideal situation where both voltage and current consist of fundamental positive-
sequence components
√ √
vs = K 2V ejω1 t is = K 2Iej(ω1 t+ϕ) (2.134)
(with the voltage taken as phase reference, for convenience), the instantaneous complex power is
given by
3 s s ∗ 3 √ √ jϕ
S= v (i ) = 2V 2Ie = 3V I cos ϕ +j 3V I sin ϕ (2.135)
2K 2 2 . /0 1 . /0 1
P Q
which is constant. Space vectors allows this fact to be concluded with considerably less work
involved than the phase-quantity calculations needed to obtain the identical expression (2.97) for
P.

2.6.10 Power of Multifrequency Waveforms

Suppose that the balanced voltage vs = K V! ejω1 t is applied to an unbalanced and nonlinear
load (this could be a diode rectifier or an ac machine connected to the grid via an unbalanced
impedance), which draws a current containing a fundamental negative-sequence component plus
balanced harmonics, cf. (2.107)
A B
is = K I!+1 ej(ω1 t+ϕ+1 ) + I!−1 ej(−ω1 t+ϕ−1 ) + I!−5 ej(−5ω1 t+ϕ−5 ) + I!+7 ej(7ω1 t+ϕ+7 ) + · · · . (2.136)

The resulting active power is given by

3 3 3 ! J K
P = 2
Re {vs (is )∗ } = 2
Re {(vs )∗ is } = V Re e−jω1 t is
2K L 2K 2K M
jϕ+1 j(−2ω1 t+ϕ−1 )
= 3V Re I+1 e + I−1 e + I−5 ej(−6ω1 t+ϕ−5 ) + I+7 ej(6ω1 t+ϕ+7 ) + · · ·
(2.137)
2.6. Space Vectors 67

√
where V = V! / 2, etc. It may be noted that the factor (vs )∗ in effect gives a dq transformation of
is , which results in multifrequency components similar to (2.124). The following observations can
be made.
1) Only the fundamental positive-sequence current component (with rms value I+1 ) contributes
to the constant term; the fundamental negative-sequence component and the harmonics just
cause pulsations.
2) The pulsations are of the angular frequencies 2ω1 and 6nω1 , n = 1, 2, . . .
Pulsations at 2ω1 are generally more disturbing than those at higher frequencies. In a machine,
power pulsations translate to torque pulsations, and low-frequency torque pulsations propagate
easier and are more prone to excite mechanical resonances than high-frequency torque pulsations.
Assuring a balanced impedance in the grid connection gives I−1 = 0. The lowest angular frequency
of pulsation then becomes 6ω1 rather than 2ω1 . If a 3rd harmonic had existed—which might have
been the case for a system with, say, two or five phases—the lowest angular frequency of pulsation
could yet have been 2ω1 . Hence, using either less or more phases than three would aggravate the
problem of power pulsations due to harmonics, which provides yet more evidence that the choice
to use three phases for electric power is wise.
Remark 2.5 There is a similarity between the pulsating versus ideally constant powers of single-
versus three-phase systems and the smoother torque of a three-cylinder combustion engine as
compared to a single-cylinder engine. Unlike the number of phases, however, adding yet more
cylinders gives even smoother torque.

2.6.11 Selection of the Space-Vector Scaling Constant

The scaling constant K can be chosen arbitrarily. Depending on the situation, one choice may be
more convenient than another. There are three standard selections. Selecting K = 1 we obtain for
a fundamental positive-sequence component
vs = V! ej(ω1 t+ϕ) v = V! ejϕ (2.138)
√
so K = 1 gives peak-value scaling. K = 1/ 2 yields
V! V!
vs = √ ej(ω1 t+ϕ) v = √ ejϕ . (2.139)
2 2
That is, rms-value scaling. The expression
" (2.132) for the instantaneous complex power then reads
S = 3vi∗ . To remove the factor 3, K = 3/2 should instead be chosen, giving
S = vi∗ . (2.140)
"
K = 3/2 is therefore called the choice for power invariance.
To reiterate.
Peak-value scaling: K =1 √
RMS-value scaling: " 2
K = 1/
Power-invariant scaling: K = 3/2.
Although all formulas derived henceforth are invariant of the choice of K, rms-value scaling is used
in most examples.
" √ √
Remark 2.6 Because 3/2 is 3 times larger than 1/ 2, power-invariant scaling of voltages can
be regarded also as scaling with respect to the line-to-line voltage.
68 Chapter 2. Systems and Circuits in Power Electronics and Drives

2.7 Symmetrical Linear Systems

In Paragraph 2.6.5, the symmetry property of a complex gain was mentioned. In this section the
symmetry property is extended to linear dynamic systems with space vectors as input and output
signals.

2.7.1 Complex System Models

In Paragraph 2.6.3, one of the noted differences between complex space vectors and phasors was
that a complex space vector can indeed be considered as a complex signal, not just as a tool
for calculation. As such, it will pass through dynamic systems which can be described with
complex models. These are, on the one hand, models of three-phase circuits, machines, etc. On
the other hand, they may also be controllers, filters, and other algorithms for processing of space-
vectors signals in a converter control system. Let us consider the perhaps most important example,
a balanced impedance [see Figure 2.15(a)] with inductance L per phase. The phase quantities
obviously obey the law of induction
dia dib dic
va = L vb = L vc = L . (2.141)
dt dt dt
This would seem to form a third-order dynamic system. However, as the phase currents normally
add up to zero, they are not linearly independent. As a result, the effective system order is reduced
to two, which makes the following equivalent two-phase representation an appropriate model:
diα diβ
vα = L vβ = L . (2.142)
dt dt
In complex space-vector form, the model reduces further to
dis
vs = L (2.143)
dt
which also can be represented by the complex Laplace impedance
Zs (s) = sL. (2.144)

Because this impedance determines the relationship between the complex space vectors for voltage
and current, we denote it as and call it a complex impedance, even though it is real valued for a real
s. Furthermore, although (2.144) appears to be a system of order one, it is really a second-order
system, being but a compact form of (2.142). This applies to all complex system models.

The true system order is twice the order of the complex system model.
Figure 2.22 illustrates the steps in the modeling process.

Example 2.7 We wish to calculate the response in the current is when the voltage

vs = V ejω1 t
is applied at t = 0 to a series connection of an inductor L and a resistor R per phase (connected
in a Y). Assuming that is (0) = 0 and taking the Laplace transform of vs yields
V
Vs (s) = .
s − jω1
2.7. Symmetrical Linear Systems 69

L
va
L
L vα L
vb vs
L vβ L
vc
(a) (b) (c)
Figure 2.22. (a) Balanced inductors with ungrounded Y point. (b) Equivalent two-phase model. (c)
Equivalent complex-space-vector model.

The Laplace transform of the current is found by division by the impedance R + sL

, -
V V 1 1
Is (s) = = − .
(R + sL)(s − jω1 ) R + jω1 L s − jω1 s + R/L
Inverse transformation results in
V 2 3
is = ejω1 t − e−Rt/L , t ≥ 0.
R + jω1 L
V
The steady-state part is ejω1 t , which can be obtained also from phasor calculation simply
R + jω1 L
by dividing by the complex impedance R + jω1 L.
♣

2.7.2 Complex Transfer Functions

The Laplace impedance (2.144) is a special case of a general complex transfer function. Such are
used for modeling of linear and time-invariant dynamic systems which have space vectors as input
and output signals. A general complex transfer function may be written (if referred to the αβ
frame) as

Hs (s) = Hα (s) + jHβ (s) (2.145)

and it may operate on the input space phasor us , giving the output space vector ys as

ys = Hs (p)us . (2.146)

In component form, (2.146) can be expressed as

ys = [Hα (p) + jHβ (p)](uα + juβ )
= Hα (p)uα − Hβ (p)uβ +j [Hα (p)uβ + Hβ (p)uα ] (2.147)
. /0 1 . /0 1
yα yβ
which implies the following correspondence for real space vectors:
( ) ( )( )
yα Hα (p) −Hβ (p) uα
= . (2.148)
yβ Hβ (p) Hα (p) uβ
. /0 1 . /0 1 . /0 1
ys H s (p) us
70 Chapter 2. Systems and Circuits in Power Electronics and Drives

uα yα
Hα(s)
us ys Hβ(s)
H s(s)
Hβ(s)
(a) uβ yβ
Hα(s)
(b)

Figure 2.23. Block diagrams showing the complex transfer function Hs (s) = Hα (s) + jHβ (s). (a)
Complex space-vector schematic. (b) Component-wise schematic.

Figure 2.23 illustrates the principles. Notice the similarity to Figure 2.18.

From (2.148), two properties of complex transfer functions can be found.

1) Systems which are described by complex transfer functions are multivariable [77] with two
inputs and two outputs, the respective α and β components.
2) H s (s) is a special case of a general 2 × 2 transfer function matrix [77] with the restriction
that the diagonal elements are equal, whereas the antidiagonal elements are equal but with
a sign change (similarly to a complex gain).
A multivariable system that has these properties is referred to as symmetrical (also known as
isotropic or rotational invariant). Examples of symmetrical systems are balanced impedances and
round-rotor ac machines (see further Chapter 5). Even though they are multivariable, symmetrical
systems can always be modeled using complex transfer functions. This significantly reduces the
complexity of modeling, analysis, and design (e.g., of controllers) as compared to using transfer
function matrices. Complex transfer functions can be handled with almost the same ease as real
transfer functions, but they have some different properties.
• Unlike a real transfer function, the poles and zeros of a complex transfer function do not
necessarily appear in conjugated pairs.
• Unlike a real transfer function H(s), the frequency response of a complex transfer function
is not necessarily symmetrical about ω = 0 as H(−jω) = H ∗ (jω). This property implies
that positive- and negative-sequence components of the same frequency not necessarily have
equal gains.
These properties are illustrated in the following example.
Example 2.8 Complex bandpass filter. Suppose that a three-phase current contains multi-
frequency components as
A B
is = K I!+1 ej(ω1 t+ϕ+1 ) + I!−1 ej(−ω1 t+ϕ−1 ) + I!−5 ej(−5ω1 t+ϕ−5 ) + I!+7 ej(7ω1 t+ϕ+7 ) + · · · .

We are interested in monitoring the fundamental positive-sequence component and therefore want
to filter out all other components, giving the filtered space vector ys as shown in Figure 2.24.
The obvious solution would be to insert identical bandpass filters, with center frequency ω1 , in
the α and β channels. However, while this filtering would reject the harmonics, it would admit the
2.7. Symmetrical Linear Systems 71

ia
ib
ic

3 is ys
2 filter

Figure 2.24. Filtering of measured current signal.

fundamental negative-sequence component. To reject also this component, a complex bandpass

filter is needed. The simplest possible such filter is obtained by first considering a first-order
low-pass filter
a
HLP (s) = .
s+a
This filter attenuates frequencies outside the passband −a < ω < a by 20 decibels per decade. The
desired bandpass filter is obtained simply by translating the passband to be centered at ω = ω1
instead of ω = 0. This translation is obtained by substituting s → s − jω1 , giving
a
HBP (s) = HLP (s − jω1 ) = .
s + a − jω1
We now have HBP (jω1 ) = 1, whereas frequencies outside the passband ω1 − a < ω < ω1 + a
are attenuated by 20 decibels per decade. Note also that there is one complex pole—not a pole
pair—at s = −a + jω1 . It is instructive to find a realization of the filter in the form of a block
diagram consisting of adders, multipliers, and integrators. (Such a realization is the basis for
analog implementation using resistive–capacitive networks built around operational amplifiers. It
can also be the basis for digital implementation, as will be shown in Paragraph 2.7.4.) For this
sake, we rewrite the input–output-signal relation as
a
ys = is ⇒ (p + a − jω1 )ys = ais ⇒ pys = a(is − ys ) + jω1 ys .
p + a − jω1
Finally, we divide the relation by p, obtaining
1
ys = [a(is − ys ) + jω1 ys ].
p
The block diagram which corresponds to this equation is shown in Figure 2.25(a). The filter consists
of two adders, two multipliers (one real, by a, and one imaginary, by jω1 ), and one integrator, all
operating on complex space-vector signals. The corresponding component-wise block diagram is
shown in Figure 2.25(b). The numbers of adders, multipliers, and integrators have now doubled.
Because there are now two integrators, the true system order is revealed to be two, whereas that
of the complex model is just one.

♣
Example 2.9 Unbalanced inductances. Suppose that instead of the balanced inductances
given by (2.141) we have unbalanced inductances as
dia dib dic
va = La vb = Lb vc = Lc
dt dt dt
72 Chapter 2. Systems and Circuits in Power Electronics and Drives

a
is ys
1/s

j ω1

(a)

iα a yα
1/s

ω1

ω1

iβ a yβ
1/s
(b)

Figure 2.25. Block diagrams of complex bandpass filter with angular center frequency ω1 . (a) Complex
space-vector schematic. (b) Component-wise schematic.

or in vector form ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
va La 0 0 ia
⎣ vb ⎦ = ⎣ 0 d
Lb 0 ⎦ ⎣ ib ⎦ .
dt
vc 0 0 Lc ic
Transformation to a two-phase representation using (2.109) and (2.110) yields
⎡ ⎤
⎡ ⎤ 1 0
2 1 1 ⎡ ⎤ ⎢ √ ⎥
⎢ − − ⎥ La 0 0 ⎢ 1 3 ⎥
3 3 3 1 ⎢ − ⎥ dis
vs = K ⎢ ⎥ ⎣ 0 Lb 0 ⎦ ⎢ 2 ⎥ dt
⎣ 1 1 ⎦ K⎢ 2 √ ⎥
0 √ −√ 0 0 Lc ⎣ ⎦
3 3 1 3
. /0 1 − −
. 2/0 2 1
T32
T23
which can be simplified to
⎡ 4L + L + L Lc − Lb ⎤
a b c
√
s ⎢ 6 2 3 ⎥ dis
v =⎣ Lc − Lb Lb + Lc ⎦ .
√ dt
2 3 2
This inductance matrix fulfills the requirements for a symmetrical system only when La = Lc = Lc ,
i.e., for balanced inductances, when it becomes diagonal.
♣
2.7. Symmetrical Linear Systems 73

2.7.3 Coordinate Transformations of Symmetrical Systems

Transformation of αβ-frame quantities to the dq frame is convenient, as previously noted, since the
quantities then become constant—rather than constantly oscillating—in the steady state. Con-
sequently, it is useful to model and also to implement symmetrical systems that are dq-frame
equivalents of αβ-frame systems. Let us start by transforming the time derivative of a space vector
from stationary coordinates to synchronous coordinates. Let ys be a general αβ-transformed space
vector

ys = ejθ1 y. (2.149)

Taking the time derivative yields, using the chain rule

, -
dys d(ejθ1 y) jθ1 dy
= =e jω1 y + . (2.150)
dt dt dt
That is, a term jω1 y is added in the dq frame, which introduces a cross coupling between the
d and q axes, cf. Figure 2.19(b) with b = 1. For a constant ω1 the Laplace transform can be
applied, showing that when transforming from stationary to synchronous coordinates the following
substitution thus should be made:

s → s + jω1 . (2.151)

This is a logical result; the added term accounts for the synchronous rotation associated with the
dq transformation. One implication thereof is that the Laplace impedance of an inductor—cf.
(2.144)—when referred to the dq frame is

Z(s) = (s + jω1 )L (2.152)

which is illustrated in Figure 2.26.

s
Z = sL Z = (s + jω1) L
vs is v i
(a) (b)
Figure 2.26. Complex inductor impedance in (a) stationary coordinates and (b) synchronous coordinates.

Example 2.10 Consider again the series-connected three-phase inductor and resistor in Example
2.7. We wish to verify the result by calculating in synchronous coordinates the current is resulting
from the application of the voltage vs = V ejω1 t at t = 0. Transformation to synchronous coordi-
nates yields v = V , t ≥ 0, whose Laplace transform is V(s) = V /s. The impedance is in this case
R + (s + jω1 )L, giving
, -
V V 1 1
I(s) = = − .
s[R + (s + jω1 )L] R + jω1 L s s + jω1 + R/L
Inverse Laplace transformation yields
V A B
i= 1 − e−(jω1 +R/L)t , t ≥ 0
R + jω1 L
74 Chapter 2. Systems and Circuits in Power Electronics and Drives

and αβ transformation finally gives the same result as in Example 2.7

V 2 3
is = ejω1 t i = ejω1 t − e−Rt/L , t ≥ 0.
R + jω1 L
♣
The property given by (2.151) implies that a dq-frame correspondence to any symmetrical αβ-
frame dynamic system is straightforward to obtain. Consider a symmetrical system with complex
transfer function Hs (s), whose input and output signals respectively are us and ys . If another
system is constructed as the dq transformation of Hs (s) according to (2.151)

H(s) = Hs (s + jω1 ) (2.153)

and whose input signal is the dq transformation of us , then the system’s αβ-transformed output
signal is identical to ys . This is illustrated in Figure 2.27; the configuration in Figure 2.27(b) is
sometimes called the rotator-accompanied equivalent to the configuration in Figure 2.27(a), owing
to the two coordinate transformations.

us ys us u y ys
H s(s) H (s)
jθ1
e e jθ1
(a) (b)

Figure 2.27. (a) Symmetrical system in the αβ frame and (b) its dq-frame equivalent [provided that
H(s) = Hs (s + jω1 ), θ̇1 = ω1 ].

An αβ-frame equivalent to a dq-frame symmetrical system can obviously be obtained as

Hs (s) = H(s − jω1 ). (2.154)

Example 2.11 Complex dq-frame-equivalent bandpass filter. In Example 2.25 we derived

a complex αβ-frame bandpass filter by first considering a low-pass filter, and then realizing that it
can be modified into the desired bandpass filter by substituting s → s − jω1 . That is, we effectively
applied (2.154) without yet having derived the relation. Instead of making this substitution, the
low-pass filter HLP (s) = a/(s + a) can be realized in the dq frame and accompanied with rotators,
as shown in Figure 2.28(b).

a a
is ys is i y ys
1/s 1/s
jθ1
j ω1 e e jθ1
H LP(s)
(a) (b)

Figure 2.28. (a) αβ-frame complex bandpass filter and (b) its rotator-accompanied dq-frame equivalent.
2.7. Symmetrical Linear Systems 75

The original αβ-frame filter, shown in Figure 2.28(a), is less costly to implement than its dq-
frame equivalent, however. As found in Example 2.25, the former requires four adders and four
multipliers. The latter, on the other hand, requires six adders and ten multipliers. This is because
each complex multiplication, i.e., in this case the two coordinate transformations, require four
multiplications and two additions, cf. Figure 2.18(b).

dq Transformation of a Time Delay

Another important implication of (2.151) is how an αβ-frame time delay

ys (t) = us (t − Td ) ⇔ ys (t) = e−pTd us (t) (2.155)

transforms to the dq frame. Applying (2.151) to (2.155) yields

y(t) = e−(p+jω1 )Td u(t) ⇔ y(t) = e−jω1 Td u(t − Td ) (2.156)

i.e., the time delay is transformed to a time delay and a complex gain e−jω1 Td , which effectively
causes a rotation of the dq frame by the angle −ω1 Td . If the time delay is known, this rotation
can be compensated by pre-multiplying u or post-multiplying y by the complex gain ejω1 Td .

2.7.4 Discrete-Time Symmetrical Systems

So far, we have considered solely continuous-time symmetrical systems. Extension of the principles
to discrete-time symmetrical systems is straightforward.

Discretization
An discussed in Paragraph 2.4.7, Euler discretization has to be applied with caution, since poorly
damped continuous-time systems may result in unstable discretizations and an accurate mapping
of the frequency response is obtained only for lower frequencies. We illustrate by an example.

Example 2.12 Euler discretized complex bandpass filters. Let us discretize the complex
αβ-frame in Example 2.25 together with its rotator-accompanied dq-frame equivalent in Example
2.11. Direct substitution s → (z − 1)/Ts in the block diagrams of Figure 2.28 yields the block
diagrams shown in Figure 2.29.

a a
is Ts ys is i Ts y ys
z−1 z−1
jθ1
j ω1 e e jθ1
H LP(z)
(a) (b)

Figure 2.29. Euler discretizations of (a) αβ-frame complex bandpass filter and (b) its rotator-accompanied
dq-frame equivalent.
76 Chapter 2. Systems and Circuits in Power Electronics and Drives

Let us analyze the obtained discrete-time systems. Transfer function HLP (z) in Figure 2.29(b)
is given by
aTs
aTs
HLP (z) = z−1 aTs
=
1 + z−1 z − 1 + aTs

whose pole is located at z = 1 − aTs . Thus, HLP (z) is asymptotically stable as long as a < 2/Ts ,
which agrees with the previously found stability criterion for the Euler method: the poles of the
system to be discretized must be located within a circle with radius 1/Ts centered at s = −1/Ts .
The transfer function HBP (z) from is to ys in Figure 2.29(a) is given by
aTs
z−1 aTs
HBP (z) = (a−jω1 )Ts
= .
1+ z − 1 + (a − jω1 )Ts
z−1

The pole of HBP (z) is located at z = 1 − aTs + jω1 Ts . That is, it is located within the unit circle
if

|1 − aTs + jω1 Ts | < 1 ⇒ (1 − aTs )2 + (ω1 Ts )2 < 1 ⇒

1" 1"
ω1 < 1 − (1 − aTs )2 = 2aTs − (aTs )2 . (2.157)
Ts Ts
This criterion is often not difficult to fulfill. Suppose that a sampling frequency of 10 kHz is used,
i.e., Ts = 0.1 ms, that ω1 = 314 rad/s, and a = 0.1ω1 = 31.4 rad/s. Then, the right-hand side
of (2.157) equals 792 rad/s, so the filter is asymptotically stable. Remaining is to evaluate the
frequency response. The fundamental frequency should ideally be admitted with unity gain and
zero phase shift, but we get

aTs
HBP (ejω1 Ts ) = .
ejω1 Ts − 1 + (a − jω1 )Ts
◦
If the given numerical values are plugged into the expression, then HBP (ejω1 Ts ) = 1.19ej0.11 ,
which shows that Euler discretization in this case is inappropriate, even though the discretized
system is rendered stable.

Coordinate Transformations
We know from (2.151) that a dq transformation of an αβ-frame symmetrical system is obtained by
substituting s → s + jω1 . This substitution can be directly applied to the relation (2.66) between
the variables s and z, i.e., z = esTs . Applying (2.151) to (2.66), we get

esTs .
z → e(s+jω1 )Ts = ejω1 Ts ./01 (2.158)
z
That is, dq transformation of an αβ-frame discrete-time symmetrical system is obtained by the
substitution

z → ejω1 Ts z (2.159)
2.8. Normalized (Per-Unit) Values 77

whereas αβ transformation of an dq-frame discrete-time symmetrical system is obtained by the

substitution
z → e−jω1 Ts z. (2.160)
We illustrate the principle by an example.
Example 2.13 αβ transformation of an Euler discretized lowpass filter. We now continue
Example 2.12 with the objective of rectifying the partially failed Euler discretization of the αβ-
frame system, see Figure 2.29(a). Instead of direct Euler discretization in the αβ frame, we make
an αβ transformation of the Euler discretized dq-frame low-pass filter shown in Figure 2.29(b).
Applying (2.160) to HLP (z) yields
aTs aTs ejω1 Ts
HBP (z) = HLP (e−jω1 Ts z) = = .
e−jω1 Ts z− 1 + aTs z − (1 − aTs )ejω1 Ts
The pole is now located at z = (1 − aTs )ejω1 Ts , which is within the unit circle as long as a < 2/Ts .
Moreover, HBP (ejω1 Ts ) = 1, as desired. To find a realization of the filter, we write the input–output
relation in the shift operator
aTs ejω1 Ts
ys = is ⇒ qys = ejω1 Ts [aTs is + (1 − aTs )ys ] ⇒
q − (1 − aTs )ejω1 Ts
ys = q −1 ejω1 Ts [aTs (is − ys ) + ys ]
which corresponds to the block diagram shown in Figure 2.30. This is an equivalent to the αβ-
frame realization shown Figure 2.29(b), whereas that shown in Figure 2.29(a) is not. Magnitude
and phase diagrams for ω1 = 314 rad/s, a = 0.1ω1 = 31.4 rad/s, and Ts = 0.1 ms are shown
in Figure 2.31. It can be observed that the αβ-transformed Euler discretized low-pass filter has
frequency-domain properties which are near identical to those if the underlying continuous-time
filter.

aTs e jω1Ts
is ys
z−1

Figure 2.30. Complex bandpass filter obtained by αβ transformation of an Euler discretized lowpass
filter.

2.8 Normalized (Per-Unit) Values

It is common practice to use normalized or per-unit (pu) quantities in electric power engineering.
Normalization is done with respect to a set of base values, which account for the nominal (rated)
operation of a system.
78 Chapter 2. Systems and Circuits in Power Electronics and Drives

1.4

1.2

0.8
|
BP
|H
0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
ω/ω1

100

50
(degrees)

0
BP
arg H

−50

−100
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
ω/ω
1

Figure 2.31. Magnitude and phase functions for (solid) continuous-time bandpass filter, (dashed) Euler
discretized bandpass filter, and (dash-dotted) αβ-transformed Euler discretized low-pass filter.

2.8.1 Single-Phase and DC Systems

The fundamental base values are selected equal to the nominal values (the latter denoted by the
subscript N ):
Voltage: Vbase = VN
Current: Ibase = IN (2.161)
Frequency: ωbase = ωN .
Using the fundamental base values, normalized quantities (denoted by the subscript n) are intro-
duced as follows:
Voltage: vn = v/Vbase
Current: in = i/Ibase
(2.162)
Angular frequency: ωn = ω/ωbase
Time: tn = t/tbase = ωbase t.
Then, in = 1 pu, for example, tells that nominal current is flowing. From (2.161), the following
additional base values (and others) can be derived:
Power: Pbase = Vbase Ibase
Impedance: Zbase = Vbase /Ibase
(2.163)
Inductance: Lbase = Zbase /ωbase
Capacitance: Cbase = 1/(ωbase Zbase ).
2.8. Normalized (Per-Unit) Values 79

For an ac system, Pbase = Qbase = Sbase , where the latter respectively are the base values for
reactive and apparent powers.
Although normalization should not be overdone (for example, equations which are valid only
in per-unit quantities are not recommended, or should at the very least be used sparingly), there
are several benefits. Regardless of power rating, most circuit parameters tend to be constrained
to a range about 1 pu: from 0.01 pu up to 5 pu, i.e., a spread of 500. For example, inductances
are typically in the range from 0.05 pu to 4 pu, and resistances from 0.01 pu to 0.1 pu. Variables
(voltages, currents, etc.) are often confined to the range between −1 pu and 1 pu, since an absolute
value larger than 1 pu is larger than the nominal value.
To have variables that are confined to a range from 0 to 1 is also beneficial for the implemen-
tation of control algorithms on a DSP. This is particularly important for fixed-point arithmetics,
where the available number range is limited.
Example 2.14 The differential equation describing an RL series circuit is given by
di
L + Ri = v
dt
where i is the current through and v is the voltage across both the resistance and the inductance.
Let us first normalize v, i, and t:

d(in Ibase )
L + R(in Ibase ) = (vn Vbase )
d(tn /ωbase )

which can be rewritten as

ωbase Ibase din Ibase
L +R in = vn
Vbase dtn Vbase
. /0 1 . /0 1
1/Lbase 1/Zbase
and
din
Ln + Rn in = vn . (2.164)
dtn
This differential equation is formally equal to the original one; the equation is formally unchanged
regardless whether it is normalized or not. For this reason, the subscript n is normally omitted, as
it should be evident anyhow whether a quantity is normalized or not.
♣

Remark 2.7 Note that in per-unit value the reactance (at base frequency) of an inductor is the
same as the inductance:
XL ωbase L
XLn = = = Ln .
Zbase Zbase
So, we could write (2.164) also as

din
XLn + Rn in = vn .
dtn
Although this notation sometimes occurs in the literature, its usage is discouraged. It is good
practice to stay with a notation which holds regardless if the quantities are normalized or not.
80 Chapter 2. Systems and Circuits in Power Electronics and Drives

PROBLEM 2.24
A passive LC filter is tuned to remove the 3rd harmonic from the current in a 230-V, 50-Hz
network.
a) If Cn = 20 pu, what is then most likely Ln ? Hint: Make reasonable assumptions,
including that the resonant frequency of the LC circuit is equal to the frequency that
is to be removed.
b) If C = 56 µF, what are then L and Ibase ?

2.8.2 Three-Phase Systems

For space vectors, the selection of base values should include the space-vector scaling constant K
as follows:
√
Voltage: Vbase = √2KVN
Current: Ibase = 2KIN (2.165)
Frequency: ωbase = ωN .

where VN and IN are the nominal rms phase voltage and current, respectively. The quantities are
then normalized as in the single-phase case:

Voltage: vn = v/Vbase
Current: in = i/Ibase
(2.166)
Angular frequency: ωn = ω/ωbase
Time: tn = ωbase t.

A space-vector modulus of 1 pu then corresponds nominal voltage or current, irrespective of how

the scaling constant is selected. From (2.165), the following additional base values (and others)
can be derived:
3
Power: Sbase = Pbase = Qbase = 3VN IN = Vbase Ibase
2K 2
Impedance: Zbase = VN /IN = Vbase /Ibase (2.167)
Inductance: Lbase = Zbase /ωbase
Capacitance: Cbase = 1/(ωbase Zbase ).

This has the effect that the relation (2.131) for the complex power simplifies to
3 3
S = vi∗ = Vbase vn Ibase i∗n = Sbase vn i∗n
2K 2 2K 2
⇒ Sn = vn i∗n (2.168)
so Sn = 1 pu, quite logically, corresponds to nominal power. This is an exception to the rule that
the normalized equation is formally identical to the original one. Caution should be exercised when
using equations involving normalized power quantities!
CHAPTER 3

Voltage-Source Converters

This chapter focuses mainly on operating principles and pulsewidth-modulation (PWM) techniques
for single-phase and (primarily) three-phase VSCs. VSCs build on the same principle as dc–dc
converters. Therefore, to enhance the understanding of the operating principle of a VSC, we begin
by a brief introduction to the dc–dc buck converter and the principle of hysteresis current control.

3.1 Buck Converters

DC–dc converters can be considered as “dc transformers.” There are several dc–dc converter types,
of which we shall make a brief introduction to the buck (or step-down) converter. Its circuit diagram
is shown in Figure 3.1. The converter is fed from a dc link, whose voltage Vdc is assumed constant,
as is the load voltage V .

L
idc
iD i
Vdc v V

Figure 3.1. Circuit diagram for the buck converter.

3.1.1 Switch-Mode Operation

The device shown in Figure 3.1 as a diode symbol to which two gates symbols are added is a
turn-on–turn-of switch, which generally is transistor. In the ON state the voltage drop across the
switch is small, whereas in the OFF state no current is conducted by the switch. The duty cycle D,
which is the ratio between the ON time t+ and the switching period Tsw : D = t+ /Tsw , determines
the mean value of voltage v. To keep the mean value of current i constant, the mean value of v
must equal the load voltage V . Thus, the following relation must hold:

V = DVdc , 0 ≤ D ≤ 1. (3.1)

The dc-link voltage is “stepped down” by the factor D; hence, the name of the converter. Due
to the switch-mode operation, the waveform of v becomes “chopped” or pulsewidth modulated; a
train of pulses. Typical switching frequencies fsw = 1/Tsw are from 1 kHz up to about 50 kHz.
The chopped waveform of v is acceptable in many applications (for example dc motors), as long
as the output current i is fairly smooth. This is why a sufficiently large inductance L has to be
82 Chapter 3. Voltage-Source Converters

included in the circuit. Furthermore, an alternative path for the output current when the transistor
is switched OFF must be provided, which is the reason for including the so-called freewheeling diode
in Figure 3.1.
Referring to Figure 3.2, when the switch is turned ON, i increases, flowing through the switch;
when turned OFF, i decreases, flowing through the diode. If the switching is made fast enough,
then the resulting ripple ∆i will be low (the ripple in Figure 3.2 is exaggerated).

vGS iD

t t
t+ t
i idc
∆i

t t
.
Tsw

Figure 3.2. Output voltage and current waveforms in the step-down converter.

3.1.2 Load-Current Ripple

Let us calculate the ripple ∆i by finding its relation to the switching frequency fsw = 1/Tsw and
the duty cycle D. In order for i in Figure 3.1(b) to remain constant, DVdc = V . When the
switching valve is turned ON, for 0 ≤ t < t+ , we have
di
L = Vdc − V = (1 − D)Vdc (3.2)
dt
giving
$
1 t+ (1 − D)Vdc t+
∆i = (1 − D)Vdc dt = = {D = t+ /Tsw = t+ fsw }
L 0 L
D(1 − D)Vdc
= . (3.3)
fsw L
This function has its maximum for D = 1/2, i.e., V = Vdc /2
Vdc
∆imax = . (3.4)
4fsw L
As could be conjectured, increasing the inductance and/or the switching frequency reduces the
ripple.
PROBLEM 3.1
A buck converter in an automobile using Vdc = 12 V feeds, via a series-connected inductance,
a resistive load of 12 Ω. There are three choices of the switching frequency and the inductance:
4 kHz, 8 kHz, and 16 kHz and 10 mH, 20 mH, and 30 mH, respectively. It is desired that
3.1. Buck Converters 83

the switching frequency and the inductance both should be as low as possible. Are there any
combinations that give a ripple less than 0.5% of the maximum current?

PROBLEM 3.2
A buck converter with Vdc = 100 V supplies variable voltage to a dc motor. The dc motor can
be modeled as a 40-mH inductance in series with a back emf. The back emf is proportional
to the motor speed, which is variable. The load torque is constant, however, so the motor
current is also constant: 8 A.
a) Select the switching frequency such that the ripple in the motor current never exceeds
0.2 A.
b) What is the current ripple when the motor operates at 80% of its maximum attainable
speed?

3.1.3 Hysteresis Current Control

In many applications it desired to control the output current i in a closed loop. Methods for
current control are discussed in Chapter 4 (when the output current is dc) and in Chapter 6 (when
the output current is three-phase ac). However, it is instructive to study already at this stage a
simple and popular method for current control of switched converters.
Temperatures in ovens and electric radiators are controlled by switching the electric power on
and off. When the temperature exceeds a certain level, power is switched off, whereas power is
switched on again when the temperature dips below a certain level. The span within which the
temperature is allowed to vary is called the tolerance band.
PROBLEM 3.3
Suggest at least two more applications where this type of control is used.
This method is called hysteresis control, alternatively on/off control or tolerance-band control.
Although the switching frequency is quite low (fraction of 1 Hz) in the mentioned applications, the
same principle can be applied also to power electronic converters switching at 1 kHz and above,
simply by choosing a sufficiently narrow tolerance band.

L
Vdc v

E
e i

.
iref

Figure 3.3. Hysteresis control of the buck converter.

Let us consider a buck converter feeding a load with a constant voltage E, as shown in Figure
3.3, where the semiconductor switch and the freewheeling diode are replaced by a so-called change-
over switch model. The block which controls the switch is called a hysteresis comparator. Suppose
84 Chapter 3. Voltage-Source Converters

that the switch is initially turned off, i.e., v = 0. The current i then decreases, and once it dips
below its reference iref , the control error e = iref − i turns positive. When e exceeds the tolerance
band, e > ∆i/2, the output of comparator goes high, causing the switch to turn on. Now, v = Vdc ,
so the current increases until e < −∆i/2, when the switch turns off. Typical curves are illustrated
in Figure 3.4.

i iref i ref +∆i/2

iref −∆i/2
Vdc
v
0
t t
Tsw
.

Figure 3.4. Voltage and current waveforms in hysteresis control.

Obviously, by decreasing the tolerance band ∆i, the current ripple is decreased accordingly,
while the mean switching frequency increases. For buck converters, the mean switching frequency
is found from (3.3)

D(1 − D)Vdc
fsw = (3.5)
L∆i

where the maximum switching frequency is obtained for D = 1/2

max Vdc
fsw = . (3.6)
4L∆i

Hysteresis current control has some benefits. The method is simple to implement, and the response
to a reference step is very quick. The main drawback is the varying switching frequency. This is
particularly troublesome for D ≈ 0 and D ≈ 1, when fsw will be low. A low switching frequency
gives a low-frequency current ripple. In a dc motor drive, this gives low-frequency torque ripple,
which may excite mechanical resonances. Switching of power semiconductors also leads to the
emission of electromagnetic interference (EMI), a problem which is more pronounced at higher
switching frequencies than at lower. It is more difficult to satisfactorily protect sensitive equipment
against EMI if the switching frequency varies than if it is held constant.
The varying switching frequency problem can be circumvented by tuning ∆i on-line. If ∆i is
selected proportionally to D(1 − D), then it is seen in (3.5) that fsw will stay constant. A variable
∆i can be achieved by implementing the hysteresis control system digitally. This requires a very
high sampling frequency to prevent timing errors. Another way is to use analog implementation
but to adjust ∆i digitally via a D/A converter.
3.2. Single-Phase Voltage-Source Converters 85

3.2 Single-Phase Voltage-Source Converters

The buck converter in Figure 3.2 is capable only of supplying positive output voltage and current.
It is straightforward to improve the configuration so that positive output voltage and/or current
can be handled. In the circuit depicted in Figure 3.5(a), three more switches have been added,
together with antiparallel-connected (freewheeling) diodes. The configuration has two legs, A and
B, as indicated in Figure 3.5(a). The switching modes are illustrated by the change-over switch
model in Figure 3.5(b). By switching the legs pairwise, such that the upper valve of leg A is turned
on at the same time as the lower valve of leg B, we get v = Vdc . This is called the positive switching
mode: SW+. Switching the legs vice versa (i.e., upper valve of leg B ON, lower valve of leg A
OFF), we get v = −Vdc : SW−. If both legs are switched the same way (i.e., both upper valves
ON or OFF), we get v = 0: SW0.
By suitable combination of these three switching modes, any switched waveform can be gen-
erated, dc or ac. The converter is therefore called a single-phase inverter, since it “inverts” the
dc-link voltage into any waveform. Other names are, full-bridge converter, H-bridge converter and,
as the title of Section 3.2 indicates, single-phase VSC. Owing to the antiparallel-connected diodes,
the single-phase VSC is capable of handling also negative current. The four modes of operation
excluding SW0 are illustrated in Figure 3.6(a)–(d). Being able to use positive and negative volt-
age and current (which for dc motor drives enables forward and reverse speed as well as electric
braking) is called four-quadrant operation, cf. Figure 3.6(e).
In addition to four-quadrant dc motor drives, an important application for single-phase VSCs
is as active front ends in ac rail traction. The converter then operates in rectifier mode, feeding
power from the catenary (alternating voltage v) to the dc link. Three-phase VSCs are connected
to the dc link, feeding traction ac motors. Usage of a single-phase VSC instead of a diode rectifier
permits regeneration of power to the catenary when the rail vehicle is braked.

leg A leg B

Vdc SW+
i
Vdc

SW0
Vdc SW
v

(a) (b)
Figure 3.5. Single-phase VSC. (a) Circuit diagram. (b) Change-over switch model.

3.2.1 Pulsewidth Modulation: Bipolar Switching

Due to the switch-mode operation, the desired voltage is obtained only in mean. Let us begin by
using only SW+ and SW−, not SW0, which is called bipolar switching. Referring to Figures 3.2
and 3.6, it is obvious that the mean value of v can be expressed as
1 t+
V = [t+ Vdc + t− (−Vdc )] = (2D − 1)Vdc , D= (3.7)
T Tsw
86 Chapter 3. Voltage-Source Converters

OFF

OFF
ON

ON
OFF

OFF
ON

ON
Vdc Vdc

(a): SW+,i >0 (b): SW , i >0

OFF

OFF
ON

ON
OFF

OFF
ON

ON
Vdc Vdc

(c): SW+,i <0 (d): SW ,i <0

i
Reverse Forward
generation motoring
v
Reverse Forward
motoring generation

.
(e)

Figure 3.6. Single-phase VSC: four switching modes, (a)–(d), with v and i positive or negative. (Con-
ducting devices are shaded.) (e): The four quadrants of operation for a dc motor drive.

where t+ and t− are the intervals in which SW+ and SW−, respectively, are used. Thus, if
D = 1 ⇔ t+ = Tsw (only SW+ is used), we get V = Vdc ; if D = 1/2 ⇔ t+ = t− = Tsw /2 (SW+
and SW− are used equally long times), we get V = 0, and if t+ = 0 (only SW− is used), we
get V = −Vdc . Quite similarly to the output voltage of a buck converter, the resulting voltage
waveform becomes pulsewidth modulated; see Figure 3.7.
If the right-hand side of (3.7) is made equal to the reference (desired value) vref for the output
voltage
vref /Vdc + 1
D= (3.8)
2
then the instantaneous output voltage can be expressed as

v = vref + switching harmonics. (3.9)

The switching harmonics (see Paragraph 3.2.4) cause ripple in i. As for the buck converter, the
load must be inductive for the output current to be smooth enough. The higher the switching
frequency, the lower the current ripple. From a control standpoint the switching harmonics can
3.2. Single-Phase Voltage-Source Converters 87

v
Vdc

t
t t

Vdc
Tsw

Figure 3.7. Pulsewidth-modulated voltage waveform (solid) and reference value (dashed).

usually be disregarded, and the converter treated as an actuator that, with a time delay which
often can be neglected, see Paragraph 3.12, generates the desired voltage: v = vref .

PROBLEM 3.4
Derive a formula similar to (3.3) for the output current ripple of the single-phase VSC using
bipolar switching.

PROBLEM 3.5
Hysteresis control is useful also for single-phase VSCs. Replacing the buck converter by
a single-phase VSC simply implies changing the output voltage when the switch is aligned
downward in Figure 3.3 from 0 to −Vdc . Express the switching frequency in a formula similar
to (3.5), assuming R ≈ 0.

3.2.2 Suboscillation Method: Bipolar Switching

We now wish to generate appropriate converter leg switching signals from the reference signal vref .
The classical method for this is the suboscillation (or triangle comparison) method. The idea is
to compare a reference signal s = vref /Vdc to a triangular carrier signal of amplitude 1. When s
is larger than the carrier, SW+ is used; otherwise SW−. This is illustrated in Figures 3.8. The
only difference as compared to Figure 3.7 is that the switchings are made symmetrically within
the period 0 < t < Tsw , not at the beginning of the period. Simple geometry gives

(1 + s)Tsw (1 − s)Tsw
t+ = t− = Tsw − t+ = . (3.10)
2 2
Thus, the mean voltage over one switching period is found to be
$ Tsw
1 1
<v> = v dt = [t+ Vdc + t− (−Vdc )] = sVdc = vref (3.11)
Tsw 0 Tsw

just as desired.
The system in Figure 3.8 is straightforward to implement using analog circuitry. Custom-made
triangle-wave generator chips are available, and an analog comparator is easily constructed using
an operational amplifier. Analog PWM implementation was state of the art up until at least the
early 1990s.
88 Chapter 3. Voltage-Source Converters

vref/Vdc SW+
t
Tsw/2 Tsw
SW

vref/Vdc
1
t /2 t+ t /2
v
Vdc

t
Tsw

.
Vdc

Figure 3.8. Suboscillation PWM method.

PROBLEM 3.6
A single-phase bipolar-switched VSC is used to power the dc motor driving a forklift truck.
The battery gives Vdc = 50 V, the motor’s back emf is proportional to the speed, its induc-
tance is 50 mH, and its current may be at most 100 A.

a) Select the switching frequency such that the current ripple never exceeds 1 A. Hint:
Problem 3.4.
b) At half the base speed, the motor is braked with the current −8 A. For this operating
condition, sketch the switching waveforms as well as the currents through the upper
switching valve and the antiparallel-connected diode in leg A during one switching pe-
riod.

PROBLEM 3.7
Using
√ a single-phase bipolar-switching VSC with Vdc = 513 V, a sinusoidal voltage v =
110 2 sin(120πt) is to be generated in order to supply electrical equipment imported to
Europe from the United States. The switching frequency is 5 kHz and the load can be
assumed resistive–inductive with 2 Ω och 100 mH.

a) Sketch for one switching period the switching waveform as well as the currents through
the upper switching valve and the antiparallel-connected diode in leg A at t = 18 ms.
b) Determine the current ripple at t = 18 ms. (With current ripple it is here meant the
deviation from the ideal sinusoidal waveform.) Hint: Problem 3.4.
3.2. Single-Phase Voltage-Source Converters 89

c) Suppose that the VSC’s dc link is supplied from a 400-V (line-to-line voltage), 50-Hz
grid via a three-phase diode rectifier. What is then the line inductance?

3.2.3 Pulsewidth Modulation: Unipolar Switching

The performance of a pulsewidth-modulated converter can be improved if SW0 is also used. If
positive voltage is desired, SW+ and SW0 should be combined rather than SW+ and SW−.
Likewise, for negative voltage, SW− and SW0 should be combined. The resulting PWM scheme
is called unipolar switching, because SW+ and SW− are not combined.
Unipolar switching is accomplished by skewing the switching events, such that legs A and
B are not switched simultaneously. The method is perhaps best understood by considering its
implementation using the suboscillation method. Leg A is switched as in bipolar switching: the
normalized reference signal vref /Vdc is compared to a triangular carrier signal with amplitude 1.
Leg B, however, is switched by comparing −vref /Vdc to the triangular carrier signal. With the
vector (A, B) indicating the switching state—(A, B) = (1, 0) indicates that leg A is switched on
and leg B is switched off, etc.—the sequence in bipolar switching is simply

· · · − (1, 0) − (0, 1) − (1, 0) − · · ·

whereas in unipolar switching, the sequence is

· · · − (1, 1) − (1, 0) − (0, 0) − (1, 0) − (1, 1) − · · · .

That is, (0, 1) is not used at all, so the pulsewidth-modulated voltage is always positive. For
negative voltage, the sequence is instead

· · · − (1, 1) − (0, 1) − (0, 0) − (0, 1) − (1, 1) − · · ·

and now (1, 0) is never used.

Unipolar switching is preferable to bipolar switching. One reason is that the voltage step at
each switching event is smaller: Vdc instead of 2Vdc , resulting in a smaller current ripple.

PROBLEM 3.8
This problem is identical to Problem 3.7, but unipolar switching is used.

3.2.4 Frequency Spectra

Switched voltage waveforms contain harmonics. If the reference voltage vref is a sinusoid, analytic
expressions for the harmonic content of v unfortunately cannot be derived; numerical calculations
must be relied upon. In Figures 3.9 and 3.10, voltage and current waveforms as well as voltage
frequency spectra are depicted for a sinusoidal reference voltage of peak value 100 V and frequency
50 Hz. The load is an RL series circuit with R = 1 Ω and L = 10 mH, and the switching frequency
is 1 kHz. The following conclusions can be drawn.

• For bipolar switching there are clusters of harmonics around nfsw , n = 1, 2, 3, . . . The am-
plitude of the strongest harmonic of each cluster decreases as 1/n.

• For unipolar switching there are clusters of harmonics around 2nfsw , n = 1, 2, 3, . . . That
is, the switching frequency is effectively doubled. Again, the amplitude of the strongest
harmonic of each cluster decreases as 1/n.
90 Chapter 3. Voltage-Source Converters

• Due to the lower harmonic content, the output current has much less ripple for unipolar
switching than for bipolar switching.
Thus, we have verified that unipolar switching is preferable. This is particularly true for low
switching frequencies: fsw < 1 kHz, when the lowest-order harmonic is in the range of hundreds
of hertz.

100
v(t) (V)

−100

0 10 20 30 40 50 60 70 80
t (ms)
100
|V(f)| (V)

50

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
f (Hz)
50
i(t) (A)

−50
0 10 20 30 40 50 60 70 80
t (ms)

Figure 3.9. Bipolar switching with Vdc = 100 V and fsw = 1 kHz: voltage waveform, voltage frequency
spectrum, and current waveform.

3.2.5 Digital Implementation Using Synchronous Sampling

In a digital current control system, i is sampled with the sampling period Ts . To avoid EMI from
spikes that occur due to switched-mode operation, it is preferable to synchronize sampling and
switching, so that samples of i are taken in between switchings. This coincides with the positive
and negative peak values of the suboscillation triangular carrier waveform, see Figure 3.11. The
sampling frequency is thus selected as twice the switching frequency
fs = 2fsw . (3.12)
Using synchronous sampling, approximately the mean value of the current is obtained at the
sampling instants. Thus, not only is EMI avoided, but so is also the ripple in i. This warrants us
neglecting the switching harmonics in (3.9). Quite often, synchronous sampling is effective enough
so that prefiltering of i (see Paragraph 2.4.1) is unnecessary.
3.2. Single-Phase Voltage-Source Converters 91

100

v(t) (V)
0

−100

0 10 20 30 40 50 60 70 80
t (ms)
100
|V(f)| (V)

50

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
f (Hz)
50
i(t) (A)

−50
0 10 20 30 40 50 60 70 80
t (ms)

Figure 3.10. Unipolar switching with Vdc = 100 V and fsw = 1 kHz: voltage waveform, voltage frequency
spectrum, and current waveform.

Digital PWM can be implemented using only one timer. When the timer runs out, a processor
interrupt is triggered and the proper switching is executed. (Alternatively, the switching logic can
be implemented directly in hardware, e.g., using an FPGA.) As shown in Figure 3.12, the timer can
be reloaded with the proper countdown time ∆tk at the sampling instants, i.e., at time t = kTs . If
the switching state is SW+ when the timer is reloaded, then ∆tk = t+ /2, otherwise ∆tk = t− /2.
From (3.10) with Tsw = 2Ts and s = sk , we obtain
*
(1 + sk )Ts /2 if switching state=SW+
∆tk = (3.13)
(1 − sk )Ts /2 if switching state=SW−.

PWM Time Delay

The countdown time is confined to the interval [0, Ts ]. If ∆tk = 0, then the switching is executed
immediately, and if ∆tk = Ts , then the timer is reloaded immediately after it has run out. These
extremes correspond to short-duration pulses, such as that centered at t = (k + 3)Ts in Figure 3.12.
Except for such sudden changes (as that exemplified in Figure 3.12), the switching and sampling
frequencies are normally high enough in relation to the system dynamics, so that the change in the
reference signal between two consecutive sampling instants is small, i.e., sk+1 ≈ sk . Consequently,
since the switching state alternates between SW+ and SW+ with every sampling instant, the mean
countdown time for two consecutive switching events is given by
∆tk + ∆tk+1 Ts (sk − sk+1 )Ts Ts
= ± ≈ . (3.14)
2 2 4 2
92 Chapter 3. Voltage-Source Converters

t
Ts 2Ts
v

t−/2 t+/2 t+/2 t−/2

Figure 3.11. Synchronous sampling with the suboscillation method: current i is sampled in between
switchings, at t = {0, Ts , 2Ts , . . .}. The switching frequency is 1/(2Ts ).

Thus, seen over one switching period the mean time delay from timer reloading to switching
execution—which we shall call the PWM time delay—is approximately Ts /2.

Controller Time Delay

Reference sk is generally computed by a current controller, see further Section 4.4. The current
controller must include a one-sample delay, i.e., the most recent current sample used in the com-
putation of sk by the current controller is ik−1 , as shown by the dotted lines in Figure 3.12. If this
time delay were not included, then timing errors would result for short-duration pulses. This is
exemplified in Figure 3.13, which shows a similar switching sequence as in Figure 3.12, but without
the controller time delay. The reload time instants are now postponed by the computational time
delay Tc , and consequently the countdown time is adjusted as

∆t′k = ∆tk − Tc . (3.15)

Since Tc ≪ Ts , this scheme works flawlessly until the short-duration pulse centered at t = (k + 3)Ts
is reached. Here, ∆tk+3 < Tc , so the control system resorts to setting ∆t′k+3 = 0, i.e., the switching
is executed immediately. The delayed switching gives slightly higher mean voltage and current, as
can be observed by comparing with the ideal dashed curves in Figure 3.13.

Total Time Delay

It is found that the mean total time delay for PWM using synchronous sampling is given by

Td = 1.5Ts (3.16)
3.3. Three-Phase Voltage-Source Converters 93

t
k Ts (k+1)Ts (k+2)Ts (k +3)Ts (k +4)Ts
v
t

∆tk ∆ tk+1 ∆tk+2 ∆tk +3

Figure 3.12. Digital PWM with a one-sample controller time delay.

t
k Ts (k+1)Ts (k+2)Ts (k +3)Ts (k +4)Ts
v
t
Tc

∆tk ∆ tk+1 ∆tk+2 ∆tk+3 =0

Figure 3.13. Digital PWM without a controller time delay: timing errors result for short pulses.

of which Ts results from the controller time delay and 0.5Ts results from the PWM time delay.

3.3 Three-Phase Voltage-Source Converters

Adding one more leg to the single-phase VSC in Figure 3.5, we obtain the three-phase VSC shown
in Figure 3.14. The corresponding change-over switch model is shown in Figure 3.15(a).
The circuit shown in Figure 3.14 is ungrounded, but for convenience the middle of the dc link
is selected as neutral point in the calculations to follow. As for the single-phase VSC, we introduce
a switching vector (a, b, c). For example, (a, b, c) = (1, 0, 0) means that leg a is switched on, giving
va = Vdc /2, and legs b and c are switched off, giving vb = vc = −Vdc /2, corresponding to Figure
3.15(a). Let us now introduce the space vector for the converter output voltage, as defined in
(2.101)
2K
vs = (va + ej2π/3 vb + ej4π/3 vc ). (3.17)
3
Because the phase voltages only can attain ±Vdc /2, vs can attain only the following values:
* + * +
KVdc 2KVdc KVdc
vα = 0, ± ,± , vβ = 0, ± √ . (3.18)
3 3 3
94 Chapter 3. Voltage-Source Converters

leg a leg b leg c

Vdc/2

Vdc/2
va vb vc

Figure 3.14. Three-phase VSC.

The resulting vector diagram is depicted in Figure 3.15(b). The maximum modulus of the voltage
vector is
2KVdc
|vs |max = (3.19)
3
which is attained for all switching states except (1, 1, 1) and (0, 0, 0), which both yield the zero
vector. A rotating voltage vector of maximum modulus is obtained by repeating the switching
sequence
(1, 0, 0) − (1, 1, 0) − (0, 1, 0) − (0, 1, 1) − (0, 0, 1) − (1, 0, 1)
i.e., the zero vector is not used at all. This switching mode is known as six-step.
PROBLEM 3.9
What is a binary sequence of this type called?
It is rather obvious that by combining all switching states appropriately using PWM—including
the zero vector—any location within the hexagon in Figure 3.15(b) can be accessed, i.e., any
voltage vector within the hexagon can be created in mean. For instance, [vα , vβ ] = [KVdc /3, 0] can
be obtained by switching (0, 0, 0) and (1, 0, 0) equally long times. However, the circle in Figure
3.15(b), which touches the sides of the√hexagon, is the limit for linear modulation. Simple geometry
shows that this circle has a radius of 3/2 times the maximum voltage vector
√
3 s KVdc
circle radius = |v |max = √ . (3.20)
2 3
Therefore, as shown in Figure 3.15(b), it is logical to select the voltage base value Vbase (see Section
2.8) as
KVdc
Vbase = √ . (3.21)
3
√
The relation to the nominal rms phase voltage VN is obtained for K = 1/ 2
Vdc
VN = √ . (3.22)
6
3.3. Three-Phase Voltage-Source Converters 95

(a,b,c) = (1,0,0)
Vdc/2
a b c

Vdc/2

va vb vc

.
(a)
vβ
(0,1,0) KVdc (1,1,0)
3

(0,1,1) (0,0,0) Vbase (1,0,0)

vα
(1,1,1)
2KVdc 2KVdc
3 3
KVdc
(0,0,1) 3 (1,0,1)
.
(b)

Figure 3.15. (a) Change-over switch model for a three-phase VSC. (b) Space-vector diagram.

PROBLEM 3.10
Suppose that a VSC is fed from a three-phase grid via a diode rectifier. What is the maximum
relation between the converter output phase voltage and the grid phase voltage?

PROBLEM 3.11
The vector
KVdc j30◦
vs = √ e
2 3
can be created in mean by the switching sequence

(0, 0, 0) − (1, 0, 0) − (1, 1, 0) − (1, 1, 1) − (1, 1, 0) − (1, 0, 0) − (0, 0, 0).

Determine the percentage of the switching period each switching state should be applied.
Hint: As we shall soon see, it is useful to create the zero vector by switching (0, 0, 0) and
(1, 1, 1) equally long times.
96 Chapter 3. Voltage-Source Converters

PROBLEM 3.12
Suggest a switching sequence by which the vector
3KVdc j210◦
vs = √ e
4 3
can be created in mean. Also determine the percentage of the switching period each switching
state should be applied.
PROBLEM 3.13
Sketch within the space-vector hexagon the vector obtained in mean by the switching se-
quence (where the percentages are the applications of each vector relative the switching
period)
(0, 0, 0) [10%] − (0, 0, 1) [20%] − (1, 0, 1) [10%] − (1, 1, 1) [20%] −
(1, 0, 1) [10%] − (0, 0, 1) [20%] − (0, 0, 0) [10%].

3.3.1 Suboscillation Method

In Section 3.2 we used the suboscillation method for PWM of a single-phase VSC. Extension to
a three-phase VSC, given the phase reference signals sa,b,c , is straightforward. As for the single-
phase VSC, these reference signals are confined to the interval [−1, 1]. Three comparators, one for
each phase leg, are constructed. The mean phase voltages that result can be computed similarly
to (3.11), but since the phase voltages are defined relative the neutral point, we need to substitute
Vdc → Vdc /2
( , -)
1 a,b,c Vdc Vdc sa,b,c Vdc
va,b,c = t+ + ta,b,c
− − = . (3.23)
T 2 2 2
This results in the following voltage space vector:
⎡ ⎤ ⎡ ⎤
2 1 1 ⎡ ⎤ 2 1 1 ⎡ ⎤
( ) − − − −
⎢ 3 ⎥ va ⎢ 3 3 ⎥
sa
vα
=K⎢
3 3 ⎥ ⎣ vb ⎦ = KVdc ⎢ 3 ⎥ ⎣ sb ⎦ . (3.24)
vβ ⎣ 1 1 ⎦ 2 ⎣ 1 1 ⎦
0 √ −√ vc 0 √ −√ sc
3 3 3 3
So, if the reference signals sa,b,c are calculated from the voltage reference vector vsref = vαref + jvβref
as
⎡ ⎤ ⎡ ⎤
1 0 1 0
⎡ ⎤ ⎢ √ ⎥( ⎢ √ ⎥(
sa ⎢ 1 3 ⎥ ref
) ⎢ 1 3 ⎥ )
⎣ sb ⎦ = 2 ⎢ − ⎥ v α 2 ⎢ − ⎥ vαref
⎢ 2 2 ⎥ = √ ⎢ 2 2 ⎥
KVdc ⎢ √ ⎥ vβref 3Vbase ⎢ √ ⎥ vβref
sc ⎣ ⎦ ⎣ ⎦
1 3 1 3
− − − −
⎡ 2 2 ⎤ 2 2
2
√ 0
⎢ 3 ⎥
⎢ ⎥( )
⎢ ⎥
1 ⎢ 1 ⎥ vαref
= ⎢ −√ 1 ⎥ (3.25)
Vbase ⎢ 3 ⎥ vβref
⎢ ⎥
⎣ 1 ⎦
−√ −1
3
we obtain vs = vsref in mean.
3.3. Three-Phase Voltage-Source Converters 97

3.3.2 Symmetrical Suboscillation Method

The three-phase suboscillation method √ is associated with a problem.
√ Suppose that vsref = Vbase .
Equation (3.25) then yields sa = 2/ 3 and sb = sc = −1/ 3. But sa > 1, so the reference
signal exceeds the triangular carrier signal, leading to so-called overmodulation,
√ which gives highly
distorted voltages. Apparently, the maximum permissible vector modulus is 3Vbase /2 = 0.87Vbase ,
since then sa = 1. The area which can accessed by the suboscillation method is depicted in Figure
3.16.
vβ

Vbase
vα
3V
2 base

Figure 3.16. Voltage limitation for√the suboscillation method: only vectors within the dashed circle can
be generated (maximum modulus: 3Vbase /2).

To circumvent the problem of reduced output voltage associated with the suboscillation method,
zero-sequence components can be added to the reference signals. As discussed previously, this does
not result in zero-sequence currents as long as the Y point is ungrounded. Several variants for
zero-sequence addition have been proposed. One variant that is intuitive as well as simple to
implement is known as the symmetrical suboscillation method, which will now be described.
First, we observe that if the same deviation ∆ is subtracted from all reference signals, giving
s′a,b,c = sa,b,c − ∆, a zero-sequence component is added. The resulting voltage space vector is then
not altered. The key to extending the modulation range to the entire hexagon is to select ∆ so
that

max(s′a , s′b , s′c ) = − min(s′a , s′b , s′c ). (3.26)

This implies selecting

max(sa , sb , sc ) + min(sa , sb , sc )
∆= s′a,b,c = sa,b,c − ∆. (3.27)
2
The method makes the reference signals symmetric with respect to the maximum and minimum
values; hence, its name [79].
√
Example 3.1 Let us consider the case vsref = 3Vbase /2, i.e., the maximum space vector which
can be generated using the standard suboscillation method. This yields the reference signals
sa = 1, sb = sc = −1/2. From (3.27) we get ∆ = (1 − 1/2)/2 = 1/4 and s′a = 3/4, s′b = s′c = −3/4.
The reference signals are now symmetrized. Figure 3.17 illustrates the method in the time domain.
98 Chapter 3. Voltage-Source Converters

It is seen that (sa , sb , sc ) and (s′a , s′b , s′c ) both yield the same space vector (in mean). The difference
between them is how the zero vector is generated. For (sa , sb , sc ), the zero vector is applied by
switching (1, 1, 1) in the middle of the switching interval. For (s′a , s′b , s′c ), three zero vectors are
generated: at the beginning and the end of the switching interval by switching (0, 0, 0), and in the
middle by switching (1, 1, 1). The total time the zero vector is applied is equal in the two cases.
However, for the choice (s′a , s′b , s′c ), we see that the reference signals can be further increased by
1/4 before the edges of the triangular carrier signal are touched. Then (s′a , s′b , s′c ) = (1, 0, 0), and
the maximum voltage vector is obtained.

Standard suboscillation Symmetrical suboscillation

method method
+1 sa +1 sa’

t t
sb sc Tsw Tsw
sb’ sc’
−1 −1
va v’a

t t
Tsw Tsw

vb v’b

t t
Tsw Tsw

vc v’c

t t
Tsw Tsw

Figure 3.17. Two choices of the reference signals which yield the same space vector in mean.

3.3.3 Prevention of Overmodulation

The symmetrical suboscillation method allows the entire hexagon to be accessed. Yet, overmod-
ulation will result should any of the reference signals (s′a , s′b , s′c ) exceed the interval [−1, +1]. In
3.3. Three-Phase Voltage-Source Converters 99

that case, (s′a , s′b , s′c ) must be scaled such that the maximum value equals 1

m = max(s′a , s′b , s′c ); if m > 1 : s′a,b,c → s′a,b,c /m (3.28)

or simpler

s′a,b,c
s′a,b,c → . (3.29)
max(1, s′a , s′b , s′c )

This has the effect of reducing the modulus of the resulting voltage space vector without modifying
its angle. This is called the minimum phase error method and is one of many ways of selecting the
voltage vector in order to prevent overmodulation. One benefit of this method is that it obviously
is very simple to implement. An alternative is the minimum magnitude error method. Both are
illustrated in Figure 3.18; for more details, see [37]. The voltage reference vector that results from
application of the described method is denoted as v̄sref and is called the realizable or saturated
reference. Since the algorithm which performs the vector modulus reduction is an integral part of
the PWM generation, the following notation is used:

v̄sref = PWM (vsref ). (3.30)

Reduced vector modulus, i.e., |v̄sref | < |vsref |, is known as operation in the saturated region.

vβ Ideal vector vβ

(a) Ideal path

(b)

vα vα
Vbase

Realizable path
(c)

Figure 3.18. Prevention of overmodulation. Realizable vectors for (a) minimum phase error method, and
(b) minimum magnitude error method. (c) Ideal (circular) and realizable (distorted) vector paths.

Algorithm 3.1 shows PWM using the symmetrical suboscillation and minimum phase error
methods. In addition to the modified reference signals (s′a , s′b , s′c ) it also computes the components
of the realizable space vector v̄sref = v̄αref + jv̄βref .
100 Chapter 3. Voltage-Source Converters

Algorithm 3.1
Three-Phase PWM Using the Symmetrical Suboscillation and Minimum Phase Error
Methods
⎡ ⎤
2
√ 0
⎡ ⎤ ⎢ 3 ⎥
⎢ ⎥( )
sa ⎢ ⎥
⎣ sb ⎦ = 1 ⎢ − √
1
1 ⎥ vαref
⎢ ⎥
Vbase ⎢ 3 ⎥ vβref
sc ⎢ ⎥
⎣ 1 ⎦
−√ −1
3
max(s a , sb , sc ) + min(sa , sb , sc )
s′a,b,c = sa,b,c −
2
′
s a,b,c
s′a,b,c =
max(1, s′a , s′b , s′c )
⎡ ⎤
1 1 1 ⎡ ′ ⎤
( ref ) √ √ √
v̄α ⎢ 3 − 2 3 − 2 3 ⎥ sa′
= Vbase ⎢ ⎣
⎥⎣ s ⎦
v̄βref 1 ⎦ s′
b
1
0 − c
2 2

3.3.4 Digital Implementation Using Synchronous Sampling

PWM with synchronous sampling for a three-phase VSC can be implemented on a per-phase
basis, following the principles outlined in Paragraph 3.3.4. That is, three timers, one for each
phase modulator, are needed. Implementing the symmetrical suboscillation method digitally yields
a PWM scheme which is equivalent or closely related to the space-vector-modulation method
proposed by Holtz, see [10, Chapter 3]. However, by using both space vectors and three-phase
quantities for computation of the countdown times, a much simpler algorithm is obtained compared
to a direct implementation of space vector modulation.

PROBLEM 3.14
In Figure 3.19, switching waveforms of a three-phase PWM converter are shown.

a) What is the switching frequency?

b) The switching waveforms reveal that three different space vectors are generated, each
one during a couple of switching periods. Then, the vector is changed. Determine at
which time instants this happens, and plot the vectors in an αβ plane (PWM hexagon).

PROBLEM 3.15
A three-phase VSC employs the symmetrical subsocillation method for PWM. The switching
frequency is 1 kHz. In Figure 3.20, the instantaneous output voltage vector in the αβ plane is
shown (as dots, one dot per vector generated over one switching period). Peak-value scaling
is used (K = 1).
3.3. Three-Phase Voltage-Source Converters 101

0.5

leg a
0

−0.5

−1
0 5 10 15

0.5
leg b

−0.5

−1
0 5 10 15

0.5
leg c

−0.5

−1
0 5 10 15
t (ms)

Figure 3.19. Switching waveforms.

a) In Figure 3.20(a), the output voltage is a stationary sinusoid (considering the funda-
mental frequency component). What are its frequency and rms value?
b) Figure 3.20(b) illustrates a vector diagram for an ac motor drive, where the motor is
accelerating. Determine approximately the start and stop frequencies, and how long the
acceleration takes.
c) What is the converter’s dc-link voltage?
d) Sketch the switching waveforms during one switching period for the marked vector!

PROBLEM 3.16
For a three-phase VSC, Figure 3.21 shows the reference signals (to the triangular-carrier
comparator) during one switching period at three different time instants.

a) What is the switching frequency?

b) For each of the three time instants, sketch the switching waveforms over one switching
period, as well as the voltage vector which is generated in mean.
c) What is the fundamental frequency of the voltage?
102 Chapter 3. Voltage-Source Converters

(a) (b)
150 150

100 100

50 50
vβ (V)

vβ (V)
0 0

−50 −50

−100 −100

−150 −150
−150 −100 −50 0 50 100 150 −150 −100 −50 0 50 100 150
vα (V) vα (V)

Figure 3.20. Vector diagrams.

(a) (b) (c)

1 1 1
sa
sb
sa
0.5 0.5 0.5
sb
s
a
0 0 0
sb = sc
−0.5 −0.5 sc −0.5 sc

−1 −1 −1
0 0.5 1 20 20.5 21 40 40.5 41
t (ms) t (ms) t (ms)

Figure 3.21. Reference signals.

3.3.5 Switching Harmonics

The amount of subharmonics that result from PWM is significantly reduced if the pulse number
mf = fsw /f1 , i.e., the quotient between the switching and fundamental frequencies, is selected
as an integer. This is called synchronized switching. For even-order harmonics to be avoided the
pulse number should be odd. It is particularly important to use synchronous switching when the
switching frequency is low; as a rule of thumb for mf < 21.
Synchronized switching relatively straightforward to implement for grid-connected VSCs, where
the fundamental frequency is near constant. For variable-speed ac drives on the other hand, a
varying switching frequency is required. The switching frequency should be confined between a
lower limit and an upper limit. The solution is to vary the pulse number on line. Figure 3.22 shows
a typical example thereof. For low f1 , a constant switching frequency is used: fsw = 500 Hz. This
is followed by mf = 21, mf = 15, and mf = 9 as the fundamental frequency increases. The change
in modulation ratio occurs when fsw reaches the maximum switching frequency, in this example
1000 Hz.
3.3. Three-Phase Voltage-Source Converters 103

1000
mf = 15
m = 21 m =9
800 f f

fsw (Hz) 600

400

200

0
0 10 20 30 40 50 60 70 80 90 100
f1 (Hz)

Figure 3.22. Synchronized switching: the switching frequency is selected as a multiple of the fundamental
frequency (except for low fundamental frequencies).
104 Chapter 3. Voltage-Source Converters

.
CHAPTER 4

DC Motor Drives

DC motor drives may today be somewhat obsolete, but they are not unimportant to study. One—
and perhaps the strongest—reason is that dc motor drives are conceptually much easier to grasp
than ac motor drives. The models involve scalar quantities, whereas for ac drives vector models
have to be used. For this reason it convenient to introduce topics that are important also for
ac motor drives, such as dynamic models, current control, speed control, etc., in the context of
dc motor drives. Many of the introduced models and control methods can be extended to ac
motor drives by making generalizations from scalar to vector quantities, as will be shown in the
subsequent chapters. Much can be gained from a pedagogical standpoint by starting at the dc
motor drives end.
After a brief study of the mechanics for rotational motion, we in this chapter derive from
electromagnetic theory a model for the dc motor. The open-loop dynamics are studied, and the
benefit of using closed-loop current control in variable-speed drives is explained. We then proceed
with current and speed controller designs.

4.1 Mechanics for Rotational Motion

From basic mechanics we know that to make a body with mass m move, a force F must be applied.
An acceleration a, given by

F
a= (4.1)
m

then results. This sets the body in motion with a velocity (speed) v, given by
$ $
1 dv
v= a dt = F dt ⇔ m =F (4.2)
m dt

yielding in turn the position x

$
dx
x= v dt ⇔ = v. (4.3)
dt

The power P is the incremental work F dx performed over the incremental time dt

dx
P =F = F v. (4.4)
dt
106 Chapter 4. DC Motor Drives

The above equations are valid for a system in linear motion. For a system in rotational motion,
the following closely related relations apply:
τ
ar = (4.5)
J
$ $
1 dω
ω = ar dt =τ dt ⇔ J =τ (4.6)
J dt
$
dθ
θ = ω dt ⇔ =ω (4.7)
dt
dθ
P = τ = τω (4.8)
dt
where τ is the torque, ar = ω̇ is the angular acceleration, ω is the angular speed, θ is the angle,
and J is the moment of inertia. The correspondences between quantities in linear and rotational
motion are summarized in Table 4.1.

Table 4.1. Correspondences between quantities for linear and rotational motion.

Linear SI unit Rotational SI unit

F N=kgm/s2 τ Nm=kgm2 /s2
a m/s2 ar rad/s2
v m/s ω rad/s
x m θ rad
m kg J kgm2

The moment of inertia for a rotating body is given by the volume integral
$
J = r2 dm. (4.9)

That is, every mass element of the body is weighted by the square of the radius (relative the center
of rotation) for that element, and integrated over the entire volume.
Example 4.1 In electrical drives, the most common rotational object is a solid cylinder with an
approximately uniform mass density ρ. The rotor of the electrical motor itself, as well as many
mechanical loads, are shaped like this. Let the cylinder have radius r and axial length l. To
calculate J, we sectionalize the cylinder as shown in Figure 4.1. As dm = ρ dV and dV = 2πr′ l dr′ ,
the following expression is obtained:
$ r
π
J = 2πρl r′ 3 dr′ = ρlr4 .
0 2

Alternatively, with the total mass of the cylinder given as m = ρV , where V = πr2 l is the total
volume, we obtain

mr2
J= . (4.10)
2
4.1. Mechanics for Rotational Motion 107

To obtain a small J, it is seen that the radius should be as small as possible. Long and slender
motors (l > r), such that J is kept small, are therefore preferable in applications which use high
speeds and/or rapid accelerations. Motors that rotate slowly, on the other hand, are often designed
so that r > l, since it is then not important to keep J small.
To appreciate how J varies with the size of the motor, consider first a small motor, whose rotor
weighs 1 kg and has a radius of 3 cm: this yields J = 4.5 · 10−4 kgm2 . The rotor of a medium-sized
motor may have m = 100 kg and r = 1 dm, giving J = 0.5 kgm2 . A large motor, finally, may
weigh 1 ton and have r = 0.5 m, yielding J = 125 kgm2 .

l
r

r’
ω
dr’

Figure 4.1. Rotating cylinder.

PROBLEM 4.1
In order for a dc motor to produce a nominal torque of 20 Nm, it has been determined that
the rotor’s radial surface area must be 0.3 m2 to obtain sufficiently high flux. Select the
radius and axial length of the rotor so that the nominal acceleration when the motor is not
loaded mechanically is at least 50 rad/s2 . The rotor can be assumed to have a mass density
of 7500 kg/m3 .

PROBLEM 4.2
Hollow cylinder. Find an expression similar to (4.10) for the inertia of a hollow cylinder
with inner radius r1 , outer radius r2 , and mass m.

PROBLEM 4.3
Gearing. In many applications, a motor drives a load which consists of two or more rotating
bodies interconnected by geared wheels, as in Figure 4.2(a). The gears engage at the point
P. The wheels are axially coupled to rotating bodies with inertias J1 and J2 , respectively.
Suppose that, as shown in the figure, an electrical motor applies a torque τe on the wheel
with radius r1 .
a) Find the relation between the angular speeds ω1 and ω2 .
b) Show that, as “seen” from the applied torque, the total inertia of the system is
, -2
r1
J = J1 + J2 .
r2

Hint: At the point P, the peripheral speeds v1 and v2 are equal, as are the tangential forces
F1 and F2 . Furthermore: τ = F r, v = ωr.
108 Chapter 4. DC Motor Drives

PROBLEM 4.4
Hoist drive. In Figure 4.2(b), a motor with torque τe drives a hoist with mass m using a
wheel of radius r and inertia J. The gravitational force is mg. The dynamics of the drive
can be described as
dω
J′ = τe − τl .
dt
Find expressions for the equivalent inertia J ′ (which is larger than J due to the mass m) and
the load torque τl . Hint: τ = F r, F = mv̇, v = ωr.

ω2 r
ω1
J
r1 P r2
τe
J1
J2
τe m

.
(a) (b) mg

Figure 4.2. (a) Two rotating bodies interconnected by geared wheels. (b) Hoist drive.

PROBLEM 4.5
In an electrical motor drive, the mechanical load consists only of an inertia J = 1.0 kgm2 .
It is specified that the speed shall vary sinusoidally: ω = ω
! cos ω0 t, where ω
! = 10 rad/s and
ω0 = 200π rad/s.

a) Determine in degrees the amplitude of the oscillation in the shaft angle θ.

b) Determine the maximum torque that has to be supplied from the motor.
c) Determine also the maximum and mean powers.

4.2 DC Machine Fundamentals

DC machines can be used as both motors and generators. Although the focus of the chapter is dc
motor drives, we in this section nevertheless use the term “machine,” for generality.

4.2.1 Principle
The rotor, or armature, of a dc machine is cylindrical. Slots are machined in the surface, where
the armature winding is placed. In Figure 4.3, two opposite conductors of the armature winding
are shown. Around the rotor a magnetized stator is placed. Both the stator and the rotor can be
constructed either as a solid steel forging or by using laminated steel sheets. Figure 4.3 shows a
two-pole stator, magnetized by a permanent magnet. Alternatively, magnetization can be made
by a separate field winding, as is further discussed in Paragraph 4.2.4.
4.2. DC Machine Fundamentals 109

rotor brush

I F

stator
F

N S
Figure 4.3. DC machine principle.

The density of the magnetic flux that crosses the airgap between the stator and the rotor is
B. When a current I flows through one rotor conductor, as shown in Figure 4.3, a force F = BIl
starts to act on the conductor, where l is the axial length of the rotor. This force turns the rotor
clockwise. Because the flux crosses the airgap perpendicularly to the rotor surface, the force on
the conductor stays constant as long as the conductor is enclosed by either one of the stator poles,
as shown in Figure 4.4(a). The rotor sectors which are not enclosed by the two stator poles (i.e.,
around each brush) are unmagnetized. When the conductor passes an unmagnetized sector, see
Figure 4.4(b), then no force acts upon it. Once the conductor by the rotation of the rotor moves
within the opposite stator pole, if the direction the current is kept, then a force that counteracts the
rotation will result. Therefore, the direction of the current must be changed when the conductor
passes the brush. This is called commutation and is illustrated in Figure 4.4(c).
In a dc motor, commutation is made by the commutator, which typically is constructed as
illustrated in Figure 4.5. The armature winding consists of several coils that are placed in the
rotor slots. Each coil is connected between two opposite segments of the commutator. The brushes,
through which the armature current is carried, slide on the commutator, commutating as many
times per revolution as there are coils. The brushes are usually made of carbon; thus they wear
down, but can be replaced inexpensively. This allows the commutator—which is made of copper
and therefore is costly to replace—to be kept free of wear. The need for replacing brushes regularly
makes dc machines less attractive than ac machines (which do not use brushes). Also, there is a
voltage drop of about 1 V across each brush, giving extra losses. (This effect is neglected in the
following, however.)

4.2.2 Electrical Model

Let the stator and rotor dimensions be given by Figure 4.6. Furthermore, let us assume that
the rotor has a total of N turns in its series-connected coils. There are two parallel paths in
the winding, so each of the 2N conductors (exemplified by 2N = 32 in Figure 4.6) carries the
110 Chapter 4. DC Motor Drives

N S N S N S
(a): Conductors magnetized, forces (b): Conductors not magnetized, (c): Commutation done: forces
.
turn rotor clockwise no force on rotor still turn rotor clockwise

Figure 4.4. Principle of commutation.

Brush

Commutator
segment

Rotor
.

Shaft
Rotation

Insulator
.

Figure 4.5. Commutator.

current i/2.1 A force Bil/2 thus acts on each magnetized conductor. Assuming that the number
of conductors is fairly large, we can calculate the number of magnetized conductors as the quotient
between total pole span 2w and the rotor circumference 2πr. Hence, the total force F that acts
on the rotor is given by
2w i N Bwl
F = 2N B l = i. (4.11)
2πr 2 πr
But wl is the cross-sectional area of each pole, and, since magnetic flux Φ is flux density times
cross-sectional area, we have Φ = Bwl. This yields the simplified expression
NΦ
F = i. (4.12)
πr
The electrical torque τe produced by the machine is related to the total force as τe = F r, so
NΦ
τe = ψi, ψ= . (4.13)
π
Here, we have introduced a new quantity: the flux linkage ψ, which is the actual flux scaled by the
number of turns (and 1/π; a difference compared to the standard definition of flux linkage). By
1 We from now on assume that the armature current is time varying (but constant in the steady state), so i is

used rather than I.

4.2. DC Machine Fundamentals 111

using flux linkage rather than actual flux, we avoid dragging the factor N/π along in the equations.
Flux linkage (in various forms) will be used frequently in the following, also for ac machines. Quite
often, we for simplicity drop “linkage” and refer to ψ only as “flux.”
It is also well known that, if a conductor of length l is moved with velocity v in a magnetic
field with flux density B, then a voltage Blv is induced across the conductor. Therefore, once
the machine starts rotating, an induced voltage will appear across the armature winding. Let
the angular rotor speed be ωr . Then, v = ωr r, so across each magnetized conductor, an induced
voltage Blωr r appears. No voltage is induced in the conductors that are not magnetized, so the
total induced voltage—which is also known as the back emf—is given by

2w
E= N Blωr r = ψωr . (4.14)
2πr
Note the similarity to the torque equation (4.13): the only difference is that i is replaced by ωr .

w i /2 i /2

Figure 4.6. DC machine with 32 rotor slots and stator/rotor dimensions.

PROBLEM 4.6
For the machine in Problem 4.1, w/πr = 80%. How many turns N are required for the
machine to produce the desired torque 20 Nm if the nominal armature current is 20 A and
B = 1.5 T?

PROBLEM 4.7
E = ψωr can also be derived from the relation τe = ψi and “power = angular speed ×
torque.” Do that!

4.2.3 Dynamic Model

We are now ready for the final model of the dc machine, which is simply the back emf in series with
an inductance and a resistance. This is simply because coils—which have resistive losses—wound
around an iron core also become inductive. So, the complete dc machine model is that given by
the equivalent circuit of Figure 4.7.
112 Chapter 4. DC Motor Drives

R L

v E=ψωr ψ

i ωr
τe = ψi

Figure 4.7. Equivalent circuit for the dc machine.

From the equivalent circuit it is straightforward to write down the dynamic equations for the
machine. Suppose that a voltage v is applied across the machine terminals. Then the electrical
dynamics are given by
di
v − Ri − L = E = ψωr . (4.15)
dt
Furthermore, the mechanical dynamics are
dωr
J = τe − τl = ψi − τl (4.16)
dt
where τl is the load torque and J is the total inertia of the rotor and the mechanical load. Normally,
the load torque increases with the speed. As a simplification, let us temporarily assume that the
load torque is proportional to the speed as τl = bωr , where b is called the viscous damping constant.
The machine model then becomes linear
di
L = v − Ri − ψωr (4.17)
dt
dωr
J = ψi − bωr . (4.18)
dt
We see that the electrical and mechanical dynamics are interlinked: the back emf ψωr links (4.18)
to (4.17) and the electrical torque ψi vice versa. How this affects the total dynamics will be
investigated in Section 4.3.

4.2.4 Magnetization by Field Winding

Figure 4.3 shows the stator to be magnetized by a permanent magnet. Although this is common
for machines of lower power ratings (1 kW and below), the flux is more commonly supplied by
a separate field winding, carrying the field current (or magnetizing current) if , as illustrated in
Figure 4.8(a). The relation between if and ψ can be expressed as
f (if )
ψ = f (if ), = Lf (4.19)
if
where Lf is called the field-winding inductance. For smaller values of if the relation to ψ is linear,
i.e., Lf is constant. However, as if increases the iron enters saturation, leading to a decreasing
4.2. DC Machine Fundamentals 113

Lf . The field current is normally selected such that the iron gets only slightly saturated, as shown
in Figure 4.8(b), giving the base flux ψbase ; also called nominal or rated flux, see further below.
(Driving the iron further into saturation would create excessive losses due to a large if and thus
substantial heating in the field circuit.)

ψ
ψbase

ψ
Slope:
if Lf
if
if,base
(a) (b)
Figure 4.8. (a) Magnetization by field winding. (b) Relation field current–flux linkage.

The field winding usually has more turns than the armature winding, so in most cases, Lf > L
(Lf ≈ 10L is not unreasonable). Fast changes in the field current (and, in turn, in ψ) are therefore
difficult to achieve. For quick torque changes, ψ should be kept constant and i be modified.
PROBLEM 4.8
A dc motor is driving a fan, whose load torque is 5 Nm at the nominal speed 1800 rpm. The
motor is fed from a constant 200-V source. Speed control is accomplished by varying the
magnetization. The inductance of the field winding is 2 H and the armature resistance is 2
Ω. Determine the field current needed to give nominal speed.

4.2.5 Rating and Base Values

Rather obviously, a dc machine of larger size will be able to produce more power than a smaller
one. If one were to try squeezing significantly more power out of a machine than intended in the
design, the heating due to losses in the armature winding would be excessive. Simply put, the
machine would burn up. The nominal or rated (maximally allowed in the state state) operating
conditions are specified by the manufacturer on data sheets and/or the machine’s so-called name
plate (which is mounted on the machine housing). Let us go through the steps by which the rating
in principle is set.
First, the base armature voltage Vbase is determined. This is given by the voltage allowed be-
tween adjacent commutator segments (for machines without commutators, the winding insulation
determines the limit). Clearly, bigger-sized machines can withstand higher voltage levels than
smaller. If the machine is supplied from a power electronic converter, it is natural to set Vbase
to the maximum output voltage of the converter, unless this exceeds the maximum permissible
armature voltage. For a single-phase VSC—see Section 3.2—this is given by the dc-link voltage,
so it is logical to set Vbase = Vdc .
Second, the base flux linkage ψbase is determined, as described above.
Next, we can determine the base speed. Considering (4.17) in the steady state, we have
v = Ri + ψωr . (4.20)
Normally, the resistive voltage drop Ri is much smaller than the back emf ψωr , especially so when
the torque is low, yielding a small armature current i. Thus, we may approximate v ≈ ψωr . From
114 Chapter 4. DC Motor Drives

Table 4.2. DC-machine base values.

Voltage Vbase (Limited by commutator or converter)

Current Ibase (Limited by power losses or converter)
Flux linkage ψbase (Limited by magnetic saturation)
Speed ωbase = Vbase /ψbase
Time tbase = 1/ωbase
Power Pbase = Vbase Ibase
Torque τbase = Pbase /ωbase
Impedance Zbase = Vbase /Ibase
Inductance Lbase = Zbase /ωbase
2
Inertia Jbase = τbase /ωbase
Viscous damping constant bbase = τbase /ωbase

this relation the base speed is obtained as

Vbase
ωbase = . (4.21)
ψbase

Next to be determined is the base current. Multiplying (4.20) by i, we obtain the input–output
power balance

./01 Ri2 + ψωr i .

vi = ./01 (4.22)
. /0 1
Pin Ploss Pout

In general, the efficiency of a dc machine is fairly high (although not as high as for an ac machine),
with η = Pout /Pin up to 80%. Still, it is the power loss Ploss = Ri2 that is the limiting factor, since
this is converted to heat in the armature circuit. The base current Ibase is given by
=
max
Ploss
Ibase = (4.23)
R
max
where Ploss is the maximum acceptable power loss. Furthermore, if supplied from a power elec-
tronic converter which is rated at a lower current than the machine, then this is of course what
determines Ibase . From the fundamental base values for voltage, current, and flux, an extended
list of base values is shown in Table 4.2.

4.2.6 Normalization of the DC Machine Model

The dc machine model is normalized in the same fashion as described in Section 2.8. Normalization
of the electrical dynamics is straightforward. Consider (4.17)

di
L = v − Ri − ψωr . (4.24)
dt
4.2. DC Machine Fundamentals 115

Let us introduce normalized values as defined in Table 4.2

, -
Zbase d(in · Ibase )
Ln · = (vn · Vbase ) − (Rn · Zbase )(in · Ibase )
ωbase d(tn /ωbase )
−(ψn · ψbase )(ωrn · ωbase ). (4.25)
But Zbase Ibase = Vbase and ψbase = Vbase /ωbase , so we obtain
din
Ln = vn − Rn in − ψn ωrn . (4.26)
dtn
As in Section 2.8, the differential equation is formally unchanged by the normalization. We now
introduce normalized quantities in the mechanical dynamics (4.18)
d(ωrn · ωbase )
J = (ψn · Vbase /ωbase )(in · Ibase ) − b(ωrn · ωbase ) (4.27)
dtn /ωbase
giving
, 3
- , 2
-
ωbase dωrn ωbase
J· = ψn in − b · ωrn (4.28)
Vbase Ibase dtn Vbase Ibase
which verifies the base values for inertia and viscous damping constant in Table 4.2 and gives the
following normalized equation:
dωrn
Jn = ψn in − bn ωrn . (4.29)
dtn
Again, the normalized equation is formally equal to the original equation.
Example 4.2 Suppose that we have a dc machine with the following data.

Base voltage: Vbase = 100 V

Base current: Ibase = 10 A
Base flux: ψbase = 2 Vs
Armature inductance: L = 60 mH
Armature resistance: R=1Ω
Inertia: J = 2.4 kgm2
Viscous damping constant: b = 0.4 kgm2 /s

The base power is then 1 kW and the base speed is

Vbase
ωbase = = 50 rad/s.
ψbase
We obtain the impedance base value as Zbase = Vbase /Ibase = 10 Ω, giving
R ωbase L
Rn = = 0.1 pu Ln = = 0.3 pu.
Zbase Zbase
The mechanical base values are
Vbase Ibase Vbase Ibase
Jbase = 3 = 8 · 10−3 kgm2 bbase = 2 = 0.4 kgm2 /s
ωbase ωbase
116 Chapter 4. DC Motor Drives

giving
Jn = 300 pu bn = 1 pu.
Note the large value of the normalized inertia; one notable exception to the rule that the per-unit
parameter values are in the range of 1 or below.
♣
PROBLEM 4.9
A dc motor with base voltage, current, and flux 400 V, 50 A, and 5 Vs, respectively, has a
nominal efficiency of 78%.
a) Determine the armature resistance in pu and in ohms.
b) Find the base speed and the speed at nominal load in rad/s.

4.3 Open-Loop Dynamics

It is now time to revisit the electrical and mechanical dynamics (4.17)–(4.18), rewritten here for
clarity
di
L = v − Ri − ψωr (4.30)
dt
dωr
J = ψi − bωr . (4.31)
dt
These equations correspond to the block diagram in Figure 4.9. (As we henceforth discuss only
motors, we now return to the term “motor” rather than “machine.”)

electrical
dynamics "torque gain"
v 1 i τe 1 ωr
ψ
sL sJ
E
R b
ψ
"back-emf feedback"
Figure 4.9. Block diagram of the dc-motor dynamics.

From this block diagram it is quite obvious that the flux ψ has the effect of interlinking the
electrical and mechanical dynamics, as we already found by inspecting the equations. As feedfor-
ward there is the gain in the electrical torque, ψi, and as feedback we have the back emf E = ψωr .
Suppose that the flux is reduced to zero, ψ = 0. The couplings between the electrical and the
mechanical dynamics then disappear, breaking the system into two noninteracting first-order sub-
systems. Expressing both in the form T dy/dt = ku − y, we have
L di 1
= v−i (4.32)
R dt R
J dωr
= −ωr . (4.33)
b dt
4.3. Open-Loop Dynamics 117

Hence, the electrical and mechanical time constants are, respectively, given as

L J
Te = Tm = . (4.34)
R b
The electrical time constant is much smaller than the mechanical: for the parameter values of
Example 4.2 we have Te = 3 pu and Tm = 300 pu, corresponding to 9.6 ms and 0.96 s, respectively,
for ωbase = 50 rad/s. Therefore, the electrical and the mechanical dynamics are on different time
scales when the motor is unmagnetized. (Although we here used a particular example, this is
almost always the case.)
What happens when the flux is restored to its nominal value? Based on Figure 4.9, we make
the following conjectures:

• The mechanical dynamics will be speeded up as ψ increases, due to the feedback formed by
the back emf.

• The electrical dynamics—forming via R an inner closed-loop system—will remain roughly

the same regardless of ψ.

Let us verify this, and also find out really how fast the mechanical dynamics become. With
x = [i, ωr ]T , (4.30)–(4.31) can be described in state-space form as ẋ = Ax + Bv, where
⎡ ⎤ ⎡ ⎤
R ψ 1
− −
⎢ L L ⎥
A=⎣ B = ⎣ L ⎦. (4.35)
ψ b ⎦
− 0
J J
Matrix A has the following characteristic polynomial:
, -, -
R b ψ2
det(sI − A) = s + s+ +
L J LJ
, -
R b Rb + ψ2
= s2 + + s+ . (4.36)
L J LJ
In Figure 4.10, loci for the poles are depicted, again for the parameter values of Example 4.2, as
ψ varies from 0 to 1 pu. For ψ = 0, there is a separation of two decades between the electrical
and the mechanical dynamics; the respective poles are located at s = −1/Te = −R/L = −0.33
and s = −1/Tm = −b/J = −0.0033. As ψ increases, the slow (mechanical) pole moves away from
the origin, becoming faster, whereas the fast (electrical) pole moves toward the origin, becoming
slower. However, the relative movement of the fast pole is not significant; the pole stays in the
vicinity of s = −0.3. On the other hand, the relative movement of the slow pole is as much as
one decade, starting at s = −0.0033 and ending at s = −0.04. Thus, we have verified both above
conjectures as true.
For most dc motors, the electrical dynamics are significantly faster than the mechanical dy-
namics (note in Figure 4.10 that even for ψ = 1 pu, the separation is one decade). Therefore, we
may consider the electrical dynamics in the steady state as seen from the mechanical dynamics.
The transfer function Ge (s) of the electrical dynamics, from v to i in Figure 4.9, is

1
Ge (s) = . (4.37)
sL + R
118 Chapter 4. DC Motor Drives

(a) (b)
0.2

0.15 0.02

0.1
0.01
0.05
Im(s)

Im(s)
0 0

−0.05
−0.01
−0.1

−0.15 −0.02
−0.2
−0.4 −0.3 −0.2 −0.1 0 −0.05 −0.04 −0.03 −0.02 −0.01 0
Re(s) Re(s)

Figure 4.10. Loci for the poles of the dc motor model for 0 ≤ ψ ≤ 1 pu. (a) Fast (electrical) and slow
(mechanical) poles. (b) Enlargement of the slow pole.

v 1 i τe 1 ωr
ψ
R sJ
E
b

Figure 4.11. Block diagram with the electrical dynamics considered in the steady state as seen from the
mechanical dynamics, obtaining a reduced-order model.

Steady state implies letting s = 0, so the electrical dynamics can be replaced by the steady-state
gain 1/R. See Figure 4.11.
We then arrive at the following reduced-order model for the dc motor:
, -
dωr ψv ψ2
J = − b+ ωr (4.38)
dt R R

and, since b ≪ ψ 2 /R normally,

dωr
RJ = ψv − ψ 2 ωr (4.39)
dt
giving the transfer function

ψ
Gv (s) = (4.40)
sRJ + ψ 2

(where the subscript v stands for “voltage input”) from v to ωr . The pole is located at s = −ψ 2 /RJ.
At nominal flux and with R = 0.1 pu and J = 300 pu, this is equal to s = −0.033, which agrees
with Figure 4.10(b).
4.3. Open-Loop Dynamics 119

Let us finally simulate a start of the dc motor in Example 4.2, with ψ = 1 pu. Figure 4.12
shows the result. We see that the the speed rises with approximately the time constant 0.6 s, i.e.,
30 pu. This is the inverse absolute value of the pole at s = −0.033, and thus agrees with the
theoretical result. It can also be noted that the current rises fast: the time constant here is about
0.06 s (3 pu), corresponding to the electrical pole at s = −R/L = −0.3.
Notable is the large current peak. The reason for this is obvious: when the motor is standing
still, there is no back emf. Only the armature resistance then limits the current. The peak is
undesirable, however. Even though the motor itself can withstand a starting current much larger
than base current—provided that the duration is short enough—it may cause the tripping of an
overcurrent relay. To avoid this problem, closed-loop current control is preferably used for variable-
speed dc drives. This is the topic of the next section.

10

8
i (−) ωr (−−) (pu)

0
0 0.5 1 1.5 2 2.5 3
t (s)

Figure 4.12. Open-loop start of a dc motor.

PROBLEM 4.10
A dc motor with armature resistance 0.3 Ω is started by direct connection to a 100-V source.
After 5.5 s the speed has reached 120 rad/s. The speed then settles to 150 rad/s in the steady
state. Determine the total inertia and the motor’s flux linkage.

PROBLEM 4.11
A dc motor with total inertia 1.8 kgm2 and armature resistance 0.5 Ω is fed from a 200-V
source and is rotating with 1500 rpm. An external load torque τL = 30 Nm is applied at
t = 0. Calculate the speed at t = 1.5 s. Hint: The dynamics (4.39) are modified to

dωr
RJ = ψv − RτL − ψ 2 ωr .
dt

PROBLEM 4.12
A dc motor with R = 4 Ω and ψ = 2 Vs is driving a load with a relatively high total inertia:
J = 8 kgm2 . The viscous damping is not negligible, but the constant b is unknown. To find b,
an experiment is conducted. A 48-V source is connected to the motor’s armature terminals,
and the motor is allowed to accelerate, taking 15.3 s to attain 90% of the steady-state speed.
Find b and the steady-state speed.
120 Chapter 4. DC Motor Drives

4.4 Current Control

We shall in this section study armature current control of dc motor drives. High-performance
control is considered, which implies that the motor is fed from a PWM converter. Such dc motors
drives are today largely obsolete, having been replaced by ac motor drives in most applications.
The section thus primarily serves as an introduction to the concept of current control that will be
built upon in Chapter 6, where vector current control for ac motor drives is considered.
The reasons for using closed-loop current control are to further improve the response time of
the armature current i, to allow precise torque control, and (as mentioned above) to prevent i
from exceeding a certain maximum allowed value (which in the steady state is Ibase , whereas at
transients a larger value Imax may be allowed). We have, from (4.17) (where E = ψωr )
di
L = v − Ri − E. (4.41)
dt
Using the transfer function (4.37), this can be expressed as the block diagram in Figure 4.13(a).
Note that the back emf E enters Ge (s) in the same way as the input signal v (but with opposite
sign). Since the mechanical dynamics are slow, these can be disregarded as seen from the current
control loop, and E treated as a disturbance. A disturbance entering the process in the same way
as the control input is known as a load disturbance.

E E
v i iref v i
Ge (s) CC Ge (s)

(a)
(b)

iref e v
Fc (s)

i (c)

Figure 4.13. Block diagram of the electrical dynamics with Ge (s) = 1/(sL + R) and the back emf E
acting as load disturbance. (a) Open-loop. (b) Closed-loop (“CC” is the current controller). (c) One-
degree-of-freedom PI controller.

We now close the loop around the electrical dynamics, as shown in Figure 4.13(b). The armature
current i is measured and compared to the reference (setpoint) iref . The error signal e = iref − i
forms the input to the current controller—see Figure 4.13(c)—which manipulates the voltage v via
a power electronic converter, such that i follows iref quickly and accurately.

4.4.1 One-Degree-of-Freedom Controller Design

Since the electrical dynamics are of order one, a simple PI controller is appropriate (the order of
the controller need not be higher than that of the controlled process)
ki
Fc (s) = kp + . (4.42)
s
4.4. Current Control 121

As the controller operates on the error e only, as shown in Figure 4.13(c) (not also on i individually),
this is called a one-degree-of-freedom PI controller. (This is by far the most common type of
controller used in practice. For example, it is used in more than 90% of all control loops in the
process industry.)
Let us now select the controller parameters kp and ki . It is not uncommon that this is done
by trial and error. This, however, is inconvenient. Since we know (or at least have good estimates
of) the motor parameters, it is logical to use this knowledge for selecting suitable kp and ki . A
convenient method is direct synthesis [73]. The idea is as follows. Let us start with a specification
for the 10%–90% rise time trc of the closed-loop system. Typically, rise times in the millisecond
range can be obtained using state-of-the-art dc–dc converters or single-phase VSCs. With Ge (s)
being of order one, it is apparent that a closed-loop transfer function of order one also can be
obtained. Let Gcc (s) be the closed-loop transfer function from iref to i. We now specify that
Gcc (s) (ideally) should be

αc αc /s
Gcc (s) = = (4.43)
s + αc 1 + αc /s
where αc is the closed-loop system bandwidth. In the steady state, there should then be no control
error, since Gcc (0) = 1. From (2.37) it is known that the relation between bandwidth and rise time
is αc trc = ln 9, so the specification for trc can be directly translated to a specification for αc . But
Fc (s)Ge (s)
Gcc (s) = (4.44)
1 + Fc (s)Ge (s)

so if Fc (s)Ge (s) = αc /s, the desired closed-loop system (4.43) is obtained. This yields
αc −1 αc αc R
Fc (s) = G (s) = (sL + R) = αc L + . (4.45)
s e s s
That is,
!
kp = αc L !
ki = αc R (4.46)

where “hats” are introduced to denote model motor parameters (which should be as close estimates
of the true motor parameters as possible). The controller parameters are now expressed directly
in the desired closed-loop bandwidth and the model motor parameters L ! and R,
! so trial-and-error
steps are avoided.
Since the controller is designed using the inverse of Ge (s)—(4.45)—the pole of Ge (s) is canceled
! = L and R
(ideally, if L ! = R). The motor dynamics are “replaced by something better.” Direct
synthesis is a special case of internal model control (IMC) [26, 62]. We may therefore refer to the
selection (4.46) also as “IMC design.”

Example 4.3 Let us design and evaluate a current controller for the dc motor drive in Example
4.2. This time, however, the mechanical load is lighter, J = 50 pu. The mechanical time constant
is then J/b = 50 pu, i.e., 1 s, giving a rise time in the rotor speed of 2.2 s. If we select the rise time
in the current control loop as 22 ms (which is rather undemanding for modern power electronic
converters), there is a two-decade separation between the electrical and the mechanical dynamics.
Thus, in per-unit quantities, trc = 1.1, giving αc = 2.2/1.1 = 2. The controller parameters are
obtained as
kp = 0.6 pu ki = 0.2 pu.
122 Chapter 4. DC Motor Drives

We evaluate the controller by starting the motor at t = 0 by stepping iref from 0 to 1 pu, see Figure
4.14(a). Although the desired rise time is obtained, the result is far from satisfactory: i does not
reach the reference (at least not over the time interval displayed). Below, it is explained why.

1
(a): i (−) ωr (−−) (pu)

0.8

0.6

0.4

0.2

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
t (s)

1
(b): i (−) ωr (−−) (pu)

0.8

0.6

0.4

0.2

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
t (s)

Figure 4.14. Start of a dc motor with (a) one-degree-of freedom PI current control and (b) two-degrees-
of-freedom PI current control.

4.4.2 Load-Disturbance Rejection

The reason for the poor accuracy that can be observed in Figure 4.14(a) is the load disturbance
caused by the back emf [cf. Figure 4.13(b)]. Let us analyze the ability of the control system to
suppress a load disturbance. From Figure 4.13 the transfer function from the disturbance E to the
control error e can be calculated as
Ge (s) s s 1
GEe (s) = = Ge (s) = . (4.47)
1 + Fc (s)Ge (s) s + αc s + αc sL + R

GEe (s) has a zero at the origin, which implies differentiation. Therefore, a constant disturbance E
will be fully rejected, but not one that is time varying. The steady-state gain from dE/dt to the
control error is [GEe (s)/s]s=0 = 1/αc R. That is, if dE/dt is approximated as constant—implying
that the speed varies as a ramp, which is a reasonable approximation seen over a limited time
4.4. Current Control 123

interval—we obtain
1 dE ψ dωr
e= = . (4.48)
αc R dt αc R dt
The mechanical dynamics (4.18) give
dωr 1
= (ψi − bωr ) (4.49)
dt J
and
ψ
e= (ψi − bωr ). (4.50)
αc RJ
Returning now to Example 4.3, approximating i ≈ 1 pu and ωr ≈ 0.2 pu (the mean value over the
displayed interval), we get
1
e= (1 − 0.2) = 0.08 pu. (4.51)
2 · 0.1 · 50
This agrees well with Figure 4.14(a).
Not until ωr reaches a steady state does the current control error fully disappear. In some cases,
a control error of this size (nearly 10% of the 1-pu reference step) is unimportant. Then again,
in other cases—for example, when high-accuracy torque control is required—this is too large an
error. Equation (4.50) allows straightforward checking.

4.4.3 Feedforward of the Back EMF

If the speed ωr is measured, one remedy of poor load-disturbance rejection is to add a term

E ! r
! = ψω (4.52)

to the current controller output, as shown in Figure 4.15(a). As the flux linkage is usually known
with good accuracy, E! should be a good estimate of E = ψωr . The influence of E is thus cancelled.
Unfortunately, many dc motor drives do not use a speed sensor, because the speed easily can be
estimated, as we shall see in Paragraph 4.6.1. Another method for improved disturbance rejection
is then needed.

4.4.4 Two-Degrees-of-Freedom Controller Design

Poor rejection of load disturbances is characteristic for designs that involve pole cancellation, such
as direct synthesis. The performance can be improved somewhat by modifying (4.46) so that the
integral action increases. This, however, has the effect of giving overshoots when the reference is
changed.
It is seen in (4.50) that if R could be increased, the control error would decrease. Rather than
actually adding more resistance (which of course is highly undesirable, due to increased losses),
we introduce an inner feedback loop as shown in Figure 4.15(a), using an “active resistance” Ra .
This technique was proposed in [12].
The closed-loop system G′e (s) from v ′ to i is now
Ge (s) 1
G′e (s) = = . (4.53)
1 + Ra Ge (s) sL + R + Ra
124 Chapter 4. DC Motor Drives

Ge’(s)
CC E E kp
iref e v’ v i e v’
Fc(s) Ge (s) ki
I
1/s
Ra (b)

(a)

Figure 4.15. (a) Two-degrees-of-freedom current controller (CC) using an “active resistance” Ra and
! Ge (s) = 1/(sL + R). (b) Detail of PI controller
possibly including feedforward of the back-emf estimate E.
Fc (s), where I is the integrator state variable.

The resistance has in effect been increased from R to R + Ra , but since this is done just using
signals—involving no energy transfer—there will be no additional power loss. The PI controller
Fc (s)—see Figure 4.15(b)—can now be designed using direct synthesis
αc ′ −1 sL + R + Ra ki
Fc (s) = G (s) = αc = kp + . (4.54)
s e s s
That is,
!
kp = αc L ! + Ra ).
ki = αc (R (4.55)

Regarding the selection of Ra , it is useful to make the inner feedback loop G′e (s) as fast as the total
closed-loop system, i.e., with bandwidth αc . Then, Ra should be selected such that (R + Ra )/L =
αc , giving
! − R.
Ra = αc L ! (4.56)

This in turn yields the controller parameters

!
kp = αc L !
ki = α2c L. (4.57)

The described method yields a controller that has two inputs: the control error, but also i directly
via the “active resistance.” Therefore, it is called a two-degrees-of-freedom PI controller. Note
that the controller [i.e., the dotted box in Figure 4.15(a)] still fits within the structure of Figure
4.13(b).
It is straightforward to show that the control error formula (4.50) is unchanged, except that R
is replaced by R + Ra
ψ
e= (ψi − bωr ) (4.58)
αc (R + Ra )J
! = R and L
and, with Ra given by (4.56) (if R ! = L),

ψ
e= (ψi − bωr ). (4.59)
α2c LJ
4.4. Current Control 125

Evaluating this for Example 4.3, we find that

1
e= (1 − 0.2) = 0.013 (4.60)
22 · 0.3 · 50
which is much more acceptable a value than (4.51). We also repeat the simulation of Example
4.3, obtaining the result shown in Figure 4.14(b). As can be seen, the performance is now much
improved, and in agreement with the theory.

PROBLEM 4.13
Show (4.58)!

4.4.5 Voltage Saturation and Antiwindup

So far, we have treated the current control loop as an ideal, linear system. In practice, this
is not correct. The armature voltage v cannot be made arbitrarily large; it is limited by an
upper and a lower value. Both are determined by the rating of the power electronic converter
that supplies the voltage. For a single-phase VSC the lower and upper limits are −Vbase and
Vbase , respectively. In Example 4.3—although not shown—the voltage never reached 1 pu, so no
problems were encountered. However, for large steps in iref , the current controller’s output signal
often exceeds Vbase , especially for higher rotor speeds when the back emf is large, requiring in turn
a large armature voltage. The true voltage v is therefore becomes a saturation of the ideal voltage.
We shall denote the saturation function as
⎧ ⎫
⎨ −a, x < −a ⎬
x̄ = sat (x, a) = x, −a ≤ x ≤ a = min(|x|, a) sgn (x). (4.61)
⎩ ⎭
a, x > a

The saturation function is illustrated in Figure 4.16(a). With vref as the ideal voltage reference
(i.e., the output signal of the current controller) and v̄ref as the saturated voltage reference, we
thus have

v̄ref = sat (vref , Vbase ). (4.62)

Obviously, v̄ref rather than vref should be used as reference to the pulsewidth modulator (so
v = v̄ref + switching harmonics), see Section 3.2. Because of the saturation the current control
loop contains a nonlinearity, as shown in Figure 4.16(a).

Example 4.4 Let us again simulate the system of Example 4.3, this time with two-degrees-of-
freedom controller design. As previously, the motor is started at t = 0, but now overcurrent is
allowed: iref is stepped from 0 to 4 pu (it is assumed that the duration of the application of
overcurrent is short enough, so that nothing goes wrong). Figure 4.17(a) shows the result. As we
can see, i makes a fairly large overshoot. This occurs due to integrator windup: the integral term of
the PI controller keeps accumulating the control error during the time of maximum voltage output.
When i starts to get close to iref , the integrator has wound up, so vref remains large. Therefore, i
shoots over iref until the windup has been worked off by accumulation of negative control error.

♣
126 Chapter 4. DC Motor Drives

sat( x ,a)
a
E
iref i
x CC vref sat vref Ge (s)
−a a

−a
(a) (b)

Figure 4.16. (a) Saturation function. (b) Current control loop with voltage saturation, where dashed
lines indicate signal paths for “back calculation.”

“Back Calculation”

Example 4.4 illustrates well the performance degradation that typically results from integrator
windup. Clearly it is important to prevent that windup occurs, and several methods have been
proposed for this purpose. We shall here consider a method where the integrator input is altered
from the actual control error e to a modified control error ē. Including this alteration the controller
shown in Figure 4.15 can be described by the control law

dI
= ē (4.63)
dt
!
vref = kp e + ki I − Ra i + E (4.64)
v̄ref = sat (vref , Vbase ). (4.65)

How should ē be chosen in order to prevent windup? Notice that e is still used in the proportional
part, i.e., in (4.64). Yet, let us suppose for a moment that ē were used also in (4.64). Let us further
assume that ē is chosen such that (4.64) with e → ē gives v̄ref

!
v̄ref = kp ē + ki I − Ra i + E. (4.66)

In other words, the modified control error ē is assumed to be such that saturated operation never
is entered, instead the linear law (4.66) balances on the saturation boundary. Obviously, integrator
windup then would not occur! Subtracting (4.64) from (4.66) allows us to find an expression for
the modified control error as

1
ē = e + (v̄ref − vref ). (4.67)
kp

The method is called “back calculation,” because given the saturated reference v̄ref , (4.67) cal-
culates “backwards” to obtain the modified control error ē that would be needed to balance on
the saturation boundary. Notice that the modified control law (4.66) is never implemented, only
the “back-calculation” algorithm (4.67), which is indicated by the dashed lines in Figure 4.16.
Let us for clarity rewrite (4.63)–(4.65) with (4.67) incorporated. The control law including “back
4.4. Current Control 127

i (−) v (−−) (pu) 3

0
0 50 100 150
t (ms)

4
i (−) v (−−) (pu)

0
0 50 100 150
t (ms)

Figure 4.17. Start of dc motor with two-degrees-of freedom PI current control. (a) Without “back
calculation.” (b) With “back calculation.”

calculation” is given by
dI 1
= e + (v̄ref − vref ) (4.68)
dt kp
vref = kp e + ki I − Ra i + E! (4.69)
v̄ref = sat (vref , Vbase ). (4.70)

Example 4.5 We repeat the simulation of Example 4.4 that gave Figure 4.17(a), this time with
“back calculation” included. Performance is now acceptable, as can be observed in Figure 4.17(b).
♣

4.4.6 Digital Implementation

Up until the early 1990s, analog implementation of control algorithms for electrical drives was state
of the art. At that time, the development of microprocessors and DSPs started reaching levels
that permitted digital implementation at reasonable costs. Naturally, digital implementation is
preferable to analog, since the algorithms can be modified easily and nonlinear algorithms (which
are frequently used in ac drives, as we shall see later) can be implemented without difficulties.
Digital implementation calls for discretization of the control algorithms derived in continuous
time, see Paragraph 2.4.7. Applying the Euler method (2.79) for discretization of the integrator
128 Chapter 4. DC Motor Drives

(4.68), control law (4.68)–(4.70) results in Algorithm 4.1 and the block diagram shown in Figure
4.18.

Algorithm 4.1
Current Controller for DC Drives

e = iref − i
!
vref = kp e + ki I − Ra i + E
v̄ref = sat(vref , Vbase )
( )
1
I = I + Ts e + (v̄ref − vref )
kp

1/kp

Ts
D

I
ki
iref
e kp vref sat vref

i Ra E =ψωr

Figure 4.18. Digitally implemented two-degrees-of-freedom current controller.

4.4.7 Bandwidth Selection

The closed-loop bandwidth αc must not be selected too high in relation to the sampling frequency,
or system performance will degrade, possibly so far that instability results. In the current control
loop the total time delay Td = 1.5Ts , which is the sum of the controller and PWM time delays,
see (3.16), must be taken into account. For a worst-case scenario, we assume a small armature
resistance. The electrical dynamics, including the time delay, are then given by the transfer function

e−sTd
Ge (s)e−sTd = . (4.71)
sL

One-Degree-of-Freedom PI Control
Stability of the closed-loop system can be analyzed by applying the Nyquist criterion. The con-
troller is given by (4.45) as Fc (s) = (αc /s)G−1
e (s). This yields the open-loop transfer function

αc e−sTd
Gk (s) = Fc (s)Ge (s)e−sTd = . (4.72)
s
4.4. Current Control 129

|Gk (jαc )| = 1, i.e., αc is the crossover frequency of the open-loop system. In order for the Nyquist
curve not to encircle the critical point −1, the phase angle at the crossover frequency must be
larger than −π, i.e., arg Gk (jαc ) = −π/2 − αc Td > −π ⇒ αc Td < π/2. For Td = 1.5Ts = 3π/ωs
we thus obtain the criterion αc < ωs /6 for asymptotic stability. The following recommendation
for bandwidth selection is given:

αc < 0.1ωs (4.73)

which gives the phase and gain margins 36◦ and 1.7, respectively, see Figure 4.19(a).

Two-Degrees-of-Freedom PI Control
In this case, the integral gain is set as ki = α2c L, whereas the proportional gain is effectively
doubled owing to the “active resistance” inner feedback loop. This yields the following open-loop
transfer function:
, -
−sTd α2c L e−sTd αc (2s + αc )e−sTd
Gk (s) = [Fc (s) + Ra ]Ge (s)e = 2αc L + = . (4.74)
s sL s2
" √
For Td = 1.5Ts = 3π/ωs it is found that the crossover frequency is now αk = 2 + 5αc ≈ 2.06αc
and that αc < 0.07ωs is required for asymptotic stability. The following recommendation for
bandwidth selection is given:

αc < 0.04ωs (4.75)

which gives the phase and gain margins 32◦ and 1.9, respectively, see Figure 4.19(b).

(a) (b)
0.6 0.6

0.4 0.4

0.2 0.2
Im{Gk(jω)}

Im{Gk(jω)}

0 0

−0.2 −0.2

−0.4 −0.4

−1 −0.8 −0.6 −0.4 −0.2 0 −1 −0.8 −0.6 −0.4 −0.2 0

Re{G (jω)} Re{G (jω)}
k k

Figure 4.19. Nyquist curves for (a) one-degree-of-freedom PI control with (solid) αc = 0.1ωs and (dashed)
αc = 0.17ωs ; and (b) two-degrees-of-freedom PI control with (solid) αc = 0.04ωs and (dashed) αc = 0.07ωs .

PROBLEM 4.14
A dc motor has the following per-unit parameters: L = 0.2, R = 0.1, ψ = 1, J = 200, and
b = 0.1. The base speed is 100 rad/s.
130 Chapter 4. DC Motor Drives

a) Determine the efficiency η = output power/input power at nominal speed and load.
b) Determine the armature current and speed in per-unit values at nominal load if the flux
is reduced such that ψ = 0.2 pu.
c) The drive control system for the dc motor, which uses a PWM converter with a 1-kHz
switching frequency and a DSP sampling at 2 kHz, the current control law is pre-coded
in the following way (excluding antiwindup and saturation):
$
vref = Kp (βiref − i) + Ki (iref − i) dt.

This is a two-degrees-of-freedom PI controller that uses the so-called “setpoint weight-

ing” algorithm (factor β). The parameters are free to select by the operator. Determine
the parameters such that performance identical to the two-degrees-of-freedom PI con-
troller presented in this section is obtained.

4.5 Field Weakening

When neglecting the armature resistance, then v = E = ωr ψ in the steady state. Since |v| cannot
exceed Vbase , |ωr | cannot exceed Vbase /ψbase = ωbase . The dc motor is prevented by the back emf
E = ψωr from operating at speeds higher than base speed for a constant ψ. The operating range
can be extended to speeds higher than |ωbase | by lowering ψ, which is called field weakening. (The
method is of course not applicable to dc motors magnetized by permanent magnets.) By selecting
⎧
⎨ ψbase , |ωr | ≤ ωbase
ψ= ωbase (4.76)
⎩ ψbase , |ωr | > ωbase
|ωr |

(see also Figure 4.20), |ωr | can in theory increase arbitrarily much. Since the electrical torque is
given by τe = ψi, the torque capability is gradually reduced as the motor goes further into the
field-weakening region. Therefore, a point is eventually reached when the speed cannot increase
further due to lack of torque. Also, sparking in the commutator is a problem which puts a limit
on the speed in the field-weakening region. This in practice limits the maximum speed to about
3ωbase .

ψ
ψbase Field−weakening region
=> Reduced torque
Maximum
torque region

ωbase ωr

Figure 4.20. Principle of field weakening.

4.6. Speed Control 131

Remark 4.1 Analogy to field weakening: vehicle drive. An interesting analogy to the
relation τe = ψi and the concept of field weakening can be made for an vehicle (e.g., an automobile)
drive, where a gear box is used. Higher gear allows higher speed, but reduces the torque capability.
(When high torque is needed, e.g., for sufficient acceleration when overtaking another vehicle, the
driver often drops a gear.)
Thus, the gear box corresponds to ψ: applying a higher gear is analogous to reducing the flux.
But unlike a dc motor, the “field weakening” of the gear box is made in steps.2 The power delivered
from the combustion engine corresponds to i, and is controlled by the accelerator.
There is another interesting analogy. As concluded earlier, the armature current can be changed
quickly, owing to a relatively small armature inductance. The flux linkage on the other hand
takes longer to change, because the field-winding inductance is much larger than the armature
inductance. Similarly, the power of the combustion engine is quickly changed, simply by depressing
or relieving the accelerator, whereas switching gears takes longer time.
PROBLEM 4.15
In commercial propeller aircraft, the propeller blades can be pitched. Such a plane is normally
operated in the following way. On the ground, the propellers are revved up to their nominal
speed. Since the blades are not pitched, no torque is produced, and the plane does not move.
A small pitch gives enough torque for the plane to taxi to the runway. There, maximum
pitch is applied, the plane accelerates, and takes off. At cruising altitude, the propeller speed
is reduced in order to save fuel.
a) What are the correspondences to i and ψ in a propeller aircraft?
b) When landing, the propeller speed is again increased to the nominal, even though only
light torque is required. Why? Is this similar to how a dc motor is operated?

4.6 Speed Control

In addition to a current control loop, many drives also use an outer speed control loop. The speed
is measured or estimated—the latter called speed-sensorless control, see Paragraph 4.6.1—and fed
back to a speed controller, whose output is the armature current reference.

In a cascade control structure, the bandwidth of the outer loop should be 1/10 or less of
the bandwidth of the inner loop. With αs as bandwidth of the speed loop, we thus have the
recommendation

αs ≤ 0.1αc . (4.77)

Consider again the mechanical dynamics (4.18). So far it has been assumed that the load torque is
proportional to the speed: τl = bωr . As mentioned previously, this is a useful but not always correct
simplifications. In many applications, such as pumps and fans, the load torque is proportional to
the square of the speed. In other applications, such as rail vehicles, hoists (see Problem 4.4), and
escalators, a large part of the load torque can be attributed to gravity and is speed independent.
A more accurate model for the load torque is thus

τl = bωr + τL . (4.78)
2 One notable exception was the belt drive developed by the Dutch automobile manufacturer DAF (acquired by

Volvo in 1975).
132 Chapter 4. DC Motor Drives

ωref Speed iref Current vref

controller controller

i
ωr

Speed
estimator

speed−sensorless ωr

speed−sensored

Figure 4.21. Cascaded current and speed control loops.

That is, a viscous part plus a part due to external conditions (the “external load torque”). This
gives the following model for the mechanical dynamics:
dωr
J = ψi − bωr − τL . (4.79)
dt
Let us now design the speed controller, assuming that cascaded current control is used, with the
speed controller’s output being the reference to the current controller. The external load torque
τL enters the system as a load disturbance, much like the back emf in the current control loop.
Load disturbances (in the form of load torque variations) tend to be more pronounced in the
speed control loop than in the current control loop. It is therefore advisable to incorporate a
correspondence to the “active resistance” used in the current controller also in the speed control
loop. By introducing an inner feedback loop in the form of an “active viscous damping,” as
i = i′ − ba ωr (4.80)
the pole of (4.79) can be moved from −b/J to −αs , where αs is the desired closed-loop bandwidth.
The dynamics are then
dωr ψ b τL
= (i′ − ba ωr ) − ωr −
dt J J J
ψ ′ ba ψ + b τL
= i − ωr − . (4.81)
J . /0J 1 J
αs
Hence, if

αs J! − !b
ba = (4.82)
ψ!
(where “hat” as usual indicates model parameter), the dynamics of the inner loop created by the
“active viscous damping” will have bandwidth αs , assuming accurate model parameters, giving
4.6. Speed Control 133

the following transfer function from i′ to ωr :

ψ/J
Gs (s) = . (4.83)
s + αs

For fast speed control loops, αs J ≫ b, so ba ≈ αs J/ ! the viscous damping constant is then
! ψ;
unimportant for the controller design. Having designed the inner loop, the outer control loop is
then closed with a PI controller. With controller design using direct synthesis and model parameters
replacing true ones, we obtain

αs −1 kis
Fs (s) = Gs (s) = kps + (4.84)
s s
where

αs J! α2s J!
kps = kis = . (4.85)
ψ! ψ!

The output of the speed controller is the reference for the current controller. The final two steps
in the controller design remain. First, limitation of iref is required, so that

|iref | ≤ Imax

where Imax = Ibase in the steady state, whereas overcurrent, Imax > Ibase , may be allowed short
term (seconds, in some cases up to one minute, depending on the thermal characteristics of the
motor). Second, “back calculation” should be incorporated. We obtain the algorithm shown below,
which corresponds to the block diagram shown in Figure 4.22.

Algorithm 4.2
Speed Controller for DC Drives

es = ωref − ωr
inom
ref = kps es + kis Is − ba ωr
iref = sat(inom
ref , Imax )
( )
1 nom
Is = Is + Ts es + (iref − iref )
kps

4.6.1 Speed Estimation for Sensorless Control

Unless very precise speed control is required, a speed sensor is not needed in a dc motor drive.
Because the back emf is proportional to the speed and differs from the armature voltage only by
the voltage drop across the armature resistance: v = Ri + ψωr in the steady state, a speed estimate
is readily computed as

!
v − Ri
ω
!r = (4.86)
ψ!
134 Chapter 4. DC Motor Drives

1/k ps

Ts
D

Is
kis
ωref
es sat
kps inom
ref iref
ωr ba
Figure 4.22. Digitally implemented two-degrees-of-freedom speed controller.

where R ! is an estimate of the armature resistance. When ω !r is fed back to the speed controller,
an algebraic loop is created, however, because (the reference for) i is the speed controller’s output.
This can be resolved by embedding (4.86) in a low-pass filter
> ?
v̄ref − ! ref
Ri
!˙ r = αl
ω −ω
!r (4.87)
ψ!

where αl is the filter bandwidth. Note that instead of the measured voltage and current, their
reference values can preferably be used (the saturated reference v̄ref is the output of the current
controller, cf. Algorithm 4.1). Using references rather than measured signals reduces the noise
content of the speed estimate. The filter bandwidth shall be selected significantly larger than the
speed-control-loop bandwidth, i.e.,

αl ≫ αs . (4.88)

No undesirable lagging effect due to the low-pass filter then occurs.

PROBLEM 4.16
Feedforward of the back emf in the current controller cannot be used when the speed is
estimated. Explain why!

PROBLEM 4.17
! in (4.87) is neglected, being small. Would this resolve the algebraic
Suppose that the term Ri
loop? Explain carefully!
CHAPTER 5

AC Motor Drives

The objective of this chapter is to derive models for, and study operating principles of, ac motors.
As much as possible we rely on the theory developed for the dc motor in Chapter 4. Traditional
texts on ac motors tend to have their main focus on line-connected motors. In such, the stator
(i.e., fundamental) frequency is constant (e.g., 50 Hz in Europe and 60 Hz in North America),
and steady-state analysis is often sufficient. The classical tools in this are phasor diagrams and
equivalent circuits.
As our focus is variable-speed drives with converter-fed motors, the picture is more complicated.
Unlike when directly fed, the converter allows variable stator frequency, and also variable amplitude
and phase of the stator voltage. When the amplitude, phase, or frequency of the stator voltage is
changed, the motor enters a transient mode which cannot be described by a phasor diagram or a
traditional equivalent circuit.
Space vectors are the main tool for dynamic analysis of ac motors. Dynamic space-vector
models for the induction motor and the permanent-magnet synchronous motor are introduced,
and the classical volts-per-hertz control method is discussed. For the sake of completeness, we
also briefly study line-connected induction motors. The chapter is finished by discussions on speed
control and determination of motor parameters.

5.1 AC Machine Fundamentals

In this section, fundamental properties for ac machines1 are presented. Figure 5.1—which will be
explained in more detail below—shows the basic physical principles of three-phase synchronous
and induction machines. The stator is modeled as an equivalent two-phase winding with two
perpendicular coils. The SM uses a directly magnetized rotor, oriented along the d axis of the
synchronously rotating dq coordinate system. The IM instead has a rotor winding, here modeled
as two short-circuited coils.

5.1.1 Synchronous Machines

An SM is, in principle, a dc machine turned inside out. The magnetization (rotor flux) is produced
from inside the rotor, as shown in Figure 5.1(a), either by permanent magnets or by a field winding
supplied from an external dc source. The former is called a permanent-magnet SM and the latter
an electrically excited SM.
In the latter case, if the rotor is round—see further below—the field winding is placed in slots
in the rotor surface, as shown in Figure 5.2. Unlike the armature current in a dc motor, the current
of the field winding needs no commutation, so the commutator is replaced by two slip rings, on
1 As the properties are general for both motors and generators, we use the term “machine” in this section, even

though “motor” is preferred elsewhere.

136 Chapter 5. AC Motor Drives

β β

. .
ωr = θ ωr = θr
d d θ
q
θ=θr θr
N α α
S
q

(a) (b)

Figure 5.1. (a) Synchronous machine. (b) Induction machine. Symbols: θ—flux angle, θr —rotor position,
ωr —angular rotor speed.

which metal—not carbon, as in dc motors—brushes slide. The brushes therefore do not wear down
as in a dc motor, so the SM requires significantly less maintenance. SMs of high power ratings
often have a brushless exciter instead of slip rings. This is a smaller three-phase machine fitted on
the same shaft as the SM, whose rotor phase voltages are rectified to supply the field winding. As
the field current cannot be directly measured, brushless-exciter SMs are somewhat more difficult
to control than SMs with slip rings.
Corresponding to the armature winding of the dc motor—which carries the torque-producing
armature current—is the stator winding. This is placed in slots in the stator, as shown in Figure
5.2, and is wound with several turns of insulated copper wire. To prevent large core losses resulting
from the ac excitation, the stator core is made of laminated steel.

stator stator
stator slots
slip
shaft rings
rotor
airgap
rotor slots
rotor

.
(a) (b)

Figure 5.2. Constructional principle of a round-rotor SM. (a) Stator–rotor cross-sectional view. (b) View
from the side.

Torque is produced by the interaction of the airgap flux, i.e., the part of the rotor flux that
crosses the airgap, see Figure 5.3, and the stator current. The rotor leakage flux, i.e., the part
of the rotor flux that closes back into the rotor without crossing the airgap (not shown in Figure
5.3), does not contribute to the torque and should be kept low. Therefore, the airgap should be
narrow. The airgap width has to be a compromise, however, as manufacturing as well as cooling
5.1. AC Machine Fundamentals 137

(during operation) become easier for wider airgaps. Typical airgap widths of SMs are between 1
and 3 mm.
Electrically excited SMs are used as generators in large-scale generating stations. As motors
they are used in for example pumps, compressors, and rolling mills, typically in applications which
require very high power, up into the megawatt range. Having no commutator the SM has low
losses, with efficiencies typically between 90% and 95%.

β
stator
q d
Φ/2 N
F θ

S α
F rotor

Φ/2

Figure 5.3. Two-pole SM with flux paths.

Number of Pole Pairs

The term synchronous refers to the rotation of the rotor, which is synchronous with the flux
generated by the stator current. An SM rotates with a mechanical rotor speed ωm , which is
determined by the electrical rotor speed ωr [which in the steady state equals the angular frequency
ω1 of the applied stator voltage, cf. Figure 5.1(a)] and the number of pole pairs np

ω r = np ω m . (5.1)

A four-pole SM (two pole pairs) rotates with half the speed of a two-pole machine (one pole pair),
assuming that both have the same stator frequency. A two-pole machine connected to a 50-Hz
grid rotates with 3000 rpm, a four-pole machine rotates with 1500 rpm, and so on. Figure 5.4
illustrates the difference between two-pole and four-pole machines.

Operating Principle
Consider the two-pole SM in Figure 5.3. A synchronously rotating dq coordinate system, aligned
with the rotor flux, is introduced. This coordinate system is displaced from the stationary αβ
coordinate system by the flux angle θ. The d axis is the magnetic axis. When current is applied
in the stator (in Figure 5.3 only two conductors are shown, for clarity), forces F perpendicular to
the d axis appear in the stator. But since the stator is fixed, counterforces appear which instead
drive the rotor counterclockwise (opposite the direction of the stator forces).
Clearly, for the forces (i.e., the torque) to be constant, the stator current must be ac with
the same angular frequency ω1 as the rotation of the rotor: ω1 = θ̇. Furthermore, in order to
138 Chapter 5. AC Motor Drives

N S N

S N S

2 poles 4 poles
( np = 1) (n p = 2 )

Figure 5.4. Different numbers of poles.

obtain maximum production of torque, the stator current has to be properly vectorially placed
in relation to the rotor flux. The flux angle θ therefore has to be known with good accuracy by
the control system. Since the flux angle coincides with the physical rotor position—cf. Figure
5.1(a)—measuring this (using a digital position encoder or a resolver, see Chapter 8) directly gives
the flux position. This makes control of an SM fairly straightforward, as long as a position sensor
is used. If not—a case which is called sensorless control—the flux angle has to be estimated.

Locations of the d and q Current Components

Suppose that it is desired to generate the flux shown in Figure 5.3 from the stator rather than
from within the rotor. (This is done in an induction machine, see below.) Current then has to be
applied as shown in Figure 5.5. Although the current is geometrically applied in the q direction,
we define this as a current applied in the d direction, since then a d-direction flux is obtained.

The dq current coordinate system is determined by the corresponding flux co-

ordinate system.

The current applied in Figure 5.3 is therefore a full q-direction current.

q d

Figure 5.5. Flux-producing stator current component: by definition in the d direction.

5.1. AC Machine Fundamentals 139

Round and Salient Rotors

Rotors for SMs are either round (also called cylindrical or nonsalient) or salient (with salient poles),
as shown in Figure 5.6. Round-rotor SMs are used in applications with higher speeds, for example
generators driven by steam turbines in fossil-fuel or nuclear generating stations. Such machines
have low pole number, typically between four and eight. Salient SMs are designed for slower
rotation, and have higher pole number, up to 120. Typical applications of such are generators
driven by hydro turbines in hydroelectric generating stations and motors for rolling mills. Note
in Figure 5.6 that the field winding in a salient SM is placed around the poles, not in the rotor
surface as in a round-rotor SM.
A round-rotor SM has a smooth airgap, i.e., the mutual inductance between the stator and the
rotor is constant all along the rotor periphery. In a salient SM the airgap width is minimal at the
poles (N and S in Figure 5.6) and maximal in between the poles. Hence, the mutual inductance—
which decreases with the airgap width—is maximal at the poles and minimal in between. In
salient machines, nonlinear effects tend to be more pronounced than in round-rotor machines,
giving a higher amount of harmonics—so-called space harmonics—in the airgap flux, see Remark
5.2 below. Dynamic models which include these effects are quite complicated. Therefore, it is
normally assumed that the mutual inductance Lm varies sinusoidally with the rotor position:

1
Lm (θ) = (L0 + Lg cos 2θ) where L0 = Lmd + Lmq and Lg = Lmd − Lmq . (5.2)
2
On the d axis, Lm (0) = Lm (180◦ ) = Lmd , whereas on the q axis, Lm (±90◦ ) = Lmq , where
Lmd > Lmq . This approximation gives a model in which the space harmonics are disregarded, but
also one that is only slightly more complicated than that of a round-rotor machine. See further
Section 5.3.

N S N S

S N S N
(a) (b)

Figure 5.6. Four-pole SM rotors showing flux paths as well as field- and damper-winding placements. (a)
Round rotor. (b) Salient rotor.

Damper Winding
The SM rotor in Figure 5.6(b) is outfitted with a damper winding. This is a so-called cage winding,
consisting of bars placed in the surface of the poles, which are interconnected at each axial end
of the rotor by a short-circuit ring, as shown in the figure. The bars are slots that are machined
in the rotor and filled with aluminum, copper, or bronze. The rotor itself is made of laminated
steel. Alternatively, the poles or the entire rotor core can be made as a solid steel forging. This
140 Chapter 5. AC Motor Drives

is feasible since the SM rotates synchronously in the steady state; the core losses will be low for
normal operation. At transients, the core losses give an effect similar to that of a damper winding.
Most electrically excited SMs have a damper winding, for two reasons. In a synchronous motor,
the damper winding allows starting by direct line connection (in a similar fashion as an induction
motor, as will be seen). Called a line start, the field winding is short circuited until the speed
reaches about 95% of synchronous speed, when it is excited. In a synchronous generator the
damper winding reduces angular oscillations at transients, which is important from the standpoint
of power system stability.

Permanent-Magnet Synchronous Machines

The alternative to supplying the fundamental magnetization from an external dc source is to use
permanent magnets. These are made of high-remanent magnetic materials, such as SmCo or
NdBFe, and can either be mounted on the rotor surface, set in openings in the rotor surface, or be
buried (mounted inside the rotor), as shown in Figure 5.7. The permeability of magnet material
is about the same as for air. Therefore, the rotor of a PMSM with inset or buried magnets is
salient: the flux “sees a wider airgap” in the d direction than in the q direction. But, unlike the
salient rotor in Figure 5.6(b), the mutual inductance between the stator and the rotor in Figure
5.7(b, c) is minimal in the d direction: Lmd < Lmq . The rotor with surface-mounted magnets in
Figure 5.7(a) is, on the other hand, magnetically round. Benefits of PMSMs are that a very high
efficiency is obtained (90%–98%), and that no field current supply is needed. High cost of magnet
materials and constructional difficulties are the main drawbacks. Traditionally, PMSMs have been
used as motors in applications requiring lower power, such as industrial robots and in aerospace.
There are today PMSMs rated several hundred kilowatts on the market, however.
PMSMs are normally converter fed, but in some applications—such as pumps—line-connected
PMSMs are used. Such motors must be equipped with a damper winding to allow asynchronous
line starting. To design a line-connected PMSM is a nontrivial task. On the one hand, the magnets
will give a strong pulsating torque during the asynchronous start, which might cause lockup at
a low speed when the startup load torque is high. On the other hand, for low load torques
synchronization may not be accomplished; the rotor instead oscillates around a speed slightly
lower than synchronous. (As its field winding is short circuited during the asynchronous start,
these problems normally do not occur for an electrically excited SM.)
For more information about analysis and design of PMSMs, see [68].

q q q

d d d

(a) (b) (c)

Figure 5.7. Two-pole PMSM rotors with (a) surface-mounted magnets, (b) inset magnets, and (c) buried
magnets.
5.1. AC Machine Fundamentals 141

5.1.2 Induction Machines

An induction machine is similar to a round-rotor SM with a cage damper winding, but no field
winding or permanent magnets, as shown in Figure 5.1(b). The IM is most common as motor
in industrial applications and electric traction (mainly rail vehicles), but is also used in some
appliances such as washing machines. The IM is robust, inexpensive, and durable, which accounts
for its popularity and wide use. It is also cheaper and simpler to use in both line-connected and
converter-fed drives than electrically excited SMs, as no field winding supply is needed.
On the other hand, the IM has a fairly low starting torque when used as a line-connected
motor (which is a problem also for the SM). When fed from a converter, the control is much more
complicated than for a dc motor. Performance limitations in the performance of power electronic
converters and control equipment prevented the IM from becoming widely used in variable-speed
drives up until the mid-1980s, but is today commonplace.
IMs are more efficient than dc machines, but less efficient than SMs. Typical efficiencies are
from 85% to 90%.
Although most IMs have a cage rotor, some are instead equipped with a three-phase rotor
winding, connected to three slip rings (which are similar to the slips rings that supply the field
winding in an SM). One application for the slip-ringed IM is as wind-turbine generator, where the
stator is connected directly to the mains, whereas the rotor is connected to the mains via a power
electronic converter (normally a back-to-back VSC configuration). This is called a doubly-fed IM.
The converter allows full control, whereas most of the power goes directly to the grid via the stator,
giving a system with higher efficiency as compared to a standard converter-connected IM with a
cage rotor. Slip-ringed IMs will not be treated further in this book, however.
For more information of the physics and design principles of IMs, [68] is recommended.

Operating Principle
Unlike the SM, the IM is magnetized from the stator. As long the IM rotates synchronously, no
current is induced in the rotor. Once the IM is loaded mechanically, the rotor starts lagging behind
the synchronous rotation of the flux. Current is then induced in the rotor winding—according to
Lenz’ law—giving forces that counteract the lagging effect. This gives the electrical torque. The
slip s is defined as2
ω1 − ωr
s= (5.3)
ω1
and
ω2 = sω1 = ω1 − ωr (5.4)

is called the angular slip frequency. At normal operation the slip is fairly small, only up to about
5%.
As discussed above, measuring the physical position of the rotor of the SM immediately gives
the flux angle, since θ = θr , cf. Figure 5.1(a). This is not the case for the IM due to the slip.
Flux-angle measurement of the IM requires sensors such as search coils or Hall elements placed in
the airgap. These are expensive and delicate, and should therefore be avoided. Hence, the rotor
flux of the IM is not readily measurable; it has to be estimated. This is greatly complicates the
control of the IM.
2 We in this chapter use the derivative operator p also to denote the Laplace transform variable, to avoid risk for

confusion with the slip.

142 Chapter 5. AC Motor Drives

Constructional Principle
Figure 5.8 shows the constructional principle of the IM. The rotor cage is similar to the damper
winding of an SM, except that the rotor slots normally are skewed. The method reduces nonlinear
effects, and in turn the harmonic content of the flux, with less torque pulsations and lower acoustic
noise as result. Due to the asynchronous rotation, the rotor must be made of laminated steel in
order to keep the core losses low.
Since the magnetization is done from the stator, it is more important than for SMs to make the
airgap narrow. A wide airgap translates to a small mutual inductance, demanding a large stator
magnetizing current in order to produce the desired rotor flux. This causes higher losses in the
stator winding, giving lower efficiency. IMs rated 100 kW or below have airgaps less than 1 mm
wide.
We finally note that also induction machines can be constructed with a number of pole pairs
higher than 1. In this case, there is no difference of the rotor design principle as compared to
np = 1, but rather in the way the stator is wound.

stator stator
stator slots
shaft

airgap

rotor slots rotor short−circuit

ring
(a) (b)
.

Figure 5.8. Constructional principle of an induction machine. (a) Stator–rotor cross-sectional view. (b)
View from the side.

5.2 Dynamic Model for the Induction Motor

Let us first consider the stator circuit. The resistance Rs of the stator winding is (for all practical
purposes) equal in all three phases. From the law of induction it follows that the part of the stator
voltage which is not dissipated in the stator resistance will build up a flux in the stator winding.
Hence, with vss as the stator-voltage space vector, the following relation must hold:
dψss
vss − Rs iss − =0 (5.5)
dt
where iss and ψss are the space vectors for the stator current and the stator flux linkage, respectively.
The rotor circuit, with winding resistance Rr , can be considered in a similar way. Suppose that the
rotor is observed from a coordinate system—rotor coordinates—which rotates with the same speed
as the rotor: ωr . Let us denote rotor coordinates with the superscript r. As the coordinate system
is rotor fixed, there will be no induced voltage due to the rotation, so the rotor-flux dynamics as
expressed in rotor coordinates are formally identical to the stator-flux dynamics (5.5)
dψrr
vrr − Rr irr − =0 (5.6)
dt
5.2. Dynamic Model for the Induction Motor 143

where vrr , irr , and ψrr are the rotor-voltage, rotor-current, and rotor-flux space vectors, respectively.
r
But the rotor winding is short-circuited, so vrr = 0. Now, let us transform r
Q ir and ψr to stationary
coordinates. This is an αβ transformation using the rotor position θr = ωr dt

isr = ejθr irr ψsr = ejθr ψrr . (5.7)

Equation (5.6) is transformed as

d(e−jθr ψsr )
0 − Rr e−jθr isr − = 0
, dt -
dψs
⇒ −Rr e−jθr isr − −jωr e−jθr ψsr + e−jθr r = 0
dt
dψsr
⇒ jωr ψsr − Rr isr − = 0. (5.8)
dt
The induction motor is thus described by the following equations:
dψss
= vss − Rs iss (stator) (5.9)
dt
dψsr
= jωr ψsr − Rr isr (rotor). (5.10)
dt
Let us now find a relation between the stator and rotor flux linkages. The rotor winding is referred
to the stator, i.e., the rotor winding is represented by coils in the α and β directions, cf. Figure
5.1(b). Assuming linear magnetic conditions, the airgap flux ψsa can then be expressed as

ψsa = Lm ism , ism = iss + isr (5.11)

where Lm is the mutual inductance between the stator and the rotor, which is also called the
magnetizing inductance, and ism is the magnetizing current. The stator flux is the sum of the airgap
flux and the stator leakage flux, the latter which under linear magnetic conditions is proportional
to the stator current only. Similar reasoning for the rotor flux yields
ψss = Lm ism + Lsl iss (5.12)
ψsr = Lm ism + Lrl isr (5.13)
where Lsl and Lrl are the stator and rotor leakage inductances, respectively. The leakage induc-
tances are typically 10% of Lm or less. Alternatively, with Ls = Lm + Lsl and Lr = Lm + Lrl as
the stator and rotor self inductances, respectively, the relations can be expressed as
ψss = Ls iss + Lm isr (5.14)
ψsr = Lm iss + Lr isr . (5.15)
Combining (5.12)–(5.13) with (5.9)–(5.10), assuming constant inductances, yields
diss dis
vss − Rs iss − Lsl − Lm m = 0 (5.16)
dt dt
s s disr dism
jωr ψr − Rr ir − Lrl − Lm = 0. (5.17)
dt dt
These equations describe the dynamic equivalent circuit depicted in Figure 5.9. As there are three
inductances configurated in a “T,” this is known as the T-equivalent circuit.
144 Chapter 5. AC Motor Drives

Rs L sl L rl
i ss i sr

i sm
Rr
v ss Lm
j ωrψrs ωr

Figure 5.9. Dynamic T-equivalent circuit for the induction motor.

This is the traditional equivalent circuit for the IM commonly found in literature on electrical
machines, with two exceptions. First, rather than a resistance Rr /s in the rotor circuit, we find
the rotor resistance Rr in series with a voltage source. This voltage source—to be called the
rotor emf—corresponds to the back emf of the dc motor, and is the device that converts between
electrical and mechanical power. The power developed in the rotor emf goes to the shaft. Second,
electric power engineers often tend to become so fascinated with the equivalent circuit that they
seem to forget completely that an electrical motor is also mechanical. In an attempt to prevent
this, we shall draw the equivalent circuit with a rotating shaft sticking out of the rotor-emf symbol.
As the name suggests, the dynamic equivalent circuit is valid also at transient conditions. If
merely steady-state conditions are considered, the equivalent circuit can be simplified. In the
steady state, i.e., with p = jω1 , the voltage across Lm and Lrl is given by

jω1 (Lm ism + Lrl isr ) = jω1 ψsr . (5.18)

But this is also the voltage across Rr and the rotor emf, so

jω1 ψsr = jωr ψsr − Rr isr ⇒ jω2 ψsr = −Rr isr

ω1 Rr
⇒ jω1 ψsr = − Rr isr = − isr . (5.19)
ω2 s
This equation corresponds to the standard equivalent circuit with a total resistance Rr /s on the
rotor side. Of this, Rr is the actual rotor resistance, whereas the remaining part, (1 − s)Rr /s,
corresponds to the rotor emf. The rotor emf of the IM happens to be proportional to and in phase
with the rotor current in the steady state. Hence, it can be modeled as a resistance. Note that
the equivalent circuit in Figure 5.10 is valid only in the steady state, so it cannot be used to draw
any conclusions regarding the behavior of the induction motor in a variable-speed drive.

Remark 5.1 Stator/rotor turns ratio. In the model derived, we have not taken into account
that the stator and rotor windings have different numbers of turns per phase Ns and Nr , where
Ns > Nr normally (especially if the rotor winding is a short-circuited cage). The rotor current isr
is therefore not the actual measurable rotor current, which we may denote as isr ′′ , but the rotor
current referred to the stator. From the relation between the primary and secondary currents in a
transformer, we have
Ns s
isr ′′ = i .
Nr r
5.2. Dynamic Model for the Induction Motor 145

Rs L sl L rl
i ss Rr

i sm

v ss 1 sR ωr
Lm s r

i sr

Figure 5.10. Steady-state T-equivalent circuit for the induction motor.

Similarly, Rr is the rotor resistance reflected to the stator. The physical rotor resistance Rr′′ is
given by
, -2
′′ Nr
Rr = Rr
Ns

and is thus normally much smaller than Rr (and Rs ).

Remark 5.2 Sinusoidally distributed winding. Throughout this chapter it is implicitly as-
sumed that the airgap flux is sinusoidally distributed. In the theory for ac machines it is seen that
this requires a so-called sinusoidally distributed winding in the stator. In practice, this is not the
case. The airgap flux will deviate somewhat from the ideal sinusoid; it is said to contain space
harmonics, which are neglected in our models. The number of turns per phase Ns in the stator is
not the actual number of turns, but the equivalent number of turns which would produce the ideal
airgap flux by a sinusoidally distributed winding [68].

5.2.1 Inverse-Γ Model

Although the T-equivalent dynamic model is physically relevant, it is less good for dynamic analysis
and controller design due to the fact that it is over parametrized. The three inductor currents are
not linearly independent, as ism = iss + isr . One leakage inductance—not two—is sufficient. This
simplification is achieved by defining new rotor variables as

isr
ψsR = bψsr isR = (5.20)
b

where b is a transformation factor to be chosen. (All transformed quantities have upper-case

subscripts.) Substitution of these definitions in the flux equations (5.14)–(5.15) yields

ψss = Ls iss + bLm isR (5.21)

ψsR = bLm iss + b2 Lr isR . (5.22)

By choosing b such that the stator and rotor currents have equal coefficients in (5.22), i.e., bLm =
b2 Lr ⇒ b = Lm /Lr , the leakage inductance on the rotor side is eliminated; the leakage is referred to
the stator side, and the inverse-Γ model is obtained. When using this definition, the flux equations
146 Chapter 5. AC Motor Drives

(5.21)–(5.22) are transformed to

L2m s
ψss = Ls iss + i (5.23)
Lr R
L2m s
ψsR = (i + isR ) (5.24)
Lr s
which, with isM = iss + isR as the magnetizing current of the inverse-Γ model, also can be expressed
as

ψss = Lσ iss + LM isM (5.25)

ψsR = LM isM (5.26)

where new parameters, LM and Lσ , have been introduced. These are the transformed magnetizing
inductance and total leakage inductance, respectively
L2m
LM = (5.27)
Lr
L2m Lm (Lsl + Lrl ) + Lsl Lrl
Lσ = Ls − LM = Ls − = ≈ Lsl + Lrl . (5.28)
Lr Lm + Lrl
The stator equation (5.9) remains unchanged, and the transformed rotor equation is formally equal
to (5.10)
dψss
= vss − Rs iss (stator) (5.29)
dt
dψsR
= jωr ψsR − RR isR (rotor) (5.30)
dt
where the transformed rotor resistance
, -2
Lm
RR = Rr (5.31)
Lr

has been introduced. Combining (5.29)–(5.30) with (5.25)–(5.26), assuming constant inductances,
yields
diss dis
vss − Rs iss − Lσ − LM M = 0 (5.32)
dt dt
s
s di
jωr ψsR − RR iR − LM M = 0. (5.33)
dt
These equations form the dynamic equivalent circuit depicted in Figure 5.11. They define the
motor model in terms of stator and rotor currents. Many drives incorporate a control loop for the
stator current, see further Chapter 6. Therefore, it is natural to take iss as a state variable. The
rotor current, however, usually cannot be measured. It is therefore convenient to eliminate the
rotor current from the equations and use the rotor flux linkage ψsR as the second state variable.
Substitution of (5.32)–(5.33) together with

ψsR
isR = isM − iss = − iss (5.34)
LM
5.2. Dynamic Model for the Induction Motor 147

in (5.29)–(5.30) yields

diss dψsR
Lσ = vss − Rs iss − (5.35)
dt dt 1
. /0
Es
, -
dψsR RR
= RR iss − − jωr ψsR (5.36)
dt LM

where we introduce the symbol Es for the time derivative of the rotor flux. Es —to be called the
flux emf—can be eliminated from (5.35) by using (5.36), giving
, -
diss RR
Lσ = vss − (Rs + RR )iss − jωr − ψsR (5.37)
dt LM
. /0 1
Esb

where Esb is the back emf. Except at low speeds, the back emf is predominantly determined by the
term jωr ψsR . It is thus approximately equal to the rotor emf, and differs from the flux emf only
by the term RR iss , which is small at higher speeds
, - , -
RR RR dψsR
Esb = jωr − ψsR ≈ jωr ψsR ≈ RR iss − − jωr ψsR = = Es (5.38)
LM LM dt

or in words
back emf ≈ rotor emf ≈ flux emf.

We finally note that just as for the T model, the rotor emf is proportional to and in phase with
the rotor current in the steady state, giving the equivalent circuit depicted in Figure 5.12.

Rs Lσ
i ss i sR
s
iM
RR
v ss LM
jωrψRs ωr

Figure 5.11. Dynamic inverse-Γ-equivalent circuit for the induction motor.

PROBLEM 5.1
It is also possible to eliminate the leakage inductance on the stator side, referring the leakage
to the rotor side instead, giving the so-called Γ model. Find the correspondences of (5.35)
and (5.36) for the Γ model! Hint: Now the stator and rotor current should have equal
coefficients in (5.21).
148 Chapter 5. AC Motor Drives

Rs Lσ RR
i ss
s
iM

v ss LM 1 sR ωr
s R

i sR

Figure 5.12. Steady-state inverse-Γ-equivalent circuit for the induction motor.

5.2.2 Transformation to Synchronous Coordinates

It is often convenient to express ac motor models in synchronous coordinates. We can use (2.151),
i.e., p → p + jω1 , to immediately obtain the differential equations in synchronous coordinates
corresponding to (5.35)–(5.36)
dis dψR
Lσ = vs − (Rs + jω1 Lσ )is − jω1 ψR − (5.39)
dt , - dt
dψR RR
= RR is − + jω2 ψR (5.40)
dt LM
or, with (5.40) substituted in (5.39)
, -
dis RR
Lσ = vs − (Rs + RR + jω1 Lσ )is − jωr − ψR (5.41)
dt LM
. /0 1
back emf Eb
, -
dψR RR
= RR is − + jω2 ψR . (5.42)
dt LM
Equations (5.41)–(5.42) can be illustrated by the block diagram in Figure 5.13. There are both
similarities and differences to the block diagram of the dc motor, Figure 4.9. In both block diagrams
there are three feedback loops, where the outer loop is that of the back emf. The flux dynamics
act similarly to the mechanical dynamics of the dc motor. The mechanical dynamics of the IM
(which for simplicity are represented only as a box with current and flux input; see the sequel for
details) are connected in parallel to the flux dynamics, and interact nonlinearly with the electrical
subsystem. A nonlinear fifth-order system with the state variables id , iq , ψd , ψq , and ωr is formed.

Even though the flux dynamics are slow (sometimes comparable to the mechanical dynamics),
the electrical dynamics can often yet be considered linear, but with varying coefficients: jωr
multiplied with ψR . However, as we shall see in Chapter 9, many IM control methods indirectly
involve a nonlinear feedback of the flux to the angular stator frequency ω1 . The flux dynamics then
become highly nonlinear, and the system may turn unstable under some circumstances. Careful
analysis is needed in the design of the control system.

PROBLEM 5.2
In Figure 5.13 it is seen that the IM has three inputs in synchronous coordinates: vd , vq
5.2. Dynamic Model for the Induction Motor 149

Current dynamics Flux dynamics

vs 1 is 1 ΨR
RR p
pL σ

Rs RR jω1Lσ RR
jω2
LM
RR
jωr
"Back−emf feedback" LM ωr
ω1 ωr Mechanical
dynamics

Figure 5.13. Block diagram of the induction motor dynamics. Solid and dashed lines indicate space-vector
and scalar signal paths, respectively.

(components of vs ), and ω1 . In stator coordinates, only two inputs are available: vα and vβ
(components of vss ). This appears to be a discrepancy, but is it? Please clarify!

5.2.3 Perfect Field Orientation

Quite similarly to the orientation of the dq coordinate system along the rotor flux of the SM—cf.
Figures 5.1 and 5.3—it is particularly useful to select the dq transformation angle equal to the flux
angle:
θ1 = arg ψsR . (5.43)
The dq reference frame will then be aligned with the rotor flux also for the IM, so that ψR becomes
real: ψR = ψR . We shall call this perfect field orientation. With is = id + jiq , (5.40) then reduces
to
, -
dψR RR
= RR (id + jiq ) − + jω2 ψR . (5.44)
dt LM
Identifying the real and imaginary parts and solving for ω2 in the latter yields
dψR RR
= RR id − ψR (5.45)
dt LM
RR iq
ω2 = ω1 − ωr = . (5.46)
ψR
Under perfect field orientation, the flux modulus has first-order dynamics, with input id and time
constant Tr = LM /RR , called the rotor time constant. In the steady state,
ψR = LM id . (5.47)
Thus, id controls the flux level, and plays the part of the field current in a dc motor. Therefore, id
is called the flux-producing current component, and should be controlled such that
ψref
id = (5.48)
!M
L
150 Chapter 5. AC Motor Drives

! M is an estimate of the magnetizing

where ψref is the flux reference, i.e., the desired flux level, and L
inductance.
Furthermore, we have found a simple relation, (5.46), between the angular slip frequency, the
flux modulus, and the current component iq . This is known as the slip relation.
If perfect field orientation were possible to achieve, control of the IM would only be slightly
more difficult than of the dc motor. The problem is that the rotor flux is not readily available for
measurement, but has to be estimated. See Chapter 9.

5.2.4 Torque Production

Active power is developed at three3 locations in the equivalent circuit (Figure 5.11): in the stator
and rotor resistances and in the rotor emf. The first two give losses—so-called copper losses—
whereas the latter is converted to mechanical power. Following (2.132), we can calculate this as
(where the subscript e stands for electric power to the shaft)

3
Pe = − Re {jωr ψR i∗R } (5.49)
2K 2
the minus sign as iR is defined with a direction opposite to the rotor emf. But for a complex
number z = x + jy we have Re{jz} = Re{j(x + jy)} = −y = − Im{z}, so
* , -∗ +
3ωr ∗ 3ωr ψR 3ωr
Pe = Im {ψ i
R R } = Im ψR − is =− Im {ψR i∗s }
2K 2 2K 2 LM 2K 2
3ωr
= Im {ψ∗R is } . (5.50)
2K 2
Because power = torque × angular speed and ωr = np ωm , we obtain the following relation for the
electrical torque:

Pe np Pe 3np
τe = = = Im {ψ∗R is } . (5.51)
ωm ωr 2K 2

In component form with ψR = ψd + jψq , we have

3np
τe = (ψd iq − ψq id ). (5.52)
2K 2
It is interesting to note that the relation holds also for the stator flux. We have

3np 3np
τe = 2
Im {ψ∗R is } = {ψR = ψs − Lσ is } = Im {ψ∗s is } . (5.53)
2K 2K 2
Because the power formula is invariant of the coordinate system used, the torque can also be
expressed as

3np 3np
τe = Im {(ψsR )∗ iss } = Im {(ψss )∗ iss } . (5.54)
2K 2 2K 2
3 Neglected here are the core losses: magnetic hysteresis and eddy-current losses, which occur mainly in the rotor.

These can be modeled as a resistance in parallel with the magnetizing inductance in the equivalent circuit.
5.2. Dynamic Model for the Induction Motor 151

Perfect Field Orientation

For the special—but important—case of perfect field orientation, ψR = ψd = ψR , the torque is
given by
3np 3np
τe = 2
Im{ψR (id + jiq )} = ψR iq . (5.55)
2K 2K 2
The q-direction stator current component then controls the production of torque, just like the ar-
mature current of a dc motor. For this reason, iq is called the torque-producing current component.
Note in the slip relation (5.46) that ω2 = 0 ⇔ ω1 = ωr when iq = 0; the IM then rotates syn-
chronously. This is a natural consequence of field orientation: since there is no torque production
when iq = 0, there is no induced rotor current either. The IM must then rotate synchronously!

Impact of Harmonics
Let us briefly discuss to the case of multifrequency waveforms. In Paragraph 2.6.9 we considered
the power produced by a perfectly sinusoidal voltage and a multifrequency current. In (5.54) it
is the current iss that is (almost) perfectly sinusoidal. This is particularly true when closed-loop
current control is used, see Chapter 6. The flux, on the other hand, will contain harmonics due to
nonlinearities; the amount depending on motor design features. Hence, the conclusion of Paragraph
2.6.9, that the pulsations will contain the angular frequencies 6nω1 , n = 1, 2, . . ., clearly holds for
an IM.

5.2.5 Steady-State Relations

In the steady state, (5.39) can be expressed as

vs = (Rs + jω1 Lσ )is + jω1 ψR . (5.56)

It is of particular interest to consider higher speeds—which allows the term Rs is to be neglected—

and perfect orientation (which is a reasonable assumption at higher speeds, see Chapter 9). In
component form we then have

vd = −ω1 Lσ iq (5.57)
vq = ω1 (Lσ id + ψR ) = {(5.47)} = ω1 (Lσ + LM )id . (5.58)

Thus, at low loads when iq is small, vq dominates the stator voltage. For a constant id , it is seen
the stator voltage must increase with the stator frequency (i.e., with the rotor speed).

5.2.6 Field Weakening

The modulus of the stator voltage vector is limited to Vbase . When
S , -2
R Lσ + LM
2 2
|vs | = vd + vq = {(5.57), (5.58)} = |ω1 |Lσ iq +2 i2d = Vbase
Lσ

then |ω1 | can increase further only if id is decreased. As for the dc motor, this is called field
weakening. This way, the IM can normally be operated at speeds up to four or five times base
speed. Field-weakening operation of the IM is studied in Section 9.3.
152 Chapter 5. AC Motor Drives

5.2.7 Motor Parameters

Typical per-unit values for induction motors rated less than 100 kW are

Magnetizing inductance LM : 1.5 − 3

Leakage inductance Lσ : 0.15 − 0.3
Resistance Rs , RR : 0.01 − 0.1.

A motor of low power rating (say, 1 kW) tends to have a smaller per-unit magnetizing inductance
and a larger per-unit stator resistance than a motor of higher rating. Yet, the ranges are quite
narrow, so it is fairly obvious that similar or even identical control algorithms can be applied to
any induction motor, regardless of rating. The theory and methods for PWM, vector control, and
current control discussed in the preceding and following chapters are therefore generally applicable.
It is not unusual that a control system designed for a 1-kW drive can be modified for a 100-kW
drive simply by altering the controller parameters and using a more powerful converter, often with
a lower switching frequency. This is a privilege of the ac drives control engineer which is seldom
enjoyed by a control engineer in another field of application. Imagine trying to apply a control
system designed for a single-engine propeller aircraft to a supersonic fighter!

5.2.8 Nominal Flux and Torque

Suppose that the IM is operated at nominal conditions, i.e., |vs | = Vbase , |is | = Ibase , and ω1 =
ωbase . The steady-state voltage–current relations (5.57)–(5.58) can then be expressed as
vd = −ωbase Lσ iq (5.59)
vq = ωbase (Lσ + LM )id . (5.60)
Because vd2 + vq2 = Vbase
2
, we obtain
, -2
Vbase v 2 + vq2
= d2 = {(5.59), (5.60)} = (Lσ iq )2 + (Lσ + LM )2 i2d
ωbase ωbase
= L2σ (i2d + i2q ) +LM (LM + 2Lσ )i2d . (5.61)
. /0 1
2
Ibase
Solving for id gives the nominal flux-producing current component and, in turn, the flux level as
follows:
T2 32
U
U Vbase − (L I 2
V ω σ base )
iN
base
d = (5.62)
LM (LM + 2Lσ )
T 2 32 T 2 32
U U
U 1 − ωbase Lσ Ibase U 1 − ωbase Lσ
N V base V V Vbase V Z
= LM iN
base base
ψR d = = . (5.63)
ωbase 1 + 2Lσ /LM ωbase 1 + 2Lσ /LM
But Vbase /ωbase and Zbase /ωbase are the base values for flux and inductance, respectively, so the
nominal flux can also be expressed in normalized form as
S
N 1 − L2σn
ψRn = . (5.64)
1 + 2Lσ /LM
5.2. Dynamic Model for the Induction Motor 153

For the typical per-unit values Lσ = 0.2 and LM = 2 we have ψR = 0.89 pu. Unlike a dc motor
(and, as we shall see, an SM) nominal rotor flux of the IM is not 1 pu!

PROBLEM 5.3
What is the nominal stator flux?

Furthermore, the nominal torque is given by

3np N N 3np Vbase Ibase N N

τeN = ψR iq = ψ iqn (5.65)
2K 2 2K 2 ωbase . Rn
. /0 1 /0
κ
1
τbase

where the factor κ < 1 and τbase = np Pbase /ωbase (note that this is different from the base
torque used for the dc motor in Table 4.2). R For the same typical per-unit values Lσ = 0.2 and
N
"
LM = 2 as previously, we obtain iq = 2 N
Ibase − (id ) =2 1 − (0.89/2)2 = 0.90 pu, giving
κ = 0.89 · 0.90 = 0.80. A closed-form expression for κ can be derived:
R
ωbase N N ωbase N 2
κ = ψR iq = ψR Ibase − (iN
d )
2
Vbase Ibase Vbase Ibase
T ⎡ ⎤
U 2 32
U Vbase 2
U
1 U 1 − Lσn ⎢ 22 ωbase − (L σ Ibase ) ⎥
= ⎣Ibase − ⎦
Ibase V 1 + 2Lσ /LM LM (LM + 2Lσ )
T W, I
U -2
1 U Lσ 1
= V 2
(1 − Lσn ) 1 + − 2 . (5.66)
1 + 2Lσ /LM LM LMn

Note that when the base torque is selected as in (5.65), then τen = κ < 1 pu for nominal operation
of the IM.

5.2.9 Steady-State Relations for Line-Connected IMs

Although we here focus on variable-speed drives, most induction motors are in fact operated at
constant or near-constant speeds in line-connected applications. Because the magnetization of the
IM is made from the stator terminals, the IM can be started by direct line connection. Before
going into details about dynamics and control of the IM, it is instructive first to have a brief look
at some steady-state theory for line-connected IMs. We do so using synchronous coordinates.
Obviously, for a line-connected motor the stator frequency is constant, ω1 = ωbase = 1 pu, so the
voltage drop across the stator resistance can be neglected. So may also the magnetizing current, if
Lσ ≪ LM , when we are concerned with the torque-producing rotor current only. Consider Figure
5.12; the impedance seen from the stator is then

RR
Z(jω1 ) = jω1 Lσ + (5.67)
s
so the rotor current is given by
vs vs
iR = − =− . (5.68)
Z(jω1 ) jω1 Lσ + RR /s
154 Chapter 5. AC Motor Drives

The power to the shaft is the power developed in the “resistance” corresponding to the rotor emf

1−s 1 − ω1 −ωr ωr
RR = ω1 −ωωr1 RR = RR
s ω1
ω 2

3 ωr 2
giving Pe = 2K 2 ω2 RR |iR | , and the following expression for the electrical torque:

np Pe 3np RR 3np RR |vs |2

τe = = 2
|iR |2 =
ωr 2K ω2 2K ω2 (ω1 Lσ ) + (RR /s)2
2 2

3np s |vs |2
= (5.69)
2K ω1 RR (sXσ /RR )2 + 1
2

where Xσ = ω1 Lσ .

Maximum Torque
Maximum torque τemax is also called pull-out torque (or breakdown torque), as the IM is pulled out
of normal operation (and stops) if the load torque exceeds τemax . Equation (5.69) has its maximum
for the pull-out slip s = sp = RR /Xσ ; typical values of sp are between 10% and 20%. This is much
larger than the nominal slip, which is only a couple of percents. We obtain

3np |vs |2
τemax = . (5.70)
2K 2 2ω1 Xσ

With (5.70), we can also express (5.69) as

s 2τemax 2sp s max

τe = = 2 τ . (5.71)
2
sp (s/sp ) + 1 s + s2p e

Low-to-Nominal Torque
At normal operating conditions, i.e, low-to-nominal torque, s2 ≪ s2p , so (5.69) can be approximated
as
3np s 2s
τe ≈ 2
|vs |2 = τemax . (5.72)
2K ω1 RR sp

Thus, there is an approximately linear relationship between slip and torque. For a typical ratio
s/sp = 1/5, it is seen that the nominal torque is only 40% of the maximum torque. Thus, the
IM has the capability to deliver torque above the nominal. But, since this requires overcurrent,
|is | > Ibase , higher-than-nominal torque can be allowed only during shorter intervals (seconds, up
to one minute).

Starting Torque
At starting, ωr = 0, so s = (ω1 − 0)/ω1 = 1, giving
X, -2 Y
2
3n p 1 |v s | X σ 3np RR
τestart = 2 3 = ≫1 ≈ |vs |2 = 2sp τemax . (5.73)
2K 2 ω1 RR Xσ 2 RR 2K 2 ω1 Xσ2
RR +1
5.2. Dynamic Model for the Induction Motor 155

We find that the starting torque is typically about 30% of the maximum torque (and lower than
the nominal torque). That the starting torque is fairly low is often a problem. It is interesting to
note that the starting torque is proportional to the rotor resistance, whereas the nominal torque
is inversely proportional to the rotor resistance. One method for increasing the starting torque is
therefore to use a rotor with slip rings (i.e., the rotor winding is not short-circuited) connected to
a set of external resistances, de facto giving a larger RR . As the motor accelerates, the external
resistances are automatically disconnected, and the slip rings are short circuited.
A more elegant solution is to design the rotor slots such that the rotor resistance varies with the
speed. The induction of rotor current is due to the relative movement between the stator current
vector and the rotor, so the frequency of the actual rotor current (in rotor coordinates) is the slip
frequency. At starting, when ωr is small, the slip frequency is 50 Hz (in Europe). At higher speeds,
the slip frequency is normally less than 2 Hz (corresponding to a slip around 4%). If the rotor slots
are designed—using the skin effect—so that the resistance at higher frequencies is larger than the
dc resistance, the desired high starting and low nominal-speed rotor resistance is obtained. Such
a rotor slot design is shown in Figure 5.14. The resulting design is called a double-cage rotor. At
higher frequencies, the current penetration is poor; current is conducted only in the outer cage,
which has a small cross section and high resistance. At low frequencies, the current penetrates
into the inner cage, which has much lower resistance.

(a) (b)

Figure 5.14. Rotor slot design (double-cage rotor) giving (a) concentrated current distribution (high re-
sistance) at higher rotor frequencies, and (b) uniform distribution (low resistance) at low rotor frequencies.

Summary
⎧
⎪ max 3np |vs |2
⎪
⎪ τe = at pull-out, s = sp
⎪
⎨ 2K 2 2ω1 Xσ
τe ≈ 2s max (5.74)
⎪
⎪ τ at normal operation
⎪
⎪ sp e
⎩
2sp τemax at starting
where sp = RR /Xσ .

Stator Current
From the equivalent circuit, we find the following expression for the stator current, again neglecting
the magnetizing current:

vs s Xσ vs
is = = (5.75)
jXσ + RR /s Xσ RR jsXσ /RR + 1
156 Chapter 5. AC Motor Drives

giving

s |vs |
|is | = " . (5.76)
sp Xσ (s/sp )2 + 1

From this relation we obtain the special cases of low-to-nominal torque [(s/sp )2 ≪ 1], maximum
torque (s = sp ), and starting [s = 1, (s/sp )2 ≫ 1] as
⎧
⎪ s |vs |
⎪
⎪ , low-to-nominal torque
⎪
⎪ sp X σ
⎪
⎪
⎨
1 |vs |
|is | = √ , maximum torque (5.77)
⎪
⎪ 2 Xσ
⎪
⎪
⎪
⎪
⎩ |vs | ,
⎪
starting.
Xσ

With s/sp = 1/5 at nominal torque, and Xσ = 0.2 pu, we obtain the per-unit values 1, 3.5, and
5 at nominal torque, maximum torque, and starting, respectively. The large starting current is
typical for line-connected induction motors (and quite similar to the dc motor case, see Figure
4.12). This may be a problem, because the drive may, as a dc motor, trip due to overcurrent. One
method√ for solving the problem is to start the motor Y-connected (thereby reducing the current
by 1/ 3) and switch to a ∆ connection at a sufficiently high speed.

Numerically Calculated Relations

Since the above expressions for torque and current were derived using a simplified model with the
magnetizing current and stator resistance neglected, they are approximations. In Figure 5.15 we
compare the results to those for the full model for the parameters np = 1, Rs = 0.04, RR = 0.02,
Lσ = 0.2, and LM = 2 pu, and with power-invariant vector scaling. Note that the torque is
normalized with respect to the nominal torque τeN = κ = 0.8 pu given by (5.65). As can be seen,
the correspondence is good, except that the simplified model over estimates the maximum torque
and that the true model gives a nonzero stator current also at ωr = 1 pu: the magnetizing current.
We also in Figure 5.15 plot the rotor flux modulus. It is seen that when the torque is nominal,
τe = τeN , then ψR = 0.89 pu, i.e., the nominal value according to (5.64). For small slips (ωr ≈ 1
pu), the flux is still approximately nominal, whereas it is much lower when the slip is large (speed
much lower than 1 pu).

PROBLEM 5.4
A fan is driven by an induction motor which is directly connected in a ∆ to a 400-V (line-to-
3
line), 50-Hz three-phase voltage. The power drawn by the fan is given by Pm = kωm , where
2
k = 0.0024 kgm . Suppose that Pm ≈ 1 kW; what is then likely the number of pole pairs of
the IM?

5.2.10 Transient Impedance

Let us also briefly analyze the transient behavior of the IM. Suppose that a transient voltage—e.g.,
a step—is applied at the stator terminals. To calculate the immediate transient response of the
stator current to the voltage transient, it can safely be assumed that the magnetizing current is
constant, as the magnetizing inductance LM is much larger than the total leakage inductance Lσ .
5.2. Dynamic Model for the Induction Motor 157

τe/τN
e
2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

4
is (pu)

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

1
ψR (pu)

0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ω (pu)
r

Figure 5.15. Numerically calculated torque, current modulus, and flux modulus as functions of the speed
for the full IM model (solid) and the approximate model (dashed).

The entire transient response therefore goes through the rotor circuit. Approximating the rotor
emf as constant during the transient, the transient impedance seen from the stator terminals is

Zsσ (p) = Rσ + pLσ (5.78)

where Rσ = Rs + RR . As the transfer function from voltage to current is the inverse of the
impedance, the Laplace transform of the impulse response is
1
Zsσ (p)−1 = (5.79)
Rσ + pLσ

which has a pole at p = −Rσ /Lσ . The impulse response thus converges with the transient time
constant
Lσ
Tσ = . (5.80)
Rσ
One should keep in mind, though, that a characterization of the transient response using this type
of model holds only for the immediate response. If a voltage step is applied, the magnetizing
current will eventually change, resulting in a slow transient response with the rotor time constant
Tr = LM /RR . The rotor time constant is significantly larger than the transient time constant.
Let us consider again the per-unit values Rs = 0.04, RR = 0.02, Lσ = 0.2, and LM = 2. Then
158 Chapter 5. AC Motor Drives

Tσ = 3.3 pu and Tr = 100 pu, which for a base frequency of 50 Hz translate to 11 ms and 318 ms,
respectively.
Transients are important to consider in variable-speed drives which use closed-loop stator cur-
rent control. As will be seen in Chapter 6, current control is preferably made in synchronous
coordinates, so we should also pay attention to the synchronous-frame transient impedance. Since
dq transformation implies replacing p by p + jω1 , we immediately obtain

Zσ (p) = Rσ + (p + jω1 )Lσ . (5.81)

This time, the inverse transient impedance has a single complex pole at p = −1/Tσ −jω1 . Therefore,
the immediate transient response will be a damped oscillation of angular frequency ω1 .

5.2.11 Starting Dynamics of Line-Connected IMs

In both variable-speed and line-connected drives, it is important to study the transient response
of the motor. The classical book by Kovács [51] is devoted to the topic of transient analysis of ac
motors. Since we already have discussed the steady-state characteristics of line-connected IMs, it
is insightful to make a brief discussion also of the dynamics.

Instantaneous Maximum Starting Current

In (5.77) we calculated the starting current using a steady-state motor model. This is not entirely
correct; as we shall see below, the stator current is pulsating during the startup. The starting
current 5 pu obtained from (5.77) in the example illustrated in Figure 5.15 is the mean value.
The instantaneous maximum starting current is even larger. To calculate this, it can again be
assumed that the magnetizing current is equal to zero: the immediate stator current response to
an application of stator voltage goes through the rotor circuit. The relation between stator voltage
and current is then the transient impedance. Let the applied stator voltage vector be

vss = V ej(ω1 t+ϕ) ⇔ vs = V ejϕ , t ≥ 0. (5.82)

Taking the Laplace transform yields

1 1 V ejϕ
Is (p) = Vs (p) = (5.83)
Zσ (p) (p + jω1 )Lσ + Rσ p
giving
V ejϕ 2 3
is = L−1 {Is (p)} = 1 − e−(jω1 +Rσ /Lσ )t , t ≥ 0 (5.84)
jω1 Lσ + Rσ
and (again with Xσ = ω1 Lσ )
V2 2 3
|is |2 = 1 + e−2Rσ t/Lσ − 2e−Rσ t/Lσ cos ω1 t , t ≥ 0. (5.85)
Xσ2 2
+ Rσ
As Rσ /Xσ < 1 (Rσ /Xσ ≪ 1 for large motors which have small per-unit resistances), the instan-
taneous maximum value of (5.85) occurs approximately when cos ω1 t = −1, i.e., when t = π/ω1 .
Substituting this value in (5.85) yields
V 2 3
|is |max ≈ 1 + e−πRσ /Xσ (5.86)
Xσ
5.2. Dynamic Model for the Induction Motor 159

where Rσ2 has been neglected. With the same parameter values as before, Rσ = 0.04 + 0.02 pu =
0.06 pu and Lσ = 0.2 pu, we have |is |max ≈ 6.9 pu for V = ω1 = 1 pu, to be compared to the
previously obtained 5 pu.

Starting Transient
From the steady-state theory we know to expect a fairly low starting torque. The torque then
increases with the speed, reaches the maximum (pull-out) torque when s = sp , and finally settles
to a value proportional to the slip. Let us simulate a two-pole IM using a steady-state electrical
model and the following mechanical model (cf. Paragraph 4.2.3), with a load torque proportional
to the speed:
dωr
J = τe − bωr (5.87)
dt
where J = 100 pu. The motor parameters are Rs = 0.04, RR = 0.02, Lσ = 0.2, and LM = 2 pu,
giving κ = 0.80, whereas ωbase = 314 rad/s. The load torque is nominal (τeN = κ = 0.8 pu) at base
speed, i.e., b = 0.8 pu.
The simulation depicted in Figure 5.16 (dashed lines) looks nice enough; it shows precisely
what can be expected from the steady-state theory. However, we have not taken into account the
electrical dynamics, so the result is incorrect. This can be seen in the stator current: the maximum
is somewhat lower than 5 pu—a value also found by the steady-state theory—which is less than
the instantaneous maximum of 6.9 pu obtained from (5.86).
A simulation using the true dynamic electrical model of the IM has been overlayed (solid lines)
in Figure 5.16. Now, the instantaneous maximum stator current of the simulation agrees with
the theoretical 6.9 pu. Furthermore, it can be observed that the starting torque is pulsating and
that the mean torque over the startup is somewhat lower than the torque of the steady-state
simulation. Hence, in reality it takes longer time for the speed to reach the steady-state value of
(slightly below) 1 pu than in the incorrect simulation.
The reason for the erroneous result of the simulation using a steady-state electrical model is
that flux dynamics are slow—in the same range as the mechanical dynamics—due to a magnetizing
inductance which is much larger than the leakage inductance. The electrical dynamics therefore
cannot be neglected as seen from the mechanical dynamics. Thus, one should be cautious about
drawing too general conclusions from the steady-state electrical model for the induction motor,
i.e., the standard equivalent circuit of Figure 5.12.
The starting transient of an induction motor is very dependent on the relation between the
inertia J, the load torque, and the damping of the electrical dynamics. Light rotors and mechanical
loads—having small J—of course give shorter speed rise time, but also give a system that is not as
well damped. It is illustrative to present two additional simulations. In Figure 5.17, the inertia
has been lowered from 100 to 30 pu, and the load torque is set to zero. As can be seen, the speed
rise time is now significantly shorter than previously, whereas the dynamics are obviously much less
damped. Note the tendency to oscillations, particularly in the torque, which is not at all captured
in the incorrect simulation.
Lowering the inertia further to 10 pu (which is perhaps unreasonably small for a practical
IM with base frequency 50 Hz) yields the simulation results depicted in Figure 5.18. Now, the
rotor speed and the electrical quantities do not settle to constant values, but keep oscillating with
constant amplitudes. This is a well-known phenomenon which has been acknowledged ever since
the invention of the IM. The phenomenon is known as hunting, and is a result of the nonlinear
characteristics of the IM dynamics. As mentioned in Paragraph 5.2.2, the electrical dynamics can
160 Chapter 5. AC Motor Drives

often be regarded linear, with some coefficients that vary with ωr . This simplification is not valid
when the mechanical dynamics are fast.
With this we conclude the discussion on the open-loop dynamics of the induction motor, and
note that closed-loop control reduces but does not eliminate the risk for oscillations. The flux
dynamics of the IM make this motor type more difficult to analyze and control than other motor
types.

1
ωr (pu)

0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

2
N
τe/τe

−2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

6
is (pu)

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
t (s)

Figure 5.16. Startup simulations of a line-connected IM drive using dynamic (solid) and steady-state
(dashed) electrical models, respectively, for J = 100 pu.

5.2.12 Volts-per-Hertz Control

The traditional method for controlling converter-fed induction motors is to linearly vary both the
stator voltage magnitude and the stator frequency with the speed reference ωref as follows:

ω1 = ωref + ω
!2 (5.88)
⎧
⎨ |ω1 | V
base , |ω1 | ≤ ωbase
|vs | = ωbase (5.89)
⎩
Vbase , |ω1 | > ωbase
5.2. Dynamic Model for the Induction Motor 161

ωr (pu)
0.5

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

2
N
τe/τe

−2
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

6
i (pu)

4
s

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
t (s)

Figure 5.17. Startup simulations of a line-connected IM drive using dynamic (solid) and steady-state
(dashed) electrical models, respectively, with lower inertia (J = 30 pu) and zero load torque.

where ω
!2 is an estimate of the angular slip frequency. (If precise speed control is not necessary, ω
!2
can be set to zero.) This is known as volts-per-hertz control, and is an open-loop control scheme;
measurement of neither current nor speed is needed. It is also a scalar control scheme, as only the
modulus—not the phase—of the stator voltage is controlled using a VSC.
PROBLEM 5.5
How is field weakening obtained in volts-per-hertz control?
Due to its simplicity, volts-per-hertz control is frequently used in practice, but the control scheme
is associated with several drawbacks.
• Because the stator current is not controlled in closed loop, there will be transients similar
to that when starting a line-connected IM. For steps in the speed reference, there will be
transient reductions in the flux and transient overcurrents. These effects can be reduced or
eliminated by low-pass filtering the speed reference before it is used in the control algorithm.
See Figure 5.19.
• These transients are fairly slow, typically with large oscillations in the torque, as seen in
Figure 5.19. Fast and precise torque control is not possible.
• The method cannot be used for synchronous motors, as these require proper vectorial place-
ment of the stator current relative the rotor flux in order to obtain a non-oscillating torque,
as will be seen in the next section.
162 Chapter 5. AC Motor Drives

ωr (pu)
0.5

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

2
N
τe/τe

−2
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

6
i (pu)

4
s

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
t (s)

Figure 5.18. Startup simulations of a line-connected IM drive using dynamic (solid) and steady-state
(dashed) electrical models, respectively, with even lower inertia (J = 10 pu) and zero load torque.

PROBLEM 5.6
A two-pole induction motor with the parameters LM = 200 mH, Lσ = 20 mH, Rs ≈ 0,
and RR = 0.3 Ω is fed from a converter using volts-per-hertz control. Determine the stator
voltage and the stator frequency when the mechanical speed is 1800 rpm and the load torque
is 100 Nm. Also determine the efficiency of the motor at this operating point.

5.3 Dynamic Model for the Permanent-Magnet Synchronous Motor

We begin by studying the model for round-rotor PMSMs. The (more complicated) model for
salient PMSMs then follows in Paragraph 5.3.8. The stator circuit of the PMSM is similar to that
of the IM, so the same equation, (5.5), must hold

dψss
vss − Rs iss − = 0. (5.90)
dt
In this case, the stator flux is given by

ψss = Ls iss + ψsR (5.91)

where Ls is the stator inductance and ψsR is the rotor flux linkage, which is produced by the
permanent magnets. The rotor flux defines the d axis of the rotor, which is displaced by the
5.3. Dynamic Model for the Permanent-Magnet Synchronous Motor 163

ωr (pu)
0.5

0
0 0.5 1 1.5
6

4
i (pu)
s

0
0 0.5 1 1.5
1
ψR (pu)

0.5

0
0 0.5 1 1.5
5
N
τe/τe

−5
0 0.5 1 1.5
t (s)

Figure 5.19. Volts-per-hertz control of an IM drive with J = 100 pu. Steps in the speed reference are
made at t = 0, t = 0.5 s, and t = 1 s. Solid curves show response with low-pass-filtered speed reference,
dashed curves without.

flux angle θ relative the α axis, as shown in Figure 5.3. Hence, the rotor flux vector is given by
ψsR = ψR ejθ . The flux modulus ψR can be assumed constant, so for a constant Ls
dψss dis
= Ls s + jωr ψR ejθ (5.92)
dt dt
where ωr = θ̇. Now, (5.90) can be written as
diss
Ls = vss − Rs iss − jωr ψR ejθ (5.93)
dt . /0 1
Es
where the term Es = jωr ψR ejθ is the back emf (unlike the IM, there is no difference between
the rotor, flux, and back emfs for the PMSM). This equation can be represented by the dynamic
equivalent circuit depicted in Figure 5.20.

5.3.1 Transformation to Synchronous Coordinates

For the PMSM—perhaps even more so than for the IM—it is insightful to transform the equations
to synchronous coordinates. Transforming (5.93) with the angle θ1 we obtain
d(ejθ1 is )
Ls = ejθ1 vs − Rs ejθ1 is − jωr ψR ejθ (5.94)
dt
164 Chapter 5. AC Motor Drives

Rs Ls
i ss

v ss jωrψR e jθ ωr

Figure 5.20. Dynamic equivalent circuit for the round-rotor PMSM.

and, again with ω1 = θ̇1 ,

dis
Ls = vs − (Rs + jω1 Ls )is − jωr ψR ej(θ−θ1 ) (5.95)
dt
= vs − (Rs + jω1 Ls )is − jωr ψR ej θ̃ (5.96)
. /0 1
E
where θ< = θ − θ1 is the error angle between the actual flux angle and the synchronous coordinate
system used in the dq transformation. In this equation we see that in order for the back-emf
term jωr ψR ej θ̃ to be constant, θ< has to be constant. The applied stator voltage must rotate
synchronously with the rotor of the PMSM. When θ1 = θ, i.e., the field orientation is perfect, then
ω1 = ωr and (5.96) reduces to
dis
Ls = vs − (Rs + jωr Ls )is − jωr ψR . (5.97)
dt

5.3.2 Motor Parameters

The stator resistance of a PMSM is in the same range as that of an IM, i.e., 0.01 pu to 0.1 pu.
On the other hand, the stator inductance is typically much smaller for a PMSM than for an IM:
0.2 pu to 0.4 pu for magnetically round rotors with surface-mounted magnets (see Paragraph 5.3.8
for salient PMSMs). This is in the range of the leakage inductance of the IM. The reason is that
the permeability of the magnets is approximately equal to the permeability of the airgap. The
magnets thus affect the magnetic flux path in the same way as widening the airgap by the magnet
thickness. Also, the airgap of a PMSM is usually made wider than that of an IM of similar rating,
as previously mentioned.

5.3.3 Torque Production and Control Principle

The torque equation (5.53) applies also to the PMSM
3np
τe = Im{ψ∗R is } (5.98)
2K 2

and with ψR = ψR ej θ̃ we obtain

3np L2 3∗ M 3np 2 3
τe = Im ψ R e j θ̃
i s = ψR i q cos < − id sin θ< .
θ (5.99)
2K 2 2K 2
5.3. Dynamic Model for the Permanent-Magnet Synchronous Motor 165

Hence, if θ< is not constant, the torque of the PMSM will pulsate. If a PMSM is line connected,
then—unlike an IM—it generally will not start as the torque will be pulsating. We have thus shown
that, in order to control a PMSM with high performance, the flux angle θ must be known. As θ
is also the physical rotor position, a position sensor—a digital position encoder or a resolver—has
to be used, or θ has to be estimated. See Chapter 8.

Perfect Field Orientation

With exact knowledge of the flux angle, θ1 = θ, (5.99) reduces to
3np
τe = ψR iq . (5.100)
2K 2
This relation is similar to (5.55) for the IM. It is again only the current component iq that con-
tributes to the torque, so it is apparent that the two motor types can be controlled using similar
principles. Since the flux is created from inside the rotor, id can be controlled to zero, giving max-
imum torque for a certain stator current modulus. A PMSM therefore requires a smaller stator
current modulus than an IM of similar power rating.
For the PMSM it can be said that the magnetization is “paid once and for all” when the motor
is purchased. For the IM, the magnetization has to be supplied from the stator, which is paid for
by a larger stator current (which may require a converter of higher current rating), giving increased
losses in the stator resistance as well. This may offset the extra investment cost of a PMSM as
compared to an IM of similar rating.

5.3.4 Rotor Flux and Base Frequency

For id = 0 we obtain the following steady-state relations from (5.97):
vd = −ωr Ls iq (5.101)
vq = Rs iq + ωr ψR . (5.102)
When the torque is low, iq ≈ 0, we have |vs | = |vq | = |ωr |ψR . Hence, at base speed,
Vbase ≈ ωbase ψR (5.103)
so it is logical to select the base frequency as
Vbase
ωbase = . (5.104)
ψR
Then, ψR = 1 pu (unlike for the IM, which has ψR < 1 pu).

5.3.5 Field Weakening

As ψR is constant, field weakening requires making id negative. In the steady state with a small
iq , vd = Rs id − ωr Ls iq is also small, whereas
vq = ωr (Ls id + ψR ) ≈ Vbase . (5.105)
But |id | must not be made larger than Ibase , so
Vbase
ωrmax = . (5.106)
ψR − Ls Ibase
For Ls in the interval 0.2 to 0.4 pu we get ωrmax from 1.3 to 1.7 pu. The field-weakening range for
round-rotor PMSMs is thus normally much smaller than for IMs.
166 Chapter 5. AC Motor Drives

5.3.6 Transient Impedance

The impedance of the PMSM, as observed from the stator terminals, is given by
Zss (p) = Rs + pLs Zs (p) = Rs + (p + jω1 )Ls . (5.107)
Since there is no magnetizing branch in the equivalent circuit of the PMSM, this is also the
transient impedance. Comparing to the transient impedance of the IM—(5.78)—we see that the
resistive part is roughly half that of the IM (R = Rs + RR ). The stator inductance is, as mentioned
previously, in the same per-unit range as the leakage inductance of the IM. Thus, the time constant
of the PMSM
Ls
Ts = (5.108)
Rs
is in per-unit value typically larger than the transient time constant Tσ the IM.

5.3.7 Power Factor

One benefit of the PMSM as compared to the IM is that it can be operated with unity power factor.
In the steady state at higher speeds (when the stator-resistance voltage drop can be neglected),
and with perfect field orientation, we have
vs = jωr (Ls is + ψR ) (5.109)
giving
: ;
vs i∗s = ωr ψR iq + jωr ψR id + Ls (i2d + i2q ) . (5.110)
The active and reactive powers—cf. (2.132) and (2.133)—are proportional to the real and imaginary
parts of (5.110)
: ;
P ∼ ωr ψR iq Q ∼ ωr ψR id + Ls (i2d + i2q ) (5.111)
so by controlling id such that ψR id + Ls (i2d + i2q ) = 0, rather than just letting id = 0, unity power
factor is obtained. However, even when id = 0 the power factor
P 1
cos ϕ = " = S , -2 (5.112)
P 2 + Q2 Ls iq
1+
ψR
is fairly high. At base flux (ψR = 1 pu) and base current (iq = 1 pu), and with a typical inductance
Ls = 0.5 pu, a power factor of 0.89 is obtained. Let us compare this to an induction motor. The
relation for the IM corresponding to (5.109) is given by
vs = jω1 (Lσ is + ψR ). (5.113)
For perfect field orientation we have
vs i∗s = jω1 [Lσ (i2d + i2q ) + ψR id ] + ω1 ψR iq (5.114)
which yields
P 1
cos ϕ = " = S . (5.115)
2
P +Q 2 ( )2
Lσ (i2d + i2q ) + ψR id
1+
ψR iq
5.3. Dynamic Model for the Permanent-Magnet Synchronous Motor 167

PROBLEM 5.7
Suppose that LM = 2 pu and Lσ = 0.2 pu. What is then the power factor of the IM for a)
the nominal load and b) half the nominal load?

PROBLEM 5.8
Find an expression for how id should be selected in order to give unity power factor for the
PMSM.

PROBLEM 5.9
A two-pole, Y-connected PMSM with surface-mounted magnets has the nominal phase volt-
age and speed 230 V and 4000 rpm, respectively. The stator resistance is negligible and the
stator inductance is 20 mH.

a) What is the modulus of the space vector for the rotor flux linkage assuming rms-value
scaling?
b) The converter is rated 20 A (phase current). What is the maximum attainable speed in
the steady state? Hint: Field weakening; assume perfect field orientation.
c) The load torque reaches 20 Nm over a short period of time. How should iq be chosen
if the load torque is to be balanced by the electrical torque? Is the converter’s current
rating exceeded?
d) Can unity power factor (seen from the converter, i.e., the stator terminals) be obtained
for the operating condition in c)?
e) Determine the maximum obtainable steady-state electrical torque for unity power factor.

PROBLEM 5.10
En eight-pole PMSM is fed from a converter. The mechanical speed is 1500 rpm and the load
torque is 100 Nm. The stator inductance is 5 mH and the stator resistance can be neglected.
By closed-loop control, the stator current is regulated such that is it in phase with the stator
voltage, whose rms phase voltage is 100 V.
The current references are changed, such that id = 0 is obtained. Which rms phase stator
voltage is now required in order to maintain the torque and speed?

PROBLEM 5.11
An converter-fed two-pole PMSM operates with the stator frequency 17 Hz, producing a
power of 2.0 kW to the mechanical load. The stator current is controlled such that id =
0, rms-value space-vector scaling is used, the rotor position is measured, and the motor
parameters are Rs = 0.5 Ω, Ls = 6.0 mH, and ψR = 0.312 Vs.

a) Find an expression for the stator current vector in stationary coordinates.

b) The control unit ceases to work, locking all three converter switches in the “1” position.
Determine the stator current vector in synchronous coordinates as a function of time
after the occurrence of this fault. The rotor speed can be assumed constant.

5.3.8 Salient PMSMs and Reluctance SMs

As the model otherwise would become extremely complicated, the mutual inductance of a salient
SM is assumed to vary sinusoidally, cf. (5.2). Yet, complex space vectors cannot be used to model
168 Chapter 5. AC Motor Drives

salient PMSMs. The good thing is that equivalent real space vectors can be introduced instead,
keeping essentially the same notation as in complex quantities. Restricting the study to the case
of perfect field orientation, replacing complex space vectors by real vectors and replacing j by the
matrix J in (2.114), (5.97) is transformed to
( )
dis ψR
Ls = vs − (Rs I + Jωr Ls )is − Jωr ψR , ψR = (5.116)
dt 0

where I is here the 2 × 2 identity matrix. As the stator inductance is not constant around the
airgap, but equal to Ld and Lq along the d and q axes, respectively, Ls is a matrix
( )
Ld 0
Ls = . (5.117)
0 Lq

In component form we obtain

did
Ld = vd − Rs id + ωr Lq iq (5.118)
dt
diq
Lq = vq − Rs iq − ωr (Ld id + ψR ). (5.119)
dt
Also in this case, the equations can be represented as an equivalent circuit. However, due to the
saliency, separate circuits are required for the d and q axes; see Figure 5.21. Thus, the dynamics
are different in the two directions. Disregarding the cross coupling between the d and q axes, the
transient impedances are, respectively, given by

Zd (p) = Rs + pLd Zq (p) = Rs + pLq . (5.120)

Ld ωrLqiq
id Rs

vd

Lq ωrLdid
iq Rs

vq ωrψR

Figure 5.21. Dynamic equivalent circuit for the salient PMSM.

The mechanical power can be found by calculating the total active power developed in the emfs
in Figure 5.21. With Ed = −ωr Lq iq and Eq = ωr (Ld id + ψR ), we have

3 3
(Ed id + Eq iq ) = ωr [ψR iq + (Ld − Lq )id iq ]. (5.121)
2K 2 2K 2
5.3. Dynamic Model for the Permanent-Magnet Synchronous Motor 169

Dividing this expression by ωm = ωr /np , a formula for the electrical torque is obtained as

3np
τe = [ψR iq + (Ld − Lq )id iq ]. (5.122)
2K 2
3np
Compared to (5.100), there is now an additional term: (Ld − Lq )id iq , which is due to the
2K 2
saliency of the rotor. This is the reluctance torque. When the saliency Ld − Lq is relatively large
it may be a good idea to use a nonzero id in order to maximize the torque R production for a given
stator current modulus, by utilizing the reluctance torque. Assuming that i2d + i2q = Ibase it can
be shown that (5.122) is maximized for
"
id ( 1 + 8χ2 − 1) sgn (χ) (Ld − Lq )Ibase
= √ R " ≈ χ, χ= (5.123)
iq ψR
2 4χ2 + 1 + 8χ2 − 1
. /0 1
r(χ)

where the approximation holds for small χ. This is illustrated in Figure 5.22; notice that the linear
approximation holds quite well also for the relatively large saliency χ = −0.4.

1.1

1.05
τ (pu)

1
e

0.95

0.9
−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1
id/iq

Figure 5.22. Maximum torque of a salient PMSM for (solid) χ = 0, (dashed) χ = −0.2, and (dash-dotted)
χ = −0.4.

Inductances and Field Weakening

Mounting the magnets inside the rotor allows the airgap width to be smaller. The d and q direction
stator inductances are therefore larger in per-unit value than the stator inductance of a PMSM with
surface-mounted magnets. Characteristic per-unit values are 0.4 < Ld < 0.7 and 0.7 < Lq < 1.
This allows a larger field-weakening range. For salient PMSMs, it is easily seen that Ld replaces
Ls in (5.106)

Vbase
ωrmax = . (5.124)
ψR − Ld Ibase

Thus, ωrmax between 1.7 pu and 3.3 pu can be obtained with a salient PMSM.
170 Chapter 5. AC Motor Drives

Reluctance SMs
A motor with Ld ̸= Lq produces torque even when there is no magnetization in the rotor: ψR = 0.
This is called a reluctance synchronous motor (RSM) or a synchronous reluctance motor (SynRM).
Clearly, to obtain a reluctance torque, neither id nor iq must be zero. It is not difficult to deduce
that |i
√d | = |iq | yields maximum
√ torque for a certain current modulus. This can be shown by letting
id = 2I cos δ and iq = 2I sin δ. Then

id iq = 2I 2 cos δ sin δ = I 2 sin 2δ (5.125)

which is maximized for δ = 45◦ .

The drawback of RSMs as compared to PMSMs is that a large id is required. This implies that
the power factor will be lower, in the same range as for an IM. This is quite natural, since in both
cases magnetization is made from the stator.

5.4 Unified IM/PMSM Model

Having derived dynamic models for the IM and PMSM, it will now be shown that these can be
unified as one common model [32], assuming a round-rotor PMSM. One purpose of the unified
model is to give better insight in the similarities and differences of sensorless control of IMs and
PMSMs, see further Chapters 8 and 9.
Consider the IM model given by (5.35)–(5.36)
diss dψsR
Lσ = vss − Rs iss − (5.126)
dt , dt -
dψsR s RR
= RR is − − jωr ψsR . (5.127)
dt LM
The key to obtaining a unified IM/PMSM model is to express the rotor flux in polar form also for
the IM

ψsR = ψR ejθ . (5.128)

Transforming (5.127) to synchronous coordinates using the transformation angle θ1 , is = e−jθ1 iss ,
results in
, -
d(ψR ejθ ) jθ1 RR
= RR e is − − jωr ψR ejθ (5.129)
dt LM
giving
, -
RR
(ψ̇R + j θ̇ψR )ejθ = RR ejθ1 is − − jωr ψR ejθ
LM
, -
RR
⇒ ψ̇R + j θ̇ψR = RR e−j θ̃ is − − jωr ψR (5.130)
LM

where, as for the SM, θ< = θ − θ1 . Equation (5.126) is transformed to synchronous coordinates in
a similar way
dis
Lσ = vs − (Rs + jω1 Lσ )is − (ψ̇R + j θ̇ψR )ej θ̃ . (5.131)
dt
5.5. Mechanical Dynamics and Speed Control 171

Now, splitting (5.130) into real and imaginary parts yields

RR
ψ̇R = RR (id cos θ̃ + iq sin θ̃) − ψR (5.132)
LM
RR iq
θ̇ = ωr + (iq cos θ̃ − id sin θ̃). (5.133)
ψR

Not surprisingly, for perfect field orientation, θ< = 0, these two relations simplify to the ideal flux
modulus dynamics (5.45) and the slip relation (5.46), respectively
RR
ψ̇R = RR id − ψR (5.134)
LM
RR iq
θ̇ = ωr + (5.135)
ψR
giving ω1 = θ̇1 = θ̇.
We now adapt this model to the round-rotor PMSM. Since this motor has no rotor resistance,
it is logical to start by letting RR = 0 in (5.132) and (5.133), yielding
ψ̇R = 0 (the PMSM’s flux modulus is constant) (5.136)
θ̇ = ωr (the PMSM rotates synchronously). (5.137)
Hence, quite interestingly, the flux dynamics (5.132)–(5.133) are valid also for the PMSM, simply
by letting RR = 0. Substituting these the relations in (5.131) yields
dis
Lσ = vs − (Rs + jω1 Lσ )is − jωr ψR ej θ̃ . (5.138)
dt
But this relation is identical to the stator current dynamics of the PMSM, (5.96), except that the
leakage inductance Lσ should be replaced by the stator inductance Ls .

5.5 Mechanical Dynamics and Speed Control

The mechanical dynamics (4.16) for the dc motor also apply to the IM and the PMSM, assuming
perfect field orientation and that the reluctance torque can be neglected for salient SMs. To take
the number of pole pairs into account, ωr should be replaced by the mechanical rotor speed ωm ,
however
dωm 3np
J = ψR iq − τl . (5.139)
dt 2K 2
The load torque τl can be modeled as in (4.78) as a viscous part plus a part due to external
conditions:

τl = bωm + τL . (5.140)

Since ωr is the variable appearing in the electrical dynamics of the IM and the PMSM, it is
convenient to express (5.139) in ωr rather than ωm :
J dωr 3np ωr
= 2
ψR iq − b − τL (5.141)
np dt 2K np
172 Chapter 5. AC Motor Drives

or
dωr 3np
J′ = ψR iq − b′ ωr − τL (5.142)
dt 2K 2
where J ′ = J/np and b′ = b/np . This mechanical model is formally equal to that for the mechanical
dynamics of the dc motor, (4.79). Speed controllers for IM and PMSM drives can therefore be
designed as in Section 4.6, simply by making the following substitutions:
J b 3np ψ!R
J→ b→ ψ→ i → iq (5.143)
np np 2K 2
where ψ!R is an estimate of the rotor flux for IMs (in Chapter 9 it is shown how this can be
obtained), whereas it is an estimate of the permanent-magnet flux for PMSMs. This yields the
following controller parameter selections:
2K 2 (αs J! − !b) 2K 2 αs J! 2K 2 α2s J!
ba = kps = kis = . (5.144)
3n2 ψ!R
p 3n2 ψ!R
p 3n2 ψ!R
p

In the speed controller algorithm, it must be taken into account that limitation is necessary such
that i2d + i2q ≤ Imax
2
, where Imax is the maximum allowed short-term stator current (which should
be reduced to Ibase after the maximum time that overcurrent is allowed). This yields Algorithm
5.1 and the block diagram shown in Figure 5.23.
Algorithm 5.1
Speed Controller for AC Drives
es = ωref − ωr
iref
q,nom = kps es + kis Is − ba ωr
, R -
iref
q = sat i ref
q,nom , I 2
max − (i ref )2
d
( )
1 ref
Is = Is + Ts es + (iq − iref
q,nom )
kps

5.5.1 Normalization
Starting with base value for torque given in (5.65)
3np Vbase Ibase
τbase = (5.145)
2K 2 ωbase
it is logical to select the base values for inertia and viscous damping constant as
np τbase np τbase
Jbase = 2 bbase = . (5.146)
ωbase ωbase
(Note that unlike the dc-motor base values in Table 4.2, these base values include the number of
pole pairs.) This gives the following normalization of the mechanical dynamics:
dωrn
Jn = ψRn iqn − bn ωrn − τLn . (5.147)
dtn
This equation is not formally equal to (5.142), as the factor 3np /2K 2 has been removed. Exercise
caution when using it!
5.6. Determination of Motor Parameters 173

1/k ps

Ts
D

Is
kis
ωref
es nom sat
kps iq,ref iq,ref
ωr ba
Figure 5.23. Digitally implemented two-degrees-of-freedom speed controller for ac drives.

5.6 Determination of Motor Parameters

As will be shown in the subsequent chapters, control of ac motor drives requires (at least fairly)
accurate estimates of the motor parameters. Therefore, we in this section study methods for deter-
mining such from provided data and tests. We focus√on IMs and round-rotor PMSMs. Throughout
the section, rms-value space-vector scaling, K = 1/ 2, is assumed. This is merely done for conve-
nience, and does not dictate the same choice in the control system designed based on the obtained
motor parameters.

5.6.1 Induction Motors

Name-Plate Data
Most induction motors are equipped with a name plate, on which important data are given. The
following data for nominal operation are usually provided.
• RMS phase voltage and current for Y connection VN and IN .
N
• Output power Pout .

• Stator frequency in Hz fN .
• Mechanical rotor speed in rpm nN .

• Power factor cos ϕN .

From these data, a number of important quantities and relations, which also hold for nominal
operation, can be derived:
N
• Apparent input power: Sin = 3VN IN .
N N
• Input power (active): Pin = Sin cos ϕN .
N N
• Efficiency: ηN = Pout /Pin .

• Angular stator frequency: ωN = 2πfN .

174 Chapter 5. AC Motor Drives

N
• Mechanical rotor speed in rad/s: ωm = 2πnN /60.
N
• Number of pole pairs np : If the IM had rotated synchronously, then np = ωN /ωm . Due to
N N
the slip, ωm is lower than the synchronous speed, though, so we get np = int(ωN /ωm ) where
N
int(·) indicates integer part. Example: ωN = 314 rad/s (50 Hz) and ωm = 100 rad/s ⇒
np = int(3.14) = 3 (a six-pole motor).
• Electrical rotor speed in rad/s: ωrN = np ωm
N
.
• Slip: sN = (ωN − ωrN )/ωN .
• Torque: τeN = Pout
N N
/ωm .

Locked-Rotor and No-Load Tests

From the name-plate data obtained, only one motor parameter can be determined with good
accuracy: the rotor resistance. Equation (5.72) gives the following relation:
3np s|vs |2
RR = .
2K 2 ω1 τe
√
With K = 1/ 2 we get, at nominal operating conditions,
2
!R = 3np sN VN .
R (5.148)
ωN τeN

To obtain the remaining parameters, tests must be made. The stator resistance is particularly
straightforward to obtain, as it is found simply by ohm measurement across the stator terminals.
In order to determine the inductances, two tests are useful. (Both resistances are also obtained
in the process, giving alternative ways of determining them.)
• In the locked-rotor test, the rotor is barred from moving (giving ωr = 0 ⇔ s = 1).
Using the nominal frequency, the stator voltage is then increased to a value Vlr (rms phase
voltage) when the phase rms current reaches the base value. The (active) input power is
then measured (using a watt meter): Plr . Because the total resistance in the rotor circuit is
only RR (as compared to RR /s normally), the magnetizing branch is for practical purposes
short circuited, cf. Figure 5.10. The total impedance is then the transient impedance (5.81)
Zσ (jωN ) = Rs + RR + jωN Lσ .
Since the rotor is locked, all of the active input power is dissipated in the total resistance
Rs + RR

Plr = 3(Rs + RR )|is |2 = 3(Rs + RR )IN

2
⇒R !R = Plr .
!s + R (5.149)
2
3IN
The power developed in the leakage inductance is, of course, reactive and equal to Qlr =
"
Srl2 − Prl2 , where the apparent power Srl is given by Srl = 3Vlr IN . Hence,

Qlr = 3ωN Lσ |is |2 = 3ωN Lσ IN 2

S , -2
! σ = Vlr
⇒L 1−
Plr
. (5.150)
ωN IN 3Vlr IN
5.6. Determination of Motor Parameters 175

• Remaining to be determined is the magnetizing inductance. In the no-load test, base

voltage with base frequency is applied, and the motor is allowed to run light (no mechanical
load). The slip is then approximately zero, so the rotor circuit is broken up. The IM then
draws approximately the nominal magnetizing current as stator current: |is | = Inl (which is
measured), and the following steady-state relation holds:

vs = [Rs + jωN (Lσ + LM )]is .

The voltage drop across the stator resistance is negligible, so

! M = VN − L
VN = ωN (Lσ + LM )Inl ⇒ L !σ . (5.151)
ωN Inl

If the power Pnl is measured using a watt meter, the stator resistance can be obtained as

!s = Pnl .
R (5.152)
2
3Inl

Example 5.1 An IM with the following name-plate data is provided:

1.6 kW 220 V/3.8 A (Y) 1430 rpm 50 Hz cos ϕ = 0.76

and the results of locked-rotor and no-load tests are

Plr = 377 W Vlr = 51.5 V Inl = 2.0 A.

A VSC with the dc-link voltage Vdc = 540 V is used.

Check whether the converter is correctly selected for this motor, then find the nominal torque,
efficiency, and slip, as well as the base values and per-unit motor parameters for rms-value scaling.
√ √
Solution. From (3.21) we obtain Vbase = 540/ 6 = 220 V for K = 1/ 2, which shows that
the converter is perfectly matched to the motor.
N
The nominal speed nN = 1430 yields ωm = 2π · 1430/60 = 150 rad/s. This being a little
less than half the nominal angular frequency, 314 rad/s, indicates that this is a four-pole motor
N
(np = 2). Nominal slip: sN = (314 − 2 · 150)/314 = 4.8%. With Pout = 1.6 kW we have
N N
τe = 1600/150 = 10.7 Nm. We find that the nominal input power is Pin = 3VN IN cos ϕN = 1.9
kW, so ηN = 1.6/1.9 = 84%. Using now (5.148)–(5.151), we obtain

!R = 4.1 Ω
R !s = 4.6 Ω
R ! σ = 33 mH
L ! M = 0.32 H.
L
√
For K = 1/ 2 we get Ibase = 3.8 A. With ωbase = 314 rad/s, we find the base values for impedance
and inductance:
Zbase = 58 Ω Lbase = 0.18 H
giving in turn the following per-unit parameters:

! M = 1.8
L ! σ = 0.18
L !s = 0.079
R !R = 0.071.
R

♣
176 Chapter 5. AC Motor Drives

PROBLEM 5.12
The following measurements were taken from a Y-connected induction motor for a stator
frequency of 50 Hz.

No-load test Locked-rotor test

Active power (W) 170 750
Phase voltage (V) 230 70
Phase current (A) 5.0 9.0

a) Calculate the motor parameters as given by the inverse-Γ model.

b) What are the base values for voltage, current, and frequency, provided that the tests
were conducted as recommended and that rms-value space-vector scaling is used?
c) Determine the motor parameters in per-unit values.

5.6.2 Permanent-Magnet Synchronous Motors

Some—but not all—PMSMs are equipped with a name plate similar to that of IMs. With VN , IN ,
N
Pout , fN , nN , and cos ϕN given on the name plate, other quantities can be derived in exactly the
N
same way as for IMs in Paragraph 5.6.1, except that np = ωN /ωm , since there is no slip.
Unlike the IM, all motor parameters can be determined from the name plate, however. Obvi-
ously, we have

VN
ψ!R = . (5.153)
ωN

Furthermore, by taking (5.96) in the steady state and multiplying by i∗s , the input–output complex
power balance is obtained as follows:

vs i∗s = (Rs + jω1 )|is |2 + Ei∗s

so, with ω1 = ωr = ωN
N
Pin = 3VN IN cos ϕN = 3 Re{vs i∗s } = 3Rs IN
2
+ 3 Re{Ei∗s }
. /0 1
N
Pout
QN ∗ 2 ∗
in = 3VN IN sin ϕN = 3 Im{vs is } = 3ωN Ls IN + 3 Im{Eis } .
. /0 1
QN
out

This immediately gives

N
!s = VN cos ϕN − Pout .
R (5.154)
IN 2
3IN

To find Ls , we have E = jωr ψR ej θ̃ ⇒ |E| = ωN ψR = VN , so

S
, N
-2
N 2 Pout
(Pout ) + (QN
out )
2 2
= (3VN IN ) ⇒ QN
out = ±3VN IN 1−
3VN IN
5.6. Determination of Motor Parameters 177

where only the minus sign is relevant, since reactive power must be produced in the back emf in
order to get the desired high power factor. Hence, we obtain Im{Ei∗s } ≤ 0, and
⎛ S ⎞
, N
-2
! s = VN ⎝sin ϕN + 1 −
L
Pout ⎠. (5.155)
ωN IN 3VN IN

If full name-plate data are not provided, there should at least be motor data available that spec-
ify nominal stator voltage and current, VN and IN . These given, the motor parameters can be
determined from the following tests.

• Rotation test. Connect mechanically the PMSM to a suitable motor, and rotate the shaft
with a constant speed ωm . Leave the stator terminals open and measure the resulting stator
voltage and frequency, Vs (rms value) and ωs . This yields ψ!R = Vs /ωs and np = ωs /ωm .
The nominal stator frequency is then obtained as ωN = VN /ψ!R .

• Locked-rotor test. Bar the rotor from moving and excite the motor with a variable stator
voltage of angular frequency ωN . Increase the voltage to the value Vlr , at which nominal
stator current is obtained. Measure the input power Plr . With RR = 0 and Ls replacing Lσ ,
(5.149) and (5.150) can then be used to determine Rs and Ls as

Plr = 3Rs |is |2 = 3Rs IN

2 !s = Plr
⇒R (5.156)
2
3IN
Qlr = 3ωN Ls |is |2 = 3ωN Ls IN2
S , -2
! s = Vlr
⇒L 1−
Plr
. (5.157)
ωN IN 3Vlr IN

5.6.3 Parameter Variations

The motor parameters obtained from name-plate data and/or the tests described above are subject
to variation. Parameter variations must be taken into account in the design of the control system;
details are given in Chapters 8 and 9. The motor parameters for IMs and PMSMs vary as follows.

LM : The inductance obtained in the no-load test or from name-plate data is that for the
nominal flux. As illustrated in Figure 4.8(b) for the field-winding inductance of the dc motor,
this is a slightly saturated inductance. In the field-weakening range when the flux is reduced,
LM may increase up to 20% from the nominal value.

Lσ : Also for this parameter, a slightly saturated value is obtained in the locked-rotor test
or from name-plate data. The effect of saturation tends not to be as significant as for the
magnetizing inductance, though this is very dependent on the rotor slot design. Generally
speaking, a fairly constant Lσ can be assumed, however.

Rs and RR : The resistances are often measured with the motor at room temperature. It is
said that the “cold values” of the resistances are obtained. During operation, the temperature
increases, often to above 100◦ C, see [68]. The stator and rotor resistances then generally
increase: between 50% and 100% above the “cold values.” The increase of the stator resistance
is particularly troublesome for low-speed operation; see Chapters 8 and 9.
178 Chapter 5. AC Motor Drives

Ls : The situation is similar to that of Lσ for an IM, i.e., only small variations.

PROBLEM 5.13
On the name plate of a PMSM, the following data are provided: 230 V, 25 A, 16.5 kW,
50 Hz, 1000 rpm. That is, the nominal power factor is missing. After operating the motor
under nominal conditions for about an hour, it is stopped and the resistance across the stator
terminals is measured, giving 0.3 Ω.
a) Why was the motor operated under nominal conditions before the measurement was
made?
b) Find the stator inductance and the nominal power factor from the provided data.
CHAPTER 6

Vector Control Fundamentals

Vector control implies that a three-phase VSC is controlled using space vectors. The concept was
invented in the late 1960s by a research group in Braunschweig, Germany, led by Prof. Werner
Leonhard. The development was spurred by the desire to begin applying the emerging technology
of converter-fed induction motor drives to applications where thus far dc motor drives had been
used. Methods traditionally used for induction motor control (see Paragraph 5.2.12) had been
found inadequate; the fast dynamic response of a controlled dc motor drive was not achieved.
In vector control, which is also known as field-oriented control or simply field orientation, a dq
frame that rotates synchronously with the flux of the motor is constructed, and current control is
often performed in this frame. This allows ac motors to be controlled using similar algorithms as
for dc motors—see Section 4.4—enabling fast dynamic response. Two variants of the concept were
proposed in the theses of Blaschke and Hasse [6, 36].
Whereas vector control initially was developed for induction motor drives, the control concept
was subsequently applied also to synchronous motor drives, and to VSCs connected to the grid.
These three applications are considered in the reverse order, in Chapters 9, 8, and 7, respectively.
In this chapter we discuss the general principles of vector control before going into details. We
also extend the methods for current control of the dc motor—using single-phase VSCs or buck
converters—of Section 4.4 to three-phase VSCs. As much as possible, the theory from Section 4.4
is relied upon and extended. Also included is an introduction to synchronization control using a
phase-locked loop (PLL).

6.1 Principle of Vector Control

Vector control is often introduced as applied to induction motors. This application is technically
the most challenging one and also the one most difficult to grasp. Therefore, we postpone the
details of vector control of induction motors until Chapter 9, and start by applying the concept to
converters connected to a general load or source, which may be either an ac motor or a three-phase
grid. In the latter case there is a simplification to be obtained: the grid frequency is near constant,
whereas the stator frequency of an ac motor drive varies with the speed.
The general load or source is modeled as a series connection of a resistance, an inductance, and
a voltage source, as shown in Figure 6.1. This is a model for a
• permanent-magnet synchronous motor, see Figure 5.20, with

vs = vss is = iss R = Rs L = Ls (6.1)

and back emf

dψsR
Es = = jωr ψR ejθ (6.2)
dt
180 Chapter 6. Vector Control Fundamentals

an
• induction motor, see Figure 5.11, with

vs = vss is = iss R = Rs L = Lσ (6.3)

and flux emf

dψsR
Es = (6.4)
dt
and a
• grid-connected converter, where R and L respectively are resistance and inductance of
the converter’s input filter, whereas Es is the voltage at the point of common coupling (PCC).
As mentioned, unlike the back emf for a PMSM or the flux emf for an IM, the PCC voltage
can be assumed to have a near-constant frequency. On the other hand, the PCC voltage is
not necessarily constant to its amplitude and phase; particularly during grid fault situations,
large transients—often including significant negative-sequence voltages—may occur.

R L
is

s
vs E

Figure 6.1. Converter connected to a circuit consisting of a resistive–inductive impedance and a voltage
source.

In the stationary reference frame, the dynamic model for the system in Figure 6.1 is given by
dis
L = vs − Ris − Es . (6.5)
dt
In Chapter 5 we introduced the concept of field orientation for ac motors, which implies orienting
a rotating dq reference frame along the rotor-flux vector. The rotor flux serves as phase reference
for the dq frame. Because Es is the derivative of the rotor flux according to (6.2) and (6.4), it
is natural for the general case (i.e., including grid-connected converters) to let the dq frame be
perpendicular to Es , as this results in rotor-flux orientation for ac drives. This is accomplished by
considering Es in polar form as

Es = jEejθ . (6.6)

6.1.1 Transformation to Synchronous Coordinates

Let us transform (6.5) to a synchronously rotating reference frame, using the dq transformation

v = e−jθ1 vs i = e−jθ1 is . (6.7)

6.2. Introduction to Vector Current Control 181

With ω1 = θ̇1 we obtain

di
L = v − (R + jω1 L)i − jEej θ̃ (6.8)
dt . /0 1
E

where θ< = θ − θ1 . Suppose now that θ is measurable, so that θ1 = θ can be selected. As noted in
Section 5.2, this is known as perfect field orientation. In the steady state, (6.8) then reduces to

v = (R + jω1 L)i + jE. (6.9)

The complex output power at voltage E is

3 3
S = Ei∗ = {E = jE, i = id + jiq } = E(iq + jid ) ⇒
2K 2 2K 2
3 3
P = Eiq Q= Eid . (6.10)
2K 2 2K 2
Thus, the active and reactive powers P and Q can be controlled simply by controlling appropriately
the current components iq and id . Unity power factor is obtained by letting id = 0. For a
grid-connected converter, iq and id thus are the active- and reactive-power-producing current
components, respectively, whereas for an ac motor drive (cf. Chapter 5) they are referred to as the
torque- and flux-producing current components, respectively.

6.2 Introduction to Vector Current Control

The load or source model in Figure 6.1, applied for the definition of vector control, is directly
applicable also for current control. That is, the controlled process is the series admittance with
resistance R and inductance L. For a grid-connected converter, R and L are respectively the
converter input filter’s resistance and inductance, as discussed previously. For a round-rotor PMSM
drive, the correspondences (6.1) are valid also for vector control control. For an IM drive there is
a difference from (6.3), though. For current control, the electrical dynamics are determined by the
IMs transient impedance Zsσ (p) = Rσ + pLσ [cf. (5.78)], so the parameter correspondences are in
this case

R = Rσ = Rs + RR L = Lσ . (6.11)

That said, owing to the usage of an “active resistance” (see further below), the controller design
is fairly insensitive to R, so letting R = Rs by mistake makes little practical difference.
The differential equation for the system depicted in Figure 6.1 is in the synchronous reference
frame given by (6.8). This complex dynamic model is of order one. Splitting the real and imaginary
parts, we obtain the following second-order model:
did
L = vd − Rid + ω1 Liq − Ed (6.12)
dt
diq
L = vq − Riq − ω1 Lid − Eq . (6.13)
dt
These are two cross-coupled first-order subsystems. The cross-coupling is initiated by the terms
ω1 Liq and ω1 Lid .
182 Chapter 6. Vector Control Fundamentals

As an alternative to the circuit model in Figure 6.1 a block diagram can also be used, see Figure
6.2. Here, the complex transfer function G(s) is obtained by applying the Laplace transform to
(6.8)

1
G(s) = . (6.14)
(s + jω1 )L + R

This corresponds to the dc motor transfer function (4.37). The new twist is the term jω1 L, which
is caused by the transformation to synchronous coordinates: s → s + jω1 . This is the reason for
the cross-coupling between (6.12) and (6.13). As before, E is modeled as a load disturbance that
subtracts from the input signal v, i.e., the voltage impressed by the converter. The output signal
is, of course, the current vector.

v i
G (s)

Figure 6.2. Block diagram for the three-phase load.

Due to the relative simplicity of the dynamic model, it is not very difficult to design a well-
performing current controller. After briefly reviewing some strategies that have been proposed for
the purpose, we shall study synchronous-frame current controllers in some detail. This is mainly
an extension of the theory developed in Section 4.4.

6.3 Review of Methods for Three-Phase Current Control

Several different methods for current control of VSCs—particularly as applied to variable-speed
ac motor drives—have been suggested. In the years around 1985, many good papers on the topic
appeared, adding new and important results. Some of these papers, as well as more recent ones,
are referenced below. A good source for more information is the book by Bose [10, Chapters 4, 5].

6.3.1 Hysteresis Control

This scheme is illustrated in Figure 6.3. It is a direct extension of the hysteresis control scheme for
buck converters described in Section 3.2. Each phase current is measured, fed back, and subtracted
from its reference (a sinusoid). The error signal is then fed to a hysteresis comparator, the output
of which forms the switching signal to the respective converter leg. As in the single-phase case, one
advantage of this control method is that it is simple to implement (at least in analog electronics),
since control and PWM are merged. By decreasing the tolerance band, the switching frequency is
increased. Another benefit is the quick response to a reference change.
There are also some drawbacks associated with the method, however.

• The switching frequency is not constant. The variations are opposite to those desired for
good control; the highest switching frequency is associated with low fundamental frequencies.
This can be mitigated by varying the tolerance band.
6.3. Review of Methods for Three-Phase Current Control 183

ic
ib
ia

iaref ibref icref

Figure 6.3. Three-phase hysteresis current control.

• The actual band to which the current is confined is, somewhat surprisingly, twice the dead-
band. The phenomenon occurs because the phase currents interact in the load. (The problem
can be circumvented by using more elaborate hysteresis schemes.)
• The PWM may enter a state of very fast switchings at low fundamental frequencies, or even
chaotic behavior [63]. This is highly undesirable.
• The method is simple to implement in analog electronics but more difficult in digital elec-
tronics. If the hysteresis comparator is implemented on a DSP (or in hardware, e.g., an in
FPGA), a very high sampling rate must be used in order to prevent timing errors.
Since ease of implementation (in analog electronics) no longer is an objective—owing to powerful
DSPs—the method is not as attractive as it once was.

6.3.2 Stationary-Frame PI Control

In this scheme, iα and iβ are controlled directly in the stationary reference frame. One coordinate
transformation is still needed for synchronization, namely, from the synchronous-frame reference
vector iref = iref
d + jiq
ref
to the stationary-frame equivalent isref . This is required because iref
d
ref
and iq still must be the inputs to the current controller, as they are set proportionally to the
desired reactive and active powers or flux and torque, respectively, see (6.10). Most commonly,
two standard PI controllers (one each for the α and β axes) are used. The controller output vector
forms command signals to the PWM algorithm. See Figure 6.4.
Unlike hysteresis control, this scheme uses two stages: control and modulation. The drawback
of this control scheme is that the actual current will not track the reference in the steady state,
which was pointed out in [71]. Somewhat surprisingly, the origin of this phenomenon was debated
for several years. The explanation was given in [67]. To quote [67, p. 679].
“The fact that this error exists in a current regulated R − L load suggests that the error
observed in ac drives at higher frequencies previously attributed to so-called back EMF
184 Chapter 6. Vector Control Fundamentals

i ref dq i sref Current s

vref 2
PWM
αβ controller 3

is 2
Es
3

θ1 Synchroni−
zation
controller

Figure 6.4. Stationary-frame PI control.

or ‘running up against the bus’ is neither, but rather a consequence of the stationary
regulator itself. Therefore, whether the load is static or dynamic is immaterial from the
standpoint of phase shift and magnitude error present in the steady state currents.”

PI controllers are inherently incapable of giving zero steady-state control error for a sinusoidal
reference. The integral action removes the error only in case the reference is constant in the steady
state. In fact, there is little reason to use integral action at all if a sinusoid is to be tracked! The
pole at the origin of the PI controller should be removed and replaced by a pole pair at ±jω1 , i.e.,
the controller de facto becomes a resonator.

6.3.3 Synchronous-Frame PI Control

The disadvantage of stationary-frame PI control can easily be remedied. Using a DSP, the αβ
sinusoidal signals can be dq transformed to the synchronously rotating reference frame—i.e., to dc
signals in the steady state—where standard PI control works properly. This is illustrated in Figure
6.5, where a PLL is used for synchronization, see Section 6.5. Only one additional coordinate
transformation is required as compared to stationary-frame current control. Synchronous-frame
PI (or PI-type) controllers are therefore often the most attractive choice for current control, due
to their simplicity and ability to provide good performance. Therefore, the rest of the chapter will
focus on this control scheme.
If coordinate transformations yet are undesirable, the control algorithm can be transformed to
and implemented in stationary coordinates, as shown in [67]. This in effect moves the integrator
dq-frame pole at s = 0 to an αβ-frame pole at s = −jω1 . It should be stressed that if one desires
to implement stationary-frame current control, this scheme should be used rather than standard
PI control.
6.4. Design of Synchronous-Frame Current Controllers 185

vref s
vref
dq 2
αβ 3
iref Current vref s
vref
dq 2
controller αβ 3 PWM

i dq is 2
αβ 3
θ1
PLL
E dq Es 2
αβ 3

Figure 6.5. Vector control system with synchronous-frame current control using measurement of Es for
synchronization.

6.4 Design of Synchronous-Frame Current Controllers

Having motivated in the previous section why current control preferably should be made in syn-
chronous coordinates, it will now be seen how such controllers can be designed. We here follow,
but also add to, the method presented in [27].
Consider (6.8) and the real equivalents, (6.12)–(6.13). The latter two equations represent two
cross-coupled first-order subsystems. The cross coupling is initiated by the term jω1 Li in (6.8),
since multiplication by j maps the d axis on the q axis and vice versa

i′ = ji ⇒ i′d + ji′q = j(id + jiq ) = −iq + jid . (6.15)

The first step in the controller design is to cancel this cross coupling. This is easily done—provided
that the inductance L is known with fairly good accuracy—by adding a decoupler jω1 Li ! to the
inner “active resistance” feedback loop (which was introduced in Section 4.4 in order to improve
the disturbance rejection capability). We get

! − Ra )i + E
v = v′ + (jω1 L ! (6.16)

where Ra is the “active resistance.” Term E ! is an optional feedforward of the PCC voltage and
the back emf, respectively for grid-connected VSCs and ac drives. See Figure 6.6.
The optional feedforward term is selected differently depending on the application.

• For grid-connected VSCs, the feedforward term can be set as a first-order low-pass filtering
of the measured E

!= αf
E E (6.17)
p + αf
186 Chapter 6. Vector Control Fundamentals

E G’(s)
i ref e v’ i
F(s) G(s)
E
j ω1L R a

Figure 6.6. Current control with an inner decoupling and “active damping” loop. F(s) and G(s) are the
complex-valued transfer functions of the PI controller and the load, respectively.

where typically αf < αc . This result in

, -
< ! αf p
E=E−E= 1− E= E. (6.18)
p + αf p + αf
< = 0, i.e., the influence of the PCC voltage is mitigated.
So, in the steady state (i.e., p = 0), E
• For ac drives, the feedforward term can be set as an estimate of the back emf for a PMSM
and the flux emf for an IM
! = jω1 ψ
E ! R. (6.19)

In both cases, though, only a partial cancellation is obtained, so an “active resistance” is often
! = L) from (6.8)
useful to improve the rejection of variations in E. We obtain (ideally, if L
di <
L = v′ − (R + Ra )i − E. (6.20)
dt
This equation has no complex coefficients, so the cross coupling has been removed. Also, the
dynamics have been speeded up from R/L to (R + Ra )/L. A current controller having v′ as
output is then used in an outer loop, see Figure 6.6. The decoupled system from v′ to i has the
transfer function
1
G′ (s) = . (6.21)
sL + R + Ra
As this is a first-order complex system (representing two noninteracting first-order systems in the
d and q paths, respectively), PI control is appropriate
ki
F(s) = kp + . (6.22)
s
Following Section 4.4, we select
!
kp = αc L ! + Ra )
ki = αc (R (6.23)
where αc is the closed-loop-system bandwidth. If the inner feedback loop is made as fast as the
closed-loop system, i.e.,
R + Ra !−R
!
= αc ⇒ Ra = αc L (6.24)
L
6.4. Design of Synchronous-Frame Current Controllers 187

! On the other hand, if one-degree-

then the selection for the integral gain simplifies to ki = α2c L.
! !
of-freedom PI control is used by letting R = 0, then ki = αc R. The controller is described by

e = iref − i (6.25)
dI
= e (6.26)
dt
! − Ra )i + E
vref = kp e + ki I + (jω1 L ! (6.27)

where I = Id + jIq is the complex integrator state variable.

6.4.1 Bandwidth Selection

Selection of the closed-loop-system bandwidth can be made based on the desired closed-loop-
system rise time trc , as αc = ln 9/trc; typical rise times for state-of-the-art VSCs are in the
millisecond range or below, giving αc in the range of thousands of radians per second. Yet, to gain
stability robustness—see Section 4.4 for details—the bandwidth should not exceed 10% and 4% of
the angular sampling frequency, respectively for one-degree-of-freedom and two-degrees-of-freedom
control, cf. (4.73) and (4.75). That is,
*
0.1ωs , if Ra = 0
αc ≤ ! − R.
! (6.28)
0.04ωs , if Ra = αc L

Example 6.1 Design a current controller for an induction motor having the following data:

Base voltage: 230 V

Base current: 10 A
Base frequency: 50 Hz
Total leakage inductance: L = Lσ = 20 mH
Total resistance: R = Rs + RR = 3 Ω
Desired stator current rise time: 1 ms.

The inner feedback (“active resistance”) loop should be as fast as the closed-loop system.

Solution. The impedance base value is 230/10 = 23 Ω, so in per-unit values the total leakage
inductance is 314 · 10 · 10−3 /23 = 0.28 and the total resistance 3/23 = 0.13. For a current rise time
of 1 ms, a closed-loop bandwidth of 2.2/10−3 = 2200 rad/s is needed, i.e., αc = 7.0 pu. This yields
the following controller parameters in per-unit values:

Ra = αc L − R = 1.8 kp = αc L = 2.0 ki = α2c L = 13.7.

6.4.2 PWM, Time-Delay Compensation, and Antiwindup

The current controller’s output signal vref transformed to stationary coordinates, vsref = ejθ1 vref ,
forms the command voltage vector to the PWM algorithm. Algorithm 3.1 is used to generate
the appropriate reference signals from vsref . From Paragraph 3.12 we know that PWM using
synchronous sampling gives a time delay Td = 1.5Ts . In addition, we know from Paragraph 2.7.3
188 Chapter 6. Vector Control Fundamentals

that the dq transformation of a time delay gives an undesired rotation of the dq frame by the
angle −ω1 Td . This angle can be compensated for in the αβ transformation by adding ω1 Td to the
transformation angle θ1 as

vsref = ej(θ1 +ω1 Td ) vref , Td = 1.5Ts . (6.29)

Using vsref , the PWM algorithm computes the PWM phase-reference and also the realizable vector
v̄sref (which differs from vsref in the saturation region). As for dc-drive current control, the ideal
control algorithm (6.25)–(6.27) has to be modified using “back calculation” in order to avoid
integrator windup in the saturation region. This implies adding a term to (6.26) [cf. (4.68)–(4.70)]
as follows:

e = iref − i (6.30)
dI 1
= e+ (v̄ref − vref ) (6.31)
dt kp
! − Ra )i + E
vref = kp e + ki I + (jω1 L ! (6.32)
v̄ref = PWM(vref , θ1 + ω1 Td ). (6.33)

“PWM” here indicates the PWM algorithm including coordinate transformations. Euler discretiza-
tion of the integrator results in Algorithm 6.1, which is expressed in real vector form. The corre-
sponding vector block diagram is depicted in Figure 6.7.

1/ kp

Ts
D
I
ki
iref e
kp vref PWM vref

i E
jω1L Ra

Figure 6.7. Digitally implemented current controller in complex space-vector form.

Example 6.2 In this example, we simulate current control of a PMSM with the parameter values
L = 0.5 pu and R = 0.05 pu. In the current controller design procedure, slightly inaccurate
parameter values are available: L ! = 0.6 pu and R ! = 0.08 pu. The controller is designed for a
current rise time of 1 ms for a 50-Hz base frequency, which yields αc = 7 pu. Back-emf feedforward
is not used.
Step responses of the stator current are depicted in Figure 6.8. At ωr = 0.5 pu, irefq is first
stepped up from 0 to 0.6 pu at t = 0, and down to 0.1 pu at t = 5 ms. At the first step, maximum
available converter voltage is reached. Therefore, the rise time is longer than the desired 1 ms.
At the second step, the maximum voltage is only touched, so the rise time (or, in this case more
6.5. Synchronization Control Using a Phase-Locked Loop 189

appropriately, fall time) equals the ideal 1 ms. The controller is robust to the slightly inaccurate
parameter values used. Note particularly that id hardly deviates at all from its reference iref
d = 0.
Figure 6.9 shows the difference in performance when no “back calculation” is used. Now, the
overshoot for the first step response is significant.

Algorithm 6.1
Vector Current Controller
( ) ( )( )
id cos(θ1 ) sin(θ1 ) iα
=
iq − sin(θ1 ) cos(θ1 ) iβ
( ) ( ref ) ( )
ed id id
= −
eq iref
q iq
( ref ) ( ) ( ) W I( ) W I
vd ed Id Ra !
ω1 L id !d
E
= kp + ki − +
vqref eq Iq ! Ra
−ω1 L iq !q
E
θ1′ = θ1 + ω1 Td
( ref ) ( ) ( ref )
vα cos(θ1′ ) − sin(θ1′ ) vd
=
vβref sin(θ1′ ) cos(θ1′ ) vqref
[v̄α , v̄β ] = PWM([vα , vβ ])
( ref ) ( ) ( ref )
v̄d cos(θ1′ ) sin(θ1′ ) v̄α
=
ref
v̄q ′ ′
− sin(θ1 ) cos(θ1 ) v̄βref
( ) ( ) W Z [ I
Id Id ed + k1p v̄dref − vdref
= + Ts Z [
Iq Iq eq + k1p v̄qref − vqref

6.5 Synchronization Control Using a Phase-Locked Loop

Since vector control relies on aligning a rotating dq reference frame perpendicularly to the space
vector Es , one crucial component in any vector control system is a controller that performs syn-
chronization with the rotation of Es . How the synchronization problem is solved differs greatly
from one application to another, as we shall see in Chapters 7, 8, and 9. In some applications, the
phase angle θ is directly available for measurement. Usually, though, it is available only indirectly
from certain measurable signals, often the components of Es = jEejθ = Eα + jEβ

Eα = −E sin θ Eβ = E cos θ. (6.34)

The synchronization controller shall be designed so that θ1 tracks θ, given the information available–
i.e., characteristically (6.34)—with sufficient accuracy and speed. In addition, various disturbances
that are always present in the signals—but not explicitly shown in (6.34)—should be suppressed.
At a glance, this task may seem simple. From (6.34), just compute −Eα /Eβ = tan θ and then
apply the inverse tangent function. This method is impractical, however, since Eβ = E cos θ = 0
two times per each fundamental period, leading to division by zero, or by a small number at the
190 Chapter 6. Vector Control Fundamentals

0.8

(pu)
0.6
id (−−)
0.4
iq (−)

0.2

0 1 2 3 4 5 6 7 8
t (ms)

1
(pu)

0.5
vd (−−)

0
vq (−)

−0.5

−1
0 1 2 3 4 5 6 7 8
t (ms)

Figure 6.8. Current control with “back calculation” of a PMSM.

very least. The obtained estimate θ1 would very likely be quite distorted by noise. An improvement
can be obtained by applying the two-argument inverse tangent function as

θ1 = atan2(−Eα , Eβ )

but yet a noisy estimate is likely to be obtained. In addition, to obtain ω1 , differentiation of θ1 is

required, resulting in further noise amplification. To avoid these problems, most synchronization
controllers use some variant of a phase-locked loop, the fundamentals of which we now shall present.
Transforming Es to synchronous coordinates using the angle θ1 yields, as shown in (6.8)

E = e−jθ1 Es = jEej θ̃ (6.35)

whose components are

Ed = −E sin θ< <

Eq = E cos θ. (6.36)

Synchronism implies that θ< = 0 ⇒ Ed = 0. Consequently, θ1 can in principle be obtained by

integrating Ed . Provided that stability is obtained, adjustment of θ1 will then continue through
the integration until sin θ< = 0, which in PLL terminology is known as phase lock. Let us construct
a general PLL algorithm by starting with an estimate ω !g of ωg = θ̇. For a grid-connected VSC
this is simply the nominal angular grid frequency, e.g., ω!g = 2π · 50 rad/s or ω
!g = 2π · 60 rad/s.
6.5. Synchronization Control Using a Phase-Locked Loop 191

0.8

(pu)
0.6
id (−−)
0.4
iq (−)

0.2

0 1 2 3 4 5 6 7 8
t (ms)

1
(pu)

0.5
vd (−−)

0
vq (−)

−0.5

−1
0 1 2 3 4 5 6 7 8
t (ms)

Figure 6.9. Current control without “back calculation” of a PMSM.

To ω
!g , corrective terms proportional to −Ed and the integral thereof are added
$
ω!g − kpp Ed −kip Ed dt . (6.37)
. /0 1
ωi

This signal is then filtered through a low-pass filter in order to suppress noise. The filter output is
the instantaneous fundamental angular frequency ω1 . Often, a first-order filter is sufficient
αl
ω1 = (!
ωg − kpp Ed + ωi ). (6.38)
p + αl

Finally, the aforementioned integration is applied, i.e., ω1 is integrated into the transformation
angle θ1 as
$
θ1 = ω1 dt. (6.39)

A block diagram showing this PLL is depicted in Figure 6.10(a). As the dq and αβ transfor-
mations require taking cosine and sine of θ1 for calculating the transformation factors e±jθ1 =
cos θ1 ± j sin θ1 , such computations have to be added, but are not shown in the figure. Because the
transformation factors are obtained indirectly via integration of the fundamental angular frequency
followed by trigonometric operations, this type of synchronization method is called indirect field
192 Chapter 6. Vector Control Fundamentals

orientation (IFO). Algorithm 6.2 illustrates a software implementation of the PLL using Euler
discretization of the integrators and the low-pass filter. In the integrator whose output is θ1 the
output is computed modulo 2π, as illustrated Figure 6.10(b), to avoid endless ramping of θ1 , which
eventually might lead to numerical overflow.

Algorithm 6.2
Phase-Locked Loop

ωi = ωi − Ts kip Ed
ω1 = ω1 + Ts αl (!
ωg − kpp Ed + ωi − ω1 )
θ1 = mod(θ1 + Ts ω1 , 2π)

ωg
Ed kpp ω1 θ1
LPF
kip
ωi
(a)

θ1
π

−π π φ

−π
(b)

Figure 6.10. (a) PLL block diagram. (b) Computation modulo 2π: θ1 = mod(φ, 2π).

6.5.1 Analysis

Suppose that the low-pass-filter bandwidth αl is high enough, so that impact of the low-pass filter
can be neglected. From (6.37)–(6.39) we then obtain

ω̇i = −kip Ed (6.40)

θ̇1 = ω1 = ω
!g + ωi − kpp Ed . (6.41)
6.5. Synchronization Control Using a Phase-Locked Loop 193

This system can be analyzed by introducing ω !g , where ω̇g ≈ 0 is assumed, and θ< = θ−θ1 .
< g = ωg − ω
Since Ed = −E sin θ, < we obtain the following nonlinear dynamic system:
ω̇i = kip E sin θ< (6.42)
˙
θ< = ω <
<g − ωi − kpp E sin θ. (6.43)
A number of observations can be made from inspecting (6.42)–(6.43).

Local Stability and Pole Placement

< such that sin θ< ≈ θ.
The local stability can be investigated by assuming a small θ, < This allows
(6.42)–(6.43) to be linearized as
ω̇i = kip E θ< (6.44)
˙
θ< = ω
<g − ωi − kpp E θ< (6.45)
which can be expressed as the state-space model ẋ = Ax + B< <T
ωg , where x = [ωi , θ] and
( ) ( )
0 kip E 0
A= B= . (6.46)
−1 −kpp E 1
The characteristic polynomial of A is given by
det(sI − A) = s2 + kpp Es + kip E. (6.47)
The system is locally asymptotically stable for kpp > 0, kip > 0, and E > 0. Its poles (i.e., the
eigenvalues of A) can be placed arbitrarily by properly selecting kpp and kip . For example, the
selection
2αp α2p
kpp = kip = (6.48)
!
E E!
where αp is the synchronization-loop bandwidth and E ! is an estimate of E, yields a double pole
!
at s = −αp if E = E. This selection gives a relatively high integral gain kip , which is useful for
variable-speed drives when the synchronous frequency may vary significantly and relatively quickly.
On the other hand, for grid-connected VSCs the synchronous frequency only varies slightly about
its nominal value and on a slower time scale. Then, weaker integral action is sufficient, which is
obtained by selecting
αp αp αip
kpp = kip = (6.49)
!
E !
E
where αip ≪ αp . In this case, for E ! = E the characteristic polynomial is s2 + αp s + αp αip ≈
(s + αp )(s + αip ). Regardless whether (6.48) or (6.49) is used, it is normally appropriate to make
the synchronization loop at least one decade slower than the current control loop, i.e.,
αp ≪ αc . (6.50)
A separation of two decades or more is not uncommon, though; whereas αc typically is in the range
of thousands of radians per second, αp is typically in the range of tens to hundreds of radians per
second. In order not to slow down the PLL dynamics, the low-pass-filter bandwidth should be
selected as
αl ≫ αp . (6.51)
194 Chapter 6. Vector Control Fundamentals

PROBLEM 6.1 √
Select kpp and kip such that the poles instead are placed at s = (−1 ± j)αp / 2 (a conjugated
pole pair placed with a 45◦ angle relative the imaginary axis and the distance αp from the
origin).

Steady-State Performance
˙
The steady-state properties of (6.42)–(6.43) can be investigated by setting ω̇i = θ< = 0. From (6.42)
we then get sin θ< = 0. Substituting this in (6.43) yields ωi = ω <g . This shows that a deviation
ω
<g of the estimated ω!g from the actual ωg does not result in a static angular error, owing to the
integrator (with output ωi ). However, if kip = 0 and ωi = 0, then from (6.43) we get the static
angular error
, -
< ω
<g
θ = arcsin . (6.52)
kpp E

Global Stability
The nonlinear dynamics (6.42)–(6.43) have an infinite number of equilibrium points: ω <g⋆ = 0 and
θ<⋆ = 2nπ, n integer. As shown in Appendix A, the PLL is globally stable; that is, regardless
of the starting point [< < T , ωi converges to ω
ωg (0), θ(0)] <g and θ< converges to 2nπ. Unfortunately,
< ≈ 0, there is no guarantee that n = 0 if ω
even if θ(0) <g (0) is large; the estimator may fall several
revolutions, n, behind the motor. This is known as cycle slips in PLL terminology and pole slipping
in synchronous machine terminology. In some cases, slipping can be prevented by modifying the
PLL [86].
CHAPTER 7

Vector Control of Grid-Connected VSCs

196 Chapter 7. Vector Control of Grid-Connected VSCs

.
CHAPTER 8

Vector Control of PMSM Drives

If a position sensor—e.g., a digital position encoder—is mounted at the shaft of a PMSM, then
the flux angle θ becomes readily measurable. This allows setting the dq-frame angle θ1 equal to
the measured θ in the PMSM vector control system, which effectively reduces to just the current
controller. Vector control of a position-sensored PMSM drive is thus relatively straightforward.
Somewhat more care is needed when a resolver is used as position sensor. A resolver is a
position sensor that allows measurement of cosine and sine of the rotor position. A variant of the
PLL presented in Section 6.5 can be applied for extracting θ1 from the resolver signals, as shown
in Section 8.1.
Position sensors are expensive, delicate, prone to failure, often difficult to mount, and require
extra cabling. Therefore it is desirable to avoid them, relying instead on position estimation. This
is called sensorless control, or, to be more precise, position-sensorless control.
In Section 8.2, sensorless PMSM drives with a (magnetically) round rotor are considered, i.e.,
PMSMs with magnets mounted on the surface of the rotor. In this case, estimation of the rotor
position is made using a variant of the PLL presented in Section 8.2.1 from information available
in the back emf. It is very difficult to gain good performance at low speeds, so particular focus in
on solving this problem.
In Section 8.3 we consider sensorless drives with PMSMs that have rotors with buried or inset
magnets, giving a salient rotor. This allows position estimation using injection of a high-frequency
signal, providing good performance at all speeds.
The chapter is finished by discussions of vector current control for salient PMSMs and field-
weakening control, in Sections 8.4 and 8.5, respectively.

8.1 Position-Sensored Control Using Resolvers

A resolver is a miniature ac machine. Its rotor winding is supplied by a low-voltage signal (typically
with a peak value of 5 V) at, usually, 5 kHz. The stator is wound with two perpendicular coils,
in which voltages are induced. The amplitudes of the induced voltages are functions of the rotor
position. See Figure 8.1. (An alternative is the “variable-reluctance” resolver which does not have
a rotor winding [25].) If a resolver is mounted on the same shaft as a PMSM, then the resolver’s
rotor position will coincide with the PMSM’s rotor position, provided that they have the same
number of pole pairs and that they are correctly aligned.

8.1.1 Excitation
Let the resolver excitation signal be u = U cos ωe t. As the voltage drop across the rotor resistance
can be neglected, u is approximately the voltage across the rotor winding. The voltage components
198 Chapter 8. Vector Control of PMSM Drives

uβ

d from sinewave
generator uα
1
θ u
1
to ADC
u
α u uβ
uα 1

Physical outline "Equivalent circuit"

Figure 8.1. Principle of a resolver.

induced in the stator winding are then, ideally,

uα = U cos θ cos ωe t uβ = U sin θ cos ωe t. (8.1)

That is, cosine and sine, respectively, of the rotor position modulated with the carrier frequency ωe .
The algorithm or circuit that performs this extraction is known as a resolver-to-digital converter.
It is desirable that the excitation signal u should be as ideal a sinusoidal voltage as possible.
Demodulation of (8.1) can be performed by multiplying both uα and uβ by the excitation signal,
followed by low-pass filtering [88]. As an alternative, we here present a very simple resolver
excitation and tracking scheme which is proposed in [26]. The idea is based on utilizing synchronous
sampling; samples of uα and uβ are taken once each period of the excitation signal. The resolver
algorithm should thus use the same sampling frequency as the excitation frequency, i.e., 5 kHz,
normally. Via a one-bit D/A converter, the algorithm generates a squarewave signal of angular
frequency ωe . This signal is filtered through an analog (active) low-pass filter, such that the
fundamental frequency component at ωe is phase shifted −90◦ , whereas harmonics at 3ωe , 5ωe ,
etc. are attenuated. Sampling is then made ideally at (in practice near) the peak values of uα
and uβ . The principle is illustrated in Figure 8.2. Thus, no signal demodulation is needed, which
reduces the complexity of the system.

Filter Design
To achieve a phase shift of −90◦ , at least a second-order filter is required. At this low filter order,
the difference in stopband attenuation between various filter types is not tremendous, so let us pick
a Butterworth filter [74] for simplicity. With the Laplace variable s normalized with the angular
cutoff frequency, the second-order Butterworth filter has the following transfer function:
1
HBW (s) = √ (8.2)
s2 + 2s + 1
giving the phase function
√
ω 2
arg HBW (jω) = − arctan . (8.3)
1 − ω2
8.1. Position-Sensored Control Using Resolvers 199

DC bias
removal
uα
ADC
1−bit x LPF u
Res.
DAC −90
uβ ADC

uα

Figure 8.2. Principle of proposed resolver excitation and sampling.

So, the cutoff frequency of the filter should be set to the resolver excitation frequency (5 kHz
normally) to obtain a phase shift of −90◦. An active RC realization is shown in Figure 8.3. Figure
8.4 shows two periods of the input and output signals. Note that sampling is not made exactly at
the peaks. This is due to a small harmonic distortion; the output signal is not a pure sinusoid.

8.1.2 Tracking

The synchronously sampled resolver output signals are ideally

uα = U cos θ uβ = U sin θ. (8.4)

With θ1 as the position estimate, the following error signal to be used as PLL input can be obtained:

Ed = uα sin θ1 − uβ cos θ1 = −U sin(θ − θ1 ) = −U sin θ< (8.5)

allowing (6.40)–(6.41) to be applied to extract θ1 . Equation (6.48) with E ! →U ! , where U

! is an
estimate of U (i.e., 5 V, normally) is useful for selection of the PLL gains. Algorithm 8.1 shows
an implementation of the proposed tracking scheme. The sampling period Ts is the inverse of the
resolver excitation frequency. Parameters kpp and kip can be chosen according to (6.48).

Algorithm 8.1
200 Chapter 8. Vector Control of PMSM Drives

4.50 nF

10 k Ω 10 k Ω

x 2.25 nF
u

Figure 8.3. Second-order Butterworth low-pass filter with cutoff frequency 5 kHz.

5
x, u (V)

−5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
t (ms)

Figure 8.4. Squarewave input signal x (dashed) and resolver excitation signal u (solid).

PLL for Resolver Tracking

Ed = uα sin θ1 − uβ cos θ1
ωi = ωi − Ts kip Ed
ω1 = ω1 + Ts αl (−kpp Ed + ωi − ω1 )
θ1 = mod(θ1 + Ts ω1 , 2π)

8.2 Position-Sensorless Control Using Back-EMF Information

When a position sensor is not used, the rotor position must be estimated from other sources of
information. PMSMs with salient rotors allow the injection of a high-frequency test signal, as
will be shown in Section 8.3. This option is not available for round-rotor PMSMs (with surface-
mounted magnets), or is at least difficult to use. Position estimation must then rely on an estimate
of the back emf obtained using the stator voltage and current as

! = vs − (R
E !s + jω1 L
! s )is . (8.6)
8.2. Position-Sensorless Control Using Back-EMF Information 201

The back-emf estimate can be expressed in the actual back emf and the parameter errors R <s =
! < !
Rs − Rs and Ls = Ls − Ls . Solving for vs in (5.96) and substituting this in (8.6), the following
relation is obtained:
! = (R
E <s + jω1 L
< s )is + jωr ψR ej θ̃ (8.7)
. /0 1
E
where dis /dt has been neglected, since, as discussed in Chapter 6, the current control loop is
assumed to have a much higher bandwidth than the synchronization loop. For the somewhat
<s = L
optimistic assumption that R < s = 0, we get E
! = E, and in component form
!d = −ωr ψR sin θ<
E E <
!q = ωr ψR cos θ. (8.8)
Inspecting these components, the following conclusions can be drawn.
• When ωr is sufficiently large (and its sign is known), a position estimate θ1 can be obtained
using a PLL, since a deviation from θ< = 0 immediately will be reflected in E !d . Sensorless
control of round-rotor PMSMs is not difficult at higher speeds.
!d is no longer a reliable indicator, because E
• On the other hand, when ωr is small E !d will be
<
small even when θ is large. The estimator becomes susceptible to measurement noise and
erroneous model parameters, which are not taken into account in (8.8). Sensorless control at
low speeds is therefore difficult.
• Startup is particularly troublesome. Since E = 0 for ωr = 0, the estimator is then “blind.”
<
The rotor position is not revealed until the motor starts to rotate; the initial error angle θ(0)
<
may attain any value between 0 and 2π. If θ(0) happens to be small [θ1 (0) happens by luck to
be close to θ(0)], then the PMSM will start rotating in the desired direction. However, when
<
θ(0) < < 270◦ ), the PMSM will commence to rotate in the
is large (approximately 90◦ < θ(0)
wrong direction, which is highly undesirable. This cannot be prevented, but it is important
that the system yet synchronizes and the rotation reverses quickly. The estimator must be
carefully designed in order to obtain this property.
• As the back emf vanishes at low speeds, not only startup but also rotation reversal is difficult.
Instead of reversal, lockup at a low speed may occur.
It is by no means easy to design an estimator that gives good performance at low speeds. The
situation is compounded by the fact that in practice, the parameter errors R<s and L< s are generally
not negligible. Particularly, the stator resistance varies with the temperature of the motor, as we
know from Paragraph 5.6.3.

8.2.1 PLL for Sensorless PMSM Drives

The PLL of Section 6.5 can easily be adapted for the case of sensorless control of the PMSM
using back-emf information, i.e., (8.8). The first modification is to replace the constant estimate
ω
!g in (6.38) (see also Figure 6.10) by a varying estimate, since the speed of the motor is variable.
Assuming that θ< is small, such that cos θ< ≈ 1, and that the flux-modulus estimate is accurate,
ψ!R ≈ ψR , then from (8.8) it follows that E
!q /ψ!R ≈ ωr . Thus, we select

E!q
ω
!g = . (8.9)
ψ!R
202 Chapter 8. Vector Control of PMSM Drives

Without integral action in the PLL (which is not needed, because of the varying ω!g ), i.e., kip =
ωi = 0, and with kpp = λs /ψ!R introduced to allow a dimensionless gain λs , then (6.38) can be
expressed as

αl E !q − λs E
!d
θ̇1 = ω1 = (8.10)
p + αl ψ!R
where

λs = λ sgn ω1 , λ > 0. (8.11)

The low-pass-filter bandwidth should preferably be selected significantly larger than the speed-
control-loop bandwidth, i.e,

αl ≫ αs . (8.12)

If there is no speed controller, then the recommendation αl ≪ b/J ensures that the low-pass filter
has much higher bandwidth than the open-loop mechanical dynamics.
!s = Rs ,
Gain selection (8.11) ensures convergence of θ1 . For accurate model parameters, i.e., R
! !
Ls = Ls , and ψR = ψR , then, with impact of the low-pass filter neglected, (8.10) can be expressed
as

θ̇1 = ωr cos θ< − λs ωr sin θ.

< (8.13)

< such that sin θ< ≈ θ< and cos θ< ≈ 1, (8.13) can be combined with θ̇ = ωr as
For a small θ,

˙
θ< = −λs ω1 θ.
< (8.14)

Thus, as long as ωr ̸= 0 and ω1 and ωr have equal signs, then θ< converges asymptotically to zero.
In practice, to reduce noise and eliminate the need for measuring the converter output voltage,
references are preferably used instead of measured quantities in (8.6)

! = v̄ref − (R
E !s + jω1 L
! s )iref (8.15)
s s

where v̄ref
s is the limited (to the PWM hexagon) reference vector. A block diagram of the sensorless
vector control system is depicted in Figure 8.5.

8.2.2 Analysis and Parameter Selection

We shall first select gain parameter λ. A proposal for elimination of the estimator’s sensitivity to
the stator resistance is then presented, which significantly improves the capability of the sensorless
drive of reversing rotation under load.

Selection of λ for Quick Synchronization at Startup

It is relatively easy to show that, when ω1 and ωr have equal signs, the estimator synchronizes for
<
any initial error angle θ(0) [46]. However, when the motor starts rotating in the wrong direction,
the signs of ωr and ω1 are different. It is important that the properties of the estimator are such
8.2. Position-Sensorless Control Using Back-EMF Information 203

vsref dq 2
αβ 3

isref Current dq 2
controller αβ 3 PWM

Rs jω1Ls is
M

Re Im
Eq dq iss 2
Ed
αβ 3
λs θ1

ω1
1/ ψR αl

Figure 8.5. Vector control system for sensorless round-rotor PMSM drives using a PLL variant.

that ω1 then changes sign fairly quickly. The discontinuous nonlinearity sgn ω1 in (8.11) makes
an analytic proof difficult (or even impossible). Therefore, we resort to simulating the system. By
making repeated simulations for various initial values: θ̃(0) = n∆θ, n = 1, 2, . . . , 360◦ /∆θ, for a
small ∆θ, the convergence properties can be investigated for a certain value of λ.
Figure 8.6 shows simulations for ωr > 0, three different gain parameter values: λ = {1, 2, 5},
and ∆θ = 10◦ . The simulations are made in a normalized time scale τ = ωr t, where a constant ωr
for simplicity is assumed. The following conclusions can be drawn.

• Synchronization is guaranteed; θ< converges regardless of the initial error angle, for all three
choices of λ.

• Making λ < 2 slows down the convergence noticeably, as seen by comparing the plots for
λ = 1 and λ = 2.

• Selecting λ > 2 has only marginal impact, as seen by comparing the plots for λ = 2 and
λ = 5.

Selecting λ = 2 (or slightly larger) is therefore a good recommendation.

Selection of id for Stable Rotation Reversal

So far, the parameter errors in (8.7) have not been taken into account. We shall now consider
<s and L
the static impacts of R < s . If the real and imaginary parts of the right-hand side of (8.7)
! !
are substituted for Ed and Eq in (8.10), by setting ω̇1 = 0 and solving for ω1 (observing that ω1
204 Chapter 8. Vector Control of PMSM Drives

λ=5
50

Error angle (degrees) 0

−50
0 1 2 3 4 5 6 7 8 9 10
λ=2
50
Error angle (degrees)

−50
0 1 2 3 4 5 6 7 8 9 10
λ=1
50
Error angle (degrees)

−50
0 1 2 3 4 5 6 7 8 9 10
τ

Figure 8.6. Simulations for ωr > 0 and three different choices of λ.

!d and E
appears in the expressions for E !q ), the following relation is found:

<s (iq − λs id ) + ωr ψR (cos θ< + λs sin θ)

R <
ω1 = . (8.16)
! <
ψR − Ls (id + λs iq )

From this relation, the steady-state position error can be found by setting ω1 = ωr , i.e., assuming
synchronous rotation in the steady state. Under the simplifying assumption that the position error
is small, such that cos θ< ≈ 1, we obtain by solving for sin θ<

<s (λs id − iq )
R ψ<R < s (id + λs iq )
L
sin θ< = − − (8.17)
λs ωr ψR λs ψR λs ψR

where ψ<R = ψR − ψ!R . Conclusions can be drawn as follows.

• At low speeds, the first term on the right-hand side of (8.17) dominates, due to its inverse
proportionality to ωr . Nonsalient PMSMs are normally operated with id = 0, giving sin θ< ≈
−R <s iq /(λs ωr ψR ) for small ωr . A large stator resistance error R
<s and/or a large current iq
(i.e., high torque) dictates that a certain lowest operating speed must be observed in order
to keep the position error sufficiently small. Clearly, rotation reversal under load for id = 0
is difficult, unless the stator resistance is known with very good accuracy.
8.2. Position-Sensorless Control Using Back-EMF Information 205

• If the stator current is controlled such that id = iq /λs (instead of the usual id = 0), the
<s vanishes. Using a nonzero id to some extent reduces the current available
sensitivity to R
for torque production, since i2d + i2q ≤ Imax
2
, where Imax is the maximum permissible current
modulus. We get
Imax
|iq | ≤ " . (8.18)
1 + 1/λ2

For a fairly large λ, the reduction in |iq | is not significant. For example, λ = 2 yields
|iq | ≤ 0.89Imax , whereas λ = 3 yields |iq | ≤ 0.95Imax . Controlling the stator current such
that id = iq /λs is therefore a feasible strategy.

• The parameter error ψ<R is normally small, due to a fairly constant magnet flux.
< s id /(λs ψR ) = {id = iq /λs } =
• For fairly large λ, the position error due to the term L
< s iq /(λ ψR ) will be small.
L 2

• Hence, the unavoidable position error, for id = iq /λs with λ reasonably large (λ ≥ 2), is
given by

< s iq
L
sin θ< ≈ − . (8.19)
ψR

It should be noted that, for a load torque which is constant or has constant sign, id will change sign
at the moment of rotation reversal. Due to the well-known fact that id does not produce torque in
a nonsalient PMSM at accurate field orientation conditions, using a nonzero id will not affect the
operation of the motor in other ways than reducing the maximum |iq | according to (8.18). Making
id = iq /λs is, however, required only at low speeds, since the stator resistance is not sensitive at
high. Once out of the low-speed region, typically around ω∆ = 0.1ωbase , it is safe to let iref
d = 0.

Remark 8.1 Eliminating the sensitivity to the stator resistance is equivalent to using the instan-
taneous reactive power vd iq − vq id for estimation purposes [31, 65]. From (8.10) we have, with
λs = iq /id ,
E!q − λs E!d !q − iq E
E !d /id !q id − E
E !d iq
= =
ψ!R ψ!R ψ!R id
in the steady state, where now vd iq − vq id implicitly appears as a term in the numerator.

Speed Control
Closed-loop speed control of a sensorless drive is not different from a sensored drive. Speed con-
trollers are designed in identical ways in both cases, except that in the sensorless case the estimated
speed ω!r is fed back rather than the measured speed. Since an SM rotates synchronously, ω ! r = ω1 .
Even if αl and λ are chosen following the previously given recommendations, the synchroniza-
tion loop may affect the speed control loop. In [46] it is shown that if the stator inductance is
overestimated, giving L < s < 0, then the speed control loop may turn unstable at higher speeds,
especially if the bandwidth αs is high. In order to guarantee stability, one should make sure that
! s < Ls .
L (8.20)
206 Chapter 8. Vector Control of PMSM Drives

Algorithm and Evaluation

Algorithm 8.2 illustrates the proposed PLL variant. For |ω1 | < ω∆ , iref ref
d = iq /λs to eliminate
ref
sensitivity to the stator resistance, whereas for |ω1 | > ω∆ , id = 0.

Algorithm 8.2
PLL for Sensorless PMSM Control Using Back-EMF Information

λs = λ sgn (ω1 )
if |ω1 | < ω∆
iref ref
d = iq /λs
else
iref
d =0
endif
!d = v̄dref − R
E !s iref ! ref
d + ω1 L s iq
!q = v̄qref − R
E !s iref ! ref
q − ω1 L s id
> ?
!q − λs E
E !d
ω1 = ω1 + Ts αl − ω1
ψ!R
θ1 = mod(θ1 + Ts ω1 , 2π)

Example 8.1 Figures 8.7 and 8.8 show experimental results for a 4.2-kW, six-pole PMSM with
the per-unit parameters Rs = 0.03, Ls = 0.27, ψR = 1.0, and J = 3500, while in the estimator,
λ = 2 is used. In Figure 8.7 it can be observed that, even though the initial error angle is large,
once operation commences at t = 1 s, θ< converges to a small value in less than half a second. The
rotation reversals are made with hardly any transient error angle. In Figure 8.8 it can be noticed
that slow ramp-wise rotation reversals under load also can be accomplished successfully. In this
case, a fairly large transient error angle results around ω1 = 0, but synchronism is kept and the
rotation reversals are completed successfully.

8.3 Position-Sensorless Control Using Saliency and Signal Injection

For salient PMSMs, very accurate position estimation also at low speeds can be obtained using
methods based on signal injection. In the following, we review the method proposed by Corley
and Lorenz [15]. Since the injection of an extra signal will lead to at least some additional losses
and torque ripple, a smooth transition to back-emf-based estimation should be made as the speed
increases. We in this section show how a PLL can be adapted for usage also with the signal-injection
method.
8.3. Position-Sensorless Control Using Saliency and Signal Injection 207

0.1 ω ref

Speed (pu)
0.05
ωr ω1
0
−0.05
−0.1

0 1 2 3 4 5 6 7 8 9
1.5
1
Current (pu)

iq
0.5
0 id
−0.5
−1
0 1 2 3 4 5 6 7 8 9
Error angle (degrees)

100

50

0
0 1 2 3 4 5 6 7 8 9
t (s)

Figure 8.7. Startup and rotation reversals under no-load conditions of a nonsalient sensorless PMSM
drive.

8.3.1 Signal Injection and Position Extraction

The idea is to inject a high-frequency voltage in what is assumed to be the d direction. This is
made by adding a component to the d-direction voltage reference as

vdref ′ = V cos ωe t + vdref . (8.21)

Typically, the frequency of the added component is selected around 500 Hz, i.e., ωe = 2π · 500
rad/s. From the current in the assumed q direction, rotor position information can be extracted,
provided that the rotor is salient and the direction of the saliency (i.e., the sign of Lq − Ld ) is
known. Figure 8.9 illustrates the method.
At standstill, (5.116) is reduced to

is = eJ θ̃ (pL + R)−1 e−J θ̃ vs . (8.22)

. /0 1
G(p, θ̃)

With the stationary sinusoidal excitation (8.21), we need to consider p = jωe only:

< = eJ θ̃ (jωe L + R)−1 e−J θ̃ ≈ eJ θ̃ (jωe L)−1 e−J θ̃ = 1 eJ θ̃ L−1 e−J θ̃

G(jωe , θ) (8.23)
jωe
208 Chapter 8. Vector Control of PMSM Drives

0.1

Speed (pu)
0
ωr ω1
−0.1

0 5 10 15
1
Current (pu)

0.5 iq
0
id
−0.5
0 5 10 15
Error angle (degrees)

50

−50

−100
0 5 10 15
t (s)

Figure 8.8. Ramp-wise rotation reversals under load of a nonsalient sensorless PMSM drive.

the approximation since {ωe Ld , ωe Lq } ≫ Rs . We have

⎡ 1 ⎤
0 ( ) ( )
⎢ ⎥ Lq + Ld 1 0 Lq − Ld 1 0
L−1 = ⎣ Ld 1 ⎦ = + (8.24)
0 2Lq Ld 0 1 2Lq Ld 0 −1
Lq

which yields
( )
Lq + Ld Lq − Ld J θ̃ 1 0
eJ θ̃ L−1 e−J θ̃ = I+ e e−J θ̃
2Lq Ld 2Lq Ld 0 −1
W I
Lq + Ld Lq − Ld cos 2θ< sin 2θ<
= I+ . (8.25)
2Lq Ld 2Lq Ld sin 2θ< − cos 2θ<

Thus, from d voltage to q current we have the frequency function

( )
: ;
< = 0 1 G(jωe , θ)
Gdq (jωe , θ) < 1 = 1 Lq − Ld sin 2θ< (8.26)
0 jωe 2Lq Ld

yielding the measured current

< jωe t } + iq0 = V Lq − Ld sin 2θ< sin ωe t + iq0 .

iq = V Re{Gdq (jωe , θ)e (8.27)
ωe 2Lq Ld
8.4. Vector Current Control of Salient PMSMs 209

vd d
q d

Figure 8.9. Estimation of the position using saliency and signal injection.

This signal is an amplitude modulation of the carrier sin ωe t by the envelope sin 2θ, < to which the
dc component iq0 is added, which acts as a bias to the modulated signal. To remove this bias, the
measured current is first fed through a high-pass filter (HPF). A simple first-order filter is perfectly
adequate
s
HHP (s) = ⇒ ifq = HHP (p)iq . (8.28)
s + αh
The cut-off frequency αh of this filter should be selected low, typically in the range of αh /2π = 1
Hz ⇔ αh ≈ 6 rad/s. To appreciate this, note that iq0 is only dc in the steady state; due to
closed-loop current control it follows its reference iref
q , which varies with the dynamics of the speed
control loop (closed-loop speed control assumed). Hence, the cut-off frequency of the high-pass
filter should preferably be selected no lower than the speed-control-loop bandwidth αs , so that the
dynamics of iq0 fall within the stopband of the high-pass filter. With the mentioned αh = 6 rad/s,
a speed rise time of 2.2/αs = 2.2/αh ≈ 0.4 s is allowed for, showing that the cut-off frequency has
to be selected higher than 1 Hz only for very fast servo drives.
To demodulate, the filtered signal is multiplied by sin ωe t, giving
V Lq − Ld
ε = ifq sin ωe t = sin 2θ< sin2 ωe t
ωe 2Lq Ld
V Lq − Ld V Lq − Ld
= sin 2θ< − sin 2θ< cos 2ωe t . (8.29)
ωe 4Lq Ld ωe 4Lq Ld
. /0 1 . /0 1
(∗) (∗∗)
In this function the term (∗) carries the vital position error information. The other term, (∗∗),
acts as high-frequency disturbance. This disturbance is not significant, however, as it disappears
as θ< becomes small. The PLL can be used for speed and position estimation, using Ed →= −ε.

8.4 Vector Current Control of Salient PMSMs

In Chapter 6, synchronous-frame current controller design based on complex space vectors is con-
sidered. For this reason, the obtained controller is not immediately applicable to salient PMSMs.
210 Chapter 8. Vector Control of PMSM Drives

Extending the theory to such motors is straightforward, however. The dynamic model for the
salient PMSM, (5.118)–(5.119), can be expressed in real vector form as

⎡ ⎤
did ( ) ( )( ) ( )
Ld vd Rs −ωr Lq id Edb
⎢ dt ⎥
⎣ = − − . (8.30)
diq ⎦ vq ωr L d Rs iq Eqb
Lq
dt

Decoupling and an “active resistance” are introduced by letting

( ) ( ) W I( )
vd vd′ Rad !q
−ω1 L id
= − (8.31)
vq vq′ !d
ω1 L Raq iq

ideally giving
⎡ ⎤
did ( ′ ) ( )( ) ( )
Ld v Rs + Rad 0 id Edb
⎢ dt ⎥
⎣ ⎦ = vd′ − − . (8.32)
diq q 0 Rs + Raq iq Eqb
Lq
dt

The two inner feedback loops are now decoupled. By selecting

!d − R
Rad = αc L !s !q − R
Raq = αc L !s (8.33)

both loops get bandwidth αc , ideally. In operator notation

1 1
id = v′ iq = v′ . (8.34)
(p + αc )Ld d (p + αc )Lq q

The outer loops are then closed with two PI controllers. Applying direct synthesis yields

! d (p + αc )
αc L ! q (p + αc )
αc L
vd′ = (iref
d − id ) vq′ = (iref
q − iq ). (8.35)
p p

That is,

!d
kpd = αc L !d
kid = α2c L (8.36)
!q
kpq = αc L !q .
kiq = α2c L (8.37)

Incorporating “back calculation” yields Algorithm 8.3.

Algorithm 8.3
8.5. Field-Weakening Control 211

Vector Current Control of Salient PMSMs

( ) ( )( )
id cos(θ1 ) sin(θ1 ) iα
=
iq − sin(θ1 ) cos(θ1 ) iβ
( ) ( ref ) ( )
ed id id
= −
eq iref
q iq
( ref ) ( ) ( ) W I( ) ( )
vd kpd ed kid Id Rad !q
ω1 L id 0
= + − + !q
vqref kpq eq kiq Iq !d
−ω1 L Raq iq E
θ1′ = θ1 + ω1 Td
( ref ) ( ) ( ref )
vα cos(θ1′ ) − sin(θ1′ ) vd
=
vβref sin(θ1′ ) cos(θ1′ ) vqref
[v̄α , v̄β ] = PWM(vα , vβ )
( ref ) ( )( )
v̄d cos(θ1′ ) sin(θ1′ ) v̄αref
=
ref
v̄q − sin(θ1′ ) cos(θ1′ ) v̄βref
( ) ( ) W Z ref [ I
1
Id Id ed + kpd v̄d − vdref
= + Ts 1
Z ref [
Iq Iq eq + kpq v̄q − vqref

8.5 Field-Weakening Control

For operation above base speed the flux must be reduced, since the stator voltage then has reached
its maximum. This is made by manipulating the reference iref d for the flux-producing current
component. One convenient strategy for this manipulation is to introduce an integrating field-
weakening controller
$ \
: 2 ; \Inom
iref
d = kfw Vbase − (vd ) − (vq ) dt \\
ref 2 ref 2
(8.38)
Imin

where vdref and vqref are the ideal voltage references from the current controller (vector vref s =
ref ref ref ref ref
vd + jvq , see Algorithm 6.1), before the PWM algorithm is applied (giving v̄s = v̄d + jv̄q ).
The algorithm described by (8.38) works as follows. When |vref
s | > Vbase , i.e., the current controller
wants to put out a voltage that exceeds the base voltage, iref
d is decreased from its nominal value
Inom , but not below the minimum value Imin . When |vref s | < Vbase , id
ref
is increased, but not
above the nominal value. This way, the flux is reduced precisely as much as required to achieve
|vs | = Vbase in the field-weakening region. Saturation is avoided in the steady state, cf. Figure
3.15. For round-rotor PMSMs Inom = 0, normally (giving id = 0 below base speed), whereas for
salient PMSMs (5.123) can be applied to obtain maximum torque by letting

(Ld − Lq )Ibase ref

Inom = r(χ)iref
q ≈ iq . (8.39)
ψ!R

The minimum shall be selected as

−Ibase < Imin < 0 (8.40)

212 Chapter 8. Vector Control of PMSM Drives

because if Imin = −Ibase , then id = Imin implies that iq = 0 to prevent overcurrent, meaning
that zero torque is produced. A block diagram of the suggested controller is depicted in Figure
8.10. Notice that the nominal iref q,nom must be saturated as follows to avoid overcurrent in the
field-weakening range
, R -
iref
q = sat i ref
q,nom , I 2
max − (i ref )2
d (8.41)

where Imax shall be reduced to Ibase after the maximum time that overcurrent is allowed.

2
Vbase
2 vsref
Inom is
kfw/s
Imin idref vsref
ref CC PWM M
iq,nom
sat
iqref
Ibase
Figure 8.10. Field-weakening controller for PMSM drives, where “sat” is implemented according to
(8.41). “CC” is the current controller with “back calculation.” (To avoid clutter, stationary/synchronous
coordinate transformations are not shown.)

8.5.1 Gain Selection

Selecting the gain kfw in (8.38) requires analysis of the dynamics of id given by (8.38). These are
nonlinear, because of the quadratic terms in the input signal to the integrator, i.e., the integrand
in (8.38). Let us denote the integrand as
2
f (id ) = Vbase − (vdref )2 − (vqref )2 . (8.42)

In the steady state, with perfect field orientation, we have [cf. (5.58)]

vd = −ω1 Lq iq vq = ω1 (Ld id + ψR ) (8.43)

with the voltage drop across the stator resistance neglected. Substitution in (8.42) yields
2
f (id ) = Vbase − (ω1 Lq iq )2 − [ω1 (Ld id + ψR )]2 (8.44)

which can be linearized around an operating point i⋆d as

df (id )
f (id ) ≈ f (i⋆d ) + (id − i⋆d ) = −2ω1 Ld (Ld i⋆d + ψR )(id − i⋆d ) (8.45)
. /0 1 did . /0 1
0 vq⋆

As id is made negative during field-weakening operation, |iq | must be reduced to prevent overcur-
2
rent. Consequently, it is not unreasonable to approximate Vbase = vd2 + vq2 ≈ vq2 , which in turn
8.5. Field-Weakening Control 213

yields vq⋆ ≈ Vbase sgn ω1 and

df (id )
≈ −2|ω1 |Ld Vbase . (8.46)
did
This allows (8.38) to be linearized as

diref
d
= − 2kfw |ω1 |Ld Vbase (id − i⋆d ). (8.47)
dt . /0 1
αfw

Assuming that the field-weakening control is made significantly slower than the current control,
then id = iref
d can be assumed in (8.45), which shows that the closed-loop dynamics of the field-
weakening control has bandwidth αfw . Obviously, αfw should be chosen much lower than the
current-control-loop bandwidth for the assumption of time-scale separation with the current control
loop to hold, but it should preferably be chosen larger than the speed-control-loop bandwidth, i.e.,

αs < αfw ≪ αc . (8.48)

This results in the following recommendation for gain selection:

αfw
kfw = . (8.49)
!
2LdVbase max(|ω1 |, ωbase )

A constant gain is used below base speed, whereas kfw is decreased as 1/|ω1 | above base speed to
maintain the closed-loop bandwidth αfw . Algorithm 8.4 is obtained.

Algorithm 8.4
Field-Weakening Controller for PMSM Drives
Ts αfw : 2 ;
iref ref
d = id + Vbase − (vdref )2 − (vqref )2
! dVbase max(|ω1 |, ωbase )
2L
iref
d = min[max(iref
d , Imin ), Inom ]
, R -
iref
q = sat iref
q,nom , I 2
max − (i ref )2
d
214 Chapter 8. Vector Control of PMSM Drives

.
CHAPTER 9

Vector Control of IM Drives

It is well known that the flux of an IM cannot readily be measured. Consequently, a flux estimator
of some kind is fundamental to any IM vector control system. There are two archetypical flux
estimators: the current model, which is based on the rotor circuit of the IM, and the voltage
model, which is based on the stator circuit. Both estimator types can be implemented either in
the αβ frame or in the dq frame. The resulting vector control systems types are known as direct
field orientation (DFO) and indirect field orientation (IFO), respectively. We shall in this chapter
consider both the mentioned flux estimators, as well as variants of them. We shall also consider
both DFO and IFO, though our preference is IFO. The reason for this preference is the same as
the preference for current control in the dq frame rather than the αβ frame: the quantities are
constant rather than constantly oscillating in the steady state.
Vector control of IM drives is both similar to and different from vector control of PMSM drives.
The main difference is as follows.
Due to the asynchronous rotation of the IM, measurement of the rotor
position—unlike for the PMSM—does not reveal the rotor flux angle.

This complicates the flux estimator, but yet, if the rotor speed—or just the rotor speed—is mea-
sured it is a great help, because it makes estimating the flux angle accurately a lot easier, particu-
larly at low speeds. This will be shown in Section 9.1. Speed-sensored control of IM drives is more
complicated than position-sensored control of PMSM drives, but not tremendously so.
For cost and reliability reasons, speed sensors for IMs should preferably be avoided. The flux
angle then has to be estimated without the help of speed measurement. In the low-speed range this
greatly reduces the stability robustness of the drive, as we shall find in Section 9.2. Concerning
sensorless control there are two significant differences between the PMSM and the IM.

• Unlike the PMSM, the initial rotor flux angle θ̃(0) of the IM is known when the drive is started.
This is because the rotor flux is created from the stator by controlling the d-direction current
component to a constant value before a torque-producing (q-direction) current component is
applied. For this reason, synchronization at startup is not a problem, even without a speed
sensor.

• On the other hand, also because the rotor flux is created from the stator, the flux is dynamic.
Loss of accuracy in the field orientation may result in undesired demagnetization and loss of
torque. For this reason, stable rotation reversals under load are quite difficult to accomplish
with sensorless IM drives.

A discussion of field-weakening control is made in Section 9.3. The chapter is finished by a presen-
tation of the well-known scheme direct torque control (DTC), which can be found in Section 9.4.
This is a control method that effectively merges field orientation, current control, and PWM.
216 Chapter 9. Vector Control of IM Drives

9.1 Speed-Sensored Control

9.1.1 Direct Field Orientation Using the Current Model
By simulating the rotor circuit of the IM, given by (5.36), the following flux estimator is obtained:
> ?
! sR
dψ !R
R
! s
= RR is − − jωr ψ ! sR . (9.1)
dt !M
L

This estimator is called the current model. An implementation of the current model as it stands,
i.e., in the αβ frame, yields a DFO vector control system. Even though a flux estimate in Cartesian
! sR = ψ!R ejθ1 . This reveals
αβ coordinates is obtained, it is instructive to express it in polar form as ψ
±jθ1
that the αβ and dq transformation factors e can be obtained by normalizing the flux estimate
as
! sR
ψ ψ!α + j ψ!β ! sR )∗
(ψ ψ!α − j ψ!β
= R = ejθ1 = R = e−jθ1 . (9.2)
! sR |
|ψ !2 !2 ! sR |
|ψ ! 2 !2
ψα + ψβ ψα + ψβ

That is, cosine and sine of the transformation angle are calculated implicitly as

ψ!α ψ!β
cos θ1 = R sin θ1 = R . (9.3)
ψ!α2 + ψ!β2 ψ!α2 + ψ!β2

Avoiding trigonometric operations was imperative before powerful microprocessors became avail-
able, which allowed digital implementation of control algorithms. For implementation in analog
electronics, DFO thus was preferable, even though the square-root and division operations in (9.3)
yet posed difficulties that had to be circumvented. DFO tended to be more popular than IFO well
into the age of digital implementation, perhaps for pure historical reasons.

9.1.2 Properties of the Current Model

By inspecting (9.1), the following drawbacks of the current model are revealed.

• The current model is a speed-sensored flux estimator, because ωr is used in the estimator
equations.
!R and L
• The current model is sensitive to R ! M . Detuned (erroneous) model parameters affect
the estimated flux angle, giving imperfect field orientation. (This effect will be quantified
in the analysis made in the following.) The magnetizing inductance and (particularly) the
rotor resistance both change during operation, as discussed in Paragraph 5.6.3.

Outweighing these drawbacks is one strong benefit: the current model is guaranteed always to give
robustly stable operation in the low-speed region. See further Paragraph 9.1.4.

9.1.3 Indirect Field Orientation Using the Current Model

By assuming ψR real—i.e., perfect field orientation—the flux equation (5.40) reduces to a simpler
form, that can be rearranged as the ideal flux-modulus dynamics (5.45) and the slip relation
9.1. Speed-Sensored Control 217

(5.46). The current model can be transformed in an identical manner. First, let us transform (9.1)
to synchronous coordinates
> ?
dψ!R !R
R
=R !R is − + jω2 ψ !R (9.4)
dt !M
L
! R = ψ!R (as the field orientation of the flux estimator itself is
and let the flux estimate be real: ψ
always perfect). By separating the real and imaginary parts, we obtain
dψ!R !R
R
! R id −
= R ψ! (9.5)
dt !M R
L
! R iq
R
ω2 = ω1 − ωr = . (9.6)
ψ!R
Equation (9.6) is usually referred to simply as the slip relation, but later we shall generalize it and
therefore it will be named as the standard slip relation. The angular stator frequency is computed
using (9.6) and is then integrated into the transformation angle
!R iref
R q
ω1 = ωr + , θ̇1 = ω1 (9.7)
ψ!R
where reference iref
q is used rather than the measured component in order to reduce noise. Similarly,
iref
d should be used in (9.5), giving

dψ!R !R
R
!R iref
=R d − ψ!R . (9.8)
dt !
LM
This scheme is a direct IFO correspondence to the DFO variant of the current model (9.1). IFO is
generally preferable to DFO, since all quantities (except θ1 ) are constant in the steady state and ω1
is an explicit variable in the control algorithm. Figure 9.1 shows a block diagram of an IFO vector
control system. The flux estimator, whose output is ω1 , may be implemented using the equations
for the current model—see Algorithm 9.1—or a variant thereof. If (9.7)–(9.8) are employed, then
the signal path for v̄ref into the flux-estimator block is unused.
Algorithm 9.1
Current Model
!R iref
R q
ω1 = ωr +
!
ψR
θ1 = mod(θ1 + Ts ω1 , 2π)
> ?
!R
ψ
! ! !R i −
ψR = ψR + Ts R ref
d
!M
L

Remark 9.1 If ψref is constant and, from (5.48), iref ! !

d = ψref /LM , then it is seen in (9.5) that ψR
converges to ψref . Hence, for a constant ψref , i.e., if field weakening is not used, (9.5) need not be
implemented; ψ!R in (9.7) can be replaced by ψref .
218 Chapter 9. Vector Control of IM Drives

vref s
vref
dq 2
αβ 3
inom
ref i Current vref s
vref
sat ref controller
dq 2
αβ 3 PWM

Flux ω1 θ1 ωr
estimator
i dq is 2
αβ 3

Figure 9.1. IFO vector control system for IM drives.

9.1.4 Analysis of the Current Model

It is straightforward to show that the current model (9.5)–(9.6) with id = ψref /L! M gives perfect
! !
field orientation, assuming accurate model parameters. With LM = LM and RR = RR , (9.5) is
subtracted from (5.40) and ψ< R = ψR − ψ!R is introduced, giving

R
<˙ R = RR iq − R ψ
ψ < R − ψ!R ).
< R − jω2 (ψ (9.9)
LM

Substituting (9.6) in (9.9) yields the following simplification of (9.9):

, -
RR RR iq
<˙ R = −
ψ +j <R
ψ (9.10)
LM ψref

which has the solution

, $ t -
RR iq
< <
ψR (t) = ψR (0) exp − t + jRR dτ . (9.11)
LM 0 ψref

< R | = |ψ
Hence, |ψ < R (0)|e−RR t/LM , i.e., ψR converges exponentially to ψ!R with the rotor time
constant LM /RR . In turn, ψ!R converges to ψref because of the open-loop simulation (9.5). So,
convergence of ψR to the desired ψref is guaranteed for all operating conditions, including varying
iq and ψref . This is a unique and very strong stability property of the current model.
Note, however, the assumption of correct model parameters. Although is has been shown that
the current model is quite tolerant to erroneous model parameters with regard to stability [16],
detuning of the model parameters R !R and L ! M will lead to erroneous field orientation, giving
degraded performance. The steady-state rotor flux resulting from detuned model parameters is
9.1. Speed-Sensored Control 219

easy to calculate from (5.40)

RR is !
ψR = = {(9.6), id = ψref /L ! M } = RR (ψref /LM + jiq )
RR /LM + jω2 RR /LM + j R !R iq /ψref
> ?2
!M R
RR L ! R LM iq ! R LM
R
1+ 1−
LM ψref !
RR LM R R ψref RR L!M
= > ?2 +j LM iq > ?2 . (9.12)
!
LM ! R LM iq ! R LM iq
R R
1+ 1+
RR ψref RR ψref
. /0 1 . /0 1
ψd ψq
Perfect field orientation, ψq = 0, is thus obtained when

!M
L LM
= . (9.13)
!
RR RR

Otherwise, the field orientation will be inaccurate, and the inaccuracy increases with iq , i.e., with
the torque. From Paragraph 5.6.3 it is known that the rotor resistance tends to vary more than
the magnetizing inductance, so R !R is the primary critical parameter. Let us also calculate the
steady-state electrical torque. Using (9.12) we get
> ?2
! M iq
L
1+
3np 3np R!R L2 ψref iq ψref
M
τe = Im{ψ∗R is } = > ?2 . (9.14)
2K 2 2K 2 !2
RR L ! R LM iq
M R
1+
RR ψref

The ideal linear relation between τe and iq is obtained only when (9.13) holds. Figure 9.2 illustrates
(9.14) for ψref = 0.9 pu, LM = L! M = 2 pu, RR = 0.05 pu, and three values of R !R . If the resistance
parameter values are obtained from “cold” measurements, then R !R is normally an underestimate
of RR , which reduces the torque for higher |iq |. Slower speed response than for perfect field
orientation can be expected for larger speed-reference steps, when maximum current is applied.
PROBLEM 9.1
Why is a larger-than-nominal electrical torque obtained for iq = 1 pu?

Example 9.1 Equation (9.14) suggests that R !R should be obtained for a “hot” motor, giving an
!
overestimated RR for many operating conditions, and consequently higher torque, according to
Figure 9.2. Let us investigate by simulation whether overestimating rather than underestimating
the rotor resistance gives better performance.
A two-pole IM drive which uses the current model for flux estimation, ωbase = 100π rad/s, and
the per-unit parameters ψref = 0.9, LM = 2, L! M = 1.8, Lσ = 0.2, Rs = RR = 0.05, J = 1000, and
b = 0.1 is considered. The speed control loop is tuned for a rise time of 0.35 s and the load torque
is approximately 50% of the nominal. In Figure 9.3, the reference speed is first changed rampwise
around ωr = 0; it is thereafter at t = 4 s changed in a step to 1.5 pu in order to investigate the
220 Chapter 9. Vector Control of IM Drives

1.5

0.5

N
e e
τ /τ
0

−0.5

−1

−1.5
−1 −0.5 0 0.5 1
i (pu)
q

Figure 9.2. Normalized electrical torque as function of iq when the current model is used for flux
!R = 0.05 pu—correct value, (dashed) R
estimation and (solid) R !R = 0.03 pu—underestimated, and (dash-
dotted) R!R = 0.07 pu—overestimated.

performance in the field-weakening region (during which the controller to be designed in Section
9.3 has already been incorporated), and is at t = 6 s decreased to 0.5 pu. It can be observed that
the performance is good, even when R !R is detuned, except for the large reference step at t = 4
s when maximum current is used. As underestimating R !R gives lower torque, the dashed speed
curve shows a slower response. The response is also slower when R !R is overestimated, however,
because there is a transient reduction in ψd which was not predicted by the steady-state theory.
! M hardly affects the field orientation at all: ψq ≈ 0 as long
Note further that the slightly detuned L
!
as RR = RR . Clearly, good performance requires that R !R is reasonably accurate, not only that it
is overestimated.

9.1.5 Improved Flux Estimation

The current model dates back to the pioneering days of vector control of the late 1960s and early
1970s. Its sensitivity to the rotor resistance was found at an early stage, but it took ten more
years until an effective solution to the problem was presented. In 1980, Garcés proposed on-
line adaptation of the model rotor resistance, using an error function based on the instantaneous
reactive power [23]. Analog implementation was suggested, because the microprocessors available
at that time were not fast enough to allow digital implementation. During the following two
decades, a plethora of more or less similar adaptation schemes surfaced.
A few years after the proposal for rotor-resistance estimation, i.e., in the mid-1980s, alternatives
to the current model began to appear in publications. All these alternatives involve the stator
equation (5.35) [in addition the rotor equation (5.36), which forms the basis for the current model].
The stator equation can be rearranged as

diss dψsR
vss = Rs iss + Lσ + Es , Es = . (9.15)
dt dt
9.1. Speed-Sensored Control 221

1.5

ω (pu)
0.5
r
0
0 1 2 3 4 5 6 7 8

1
(pu)

0.5
d
ψ

0
q
ψ

−0.5
0 1 2 3 4 5 6 7 8

0.5
i (pu)

0
q

−0.5

−1
0 1 2 3 4 5 6 7 8
t (ms)

!R =
Figure 9.3. Simulation of an IM drive using the current model for flux estimation and (solid) R
! !
0.05 pu—correct value, (dashed) RR = 0.03 pu—underestimated, and (dash-dotted) RR = 0.07 pu—
overestimated.

A flux-emf estimate can be obtained by subtracting from the stator voltage estimates of the voltage
drops across the stator resistance and leakage inductance
s
! s = vs − R
E ! σ dis .
! s is − L (9.16)
s s
dt

This estimate can be used in various ways to improve the flux-estimation accuracy. Direct inte-
gration of E! s is one straightforward option, which yields the so-called voltage model, see further
Section 9.2. Many of the proposed alternative flux-estimation methods—including those analyzed
in the pioneering work by Verghese and Sanders [84]—incorporate (9.15) in an observer. This is
a simulation of a state-space system—in our case the IM electrical dynamics—augmented with
feedback of the error between the actual output and that of the model. The error is scaled with
the observer gain. This has the effect of driving unmeasurable state variables (in this case, the flux
components) as close as possible to their correct values. By appropriate selection of the observer
gain, an estimator that is less sensitive to the rotor resistance than the current model can be
designed. This is conceptually more straightforward than on-line adaptation of the model rotor
resistance, because the idea is simply to use a better estimator rather than manipulating the pa-
rameters of the current model. Several observer designs boil down to a flux estimator that makes
a seamless transition from the current model to the voltage model (or a variant of thereof) as the
speed rises, e.g., [44]. Most proposed observers are DFO flux estimation schemes.
222 Chapter 9. Vector Control of IM Drives

To obtain an improved IFO flux estimator, (9.16) is transformed to synchronous coordinates

! = vs − (R
E !s + jω1 L ! σ dis .
! σ )is − L (9.17)
dt
!s = Rs and L
If the model parameters were accurate, i.e., R ! σ = Lσ , then E
! = E, where
dψR
E = jω1 ψR + (9.18)
dt
which can be expressed in component form as
Ed = ψ̇d − ω1 ψq (9.19)
Eq = ψ̇q + ω1 ψd . (9.20)
Alternatively, the rotor flux can be expressed in polar form as ψsR = ψR ejθ , giving
dψsR
E = e−jθ1 Es = e−jθ1 = (ψ̇R + j θ̇ψR )ej θ̃ , θ< = θ − θ1 (9.21)
dt
and the following expressions for the components:
Ed = ψ̇R cos θ< − θ̇ψR sin θ< (9.22)
Eq = ψ̇R sin θ< + θ̇ψR cos θ.
< (9.23)
At higher speeds, |ψ̇R | ≪ |θ̇ψR | can be assumed (because the flux modulus varies with the rotor
time constant, i.e., relatively slowly), which allows the approximations
Ed ≈ −θ̇ψR sin θ< (9.24)
<
Eq ≈ θ̇ψR cos θ. (9.25)
Particularly it can be noted that perfect field orientation yields Ed ≈ 0. This property is funda-
mental for the design of an improved IFO flux estimator.

9.1.6 Modified Current Model

In Section 6.5, theory for synchronization control using a PLL was presented. The PLL input
signal that drives the dq-frame angle θ1 to its correct value is −Ed , see Figure 6.10. A comparison
shows that (9.24)–(9.25) are identical to (6.36) with E = θ̇ψR . This indicates that a PLL can be
used to enhance the accuracy of the flux estimate in the vector control system for an IM drive.
Let us revisit (6.38), which determines how ω1 is calculated by the PLL. The equation is repeated
here for clarity, with ωi = 0, i.e., without integral action in the PLL
αl
ω1 = (!ωg − kpp Ed ). (9.26)
p + αl
In an IM drive, neither the flux, nor the flux emf, can be measured, but we can use the estimated
flux-emf component E !d obtained by taking the real part of (9.17) in lieu of Ed . In addition, ω
!g has
to be replaced by an estimate of the angular flux frequency, and such we have: it is readily given
by the standard slip relation as ω ! g = ωr + R!R iq /ψ!R . This results in a modified current model
(MCM)
> ?
αl R! R iq
ω1 = ωr + !d .
− kpp E (9.27)
p + αl ψ!R
9.1. Speed-Sensored Control 223

which, in effect, is the current model embedded in a PLL. It is convenient to further modify (9.27)
by introducing kpp = λs /ψ!R , which allows a dimensionless gain λs . We also use the reference iref
q
instead of the measured current component
> ?
αl !R iref
R !
q − λs Ed
ω1 = ωr + . (9.28)
p + αl ψ!R

For λs = 0, (9.28) reverts to the standard slip relation (9.7), albeit with a low-pass filter added.
Equation (9.28) may therefore be called a generalized slip relation.
Often, because the current control loop is much faster than the flux dynamics, L! σ dis /dt can be
neglected in (9.17). In addition, reference voltage and current components should be used rather
than the corresponding measured components (cf. Figure 9.1), giving
!d = v̄ ref − R
E !s iref + ω1 L
! σ iref . (9.29)
d d q

!d . The low-pass filter in (9.28) gives further noise

These modifications reduce the noise level of E
reduction. Selecting the filter bandwidth αl may require some trial-and-error work to obtain an
adequate compromise between good noise reduction and small enough lag. Suitable selections are
within the range
RR
≪ αl < αc (9.30)
LM
i.e., much higher than the inverse rotor time constant, but lower than the current-control-loop
bandwidth.

9.1.7 Analysis and Parameter Selection of the Modified Current Model

Remaining to do is to select gain λs and to evaluate the performance of the modified current model.
If the modification is successful, then a lower parameter sensitivity than for the standard current
model should result. Finding an adequate gain selection requires some analysis. To this end, let
us express the flux dynamics (5.40) in component form
RR
ψ̇d = RR id − ψd + (ω1 − ωr )ψq (9.31)
LM
RR
ψ̇q = RR iq − ψq − (ω1 − ωr )ψd . (9.32)
LM
Accurate model parameters are assumed, as otherwise the analysis would get far too complicated.
Under this assumption we have E !d = Ed , which is given by (9.19) as Ed = −ω1 ψq + ψ̇d . Using
(9.31) and id = ψref /LM , (9.19) can equivalently be expressed as
RR
Ed = (ψref − ψd ) − ωr ψq . (9.33)
LM
Equation (9.33) is then substituted in (9.28). Owing to the bandwidth selection (9.30), the low-
pass filter in (9.28) can be considered in the steady state as observed from the flux dynamics,
giving
( )
RR iq λs RR
ω1 = ωr + − (ψref − ψd ) − ωr ψq (9.34)
ψref ψref LM
224 Chapter 9. Vector Control of IM Drives

where we also have substituted ψ!R → ψref . This substitution is adequate for analysis purposes,
since ψ!R is given by the open-loop simulation (9.5) and thus always converges to ψref . In (9.34)
it can be observed that, unlike the standard current model, ω1 is now not only a function of ωr
and iq , but also of the actual flux components. A nonlinear feedback of the flux is created via
the integration ω1 into θ1 and the usage of the latter in the coordinate transformations. This
alters the flux dynamics from those obtained when the standard current model is used, i.e., (9.10).
Substituting (9.34) in (9.31)–(9.32), it is found that this nonlinear feedback creates a nonlinear
system in the state variables ψd and ψq
( , -)
RR RR iq λs RR
ψ̇d = (ψref − ψd ) + − (ψref − ψd ) − ωr ψq ψq (9.35)
LM ψref ψref LM
( , -)
RR RR iq λs RR
ψ̇q = RR iq − ψq − − (ψref − ψd ) − ωr ψq ψd . (9.36)
LM ψref ψref LM

This system has an equilibrium point at [ψd = ψref , ψq = 0], around which the system is now
linearized. Introducing the error variable ψ<d = ψd −ψref , the linearized state-space system becomes
⎡ ⎤ ⎡ ⎤⎡ ⎤
˙ RR RR iq
ψ<d − ψ<
⎢ ⎥ ⎢ LM ψref ⎥ ⎢ d⎥
⎢ ⎥=⎢ ⎥⎢ ⎥. (9.37)
⎣ ⎦ ⎣ λs RR RR iq RR ⎦⎣ ⎦
ψ̇q − − − − λs ωr ψq
LM ψref LM
. /0 1
A
The characteristic polynomial is given by
, - , -2 , -2
2RR RR RR RR iq
det(sI − A) = s2 + λs ωr + s + λs ω1 + + . (9.38)
LM LM LM ψref

where ω1 = ωr + RR iq /ψref . At higher speeds, ω1 and ωr have equal signs. If λs is given the same
sign as ωr , then all coefficients are positive and the system is asymptotically stable. Care must
be taken to ensure that the coefficients are positive also at low speeds, though, when the signs of
ω1 and ωr may differ. For this reason, it is recommended that λs should be ramped up with the
speed as (see Figure 9.4)
⎧
⎨ λωr , |ωr | ≤ ω∆
λs = sat (λωr /ω∆ , λ) = ω∆ (9.39)
⎩
λ sgn ωr , |ωr | > ω∆

where reasonable parameter values are

1≤λ≤2 0.05ωbase ≤ ω∆ ≤ 0.2ωbase . (9.40)

In the low-speed range the properties of the standard current model thus are preserved—including
its robust stability, but also its high rotor-resistance sensitivity. In the higher-speed range, |ωr | >
ω∆ , (9.38) can be approximated by considering RR and ω2 small, giving
, -
2 RR RR
det(sI − A) ≈ s + λ|ωr |s + λ|ωr | ≈ (s + λ|ωr |) s + . (9.41)
LM LM
9.1. Speed-Sensored Control 225

There are two poles widely separated on the negative real axis. The “slow” pole at s = −RR /LM
corresponds mainly to the dynamics of ψd (and are equal to the open-loop flux dynamic obtained
using the standard current model), whereas the “fast” pole at s = −λ|ωr | corresponds mainly to
the dynamics of ψq . Gain λs has the effect of “speeding up” ψq , but not ψd . This can be regarded
as a speedup of the field orientation, i.e., the adjustment of θ1 , whereas the flux-modulus dynamics
by the physical characteristics of the IM remain trapped at the rotor time constant. This is a
expected result, since the incorporation of the current model in a PLL logically should enhance the
synchronization properties. Algorithm 9.2 illustrates an implementation of the modified current
model with the proposed selection of λs .

Algorithm 9.2
Modified Current Model
!d = v̄ ref − R
E !s iref + ω1 L
! σ iref
d d q
λs = sat (λωr /ω∆ , λ)
> ?
!R iref
R !
q − λs Ed
ω1 = ω1 + Ts αl ωr + − ω1
ψ!R
θ1 = mod(θ1 + Ts ω1 , 2π)
> ?
! ! ! ref ψ!R
ψR = ψR + Ts RR id −
!M
L

λs = sat ( λωr /ω∆ ,λ)

λ

ωr
ω∆ ω∆

Figure 9.4. Selection of gain λs .

9.1.8 Parameter Sensitivity of the Modified Current Model

In the low-speed region the modified current model has properties that are similar to those of the
standard current model, due to gain selection (9.39). The sensitive parameters are thus R !R and
!
LM . When analyzing the impact of erroneous model parameters at higher speeds, i.e., |ωr | ≥ ω∆ ,
226 Chapter 9. Vector Control of IM Drives

the voltage drop across the stator resistance can be neglected. With ψ̇d neglected in (9.19), this
!d ≈ −ω1 (L
yields E < σ iq + ψq ), where L
< σ = Lσ − L
! σ , and

! R iq
R λ|ω1 | <
ω1 = ωr + + (Lσ iq + ψq ). (9.42)
ψref ψref
Substitution of (9.42) in (9.32) yields
W I
RR ! R iq
R λ|ω1 | <
ψ̇q = RR iq − ψq − + (Lσ iq + ψq ) ψd (9.43)
LM ψref ψref

whose steady-state properties are considered, i.e., ψ̇q = 0. Assuming that ψd = ψref and RR /LM ≪
λ|ω1 |, solving for ψq in (9.43) yields
> ?
R<R
ψq ≈ − Lσ +< iq . (9.44)
λ|ω1 |

If λ is sufficiently large, the error due to R<R can be neglected. Thus, we find that the leakage
inductance is the only sensitive parameter of the modified current model at higher speeds, given
that gain λs is properly selected. Since the leakage inductance tends to vary much less than the
rotor resistance, it can be concluded that the objective for the modification of the current model
has been fulfilled.

Example 9.2 Let us see whether improved performance as compared to Example 9.1 (Figure 9.3)
is obtained by the modified current model. We again simulate a two-pole IM with the per-unit
parameters J = 1000, b = 0.1, LM = 2, Lσ = 0.2, and Rs = RR = 0.05. The speed control loop
is tuned for a rise time of 0.35 s and the load torque is approximately 50% of the nominal. The
per-unit model parameters are L ! M = 1.8, L! σ = 0.22, R !s = 0.03, and R
!R = 0.025. In the gain
selection, ω∆ = 0.2 pu and λ = 1.
As in Example 9.1, the speed reference is first changed rampwise, giving slow positive-to-
negative and negative-to-positive rotation reversals. At t = 4 s, the speed reference is changed to
1.5 pu, and is finally at t = 6 set to 0.5 pu. Figure 9.5 shows the results. Whereas at low speeds the
field orientation suffers due to the detuning of the model rotor resistance, once the speed increases
accurate field orientation is obtained. This allows faster speed response for the step to 1.5 pu than
the standard current model for an erroneous R !R , cf. Figure 9.3.

9.2 Speed-Sensorless Control

The difficulty of controlling an IM drive increases tremendously if the rotor speed cannot be
measured, i.e., the drive is speed sensorless. The challenge of speed-sensorless control of an IM
drive lies not in estimation of the rotor speed per se, but rather in accurately estimating the rotor
flux without knowledge of the speed. Neither the standard current model nor the modified current
model can be employed without modification, as they use ωr in their equations.
One such modification is to use a speed estimate ω !r in place of ωr in the (generalized) slip
relation. This speed estimate can be obtained in at least three different ways.
9.2. Speed-Sensorless Control 227

ω (−) ω (−−) (pu)

1.5

1
0.5

0
r

0 1 2 3 4 5 6 7 8

1.5
ψ (−)

1
d

0.5
ψ (−)
q

−0.5
0 1 2 3 4 5 6 7 8

1
i (−−) (pu)

0.5

0
d
i (−)

−0.5
q

−1
0 1 2 3 4 5 6 7 8
t (s)

Figure 9.5. Simulations of an IM drive using the modified current model.

1) Extraction of rotor-slot harmonics from the stator current [43].

2) Extraction of information obtained from naturally occurring or artificially created rotor

saliencies [45].

3) Extraction of information available in the fundamental-wave excitation.

The first two methods are very reliable, provided that they can be used. But rotor-slot harmonics
tend to vanish for IMs that have skewed rotors [21] and to create saliencies artificially is generally
not feasible. The effect of naturally occurring saliencies is often very small. Therefore, it is often
necessary to rely on the third option, but when doing so, the properties of, e.g., the standard
or modified current model become drastically altered [35]. Particularly, the excellent low-speed
stability robustness is lost. A better approach is therefore to completely replace the current model
as flux estimator in a sensorless IM drive. Preferable to use is a flux estimator that does not use
the rotor speed in its equations at all. This is called an inherently sensorless flux estimator [31].

9.2.1 Direct Field Orientation Using the Voltage Model With Modifications

The first inherently sensorless flux estimator proposed is the voltage model, which is obtained as
follows. Whereas the current model is a simulation of the rotor circuit of the IM, the voltage
model is a simulation of the stator circuit as given by (9.16). A flux estimate can be obtained by
228 Chapter 9. Vector Control of IM Drives

integrating (9.16), giving

$
ψR = (vss − R
! s !s iss ) dt − L
! σ iss (9.45)

! s introduced
or, in operator form with the flux-emf estimate E
1 !s ! s = vs − (R
! s + pL
! σ )is .
! sR =
ψ E , E s s (9.46)
p
Notice that ωr does not appear on the right-hand side of (9.45); the voltage model is indeed inher-
ently sensorless. However, neither does ψ! sR ; the flux estimate is obtained by open-loop integration,
which indicates that the resulting flux dynamics will be marginally stable. Analysis confirms this
conjecture. The voltage model therefore cannot be used as it stands, but needs modification to
give an asymptotically stable system. We shall, in two steps, modify the voltage model such that
stability is gained, while accuracy is not sacrificed.

Modification I: Low-Pass Filter (Approximate Integrator)

The standard modification is to replace the open-loop integration in (9.45) by a low-pass filter,
which can also be regarded as adding a “leakage term” to the integral on right-hand side of (9.45)
$
ψR = (vss − R
! s !s is − αv ψ ! σ is
! sR ) dt − L (9.47)
s s

or, in operator form

1 !s
! sR =
ψ E . (9.48)
p + αv
Obviously, this introduces an error to the flux estimate, which can be quantified by combining
(9.48) with (9.46)
p
! sR =
ψ ψs . (9.49)
p + αv R
In the steady state, with p = jω1 , we have
jω1
! sR =
ψ ψs . (9.50)
jω1 + αv R
For a small ω1 and a constant filter bandwidth αv , such that |ω1 | < αv , the resulting error would
be significant. To rectify this, it is useful to select αv proportional to the angular stator frequency
αv = λ|ω1 |. (9.51)
This was proposed in [11, 76]; there called a “programmable low-pass filter.” An error which is
constant for all stator frequencies (sign changes excluded) is obtained
jω1 1
! sR =
ψ ψsR = ψs
jω1 + λ|ω1 | 1 − jλs R
! sR
ψ ψ!R −j θ̃ 1 1
⇒ = e = =√ ej arctan λs (9.52)
ψsR ψR 1 − jλs 1 + λ2
where λs = λ sgn ω1 . Now, if λ is fairly small, so is the error angle: λ = 0.1 ⇒ θ< = ±5.7◦ and
λ = 0.2 ⇒ θ< = ±11.3◦ , whereas in both cases ψ!R is practically equal to ψR .
9.2. Speed-Sensorless Control 229

Modification II: Compensation Gain

The small value of λ required to avoid large error angles unfortunately does not give sufficient
damping. Although the system is asymptotically stable (i.e., the poles are located in the LHP, not
on the imaginary axis), analysis shows that the poles are still located close to the imaginary axis,
at
λ|ω1 |
p≈− ± jω1 (9.53)
2
for smaller λ. A poorly damped system with oscillatory response results. From Paragraph 2.2.8,
we know that pole placement with an angle relative the imaginary axis of at least 45◦ can be
recommended, in order to obtain a well-damped system. This obviously requires a fairly large gain
λ. Steps must then be taken to compensate for the resulting error angle, which otherwise would
become unacceptably large.
Consider (9.52). The error due to the low-pass filter can be perfectly compensated in the steady
! sR with the complex gain 1 − jλs , as
state by scaling ψ
$
! sR ′ = (vss − R
ψ !s is − λs ω1 ψ ! σ is
! sR ′ ) dt − L (9.54)
s s

! sR = (1 − jλs )ψ
ψ ! sR ′ (9.55)
or, in operator form
1 − jλs ! s
! sR =
ψ E . (9.56)
p + λs ω1

The result is known as the statically compensated voltage model (SCVM), whose block diagram
is shown in Figure 9.6.

λs ω1

s 1 jλ s s
vss ΨR ΨR

Rs Lσ

iss

Figure 9.6. Block diagram for the SCVM.

9.2.2 Indirect Field Orientation Using the Statically Compensated Voltage Model
For the current model we concluded that IFO is often preferable to DFO. This holds for the SCVM
as well. An IFO implementation of the SCVM is readily obtained by expressing the flux estimate
230 Chapter 9. Vector Control of IM Drives

! sR = ψ!R ejθ1 , in (9.56). We obtain

in polar coordinates, as ψ
!s ⇒
(p + λs ω1 )(ψ!R ejθ1 ) = (1 − jλs )E
˙ ! s = (1 − jλs )E
ψ!R + (jω1 + λs ω1 )ψ!R = e−jθ1 (1 − jλs )E ! (9.57)
where [cf. (9.17)]
! =E
E !d + j E
!q = vs − [R
!s + (p + jω1 L
! σ )]is . (9.58)
˙
Equation (9.57) can now be split into its real and imaginary parts, from which ψ!R and ω1 can be
solved as
˙
ψ!R = −λs ω1 ψ!R + E!d + λs E
!q (9.59)
E!q − λs E
!d
ω1 = . (9.60)
ψ!R
Equation (9.59) can be simplified by substituting (9.60) in (9.59), whose right-hand side simplifies
to
!q − λs E
E !d
−λs ψ!R + E
!d + λs E
!q = (1 + λ2s )E
!d . (9.61)
!
ψR
!d and E
Since E !q are functions of ω1 [cf. (9.58)], (9.60) should include a low-pass filter to avoid an
algebraic loop between the left- and right-hand sides. We thus get
˙
ψ!R = γ E
!d (9.62)
αl E !q − λs E
!d
ω1 = , θ̇1 = ω1 (9.63)
p + αl ψ!R
where

λs = λ sgn ω1 γ = 1 + λ2s . (9.64)

It is interesting to note that (9.63) is identical to the PLL for PMSM drives, (8.10), even though
the SCVM was derived in a completely different way than the PLL. One addition has been made,
namely, that the SCVM the flux-modulus estimate ψ!R is dynamic according to the update law
(9.62). In (9.58), references should be used rather than measured variables. In addition, since the
flux dynamics are much slower than the current dynamics, the derivative term pL ! σ is can generally
be neglected (as was made for the modified current model). In component form we thus get
!d = v̄dref − R
E !s iref ! ref
d + ω1 L σ iq (9.65)
!q = v̄qref − R
E !s iref ! ref
q − ω1 L σ id . (9.66)
Remaining is to obtain an estimate of the rotor speed. This is straightforward; the standard slip
relation (9.6) is simply used “backwards.” With ω1 obtained from (9.63), an estimate of the angular
slip frequency is subtracted
!R iref
R q
ω
! r = ω1 − . (9.67)
ψref
9.2. Speed-Sensorless Control 231

If ω̂r is used in a speed controller whose output is iref

q , then a low-pass filter has to be added to
avoid an algebraic loop between (9.67) and the speed controller
> ?
αl !R iref
R q
ω
!r = ω1 − . (9.68)
p + αl ψref

Parameters αl and λ of the SCVM are similar to those of the modified current model, and similar
selection recommendations apply
RR
≪ αl < αc 1 ≤ λ ≤ 2. (9.69)
LM
Algorithm 9.3 illustrates an IFO implementation of the SCVM, whereas Figure 9.7 shows the block
diagram of an IM vector control system utilizing the SCVM, excluding the estimator (9.67) for ω
!r
to avoid clutter. Notice that the only difference as compared to the PMSM vector control system
using a PLL variant, shown in Figure 8.5, is the added estimator for ψ!R .
Algorithm 9.3
Statically Compensated Voltage Model
!d = v̄ ref − R
E !s iref + ω1 L
! σ iref
d d q
!q = v̄ ref − R
E !s iref − ω1 L
! σ iref
q q d
λs = λ sgn (ω1 )
γ = 1 + λ2s
> ?
!q − λs E
E !d
ω1 = ω1 + Ts αl − ω1
ψ!R
θ1 = mod(θ1 + Ts ω1 , 2π)
ψ!R = ψ!R + Ts γ E
!
>d ?
R!R iref
q
ω !r + Ts αl ω1 −
!r = ω −ω
!r
ψ!R

9.2.3 Properties of the Statically Compensated Voltage Model

The SCVM does not use R !R in any of its equations for determination of ω1 . From this fact it can
be concluded that the SCVM is completely insensitive to the rotor resistance at all speeds. This
is a universal property for all sensorless IM control systems [35]. On the other hand, an erroneous
R!R affects the accuracy of the speed estimate, as can be deduced from (9.67).
Let us compare the generalized slip relations that result from the modified current model,
(9.28), and the SCVM, (9.63). Both are repeated here, somewhat differently expressed
> ?
αl !R iref
R !d
λs E
q
ω1 = ωr + − (MCM) (9.70)
p + αl ψ!R ψ!R
> ?
αl E!q λs E!d
ω1 = − . (SCVM) (9.71)
p + αl ψ!R ψ!R
232 Chapter 9. Vector Control of IM Drives

vsref dq 2
αβ 3

isref Current dq 2
controller αβ 3 PWM

Rs jω1Ls is
M

Re Im
Eq dq iss 2
Ed
αβ 3
γ λs θ1

. ω1
.
ψR αl

Figure 9.7. IFO vector control system using the SCVM for speed-sensorless IM drives.

We find that both generalized slip relations have the same structure. Equation (9.71) is effectively
obtained by substituting
! R iq
R E!q
ωr + → (9.72)
ψ!R ψ!R
!R iref
in (9.70). Here, ωr + R !
q /ψR is the angular-flux-frequency estimate obtained from the standard
slip relation. Since ωr is not available in a speed-sensorless drive, the substitution shown in
(9.72)—or something similar—has to be made. The effect of this substitution can be understood
< For accurate model parameters, E
from (9.25): Eq ≈ θ̇ψR cos θ. !q = Eq . If, in addition, we assume
! <
that ψR ≈ ψR and θ ≈ 0, then
E!q
≈ θ̇ (9.73)
ψ!R
i.e., the desired estimate of the angular flux frequency. Since E!q /ψ!R gives a good estimate of θ̇
only if certain assumptions hold, it can intuitively be understood that the stability properties of
the SCVM are radically different from those of the MCM at low speeds. Instability phenomena
may appear, whose elimination require a slight redesign of the SCVM, see Paragraphs 9.2.4 and
9.2.5.
On the other hand, because Ed ≈ −θ̇ψR sin θ< [cf. (9.24)], the common term −E !d /ψ!R in the
generalized slip relations of the MCM and the SCVM has the effect of increasing the convergence
rate of ψq , as was found in Paragraph 9.1.7, forcing θ1 to convergence to a value close to θ faster.
Consequently, the SCVM has properties similar to the MCM in the higher-speed range. That is,
the only sensitive parameter is the leakage inductance, which impacts the field-orientation accuracy
according to (9.44).
9.2. Speed-Sensorless Control 233

9.2.4 Instability Phenomena for Regeneration-Mode Operation

In a sensorless IM drive, difficulties arise at low frequencies, where most schemes—including the
SCVM—suffer from a number of instability phenomena. Typically, instabilities tend to develop
in the regeneration mode, when speed and torque have different signs. There are two particularly
detrimental instability phenomena.
• Flux collapse [29, 70, 72]. The field orientation deteriorates in a way so that the IM is
unintentionally demagnetized, which is illustrated in Figure 9.8(a) The magnetizing current
component id in the dq frame used by the control system (i.e., that of the estimated flux) is
positive. When the error angle θ< is positive and large, while iq < 0, the actual flux-producing
current component iR d —which is the stator current vector projected on the rotor flux vector—
becomes negative, demagnetizing the motor. (This may also occur when θ< < 0 and iq > 0.)
Electrical torque is lost, and uncontrolled rotation in the direction of the load torque results.
• Frequency lockup [64] (or speed estimation failure [18]). The field orientation deteriorates,
such that the electrical torque reduces to a point where it equals the load torque, causing
lockup close to zero stator frequency. The flux emf—being proportional to the stator fre-
quency, see (9.24)–(9.25)—then vanishes, leaving the flux estimator—whose inputs are E !d
!
and Eq , see (9.62)–(9.63)—“blind.” Instead of the flux collapsing, the actual flux-producing
current component iR d will be larger than id , resulting in over-fluxed operation and a satu-
rating magnetizing inductance; see Figure 9.8(b).

jq jq
iq

!R !R
iR
"! ! "! d !
idR id
id
d d

iq

(a) (b)

Figure 9.8. Vector diagrams illustrating (a) flux collapse and (b) frequency lockup.

For some flux estimators the unstable region is wide (in the worst case, all negative speeds
for positive load torques, and vice versa), whereas a well-designed estimator has only a narrow
unstable region, close to zero stator frequency, or—in the ideal case—no unstable region at all. A
sensorless IM control system that is devoid of an unstable region may be called completely stable.
In [34], general theory for complete stabilization of some different flux estimator types is presented.
It is very difficult to guarantee complete stability, particularly with model parameter inaccu-
racies taken into account. Yet, if the drive does not dwell in the low-speed region, so that the
234 Chapter 9. Vector Control of IM Drives

unstable region is passed through quickly, then there is not enough time for an instability to de-
velop fully. (Still, instability tendencies may be noticed in the form of large flux transients.) Slow
rotation reversals under constant load should therefore be avoided, if possible, for a sensorless IM
drive.

9.2.5 Complete Stabilization of the Statically Compensated Voltage Model

In [33], a redesign of the SCVM is proposed, with the aim of obtaining complete stability. The
redesign is simple; it merely involves a gain selection that is slightly different from (9.64)
!R
λR
λs = sat (λ!
ωr /ω∆ , λ) γ =1− . (9.74)
!M
ω∆ L
Notice that the modified selection of λs is identical to that of modified voltage model, (9.39), but
with ω
!r replacing ωr . By this redesign, the regeneration mode remains stable, but only if R !s is a
!R and L
very accurate estimate of the true stator resistance. [R ! M in (9.74) are not critical [35].] Yet
better performance can be obtained if the sensitivity to R!s can be reduced. Methods for achieving
this—including on-line adaptation—are discussed in [35] and papers cited therein. Algorithm 9.4
shows the redesigned SCVM.
Algorithm 9.4
Statically Compensated Voltage Model Redesigned for Complete Stability
!d = v̄ ref − R
E !s iref + ω1 L
! σ iref
d d q
!q = v̄ ref − R
E !s iref − ω1 L
! σ iref
q q d
λs = sat (λ!
ωr /ω∆ , λ)
!R
λR
γ =1−
!M
ω∆ L
> ?
E!q − λs E
!d
ω1 = ω1 + Ts αl − ω1
ψ!R
θ1 = mod(θ1 + Ts ω1 , 2π)
ψ!R = ψ!R + Ts γ E
!
>d ?
R!R iref
q
ω !r + Ts αl ω1 −
!r = ω −ω
!r
ψ!R

Remark 9.2 Algorithm 9.3 is an IFO equivalent to the DFO variant of the SCVM shown in
Figure 9.6. Algorithm 9.4, on the other hand, has no DFO equivalent for complex space vectors.

Example 9.3 Experimental results for the redesigned SCVM with a four-pole, 22-kW, 50-Hz IM
drive will now be presented. The motor parameters are LM = 2.8 pu, Lσ = 0.21 pu, Rs = 0.023 pu,
and RR = 0.034 pu. The following inaccurate (in order to evaluate the robustness to parameter
! M = 0.9LM , L
errors) model parameters are selected: L ! σ = 1.1Lσ , and R
!R = 0.9RR . The critical
model !
√ parameter Rs is explicitly stated for each experiment. In the SCVM, ω∆ = 0.05 pu and
λ = 2. Two experiments are made, where in both experiments the speed reference ωref is stepped
9.2. Speed-Sensorless Control 235

through the sequence 0.2 pu, 0, −0.2 pu, and is then slowly ramped up at t = 8 s, starting from
ωref = −0.2 pu. A constant load torque that requires iq = 0.6 pu to balance is applied.
Figure 9.9 shows results of the first experiment, with R !s = Rs . The performance is good,
!
but some “bumps” can be seen in ψR and iq for zero-speed operation and zero-frequency crossing.
Since it is unlikely that the stator resistance is perfectly modeled, a narrow unstable region is likely
present, but passage through it is made sufficiently fast to prevent instabilities from developing.

(a) 0.4
ωr, ωref (pu)

0.2
0
−0.2
−0.4
0 2 4 6 8 10 12
(b) 1.2
0.8
id, iq (pu)

0.4
0
−0.4
0 2 4 6 8 10 12
(c) 1.6
1.2
ψ̂R (pu)

0.8
0.4
0
−0.4
0 2 4 6 8 10 12
t (s)

!s = Rs .
Figure 9.9. Successful operation for the ideal case R

The second experiment uses an underestimated model stator resistance R !s = 0.7Rs , see Figure
9.10. The speed step from −0.2 pu to 0.2 pu at t = 2 s is barely completed, which indicates
stability problems. At t = 9.5 s, a flux collapse occurs for the slow speed ramping against the load
torque. The experiment is interrupted at t = 10.5 s.

♣
236 Chapter 9. Vector Control of IM Drives

(a) 0.4

ωr, ωref (pu)

0.2
0
−0.2
−0.4
0 2 4 6 8 10 12
(b) 1.2
0.8
id, iq (pu)

0.4
0
−0.4
0 2 4 6 8 10 12
(c) 1.6
1.2
ψ̂R (pu)

0.8
0.4
0
−0.4
0 2 4 6 8 10 12
t (s)

!s = 0.7Rs .
Figure 9.10. Operation giving a flux collapse with R

9.3 Field-Weakening Control

Field-weakening control of the IM can use a similar control method as for the PMSM, see (8.38)
and Figure 9.11. The algorithm for modifying the d-direction current reference is given by
$ \
: 2 ; \Inom
iref
d = kfw Vbase − (vdref )2 − (vqref )2 dt \\ . (9.75)
Imin

In this case, though, the nominal value should be selected as

ψref
Inom = (9.76)
!M
L

whereas the minimum value should be selected large enough to prevent complete demagnetization,
e.g.,

Imin = 0.1Inom . (9.77)

9.3. Field-Weakening Control 237

2
Vbase
2 vsref
Inom is
kfw/s
Imin idref vsref
ref CC PWM M
iq,nom
sat
iqref
Ibase
Figure 9.11. Field-weakening controller for IM drives, where “sat” is implemented according to (9.94).
“CC” is the current controller with “back calculation.” (To avoid clutter, the dq and αβ coordinate trans-
formations are not shown.)

9.3.1 Gain Selection

To find a selection recommendation for kfw , the IM correspondence to (8.42) for the PMSM is
considered. The voltage components are obtained by substituting Ld → Lσ and Lq → Lσ in (8.43)
vd = −ω1 Lσ iq vq = ω1 (Lσ id + ψR ) (9.78)
yielding the following correspondence to (8.44):
2
f (id , ψR ) = Vbase − (ω1 Lσ iq )2 − [ω1 (Lσ id + ψR )]2 . (9.79)
In this case, though, the flux modulus is not constant; the dynamics (??) must be taken into
account. Hence, f is a function of two variables. Linearization and making approximations similar
to those made in the analysis of field-weakening control in the PMSM case yield
∂f (id , ψR ) ∂f (id , ψR )
f (id , ψR ) ≈ (id − i⋆d ) + ⋆
(ψR − ψR )
∂id ∂ψR
≈ −2|ω1 | [Lσ (id − i⋆d ) + ψR − ψR ⋆
] Vbase (9.80)
giving the second-order system
, ⋆
-
did ⋆ ψR − ψR
= −2kfw |ω1 |Vbase Lσ id − id + (9.81)
dt Lσ
dψR RR
= RR id − ψR . (9.82)
dt LM
We introduce αfw = 2kfw |ω1 |Vbase Lσ as the bandwidth
( of (9.81). System
) (9.81)–(9.82) can be
−αfw −αfw /Lσ
expressed in state-space form with the matrix A = , giving the characteristic
RR −RR /LM
polynomial
, - , -
RR 1 1
det(sI − A) = s2 + αfw + s + αfw RR + . (9.83)
LM Lσ LM
Assuming that αfw ≫ RR /LM and approximating 1/Lσ + 1/LM ≈ 1/Lσ (since LM ≫ Lσ ), we get
αfw RR
det(sI − A) ≈ s2 + αfw s + (9.84)
Lσ
238 Chapter 9. Vector Control of IM Drives

whose roots (i.e., the poles of the system) are

S
αfw α2fw αfw RR
s=− ± − . (9.85)
2 4 Lσ
This shows that a reasonable bandwidth selection is
2R!R
αfw = (9.86)
!σ
L
since poles at s = (−1 ± j)RR /Lσ then ideally (for accurate model parameters) are obtained, i.e.,
with the desired minimum angle of 45◦ relative the imaginary axis. The resulting dynamics are
fast; the response time Lσ /RR is roughly twice the transient time constant given in (5.80), and in
the range of 1/10 of the rotor time constant LM /RR . The dynamic properties for smaller speed
changes in the field-weakening region are thus normally excellent. The corresponding gain selection
is
!R
R
kfw = (9.87)
! 2 Vbase max(|ω1 |, ωbase )
L σ

which yields a constant gain below base speed (where field-weakening control is not used).

9.3.2 Maximum Torque

In the
Rfield-weakening range, maximum torque is mostly obtained by applying the nominal current,
i.e., i2d + i2q = Ibase . This is called field-weakening region I. Interestingly, though, above a certain
stator frequency, maximum torque is obtained for a current lower than the nominal. This is called
field-weakening region II. The torque-maximizing component iq to be applied there can be found
as follows.
The electrical torque is proportional to ψR iq and, hence, to id iq in the steady state, since
ψR = LM id in the steady state under perfect field orientation. Equation (9.78) then can be
expressed as
vd = −ω1 Lσ iq vq = ω1 (Lσ + LM )id (9.88)
which shows that the electrical torque is proportional to
vd vq
id iq = − 2 (9.89)
ω1 (Lσ + LM )Lσ
R
in the steady state. If the nominal voltage is applied, i.e., vd2 + vq2 = Vbase , then maximum
electrical torque is obtained when1
Vbase
|vd | = |vq | = √ . (9.90)
2
From (9.88) it is found that this corresponds to the following relation for the current components:
Lσ + LM
|iq | = ξid , ξ= . (9.91)
Lσ
1 This can be shown in a similar fashion as (5.125).
9.3. Field-Weakening Control 239

So, provided that the model parameters are accurate, maximum torque can be obtained by selecting
the q-component reference as
!σ + L
L !M
|iref ! ref ξ! =
q | = ξid , . (9.92)
!σ
L
This selection would result in overcurrent in field-weakening region I. Calculation of the break point
where (9.91) yields nominal current gives the anglar stator frequency at which field-weakening
region II begins. We get
, - , -2 , -2
2 2 1 2 1 + ξ2 vq 1 + ξ2 Vbase 2
id + iq = + 1 iq = = √ = Ibase
ξ2 ξ2 ω1 L σ ξ2 2ω1 Lσ
S
1 + ξ 2 Vbase
⇒ |ω1 | = . (9.93)
2ξ 2 Lσ Ibase

For example, for LM = 2 pu and Lσ = 0.2 pu, (9.93) yields |ω1 | ≥ 3.55 pu for Vbase = Ibase = 1
pu. This is quite high an angular stator frequency, which shows that field-weakening region II is
used only in applications that require operation deep into the field-weakening range. Figure 9.12
shows typical steady-state characteristics for operation into the field-weakening range, again for
LM = 2 pu and Lσ = 0.2 pu. Notice the sharp drop-off in the electrical torque that results if |iq |
is not reduced according to (9.92) in field-weakening region II. The algorithm for selection of the
torque-maximizing reference can be written as
( ,R -)
iref
q = sat i ref
q,nom , min I 2
max − (i ref )2 , ξiref
d d (9.94)

where Imax shall be reduced to Ibase after the maximum time that overcurrent is allowed.
The resulting field-weakening controller is identical to that for PMSM drives, see Figure 8.10,
except that Inom , Imin , and the saturation scheme for iref
q are chosen differently. See also Algorithm
9.5.
Algorithm 9.5
Field-Weakening Controller for IM Drives
!R
Ts R : ;
iref ref
d = id +
2
Vbase − (vdref )2 − (vqref )2
! 2 Vbase max(|ω1 |, ωbase )
L σ
iref
d = min[max(iref
d , Imin ), Inom ]
!σ + L
L !M
ξ! =
L!σ
( ,R -)
iref = sat iref , min I 2 − (i ! ref
ref )2 , ξi
q q,nom max d d

Remark 9.3 During field-weakening operation LM is usually somewhat larger than during normal
operation, because it is no longer saturated. A saturation model or an on-line adaptation scheme
! M can be used to improve performance.
for L
240 Chapter 9. Vector Control of IM Drives

0.8

iq (pu) 0.6

0.4
id

0.2

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
ω1 (pu)

0.8

0.6
τe (pu)

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
ω (pu)
1

Figure 9.12. Current components and maximum obtainable electrical torque from zero stator frequency
up into field-weakening region II. The dashed curves show results obtained when |iq | is not appropriately
reduced in field-weakening region II.

Remark 9.4 Perfect field orientation is assumed in the above analysis. In [30] it is shown
that leakage inductance should be overestimated for good performance during operation in field-
weakening region II

! σ ≥ Lσ .
L (9.95)

Example 9.4 We again consider a two-pole IM, with similar electrical parameters to that in
Example 9.1, i.e., ωbase = 100π rad/s and the per-unit parameters ψref = 0.9, LM = 2, Lσ = 0.2,
and Rs = RR = 0.05. The speed control loop is tuned for a rise time of 0.35 s, but light mechanical
dynamics with the per-unit parameters J = 100 and b = 0.02 together with zero external load
torque are this time considered. The current model with accurate model motor parameters is
used for field orientation. Algorithm 5.1 (speed controller), Algorithm 9.1 (current model), and
Algorithm 9.5 (field-weakening controller) are combined as shown in Algorithm 9.6.

Algorithm 9.6
9.3. Field-Weakening Control 241

Current Model With Speed and Field-Weakening Controllers

!R
Ts R : ;
iref ref
d = id +
2
Vbase − (vdref )2 − (vqref )2
! 2 Vbase max(|ω1 |, ωbase )
L σ
iref ref
d = min[max(id , Imin ), Inom ]
> ?
ψ!R
ψ!R = ψ!R + Ts R
!R iref
d −
!M
L
2K 2 (αs J! − !b)
ba =
3n2 ψ!R
p

2K 2 αs J!
kps =
3n2 ψ!R
p

2K 2 α2s J!
kis = .
3n2p ψ!R
es = ωref − ωr
iref
q,nom = kps es + kis Is − ba ωr
!σ + L
L !M
ξ! =
!σ
L
( ,R -)
ref ref
iq = sat iq,nom , min 2 ref 2 ! ref
Imax − (id ) , ξid
( )
1 ref ref
Is = Is + Ts es + (i − iq,nom )
kps q
R!R iref
q
ω1 = ωr +
!
ψR
θ1 = mod(θ1 + Ts ω1 , 2π)

Three slightly different variants are tested.

• Variant 1. Algorithm 9.6 is used as it stands.

• Variant 2. The saturation scheme for iref

q of Algorithm 5.1
, R -
iref
q
ref 2 ref
= sat iq,nom , Imax − (id ) 2

is used by mistake. Thus, no reduction of |iq | is made in field-weakening region II.

• Variant 3. Gain scheduling of the speed-controller parameters with the varying flux estimate
is not made. Instead, the nominal flux is used in the gain selections as

2K 2 (αs J! − !b) 2K 2 αs J! 2K 2 α2s J!

ba = kps = kis = .
3n2p ψref 3n2p ψref 3n2p ψref
242 Chapter 9. Vector Control of IM Drives

Performance is evaluated by stepping the speed reference from 0 to 4 pu. The results for the three
variants are shown in Figure 9.13. Two observations can be made.

• For Variant 2, the speed setpoint is not attained, because not enough torque is produced in
field-weakening region II.

• For Variant 3, ωr makes a slight overshoot and iq makes an undershoot after the speed
reference is reached.

It can be concluded that torque maximization according to (9.94) and gain scheduling of the speed
controller parameters with the estimated flux modulus both improve performance.

3
ω (pu)

2
r

0
0 1 2 3 4 5 6 7 8 9 10
t (s)

0.8
(pu)

0.6
q
i

0.4
d
i

0.2

0
0 1 2 3 4 5 6 7 8 9 10
t (s)

Figure 9.13. Simulation of an IM drive for a speed-reference step to 4 pu. The solid, dashed, and
dash-dotted curves respectively show results for Variants 1, 2, and 3 of Algorithm 9.6.

9.4 Direct Torque Control

In vector control, motor control and converter control can usually be considered as separate func-
tions: the converter control creates the PWM pattern based on the voltage reference generated by
the motor control. When a hysteresis control scheme is used for current control, see Paragraphs
3.1.3 and 6.3.1, current control and PWM are merged, since the current control forms the switching
9.4. Direct Torque Control 243

signal to the respective inverter leg. Direct torque control (DTC) is a step further in this direction.
The DTC scheme was independently invented by Takahashi and Noguchi [80] and Depenbrock [17]
in the mid-1980s. Later, the DTC principle was successfully applied to commercial IM drives by
ABB [82].

9.4.1 Control Principle

The main idea of DTC, as originally suggested by Takahashi and Noguchi in [80], is to keep the
stator flux and the electrical torque within specified tolerance bands by always choosing the voltage
vector that is most suitable for the situation. The method departs from the idea of coordinate
transformation and the analogy with dc motor control. It also departs from the principle of rotor-
flux orientation, using instead stator-flux orientation (SFO), but without the explicit usage of a
stator-flux-oriented dq frame. The hysteresis control principle allows the torque to be controlled
at the fastest possible rate of change. In addition, there is no overshoot in the torque response.

β
v3 (0, 1, 0) v2 (1, 1, 0)
idc
+
v4 (0, 1, 1) v1 (1, 0, 0)
Vdc ha hb hc
α
v7 (1, 1, 1) v0 (0, 0, 0)
−
ia ib ic
v5 (0, 0, 1) v6 (1, 0, 1)

(a) (b)

Figure 9.14. (a) Change-over switch model of a three-phase VSC. (b) Space vectors produced (in stator
coordinates).

In the three-phase VSC model shown in Figure 9.14(a), the change-over switches are coupled
either to the negative potential (down) or the positive potential (up). The dc-link voltage is denoted
by vdc , and the switching functions ha , hb , and hc of each phase equal either 0 or 1, corresponding
to the switch positions down and up, respectively. All switch positions can be defined by the
combination (ha , hb , hc ) of the switching functions. For example, the switch positions in Figure
9.14(a) are defined by the combination (0, 1, 1) of the switching functions. The possible voltage
vectors of the converter output (and the corresponding switching functions) are shown in Figure
9.14(b) in stator coordinates. The magnitude of the nonzero vectors v1 , . . . , v6 is 2KVdc /3, and
the angle between adjacent voltage vectors is 60◦ . Two zero vectors are available: v0 and v7 .
In the DTC algorithm, the most suitable voltage vector is chosen based on the stator equation
in stator coordinates (5.29), rewritten here as

dψss
vss = Rs iss + . (9.96)
dt
During one switching combination, the stator voltage vector vn , n = {0, . . . , 7} is constant over
244 Chapter 9. Vector Control of IM Drives

the time span ∆t. The resulting change of the stator flux is
$ ∆t
∆ψss = ψss − ψss,0 = (vss − Rs iss ) dt ≈ vn ∆t. (9.97)
0

The stator-flux vector thus changes in the direction of the applied voltage vector, and the rate
of change is proportional to the dc-link voltage when the effect of the stator resistance is small.
Consequently, the flux can be changed in six directions. When a zero vector is used, the flux vector
practically stops moving. When controlling the stator flux vector, the complex plane is subdivided
into six sectors as shown in Figure 9.15. The borders of the sectors are in the middle of the angles
between adjacent voltage vectors.

Sector 3 Sector 2
v3 v2

Sector 4 v4 v1 Sector 1

v5 v6

Sector 5 Sector 6

Figure 9.15. Definition of the six sectors.

The stator flux vector is controlled to follow the flux-reference circle within a hysteresis band,
as depicted in Figure 9.16. The voltage vectors (and the time periods in which they are used) are
chosen in such a way that, in addition to the flux, the torque remains within a certain tolerance
band. From (5.25)–(5.26) we find that the stator- and rotor-flux vectors are related as

ψss = ψsR + Lσ iss . (9.98)

This relation allows, by substituting iss = (ψss − ψsR )/Lσ in (5.54), the torque expression to be
expressed as
3np
τe = Im {(ψsR )∗ ψss } . (9.99)
2K 2 Lσ
The stator flux in effect takes the place of the stator current. Quick torque changes are thus
obtained by holding the rotor flux constant and changing the stator flux. [Owing to the leakage
flux contribution Lσ iss in (9.98), the stator flux reacts quicker to changes in the stator voltage than
the rotor flux.] Expressing the flux vectors in polar form allows (9.99) to be expressed as

3np
τe = ψR ψs sin(θψs − θψR ) (9.100)
2K 2 Lσ

where θψs and θψR respectively are the stator- and rotor-flux angles, as shown in Figure 9.17(a).
According to (9.100), the torque is proportional to sine of the angle θψs − θψR between the flux
vectors as well as to their magnitudes. A fast change of the torque can be achieved by accelerating
9.4. Direct Torque Control 245

v4 v5 v4
v3
v4
v5
v4
v3
v5
v4 ω1
v6 v3
v2
v5
v3 v2 ∆ψss
ψss,0
2∆ψ
v4 v1
v6
v5
v5 v6

Figure 9.16. Path of stator flux vector showing the voltage vectors used in DTC. Zero voltage vectors
are illustrated by dots.

or decelerating the stator-flux vector, which can be accomplished by proper selection of the stator
voltage vectors, i.e., by selection of consecutive switching states. The stator-current vector is not
directly controlled, and it can be considered as a secondary quantity in the DTC. It is to be noted
that (9.100) is not used for control purposes; it is shown here only for describing the behavior of
the motor.

v3 v2
ψss
ψss,0
θψs ψsR
v5 v6
θψR
α 1 ψsR,0

(a) (b)

Figure 9.17. Illustrations of DTC principle. (a) Stator-flux and rotor-flux angles. (b) Voltage vectors
that can be chosen in Sector 1.

In Figure 9.17(b), the rotational direction and the torque of the motor are assumed to be
positive (counterclockwise) and the stator-flux vector is in Sector 1. For increasing the stator flux,
either voltage vector v2 or v6 is to be chosen. The stator flux can be decreased by choosing voltage
vector v3 or v5 . The torque can be increased by advancing the stator-flux vector, i.e., by increasing
the angle between the two flux vectors. Voltage vectors v2 or v3 can be used for this purpose.
The torque can be decreased by using voltage vectors v5 or v6 , but zero vectors v0 and v7 also
decrease the torque; the stator-flux vector is then stopped and the rotor-flux vector continues to
rotate when the rotor is rotating. Normally, zero vectors are used for decreasing the torque, except
at low speeds.
246 Chapter 9. Vector Control of IM Drives

new trajectory old trajectory

v5 v4 v3
γ v4
v6 v3
v2
v5
v3 v2 v3

v4 v1
v6
v2
v5 v3
v5 v6

Figure 9.18. Path of stator flux vector when the border between Sectors 2 and 3 is shifted for obtaining
a rapid torque change.

As illustrated in Figure 9.16, a voltage vector is used in one sector to increase the stator flux,
and the same voltage vector is used in the next sector to decrease the stator flux. The trajectory
of the stator flux vector forms a piecewise linear approximation of a circle. Figure 9.18 shows a
situation in which, following a rapid change in the torque reference, the border separating Sectors
2 and 3 is shifted backwards by the angle γ. Now, the converter chooses between voltage vectors v4
and v5 instead of v3 and v4 , respectively. It is to be noted that the linear speed of the stator-flux
vector along its trajectory is constant and is proportional to the dc-link voltage. Therefore, the
flux vector arrives to a new location in a shorter time than if it had traveled along the regular
trajectory. A rapid increase of the torque is achieved by this acceleration of the stator-flux vector,
because that vector quickly moves away from the rotor-flux vector. It is to be noted that this
simple method, based on shifting the border between two adjacent sectors, is only an illustrative
example. In practice, more elaborate methods are used for choosing the voltage vectors when rapid
changes in the torque are requested.
Depenbrock’s DSC scheme differs from the described principle in the sense that the prescribed
flux path is a hexagon rather than a circle. All but six switchings per period of the fundamental
excitation are used to reduce the torque ripple. This scheme therefore generates quite low torque
ripple. Notably, low-order torque harmonics are small or absent [2], which is important when the
switching frequency is low, as mechanical resonances otherwise may be excited. The difference
between the two schemes is illustrated in Figure 9.19.

9.4.2 Control System

A control system based on the DTC method is depicted in Figure 9.20. The stator flux can be
estimated by a variant of the voltage model, e.g., the SCVM, obtained simply by omitting all terms
! σ from the equations.
proportional to L
The only motor parameter needed for the flux estimation is the stator resistance estimate R !s .
The torque estimate is calculated in stator coordinates by means of the torque expression
3np
τ!e = Im {(ψss )∗ iss } . (9.101)
2K 2
9.4. Direct Torque Control 247

(a) (b)

1 1
(0,1,1)

0.5 0.5
(0,0,1) (0,1,0)
ψsβ (pu)

(pu)
0 0

sβ
ψ
(1,0,1) (1,1,0)
−0.5 −0.5

(1,0,0)
−1 −1

−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1

ψsα (pu) ψsα (pu)

Figure 9.19. Flux paths. (a) DSC (Depenbrock): hexagonal. (b) DTC (Takahashi–Noguchi): ideally
circular. (Note that, unlike the rotor flux, the nominal modulus of the stator flux is 1 pu.)

The difference between the flux-modulus reference ψsref and the estimated stator-flux modulus ψ!s
is fed to the hysteresis controller for the flux, whose output signal is dψ . Similarly, the difference
between the torque reference τref and the torque estimate τ!e is fed to the hysteresis controller
for the torque, whose output signal is dτ . The flux controller is usually a two-level comparator,
whereas the torque controller is a three-level comparator (taking the direction of rotation into
account). The output signals dψ and dτ of the hysteresis controllers give information about the
direction in which the flux should be changed: dψ about the flux magnitude and dτ about the flux
angle. Based on this information and the angle estimate θ!ψs of the flux (giving the sector of the
flux), the switching logic chooses the actual voltage vector to be used.
The principle of the switching logic can be explained as follows. The tolerance band of the
hysteresis controller for the stator flux linkage is denoted as 2∆ψ. The output of the flux controller
is changed to

dψ = 1 if ψ!s < ψsref − ∆ψ,

dψ = −1 if ψ!s > ψsref + ∆ψ.

The tolerance band of the hysteresis controller for the torque is denoted by 2∆τ . The output of
this controller also depends on the direction of rotation. When the flux is rotated in the positive
direction, the output of the torque controller is changed to

dτ = 1 if τ!e < τref − ∆τ ,

dτ = 0 if τ!e > τref + ∆τ .

When the flux is rotated in the negative direction, the output of the torque controller is changed
to

dτ = 0 if τ!e < τref − ∆τ ,

dτ = −1 if τ!e > τref + ∆τ .
248 Chapter 9. Vector Control of IM Drives

+
ψsref dψ
− Switching
ωref logic
Speed τref + dτ
controller −

τ!e Torque iss

estimation
θ!ψs
Flux and
ψ!s
speed vss
ω
!r estimation

M
ωr (measured)

Figure 9.20. DTC system.

Table 9.1. Voltage vectors to be selected by the switching logic. Zero voltage vectors (denoted by “0”)
are realized by means of v0 or v7 , depending on the previous voltage vector.
dτ dψ Sector 1 Sector 2 Sector 3 Sector 4 Sector 5 Sector 6
1 1 v2 v3 v4 v5 v6 v1
1 −1 v3 v4 v5 v6 v1 v2
0 1 0 0 0 0 0 0
0 −1 0 0 0 0 0 0
−1 1 v6 v1 v2 v3 v4 v5
−1 −1 v5 v6 v1 v2 v3 v4

The voltage vectors chosen by the switching logic are presented in Table 9.1 for all sectors. Zero
voltage vectors are denoted by “0” in the table; the actual choice between the zero vectors v0 and
v7 should minimize the number of switchings.
Let us consider positive direction of rotation. The torque is increased when dτ = 1. In this case,
the switching logic chooses a voltage vector that rotates the stator-flux vector forward. Therefore,
• in the motoring mode, the stator flux vector rotates further away from the rotor flux vector
and the torque increases, whereas
• in the regenerating mode (when the stator flux lags the rotor flux), the stator flux vector
gets closer to the rotor flux vector and the torque becomes less negative.
The torque is decreased when dτ = 0. In this case, the switching logic chooses a zero voltage vector
that practically stops the stator flux, and the rotor flux continues to rotate. Therefore,
• in the motoring mode, the rotor-flux vector gets closer to the stator-flux vector, and the
torque and the shaft power decrease, and
9.4. Direct Torque Control 249

• in the generating mode, the rotor-flux vector rotates further away from the stator-flux vector,
and the torque becomes more negative.

When the direction of rotation is negative, the torque is changed more negative when dτ = −1,
and more positive when dτ = 0. In addition to the hysteresis control of the torque, the magnitude
of the stator flux is increased when dψ = 1, and it is decreased when dψ = −1. When dτ = 0, a
zero vector is chosen regardless of the flux control.

9.4.3 Discussion
The hysteresis control requires a high sampling frequency and computing speed of the control
system in order to detect when the flux or the torque gets outside the hysteresis band. Typically,
the estimates of the flux linkage and torque are calculated at a sampling frequency of 40 kHz or
higher (corresponding to sampling intervals of 25 µs or less).
The hysteresis control is characterized by a somewhat chaotic behavior. The switching fre-
quency of the inverter varies depending on the operating point and motor parameters. In addition,
the switchings are not exactly repeated during subsequent rotations of the stator flux vector.
Therefore, there is a continuous variation in the number and length of voltage pulses fed to the
motor. In some cases, a slow subharmonic voltage fluctuation can occur, specially when the sta-
tor frequency is low. Subharmonic components can also be seen in the torque, which can cause
problems in the form of mechanical vibrations. In order to avoid these problems, the tolerance
bands of the hysteresis controllers must be varied so that the switching frequency is high at all
stator frequencies. On the other hand, the switching losses in the semiconductor components set
an upper limit to the switching frequency.
In the literature, various suggestions have been presented for modifying the DTC method. A
simple example is to replace the hysteresis controllers with linear controllers in the flux and torque
control loops for obtaining a stator voltage reference, and to use a PWM scheme for the realization
of the voltage reference instead of the switching logic. In this way, various problems of the DTC
method can be solved, but its ability to a fast torque response is lost. A survey of DTC-based
methods can be found in [13].

9.4.4 Stator-Flux Orientation

Unlike all other vector control schemes discussed in this chapter, DTC uses stator-flux orientation
(SFO) instead of rotor-flux orientation (RFO). Is SFO an alternative to RFO also in standard vector
control, employing an inner synchronous-frame current control loop and an outer flux estimator
and synchronization loop? Let us analyze SFO in such a system.
From (5.9) we know that the stator voltage, stator current, and stator flux are related as

dψss
= vss − Rs iss . (9.102)
dt
Thus, the voltage model for stator-flux estimation simplifies to

! ss
dψ !s iss
= vss − R (9.103)
dt
(i.e., the voltage drop across the leakage inductance vanishes, as compared to the corresponding
relation for the rotor flux). In, e.g., IFO implementation of the SCVM, the flux-emf components
250 Chapter 9. Vector Control of IM Drives

are computed as
!d = v ref − R
E !s iref !q = v ref − R
E !s iref . (9.104)
d d q q

!d and E
The problem of an algebraic loop forming because ω1 is computed from E !q , but is also used
to compute these components, disappears. Moreover, the estimator’s sensitivity to L! σ vanishes,
leaving the stator resistance as the only sensitive parameter.
Let us now consider the rotor circuit in the IM’s dynamic equivalent circuit. Substituting
ψR = ψs − Lσ is in (5.40), we obtain
, -
dψs dis RR
− Lσ = RR is − + jω2 (ψs − Lσ is ) (9.105)
dt dt LM
(, - ) , -
dψs dis Lσ RR
⇒ = Lσ + 1+ RR + jω2 Lσ is − + jω2 ψs .
dt dt LM LM
Neglecting dis /dt as seen from the slower flux dynamics, and assuming perfect field orientation,
i.e., taking the stator flux real, we obtain by splitting the real and imaginary parts
, -
dψs Lσ RR
= 1+ RR id − ω2 Lσ iq − ψs (9.106)
dt LM LM
, -
Lσ
0 = 1+ RR iq + ω2 Lσ id − ω2 ψs . (9.107)
LM
In (9.106) it is seen that the flux modulus is determined not only by id , but also (to a lesser degree)
by iq . Whereas a constant id is used in RFO (except in the field-weakening region), in SFO id
must be modified with the torque (i.e., with iq and ω2 ) in order to keep the stator-flux modulus
constant. For this purpose, an outer flux control loop—with fairly low bandwidth—can be used.
Following the theory for the SCVM, the flux modulus can be estimated as follows:

dψ!s !d .
= γE (9.108)
dt
The error ψsref − ψ!s is then fed to an integrator, whose output is iref
d . A variant of the current
model can be obtained. The slip relation in SFO is found from (9.107)
> ?
!σ
L
1+ ! R iq
R
!M
L
ω1 = ωr + , θ̇1 = ω1 (9.109)
ψ!s − L! σ id

which is more complicated than that in RFO.

To summarize and add some results not explicitly derived, SFO has been found to have the
following properties as compared to RFO.
• Inherently sensorless flux estimator (SCVM): Simpler; the low-pass filter used for
avoiding an algebraic loop can be removed.
!s is the only critical
• Accuracy of flux estimation and field orientation: Better; R
parameter.
• Flux control: More complicated; id must be varied, even below base speed.
9.4. Direct Torque Control 251

• Sensored control using the current model: More complicated slip relation.
• Speed estimation in sensorless control: More complicated. Here, the slip relation
(9.109) can be used “backwards” [cf. (9.67)].
• Low-speed stability properties for sensorless control: Similar. Analysis shows that
the problems regarding flux collapse and frequency lockup remain.
• Field weakening: Slightly simpler. The standard method of reducing the stator flux as
1/|ω1 | in the field-weakening region has been shown to give close to maximum torque [87]
(which is not true for the rotor flux [49]).
• Current control: Slightly simpler. It can be shown that the cross-coupling term jω1 Lσ is in
RFO is in SFO replaced by jω2 Lσ is . This term is small enough to be neglected, so decoupling
is not necessary for good performance.
Clearly, there are both pros and cons of SFO as compared to RFO; the choice of method appears
often to be made based on personal preferences of each researcher, design engineer, or engineering
team. We in this book prefer RFO because it is more general, being applicable to IMs, PMSMs,
and in effect also grid-connected inverters, with a minimum of modifications necessary.
252 Chapter 9. Vector Control of IM Drives

.
APPENDIX A

Lyapunov Stability Theory

Linearization, as described in Section 2.3, may establish local stability about an equilibrium point.
If possible, it is desirable to show that a nonlinear system is globally stable. That is, regardless of
the starting point x(0), it can be guaranteed than x converges asymptotically to the equilibrium
point, which we here for simplicity take as the origin. This can be accomplished by means of
Lyapunov’s second (or direct) method. A Lyapunov function is a function with the following
properties:
1. V (x) > 0, x ̸= 0; V (0) = 0.
dV (x) dV (x)
2. V̇ (x) = ẋ = f (x) ≤ 0.
dx dx
If, in addition, the function is unbounded, i.e., V (∞) = ∞, it is said to be a global Lyapunov
function. If V̇ (x) < 0, x ̸= 0, it is said to be a strict Lyapunov function.
A Lyapunov function can be considered as the energy stored in the system. Naturally, the
energy has to be positive—hence, the first property—and if energy is always dissipated—i.e.,
V̇ (x) ≤ 0—it appears logical that the system is stable. We formulate this as the following theorem,
the proof of which can be found e.g., in [48].
Theorem A.1 Suppose that there exists a Lyapunov function V (x) for the system
ẋ = f (x).
⋆
Then the equilibrium point x = 0 is stable. Furthermore, it is globally stable if V (x) is a global
Lyapunov function, and asymptotically stable if V (x) is a strict Lyapunov function.
If a strict Lyapunov function is not global, there may be several equilibrium points to which x may
converge. (Such a case will soon be encountered.) The good thing is that x cannot go into a limit
cycle or diverge.
Lyapunov theory provides an extremely powerful tool for stability analysis of nonlinear systems.
The problem is that finding a Lyapunov function is not always trivial. In many cases, the energy
stored in the system can be used as a Lyapunov function, though.
Example A.1 Circuit with nonlinear resistor. Consider the RLC circuit in Figure A.1. The
resistor is a nonlinear one with the characteristic that the voltage–current relationship is given by
vR = ki3R .
The circuit is therefore nonlinear. However, since the power developed in the resistor, vR iR = ki4R ,
is non-negative, energy is always dissipated in the circuit, so it should be stable. To prove this, we
let the energy stored in the reactive components be a Lyapunov function candidate:
1 1 1
V (x) = L1 x21 + C2 x22 + L3 x23 .
2 2 2
254 Appendix A. Lyapunov Stability Theory

This is a global function, since V (∞) = ∞ regardless of which one of the variables x1 , x2 , or x3
that is made infinitely large. Applying Ohm’s and Kirchhoff’s laws yields the following equations:
k 3 1
ẋ1 = − x1 − x2 (A.1)
L1 L1
1
ẋ2 = (x1 − x3 ) (A.2)
C2
1
ẋ3 = x2 . (A.3)
L3
Differentiating the Lyapunov function candidate then yields
V̇ (x) = L1 x1 ẋ1 + C2 x2 ẋ2 + L3 x3 ẋ3
, -
k 3 1 1 1
= L1 x1 − x1 − x2 + C2 x2 (x1 − x3 ) + L3 x3 x2
L1 L1 C2 L3
= −kx41 ≤ 0.
Quite naturally, the time derivative of the stored energy is the power dissipated in the nonlinear
resistor. The energy is thus a Lyapunov function and the circuit is, indeed, stable.
Is the circuit asymptotically stable? Since V̇ is not a function of x2 and x3 , V is not a strict
Lyapunov function. It is only guaranteed that x1 converge to 0. Once x1 = 0, then V̇ = 0, i.e.,
1 1
V (x)|x1 =0 = C2 x22 + L3 x23 = constant. (A.4)
2 2
Hence, although x2 and x3 cannot diverge, it cannot immediately be guaranteed that x2 and x3
will not enter a limit cycle once x1 has converged. The system is asymptotically stable, however.
This can be deduced by the following reasoning.
1. V̇ < 0 as long as x1 ̸= 0; thus, x1 converges asymptotically to 0. Clearly, when x1 ≡ 0, then
ẋ1 ≡ 0.
2. But x1 ≡ ẋ1 ≡ 0 requires—see (A.1)—that x2 ≡ 0. Thus, ẋ2 ≡ 0.
3. From (A.2) it follows that when ẋ2 ≡ 0 and x1 ≡ 0, then x3 ≡ 0.
Thus, we have shown that the equilibrium point x⋆1 = x⋆2 = x⋆3 = 0 is globally asymptotically
stable. This method for checking asymptotic stability is known as LaSalle’s invariance principle
[48].

L1
x1 x3

C 2 x2 L3

Figure A.1. RLC circuit with nonlinear resistor.

 255

Example A.2 Standard slip relation derived using Lyapunov theory. In some cases,
Lyapunov theory is useful for finding a nonlinear control law. A Lyapunov function candidate V
is selected. A control law which assures that V̇ ≤ 0 is then derived. Consider the differential
equations for the rotor flux of the induction motor in synchronous coordinates
RR
ψ̇d = RR id − ψd + ω2 ψq
LM
RR
ψ̇q = RR iq − ψq − ω2 ψd .
LM
We want ψd to converge to ψref and ψq to 0. Let us assume that the motor parameters are known
and id = ψref /LM , where ψref is constant. Introducing ψ<d = ψd − ψref yields
˙ RR <
ψ<d = − ψd + ω2 ψq
LM
RR
ψ̇q = RR iq − ψq − ω2 (ψ<d + ψref ).
LM
Now, let us take
1 <2
V = (ψ + ψq2 ) (A.5)
2 d
as a Lyapunov function candidate. This yields

˙ RR <2
V̇ = ψ<d ψ<d + ψq ψ̇q = − (ψ + ψq2 ) + (RR iq − ω2 ψref )ψq .
LM d

Thus, it is not guaranteed that V̇ ≤ 0, so V is not automatically a Lyapunov function. But since
ω2 is an input signal available for manipulation (via ω1 = ωr + ω2 , if ωr is measured), we can
choose
RR iq
ω2 = (A.6)
ψref
in order to cancel the last two terms, and obtain
RR <2
V̇ = − (ψ + ψq2 ) ≤ 0.
LM d
The resulting system is globally asymptotically stable. This example has shown that the standard
slip relation (A.6) can be derived as a step in making V a Lyapunov function.

Example A.3 Global stability of the PLL, which was claimed in Section 6.5.1, will now be
proven. Let us consider (6.42)–(6.43) with ω
<i = ω
< s − ωi :

<˙ i = −kip E sin θ<

ω
˙
θ< = ω
<i − kpp E sin θ<
256 Appendix A. Lyapunov Stability Theory

and take
1 2
< = <
V (<
ωi , θ)
ω
< + k(1 − cos θ)
2 i
as Lyapunov function candidate, where k is a positive constant to be selected. Clearly, V (0, 2nπ) =
0 (n integer) and V > 0 otherwise. Differentiating yields
˙
V̇ = ω <˙ i + k sin θ<θ< = −kip E ω
<i ω <i sin θ< + k sin θ(<
< ωi − kpp E sin θ)
<
= −kkpp E sin2 θ< + (k − kip E)< ωi sin θ< = {k = kip E}
= −kip kpp E 2 sin2 θ< ≤ 0.
This shows that the PLL is stable, but what about global and asymptotic stability? Because there
is an infinite number of equilibrium points: [< < = [0, 2nπ], the system is not globally stable in
ωi , θ]
a strict sense. However, if all equilibrium points are considered equal—which is logical, as they
differ only by one revolution (2π) along the θ< axis—the system can be considered globally stable.
There is no guarantee for convergence to the closest equilibrium point, however, which causes the
mentioned slipping phenomenon.

♣
APPENDIX B

CORDIC Algorithm

In most vector control systems, at least two coordinate transformations have to be performed:
transformation of the measured stator current vector from stator to synchronous coordinates and
of the converter voltage command vector from synchronous to stationary coordinates. These
coordinate transformations can all be expressed as
( ′ ) ( )( )
x cos θ − sin θ x
= (B.1)
y′ sin θ cos θ y
. /0 1 . /0 1 . /0 1
v′ eJ θ v

which is an αβ transformation. The dq transformation is obtained by substituting θ ⇒ −θ. While

coordinate transformations were difficult to handle when implemented in analog electronics, they
present little difficulties if implemented digitally. The tool which allows coordinate transformations
(or, perhaps more accurate, coordinate rotations) to be performed with ease is known as the
coordinate rotation digital computer (CORDIC) algorithm; invented by Volder in the 1950s [85].
The algorithm has since been used in almost all scientific calculators to calculate cosine and sine,
since this is merely a special case of (B.1) with x = 1, y = 0, yielding
( ′ ) ( )
x cos θ
= . (B.2)
y′ sin θ
Let us derive the CORDIC algorithm and see how it can be used to implement the αβ transfor-
mation.
It is obvious that the coordinate transformation (B.1) can be broken down into a series of N
coordinate rotations:
v ′ = eJθ v = eJα0 eJα1 · · · eJαN −1 v (B.3)
if α0 + α1 + · · · + αN −1 = θ. This can also be expressed as the iteration
vn+1 = eJαn vn , n = 0, 1, . . . , N − 1, v0 = v (B.4)
′
and v = vN . Equation (B.4) can be expressed as
( ) ( )
xn+1 xn cos αn − yn sin αn
=
yn+1 xn sin αn + yn cos αn
( )
xn − yn tan αn
= cos αn , n = 0, 1, . . . , N − 1. (B.5)
yn + xn tan αn
The key to obtaining a simple algorithm is to select the angles αn such that
*
2−n , αn > 0
tan αn = (B.6)
−2−n , αn < 0.
258 Appendix B. CORDIC Algorithm

This allows transformation angles in the range −99.9◦ < θ < 99.9◦ , since
, - , -
1 1
arctan(1) + arctan + arctan + · · · = 99.883◦. (B.7)
2 4

Then the multiplication by tan αn can be implemented merely as n binary right shifts, so the
algorithm now reads
( ) ( )
xn+1 xn − sgn (αn )yn 2−n
= kn , n = 0, 1, . . . , N − 1. (B.8)
yn+1 yn + sgn (αn )xn 2−n

where kn = cos(arctan 2−n ). The sign of αn is determined as follows. Suppose that a coordinate
rotation of θ = 60◦ is desired. For n = 0, the angle is advanced by arctan 1 = 45◦ , and for n = 1
by arctan 1/2 = 26.6◦. This yields a sum of 71.6◦ , so for n = 2, the angle has to be negative:
α2 = − arctan 1/4 = −14.0◦. The total angle after these three iterations is 57.6◦ (we are rapidly
converging to the correct angle!), so for n = 3, a positive angle is required. These steps can be
expressed as follows, using an error angle θn (which converges to zero):

θ0 = θ θn+1 = θn − αn sgn (αn+1 ) = sgn (θn+1 ). (B.9)

The only multiplication in the algorithm is that by kn . However, this multiplication need not be
made at each step; it can be implemented as a scaling by the constant
N
] −1
KN = cos(arctan 2−n ) (B.10)
n=0

made when initializing the algorithm, i.e.,

( ) ( )
x0 x
= KN
y0 y
( ) ( )
xn+1 xn − sgn (αn )yn 2−n
= , n = 0, 1, . . . , N − 1. (B.11)
yn+1 yn + sgn (αn )xn 2−n

KN is known as the aggregate constant and it converges to 0.6073... as N → ∞. (The convergence

is fast; the first four decimal positions remain unchanged for N > 7.)
The scaling by KN is thus the only multiplication needed when implementing the CORDIC
algorithm. (If only cosine and sine are to be computed, no multiplications at all are needed, since
then x = 1, y = 0.) To keep track of the angle increments αn , a look-up table for the function
arctan 2−n is needed. This table only requires N positions, however. The number of iterations N
does not have to be particularly large in order to obtain a good resolution. Eight iterations yields
cos 2−8 = 0.22◦ and ten iterations cos 2−10 = 0.056◦ . Clearly, N need not be selected larger than
the number of bits used in the A/D and D/A conversions of the stator current and voltage signals,
as this already puts a limit on the resolution.
We are now ready to present the complete CORDIC algorithm.

Algorithm B.1
CORDIC Algorithm
 259

( ) ( )
x x
= KN
y y
for n = 0 to N − 1
α = sgn (θ) arctan(2−n )
( ) ( ) ( )
x x −y
= + sgn (θ) 2−n
y y x
θ =θ−α
end
return [x, y]
260 Appendix B. CORDIC Algorithm

.
Bibliography

[1] M. Alaküla, “On the control of saturated synchronous machines,” Ph.D. Dissertation, Lund
Institute of Technology, Sweden, May 1993.

[2] E. Ch. Andresen and A. Haun, “Influence of the pulsewidth modulation control method on
the performance of frequency inverter induction motor drives,” Eur. Trans. Electr. Power,
Vol. 3, No. 2, Mar./Apr. 1993, pp. 151–161.

[3] G. J. Armstrong and D. J. Atkinson, “A comparison of model reference adaptive system

and extended Kalman filter estimators for sensorless vector drives,” Proc. Eur. Conf. Power
Electron. Appl., Trondheim, Norway, Vol. 1, Sept. 1997, pp. 424–429.

[4] K. J. Åström and B. Wittenmark, Computer-Controlled Systems: Theory and Design, 2nd
Ed., Prentice Hall, Englewood Cliffs, NJ, 1990.

[5] K. J. Åström and B. Wittenmark, Adaptive Control, 2nd Ed., Addison-Wesley, Reading, MA,
1995.

[6] F. Blaschke, “The principle of field orientation as applied to the new TRANSVECTOR closed-
loop control system for rotating-field machines,” Siemens Review, No. 2, 1972, pp. 217–223.

[7] R. Blasco-Giménez, G. M. Asher, M. Sumner, and K. J. Bradley, “Dynamic performance

limitations for MRAS based sensorless induction motor drives. Part 1: Stability and analysis
for the closed loop drive,” IEE Proc.–Electr. Power Appl., Vol. 143, No. 2, Mar. 1996, pp.
113–122

[8] R. Blasco-Giménez, G. M. Asher, M. Sumner, and K. J. Bradley, “Dynamic performance

limitations for MRAS based sensorless induction motor drives. Part 2: Online parameter
tuning and dynamic performance studies,” IEE Proc.–Electr. Power Appl., Vol. 143, No. 2,
Mar. 1996, pp. 123–134

[9] M. H. J. Bollen, Understanding Power Quality Problems: Voltage Sags and Interruptions,
IEEE Press, 2000.

[10] B. K. Bose (Ed.), Power Electronics and Variable Frequency Drives, IEEE Press, Piscataway,
NJ, 1996.

[11] B. K. Bose and N. R. Patel, “A programmable cascaded low-pass filter-based flux synthesis for
a stator flux-oriented vector-controlled induction motor drive,” IEEE Trans. Ind. Electron.,
IEEE Trans. Ind. Electron., Vol. 44, No. 2, Feb. 1997, pp. 140–143.

[12] F. Briz, M. W. Degner, and R. D. Lorenz, “Dynamic analysis of current regulators for ac
motors using complex vectors,” IEEE Trans. Ind. Appl., Vol. 35, No. 2, Nov./Dec. 1999, pp.
1424–1432.
262 BIBLIOGRAPHY

[13] G. S. Buja and M. P. Kazmierkowski, ‘Direct torque control of PWM inverter-fed ac motors—
A survey,” IEEE Trans. Ind. Electron., Vol. 51, No. 2, Aug. 2004, pp. 744–757.
[14] J.-W. Choi and S.-K. Sul, “New control concept–minimum time current control in the three-
phase PWM converter,” IEEE Trans. Power Electron., Vol. 12, No. 2, Jan. 1997, pp. 124–131.
[15] M. J. Corley and R. D. Lorenz, “Rotor position and velocity estimation for a permanent
magnet synchronous machine at standstill and high speeds,” IEEE Trans. Ind. Appl., Vol. 34,
No. 2, July–Aug. 1998, pp. 784–789.
[16] P. A. S. De Wit, R. Ortega, and I. Mareels, “Indirect field-oriented control of induction motors
is robustly globally stable,” Automatica, Vol. 32, No. 2, 1996, pp. 1393–1402.
[17] M. Depenbrock, “Direct self-control (DSC) of inverter-fed induction machine,” IEEE Trans.
Power Electron., Vol. 3, No. 2, Oct. 1988, pp. 420–429.
[18] M. Depenbrock, F. Hoffman, and S. Koch, “Speed sensorless high performance control for
traction drives,” in Proc. Eur. Conf. on Power Electron. and Applicat., Trondheim, Norway,
Sept. 1997, pp. 418–423.
[19] M. Depenbrock, C. Foerth, and S. Koch, “Speed sensorless control of induction motors at
very low stator frequencies,” in Proc. Eur. Conf. on Power Electron. and Applicat., Geneva,
Switzerland, Sept. 1999, CD-ROM.
[20] W. C. Duesterhoeft, M. W. Schulz, and E. Clarke, “Determination of instantaneous currents
and voltages by means of alpha, beta, and zero components,” Trans. Amer. Inst. Electr. Eng.,
vol. 70, no. 2, July 1951, pp. 1248–1255.
[21] A. Ferrah, K. J. Bradley, P. J. Hogben-Laing, M. S. Woolfson, G. M. Asher, M. Sumner, J.
Cilia, and J. Shuli, “A speed identifier for induction motor drives using real-time adaptive
digital filtering,” IEEE Trans. Ind. Appl., Vol. 34, No. 2, Jan./Feb. 1998, pp. 156–162.
[22] B. Gallwitz, F. Hillenbrand, and C. Landgraf, “A proposal for avoiding the direct measurement
of speed and angular position of the synchronous machine,” Proc. IFAC Control in Power
Electron. and Electrical Drives, Lausanne, Switzerland, 1983, pp. 63–68.
[23] L. J. Garcés, “Parameter adaptation for the speed-controlled static ac drive with a squirrel-
cage induction motor,” IEEE Trans. Ind. Appl., Vol. 16, No. 2, Mar./Apr. 1980, pp. 173–178.
[24] S. H. Hahn, Hilbert Transforms in Signal Processing, Artech House, 1996.
[25] D. C. Hanselman, “Resolver signal requirements for high accuracy resolver-to-digital conver-
sion,” IEEE Trans. Ind. Electron., Vol. 37, No. 2, Dec. 1990, pp. 556–561.
[26] L. Harnefors, “On analysis, control and estimation of variable-speed drives,” Ph.D. Disserta-
tion, Electrical Machines and Drives, Dept. of Electric Power Eng., Royal Institute of Technol.,
Stockholm, Sweden, 1997.
[27] L. Harnefors and H.-P. Nee, “Model-based current control of ac machines using the internal
model control method,” IEEE Trans. Ind. Appl., Vol. 34, No. 2, Jan./Feb. 1998, pp. 133–141.
[28] L. Harnefors and H.-P. Nee, “A general algorithm for speed and position estimation of ac
motors,” IEEE Trans. Ind. Electron., Vol. 47, No. 2, Feb. 2000, pp. 77–83.
BIBLIOGRAPHY 263

[29] L. Harnefors, “Instability phenomena and remedies in sensorless indirect field oriented con-
trol,” IEEE Trans. Power Electron., Vol. 15, No. 2, July 2000, pp. 733–743.

[30] L. Harnefors, K. Pietiläinen, and L. Gertmar, “Torque-maximizing field-weakening control:

design, analysis, and parameter selection,” IEEE Trans. Ind. Electron., Vol. 48, No. 2, Feb.
2001, pp. 161–168.

[31] L. Harnefors, “Design and analysis of general rotor-flux-oriented vector control systems,” IEEE
Trans. Ind. Electron., Vol. 48, No. 2, Apr. 2001, pp. 383–390.

[32] L. Harnefors, M. Jansson, R. Ottersten, and K. Pietiläinen, “Unified sensorless vector control
of synchronous and induction motors,” IEEE Trans. Ind. Electron., Vol. 50, No. 2, Feb. 2003,
pp. 153–160.

[33] L. Harnefors and R. Ottersten, “Regeneration-mode stabilization of the ‘statically compen-

sated voltage model’,” IEEE Trans. Ind. Electron., Vol. 54, No. 2, Apr. 2007, pp. 818–824.

[34] L. Harnefors and M. Hinkkanen, “Complete stability of reduced-order and full-order observers
for sensorless IM drives,” IEEE Trans. Ind. Electron., Vol. 55, No. 2, Mar. 2008, pp. 1319–
1329.

[35] L. Harnefors and M. Hinkkanen, “Stabilization of sensorless inductor motor drives—A survey,”
IEEE J. Emerg. Select. Topics Power Electron., Vol. 2, No. 2, Jun. 2014, pp. 132–142.

[36] K. Hasse, “Zur Dynamik drehzahlgeregelter Antriebe mit Stromrichter gespeisten Asynchron-
Kurzschluss laufermaschinen,” Ph.D. Dissertation (in German), Technische Hochschule Darm-
stadt, Darmstadt, Germany, 1969.

[37] A. M. Hava, S.-K. Sul, R. J. Kerkman, and T. A. Lipo, “Dynamic overmodulation charac-
teristics of triangle intersection PWM methods,” IEEE Trans. Ind. Appl., Vol. 35, No. 2,
July/Aug. 1999, pp. 896–907.

[38] R. C. Hilborn, Chaos and Nonlinear Dynamics: An Introduction for Scientists and Engineers.
London, U.K.: Oxford Univ. Press, 1994.

[39] M. Hinkkanen, “Analysis and design of full-order flux observers for sensorless induction mo-
tors,” IEEE Trans. Ind. Electron., Vol. 51, No. 2, Oct. 2004, pp. 1033–1040.

[40] M. Hinkkanen and J. Luomi, “Stabilization of regenerating-mode operation in sensorless in-

duction motor drives by full-order flux observer design,” IEEE Trans. Ind. Electron., Vol. 51,
No. 2, Dec. 2004, pp. 1318–1328.

[41] J. Holtz, “Sensorless control of induction motor drives,” Proc. IEEE, vol. 90, no. 8, pp. 1359–
1394, Aug. 2002.

[42] G.-C. Hsieh and J. C. Hung, “Phase-locked loop techniques—a survey,” Invited Paper, IEEE
Trans. Ind. Electron., Vol. 43, No. 6, pp. 609–615, Dec. 1996.

[43] K. D. Hurst and T. G. Habetler, “Sensorless speed measurement using current harmonics
spectral estimation in induction machine drives,” IEEE Trans. Power Electron., Vol. 11, No.
2, Jan. 1996, pp. 66–73.
264 BIBLIOGRAPHY

[44] P. L. Jansen and R. D. Lorenz, “A physically insightful approach to the design and accuracy
assessment of flux observers for field oriented induction machine drives,” IEEE Trans. Ind.
Appl., Vol. 30, No. 2, Jan./Feb. 1994, pp. 101–110.
[45] P. L. Jansen and R. D. Lorenz, “Transducerless position and velocity estimation in induction
and salient ac machines,” IEEE Trans. Ind. Appl., Vol. 31, No. 2, pp. 240–247, Mar./Apr.
1995.
[46] M. Jansson, L. Harnefors, O. Wallmark, and M. Leksell, “Synchronization at startup and
stable rotation reversal of nonsalient PMSM drives,” accepted to IEEE Trans. Ind. Electron.
[47] M. K. Kazimierczuk and D. Czarkowski, Resonant Power Converters. Wiley, 1995.
[48] H. K. Khalil, Nonlinear Systems, Macmillan Publishing, New York, 1992.
[49] S.-H. Kim and S.-K. Sul, “Maximum torque control of an induction machine in the field
weakening region,” IEEE Trans. Ind. Appl., Vol. 31, No. 2, July/Aug. 1995, pp. 787–794.
[50] S.-H. Kim and S.-K. Sul, “Voltage control strategy for maximum torque operation of an
induction machine in the field weakening region,” IEEE Trans. Ind. Electron., Vol. 44, No. 2,
Aug. 1997, pp. 512–518.
[51] K. P. Kovács, Transient Phenomena in Electrical Machines, Elsevier, Amsterdam, 1984.
[52] H. Kubota, K. Matsuse, and Y. Hori, “Behavior of sensorless induction motor drives in re-
generation mode,” in Proc. Power Conversion Conf., vol. 2, pp. 549–552, Nagaoka, Japan,
1997.
[53] H. Kubota, I. Sato, Y. Tamura, K. Matsuse, H. Ohta, and Y. Hori, “Regeneration-mode low-
speed operation of sensorless induction motor drive with adaptive observer,” IEEE Trans.
Ind. Applicat., vol. 38, no. 4, pp. 1081–1086, July/Aug. 2002.
[54] H. Kubota and Y. Tamura, “Stator resistance estimation for sensorless induction motor drives
under regenerating condition,” in Proc. IEEE IECON’02, vol. 1, pp. 426–430, Nov. 2002.
[55] J.-W. Lee, “An intelligent current controller using delay time compensation for PWM con-
verter”, in Proc. Eur. Conf. on Power Electron. and Applicat., Trondheim, Norway, pp. 342–
347, Sept. 1997.
[56] W. Leonhard, Control of Electrical Drives, 2nd Ed., Springer-Verlag, Berlin, 1996.
[57] T. Matsumoto, L. Chua, and M. Komuro, “The double scroll,” IEEE Trans. Circuits Syst.,
Vol. 32, No. 2, Aug. 1985, pp. 797–818.
[58] P. Mattavelli and F. P. Marafao, “Repetitive-based control for selective harmonic compen-
sation in active power filters,” IEEE Trans. Ind. Electron., Vol. 51, No. 2, Oct. 2004, pp.
1018–1024.
[59] R. H. Middleton and G. C. Goodwin, Digital Control and Estimation: A Unified Approach,
Prentice Hall, Englewood Cliffs, NJ, 1990.
[60] T. J. E. Miller, Switched Reluctance Motors and Their Control, Clarendon Press, Oxford,
1993.
BIBLIOGRAPHY 265

[61] N. Mohan, T. M. Undeland, and W. P. Robbins, Power Electronics: Converters, Applications

and Design, 3rd Ed. Wiley, 2003.

[62] M. Morari and E. Zafiriou, Robust Process Control, Prentice Hall, Englewood Cliffs, NJ, 1989.

[63] I. Nagy, “Chaotic processes in PWM using tolerance band control for voltage source inverters,”
Proc. Fifth IEE Int. Conf. Power Electron. Variab. Speed Drives, London, UK, Oct. 1994, pp.
66–71.

[64] R. Ottersten, “On control of back-to-back converters and sensorless induction machines
drives,” Ph.D. Dissertation, Department of Electric Power Eng., Chalmers Univ. of Tech-
nol., 2003.

[65] F.-Z. Peng and T. Fukao, “Robust speed identification for speed-sensorless vector control of
induction motors,” IEEE Trans. Ind. Appl., Vol. 30, No. 2, Sept./Oct. 1994, pp. Sept./Oct.
1994.

[66] K. Pietiläinen, L. Harnefors, A. Petersson, and H.-P. Nee, “DC-link stabilization and voltage
sag ride-through of inverter drives,” accepted to IEEE Trans. Ind. Electron.

[67] T. M. Rowan and R. J. Kerkman, “A new synchronous current regulator and an analysis of
current-regulated PWM inverters,” IEEE Trans. Ind. Appl., Vol. 22, No. 2, July/Aug. 1986,
pp. 678–690.

[68] C. Sadarangani, Electrical Machines – Design and Analysis of Induction and Permanent Mag-
net Motors, Electr. Machines and Power Electron., Dept. of Electr. Power Eng., Royal Inst.
of Technol., 2000.

[69] M. Saitou, M. Matsui, and N. Shimizu, “A control strategy of single-phase active filter using
a novel d-q transformation,” in Conf. Rec. IEEE Ind. Applicat. Soc. Annu. Meeting, Vol. 2,
Oct. 2003, pp. 1222–1227.

[70] S. Sangwongvanich, “Speed-sensorless vector torque control of induction motors—Stability

analysis and realization,” in Proc. IPEC, Yokohama, Japan, 2001, pp. 310–315.

[71] C. D. Schauder and R. Caddy, “Current control of voltage-source inverters for fast four-
quadrant drive performance,” IEEE Trans. Ind. Appl., Vol. 18, No. 2, 1982, pp. 163–171.

[72] C. Schauder, “Adaptive speed identification for vector control of induction motors without
rotational transducers,” IEEE Trans. Ind. Appl., Vol. 28, No. 2, Sept./Oct. 1992, pp. 1054–
1061.

[73] D. Seborg, T. Edgar, and D. Mellichamp, Process Dynamics and Control, Wiley, 1989.

[74] A. S. Sedra and P. O. Brackett, Filter Theory and Design: Active and Passive, Pitman
Publishing Ltd, 1978.

[75] W. Shepherd and L. Zhang, Power Converter Circuits. Marcel Dekker, 2004.

[76] M.-H. Shin, D.-S. Hyun, S.-B. Cho, and S.-Y. Choe, “An improved stator flux estimation
for speed sensorless stator flux orientation control of induction motors,” IEEE Trans. Power
Electron., Vol. 15, No. 2, Mar. 2000, pp. 312–318.
266 BIBLIOGRAPHY

[77] S. Skogestad and I. Postlethwaite, Multivariable Feedback Control, John Wiley & Sons, Chich-
ester, England, 1996.
[78] K. Stockman, F. D’hulster, and R. Belmans, “Torque capability of a field oriented induction
motor drive under voltage sag conditions,” in Proc. 10th Eur. Conf. on Power Electron. and
Applicat., Toulouse, France, 2003, CD-ROM.
[79] T. Svensson, “On modulation and control of electronic power convertors,” Ph.D. Dissertation,
Chalmers Univ. of Technol., Gothenburg, Sweden, 1988.
[80] I. Takahashi and T. Noguchi, “A new quick-response and high-efficiency control strategy of
an induction motor,” IEEE Trans. Ind. Appl., Vol. 22, No. 2, Sept./Oct. 1986, pp. 820–827.
[81] R. Teodorescu, F. Blaabjerg, M. Liserre, and P. C. Loh, “Proportional-resonant controllers
and filters for grid-connected voltage-source converters,” IEE Proc.–Electr. Power Appl., Vol.
153, No. 5, Sep. 2006, pp. 750–762
[82] P. Tiitinen, P. Pohjalainen, and J. Lalu, “The next generation motor control method: Direct
Torque Control (DTC),” EPE Journal, Vol. 5, No. 1, Mar. 1995, pp. 14–18.
[83] E. Twining and D. G. Holmes, “Grid current regulation of a three-phase voltage source inverter
with an LCL input filter,” IEEE Trans. Power Electron., Vol. 18, No. 2, May 2003, pp. 888–
895.
[84] G. C. Verghese and S. R. Sanders, “Observers for flux estimation in induction machines,”
IEEE Trans. Ind. Electron., Vol. 35, No. 2, Feb. 1988, pp. 85–94.
[85] J. E. Volder, “The CORDIC trigonometric computing technique,” IRE Trans. Electron. Com-
put., Vol. 8, No. 3, pp. 330–334, 1959.
[86] O. Wallmark, L. Harnefors, and O. Carlson, “An improved speed and position estimator for
salient permanent-magnet synchronous motors,” IEEE Trans. Ind. Electron., Vol. 52, No. 2,
Feb. 2005, pp. 255–262.
[87] X. Xu, R. de Doncker, and D. W. Novotny, “Stator flux orientation control of induction ma-
chines in the field weakening region,” in Conf. Rec. IEEE Ind. Applicat. Soc. Annu. Meeting,
Vol. 1, Oct. 1988, pp. 437–443.
[88] C.-H. Yim, I.-J. Ha, and M.-S. Ko, “A resolver-to-digital conversion method for fast tracking,”
IEEE Trans. Ind. Electron., Vol. 39, No. 2, Oct. 1992, pp. 369–378.
BIBLIOGRAPHY 267