The Design of

High Performance Mechatronics

This page intentionally left blank
 The Design of
High Performance
Mechatronics
High-Tech Functionality by
Multidisciplinary System Integration

Robert Munnig Schmidt

Georg Schitter
Jan van Eijk

Delft University Press

© 2011 The authors and IOS Press. All rights reserved.

ISBN 978-1-60750-825-0 (print)

ISBN 978-1-60750-826-7 (online)
doi:10.3233/978-1-60750-826-7-i

Published by IOS Press under the imprint Delft University Press

IOS Press BV
Nieuwe Hemweg 6b
1013 BG Amsterdam
The Netherlands
tel: +31-20-688 3355
fax: +31-20-687 0019
email: [email protected]
www.iospress.nl

LEGAL NOTICE
The publisher is not responsible for the use which might be made of the
following information.

PRINTED IN THE NETHERLANDS

Contents

Preface xvii
Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . xix
Summary of the contents . . . . . . . . . . . . . . . . . . . . . . . xx

1 Mechatronics in the Dutch high-tech industry 1

1.1 Historical background . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 The Video Long-play Disk (VLP) . . . . . . . . . . . . . . 3
1.1.1.1 Signal encoding and read-out principle . . . . 4
1.1.1.2 The Compact Disc and its family members . . 6
1.1.2 The Silicon Repeater . . . . . . . . . . . . . . . . . . . . . 9
1.1.2.1 IC manufacturing process . . . . . . . . . . . . 10
1.1.2.2 The accurate wafer stage . . . . . . . . . . . . . 13
1.1.3 The impact of mechatronics on our world . . . . . . . . . 16
1.2 Definition and international positioning . . . . . . . . . . . . . 17
1.2.1 Different views on mechatronics . . . . . . . . . . . . . . 18
1.2.1.1 Cultural differences in mechatronics . . . . . . 18
1.2.1.2 Focus on precision-controlled motion . . . . . . 21
1.3 Systems engineering and design . . . . . . . . . . . . . . . . . . 22
1.3.0.3 Definitions and V-model . . . . . . . . . . . . . 23
1.3.0.4 The product creation process . . . . . . . . . . . 26
1.3.0.5 Requirement budgeting . . . . . . . . . . . . . . 28
1.3.0.6 Roadmapping . . . . . . . . . . . . . . . . . . . . 29
1.3.1 Design methodology . . . . . . . . . . . . . . . . . . . . . . 31
1.3.1.1 Concurrent engineering . . . . . . . . . . . . . . 32
1.3.1.2 Modular design and platforms . . . . . . . . . . 34

2 Electricity and frequency 37

2.1 Electricity and signals . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1.1 Electric field . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

v
vi Contents

2.1.1.1 Potential difference . . . . . . . . . . . . . . . . . 40

2.1.1.2 Electric field in an electric element . . . . . . . 42
2.1.2 Electric current and voltage . . . . . . . . . . . . . . . . . 43
2.1.2.1 Voltage source . . . . . . . . . . . . . . . . . . . . 43
2.1.2.2 Electric power . . . . . . . . . . . . . . . . . . . . 45
2.1.2.3 Ohm’s law . . . . . . . . . . . . . . . . . . . . . . 46
2.1.2.4 Practical values and summary . . . . . . . . . 47
2.1.3 Variability of electric signals . . . . . . . . . . . . . . . . 48
2.1.3.1 The concept of frequency . . . . . . . . . . . . . 49
2.1.3.2 Random signals or noise . . . . . . . . . . . . . 52
2.1.3.3 Power of alternating signals . . . . . . . . . . . 53
2.1.3.4 Representation in the complex plane . . . . . . 54
2.2 Energy propagation and waves . . . . . . . . . . . . . . . . . . . 57
2.2.1 Mechanical and acoustic waves . . . . . . . . . . . . . . . 57
2.2.2 Electromagnetic waves . . . . . . . . . . . . . . . . . . . . 59
2.2.2.1 Transferred energy and amplitude . . . . . . . 60
2.2.3 Reflection of waves . . . . . . . . . . . . . . . . . . . . . . . 61
2.2.3.1 Standing waves . . . . . . . . . . . . . . . . . . . 63
2.3 Mathematical analysis of signals and dynamics . . . . . . . . . 65
2.3.1 Fourier transform . . . . . . . . . . . . . . . . . . . . . . . 65
2.3.1.1 Triangle waveform . . . . . . . . . . . . . . . . . 67
2.3.1.2 Sawtooth waveform . . . . . . . . . . . . . . . . 68
2.3.1.3 Square waveform . . . . . . . . . . . . . . . . . . 69
2.3.1.4 Fourier analysis of non-periodic signals . . . . 70
2.3.2 Laplace transform . . . . . . . . . . . . . . . . . . . . . . . 72
2.4 Dynamic system response to a stimulus . . . . . . . . . . . . . . 74
2.4.0.1 Step response . . . . . . . . . . . . . . . . . . . . 74
2.4.0.2 Impulse response . . . . . . . . . . . . . . . . . . 76
2.4.0.3 Frequency response . . . . . . . . . . . . . . . . 78
2.4.1 Graphical representation in the frequency domain . . 79
2.4.1.1 Bode-plot . . . . . . . . . . . . . . . . . . . . . . . 79
2.4.1.2 Nyquist plot . . . . . . . . . . . . . . . . . . . . . 83

3 Dynamics of motion systems 85

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.1 Stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.1.1 Importance of stiffness for precision . . . . . . . . . . . . 88
3.1.2 Active stiffness . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.2 Mass-spring systems with damping . . . . . . . . . . . . . . . . 95
3.2.1 Compliance of dynamic elements . . . . . . . . . . . . . . 95
Contents vii

3.2.2 Combining dynamic elements . . . . . . . . . . . . . . . . 96

3.2.3 Transfer functions of the compliance . . . . . . . . . . . 101
3.2.3.1 Damped mass-spring system. . . . . . . . . . . 102
3.2.3.2 Critical damping and definition of ζ . . . . . . 106
3.2.3.3 Quality-factor Q . . . . . . . . . . . . . . . . . . . 110
3.2.3.4 Behaviour around the natural frequency . . . 112
3.2.4 Transmissibility of a damped mass-spring system . . . 114
3.3 Multi-body dynamics and eigenmodes . . . . . . . . . . . . . . . 118
3.3.1 Dynamics of a two body mass-spring system . . . . . . 119
3.3.1.1 Analytical description . . . . . . . . . . . . . . . 119
3.3.1.2 Multiplicative expression . . . . . . . . . . . . . 121
3.3.1.3 Effect of different mass ratios . . . . . . . . . . 122
3.3.2 The additive method with eigenmodes . . . . . . . . . . 125
3.3.2.1 Multiple eigenmodes and modal analysis . . . 129
3.3.2.2 Location of actuators and sensors . . . . . . . 131
3.3.2.3 Summary . . . . . . . . . . . . . . . . . . . . . . . 135

4 Motion Control 137

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.1 A walk around the control loop . . . . . . . . . . . . . . . . . . . 138
4.1.1 Poles and zeros . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.1.1.1 Controlling unstable mechanical systems . . . 140
4.1.1.2 Creating instability by active control . . . . . 141
4.1.1.3 The zeros . . . . . . . . . . . . . . . . . . . . . . . 142
4.1.2 Properties of feedforward control . . . . . . . . . . . . . . 143
4.1.3 Properties of feedback control . . . . . . . . . . . . . . . . 146
4.2 Feedforward control . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.2.1 Model based open-loop control . . . . . . . . . . . . . . . 149
4.2.2 Input-shaping . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.2.3 Adaptive feedforward control . . . . . . . . . . . . . . . . 155
4.3 PID feedback control . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.3.1 PD-control of a Compact-Disc player . . . . . . . . . . . 157
4.3.1.1 Proportional feedback . . . . . . . . . . . . . . . 158
4.3.1.2 Proportional-differential feedback . . . . . . . 161
4.3.1.3 Limiting the differentiating action . . . . . . . 163
4.3.2 Sensitivity functions of feedback control . . . . . . . . . 167
4.3.3 Stability and robustness in feedback control . . . . . . . 170
4.3.4 PID-control of a mass-spring system . . . . . . . . . . . 174
4.3.4.1 P-control . . . . . . . . . . . . . . . . . . . . . . . 175
4.3.4.2 D-control . . . . . . . . . . . . . . . . . . . . . . . 176
viii Contents

4.3.4.3 I-control . . . . . . . . . . . . . . . . . . . . . . . . 177

4.3.5 PID-control of more complex systems . . . . . . . . . . . 182
4.3.5.1 PID-control of a magnetic bearing . . . . . . . 182
4.3.5.2 Eigenmodes above the desired bandwidth . . 187
4.3.5.3 “Optimal” PID control . . . . . . . . . . . . . . . 192
4.3.5.4 Open-loop and closed-loop . . . . . . . . . . . . 193
4.4 State-space control representation . . . . . . . . . . . . . . . . . 195
4.4.1 State-space in relation to motion control . . . . . . . . . 196
4.4.1.1 Damped mass-spring system . . . . . . . . . . . 198
4.4.1.2 PID-control feedback . . . . . . . . . . . . . . . . 200
4.4.2 State feedback . . . . . . . . . . . . . . . . . . . . . . . . . 202
4.4.2.1 System Identification . . . . . . . . . . . . . . . 204
4.4.2.2 State estimation . . . . . . . . . . . . . . . . . . 205
4.4.2.3 Additional remarks on state-space control . . 207
4.5 Limitations of linear feedback control . . . . . . . . . . . . . . . 209
4.6 Conclusions on motion control . . . . . . . . . . . . . . . . . . . . 214

5 Electromechanic actuators 215

5.1 Electromagnetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
5.1.0.4 History on magnetism . . . . . . . . . . . . . . . 217
5.1.1 Maxwell equations . . . . . . . . . . . . . . . . . . . . . . . 218
5.1.2 Magnetism caused by electric current . . . . . . . . . . . 222
5.1.3 Hopkinson’s law . . . . . . . . . . . . . . . . . . . . . . . . 225
5.1.3.1 Ferromagnetic materials . . . . . . . . . . . . . 228
5.1.4 Coil with ferromagnetic yoke . . . . . . . . . . . . . . . . 229
5.1.4.1 Magnetisation curve . . . . . . . . . . . . . . . . 230
5.1.5 Permanent magnets . . . . . . . . . . . . . . . . . . . . . . 231
5.1.5.1 Thermal behaviour and Curie temperature . 235
5.1.6 Creating a magnetic field in an air-gap . . . . . . . . . . 235
5.1.6.1 Optimal use of a permanent magnet . . . . . . 239
5.1.6.2 Flat magnets to reduce stray flux . . . . . . . . 239
5.1.6.3 Low cost loudspeaker magnet configuration . 241
5.2 Lorentz actuator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
5.2.1 Lorentz force . . . . . . . . . . . . . . . . . . . . . . . . . . 243
5.2.2 Improving the force of a Lorentz actuator . . . . . . . . 247
5.2.3 The moving-coil loudspeaker actuator . . . . . . . . . . . 248
5.2.4 Position dependency of the Lorentz force . . . . . . . . . 248
5.2.4.1 Over-hung and under-hung coil . . . . . . . . . 250
5.2.5 Electronic commutation . . . . . . . . . . . . . . . . . . . 251
5.2.5.1 three-phase electronic control . . . . . . . . . . 253
Contents ix

5.2.6 Figure of merit of a Lorentz actuator . . . . . . . . . . . 254

5.3 Reluctance actuator . . . . . . . . . . . . . . . . . . . . . . . . . . 257
5.3.1 Reluctance force in Lorentz actuator . . . . . . . . . . . 257
5.3.1.1 Eddy-current ring . . . . . . . . . . . . . . . . . 258
5.3.1.2 Ironless stator . . . . . . . . . . . . . . . . . . . . 259
5.3.2 Analytical derivation of the reluctance force . . . . . . . 260
5.3.3 Variable reluctance actuator. . . . . . . . . . . . . . . . . 264
5.3.3.1 Electromagnetic relay . . . . . . . . . . . . . . . 266
5.3.3.2 Force exerted by a magnetic field . . . . . . . . 267
5.3.4 Hybrid actuator . . . . . . . . . . . . . . . . . . . . . . . . 268
5.3.4.1 Double variable reluctance actuator . . . . . . 268
5.3.4.2 Combining two sources of magnetic flux . . . . 271
5.3.4.3 Hybrid force calculation . . . . . . . . . . . . . . 273
5.3.4.4 Magnetic bearings . . . . . . . . . . . . . . . . . 276
5.4 Application of electromagnetic actuators . . . . . . . . . . . . . 279
5.4.1 Electrical interface properties . . . . . . . . . . . . . . . 279
5.4.1.1 Dynamic effects of self-inductance . . . . . . . 279
5.4.1.2 Limitation of the “jerk” . . . . . . . . . . . . . . 281
5.4.1.3 Damping caused by source impedance . . . . . 282
5.4.2 Comparison of the actuation principles . . . . . . . . . . 285
5.4.2.1 Standard coil dimension for the comparison . 286
5.4.2.2 Force of the Lorentz actuator . . . . . . . . . . 288
5.4.2.3 Force of the reluctance actuator . . . . . . . . . 288
5.4.2.4 Force of the hybrid actuator . . . . . . . . . . . 289
5.4.2.5 Dynamic differences . . . . . . . . . . . . . . . . 289
5.4.2.6 Moving mass . . . . . . . . . . . . . . . . . . . . . 290
5.5 Intermezzo: electric transformers . . . . . . . . . . . . . . . . . . 291
5.5.1 Ideal transformer . . . . . . . . . . . . . . . . . . . . . . . 292
5.5.2 Real transformer . . . . . . . . . . . . . . . . . . . . . . . . 294
5.6 Piezoelectric actuators . . . . . . . . . . . . . . . . . . . . . . . . . 296
5.6.1 Piezoelectricity . . . . . . . . . . . . . . . . . . . . . . . . . 296
5.6.1.1 Poling . . . . . . . . . . . . . . . . . . . . . . . . . 297
5.6.2 Transducer models . . . . . . . . . . . . . . . . . . . . . . 298
5.6.3 Nonlinearity of piezoelectric transducers . . . . . . . . . 301
5.6.3.1 Creep . . . . . . . . . . . . . . . . . . . . . . . . . 301
5.6.3.2 Hysteresis . . . . . . . . . . . . . . . . . . . . . . 302
5.6.3.3 Aging . . . . . . . . . . . . . . . . . . . . . . . . . 303
5.6.4 Mechanical considerations . . . . . . . . . . . . . . . . . . 304
5.6.4.1 Piezo-stiffness . . . . . . . . . . . . . . . . . . . . 304
5.6.4.2 Actuator types . . . . . . . . . . . . . . . . . . . . 305
x Contents

5.6.4.3 Actuator integration . . . . . . . . . . . . . . . . 307

5.6.4.4 Mechanical amplification . . . . . . . . . . . . . 308
5.6.4.5 Multiple directions by stacking . . . . . . . . . 309
5.6.5 Electrical considerations . . . . . . . . . . . . . . . . . . . 311
5.6.5.1 Charge vs. voltage control . . . . . . . . . . . . 311
5.6.5.2 Self-sensing actuation . . . . . . . . . . . . . . . 312

6 Analogue electronics in mechatronic systems 315

6.1 Passive electronics . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
6.1.1 Network theory and laws . . . . . . . . . . . . . . . . . . . 317
6.1.1.1 Voltage source . . . . . . . . . . . . . . . . . . . . 317
6.1.1.2 Current source . . . . . . . . . . . . . . . . . . . 318
6.1.1.3 Theorem of Norton and Thevenin . . . . . . . . 319
6.1.1.4 Kirchhoff’s laws . . . . . . . . . . . . . . . . . . . 320
6.1.1.5 Impedances in series or parallel . . . . . . . . . 321
6.1.1.6 Voltage divider . . . . . . . . . . . . . . . . . . . 322
6.1.1.7 Maximum power of a real voltage source . . . 324
6.1.2 Impedances in electronic networks . . . . . . . . . . . . . 326
6.1.2.1 Resistors . . . . . . . . . . . . . . . . . . . . . . . 326
6.1.2.2 Capacitors . . . . . . . . . . . . . . . . . . . . . . 328
6.1.2.3 Inductors . . . . . . . . . . . . . . . . . . . . . . . 333
6.1.3 Passive filters . . . . . . . . . . . . . . . . . . . . . . . . . . 335
6.1.3.1 Passive first-order RC-filters . . . . . . . . . . . 335
6.1.3.2 Passive higher-order RC-filters . . . . . . . . . 338
6.1.3.3 Passive LCR-filters . . . . . . . . . . . . . . . . . 340
6.1.4 Mechanical-electrical dynamic analogy . . . . . . . . . . 346
6.2 Active electronics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
6.2.1 Basic discrete semiconductors . . . . . . . . . . . . . . . 351
6.2.1.1 Semiconductor diode . . . . . . . . . . . . . . . . 354
6.2.1.2 Bipolar transistors . . . . . . . . . . . . . . . . . 357
6.2.1.3 MOSFET . . . . . . . . . . . . . . . . . . . . . . . 359
6.2.2 One transistor amplifiers . . . . . . . . . . . . . . . . . . . 361
6.2.2.1 Emitter follower . . . . . . . . . . . . . . . . . . . 362
6.2.2.2 Voltage amplifier . . . . . . . . . . . . . . . . . . 364
6.2.2.3 Differential amplifier . . . . . . . . . . . . . . . 367
6.2.3 Operational amplifier . . . . . . . . . . . . . . . . . . . . . 370
6.2.3.1 Basic operational amplifier design . . . . . . . 370
6.2.3.2 Operational amplifier with feedback . . . . . . 372
6.2.4 Linear amplifiers with operational amplifiers . . . . . . 373
6.2.4.1 Design rules . . . . . . . . . . . . . . . . . . . . . 373
Contents xi

6.2.4.2 Non-inverting amplifier . . . . . . . . . . . . . . 374

6.2.4.3 Inverting amplifier . . . . . . . . . . . . . . . . . 376
6.2.4.4 Adding and subtracting signals . . . . . . . . . 377
6.2.4.5 Transimpedance amplifier . . . . . . . . . . . . 379
6.2.4.6 Transconductance amplifier . . . . . . . . . . . 381
6.2.5 Active electronic filters . . . . . . . . . . . . . . . . . . . . 383
6.2.5.1 Integrator and first-order low-pass filter . . . 383
6.2.5.2 Differentiator and first-order high-pass filter 385
6.2.6 Analogue PID controller . . . . . . . . . . . . . . . . . . . 387
6.2.6.1 Transfer function . . . . . . . . . . . . . . . . . . 388
6.2.6.2 Control gains . . . . . . . . . . . . . . . . . . . . 390
6.2.6.3 High speed PID control . . . . . . . . . . . . . . 390
6.2.7 Higher-order electronic filters . . . . . . . . . . . . . . . . 391
6.2.7.1 Second-order low-pass filter . . . . . . . . . . . 392
6.2.7.2 Second-order high-pass filter . . . . . . . . . . 393
6.2.7.3 Different types of active filters . . . . . . . . . . 393
6.2.8 Ideal and real properties of operational amplifiers . . . 395
6.2.8.1 Dynamic limitations . . . . . . . . . . . . . . . . 395
6.2.8.2 Limitations on the inputs . . . . . . . . . . . . . 401
6.2.8.3 Power supply and output limitations . . . . . . 404
6.2.9 Closing remarks on low-power electronics . . . . . . . . 405
6.3 Power amplifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
6.3.1 General properties of power amplifiers . . . . . . . . . . 408
6.3.2 Linear power amplifiers . . . . . . . . . . . . . . . . . . . 411
6.3.2.1 High output impedance amplifiers . . . . . . . 413
6.3.2.2 Dynamic loads, four-quadrant operation . . . 417
6.3.3 Switched-mode power amplifiers . . . . . . . . . . . . . . 419
6.3.3.1 First example amplifier . . . . . . . . . . . . . . 419
6.3.3.2 Power MOSFET, a fast high-power switch . . 422
6.3.3.3 Pulse-width modulation . . . . . . . . . . . . . . 424
6.3.3.4 High-power output stage . . . . . . . . . . . . . 428
6.3.3.5 Preliminary conclusions and other issues . . . 432
6.3.3.6 Driving the power MOSFETs . . . . . . . . . . 432
6.3.3.7 Charge-pumping . . . . . . . . . . . . . . . . . . 434
6.3.3.8 Dual-ended configuration . . . . . . . . . . . . . 435
6.3.3.9 Output filter . . . . . . . . . . . . . . . . . . . . . 437
6.3.4 Resonant-mode power amplifiers . . . . . . . . . . . . . . 438
6.3.4.1 Switching sequence of the output stage . . . . 440
6.3.4.2 Lossless current sensing . . . . . . . . . . . . . 444
6.3.5 Three-phase amplifiers . . . . . . . . . . . . . . . . . . . . 444
xii Contents

6.3.5.1 Concept of three-phase amplifier . . . . . . . . 445

6.3.5.2 Three-phase switching power stages . . . . . . 446
6.3.6 Some last remarks on electronics . . . . . . . . . . . . . . 447

7 Optics in mechatronic systems 449

7.1 Properties of light and light sources . . . . . . . . . . . . . . . . 451
7.1.1 Light generation by thermal radiation . . . . . . . . . . 452
7.1.2 Photons by electron energy state variation . . . . . . . . 453
7.1.2.1 Light emitting diodes . . . . . . . . . . . . . . . 455
7.1.2.2 Laser as an ideal light source . . . . . . . . . . 456
7.1.3 Useful power from a light source . . . . . . . . . . . . . . 460
7.1.3.1 Radiant emittance and irradiance . . . . . . . 461
7.1.3.2 Radiance . . . . . . . . . . . . . . . . . . . . . . . 461
7.1.3.3 Etendue . . . . . . . . . . . . . . . . . . . . . . . . 464
7.2 Reflection and refraction . . . . . . . . . . . . . . . . . . . . . . . 465
7.2.1 Reflection and refraction according to the least time . 465
7.2.1.1 Partial reflection and refraction . . . . . . . . . 469
7.2.2 Concept of wavefront . . . . . . . . . . . . . . . . . . . . . 470
7.2.2.1 A wavefront is not real . . . . . . . . . . . . . . 471
7.3 Geometric Optics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
7.3.1 Imaging with refractive lens elements . . . . . . . . . . 473
7.3.1.1 Sign conventions . . . . . . . . . . . . . . . . . . 475
7.3.1.2 Real lens elements . . . . . . . . . . . . . . . . . 476
7.3.1.3 Magnification . . . . . . . . . . . . . . . . . . . . 479
7.3.2 Aberrations . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
7.3.2.1 Spherical aberration . . . . . . . . . . . . . . . . 481
7.3.2.2 Astigmatism . . . . . . . . . . . . . . . . . . . . . 483
7.3.2.3 Coma . . . . . . . . . . . . . . . . . . . . . . . . . 485
7.3.2.4 Geometric and chromatic aberrations . . . . . 485
7.3.3 Combining multiple optical elements . . . . . . . . . . . 487
7.3.3.1 Combining two positive lenses . . . . . . . . . . 488
7.3.4 Aperture stop and pupil . . . . . . . . . . . . . . . . . . . 491
7.3.5 Telecentricity . . . . . . . . . . . . . . . . . . . . . . . . . . 493
7.3.5.1 Pupil in a telecentric system . . . . . . . . . . . 494
7.3.5.2 Practical applications and constraints . . . . . 495
7.4 Physical Optics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
7.4.1 Polarisation . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
7.4.1.1 Birefringence . . . . . . . . . . . . . . . . . . . . 499
7.4.2 Interference . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
7.4.2.1 Fabry-Perot interferometer . . . . . . . . . . . . 503
Contents xiii

7.4.3
Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505
7.4.3.1 Amplitude gratings . . . . . . . . . . . . . . . . 506
7.4.3.2 Phase gratings . . . . . . . . . . . . . . . . . . . 508
7.4.3.3 Direction of the incoming light . . . . . . . . . 515
7.4.4 Imaging quality based on diffraction . . . . . . . . . . . 515
7.4.4.1 Numerical aperture and f-number . . . . . . . 519
7.4.4.2 Depth of focus . . . . . . . . . . . . . . . . . . . . 522
7.5 Adaptive optics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525
7.5.1 Thermal effects in optical imaging systems . . . . . . . 525
7.5.2 Correcting the wavefront . . . . . . . . . . . . . . . . . . . 527
7.5.2.1 Zernike modes . . . . . . . . . . . . . . . . . . . . 528
7.5.2.2 Adaptive optics as correction mechanism . . . 532
7.5.3 Principle of operation . . . . . . . . . . . . . . . . . . . . . 533

8 Measurement in mechatronic systems 537

8.1 Introduction to measurement systems . . . . . . . . . . . . . . . 539
8.1.1 Errors in measurement systems, uncertainty . . . . . . 539
8.1.1.1 The ultimate in uncertainty . . . . . . . . . . . 541
8.1.2 Functional model of a measurement system element . 542
8.2 Dynamic error budgeting . . . . . . . . . . . . . . . . . . . . . . . 544
8.2.1 Error statistics in repeated measurements . . . . . . . 544
8.2.2 The normal distribution . . . . . . . . . . . . . . . . . . . 545
8.2.3 Combining different error sources . . . . . . . . . . . . . 547
8.2.4 Power spectral density and cumulative power . . . . . . 548
8.2.5 Cumulative amplitude . . . . . . . . . . . . . . . . . . . . 550
8.2.5.1 Variations on dynamic error budgeting . . . . 551
8.2.6 Sources of noise and disturbances . . . . . . . . . . . . . 551
8.2.6.1 Mechanical noise . . . . . . . . . . . . . . . . . . 552
8.2.6.2 Electronic noise . . . . . . . . . . . . . . . . . . . 552
8.2.6.3 Using noise data from data-sheets . . . . . . . 554
8.3 Sensitive signals in measurement systems . . . . . . . . . . . . 556
8.3.1 Sensing element . . . . . . . . . . . . . . . . . . . . . . . . 557
8.3.2 Converting an impedance into an electric signal . . . . 558
8.3.2.1 Wheatstone bridge . . . . . . . . . . . . . . . . . 559
8.3.3 Electronic interconnection of sensitive signals . . . . . 565
8.3.3.1 Magnetic disturbances . . . . . . . . . . . . . . 565
8.3.3.2 Capacitive disturbances . . . . . . . . . . . . . . 568
8.3.3.3 Ground loops . . . . . . . . . . . . . . . . . . . . 569
8.4 Signal conditioning . . . . . . . . . . . . . . . . . . . . . . . . . . . 571
8.4.1 Instrumentation amplifier . . . . . . . . . . . . . . . . . . 571
xiv Contents

8.4.2 Filtering and modulation . . . . . . . . . . . . . . . . . . . 574

8.4.2.1 AM with square wave carrier . . . . . . . . . . 575
8.4.2.2 AM with sinusoidal carrier . . . . . . . . . . . . 576
8.5 Signal processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 579
8.5.1 Schmitt trigger . . . . . . . . . . . . . . . . . . . . . . . . . 579
8.5.2 Analogue-to-Digital conversion . . . . . . . . . . . . . . . 580
8.5.2.1 Gray code . . . . . . . . . . . . . . . . . . . . . . . 581
8.5.2.2 Sampling of analogue values . . . . . . . . . . . 583
8.5.2.3 Nyquist-Shannon theorem . . . . . . . . . . . . 584
8.5.2.4 Filtering to prevent aliasing . . . . . . . . . . . 587
8.5.3 Analogue-to-digital converters . . . . . . . . . . . . . . . 587
8.5.3.1 Dual-slope ADC . . . . . . . . . . . . . . . . . . . 589
8.5.3.2 Successive-approximation ADC . . . . . . . . . 590
8.5.3.3 Delta-Sigma ADC . . . . . . . . . . . . . . . . . 593
8.5.4 Connecting the less sensitive elements . . . . . . . . . . 596
8.5.4.1 Characteristic impedance . . . . . . . . . . . . . 596
8.5.4.2 Non-galvanic connection . . . . . . . . . . . . . 598
8.6 Short-range motion sensors . . . . . . . . . . . . . . . . . . . . . 599
8.6.1 Optical sensors . . . . . . . . . . . . . . . . . . . . . . . . . 599
8.6.1.1 Position sensitive detectors . . . . . . . . . . . . 600
8.6.1.2 Optical deflectometer . . . . . . . . . . . . . . . 603
8.6.2 Capacitive position sensors . . . . . . . . . . . . . . . . . 605
8.6.2.1 Linearising by differential measurement . . . 606
8.6.2.2 Accuracy limits and improvements . . . . . . . 607
8.6.2.3 Sensing to conductive moving plate . . . . . . 610
8.6.3 Inductive position sensors . . . . . . . . . . . . . . . . . . 611
8.6.3.1 Linear variable differential transformer . . . 613
8.6.3.2 Eddy-current sensors . . . . . . . . . . . . . . . 615
8.7 Dynamic measurements of mechanical quantities . . . . . . . 617
8.7.1 Measurement of force and strain . . . . . . . . . . . . . . 617
8.7.1.1 Strain gages . . . . . . . . . . . . . . . . . . . . . 618
8.7.1.2 Fibre Bragg grating strain measurement . . . 620
8.7.2 Velocity measurement . . . . . . . . . . . . . . . . . . . . 623
8.7.2.1 Geophone . . . . . . . . . . . . . . . . . . . . . . . 624
8.7.3 Accelerometers . . . . . . . . . . . . . . . . . . . . . . . . . 628
8.7.3.1 Closed-loop feedback accelerometer . . . . . . 628
8.7.3.2 Piezoelectric accelerometer . . . . . . . . . . . . 630
8.7.3.3 MEMS accelerometer . . . . . . . . . . . . . . . 638
8.8 Optical long-range incremental position sensors . . . . . . . . 641
8.8.1 Linear optical encoders . . . . . . . . . . . . . . . . . . . . 642
Contents xv

8.8.1.1 Interpolation . . . . . . . . . . . . . . . . . . . . . 646

8.8.1.2 Vernier resolution enhancement . . . . . . . . 648
8.8.1.3 Interferometric optical encoder . . . . . . . . . 650
8.8.1.4 Concluding remarks on linear encoders . . . . 654
8.8.2 Laser interferometer measurement systems . . . . . . . 656
8.8.2.1 Homodyne distance interferometry . . . . . . . 657
8.8.2.2 Heterodyne distance interferometry . . . . . . 662
8.8.2.3 Measurement uncertainty . . . . . . . . . . . . 672
8.8.2.4 Different configurations . . . . . . . . . . . . . . 678
8.8.3 Mechanical aspects . . . . . . . . . . . . . . . . . . . . . . 684
8.8.3.1 Abbe error . . . . . . . . . . . . . . . . . . . . . . 685

9 Precision positioning in wafer scanners 687

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 687
9.1.1 The wafer scanner . . . . . . . . . . . . . . . . . . . . . . . 689
9.1.2 Requirements on precision . . . . . . . . . . . . . . . . . . 691
9.2 Dynamic architecture . . . . . . . . . . . . . . . . . . . . . . . . . 695
9.2.1 Balance masses . . . . . . . . . . . . . . . . . . . . . . . . . 696
9.2.2 Vibration isolation . . . . . . . . . . . . . . . . . . . . . . . 698
9.2.2.1 Eigendynamics of the sensitive parts . . . . . 701
9.3 Zero stiffness stage actuation . . . . . . . . . . . . . . . . . . . . 705
9.3.1 The wafer stage actuation concept . . . . . . . . . . . . . 706
9.3.1.1 Wafer stepper long-range Lorentz actuator . . 706
9.3.1.2 Multi-axis positioning . . . . . . . . . . . . . . . 709
9.3.1.3 Long- and short-stroke actuation . . . . . . . . 710
9.3.2 Full magnetic levitation . . . . . . . . . . . . . . . . . . . 713
9.3.3 Limits in acceleration of reticle stage . . . . . . . . . . . 714
9.4 Position measurement . . . . . . . . . . . . . . . . . . . . . . . . . 716
9.4.0.1 The alignment sensor . . . . . . . . . . . . . . . 718
9.4.1 Keeping the wafer in focus . . . . . . . . . . . . . . . . . . 720
9.4.2 Dual-stage measurement and exposure . . . . . . . . . . 722
9.4.3 Long-range incremental measurement system . . . . . 723
9.4.3.1 Real-time metrology loop . . . . . . . . . . . . . 724
9.5 Motion control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 727
9.5.1 Feedforward and feedback control . . . . . . . . . . . . . 728
9.5.2 The mass dilemma . . . . . . . . . . . . . . . . . . . . . . . 730
9.6 Main design rules for precision . . . . . . . . . . . . . . . . . . . 731
xvi Contents

Appendix 733
Recommended other books . . . . . . . . . . . . . . . . . . . . . . . . . 733
Nomenclature and abbreviations . . . . . . . . . . . . . . . . . . . . . 737
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 746
Preface

Motivation

A world without mechatronics is almost as unthinkable as a world with-

out electric light. After its origin around the second world war the name
mechatronics has become known for all kind of mechanical systems where
mechanics and electronics are combined to achieve a certain function. The
complexity of mechatronics ranges from a simple set of electronic controlled
relay-switches to highly integrated precision motion systems. This prolifer-
ation of mechatronics has been accompanied by many books that each have
been written with a different scope in mind depending on the specific tech-
nological anchor point of the author(s) within this wide multidisciplinary
field of engineering.
The book that you are reading now distinguishes itself from other books
in several ways. First of all it is written as a balancing act between both
the industrial and the academic background of the authors. The industrial
part is based on extensive experience in designing the most sophisticated
motion systems presently available, the stages of wafer scanners that are
used in the semiconductor industry. The academic part is based on ad-
vanced research on ultra precision metrology equipment with fast Scanning-
Probe Microscopy and optical measurements with sub-nanometre accuracy.
Closely related to the industrial background is the focus on high precision
positioning at very high velocity and acceleration levels. With this focus, the
book does not include other important applications like robotics and vehicle
mechatronics. All presented material is focused on obtaining a maximum
of control of all dynamic aspects of a motion system. This is the reason for
the term “High Performance” in the title.
A second reason for writing this book next to all others is the observation,
when teaching engineering at the university, that most students are rather
well trained in applying mathematical rules but too often fail to understand

xvii
xviii Preface

the full potential of these mathematics in real mechanical designs. The

need for the education of real engineers with both theoretical and practical
skills, combined with a healthy critical attitude to the outcome of computer
simulations, became a guiding motive to finish the tedious job of writing.
The capability to swiftly switch between model and reality is one of the most
important skills of a real multidisciplinary designer. This capability helps
to quickly predict the approximate system behaviour in the concept phase
of a design, where intuition and small calculations on the backside of an
envelope are often more valuable than computer based detailed calculations
by means of sophisticated modelling software. It is certainly true that these
software tools are indispensable for further detailing and optimisation in
the later phase of a design project but more attention is needed for basic
engineering expert-knowledge to cover the concept-design phase where the
most important design decisions are taken.
In view of these main motivations to write this book, it was also decided to
focus uniquely on the hardware part of mechatronic systems. This means
that the important field of embedded software is not presented even though
software often serves as the actual implementation platform for modern con-
trol systems. The reason for this exclusion is the intended focus of this book
on the prime functionality of a mechatronic system, without the interfaces
to other systems and human operators. The logical sequence algorithm of
the controller, together with the sampling delay, is more important for this
prime functionality than the way how this algorithm is described in C-code.
When writing a book on mechatronics, the broad range of contributing
disciplines forces a limitation on the theoretical depth to which the theory on
each of these disciplines can be treated. Where necessary for the explanation
of certain effects the presented material goes somewhat deeper, but most
subjects are treated in such a way that an overall understanding is obtained
that is based on first principles rather than on specialised in depth knowledge
of all details.
Like the work of a mechatronic engineer as system designer in a team of
specialists, this book is aimed to be rather a binding factor to the related
specialised books than one that makes these redundant.
It is our sincere hope this book serves its purpose.
Also on behalf of the co-authors Georg Schitter and Jan van Eijk,
Robert Munnig Schmidt,
author/editor July 2011
Preface xix

Contributions and acknowledgements

Besides much material from our own experience, this book also includes
material created by many other people.
Several university staff members and students have contributed to and
reviewed the material. Unfortunately it is impossible to mention all without
forgetting some names so as example only the three most important students
are mentioned.
The first is Ton de Boer, who accepted the impossible task as MSc-student to
write the rough material that started this book as lecture notes by following
the lectures on Mechatronic System Design. Initially, in spite of professional
advice, this writing was done in a well-known WYSIWYG program and only
later it was transferred into LATEXwhich proved indeed the only realistic
way to create a professional technical textbook.
Leon Jabben and Jonathan Ellis have been working as PhD-students at our
laboratory in Delft and parts of their theses are used in the measurement
chapter.
Our partners from industry deserve gratitude for their support, financially,
in equipment or advice, by permission to use company illustrations or by
reviewing the material. The three most important to mention are the Dutch
high-tech company ASML and the metrology companies Heidenhain from
Germany and Agilent Technologies from the United States.
From ASML especially Hans Butler, Patrick Tinnemans and Jan Mulkens
(thanks to the volcano on Iceland!) have helped in reviewing some chapters.
We further thank all other companies and individuals that kindly gave
permission to use their illustrations. These all are separately mentioned at
the related figures.
It is true to say that this textbook is based on the knowledge of many others
as laid down in books, patents and journal articles. For reason of readability
we decided not to include references in the text but instead we included a
list of the most relevant books that we found to be applicable.
Finally also a word of respect and gratitude should be given to the many
contributors of Wikipedia. Even though this huge source of information is
not always as consistent and flawless as might be required by the scientific
community, Wikipedia has proven to be very useful to quickly find the
right physical and mathematical terms or derivations. It also provided
information about small trivia like the date of birth or the full name of a
famous scientist from the past.
xx Preface

Short summary and introduction of the contents

This book is written in such a way that it is useful both for a high-level
student who wants to learn about advanced mechatronics and for engineers
in the high-tech industry who want to learn more about adjacent special-
isations. To accommodate this dual approach, the first and last chapter
determine the environment that makes use of the material of the theoretical
chapters in between. It is not a surprise that this first and last chapter are
connected by the wafer scanners of ASML as these might well be the most
advanced mechatronic systems that are ever designed.
The nine chapters are summarised as follows:
The introduction in Chapter 1 gives the context of mechatronics in the
Dutch high-tech industry with the historical background, some general
observations on the international differences in approach towards mecha-
tronics and the close link with “Systems Engineering”. Subjects include
the development of the optical Video Long Play (VLP) disk and the wafer
stepper at Philips Research Laboratories. These developments have strongly
determined the dominant foothold of high-precision mechatronic system
design in the Netherlands and are exemplary for the specific photon-physics
oriented approach in this country, so quite different from the machining
oriented approach in most other countries. The overview on systems en-
gineering and design introduces some functional design and development
methodologies that have proved to be crucial for the success of the high-tech
industry. These methods are based on industrial practice where complex
multidisciplinary designs have to be realised. Systems Engineering is a
field closely related to mechatronics and the corresponding principles are
used in structuring the design of a mechatronic system.
Chapter 2 is the first of a series of chapters on the basic theory that is applied
in controlled motion systems. It consists of a short overview of the principles
of electricity, frequencies, waves and signal responses. The chapter
starts with basic electricity, the linking element in a mechatronic system.
Followed by signal theory this chapter explains the reason why the properties
of mechatronics are so often described in the frequency domain next to
the more mechanical oriented time-related step and impulse responses.
The chapter also introduces different graphical representations of these
responses as this material is used throughout most chapters in this book.
The hard-core of a mechatronic system is still the mechanics that represent
the real, dynamic, hardware world that has to be mastered when position-
ing objects in a controlled way. In most cases, the dynamic properties of
Preface xxi

the mechanical construction determine the control performance. Expert

knowledge of this field is a prerequisite for a mechatronic designer. For
that reason Chapter 3 deals with these dynamics of motion systems and
mainly concentrates on the uncontrolled properties of standard mechanical
elements consisting of a multitude of springs, masses and dampers. As a
first step towards active motion control this theory enables to determine
dynamic causes for observed instability issues in controlled motion systems.
Immediately related to the mechanical dynamics is the important field of
active motion control in Chapter 4. This chapter concentrates on a thor-
ough understanding of the working principle and tuning of the still widely
used PID controllers. Also a short introduction is given in more modern
model-based control approaches that are expected to play an increasing
role in mechatronic systems. A strong emphasis is put on the insight that
control both adds virtual elements from the mechanical domain like springs
and dampers and new elements like integration.
Electromechanic actuators and analogue electronics are two closely
related hardware components of a mechatronic system. Their interaction is
increasingly underestimated by system designers, because of two reasons.
Firstly the field is controlled by experts in physics and electronics. These
specialists have a fundamentally different more abstract frame of view than
the mostly concrete-mechanical visually oriented system designers. The
second reason for underestimating these related fields is caused by the
overwhelming amount of electronics and electro-motors that are around
us, giving rise to the idea that their principle is simple and mastered by
many. This idea is a dangerous delusion as the difficulty in electronics is
related to its dynamic analogue behaviour and unfortunately the number
of people that master that part is rather decreasing than increasing. It is
the analogue side of electronics that deals with measurement and actuation
that needs most of the attention of the mechatronic designer.
With this purpose in mind Chapter 5 first presents linear electrome-
chanic actuators. This chapter mainly focuses on electromagnetic actua-
tors but also piezoelectric actuators are presented as these are increasingly
applied in precision mechatronic systems. This chapter will help in the
selection process of actuation systems and creates a knowledge base for
further study on the subject. Also the relation with power-amplifier con-
straints, that are presented in the following chapter, is made clear.
Chapter 6 deals with analogue electronics for measurement and power
and starts at a very basic level with passive components because most me-
chanical engineering students have hardly any knowledge about electronics.
xxii Preface

The introduction of the active components leads to their application in the

basic design of the operational amplifier, the most universal and widely used
analogue electronic building block. The last section in this large chapter
gives an overview of the basic design of Power Amplifiers that act as the
interface between the controller and the actuators.
Optics has become a main driver of mechatronic advancement in the past
decades and for that reason Chapter 7 gives an introduction to optics from
the perspective of a mechatronic designer. Optics are important in two ways.
Firstly it is an application area where mechatronics are used to control and
correct optical properties of imaging systems and other instrumentation.
Secondly optics are used to determine distances in a plurality of sensors, that
enables us to create measurement systems with extreme precision. Starting
with basic physics on optics with sources and the duality of light, an overview
of geometrical and physical optics is presented including limiting factors
for the performance of imaging systems. The chapter concludes with an
introduction on adaptive optics.
Chapter 8 presents the basic principles of sensors for force and dynamic
position measurements based on several physical principles including
strain-, inductive-, capacitive- and optical sensors. The theory in this chapter
will enable the first selection of suitable sensors when designing a mecha-
tronic system. Laser interferometry and encoders will also be presented as
these are most frequently applied in high precision mechatronic systems.
Even though metrology in general will be shortly touched, the chapter con-
centrates on measurement for control. For this reason also the principle
of Dynamic Error Budgeting is included, a statistical method to determine
the total error in a dynamic precision system from contributions of different
error sources.
As closure of the book Chapter 9 presents the mechatronic design for pre-
cision positioning in waferscanners where all theory is applied to its
most extreme level. This chapter includes the basic design of positioning
stages, the need for and active control of vibration isolation, and the motion
control approach to achieve a position accuracy of less them a nanometre at
speeds of more than 1 m/s and accelerations of more than 30 m/s2 .
Chapter 1

Mechatronics in the
Dutch high-tech industry

This introductory chapter places the subject of this book in the context of
the rapid development of the high-tech industry in the Netherlands. This
industry has become a main driver of the economic growth in the industrial
region around Eindhoven with the multinational company “Royal Philips
Electronics” as the most prominent original source of technical innovation.
Based on their wide market scope with both consumer and industrial prod-
ucts Philips gave birth to several high-tech spin-off companies among which
ASML would become the global market leader in exposure equipment for
semiconductor manufacturing. The first part of this chapter will pay tribute
to the achievements within the research and development departments of
“Philips Gloeilampen fabrieken”, as the company was named at that time,
because these have determined the success of high performance mechatron-
ics in both high-tech and consumer equipment. After this short historical
overview, the second section will position the field of mechatronics on the
international playground where precision machining and manufacturing
appear to be the main application fields. The last section deals with systems
engineering as a design and development framework for the highly complex
mechatronic systems that are applied in the high-tech industry. It will also
give a general overview on the related development processes.

1
2 Chapter 1. Mechatronics in the Dutch high-tech industry

1.1 Historical background

High performance mechatronics originated in the Netherlands around 1970 –

1980 at Philips Research Laboratories near Eindhoven. Although already
world famous at the time, this laboratory was humbly called the “Nat-
Lab”, a name that was based on the original Dutch name “Natuurkundig
Laboratorium”. The research activities were characterised by multidisci-
plinary teamwork on subjects with an estimated large application potential.
Even though frequently this potential was far from proven and might take
a very long time to reach reality, research people with vision and ambition
got the opportunity to work on their dream.
The multidisciplinary character of the research subjects was especially ev-
ident in the Optical Research Group where in those early days a unique
combination of people worked together in close harmony on the optical Video
Long Play Disk (VLP) and the Silicon Repeater. These people were not
consciously aware of the fact that they were pioneering in a new field of
expertise. A field that later would be known under the name “mechatronics”.

Figure 1.1: The Philips “Nat Lab” research laboratories in Eindhoven where the
real breakthrough in mechatronics was initiated.
(source: Philips Technisch Tijdschrift Vol.43 nr 2/3/4)
1.1. Historical background 3

Figure 1.2: The VLP Optical disk player had a contact-less optical read-out principle
with light from a Helium-Neon laser that was focused on the track of
the disk.
(source: Philips Technisch Tijdschrift Vol.43 nr 2/3/4)

1.1.1 The Video Long-play Disk (VLP)

In those days, only the gramophone record player and cassette recorder
were known for music playback and home registration, while for video
only the video cassette recorder (VCR) existed. Even though pre-recorded
videotapes would eventually become very popular, the general opinion was
at that time that tapes would always be too expensive for the consumer
market, due to the large number of parts and expensive material in one
tape-cassette. There were already several people in the world with the vision
that a gramophone record for video would fundamentally solve that problem,
because moulding a gramophone record is not costly at all. Unfortunately,
the registration of video images in analogue technology would have required
a bandwidth of more than 4 MHz for normal Television signals and that
is more than a hundred times higher than needed for the reproduction of
sound with a maximum frequency of 20 kHz in two-channel stereo. Based
on the fact that even this 20 kHz was quite difficult to achieve with a contact
stylus with diamond tip, including the unavoidable contamination of the
grooves, it was not considered possible to register video in a comparable
way! The breakthrough in thinking, that was required for the future success,
was based in the understanding that the contact method would need to be
replaced by a contact-less read-out principle with light. With the then newly
developed laser as a light source and the application of precision optics, a far
higher density of information could be registered than would ever be possible
4 Chapter 1. Mechatronics in the Dutch high-tech industry

Figure 1.3: The signal on a VLP disk is coded in pits on a spiral wound track with
a variable length and frequency at a fixed width and mutual distance.
The frequency of the pits determined the video signal and the length of
the pitsdetermined the audio signal.

by mechanical means. The benefit of contact-less read-out of a rotating disk,

as shown in Figure 1.2, becomes even more clear with a small calculation
on the registration of high-end audio with a 22 kHz signal on a gramophone
record that rotates with 33 rounds per minute. If the signal is located on the
inner groove of the disk at a diameter of 130 mm, one period of this signal
corresponds with a track-length on the disk of approximately 10 μm. The tip
radius of the diamond stylus can not be made much smaller than this same
order of magnitude, because otherwise the lifetime would be too short. The
tip radius effectively limits the detection of a shorter wavelength. Even with
the most refined methods, as are presently applied in the recently revived
gramophone players, the maximum frequency that can be registered on a
gramophone record is limited to about 40 kHz. With suitable optics and
a Helium Neon Laser source with a wavelength of 633 nanometre it was
found to be possible to detect details on a rotating disk with just less than
a micrometre. Compared with the 10 μm of the mechanical record player,
this value is less than one hundredth of the surface area per detail, which
is sufficient for the registration of an hour of video information on a disk
with the size of a normal gramophone record.

1.1.1.1 Signal encoding and read-out principle

For reason of efficiency, it was decided to directly write the analogue video
signal on the disk as a series of pits, tiny stripes with a variable frequency
and length. While the frequency contained the video information, the sound
was registered in the length of the pits as shown in Figure 1.3.
The pits were written in a spiral-wound track with a width of 0,4 μm at
a mutual distance of 1,6 μm on a transparent disk with the same size
as an ordinary gramophone record, a diameter of 0.3 m. The rotating
speed would be a factor ten higher however, with up to 30 revolutions per
second, in order to be able to handle the high frequency. Also different to
1.1. Historical background 5

Opcal sensor Protecng layer



Laser Lens

Semi-transparent
mirror
Out of focus scratch 


Figure 1.4: The basic optical read-out principle of an optical disk by focusing a laser
beam on the track that is embedded in the optical disk and measuring
the reflected intensity with an optical sensor. A scratch on the surface
of the protecting layer is not detected because it is not in focus.

the gramophone record was the location of the track, that no longer was
embedded in a groove, but was hidden inside the disk and detected by means
of a small spot of light through the transparent covering layer, as shown
in Figure 1.4. This really appeared to be a revolutionary way of thinking.
Some of the competitors, like RCA, still tried to read the information written
in a groove on the surface with the help of a stylus. In this case the stylus
was only used to follow the track, while the high frequency signal was
detected with a local capacitive sensor on the tip of the stylus. With this
splitting of functions, following and detecting, they did succeed in getting
the mechanical system operational and it has even been some time on the
market with well protected disks in cassettes to prevent contamination.
This product was in itself a significant achievement for that time but when
compared with the contact-less optical read-out system of Philips it could
not survive. Especially the lack of sensitivity for surface scratches appeared
to be one of the big advantages of optical detection. The scratches are not
detected because they are not in the focal point of the detecting laser beam. A
problem with this principle was however the fact that no mechanical means
were available to follow the track and this created the need to look for ways
to actively control the position of the optical pick-up unit that was designed
to read the information on the disk. It was the solution to this problem
that determined the crucial breakthrough of precision mechatronics in the
Netherlands. From the signal of a special segmented photo-diode with some
additional optics, the relative position of the optical pick-up unit to the
track could be measured and with tiny linear moving-coil motors, similar
to the ones that are used in loudspeakers, it became possible to keep the
6 Chapter 1. Mechatronics in the Dutch high-tech industry

Opcal disk
Opcal Controller Drive
pick-up unit
Posion
Posion error
Loudspeaker Measurement
coils
Magnec
system

Figure 1.5: The active position control of the optical pickup-unit by measuring the
position from information in the reflected light and correcting position
errors by means of moving coil actuators.

pick-up unit right on track by means of a real fast position control system, as
shown in Figure 1.5. This so called servo-system was capable to correct the
different disturbances and imperfections in the track. Today this correction
is achieved even in the cheapest CD-players with special digital processors
in one IC that perform all the control tasks, but at that time this control
system could only be realised with operational amplifiers, RC-networks and
adjustable resistors.

1.1.1.2 The Compact Disc and its family members

The VLP was the first product that was brought on the market with optical
recording but in spite of the cooperation with MCA for the movie software,
the introduction under the name of “Laservision” was eventually not success-
ful. This was mainly caused by the lower than envisaged cost of the video
cassette, while the VCR-player could also record video signals directly from
television. The missing, at that time not sufficiently developed, possibility
to record optical disks on a consumer product, appeared to be the most
important drawback of the Laservision disk. This problem would be solved
some time later with recordable CDs but that came only after the second
breakthrough that reached almost every human being in the world, the in-
vention of the audio Compact Disc player of which one of the first commercial
products is shown in Figure 1.6. With many successful and less successful
successors like CD-Interactive, CDROM, CDRAM, DVD Superaudio CD
and Blu-Ray disk, this development really has made a major difference to
the entertainment industry. With the CD three major developments were
1.1. Historical background 7

Figure 1.6: The CD 303 was the first real high-end CD player of Philips. The
complete optical read-out mechanism with the rotating motor for the
CD and the rotary arm that carried the optical pick-up unit, was built
into a separate drawer that would enable to load the disk from the front.
This innovation would not have been easily possible with a normal
gramophone record player.

united, the digitisation of analogue signals, the optical disk principle of the
VLP and the invention of the semiconductor laser, that enabled a far smaller
design of the optical system. As an example of the miniaturisation, that
accompanied these developments, the complete mechanism of a CD-player
and the optical pick-up unit from a Blu-ray player are shown in Figure 1.7.
The design principles that are applied in these systems with the long-stroke,
short-stroke splitting of precision and range will return in the last chapter
of this book that describe the more recent developments in wafer scanners
that started as described in the following section.
8 Chapter 1. Mechatronics in the Dutch high-tech industry

Figure 1.7: The optical pick-up unit in the CD-mechanism on top of this picture is
carried by a simple and inexpensive linear moving stage that is driven
with a rotating motor and a screw-nut/gearwheel transmission. This
long-stroke linear drive roughly positions the pick-up unit to within the
capture range of the precision track-following servo-system with the
moving-coil motors of the pick-up unit. This basic long-stroke, short-
stroke positioning concept is refined in later developments like the
Blu-ray disk of which the optical pick-up unit is shown below the CD-
mechanism. This unit is designed to be compatible with all previous CD
and DVD related formats. For that reason three different laser sources
are integrated in one unit, working at three different wavelengths
while all light is optically combined to one spot by means of several
semi-reflecting cubes and lenses.
1.1. Historical background 9

Figure 1.8: The OM 200 was the first integrated circuit made by Philips in 1965.
The size was 0.75×0.75 mm2 and it integrated three transistors and
two resistors.
(source: Philips Technisch Tijdschrift Vol.43 nr 2/3/4)

1.1.2 The Silicon Repeater

Around the same time as optical recording was invented, Philips produced
its own semiconductors. To support this activity, within the Natlab several
research groups were active in the global battle to decrease the size of
the details in semiconductor based Integrated Circuits (ICs). The first
commercial integrated circuit of Philips is shown in Figure 1.8 with only a
very small number of components, but gradually the number of components
increased dramatically with sometimes millions of transistors, resistors
and capacitors that have to be realised on a single silicon substrate. These
electronic elements are all connected by a multitude of different layers with
wiring on top of the active elements, the lighter regions in Figure 1.8.
The technical support and design group within the Natlab supported this
research with in-house designed and manufactured precision equipment,
like the “Opticograph” that could write a mask for an IC by means of a
scanning focused light beam. Among these machines, the Silicon Repeater, of
which two generations are shown in Figure 1.9, was designed for exposing the
pattern of an integrated circuit. The first generation (Mark 1) was designed
together with the engineering department of Philips Semiconductor in
Nijmegen, while Silicon Repeater Mark 2 was purely made inside the Natlab.
10 Chapter 1. Mechatronics in the Dutch high-tech industry

Figure 1.9: The Silicon Repeaters Mark 1 and Mark 2 were the first wafer steppers
of Philips and achieved an imaging resolution in the micrometre range.
They were designed as laboratory tools to enable advanced research on
IC production technology.
(source: Philips Technisch Tijdschrift Vol.37 nr 11/12)

1.1.2.1 IC manufacturing process

Both for the realisation of the active electronic components in the IC and
the related wiring, it is necessary to work the silicon by means of etching,
oxidising or changing the properties locally by different chemical elements.
The related process is called lithography which name is based on the ancient
Greek words “λιϑ oς” = stone and “γραϕειν” = writing.
The first step in this process is to “print” the desired pattern by means of
an optical exposure of a light sensitive resist layer. This exposure is done
by an optical system that images the mask (also called a reticle) on the
resist. Subsequently, after the development of the resist, the silicon can be
chemically treated on the spots where the resist was illuminated, as shown
in Figure 1.10. It is essential to note that the same step has to be repeated
up to around thirty times, where each different layers has to be defined in
one exposure cycle. All exposed layers have to be positioned with a high
accuracy relative to the previous layers. In Figure 1.11 a schematic overview
of this IC manufacturing process is shown.
1.1. Historical background 11

2: Develop resist
UV light

Recle

3: Etch Oxide
Resist
Si oxide
Si wafer

1: Expose resist 4: Remove resist

Figure 1.10: The lithographic process is a local chemical treatment of the silicon
substrate. The treated areas are determined by a resist layer that
contains a pattern that is previously imaged from the pattern on a
mask by an optical exposure system.

Exposure Developing
Photoresist and baking

 

 


Slicing
Polishing

Etching and ion Repeated cycles



Removing
Completed
the photoresist
wafer  Packaging

Figure 1.11: IC production flow starting with a grown mono-crystalline Silicon

ingot cut into thin slices (wafers) that undergo a multitude of chemical
treatments. The details of the integrated circuit are determined by
the optical exposure system at step 5.
12 Chapter 1. Mechatronics in the Dutch high-tech industry

Light source+ light shaping





 ! 
"
"

#""

Figure 1.12: Schematic drawing of the main components of a wafer stepper. The
mask is imaged on the wafer with a demagnifying lens. Due to the
high opening angle that is required for the high resolution only small
areas can be exposed at one exposure at the same time. This requires
the wafer to be exposed in steps by means of a wafer stage.

The smallest details of an IC determine both its energy consumption and

functionality. Together with the continuous increase in productivity of
the manufacturing process for cost reduction, these factors are the most
important economic drivers in the semiconductor industry. Both factors
are mainly determined by the exposure system. Though all other process
steps also need to be capable of realising a high level of refinement, the
exposure step is the only one that deals with the detailing of the circuits.
For this reason the exposure step has been of overriding importance for the
developments in the semiconductor industry. Initially the exposure was
done by direct illumination through a mask that was closely positioned
above the wafer, like shown in Figure 1.10. This shadow mask exposure
principle is however limited in resolution by the laws of diffraction and
suffers from vulnerability of the previous layers for damage by touching the
surface of the mask. To solve that issue, the wafer stepper and the wafer
scanner were invented.
The main components of a wafer stepper are shown in Figure 1.12. The
optical projection system is located in the centre. Like a slide projector,
1.1. Historical background 13

that projects light through a transparency and a lens to the wall, a wafer
stepper projects light through a reticle and lens to a silicon wafer. The only
difference is the demagnification of the image on the silicon wafer which
is related to the required high resolution. To realise a high resolution, the
opening angle of the lens needs to be very high as will be explained in
Chapter 7 on optics. For this reason the image has to be very close to the
lens and light should come from all directions. Exposing the entire wafer in
one step would require a lens with a very large diameter which would be too
expensive to manufacture. It was concluded that only a small area could be
exposed in one exposure at the same time. This means that the wafer has
to be exposed in steps while it should be positioned very accurately between
the different exposures by means of a mechanism, the wafer stage. This
stepping motion has to be done extremely fast to not lose time and keep the
productivity on an acceptable level. For that reason the speed of this highly
accurate movement became one of the key aspects that drove the need for
perfection of mechatronics in these machines.

1.1.2.2 The accurate wafer stage

The most important and difficult part from a mechatronic perspective was
and still is the wafer stage. The accuracy of the positioning of the wafer in
the Silicon Repeater was in the same order of magnitude as the accuracy
of the previously described Video Long-play Disk with an error of signifi-
cantly less than a micrometre over a range of several tens of centimetres.
Instead of following a fast moving track, the wafer in a waferstepper had to
be positioned in steps followed by a perfect stand-still during exposure. This
standing-still had to be realised with a position repeatability of better than
0.1 μm on a wafer of at that time only 5 inch (≈ 12 cm), which is a better
accuracy than one over a million.
The first wafer stages were driven by hydraulic linear motors as schemat-
ically shown in Figure 1.13. Precision hydraulics was a known field of
expertise within Philips, originating from the design of high precision ma-
chining tools that were needed in the laboratory to manufacture for instance
optical parts. Hydraulic cylinders and servo-valves were a proven technology
for high precision positioning under several strict conditions. Friction had to
be avoided in the hydrostatic bearings and the temperature of the hydraulic
oil had to be actively controlled to around 100 mK in order not to influence
the measurement of the position by laser interferometers. With this wafer
stage a repeatability of 0.1 μm was realised by using special valves that
control the differential pressure over the hydraulic motors without any back-
14 Chapter 1. Mechatronics in the Dutch high-tech industry

Figure 1.13: The first wafer stages had hydraulic linear motors to achieve the
required stiffness, speed and accuracy. (source: commercial leaflet
Philips)

lash by internal friction. In spite of the good performance, hydraulic drives

with the necessary high pressure pumps and oil sweating sealings were not
the most ideal mechanical parts to be applied in a clean room of an IC fac-
tory. One incident with a bursting high pressure tube on a prototype in the
laboratory made everyone aware of the problems related to clean everything
again. For that reason an electric alternative was perceived to be a far better
solution and in parallel, research was done on an alternative electric stage,
driven with rotating servo-motors and a friction-wheel transmission. The
prototype that was built with this principle showed problems with dynamics
due to the transmissions and low stiffness connections and eventually a
direct-drive method provided the real replacement of the hydraulic drive.
The design of this direct-drive wafer stage was based on the research on the
Video Long-play Disk with its magnetically suspended optical pick-up unit.
It was recognised that those same principles could also be applied to larger
positioning systems, notwithstanding the intuitive drawback of the large
mass. At that time linear motion was mostly achieved by screw spindles
and rotating motors like the example with friction wheels. In principle the
position repeatability of such indirect drives is determined by the manufac-
turing precision of the parts and generally these are not accurate enough
for micrometre accuracy. By applying a position control system errors could
1.1. Historical background 15

Figure 1.14: With electric linear motors the wafer stage became “cleanroom-
friendly”.

be reduced to a level of approximately 1 μm, which was mainly limited by

backlash, dynamics and friction. Experiments with piezoelectric actuators,
to correct the remaining error, were not successful yet at that time, so the
only solution that remained was to adapt the direct control principle of the
VLP pick-up unit.
It is interesting to be aware of the following observation related to this step.
The frame of thought of the traditional mechanical engineer was (and often
still is!) based on the understanding that for precision something has to be
stiffly connected on an adjustment mechanism that by it self is determinis-
tically constrained in position. With the optical pick-up unit the approach
was however completely different. In principle the optical pick-up unit is not
connected to anything, while the position is determined by exerting electro-
magnetic forces that are controlled by the difference between the actual and
the desired position, thus creating a virtual active stiff connection between
the moving part and the track. This more abstract understanding that an
accurate position can be realised with virtual stiffness appeared not easy to
understand for people that are accustomed to realise more rigid mounted
objects. Possibly for that reason only one real mechanical engineer was
allowed in the optical research group at the Natlab and one of the authors of
this book was one of those! The early breakthroughs in mechatronics have
16 Chapter 1. Mechatronics in the Dutch high-tech industry

been realised mainly by physicists and electronic engineers. An interesting

observation!
Through the application of the principles from the VLP research in the
so much larger system of the wafer stage it became possible to design a
superior electric alternative for the hydraulic motors that took away one
of the largest obstacles for customer acceptance of the first Philips wafer
steppers. For this reason this electrical linear motor driven wafer stage,
as shown in Figure 1.14, is symbolic for the success of the mechatronic
discipline in the Netherlands.

1.1.3 The impact of mechatronics on our world

The described two subjects of research, in their mutual mechatronic re-

lation within the same research group, have resulted in two significant
technological developments with an influence far beyond the borders of the
Netherlands. A world without optical registration of data, images and sound
has become almost unimaginable and the Philips wafer stepper has grown
within the Philips daughter ASML to the machine that determined the
necessary conditions for the overwhelming developments in the electronic
industry. The wafer stepper and its successor, the wafer scanner will be
further presented in Chapter 9 at the end of the book where all the theory
on mechatronics comes together.
1.2. Definition and international positioning 17

1.2 Definition and international positioning

Mechatronic design is a discipline within mechanical engineering that com-

bines the classical mechanical disciplines of static and dynamic mechanics,
thermodynamics, metrology and tribology with historically non-mechanical
disciplines like electronics, software and optics. This combination is cer-
tainly more than just the sum of these different disciplines. First and
foremost, mechatronics should be considered to represent technology in-
tegration to achieve optimal system functionality. In fact a complete
mechatronic product can only achieve its desired functionality through a
process of systematic integration of all inherent disciplines, right through
from the conceptual phase. Mechatronics opens up enormous technological
possibilities, as already evidenced in the previous section by the appear-
ance of sophisticated products like wafer scanners and compact disc players.
These products would never have been realised by only a traditional single
disciplinary approach.
Mechatronics is hence a real multidisciplinary field of expertise, a funda-
mental way of realising a certain specific function where in most cases
controlled motion is determinative for the result. Being a design discipline,
mechatronics focuses on integration and synthesis rather than analysis,
though the latter is indispensable for a good understanding. Like shown in
a humorist way in Figure 1.15, mechatronics aims to give a more optimal
functionality than would be feasible by only focusing on one specialisation.

Mechanical Electronic / Electrical Pneumac Hydraulic Chemical

Thermal Opcal Acouscal So$ware

Figure 1.15: Essentially mono disciplinary solutions to a problem. They work, but
is the solution optimal?
18 Chapter 1. Mechatronics in the Dutch high-tech industry

1.2.1 Different views on mechatronics

Many definitions exist for the field of mechatronics. The first example finds
its origin in Japan where the term mechatronics was first officially deposited:

The planned application and efficient integration of

• mechanical and electronic technology

• in a multi-disciplinary and integrated approach of
• product and process design

to optimise production.

This clearly is more focused on production rather than product development

although it is both targeting application and integration.
From Europe comes the following definition which is closer to our under-
standing:

A synergistic combination of

• precision mechanical engineering

• electronic control
• systems thinking

in the design of products and processes.

This definition is however very wide and covers about every engineering
subject ranging from controlling chemical processes to air planes. The rela-
tion with precision engineering is clear but precision is a relative concept.
Present modern industries are all working towards the maximum of their
technological capabilities. It is probably true to say that these definitions in
fact define mechatronics to be the modern version of mechanical engineering.
Mechanical engineering has in its history always integrated new emerging
technologies, like for instance thermodynamics and electrical power sources.
While the term “mechatronics” originated in the seventies of the last cen-
tury, it clearly appears to have become an inextricable part of the present
mechanical engineering discipline.

1.2.1.1 Cultural differences in mechatronics

The above definitions and the international developments in the field unveil
some interesting observations regarding the differences in approach on
1.2. Definition and international positioning 19

mechatronics. The understanding of these differences is important to be

able to effectively cooperate with people from different regions and cultures.
First of all the term Precision Engineering originally was used for precision
manufacturing and machining. Research and improvements in that field
concentrated on primary processes like milling, grinding and polishing
combined with extensive dimensional and geometric metrology to check and
optimize the machining process. When electronics and control principles
came available, this background has clearly resulted in the first mentioned
Japanese definition of mechatronics. One might say that this definition
is related to the machining orientation of mechatronics where the new
technology was applied to correct flaws in the uncontrolled machines of that
time. A clear example is found in the use of piezoelectric actuators to correct
positioning errors in machine tools, where most of the linear movement was
done by means of rotating motors and precision screw spindles. Without
changing the basic concept for the long-stroke spindle positioning system,
the piezoelectric actuator had its own feedback control system where it
had to correct errors that originated from the screw-spindle actuator. This
approach is still a practised way to achieve precision in machining tools
with some clear benefits as disturbing cutting forces are absorbed by the
inherently stiff construction of the piezoelectric actuator even before the
feedback-control needs to correct them.
Meanwhile, in the essentially electronics and photon-physics oriented in-
dustry on consumer electronics and IC lithography exposure equipment,
a different approach was followed. In the previous section it was shown
how the development of contact-less optical recording with the use of optics
had forced people in finding ways to manipulate photons in following a fast
moving track. As photons are essentially without any mass or disturbing
forces, it became preferred to avoid all contact with the “dirty” interfering
vibrating environment and only control the system in respect to the track
on the optical disk that had to be read out. This design concept resulted
in a moving-coil actuated system with zero-stiffness to the surroundings
and an electronic control system with inherent high virtual stiffness to the
track on the optical disk. The resulting precision was impressive for that
time (< 0.1 μm). It had only one drawback because the possible movement
range of a Lorentz actuator is limited or high cost of big magnets and coils
would incur. This was eventually solved by adding a non critical long-stroke
mover, the described screw spindle in the later CD players that only had to
transport the stationary part of the moving-coil actuators. The demands on
the long-stroke actuator were so limited that the ultimate cost of the total
20 Chapter 1. Mechatronics in the Dutch high-tech industry

system became much lower than would be the case with using piezoelec-
tric actuators, as would have been the common approach in the machining
industry.
The signatures of these two different applications are still very visible in the
present mechatronic playground. In the USA, where the competition in the
development and manufacturing of consumer electronic products is almost
completely vanished, the machining industry determined the mechatronic
research and development. Only in the last several years fast tool servo
systems with reduced mounting stiffness and high bandwidth for increased
precision are introduced, specifically for the manufacturing of high quality
optical parts.
In Japan a mixture of both approaches is practised, as in that country both
precision machining and consumer electronics are developed. Nevertheless
the more conservative machining approach is frequently observed which
sometimes hampers progress in their developments.
Europe is in this respect interesting in its own way. The largest part of
Europe followed along the lines of the developments in the USA and fo-
cused on precision machining and related mechatronics instead of consumer
products related technology. The Netherlands however went 100 % into
the other direction. First of all there is hardly any machining industry left,
while several world class companies in the high-tech industry are active in
that country applying precision mechatronics in professional printing, in
electron optic and x-ray imaging and in optical lithographic equipment. The
related photon-physics oriented mechatronic approach has been instrumen-
tal in the success of these industries and for that reason it is unavoidable
that this photon-physics oriented Dutch school of mechatronics resonates in
this book.
1.2. Definition and international positioning 21

Informaon Energy
Feedforward Electrical
Power Heatloss
Mechanical
Error Power
Posion + 
setpoint Controller Amplier Actuator Mechanism
-

Measured posion
Sensor
Control electronics + so$ware Hardware

Figure 1.16: A mechatronic motion system consists of several elements that are all
required for the total functionality. The left side (yellow) only deals
with information. Although fully analogue electronic realisations are
still used, presently this function is mostly implemented in software
with digital electronics. At the hardware side (blue) both information
and energy is manipulated to achieve the final goal.

1.2.1.2 Focus on precision-controlled motion

In view of the aforementioned considerations it was decided to use the Eu-

ropean definition of mechatronics and focus the contents of this book on
controlled motion systems with emphasis on the precision engineering and
positioning principles that are applied in the Dutch high-tech industry.
A mechatronic system within this focus area generally consists of the follow-
ing main components:

• A movable mechanical construction.

• An electrically controllable drive to move the mechanism in its degrees

of freedom, the actuator.

• An amplifier to convert the electrical power to the power needed by

the actuator as function of the control output.

• A measurement system to monitor the movement, the sensor.

• A control part that controls the system, the software.

Figure 1.16 shows these elements in an overview. The arrows indicate the
flow of information and energy, most of which takes place in the electrical
domain. This connecting function of electricity within mechatronics is also
one of the connecting elements in this book.
22 Chapter 1. Mechatronics in the Dutch high-tech industry

1.3 Systems engineering and design

Mechatronic design is very much related to the field of systems engineer-

ing. In this section the basic principles of systems engineering and design
are introduced, because they are widely used in the high-tech industry
and present a useful framework for the multidisciplinary design of any
mechatronic system.
Systems engineering originated around World War 2 as a discipline to cope
with the continuous increase in complexity of machines. Before that time,
machines like airplanes, cars and ships were so fundamentally simple from
a concept point of view, that almost without any exception several people
knew how the total system worked. With the introduction of electronics,
software and all kind of safety, diagnostics and other surrounding informa-
tion oriented technologies in these machines, it became a hazard or even
impossible to grab the whole entity by one person. This development has
created the need to divide the capabilities in design between generalists,
that have rather a global overview of the functionality of a system, and
specialists„who have an excellent view on all details of a certain discipline.
This is visualised in Figure 1.17.
In view of the multidisciplinary character of mechatronics, a mechatronic
designer is rather a generalist than a specialist and he or she needs to keep
a complete overview of a complicated system in order to be able to act as
the lead-designer of a development team. That need has become more and

Breadth of
knowledge

Generalist
Generalist
Specialist

Specialist

Specialist

Specialist

Specialist

Specialist

Specialist

Specialist
knowledge
Depth of

Figure 1.17: Both specialists and generalists are required to design a complex
system.
(Courtesy of G. Muller, Gaudi site)
1.3. Systems engineering and design 23

more a real issue as the capability to act as a real generalist with sufficient
knowledge to effectively communicate with specialists in a multitude of
disciplines is something one can hardly learn from books, but first and for
all by experience.
Based on this trend, a continuous effort is taking place to find methods that
help to keep the problems under control that are inherent to this increasing
complexity. In the following, some basic methods are described that are
widely accepted with clear benefits for a successful completion of a complex
system design.

1.3.0.3 Definitions and V-model

As mentioned in the previous part, a consequence of the increased complexity

of systems was the rise of specialists, people who knew a lot but only about
a part of the system. With as example a car, one specialist knew all about
the engine, one was specialised in the steering, one in the brakes, one in
the suspension, and later even one for the electronics and software. One
might think that it would be a perfect solution if these specialists just
communicated with each other and by virtue of some overlap would come
with an optimally designed car. Like with the mousetrap example this
does not work in real life, unfortunately. Some people are more dominant
than others and these often force a design in the direction of their hobby
(read: specialism), which is not necessarily the optimal direction. Of course
there are positive exceptions, but in many large companies and especially
in situations where safety is an issue, methods had to be developed that
would make the designs less depending on individuals, in order to achieve
continuity in technological quality and performance. It was first in the
professional, military and aerospace industry, that the concept of Systems
Engineering was set forth. This concept is based on a large set of clear
definitions and a visualised model, the V-model of systems engineering.
The following definitions originate with a few small adaptations from the
International Council of Systems Engineering (INCOSE1 ) and the Gaudi
website2 , that provides systems architecture information and is maintained
by Gerrit Muller, professor in Systems Engineering from the Buskerud
University college.

1 www.incose.org
2 www.gaudisite.nl
24 Chapter 1. Mechatronics in the Dutch high-tech industry

System: An interacting combination of elements to accomplish a

defined objective. These include hardware, software, firmware,
people, information, techniques, facilities, services, and
other support elements.
Systems engineering: An interdisciplinary methodology to en-
able the realisation of successful systems.
Systems engineer: An engineer trained and experienced in the
field of systems engineering.
Systems engineering processes: A logical, systematic set of
process steps selectively used to accomplish systems engi-
neering tasks.
Product Creation/Generation Process (PCP/PGP): A logical,
systematic set of process steps to create/generate a product
or system.
System architecture: The arrangement of elements and sub-
systems and the allocation of functions to them to meet
system requirements.
System design: The activity where it is determined how the
system will be realised.
System requirements: The list of functionalities that the cus-
tomer would like to have in the system.
System specification: The list of unambiguously measurable
functionalities that are planned to be realised in the system.
System performance: The list of unambiguously measurable
functionalities that are realised after completion of the sys-
tem.
System property: The list of both wanted and unwanted both
unambiguously measurable and not well measurable func-
tionalities of a system.

It is important to note that in these definitions systems engineering is de-

scribed as a methodology and approach to organise the design of a complex
system. This means that the system engineer role in organising the process
is not the same as the role of a system designer, who concentrates more on
the content. In principle and even preferably these roles are united in one
person.
1.3. Systems engineering and design 25

Funconal requirements Realised funcons

Feedback loops

System design

Subsystem design

Detailed design Timeline

Figure 1.18: The V-model of systems engineering divides the design process of a
complex system both in time and in levels of complexity, in order to
organise the effort over specialists and generalists.

The V-model of systems engineering as shown in Figure 1.18 was introduced

in the design process to organise the way of working. The value of the model
is found in the methodology of dividing responsibilities and tasks of both
specialists and generalists over the complexity of the system. By working
with layers with increasing detail per layer, the design of a complex system
becomes manageable. Going down on the left leg of the V the following order
of events is followed:

• In the “System design” top-layer, the functional requirements of the

total product are determined and translated into system specifications
and subsystem requirements.

• In the “Subsystem design” layer, these subsystem requirements are

translated into subsystem specifications and requirements for the
detailed elements.

• In the “Detailed design” layer, the requirements for the detailed ele-
ments are translated into drawings, software code and realised parts
with a verified performance that can be assembled into functional
subsystems.

Then the right leg is followed upward again.

26 Chapter 1. Mechatronics in the Dutch high-tech industry

• In the “Subsystem design” layer, the engineers integrate the different

parts into a working subsystem that is tested in order to verify that
the performance of the subsystems meets their specifications.

• In the “System design”, layer the different subsystems are integrated

into a whole system and the performance of the total system is tested
and verified.

In very complex systems, the subsystem layer itself often consists of dif-
ferent layers. Each has its own requirements, specifications and realised
functionality.
The work in each layer has a different character. The deeper one goes to-
wards the bottom of the V-model the more detailed and concrete the work
will be. But also within one layer the work is different between the left and
the right leg, due to the fact that definition and realisation require different
skills. For this reason, often the people in the team are changed during
the project. This is not preferable from a learning perspective, because
experience with results of the realisation phase will improve the decisions
taken in the definition phase of a next project. Nevertheless, the work in
the left leg is in practice done primarily by innovator and creator types of
designers, while the right leg is realised by more completer and finisher
types of people. In any case, frequent feedback is necessary within a project
to overcome non compliances with the specifications that show up in the
verification stage. Communication is also necessary between projects where
experience from the past is used for continuous improvement over time.
These feedback loops over different people need continuous attention as they
do not happen automatically.
The subsystem layer is the place where many mechatronic systems are
designed, like motion stages and actively controlled optical imaging systems.
In the following section the work in these layers is described in the context
of a regular development according to the product creation process.

1.3.0.4 The product creation process

Figure 1.19 shows the time division of the product creation process that is
commonly used in industry. The phases 0, 1 and 2 correspond with the left
leg of the V-model while the phases 3, 4 and 5 correspond with the right leg.
From practical experience it is known that many mistakes are made during
the feasibility and definition phase, when not sufficient time and effort is
given to determine all factors that are at stake. In this early phase the
1.3. Systems engineering and design 27

0. 1. 2. 3. 4. 5.
Feasibility Denion System Engineering Integraon Field
design & test monitoring

Requirements

Specicaon

Design

Vericaon

Engineering

Most informaon
Core informaon
Legend: in dra$
50% available in heavier change control
concept
full under development preparing or updang work
Timeline

Yearly Cost of project

€ Financial Impact per decision

Figure 1.19: This standard time line of a product creation process is closely related
to the V-model and is more directly suitable for planning purposes.
The lower graph shows the yearly cost related to the effect of decisions
on the financial results of the project and emphasises the need to
especially control the first phases.
(Courtesy of G. Muller, Gaudi site)

decisions with often the largest impact on the success of a project are made
at the lowest relative cost. This relation is schematically shown in the lower
graph of the figure. Especially in this phase the systems engineer needs to
have a good overview on market demands and technological possibilities.
In most cases he closely cooperates with a product manager who takes
care of marketing issues and represents the end user of the product. This
combination of often only two people is to a large extent responsible for the
success or failure of a product. One might say that a badly started product
creation process can hardly be repaired while of course a good start can still
end up in a mess due to a bad execution. This duo of systems engineer and
product manager has to deal with the tension from the external market
in timing, product margins and functional requirements. They will define
the final set of requirements in a process as shown in Figure 1.20 that are
used as input for the development activities. They also have to keep these
requirements closely under change control. After the integration and test of
28 Chapter 1. Mechatronics in the Dutch high-tech industry

top-down
Key drivers
(customer, business)
Operaonal drivers
(logiscs, producon, et cetera)
Roadmap
(posioning and trends in me)
Compeon
(posioning in the market)
Connued
Regulaons Requirements Product Creaon
Process
Ideal reference design
Prototyping, simulaon
(learning vehicle) Feedback
Boom-up
(technological opportunies)
Exisng systems
boom-up

Figure 1.20: Requirements originate from a large amount of input information.

Well defined requirements are important for a successful end result.
(Courtesy of G. Muller, Gaudi site)

the product, this team is often also responsible for dealing with the market
response on the product.
At all levels in the V-model, a trade-off has to be made between specifications
of subsystems that have to work together. In the next section the related
process of requirement budgeting will be explained a bit further.

1.3.0.5 Requirement budgeting

In Figure 1.21 an accuracy budget of a positioning system is shown as an

example, indicating how the requirements and specifications are divided
over the different subsystems with their corresponding sources of errors. It
is clear that a systems engineer who has to define these requirements needs
to have a thorough insight in all subsystems that contribute to the overall
end result in order to give the largest error budget to the most difficult
subsystem. The shown calculation is only valid if the errors are random
stochastic and independent in which case one can take the square root from
the sum of squares of all contributing errors as will be further explained in
Chapter 8 on measurement.
The system engineer also has to decide which part of the budget is allowed
to the smaller frequently occurring random error sources. This is noted in
the item “Other” in the figure. This process of budgeting of requirements
is only one half of the total activity as the top-down process to split-up the
requirements is followed by a bottom-up process to determine the corre-
1.3. Systems engineering and design 29

Balls
15 μm

Bearings
18 μm

Sensors Drive Grooves

10 μm
40 μm 20 μm

Tolerances
Overall Actuaon Guiding 20 μm
60 μm 40 μm 30 μm

Stability
10 μm
Other Other
20 μm 20 μm

Other
10 μm

Figure 1.21: Even a simplified accuracy budget of a positioning system with many
independent error sources already shows the choices that have to be
made in close cooperation with the design teams that have to realise
the targeted budget.

sponding specifications. The designers that are involved in this process

will often face contradictory or conflicting requirements and in negotiation
with the system engineer the specifications are determined with as goal a
minimal effect on the original targeted full-system specifications. In this
way the expertise of the responsible subsystem designers is used to create a
balanced specification budget.

1.3.0.6 Roadmapping

Figure 1.22 shows in a graphical way, that the process of developing prod-
ucts takes place in a continuous flow of changing external influences, the
increasing demands of the market together with the also increasing techno-
logical developments. In most cases these external influences can hardly
be influenced by the design team. A good example is the development of
the Personal Computer. Initially the combination of IBM, Intel and Mi-
crosoft could determine their market approach without the need to take the
competition into account. Inevitably however, at some moment in time, the
competition like Apple and the Unix based Linux community had arisen forc-
ing the seemingly monopolistic PC league to listen to their customers. This
observation has as consequence that in general a lot of marketing research
30 Chapter 1. Mechatronics in the Dutch high-tech industry

Market demands

Products

Technological developments

Figure 1.22: The natural flow of product development.

is needed to get and keep a good feeling of the market dynamics and to make
reliable estimates of future needs and possibilities. These are key tasks of
the marketing people in close cooperation with their technological counter-
parts. For several practical reasons, developing products can never be as
continuous as the market would like. To keep the cost down and the quality
high, the diversity in parts and products has to be limited. For this reason
products are generally developed in product families, also called platforms,
that will stay in production as long as possible. As a consequence, when
the time needed for the development itself is added, the initial decisions
on requirements and specifications need to remain valid for a long time.
One of the tools to systematically support these long term decisions is the
process of road mapping. This is a strategic process that should be done at
regular intervals to act as a basis for investments in new technologies and
developments. Figure 1.23 shows a typical example of such a roadmap just
for illustration of the principle. It is a highly simplified roadmap dealing
with the engine drive-train of a car.
Of course there is a large freedom in methods of road mapping and the shown
example is just one of the many possibilities to draw a suitable roadmap
for the planning of future development activities. The roadmap consists
of several layers where, like in the V-model, the upper layer deals with
the total overview of all product introductions and the lower layer with the
actual detailed plans on subsystem level. The definition of these roadmaps is
real teamwork, where people from marketing, development, manufacturing
and other sectors all need to have their input. An additional benefit of the
process is the resulting commitment of all involved people to realise what
has been agreed upon. It is in this way also clearly communicated and
understood that some developments need more time than others, like the
electrical drive in the example that is related to the applied batteries.
1.3. Systems engineering and design 31

Development Roadmap " 

'*
6"
2007 2008 2009 2010 2011
Product ranges A B H E

Innovaon Drivers
Maximum speed (km/hr) 150 160 180 150 130

Fuel consumpon (km/l) 10 10 9 13 n.a

Acceleraon (0-100 km/hr in sec.) 10 9 8 10 10

Technologies
Diesel Improved turbo (A)
Diesel Improved turbo (B)
Hybrid Hybrid technology
Electrical All electrical

Planned acvies
Enhanced Turbo development
Development hybrid technology
Development electrical drive in H and E

= Required start date

= First shipment on a car
Improved inlet pressure turbo type A and B
Hybrid
All electrical

Figure 1.23: Roadmap that was drawn in 2007 as university lecture example, indi-
cating the different engine-drive principles that were expected to be
developed by the car industry in the following years.

A full roadmap of a product line can consist of many different roadmaps

that are all interrelated. A suitable method is again based on the V-model
with an overall roadmap on system level and more detailed roadmaps on
subsystem level.

1.3.1 Design methodology

As noted in the previous part a (system) designer has to deal with several
contradictory demands:

• Shorten the development time in order to get the product earlier in

the market (more profit!).

• Stimulate the market with a continuous flow of new products.

• Enhance product performance by adding functionality (more cost!).

• Reduce product cost to increase margins (more profit!).

32 Chapter 1. Mechatronics in the Dutch high-tech industry

Those items need to be balanced as they can not be achieved simultaneously

in a simple way.
A good example of this balancing is the observed phenomenon that, without
additional measures, a reduction of the development time will almost by
definition increase the total cost of the development process in an expo-
nential way, while the gain in profit for an earlier introduction not always
justifies such increased cost. Especially in the high-tech electronics industry,
however, an earlier introduction can be very valuable. A good example of
the first three bullets is the dramatic diversification in mobile phones and
PDAs. Being often an impulse buy where the choice is highly determined
by emotion. As a consequence of this emotive value, an old design is often
obsolete within a few months. This problem is far less the case for large
domestic appliances, like washing machines, that mainly have to be reliable
with a very long lifetime without any surprises in their user interface.
Because of these contradictory demands several methods are used of which
the following two will be presented in the last part of this chapter:

• Concurrent Engineering.

• Modular design with platforms.

1.3.1.1 Concurrent engineering

The drive for total quality in the second half of the last century has forced
development activities to become a more and more rigidly planned process.
Total quality means that not only the products are expected to function
exactly like they are promised to do, but also the process to create and
produce them needs to be fully determined. One item that was high on the
priority list at that time was the reliability of the development timing and
the reduction of scrap due to unplanned repairs. This process resulted in a
strict planning with milestones and go/no-go decisions in order not to start
with a next step until all questions of the previous step were answered. This
approach had several advantages:

• The planning became continuously more solid towards the end of the
process.

• High investments could be postponed until really needed.

• Surprises were avoided by testing prototypes in the period that long-

lead items were ordered.
1.3. Systems engineering and design 33

Non Concurrent (serial) Engineering

2007 2008 2009 2010 2011
Planning Start

Product denion
Concept design
Detailed design
Long-lead item order and delivery
Integraon and test
Producon

Concurrent (parallel) Engineering

2007 2008 2009 2010 2011
Planning Start

Product denion
Concept design
Detailed design
Long-lead item order and delivery
Integraon and test
Producon

Figure 1.24: Concurrent engineering results in shorter lead times by working in

parallel on different phases of the project. It requires a high discipline
in communication and change control.

• Manufacturing could better prepare itself.

• Development personnel could make the product and service documen-

tation during the order and delivery phase of the parts.

The process to allocate all activities essentially in series over time often
resulted in practice in a development lead time of more than three years
for a general consumer product. To partly solve this problem, the process
of concurrent engineering was introduced as shown in Figure 1.24. Under
pressure of market needs, people started to make use of the fact that in
each phase of the development process some elements will take more time
than others to be finished. It might be useful to treat these long-lead items
as off-the-shelf parts that have to be used just as they are. This approach
means that the long-lead items should be the first to design and the other
parts have to adapt to the interfaces chosen for these long-lead items. In
this way a hierarchy of parts is created that enables the designers to start
earlier with the next step in the process.
An important condition to make this method work is the use of agreed
interfaces. From the computer industry several well established interface
standards are known like PCI, USB and (serial-) ATA to name a few and
these have helped this industry to become as successful as it is. For the
34 Chapter 1. Mechatronics in the Dutch high-tech industry

design of mechatronic systems the situation is less easy due to the low
flexibility of the solid hardware but even in that field standard interfaces
are quite common with as best example the standardisation in connective
parts like bolts and nuts
In spite of the reduction of the development time, this more parallel concur-
rent engineering approach also has some drawbacks that need attention:

• Very costly repairs in case of forgotten or overlooked interface needs.

• Less time available for documenting.

• Less gradual planning of development resources.

In the industries where concurrent engineering became the “de facto” stan-
dard of operation like in semiconductor manufacturing equipment, several
methods are used to avoid these mentioned problems. First of all a formal
and intensive communication network is created, focused on exact interface
definitions. Specifications are only adjusted in a disciplinary process accord-
ing to the rules of systems engineering. Secondly it appeared to be of crucial
importance to work with a modular design that is based on a technology
platform with a lifetime of more than one product introduction.

1.3.1.2 Modular design and platforms

The first example of a modular design around platforms is seen in the

previously mentioned computer industry. A basic desktop PC consisted
around the time of writing of this book of several clearly distinguishable
modules, like the power supply, motherboard, memory, processor, hard disk,
monitor, keyboard and several other smaller items. All of these are modules
in principle interchangeable for other versions within a platform, where
a platform is in this case a certain combination of processor and related
hardware. The designers of the platform determine the interfaces that
communicate with the peripheral electronic modules. This way of working
has enabled many electronic companies to deliver parts with different levels
of functionality. Together with the use of standardised hardware-connectors
and dedicated driver-software, they were often even capable to provide
solutions for different platforms.
Another example of a modular design is found in the mobile phone industry.
By exchanging panels and windows on a standard electronic circuit, a large
variety of products can be derived without changes on the basic functionality
1.3. Systems engineering and design 35

Recle stage

Illuminator

Recle handler Projecon opcs

Wafer handler Waferstage

Figure 1.25: The main modules in a wafer scanner that is based on the “Twinscan”
platform of ASML. All modules are individually upgradable giving
the platform a virtually infinite lifetime in a market that demands a
continuous increase in performance.
(Courtesy of ASML)

and reliability in the new design. While the power supply and display are
controlled with embedded software (firmware), much flexibility is available
in showing or hiding functions depending on the market position of the
specific product. Most functions, sometimes including the camera, are
integrated and depending on the execution they are switched on or off in the
firmware or just covered by the shell. The total cost of such a design is often
even lower when some hardware in the product is not used, than would be
the case if a special hardware version without that specific function has to
be developed. This is mainly due to the lower logistic cost related to the
reduced diversity.
The last example of a modular design is the ASML wafer scanner as shown
in Figure 1.25. The “Twinscan” platform with a dual wafer stage became
the leading technology in this industry. In retrospect, many reasons for this
success can be mentioned like the team spirit, commitment and drive of
the people, the partnership with Zeiss and the strong mechatronic roots in
36 Chapter 1. Mechatronics in the Dutch high-tech industry

Philips. Nevertheless the modular design approach proved to be one of the

key elements of success for the following reasons:

• All modules can be developed on different locations.

• All modules can individually be tested to their full specifications on

test rigs without the presence of the other modules.

• All modules can be mounted and dismounted from the machine without
affecting the other modules.

• Duration of the installation times at the customer site is significantly

reduced.

• Upgrading is possible even at the customer site, which makes most

machines last almost forever.

• Testing and servicing is relatively easy because all important modules

can be dismounted.
Chapter 2

Electricity and frequency

The performance of a mechatronic system is related to the ability to gener-

ate and control a certain movement of a body. In general this means that
forces need to be applied to keep the body on track while compensating other,
external forces. In case of constant external forces, there is no real difficulty
to control these but in reality external forces are never really constant nor
are the movements that need to be controlled. These forces and movements
change in different ways both periodic and stochastic and this means that
the performance must be analysed by looking at the response of the system
to periodic signals of different frequencies. These frequencies are often
observed in both the mechanical and electric domain.
This chapter begins with the principles of electricity followed by a presenta-
tion on the concept of frequency with periodic signals and wave propagation.
The frequency and time domain related Fourier and Laplace transform
methods are presented as necessary mathematical background that is used
in the dynamic analysis of mechatronic systems. The chapter finishes with
some frequently applied graphical representations of frequency and time
responses of a dynamic system.

37
38 Chapter 2. Electricity and frequency

2.1 Electricity and signals

The phenomenon of electricity is fundamentally based on the electric charge

composition of atoms. Atoms have a core, consisting of neutrons and protons,
with a positive charge determined by the amount of protons, while neutrons
are not charged. Around the core a “cloud” of electrons with a negative
charge neutralises the charge of the core in a stable situation where the
amount of electrons equals the amount of protons.
The unit of the amount of charge is [C] (Coulomb), named after the French
physicist Charles-Augustin de Coulomb (1756 – 1806), because of his work
on electricity and magnetism. As is true for many definitions in electricity,
the value of one Coulomb has been determined much later than when the
initial definitions were made. Electrons and protons were not known when
the first electric experiments were carried out. It appeared that 1 C equals
6.25 × 1018 times the charge of one electron.
Electrons are not uniquely linked to their atoms but can more or less freely
move. The freedom to move is inversely proportional to the electric Resis-
tivity (ρ r ) [Ωm] of the material. A material can be an insulator with a very
high resistivity value like glass with ρ r ≈ 1010 Ωm and even higher with
plastics that might have a resistivity of ρ r ≈ 1020 Ωm. If the resistivity is very
low, the material becomes a conductor like Copper with ρ r = 1.68 · 10−8 Ωm.
Semiconductor materials are a special category having an electric resistiv-
ity in between a conductor and an insulator. Unique for these materials
is the possibility to change their resistivity over a large range from fully
isolating until highly conducting, either permanently by physical/chemical
modification or dynamically in an electronic circuit.
The resistance (R ) in Ohm [Ω] of an object with a cross section ( A ) and a
length (l ) is directly derived from the resistivity of the applied material and
is equal to:
l
R = ρr [Ω ] (2.1)
A
It is clear that the resistance increases proportional with length and resis-
tivity and decreases with a increase of the cross section.

2.1.1 Electric field

Charged particles like electrons and protons have as property that they
either attract each other, when their charge has an opposite sign, or repel
2.1. Electricity and signals 39

+ _

+ _

Figure 2.1: Charges have an electric field that is directed away from the positive
charge and towards a negative charge. The electric field is graphically
represented by field lines. The density of the field lines determines the
force acting on other particles in the vicinity and the arrows of the field
lines give the direction of the force in case of a positive inserted charge.
The combined field of two opposite charges results in curved lines that
start at the positive charge and end at the negative charge.

each other, when their charge has the same sign. For example an atom with
a lower number of electrons than its proton count will attract electrons until
its charge is neutralised again. The related electrostatic force is the driver
behind all important electric phenomena. The magnitude and direction of
this force is represented by the electric field E. Mathematically the electric
field at a distance r in the direction of unit vector r̂, with its origin in the
centre of a single charged particle with charge qx , is represented by the
following equation:
1 qx  
E(r) = r̂ V/m (2.2)
4π r 2 0

The term 0 is the electric permittivity in vacuum and is equal to a value

 
of 0 ≈ 9 · 10−12 As/Vm . From the 4π r 2 term in the denominator it can be
concluded that the magnitude of the electric field is related to the surface
of a surrounding sphere at distance r . This can be visualised by means
of field lines as shown in Figure 2.1. The field lines represent the direc-
40 Chapter 2. Electricity and frequency

tion of the electric field as arrows and their density is proportional to the
magnitude of the field. This also means that the field is constant orthog-
onal to the field lines. By the definition of unit vector r̂, the direction is
outward for a positive charge and inward for a negative charge. Though
the drawing is two-dimensional, in reality the space around the charge is
three-dimensional, which corresponds with the decrease in the electric field
and the corresponding density of the field lines to the distance squared. In
the lower drawing of the figure the effect on the course of the field lines of
two opposite charges is shown. The connecting curved field lines are the
result of the superposition principle of electrostatic fields, which means that
fields can be simply added together. As a consequence, the resulting field
at the equidistant plane from both charges is zero and the field lines will
cross this plane perpendicular, leading to the shown curved lines from one
charge to the other.
The magnitude and direction of the force that is acting on a charged particle
with charge qx , inserted in a field E(r) that originates from several other
charged particles, can be calculated quite straightforward as follows:

F(r) = q x E(r) [N] (2.3)

This force F will be directed in the direction of the field in case of a pos-
itive inserted charge and in the opposite direction in case of a negative
inserted charge. This direction corresponds with the known phenomenon
that negative charges are attracted in the direction of a positive charge.

2.1.1.1 Potential difference

The vectorial nature of the electric field and the related force is very useful
when determining physical effects in for instance electron microscopes,
where charged particles have to be controlled in velocity and direction. In
electric systems, where electrons are guided in conductive material like
resistors, wires and other elements, it is more easy to work with the scalar
magnitude of the related potential energy of a charged particle that is
inserted in an electric field. What happens can best be explained when
looking at two points in the electric field at a different distance from the
first charge. The potential energy is determined in the same way as with
gravitation. When a particle with charge qx moves from location r1 to r2 , the
work exerted by the electrostatic force will change its potential energy. The
energy (E) is given in the unit Joule [J] named after the English physicist
James Prescott Joule (1818 – 1889) for his important exploratory work on
2.1. Electricity and signals 41

energy. The change in potential energy of a charged particle has a negative

sign, when the movement is in the direction of the force and is given by:


r2

ΔE p = − F · dl
r1

r2

= − qx E · dl [J] (2.4)
r1

Due to the sign convention this means that a negative charge will have a
positive potential energy in the vicinity of a positive charge which increases
with the distance. This reasoning corresponds with the potential energy of
a mass under the gravity force on earth. From this difference in potential
energy between location r1 and r2 , the potential difference can be defined
in order to get a number that is independent of the charge of the inserted
particle, by dividing the previous equation by qx .

ΔE p
ΔV =
qx
r2

= − E · dl [V] (2.5)
r1

This potential difference ΔV , or just shortly V , equals the electric potential

between r1 and r2 and gives the electrostatic potential energy per unit of
electric charge between these points.
The potential V reduces the vector field E to a scalar field without losing
information. The unit Volt [V] for the potential difference is named after
the Italian physicist Count Alessandro Guiseppe Antonio Anastasia Volta
(1745 – 1827), because of his work on the first electric batteries.
As with all forms of potential energy, this potential V is a relative quantity
as it only gives a potential difference. An absolute value can be obtained by
defining the potential at a certain point to be zero. In electronics this point
is the common ground. In electrostatics often a point at infinite distance is
used. This will give the following “absolute” potential and “absolute” energy
level:
r r
V (r) = − E · dl [V] =⇒ E p (r) = − q x E · dl [J] (2.6)
∞ ∞

These are positive numbers, when the E field is created by a positive charge,
as this field is directed outwards. These expressions make clear that the
42 Chapter 2. Electricity and frequency

V1 Potenal di<erence
r1 (V1 - V2)
I
_

r2 V2
+

Figure 2.2: Two points at a different distance from a charge have a difference in
potential. When the charge is positive, the potential V1 is higher than
the potential V2 , which means V1 − V2 is positive. A negative charge,
trapped in a conducting wire between r1 and r2 , will be driven by
the electrostatic force in the direction of r1 , which corresponds with a
positive electric current flowing in the other direction.

potential difference is equal to the derivative of the potential energy over

the charge. This is important in the next section when changes in charge
represent a current.

2.1.1.2 Electric field in an electric element

A non-zero electric field determines a potential difference in space and vice

versa. This means, that in situations where a potential difference is present,
for instance over a resistor, an electric field is present inside the element.
This field exerts forces on the charges in the resistor, causing the free charges
to move. To visualise this, a thought experiment might help by assuming in
Figure 2.2, that an electron can be “trapped” at position r2 with potential
V2 in a tunnel, that only leads to r1 with potential V1 . In that case the
electron will move to r1 , because of its higher positive potential, when using
Equation (2.6) referenced to infinity. If the potential had been calculated
relative to the centre of the positive charge instead of infinity, so integrated
from r to zero, a negative potential would have resulted, increasing as
function of the distance |r|. In this thought experiment it is observed, that
the electron seeks the position with the lowest negative potential energy,
referenced to the positive charge. Indeed when |r| = 0, the energy is lost
as both opposite charges will cancel each other out. On the other hand, if
this process would be reversed by taking an electron away from a location,
this would result in a positive charge at that location and an increasing
potential difference at increasing distance. It is this principle that creates
the possibility to realise an electric circuit.
2.1. Electricity and signals 43

2.1.2 Electric current and voltage

Moving charges are known as electric current, which is given the variable-
symbol I with unit Ampère [A] named after the French physicist and math-
ematician André-Marie Ampère, because of the high value of his research
on electromagnetism. The current is equal to the flow of charge C per unit
 
of time through a surface C/s .

2.1.2.1 Voltage source

By actively separating charges, a source of potential difference is created, a

voltage source. The name voltage for potential difference is more common in
electronics, even though the name and the unit are mixed and the relative
meaning is lost. A voltage source has two interface points with the outside
world, called electrodes or terminals, of which the positive electrode has a
more positive potential than the negative electrode. The voltage source can
supply a continuous flow of electrons, running from the negative electrode
towards the positive electrode through an external conductive load. This
electron flow is represented in electric circuits by a positive current, because
electrons are negatively charged, flowing in the other direction from the
positive electrode to the negative electrode.
To achieve the separating action between the negatively charged electrons
and the stationary positively charged protons, a voltage source can be imag-
ined to possess an internal electric field, that is directed from the negative
to the positive electrode. Like explained in the previous section, this field
drives the electrons in the opposite direction, away from the protons towards
the negative electrode. The resulting surplus of electrons at the negative
electrode with the corresponding lack of electrons at the positive electrode
results in a charge and potential difference between the electrodes, with a
corresponding electric field outside the source. This external electric field is
equal to the internal “driving” field, however it is pointing in the opposite
direction.
Because of these opposite directed fields, the previous reasoning on the
potential difference is confusing as it was defined as the negative integral
of an electric field over a certain distance. This is still valid for the external
field, with a positive potential difference between the positive electrode and
the negative electrode. For the internal field this same potential difference
should be calculated by taking the positive integral of this field. To avoid
this confusion, the positive integral has been given a different name, the
44 Chapter 2. Electricity and frequency

Electromotive Force Fe .
When Ei is the internal electric field, e 1 and e 2 are the electrode locations
and dl is the inner path between the electrodes, the electromotive force
becomes:

e2

Fe = Ei · d l [V] (2.7)
e1

In Chapter 5 an example of a voltage source will be presented based on

induction of an electric field by a changing magnetic field. This is the most
common method of generating electricity. This principle will show to be
fully compatible with the mentioned integral expression. With the other
well-known method to generate electricity, based on chemical processes in
batteries, this is less straightforward to reason as the potential difference
is the result of the difference in chemical potential of the applied elements.
Especially with a battery it is more easy to just use the electromotive force
as a given.
The value of this electromotive force is equal to the potential difference of the
external field when no load is applied. In that case, due to the equilibrium
between the electromotive force and the external potential difference, no
further separation of charge can take place. This means that no current will
flow inside the voltage source, when it is not connected to a load. As soon
as an electrically conducting load is connected to the voltage source, the
potential difference will drive electrons from the negative electrode through
the load to the positive electrode. Due to the required balance between the
electromotive force and the potential difference, the amount of electrons
flowing through the load will be replenished by a flow of “new” electrons,
driven by the electromotive force. This process of continuous electron flow
is equivalent to a continuous flow of positive current, inside the source from
the negative electrode to the positive electrode and through the load from
the positive to the negative electrode.
This reasoning is valid for an “ideal” voltage source. In reality a voltage
source will also show some internal imperfections that cause the voltage at
the electrodes to be different from the electromotive force, depending on the
current level. This phenomenon will be presented further in Chapter 6 on
electronics. In that chapter also the current source will be introduced, that
drives an electric current through a load with a magnitude, that is inde-
pendent of the load. It has an infinite capability to adapt the electromotive
force to any load, such that the current remains constant.
2.1. Electricity and signals 45

The following summarises the most important relations:

• An electric current is a physical representation of a positive entity. The

real current in most electric circuits1 is the movement of the electrons
in the opposite direction of the current due to their negative charge.

• The electric field inside a source or a load is pointed in the direction of

the positive current.

• A voltage source is based on a sustained electric field, represented

by its electromotive force, that drives the electrons to the negative
electrode.

• The voltage source creates an external electric field, represented by its

potential difference, that is directed from the positive to the negative
electrode, so opposite to the internal electric field in the source. the
external potential difference drives a current through an electric load
from the positive to the negative electrode.

• The electromotive force from the source is equal to the potential differ-
ence of the external electric field and is mostly just called “voltage”.

2.1.2.2 Electric power

When a charge moves through an electric field, its potential energy E p and
its electric potential V changes. The power (P ) of a system is defined as
the time derivative of energy with unit Watt [W], named after the Scottish
mechanical engineer James Watt, because of his famous work on steam
engines. With this definition the power can be expressed as follows:
dE p
P= [W] (2.8)
dt
The electric power can be expressed in the electric potential of the charges
as given in Equation (2.5), and in the current: I = d q/d t [A]. This gives the
familiar expression for electric power:

dE p d q
P = ·
dq dt
= V · I [W] (2.9)
1 In solid material positive charge can not move. The only examples of a material flow

corresponding to a positive current is in ionised gases and the movement of positrons but the
latter do not play a role in the real engineering world.
46 Chapter 2. Electricity and frequency

Current I

+
Voltage
+
source

V Resistance R

Figure 2.3: Basic electric circuit consisting of a voltage source and a resistive load.
The current is proportional to the voltage and inversely proportional
to the resistance, following Ohm’s law. The notation of the potential
difference is given with a double tip arrow where the signs are noted to
define the direction from + to −.

The signs of the previous equations are very important. It was shown,
that an electron as being a negative charge moves towards a decreasing
negative potential while a positive charge on its turn would experience a
force in the direction of a decreasing positive potential. This means that the
electric potential energy of a charge decreases due to this movement and
the corresponding power becomes available in another form, for instance
the dissipated heat in a resistor or in mechanical work in an electro motor.
In this book the positive direction of a current is defined by the pointing
direction of a single arrow while the positive direction of the potential
difference is given by a + and − sign adjacent to a double arrow to indicate
the relative potential difference.
It should also be noted that the units after the equations will most often be
omitted when it is obvious what is meant.

2.1.2.3 Ohm’s law

Figure 2.3 shows the most basic electric circuit possible. It consists of a
voltage source and a load, connected between the electrodes, that pulls
current from the source. In this basic form of an electric circuit the load is a
resistor R with unit Ohm [Ω]. In the next section the load will be extended
into a complex load, but originally the German physicist Georg Simon Ohm
(1789 – 1854) based his well-known law on the research he did using the
electric batteries of Volta. With the constant, not time dependant, voltage
of batteries the current in the circuit is only dependent on the resistivity in
the electric circuit. For this reason Ohm’s law gives the relation between
2.1. Electricity and signals 47

the current I , voltage V and resistance R of this electric system:

V
V = IR ⇒ I= (2.10)
R
With this clear relation the electric power delivered by the source, and
dissipated as heat in the resistance, is as follows:

V2
P = IV = I 2 R = (2.11)
R
This squared, non-linear relation is an important phenomenon in mecha-
tronic systems. As an example of this phenomenon, an increase of the force
of an actuator requires an increase of the current. This increase results in
a squared increase of the power dissipated in the series resistance of the
actuator. A configuration that was just safe from a thermal point of view,
can become suddenly overheated, when only a seemingly limited increase of
the current is applied.

2.1.2.4 Practical values and summary

Practical values of voltage show a very wide range:

• MV : The level of a lightning stroke.

• KV : The level used in power distribution.

• V : Mains supply, power supply of electronics, digital electronics.

• mV : Small signals in measurement sensors.

• μV : Extremely small signals, “buried” in noise and interference.

• nV : Unmeasurable.

In practice the term nV is only found in expressions like nV/ Hz, that relates
to the density of noise in a certain frequency band.
Practical values of current show an even wider range:

• kA : Power distribution.

• A : Mains, household equipment, actuator input current.

• mA : Small signals, power supply of low power electronic circuits.

• μA : Small signals, sensors.

48 Chapter 2. Electricity and frequency

• nA : Extremely small signals, “buried” in noise and interference.

• pA : Hardly measurable, Scanning Tunnelling Microscope (STM) probe.

• fA : Unmeasurable.

In table 2.1 the most relevant definitions are summarised.

Table 2.1: SI units for electricity.

Physical quantity SI unit Variable Relation

Charge Coulomb [C] C

Current Ampère [A] I A = C/s

Potential difference Volt [V] V V = J/C = Nm/As

Resistance Ohm [Ω] R Ω = V/A

Power Watt [W] P W = VA

2.1.3 Variability of electric signals

Electric voltages and currents are generally not constant over time. The
variability of these voltages or currents contain information about their
origin, for which reason they are called electric signals.
In physics a signal represents any value with a variability over time and/or
space, the temporal or spatial variability. An example of temporal variabil-
ity is sound that changes over time in amplitude and tonal character. In
mechanical engineering a position dependent force is an example of spatial
variability. In optics, spatial variability represents the difference in inten-
sity of light as function of the location on a surface. In a mathematical sense
both spatial ( f ( x)) and temporal ( f ( t)) variability can be treated the same
way, when x is exchanged with t or vice versa in the relevant formulas.
Electric signals in mechatronic systems are always temporal even though
the signal can be derived both from temporal and spatial physics phenomena.
The latter are translated into the time domain by for instance a scanning
movement with a constant velocity and measuring with a clear time refer-
ence. For this reason, the following part of this chapter will mainly deal with
temporal varying signals while in Chapter 7 on optics the spatial variability
will mainly be used.
2.1. Electricity and signals 49

2.1.3.1 The concept of frequency

The variation of an electric signal over time can be seen as a combination of

three types of behaviour:

• Constant, unidirectional, called DC from Direct Current.

• Periodic alternating, bidirectional, called AC from Alternating Cur-

rent.

• Random, stochastic, called noise.

The DC value of an electric signal equals the average value of the current over
time. For an alternating signal the term temporal frequency is defined as
the number of occurrences of a certain event per unit of time. Consequently
one can define the time period related to this frequency as the time taken for
one event to happen. In SI units the frequency with variable-symbol ( f ) is
expressed in the unit Hertz [Hz] named after the German physicist Heinrich
Hertz (1857 – 1894). The time period of one cycle, with variable-symbol (T )
in seconds [s] is inversely proportional to f .
1
T= [s] (2.12)
f
In physics and engineering the term angular frequency with variable-symbol
(ω) is often used. This frequency is directly related to the mathematical
description of harmonic oscillations, defined by the sine or cosine of an angle
changing with a constant angular velocity ω and an amplitude A , as shown
in Figure 2.4

x( t) = A cos(ω t) y( t) = A sin(ω t) (2.13)

When the sine and cosine function are presented in a graph as function
of the angle, this results in a graphic wave shape as shown in Figure 2.5.
Though this is not a real physical wave, the term “waveform” is often also
used for this graphic representation of a frequency because of its shape.
When ω is constant, the horizontal axis is also a time line where 2π equals
the period T . This means that the angular frequency is directly related to
the temporal frequency, because one period T equals a full circle being an
angle (φ) of 2π radians:

ω = 2π f [rad/s] (2.14)

It is very important to be aware of these different units as this is one of the

traps even experienced designers encounter. A factor 6 is almost an order
50 Chapter 2. Electricity and frequency

A sin(ωt) A

ωt t=0
A cos(ωt) x

Figure 2.4: The sine and cosine of angle φ = ω t represented in a plane.

of magnitude and can easily result in wrong conclusions. In this book both
units will be used. The [Hz] units are applied with the practical examples
and graphical representation, where the relation with periodicity and mul-
tiplicity of events is more clear. It is often more related to the real world
 
and human understanding. The rad/s unit is mostly used for mathematical
analysis, equations of motion and the relation with physical effects.
When two signals with the same frequency like the sine and cosine of Fig-
ure 2.5 are compared, than the sine is shifted with an angle φ = π/2 in respect
to the cosine. This time relationship is called phase and is one of the most
important parameters in a controlled motion system because a negative
phase shift represents time delay, that can cause instability when applying
negative feedback. For practical reasons this phase shift is expressed in
degrees (◦ ), because of the small numbers involved. A 30◦ phase shift can
have a large impact while it is less than π/4. Though the degree is not an
official SI unit, it is widely accepted to be used in this context.
A simple thought experiment can be done on position, velocity and acceler-
ation with a body that moves with a sinusoidal movement x( t) = x̂ sin(ω t) =

1 Sine
0.8
0.6 Cosine
0.4
0.2
0
> 0.2
> 0.4
> 0.6
> 0.8
>1
0 π π 3π 2π 5π 3π 7π 4π
2 2 2 2
φ (ωt)

Figure 2.5: Sine and cosine function of angle φ = ω t.

2.1. Electricity and signals 51

1 ?

0.8
0.6 Velocity
0.4 @
0.2
0
> 0.2
> 0.4
> 0.6
> 0.8
>1
0   3 2 5 3 7 4
2 2 2 2


Figure 2.6: Two periods of the position, velocity and acceleration of a body with a
sinusoidal movement: x(ω t) = x̂ sin(ω t) with x̂ = 1 and ω = 1.

x̂ sin(2π f t) with an amplitude x̂ of 1. The blue line in Figure 2.6 shows that
the position at π/2 is stationary. This means that the velocity is zero, cor-
responding to the green line which was drawn at the cosine function. Also
when observing for instance the steepest down slope of the position at π it
shows that then the velocity is maximum negative, which is logical. This
reasoning can be repeated at any position in the graph. For the relation
between velocity and acceleration the same reasoning results in the red line
which was drawn as −1 times the sine function.
In fact by just looking at the graph, it is visually verified that the mathe-
matics are correct. After all, when the position ( x) can be described as:

x( t) = x̂ sin(ω t) (2.15)

then the speed (v) equals the derivative of position over time:

d x( t)
v( t) = = ẋ( t) = x̂ω cos(ω t) (2.16)
dt

and the acceleration (a) equals the derivative of velocity over time:

dv( t) d2 x
a( t) = = = ẍ( t) = − x̂ω2 sin(ω t) (2.17)
dt d t2
The visual representation is however not fully representative for the real
situation, because it only gives the right answer on the amplitude of velocity
and acceleration for a specific frequency when ω = 2π f = 1. These amplitudes
at another frequency can be found by straightforward reasoning, as a higher
frequency means a faster change. For a constant position amplitude, the
amplitude of the velocity increases proportional with the frequency and the
amplitude of the acceleration with the frequency squared.
As a second step, the phase relationship between these three parameters is
52 Chapter 2. Electricity and frequency

important. The graph is just a time sample of an endless continuous signal

and it could be concluded that velocity either advances on position with a
phase of φ = π/2 [rad] ≡ 90◦ or lags with a phase of φ = 3π/2 [rad] ≡ 270◦ . The
second option is however not logical as the position of an object is the result
of and thus comes after the velocity. In mathematics this is noted in the
following way:


t1

x( t 1 ) = x( t 0 ) + v( t) d t (2.18)
0

Where x( t0 ) is the starting position at t0 and x( t1 ) is the position at t1 .

Likewise this is true for the acceleration that advances on the velocity. This
means that the acceleration advances on the position with φ = π [rad] ≡ 180◦ .

2.1.3.2 Random signals or noise

The third kind of signals next to DC and AC is noise. Noise is a real random,
non-deterministic signal which means that the signal value at any time
of observation can have any value within a certain range. The range is
described by means of statistical distributions that give the probability that
the value is within that range. In Section 2.3 it will be explained that random
signals in the time domain consist of an infinite amount of frequencies in
the frequency domain. Also the distribution of these frequencies, called
a frequency spectrum, can only be described in statistical terms. Noise is
sometimes named with a colour term to indicate their frequency spectrum
like white noise with an equal presence of all frequencies over the entire
frequency spectrum or pink noise when the low frequency components are
more stronger present than the high frequency components
High performance Mechatronic systems exist by virtue of predictable and
well controlled dynamics of the motion systems and to that respect random
signals are by definition a part of the possible disturbances that impair the
functionality. The only thing one can do is avoid or suppress the effects of
noise but it can never be cancelled as that would require knowledge of future
values and then the signal would no longer be random. The reduction of
noise has always been an important area of research by determining the
mechanisms behind the noise in order to improve the predictability and at
least filter out the systematic effects that occur at regular intervals.
Noise and its statistical analysis is examined more in depth in Section 8.2
on Dynamic Error Budgeting where it will be explained how to deal with
random disturbances in precision measurement and positioning systems.
2.1. Electricity and signals 53

2.1.3.3 Power of alternating signals

Sometimes electronic signals are represented by their power value, being

the squared momentary value. This notation is for instance useful, when
determining the impact on accuracy of different random disturbance signals
on a system.
In case of a voltage signal, this power value would be equal to the power
dissipated in a 1 Ω resistor, as in that case the current would be equal
to the voltage. In an alternating signal the power also varies over time.
With a purely resistive load, the power is always positive, with Ps ∝ sin2 (ω t).
Although, as will be shown in several examples in this book, the power can
be negative, when the current and the voltage have a different sign.
When examining the power of a signal, as defined by the squared momentary
value, often the average value over time is taken as a representative number.
This average value of the power Ps of a signal function f ( t) is equal to:

T
1
Ps = 〈 f ( t)〉2 d t (2.19)
T
0

In the example of a voltage signal over a 1 Ω resistor this average power

level would be equal to the power dissipated in the resistor by a DC voltage
with a value of:


  T
1
Vrms =  〈 f ( t )〉2 d t (2.20)
T
0

The term “rms” refers to Root Mean Square (RMS), named from the action
of taking the root of the mean value of the squared function. The RMS value
is a well-known term in electricity to characterise the useful value of the
mains supply with a sinusoidal voltage V = V̂ sin(ω t).
The equivalent DC voltage becomes:
 
 
  T   T
1 1
Vrms =  (V̂ sin(ω t)) d t = V̂ 
2 sin2 (ω t) d t
T T
0 0

 

  T
1 1 − cos(2ω t) 1 t sin(2ω t) T
= V̂  d t = V̂ − (2.21)
T 2 T 2 4ω 0
0

By definition the RMS value has to be calculated over infinity to converge.

Fortunately with repetitive periodic signals, the value is equal to the inte-
gration over n times the full period of the signal, where n is an integer. This
54 Chapter 2. Electricity and frequency

means that the sine term will average to zero and the final result becomes:


1 t T T V̂
Vrms = V̂ = V̂ = (2.22)
T 2 0 2T 2

The 230 V AC mains supply in the Netherlands is the RMS value of the

voltage so the peak value equals V̂ = 2 · 230 = 325 V.

Similar to the voltage, the RMS value of a sinusoidal current will equal Î/ 2
and the combined power will equal VRMS I RMS = 0.5V̂ Î as is proven with the
following seasoning, when I (ω) = Î sin(ω t) and V (ω) = V̂ sin(ω t):
1 − cos(2ω t)
P = I (ω)V (ω) = P̂ sin(ω t) sin(ω t) = P̂ sin2 (ω t) = P̂ (2.23)
2

with the peak power P̂ = Î V̂ . The result is a positive number and averages
out over time to P = P̂/2 because the average value of a sinusoidal function
is zero.

2.1.3.4 Representation in the complex plane

The dynamic behaviour of mechatronic systems is generally described in

transfer functions for signals that act on the system. These functions are
derived both in the time and frequency domain. The outcome of these
transfer functions in the frequency domain describes the amplitude and
phase of the output of the system in relation to the input as function of
the frequency. It uses the mathematics of complex numbers because of its
possibility to relate the phase to an angle in the complex plane as shown in
Figure 2.7. The complex number Z = a + jb is represented in the complex
plane as a vector with coordinates [a b]. In general not the vector notation
but the following relation is used to describe a complex number with a phase
angle φ:

Z (φ) = A (cos φ + j sin φ) = Ae jφ (2.24)

The exponential function is used in the mathematical derivation of several

important transforms but for more understanding of the phenomena the
sinusoidal and graphical representation is often used. As an example
Figure 2.8 shows Ohm’s law in a more generic representation where all
values can be frequency dependant. For the load, this frequency depen-
dency automatically leads to a phase shift between the load current and the
voltage, as will be presented more in detail in the following chapters. For
this reason the load is said to have an impedance Z , with often a complex
2.1. Electricity and signals 55

Im

Z(M )

A
j sinM

M
cosM Re

Figure 2.7: Vectorial representation of complex number Z(φ) = A(cos φ + j sin φ) =

Ae jφ where the phase shift is equivalent to the angle φ.

mathematical description like Z (ω) = R + j ωL, which makes it a load with a

complex impedance.
In its generic form Ohm’s law becomes as follows:
V (ω)
V (ω) = I (ω) Z (ω) ⇒ I (ω) = (2.25)
Z (ω )

Also the relation for power becomes less trivial as a phase shift between
current and voltage has impact on the related power. Without a phase
shift, the power was shown to average out over time to P = P̂/2. In case of
a π/2 phase shift, where the current advances over the voltage, the Power
becomes:
1
P = I (ω)V (ω) = P̂ sin(ω t − π/2) sin(ω t) = P̂ cos(ω t) sin(ω t) = P̂ sin(2ω t) (2.26)
2

Current I()

+
Voltage
+ source

V () Impedance Z()

Figure 2.8: Generic representation of Ohm’s law. All variables can be frequency-
dependent. The load becomes a “complex” impedance resulting in a
possible phase shift between current an voltage.
56 Chapter 2. Electricity and frequency

1
Sine Sine
1
Power Cosine
JQ8 JQ8 Power
JQ6 JQ6

JQ4 JQ4

JQ2 JQ2

0 0

>JQ2 >JQ2

>JQ4 >JQ4

>JQ6 >JQ6

>JQ8 >JQ8

> >

0 1 2 0 1 2
Period Period

Figure 2.9: Graphical representation of power as the multiplication of two sinu-

soidal functions. At the left side the functions are in phase, resulting in
an average power of 0.5. At the right side the functions have 90◦ phase
difference, resulting in an average power over time of zero.

This is a sinusoidal function with an average value of zero. The conclusion

that can be drawn is that only the in-phase components of the current and
voltage contribute to the average power into the load which is for instance
the case with a resistor. The derived relation between power and phase is
illustrated in Figure 2.9. When the phase difference is shifted gradually
from zero to 90◦ , the average power shifts from 0.5 to zero.
It is important to note from the figure that, although the average power is
zero, the momentary value of the power is not. At half of the time the power
is positive, while in the other half the power is negative. This phenomenon
of temporary energy storage occurs in many reactive impedances, as will be
presented in the following chapters. In mechanics the mass of the body and
the stiffness of a spring are reactive impedances, while the capacitor and
inductor are the reactive impedances in an electronic system.
2.2. Energy propagation and waves 57

2.2 Energy propagation and waves

Directly related with frequency is the physical phenomenon of a wave. Waves
are well-known from real life by the waves in water. In spite of this seemingly
simple phenomenon, physical waves are not as straightforward to describe
as it may seem. A valid first observation is, that waves deal with energy
that is transferred to another place by a consecutive process that happens
over the wave. This transfer is called propagation and happens with a
certain velocity, which is called the propagation speed vp . At a certain
frequency, this propagation speed results in a spatial periodicity, that is
called the wavelength with variable-symbol (λ), related to the frequency by
the following simple relation:

vp
λ = vp T = (2.27)
f
In the following, two essentially different types of waves will be presented,
the real and observable mechanical waves, that are based on movements
of material, and the physical-model-based electromagnetic waves. The
examples are aimed to only create a basic idea of these phenomena that are
sufficient to use it in this book. For that reason a full mathematical analysis
with wave-equations is not presented.

2.2.1 Mechanical and acoustic waves

The propagation of mechanical waves through an elastic material can be

explained with the help of a simplified ideal multi-body model consisting of
a chain of springs and masses as shown in Figure 2.10. It is unsupported,
floating in outer space and contains no damping parts. In the figure two
different kind of waves are shown, longitudinal in the direction of the
propagation direction and transversal, perpendicular to the propagation
direction. In a three dimensional situation, the transversal movement can
be in any direction orthogonal to the propagation direction. In the case of
an electromagnetic wave, the transversal movement direction is called the
polarisation direction.
To explain the principle of energy transfer, the longitudinal waves are taken
as example, while the reasoning is also valid for transversal waves like in
the example of the rope.
When a movement of mass m 1 is introduced in the propagation direction
of the chain, this will first cause a compression of the elastic coupling k1 .
58 Chapter 2. Electricity and frequency

k2
k1 k n-1
m1 m2 mn
m n-1
Transversal wave

k1 k2 kn-1
m1 m2 m n-1 mn

Longitudinal wave

Figure 2.10: Transfer of energy by waves in a mechanical body can be either lon-
gitudinal or transversal. Transversal waves can be in any direction
orthogonal to the propagation direction which is always longitudinal.

The resulting compression force is transferred to mass m 2 which accelerates

resulting in its own movement which causes in its turn a compression of k2 .
This process is repeated over the total chain until the original movement
reaches m n . With this mechanism the kinetic energy from mass m1 is
converted into potential energy in k1 which in its turn is transferred into
kinetic energy of m 2 and so on until the last mass is moving. This all
under the assumption that no energy is lost. This phenomenon of transfer
of energy in an elastic body is important in mechatronic systems because
driving forces are also transported through the body as a wave and as a
consequence will experience a delay between the actuator and the sensor
when they are located separately. The propagation velocity vp is determined
by the elastic compressibility property, expressed by the Bulk elasticity B,
and the density ρ of the medium and equals approximately:

B
vp = [m/s] (2.28)
ρ

This equation is valid for gases and fluids and to a lesser extend for solid
objects where it can be used only as approximation. The shape of the object
also plays a role and the velocity for transversal and longitudinal waves is
different. As an example with steel, B ≈ 160 · 109 M/m and ρ ≈ 8 · 103 kg/m3
resulting in a calculated velocity of approximately 4500 m/s. In reality the
velocity in stainless steel varies between 3500 m/s for transversal waves and
5500 m/s for longitudinal waves. With for instance half a metre of steel this
gives a delay of about one tenth of a millisecond which results in a phase
delay of almost thirty-five degrees at one Kilohertz which can be significant
from a control point of view!
Acoustic waves that relate to sound are a special subdivision of mechan-
ical waves. In principle they can be visualised with the help of the same
Figure 2.10. However with sound the medium is gas and this means that
2.2. Energy propagation and waves 59

B
t/x
Magnec eld

Electric eld

Figure 2.11: Electromagnetic waves consist of a coupled periodic electric and mag-
netic field that propagate as function of time and position. In vacuum
the velocity is equal to the speed of light c, exactly 299,792,458 m/s.

the mass elements are very small (molecules), while the elastic elements
are also very small as they consist of the interaction between the moving
molecules. On a macro-scale, the energy in acoustic waves is transferred
by pressure and velocity where in practice only longitudinal acoustic waves
exist because of the low viscosity, which could be the only mechanism to
transfer transversal movements.
The example shown is still very much one dimensional and it is worthwhile
to consider the three dimensional case at this point as the inserted energy
is in reality transferred in all directions. To take this into account the
transferred energy is defined as the energy per unit of surface. This value
is often called intensity. Based on the law of conservation of energy this
means that the intensity value will decrease as function of the enlarging
surface which often in practice corresponds to something like a half sphere.
When motion energy is inserted at the surface of an object, the intensity will
decrease proportional with the squared distance from the point of insertion.

2.2.2 Electromagnetic waves

A special type of waves is the electromagnetic wave, because it does not use
material for its propagation. Electromagnetic waves as a physical model
were first postulated by James C. Maxwell (1831 – 1879) to describe phenom-
ena like radio transmission and light. At first, people who were accustomed
to the mechanical wave theory thought that electromagnetic waves should
also have a carrier to propagate the energy. For this reason they invented
an unknown medium called “Aether” as a wave carrier. Present theory
60 Chapter 2. Electricity and frequency

however states that electromagnetic energy is transferred by the interaction

of electromagnetic fields. In fact this postulation of electromagnetic energy
propagation as a wave is, like all physics, only a model to describe and pre-
dict behaviour of a physical phenomenon without explaining anything. It is
only valid under certain constraints. The constraints of the electromagnetic
wave theory only apply when the energy levels are so small that the energy
transfer starts to behave as separate single units of energy (quanta) that
were first postulated by Max K.E.L. Planck (1858 – 1947). The relating
theory of Quantum Electrodynamics on the interaction of photons and elec-
trons has been described by Richard P. Feynman (1918 – 1988) in his well
readable book “QED, The strange theory of light and matter”. It gives a
calculation model for electromagnetic phenomena by means of probability
statistics. Though it is in itself very interesting, especially to become aware
of the futility to search for answers why nature does the weird things it
does, reading it is not necessary to be able to work with electromagnetic
phenomena at the scale used in mechatronic systems. The wave theory is
mostly sufficient for this purpose. Nevertheless it will be shortly referenced
in the optics chapter, when presenting diffraction on gratings.
Figure 2.11 shows the transversal wave model of an electromagnetic wave
as a combination of two orthogonal periodic fields, a magnetic (B) and an
electric (E) field. This model is used in Chapter 7 on optics to describe
phenomena like interference and diffraction. Electromagnetic waves cover
a very wide spectrum as shown in Figure 2.12. It starts with the Gamma
radiation or Gamma rays at one extreme of the spectrum and radio waves at
the other. The important area of visible light is found around a wavelength
of 600 nm. Electromagnetic waves propagate with the speed of light c, which
is in vacuum exactly 299,792,458 m/s. This enormous velocity results in an
extreme unmeasurable temporal frequency of ≈ 5 · 1014 Hz for visible light.

2.2.2.1 Transferred energy and amplitude

It has been mentioned that waves represent the transfer of energy by means
of an oscillating medium. In the mechanical wave this energy is alternatingly
present in the form of kinetic energy or as potential energy. When looking
at the maximum kinetic energy E k = 1/2 mv2 at any point in a mechanical
wave it is clear that the related intensity is proportional to the amplitude
squared. Also with electromagnetic waves the intensity is proportional to
the amplitude squared, but with these waves generally the term irradiance
is used rather than intensity because in radiology the term intensity or
radiant intensity is also used as the radiated power per unit of solid angle
2.2. Energy propagation and waves 61

1015 1010 108 Frequency [Hz]

10-14 10-12 10-10 10-8 10-6 10-4 10-2 1 102 104

Wavelength [m]

400 500 600 700

Wavelength [nm]

Figure 2.12: Electromagnetic waves are observed over a vast spectrum of frequen-
cies, ranging from radio waves to gamma rays. The spectrum of visible
light covers only a very small part.

originating at the source of the wave. It is typical for waves that start at a
singular spot. The term irradiance will be frequently used in the chapters
on optics and measurement.

2.2.3 Reflection of waves

A rope is an example of a mechanical chain consisting of an infinite number

of interconnected masses and springs. Figure 2.13 shows a situation where
a man inserts a sequence of an upward and equal downward movement in a
rope causing a transversal half wave. The rope is not connected to anything
at its end point and the effects of gravity are neglected. When the wave
arrives at the end point, the momentum can not be transferred further to a
next neighbouring part of the rope. This zero force boundary condition at
the end point results in a continuation of its upward movement. This lasts
until the originally driving force from the preceding part of the rope starts
to slow down the upward movement of the end point, while transferring
the momentum back to the previous part of the rope. This effect is called
reflection and the resulting backwards moving half wave has the same sign
62 Chapter 2. Electricity and frequency

2
Sequence

Figure 2.13: A man holds a rope with a loose end and inserts a half wave by an
upwards movement followed an equal downward movement. The
reflection of the wave at the loose end has the same sign because of the
zero force boundary condition at the end point. The effect of gravity is
neglected.

as the original half wave, because it first starts with an upward momentum,
originating from the extreme upward motion of the end, followed by a down-
ward movement.
Another situation occurs when the end of the rope is connected to a rigid
stationary object like a wall. This can in realty be visualised also with
presence of gravity and is shown in Figure 2.14. This zero movement bound-
ary condition causes the force at the connection point to increase until all
energy in the wave is absorbed in the local elasticity in the rope. That
potential energy is inserted back in the rope such that its movement gains
a maximum momentum passing the centre line until its slowed down again
at the downward side where its energy will be transferred in the backward
direction. The reflection in this situation leads to a sign reversal, because
it first starts with an downward momentum, originating from the extreme
force at the connection point, followed by an upward movement.
Of course these two examples can also be calculated with real mathematics.
This would enable to precisely calculate the transfer over a discontinuity and
show the minus sign in the reflection at a rigid connection but as mentioned,
is not necessary for a sufficient comprehension of the phenomena in the
2.2. Energy propagation and waves 63

2
Sequence

Figure 2.14: The reflection of a movement at a rigidly connected end results in a

sign inversion of the reflected wave due to the zero movement boundary
condition. In this situation the force at the connection point will not
only stop the transversal wave but reflect it in the downward direction.

scope of this book. This phase reversal is also observed with the reflection
of electromagnetic waves when going from a low density to a high density
medium like with reflecting on a mirror.

2.2.3.1 Standing waves

A special situation occurs when the man inserts a sinusoidal movement in

the rope with such a frequency that the reflected wave arrives in phase with
the exerted wave. This situation is shown in Figure 2.15. The connection
point at the wall leads to sign reversal and the side of the man to the same
phase. Sign reversal of a sinusoidal wave is equal to a 180◦ or λ/2 phase
shift, which means that at least another 180◦ phase shift is necessary to be
in phase again with the inserted movement. This is accomplished when the
wave travels a distance of at least λ/2. because the wave travels twice the
distance from the man to the wall the minimal distance becomes λ/4. But
also with a distance of d = 3 · λ/4 an in phase reflection is obtained, because
a path of two times 2 · λ/4 equals a full wavelength, giving a 360◦ phase shift.
This can be written in the following generic way, when using the relation
64 Chapter 2. Electricity and frequency

Node An node

Node An node Node An node

d

Figure 2.15: A standing wave occurs with a sinusoidal excitation when the reflected
wave is in phase with the excited wave. This occurs at a frequency
equal to f = nvp /4d with n = 1,3,5,7..... The situation for n = 3 is shown
in the upper graph and n = 5 for the lower graph.

between frequency and wavelength of Equation (2.27):

vp nλ nvp
f= , d= =⇒ f= , n = 1, 3, 5, 7, ..... (2.29)
λ 4 4d
Due to the in phase reflection at these frequencies the velocity at the point
of insertion will also be in phase with the force by the man, achieving a con-
tinuous energy transfer to the rope and consequently the motion amplitude
will increase. This effect is called a resonance as will be also presented in
Chapter 3 on dynamics. In the figure, fixed positions ar shown, where the
amplitude is zero and those positions are called a “node”. At these positions
the energy is stored in elastic deformation only. The locations where the
motion amplitude is maximum are called the “anti-node” and the energy
is stored in kinetic energy at the highest velocity moment when the rope
passes the centre line.
It is important to notice that the excitation force has to be inserted at an
anti-node to be effective. This is true because insertion at a node with zero
velocity will not transfer energy and at an anti-node the velocity will be
maximal. This effect is used in any musical snare-instrument where the
musician exerts forces in an anti-node on the snare to create the sound.
Depending on the position of exertion a different tonal balance is achieved
because with each frequency another place will correspond with an anti-
node. In a guitar the sound contains more high frequencies when played
closer to the bridge, the connection of the snare with the instrument. Close
to the bridge the anti-nodes of only these high frequency harmonics are
located which is also observed in Figure 2.15 where the wall corresponds
with the bridge.
2.3. Mathematical analysis of signals and dynamics 65

2.3 Mathematical analysis of signals and dy-

namics

Several mathematical methods are used to model dynamic phenomena and

signals in mechatronic systems. In this section two related mathematical
transform principles are presented, the Fourier and the Laplace transform
that are frequently used for this purpose

2.3.1 Fourier transform

In real mechatronic systems a sinusoidal movement of one single frequency

can only occur at a resonance frequency. Real life is full of periodic events
with many different frequencies that occur simultaneously without any
correlation. Fortunately the French mathematician Jean Baptiste Joseph
Fourier (1768 – 1830) determined that any periodic signal can be seen as
a combination of sinusoidal signals with a clear harmonic frequency inter-
relation. This is true under the condition that the function contains no
discontinuities as these would lead to a non-converging series. Due to the
natural source of the periodic events of interest, these conditions are always
met.
The mathematical method is called the Fourier transform because it trans-
forms a function in the time domain to a function in the frequency domain.
It consists of a series expansion and is based on the following trigonometric
identity:
− cos(ω1 t + ω2 t) + cos(ω1 t − ω2 t)
sin(ω1 t) sin(ω2 t) = (2.30)
2
When ω1 = ω2 the first term in the numerator becomes a standard cosine
function averaging to zero over time and the second term in the numerator
becomes equal to one resulting in a non-zero (0.5) total average of the
function. In case ω1 and ω2 are not equal both terms in the numerator
would become simple cosine functions both averaging to zero over time.
Based on this sharp difference between the two situations, the average
value over time of the multiplication of an arbitrary periodic function with a
sinusoidal function of a certain frequency would be zero, unless the arbitrary
periodic function would contain a frequency component equal to, and in
phase with, the sinusoidal function.
With this mathematical conclusion the Fourier series of a periodic function
is calculated as follows. First the formula will be applied in the form where
66 Chapter 2. Electricity and frequency

the value is a function of the angle θ instead of time and the interval of one
period T is represented as an angle from −π to π. When F (θ ) represents the
Fourier transform of f (θ ) then:

∞
F (θ ) = a 0 + [a n cos( nθ ) + b n sin( nθ )] (2.31)
n=1

Where n is an integer n ≥ 1. Because the angle θ equals ω0 t where ω0

represents the fundamental frequency of the periodic function in the time
domain, the Fourier transform of this temporal periodic function can also
be written as:

∞
F ( t) = a 0 + [a n cos( nω0 t) + b n sin( nω0 t)] (2.32)
n=1

The sine and cosine terms define the phase relation between the different
components. The Fourier coefficients a n and bn are the amplitudes of the
corresponding sine and cosine terms belonging to the different frequencies
and are calculated over the mentioned interval by the next relations:
π
1
a0 = f (θ ) d θ (2.33)
2π
−π

This first coefficient represents the average value and is zero for a periodic
signal without a DC offset. The other terms become according to the above
mentioned trigonometric identity:
π
1
an = f (θ ) cos( nθ ) dθ (2.34)
π
−π

and:
π
1
bn = f (θ ) sin( nθ ) dθ (2.35)
π
−π

In equation (2.32) the frequency terms are all an integer multitude of the
fundamental frequency. This interrelation is called harmonic because in
music a waveform that consists of only harmonically related frequencies does
sound like a harmonic (not out of tune) tone. The fundamental frequency
equals the first harmonic and likewise the frequencies corresponding with
n = (1, 3, 5, ...) are called odd harmonics and the frequencies corresponding
with n = (2, 4, 6, ...) are called even harmonics2 . To illustrate and underline
2 The terms “odd” and “even” should not be confused with odd and even functions in math-

ematics that deal with symmetry relations of functions. A function is called even when the
graphic representation in the x − y plane is unchanged when mirrored around the y axis. The
graphic representation of odd functions remain unchanged when rotated 180◦ around the origin.
For example cos x is an even function while sin x is an odd function.
2.3. Mathematical analysis of signals and dynamics 67

>1

1
2
Magnitude

0 0.9
0.8
0.7

Magnitude
0.6
>1 

"! 0.5
2 X
 0.4
X
Z
0.3
X
\
0.2
!"

>1 0.1
>π > π 0 π π 0
2 2 0 1 2 3 4 5 6
Angle Harmonic

Figure 2.16: Triangle waveform, approximated by the sum of the first three har-
monic terms of the corresponding Fourier series, already comes close
to the ideal waveform.

this theory on Fourier series of periodic functions the following three wave-
forms are taken as example, the triangle, sawtooth and square waveform.

2.3.1.1 Triangle waveform

The triangle waveform is shown in Figure 2.16. It is clear that a 0 is zero as

the waveform has an average value over time of zero. Also the fundamental
frequency is most probably a cosine function as it is like the cosine an
even function. The Fourier terms can be calculated from the mathematical
expression that describes the triangle waveform as function of the angle θ .
2
f (θ ) = 1 + θ for − π < θ ≤ 0 (2.36)
π
2
f (θ ) = 1 − θ for 0<θ<π
π

When calculating the coefficients of the Fourier series of this function in

time notation (θ = ω t), the following infinite series of frequencies is found:

8 1 1
F ( t) = cos(ω t ) + cos(3ω t ) + .... + cos( n ω t ) n = 1, 3, 5, 7, ......
π2 32 n2
(2.37)

In the figure indeed the three frequencies all have a maximum at θ = 0,

which means that by summing them up, the top pushes up to shape a
68 Chapter 2. Electricity and frequency

>1

1
2
Magnitude

0 0.7


"! 0.6
X


Magnitude
0.5
X
^
>1 X
Z 0.4
2
X
 0.3
X
\
X
` 0.2
>1 !"
 0.1

>π > π 0 π π 0
2 2 0 1 2 3 4 5 6
Angle Harmonic

Figure 2.17: Sawtooth waveform, approximated by the sum of the first six harmonic
terms of the corresponding Fourier series, is still hardly recognisable
as a real sawtooth.

sharp tip if all terms would be included. It is also nice to see that between
for instance θ = π/4 and π/2 the combination of the second and third term
leads to a flattening of the curved slope of the first term in that region.
The mathematics also show, that only odd harmonics are present. Due
to the quadratic term in the numerator of the amplitude coefficients the
higher order terms are rapidly becoming negligible. The image in the figure
consists of only three terms from the Fourier series and the sum already
looks like a triangle. Nevertheless due to the sharp edges an infinite amount
of harmonics would be needed to create an ideal triangle waveform.

2.3.1.2 Sawtooth waveform

The sawtooth waveform is shown in Figure 2.17. While it might look like
another kind of triangle it is not that simple. A theoretical ideal sawtooth
waveform has an infinitely steep slope around θ = 0 which represents a
discontinuity. As a consequence many terms are required to approximate a
sawtooth waveform to an acceptable level of accuracy. Like with the triangle
a 0 is zero but in this case the fundamental frequency of the range is a sine
function as the sawtooth is like the sine an odd function. The mathematical
relation that describes the sawtooth waveform as function of the angle θ is
2.3. Mathematical analysis of signals and dynamics 69

as follows.
1
f (θ ) = −1 + θ for − π < θ < 0 (2.38)
π
1
f (θ ) = 1 − θ for 0<θ<π
π
f (θ ) = 0 for θ=0

Calculating the coefficients of the Fourier series of this function in time

notation (θ = ω t), gives the following infinite series of frequencies:

2 1 1
F ( t) = sin(ω t) + sin(2ω t) + .... + sin( nω t) n = 1, 2, 3, 4, ......
π 2 n
(2.39)

This means that both odd and even harmonics are present. Furthermore, the
amplitude coefficients of the successive harmonics only decrease proportional
with n which underlines that many terms are required to create a reliable
sawtooth. This is due to the above mentioned fact that the infinitely steep
ramp has to be created by summation of only finitely steep sine wave slopes.
The image in the figure consists of only six terms and though one faintly
recognises the final shape it is still far from a sawtooth waveform.

2.3.1.3 Square waveform

The third example is the square waveform as shown in Figure 2.18. In

this case two discontinuities are present in the signal, which also means
that more terms are required to approximate the waveform with sinusoidal
functions. Like with both previous waveforms a 0 is zero and like with the
sawtooth also the fundamental frequency of the range is a sine function as
it is an odd function. The mathematical relation that describes the square
wave as function of the angle θ is as follows.

f (θ ) = −1 for − π < θ < 0 (2.40)

f (θ ) = 1 for 0<θ<π
f (θ ) = 0 for θ = −π , 0 , π

Calculating the coefficients the Fourier series of this function in time nota-
tion (θ = ω t) gives the following infinite series of frequencies:

4 1 1
F ( t) = sin(ω t) + sin(3ω t) + .... + sin( nω t) n = 1, 3, 5, 7, ......
π 3 n
(2.41)
70 Chapter 2. Electricity and frequency

>1

1
2
Magnitude

0 1.3
1.2


"! 1.1

Magnitude
X
 1
>1 0.9
2 X
Z 0.8
X
\ 0.7
X
{ 0.6
0.5
X
| 0.4
>1
!"
 0.3
0.2
0.1
>π > π 0 π π 0
2 2 0 1 2 3 4 5 6 7 8 9 10
Angle Harmonic

Figure 2.18: Square waveform, approximated by the sum of the first five harmonic
terms of the corresponding Fourier series starts to look like a real
square.

This means that like with the triangle waveform only odd harmonics are
present. Furthermore, like with the sawtooth waveform, the amplitude
coefficients of the successive harmonics only decrease proportional with n,
which underlines that also in this case many terms are required to realise
a reliable square wave for the same reason as an ideal theoretical square
waveform has discontinuities that can only be approximated. The fact that
the summed first six terms of the Fourier series already looks more like
a real square waveform than the previous example is due to the missing
even harmonics leading to a higher frequency of the last terms that were
included in this approximation.
Based on these three examples the following observation can be made.
A periodic signal that is symmetrical around an axis in the y direction at
π/2 ± nπ will create only odd harmonics. The even harmonics all represent an
asymmetry around these axes as can be seen with the sawtooth waveform.
This is especially important with spatial frequencies in optics where in
principle only odd harmonics are present.

2.3.1.4 Fourier analysis of non-periodic signals

From the previous part it was demonstrated, that a periodic signal can be de-
composed into harmonically related frequencies. The process of calculating
the terms in the equation focused on one period only as being representative
2.3. Mathematical analysis of signals and dynamics 71

A B C D

Magnitude

Time

Figure 2.19: Graphical representation of a random waveform in the time domain.

for the entire continuous function. In reality signals often look more like
the graph of Figure 2.19 showing a random signal that was registered over a
short period of time. A real random signal consists of an infinite amount of
different sinusoidal frequencies with different (and changing!) amplitudes
and phase. When only a limited time span is observed, like shown in the
figure one can recognise areas with a low frequency like in area A – B and
with a high frequency like the area C – D. It is possible to approximate such
a random signal by a limited amount of frequencies when observing only
a limited time span. Figure 2.20 shows an example that is created by a
further execution of the above mentioned Fourier transform methodology
and it works as follows:
A sample of the signal is taken over the defined time span and this sample is
multiplied with a series of sine and cosine frequencies covering the spectrum
of interest. For example in mechatronic systems a spectrum of 0 – 1000 Hz
could be taken with steps of 1 Hz resulting in two thousand multiplications.
The outcome of each multiplication results in the amplitude of each of these
frequencies and can be represented graphically in the form of a frequency
spectrum. It is clear that a lot of computation power is necessary to do this
with a large resolution in frequencies especially when real-time information
of fast changing signals with many samples is needed. Even with the fast
computing systems of today this discrete Fourier transform (DFT) is not
used anymore as the same result can be achieved by means of a more
efficient algorithm called Fast Fourier Transform (FFT). This method of
which the mathematical details go beyond the scope of this book has become
the standard for signal analysis and is for instance applied in instruments
like a Spectrum Analyser. Next to the digital FFT operation, Spectrum
72 Chapter 2. Electricity and frequency

JQ

0.3

0.2

0.1
Magnitude

0
0.12
>JQ
0.1

Magnitude
>JQ^ 0.08

>JQZ 0.06



"! 0.04
>JQ
!"
 0.02
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0
0 10 20 30
Time Harmonic

Figure 2.20: A random waveform approximated by the sum of several harmonic

terms of the corresponding Fourier series. It shows that even a random
signal observed over a limited time span can be reliably represented
by a combination of sinusoidal waveforms.

Analysers can also be based on analogue mixing of the unknown signal with
a known frequency and creating a multiplication by non-linearity in the
amplification. This method is used in very high frequency analysers.

2.3.2 Laplace transform

The Laplace transform is a mathematical method that was created by the

French astronomer and mathematician Pierre-Simon, Marquis de Laplace
(1749 – 1827). This transform is very useful for the analytical description
of dynamic systems as it enables to solve differential equations in the time
domain by solutions in the frequency domain. It is very much related to
the Fourier transform and will be used throughout several chapters in this
book, for which reason it is included in this chapter.
In case f ( t) is a function of the time variable t, then the Laplace transform
f ( s) = L { f ( t)} is described as a function of the Laplace variable s in the
following way:

∞

f ( s) = L { f ( t)} = e−st f ( t) dt (2.42)

0
2.3. Mathematical analysis of signals and dynamics 73

The Laplace variable s is a complex number:

s = σ + jω (2.43)

where j 2 = −1 (also often noted as i ) and σ and ω are real numbers. A

complete treatise on all aspects of this transform falls beyond the scope of this
book. The most important result of this transform for time varying functions
is the possibility to replace the differential over time by the Laplace variable
s and by j ω. When investigating the frequency response of a dynamic system,
the real number σ can be neglected but it will be used in examining the
poles and zeros of a transfer function, as will be presented in Chapter 3 on
dynamics and Chapter 4 on motion control.
The Laplace transform converts a differential of a variable x( t) over time
into the following expression in the frequency domain:
 
d x( t)
L = sx( s) = j ω x( s) (2.44)
dt

The same transform can be applied to an integration action which gives the
following result:
⎧ t ⎫
⎨ ⎬ x( s) x( s) x( s)
L x( t) d t = = = −j (2.45)
⎩ ⎭ s jω ω
0

it can be concluded that the s term due to the differentiation over time of
a variable results both in a proportional increase of the magnitude of the
variable with increasing frequency by the multiplication with ω and in a
positive phase shift of 90◦ corresponding with the positive imaginary term
j . Following the same reasoning, the integration of a variable over time
gives both a proportional decrease of the magnitude of the variable with
increasing frequency because of the division with ω and a phase shift of
−90◦ corresponding with the negative imaginary term − j .
The frequency domain, when written with the s variable, is also frequently
called the Laplace domain. It is important that equations from the frequency
domain are not confused with equations from the time domain. When it
is necessary to clearly distinguish the different domains, the notation of a
function like f ( t), with the variable t between round brackets, strictly refers
to the time domain. The notations with the variable (s) for the Laplace
domain, (ω) for the radial frequency or ( f ) for the temporal frequency all
strictly refer to the frequency domain. The addition of the distinguishing
variable terms ( t), (s), (ω) or ( f ) will in many cases be only done once in every
equation at the term(s) before the equal sign in order to avoid too many
74 Chapter 2. Electricity and frequency

brackets in the equations. This limited notation is allowed and sufficient as

an equation is uniquely valid, either when the entire equation is defined
in the frequency domain or when the entire equation is defined in the time
domain.

2.4 Dynamic system response to a stimulus

In the previous part of this chapter signals are introduced in terms of

frequencies. As a next step it is necessary to link this knowledge to the
analysis of the dynamic behaviour of mechatronic systems both in the time
and frequency domain. This implies the introduction of the term stimulus
as being a specific signal, created with the purpose of stimulating a reaction
from a dynamic system. In most cases it is a force created by actuators or
by external vibrations.
The experimental investigation into dynamic properties of a mechanical
system is always done by applying external forces on different locations of the
system and by simultaneous measurement of the response of the system to
these stimuli. The frequency content of these stimuli can be quite different.
Depending on their origin, several typical responses to standardised stimuli
are distinguished of which the most important are:

• Step Response to a unit step stimulus.

• Impulse Response to a unit impulse stimulus.

• Frequency Response to a stimulus with a wide frequency spectrum

that is continuously available.

2.4.0.1 Step response

The step response is a typical time domain related phenomenon and gen-
erally the most straightforward of all responses. It is often quite easy to
understand or to visualise what happens when a force on an object increases
from zero to a certain value. For instance when putting a heavy load in
a car one sees the car sagging with a certain speed and depending on the
suspension and the shock absorbers (dampers), a slow periodic movement
with decreasing amplitude is observed. In Figure 2.21 a typical response
of a dynamic system to a sudden change of force (step) from value 0 to 1
2.4. Dynamic system response to a stimulus 75

1.5

1
F

0.5
x

0
>^ 0 ^ 4 6 8 10
t

Figure 2.21: Typical response of a damped mass spring system to a force on the
mass that changes from 0 to 1 at t = 0. The vertical axis is normalised
to 1 at t = ∞.

is shown. The dynamic system in this case is a damped mass spring sys-
tem like presented in Chapter 3. This response, if considered upside down,
corresponds to the described response of the car.
Even with the origin of the step response in the time domain its equivalent
in the frequency domain can be derived by the Fourier transform. This
shows that a step function consists of a continuous spectrum of sinusoidal
frequencies starting at 0 Hz that all are in phase at the moment of the step
and which have an amplitude that is inversely proportional to the frequency.
This can be imagined when one considers that all those frequencies will only
be correlated (in phase) at t = 0 but at any other time their correlation is
lost and the combination will average out to zero. In a mathematical sense
a step function is a special case of a square waveform with a fundamental
frequency of 0 Hz. Starting with the Fourier series of a square waveform
according to Equation (2.41), the limit of ω → 0 results in a continuous
spectrum of frequencies starting at 0 Hz.
Because of this wide frequency spectrum, a step function is a very suitable
stimulus for a passive mechanical system to investigate the occurrence
of resonances.It is important to note however that the step response is
less applicable as stimulus in an actively controlled dynamic system with
actuators as with most controllers the infinite slope will force the actuator
into saturation and cause other non-linear effects that are not observed
76 Chapter 2. Electricity and frequency

Step 2

l h

Step 1
t

Figure 2.22: An impulse function as a combination of two steps. In case of a force

step the surface defined by l and h equals 1 Ns.

under normal conditions.

Although a step is a one time event, one can as alternative approximate the
step by a low frequency square wave. This limits of course the spectrum to
the corresponding harmonic frequencies but depending on the frequency
range of interest this can help in doing multiple measurements to reduce
the impact of random noise which will be averaged out.

2.4.0.2 Impulse response

An impulse function is a special combination of two steps, one up and one

down directly after each other as shown in Figure 2.22. The ideal impulse
lasts for an infinitely short time aith an infinitely high magnitude with
a multiplied value of one. The response on such a stimulus of the same
dynamic system as was used with the step response is shown in Figure 2.23.
This impulse response looks quite similar to the corresponding step response
although it starts at x = 0 and stabilises on x = 0 again. The main reason
why this stimulus is so important and widely used is the fact that in the
frequency domain it contains all frequencies at equal amplitude. Measuring
the response to an impulse and performing a FFT analysis of this response
immediately shows all resonances of a dynamic system. Unfortunately it is
not possible to create an ideal impulse function in reality, due to limitations
at high frequencies of the physical embodiment.
2.4. Dynamic system response to a stimulus 77

JQ}

JQ`
F
JQ

JQ^
x

>JQ^

>^ 0 ^ 4 6 8 10
t

Figure 2.23: Typical response of a damped mass spring system on a force impulse
at t = 0. The vertical axis is normalised to 1 just after the impulse.

This means that the length becomes longer than zero and the peak value pro-
portionally lower than infinite while simultaneously the edges are rounded
off. To a certain extent one can compensate the corresponding lack of high
frequencies by convolution3 of the measured response with the real spectrum
of the stimulus but at higher frequencies the noise will limit the reliability
of the outcome.
The impulse response is often used for modal analysis of a mechanical sys-
tem. The impulse is applied by means of a standardised “hammer”. The
hammer is provided with a rubber tip to avoid damage by the ”infinite“
force and as a consequence, the impulse will be rounded off with a lower
amplitude of the high frequencies. By measuring the exerted impulse itself
directly on the hammer, the deviation of the ideal pulse can be taken into
account.
One important drawback of this stimulus is that all the energy is inserted
in the system in a very short time with a high force over a short time. In-
evitably this will give erroneous information in case of non-linear systems.
As with the step response, for this reason also the impulse is less suitable
for active controlled dynamic systems.

3 Convolution is a mathematical method to define the cross-correlation between different

functions.
78 Chapter 2. Electricity and frequency

2.4.0.3 Frequency response

Frequency domain related stimuli avoid the extremely high forces that are
related to the “single event stimuli” of the previous section, because they
present a more or less permanent continuous spectrum of frequencies with
sufficiently reduced amplitudes. These signals enable longer measurements
without errors due to nonlinearity.
Sinusoidal waveforms are quite easy to create in electronics. For that reason,
initially stimuli were used that consisted of a sinusoidal waveform with a
constant amplitude and a gradually changing frequency. This “frequency
sweep” was suitable for fast electronic systems. In mechanical systems
with inherent lower frequency resonances the frequency sweep method
sometimes induced errors due to the time needed for resonances to gain
sufficient energy to manifest themselves. A fast sweep would cause these
resonances to remain unnoticed. One way to overcome this problem is using
a very slow sweep frequency if time allows and this process could be made
interactive to zoom in on critical areas. A better method is to use a stimulus
containing all frequencies over a longer time than with an impulse. The first
example is to use white noise as stimulus of which the power (signal value
squared) is constant for any fixed frequency range over the total frequency
band. In practical situations, similar to the impulse, ideal white noise does
not exist due to high frequency limitations in the physical embodiment and
only the frequency band of interest is excited.
The second example of a more continuous frequency spectrum is created in
this era of high speed digital electronics and is based on the synthesis of a
stimulus signal by means of multi-sines. In principle this approach allows to
create any spectrum of signals, with correlated phase and frequency, random
phase relations or any other combination. Frequency areas of no interest can
be avoided, different amplitudes can be chosen for different frequency areas
to avoid overheating or non-linear behaviour and the frequency spectrum can
be refined in the areas of most interest. As the signals are fully deterministic,
all phase and amplitude relations in the stimulus are known and it becomes
fully straightforward to determine the dynamic system response.
2.4. Dynamic system response to a stimulus 79

2.4.1 Graphical representation in the frequency domain

Two representations are frequently used to display the response of a dynamic

system to these frequency domain related stimuli:

• The Bode-plot to show the magnitude and phase response upon a con-
tinuous frequency stimulus in two graphs as function of the frequency,
one for the magnitude and one for the phase.

• The Nyquist plot to show the magnitude and phase response on a

continuous frequency stimulus in one graph as function of the fre-
quency.

2.4.1.1 Bode-plot

The Bode-plot, named after the American engineer Hendrik Wade Bode
(1905 – 1982), visualises the frequency response of the output of a dynamic
system to an input stimulus and is originally known from analogue electronic
signal analysis. Figure 2.24 shows an example of a Bode-plot of a fourth
order coupled mass spring system as will be introduced in Chapter 3 on
dynamics.
In principle a Bode-plot consists of two parts, one above the other, that both
share the same horizontal axis with the frequency as parameter. The upper
part is the magnitude Bode-plot with the ratio between the magnitude of
the response and the magnitude of the stimulus on the vertical axis. The
second part shows the phase shift on the vertical axis relative to phase of the
stimulus. Several important aspects have to be recognised in a Bode-plot:

• In a typical Bode-plot both the frequency and the amplitude scales are
logarithmic to base 10.

• The frequency scale can be both in rad/s or in Hz. The latter is often
preferred in the mechatronic field, mainly because of the more “natural”
character of the temporal frequency.

• The magnitude scale can be different for each application because the
units of the stimulus and the response can be different. For instance
the stimulus can be a force [N] while the response can be a position
[m]. The numbers have only a meaning when this relation is known.

• The phase scale can in theory be in radians or degrees. In practice al-

ways degrees is used because it enables an easier refinement than with
80 Chapter 2. Electricity and frequency

>^

Magnitude [m/N] [dB]

10 *J

Magnitude [m/N] (abs)

>
10 *}J
*^
>`
10 *^J

>}
10 *`J

>J
10 *^JJ
0

>\
Phase [deg]

>|J

>Z\

>}J
1 ^ 3 
10 10 10 10
Frequency (Hz)

Figure 2.24: Example Bode-plot of a motion system. The upper graph gives the
magnitude of the position response divided by the magnitude of the
force stimulus. The lower graph gives the phase relation between
the position response and the force stimulus. Starting at the low
frequency side first the response is frequency independent but at 15
Hz a damped resonance determines the start of a negative slope where
the magnitude decreases as function of the frequency. Around 3 kHz a
typical resonance and anti resonance indicate the effect of a decoupling
mass as will be explained in Chapter 3.

radians while an amount of several degrees can already be important

in a control system.

• The Bode-plot represents the response of the system when all fre-
quencies are continuously present at the stimulus. It is a station-
ary representation.

The double logarithmic scales are used because most frequency responses
are either proportional or inversely proportional to the frequency f or ω
to the power n, where n is an integer. With double logarithmic scales the
function log ωn = n log ω becomes a straight line for all values of n, because
one scale shows log ωn and the other log ω. The resulting line will have an
upward slope when n is positive and a downward slope when n is negative.
It is common to talk about the order of the slope in terms of n. For example
a transfer function f (ω) = 1/ω2 , which means that n = −2, results in a line
with a slope of −2 in the Bode-plot.
2.4. Dynamic system response to a stimulus 81

In Section 2.3.2, where the Laplace transform was introduced, a +1 slope was
shown to be representative for a differentiating action with a corresponding
phase shift of +90◦ and a −1 slope is representative for an integrating action
with a corresponding phase shift of −90◦ . This direct link between phase
and slope might lead to the conclusion that the phase plot gives redundant
information. In simple “minimum phase” linear systems, the relation be-
tween the phase and amplitude is indeed unambiguously determined.
Often, however, this relation is more complicated and the same phase might
occur at different frequencies and magnitude-slope values, especially with
higher order systems. The influence of sampling, quantisation, hysteresis,
backlash and other complicating factors is another reason to not leave the
phase plot out in the analysis of the open-loop response of modern feedback
controlled mechatronic systems. The open-loop phase plot is often even more
important for dynamic system analysis than the amplitude. This is the main
reason that the Nyquist plot has been developed. Nevertheless, due to its
clear relation with the standard applied integration and differentiation
operations in motion control systems, the Bode-plot is a frequently used
graphical representation both in practice and in this book.

Absolute magnitude or deci-Bel

The earliest research on frequency behaviour of systems was about the

properties of sound. For instance the English physicist John William Strutt
(1842 – 1919), better known as Lord Rayleigh, wrote two books on this
subject, “The theory of sound Vol 1 and 2” [1], that contains most of the basic
theory on dynamics, waves and frequency dependent transfer functions,
that still are taught today, including much of the theory in this book.
Sound levels are primarily given in deci-Bel (dB). This term originally was
introduced by the American engineer Alexander Graham Bell (1847 – 1922)
to define the weakening of a signal over a long telephone line. He decided
to use a logarithmic scale, because the effect was perceived as a factor
per unit of length. As the human perception of sound intensity is also
logarithmic, a certain relative change of sound level is always perceived the
same, independent of the actual sound level before the change.
Like many signals, sound levels are expressed either in intensity (Power) or
a signal amplitude, which is pressure or velocity for sound. These variables
can be used both as they are directly interrelated. With sound the power is
the multiplication of the complex pressure and velocity.
The 10 logarithm of the ratio between two levels of Power is expressed in “Bel”
but for reasons of practicality one tenth of this value, the deci-Bel, is chosen
82 Chapter 2. Electricity and frequency

Table 2.2: The relation between the order of the slope of a Bode-plot and the mag-
nitude ratio both in deci-Bel and in amplitude and power ratios with a
frequency difference of a factor two or ten.

Slope Per octave ( f 1 = 2 f 2 ) Per decade ( f 1 = 10 f 2 )

dB Amplitude Power dB Amplitude Power
-3 -18 0.125 0.015625 -60 10−3 10−6
-2 -12 0.25 0.0625 -40 10−2 10−4
-1 -6 0.5 0.25 -20 0.1 10−2
0 0 1 1 0 1 1
1 +6 2 4 +20 10 102
2 +12 4 16 +40 102 104
3
3 +18 8 64 +60 10 106

as incremental unit.
P1
X = 10 log [dB] (2.46)
P2

In electricity the power in a circuit is directly determined by the law of Ohm

V = I ∗ R ):

V2
P = V I = I2R = [W] (2.47)
R
For that reason the relation in dB between two magnitude levels of electrical
signals is:
I1 V1
X = 20 log or X = 20 log [dB] (2.48)
I2 V2

This means that in expressing the relation between different levels of power,
the multiplication factor is ten and between different levels of signal ampli-
tude the multiplication factor is twenty.
In terms of deci-Bel, the -1,-2,+1,+2 etc term for the slope in the Bode-plot
becomes equal to a certain number of dB per frequency ratio as shown in
Table 2.2. The values are given both for a decade, representing a factor ten
between the frequencies, and for an octave with a factor two difference, as
commonly used in sound systems and music.
While the dB is frequently used for the magnitude scale of Bode-plots,
probably because it avoids the “power of ten” term with large numbers, it
is more practical to use the absolute number when analysing the dynamic
properties of mechatronic systems. The reason for this preference is that the
2.4. Dynamic system response to a stimulus 83

magnitude level is the reference for the magnitude of the different control
parameters.
In practice the use of dB can even cause confusion when analysing the
dynamic properties of a system at different frequencies at a ratio that is
neither a factor two or ten. With for instance a factor three frequency
difference one might think this is equal to 0.33 of a decade and conclude that
a -2 slope (40 dB per decade) would then give an attenuation of −0.33 · 40 =
−13, 2 dB. In reality the magnitude attenuation would be 20 log 1/32 = −19 dB.
This error is often made by students at examinations.
In control terminology the 0 dB level is often used to indicate the frequency
where the magnitude of the transfer function of a system becomes equal to
one. In case one prefers to avoid the use of dB, like in parts of this book,
this frequency can be named the unity-gain cross-over frequency.
With the Bode-plots in this book the horizontal scale is always in Hz and
the magnitude scale is mostly absolute with the exception of the electronics
chapter where both notations are used at different sides of the Bode-plot.
The main reason for this exception is the possible application of this book
by audio engineers for whom it is more logical to use the dB notation and
electronic filters are frequently used in their field.
Even though the absolute number is the preferred notation with active
control of dynamic systems, also in this field the dB notation is still fre-
quently used. For that reason a mechatronic design engineer should always
be aware of these different notations in spite of the apparent logic behind
using absolute numbers.

2.4.1.2 Nyquist plot

The Nyquist plot, named after the Swedish electrical engineer Harry Nyquist
(1889 – 1976), is a polar coordinated two dimensional vector plot where
the magnitude M is represented by the length of the vector starting in the
origin. The phase angle (ϕ) is represented by the angle of the vector relative
to the positive horizontal axis, while its sign is positive moving against the
direction of the clock. Because most dynamic systems have a response with
a phase delay, most Nyquist plots show a phase angle in the direction of the
clock, corresponding with the negative phase that is caused by the delay.
When the different vectors for all frequencies are connected, a curved line is
obtained with the frequency as parameter along the line. This plot is often
used for the stability and robustness analysis of feedback controlled systems
as will be presented in depth in Chapter 4 on motion control. A typical
84 Chapter 2. Electricity and frequency

^
0 dB
^ … Q†
Q†Q^\
  …
Q†Q` 6 dB
Q†^ Q†ZQ^
J …

Imaginary Axis
0
(-)

>
M

>^
>^Q\
~>^QZ{
€!‚Xƒ„^{\
>Z
>\ > >Z >^ > 0
Real Axis

Figure 2.25: Nyquist plot of the open-loop transfer function of a controlled mass
positioning system. The blue line connects the polar vector points
(black arrow) defined by the phase angle relative to the positive real
axis and the magnitude as length of the vector for different frequencies.
It starts at low frequencies following the blue arrow to the origin. The
phase angle of this example is negative. The Nyquist plot is used to
show the stability of a system when the feedback loop is closed.

Nyquist plot of the open-loop response of a simple closed-loop feedback

controlled system is shown in Figure 2.25. The blue line represents the
response as function of increasing frequency in the direction of the origin.
In principle all dynamic systems are not able to respond to extremely high
frequencies. As a consequence all Nyquist plots of dynamic systems end
in the origin. The distance of the blue line to the dotted circles determine
how strongly the closed-loop feedback system will show a resonance. In the
shown example the blue line touches the circle of 2 dB/Q = 1.25 which means
that with the same amplitude of the stimulus after closing the loop, the
system will show approximately a 25 % larger amplitude of the response at
its resonant frequency than at lower frequencies.
Chapter 3

Dynamics of motion
systems

Introduction
All mechatronic systems in the context of this book represent controlled
motion systems. This statementh implies, that they are all inherently
dynamic in nature. The dynamic properties of the separate parts within
mechatronic systems each have their own impact on the performance with
respect to response speed and precision. It is fully justified to say, that
the mechanical part is often in practice the most determining factor for
the final performance of the total mechatronic system. For this reason it
is more than anything else of paramount importance to fully understand
the physical properties, that are linked with items like mutually connected
springs, bodies and dampers.
The chapter starts with the relation between the stiffness of a mechanical
construction and its inherent relative position accuracy under external loads.
This will show the relation between accuracy and natural frequency and
numerical examples will give a feeling on practical design constraints. It
includes an example of an active feedback controlled positioning system as
used in a CD-player, in order to link the feedback mechanism with stiffness,
as a short introduction towards Chapter 4 on motion control.
This introduction is followed by a more theoretical analysis of the response
to a force of a mass-spring1 system with different levels of damping, the com-
1 The term “mass-spring” or “mass-spring-damper” system is a regularly used term although

85
86 Chapter 3. Dynamics of motion systems

pliance of a system to a force. Closely related is the reaction to a movement

of the support, called transmissibility. Ample use will be made of Bode-plots
and a simplified graphical way to visualise the compliance of each element
to further clarify their impact on the overall frequency response. Also a first
step towards “multi-body dynamics” will be shown in a coupled mass-spring
system. The last section will introduce mode-shapes and the way how the
location of the measurement and actuation influences the dynamic transfer
function of a mechatronic system.
For good reasons, the dynamic properties presented in this chapter are
mostly limited to movements in one direction with linearised properties of
the elements. Main argument is, that the use of vector and matrix equations
too often diverts the attention from the real physical understanding of the
phenomena in relation to the used mathematical models. This basic under-
standing is often perceived as difficult, while it is essential for mechatronic
system design, even more than the ability to master linear algebra as in
most cases the computer will do that work for us.

a spring and damper are objects, while the mass is a property of a body. To avoid confusion,
the regular naming will be used for the complete system, while the element will be named a
body with mass m as its property.
3.1. Stiffness 87

3.1 Stiffness

Most people have a rather good qualitative feeling about what is stiff and
what is not, but it becomes different when it is necessary to work with
quantitative data based on the SI units. The unit of stiffness is called
“spring constant” k and is defined in Hooke’s law of elasticity by the English
scientist Robert Hooke (1635 – 1703). It equals the incremental change in
the amount of force ( dF ), that an elastic element would produce in reaction
to an incremental deformation ( d x). When the incremental deformation is
pointing in the positive x direction the corresponding incremental force in
that same direction2 would equal:
dF
d F = − k ( x) d x =⇒ k ( x) = − N/m (3.1)
dx
The negative sign is the result of the fact, that the force is directed in the
opposite direction of the deformation, indicating that F is negative for a
positive d x resulting in a positive number for k( x).
With small deformations the stiffness is mostly independent of the defor-
mation. In that case Hooke’s law becomes linear and can be written in the
simple notation:
F
F = − kx =⇒ k=− (3.2)
x
where x is defined as the displacement from the position where no force is
exerted on the spring.
Based on the third law of the English scientist Sir Isaac Newton (1643 –
1727), stating, that the forces of interaction between two bodies are equal,
opposite and collinear, an equal force F = kx must be applied in the direction
of the deformation to the elastic element. This equation is also called Hooke’s
law, which can cause confusion when it is not clearly defined which force is
meant, the action or the reaction. For this reason in this book the second
version of Hooke’s law is called the “Hooke – Newton” law even though these
two eminent scientists would probably not have liked to be named in one
tern, in view of their not too well personal relation.
In general the term “spring constant” is associated with the spring as a
separate item. The term stiffness is preferred in the technical context
2 The force in all drawings in this book is noted with a double arrow because it acts between

two bodies according to the third law of Newton. The direction of the force at the point of
insertion is usually defined to be positive in the positive direction of the corresponding position
coordinate system defined by one or more arrows, depending on the number of directions in
the coordinate system that are presented.
88 Chapter 3. Dynamics of motion systems

because any object, whether it is a spring or a seemingly solid block has

a certain stiffness representing its resistance to deformation. In order to
get a feeling for the order of magnitude of the stiffness values in practical
mechanical systems it is useful to imagine some real objects and estimate
how much they would deform under a known load.
Take for instance a spring board in a swimming pool. Assume, that the
mass of the swimmer is 75 kg, exerting 750 N of force and the board moves
10 cm. This corresponds with a stiffness value of the board of 7.5 · 103 N/m.
This is a low value for stiffness, corresponding with the flexible character
of a spring board. As a second example in the class-room a load of 1000
kg from 15 students is standing on a strong table. This would represent a
force of approximately 104 N while the table will probably deflect less than
10 mm resulting in a stiffness of about 106 N/m, which is already a much
higher value than with the spring board. An example of a very high stiffness
is a steel rod of 20 cm long and one cm diameter, that has a stiffness of
≈ 108 N/m.
With these and other examples it can be estimated, that practical stiffness
values range between approximately 103 and 109 N/m, which differ more
than six orders of magnitude.

3.1.1 Importance of stiffness for precision

To achieve a certain level of accuracy, the stiffness of a system is a very

important property. A system with a high stiffness will deform less in re-
sponse to an applied force and in most cases that is a benefit for a precision
system. Examples of systems, that need a reduced stiffness, are for instance
a table with four legs and a car with four wheels. Both are in principle over-
constrained systems meaning that some directions are constrained more
than once. Due to the relative flexibility of the tabletop, a table is generally
able to deform a bit, making it possible to adapt to floor irregularities and
prevent wobbling. With a well designed car suspension system the problem
of over-constrain is also solved, while simultaneously increasing the comfort
in the car.
An example of a system, where an increased higher stiffness results directly
in a better system performance, is the turning lathe that is very schemati-
cally shown in Figure 3.1. Depending on the stiffness of the support, the
tip of the cutting tool will move away from the intended position due to
the cutting forces. The force and the corresponding movement is seldom
constant and accurately known so it will limit the possibility to manufacture
precise parts.
3.1. Stiffness 89

Rotang workpiece

Lathe Cu‡ng tool

F
-dx
m
x

Figure 3.1: Schematic view of a turning lathe. The force that is exerted by the
rotating work piece in the negative direction on the cutting tool will
deform the spring with stiffness k resulting in a displacement − dx from
the intended position of the tool tip.

The typical stiffness of a turning lathe is about 107 − 108 N/m. When a maxi-
mum error of one micrometre is specified, the change in the cutting force is
not allowed to be larger than ten Newton.
Another example, where high precision is required, is the inspection micro-
scope shown in Figure 3.2. In this example, the desired maximum measure-
ment error is 10 nm. This implies, that the maximum deviation xmax from
the ideal position between the microscope and the object, has to remain
smaller than that value. This requirement directly results in a minimum
value for the stiffness between the object and the microscope, because of the
influence of vibrations of the table top that the microscope is placed on.
The following assumptions are chosen for this example:
The upper part of the microscope has a mass of 5 kg and it is used in a
laboratory, where the measured vibrations on the table top show acceleration
levels up to 0.01 m/s2 . These accelerations have to be followed by the mass
of the upper part of the microscope which requires a certain force F ( t) that
will cause a deformation of the stiffness k that is not allowed to be larger
than the maximum error of x̂e = 10 nm. The required stiffness (k), to achieve
this maximum position error under these circumstances, can be calculated
using the second law of Newton:
d2 x
m
k≥
F ( t)
=
d t2 = 5 · 0.01 = 5 · 106 N/m (3.3)
x̂e ( t) x̂e 10 · 10−9
This value is directly related to the undamped natural frequency ω0 , as
90 Chapter 3. Dynamics of motion systems

Support frame

k
m Microscope
head

xe
Object
Vibraons xf

Table top

Figure 3.2: Schematic view of an inspection microscope. Vibrations in the table top
act on the entire instrument. Due to the mass and the limited stiffness
the microscope head will not be able to perfectly follow the accelerations
causing an alignment error xe with the object.

determined from the equations of motion according to the following rea-

soning. A system, that consists of a body and a spring, will resonate in its
natural frequency, when the forces inside the system are in balance and
the motion is sustained without external forces. This means, that the force
Fd ( t), resulting from the deformation of the spring, is always in equilibrium
with the force Fa ( t), that corresponds with the acceleration of the body:

d2 x d2 x
Fa ( t) + Fd ( t) = m + kx = 0, =⇒ m = − kx (3.4)
d t2 d t2
The sign can be checked by the reasoning, that a deformation of the spring
in the positive x direction causes a force and a corresponding acceleration
in the opposite direction of the deformation. It is known, that in resonance
the body follows a sinusoidal movement x( t) = x̂ sin(ω0 t) with x̂ being the
amplitude.
With this information the following equation can be formulated:

− m x̂ω20 sin(ω0 t) = − k x̂ sin(ω0 t) (3.5)

From which follows:


k
ω0 = (3.6)
m
3.1. Stiffness 91

From the calculated value of the required stiffness in Equation (3.3), the
related natural frequency of the system can be derived:

5 · 106
ω0 ≥ = 1000 [rad/s], corresponding with:
5
1
f0 ≥ 1000 = 160 [Hz] (3.7)
2π
From Equation (3.3) it can also be concluded, that a larger mass of the
upper part of the microscope requires a proportionally increased stiffness,
to achieve the same accuracy. This would mean that the natural frequency
is kept constant. It appears, that the natural frequency is directly related
to the performance with respect to the suppression of external vibration
disturbances. To show this in mathematical form, the behaviour of the
upper part is observed. This microscope head has to follow the sinusoidal
motion of the table top vibrations with position x( t) and acceleration ẍ( t)
equal to:
x( t) = x̂f sin(ω t) =⇒ ẍ( t) = − x̂f ω2 sin(ω t) (3.8)
The amplitude of the force F̂ acting on the following measurement head is
determined by the second law of Newton:
F ( t) = m ẍ( t) =⇒ F̂ = m x̂f ω2 (3.9)
The resulting maximum deformation of the connecting spring equals:
F̂ m x̂f ω2
x̂e = = [m] (3.10)
k k
With equation (3.6) k/m can be replaced by ω20 , resulting in the following
expression:
ω2
x̂e = x̂f (3.11)
ω20
The minimal value of ω0 to achieve a certain precision becomes:
 
x̂f x̂f
ω0 ≥ ω [rad/s], and f0 ≥ f [Hz] (3.12)
x̂e x̂e
This means, that the minimum natural frequency is proportional to the
frequency of excitation and to the square root of the ratio between the
excitation amplitude and the allowable error. In the next section this same
relation is shown in the performance of a feedback controlled positioning
system defined by the maximum tracking frequency, called bandwidth and
the position error ε. It will show, that these values are related to this
natural frequency in passive systems and is a preview to Chapter 4 on
motion control.
92 Chapter 3. Dynamics of motion systems

Radial tracking
Opcal disk
Focus

Hr Hf
kf Opcal
m
Pick-up
kr Unit

Figure 3.3: Schematic view of an Optical disk system with a “virtual” spring, cre-
ated by the position control system that connects the pick-up unit in
six orthogonal coordinate directions (translations and rotations) to the
track. From these directions two translation stiffness values are shown,
the radial stiffness kr and the focal stiffness kf that, together with the
mass of the pick-up unit, determine the tracking errors εr and εf under
the impact of eccentricity of the disk and external vibrations.

3.1.2 Active stiffness

In Chapter 1 the development of the optical disk has been mentioned as

an important driver for mechatronics. It required an actively controlled,
contact-less positioning of the optical pick-up unit. This example is very
appropriate to link precision to the active motion control of Chapter 4. The
basic specifications of such an optical disc system can be derived from the
requirements. In Figure 3.3 a schematic view is shown. For this example
only the radial tracking error (εr ) will be investigated, while the same
reasoning can be followed to determine the required stiffness for the focal
tracking error (εf ) and the other movement directions including the rotations.
For faultless reading of the data, the radial tracking error of the optical pick-
up unit has to be smaller than 0.2 μm and the focus error needs to be smaller
than 1 μm. The disturbances acting on the system are the eccentricity of the
optical disk, that causes a periodic motion of 200 μm in the radial direction
at a frequency of 10 Hz, and shocks that cause random movements of 200 μm
in all directions with the main frequency component at 25 Hz.
As a first step, the control system is approximated as if the optical pick-up
unit was connected with a spring of a certain radial stiffness kr to the track.
Like with the previous example, the required stiffness is calculated, based
on the accelerations that have to be followed. The largest disturbing force is
taken as basis for the calculations. In this case that are the forces, that are
caused by the movements due to the shocks, because they have the highest
3.1. Stiffness 93

frequency with the same amplitude as the eccentricity. The forces due to
the eccentricity can be neglected, because of the quadratic relation between
acceleration amplitude and frequency.
For the calculation it is further assumed, that the movements caused by
the shocks can be approximated by a sinusoidal shape. The maximum
acceleration of these movements is then described by:

ẍmax ( t) = x̂ω2 sin(ω t) = x̂(2π f )2 = 200 · 10−6 (2 · π · 25)2 ≈ 5 [m/s2 ] (3.13)

When the mass of the pick-up unit is 10−2 kg, the amplitude F̂ of the disturb-
ing force becomes 0.05 N and with this number and the required maximum
error εr of 0,2 μm the minimum radial stiffness kr can be calculated:
F̂ 0.05
kr ≥ = = 2.5 · 105 [N/m] (3.14)
εr 0.2 · 10−6
This would result in a natural frequency of:
 
ω0 1 k 1 2.5 · 105
f0 = = = ≈ 800 [Hz] (3.15)
2π 2π m 2π 0.01

When checking this frequency with equation (3.12), the same result is ob-
tained:
 
x̂ 200 · 10− 6
f0 ≥ f = 25 ≈ 800 [Hz] (3.16)
εr 0.2 · 10− 6

This frequency is called the roll-off frequency or bandwidth of the controlled

dynamic system because above this frequency the system is no longer capable
to follow the movements of the track.
In a real optical disk player there is no mechanical contact between the disk
and the lens unit, so there is also no real spring. The stiffness has to be
caused by something else. In a mechatronic system, the controller takes care
of this stiffness. For the CD-player this is a position controller, where the
deviation from the set-point position εr is measured as shown in Figure 3.4.
In the shown example of a feedback controller, the error is translated into a
force F from an actuator, opposing to the deviation (negative sign!) with a
total gain G t equal to the required spring stiffness kr :
F
F = G t εr ⇒ G t = kr = (3.17)
εr
The total proportional loop gain G t consists of the gain of four elements:

• The measurement system, that translates the deviation from the set
point into a voltage signal, Vm = G m εr .
94 Chapter 3. Dynamics of motion systems

Io Vc
Io GpVc Vc  GcVm
Power amplier Controller

x F Vm
ms 2 Hr x
m Hr x
Moving body Vm G mH r
Measurement
Actuator kr
F Ga I o

Figure 3.4: Virtual spring created by negative feedback of a deviation εr of the

position from a set point x0 to a force acting on a body in a servo
system. The equivalent spring stiffness equals the total series gain
k r = G t = G m G a G p G c of the feedback loop.

• The control system, that converts this voltage into a negative feedback
voltage, Vc = −G c Vm .

• The power amplifier that converts this voltage in to a current to the

actuator, I o = G p Vc .

• The actuator, that converts this current into a force, F = G a I o .

The negative feedback causes the force to be opposite to the direction of the
deviation, just like a passive spring opposes a deformation. This means
that a position controller with a constant gain G t , acting on a body, creates
a virtual spring that behaves fully comparable with a spring in a passive
mass-spring system.
As will be shown in the following sections, a simple system with a body and a
spring needs damping to return to stand-still after excitation by a stimulus,
which is equally true for an active system with only a virtual spring. In
Chapter 4 on motion control it will be shown, how a differentiating action
can act as damper to stabilise a feedback controlled positioning system.
3.2. Mass-spring systems with damping 95

3.2 Mass-spring systems with damping

The mechanics in positioning systems can be modelled as a combination

of mass-spring systems, because all material has mass and any structure
has a limited stiffness. This somewhat trivial statement has however large
implications, as the performance of any positioning system is dominated by
the dynamic response of the applied mechanical structure. Two important
aspects are considered in the following subsections, the compliance3 of a
mechanical system, that describes its dynamic reaction to forces acting on
it and the transmissibility, that describes the reaction to movements of the
supportive structure.

3.2.1 Compliance of dynamic elements

A mass-spring system with damping as shown in Figure 3.5 consists of three

elements, that have a different behaviour in respect to forces acting on it.
This behaviour is called compliance because it gives a value to the ability of
the element to comply, that is to move along with the force. The compliance
is given the variable C .

• The first element, the spring, complies with a force by a proportional

displacement. Its compliance is equal to the inverse of its stiffness k.
C s = 1/k.

• The second element, the damper, complies with a force by a propor-

tional velocity. Its compliance is proportional to the inverse of its
damping coefficient c. Cd ∝ 1/c.

• The third element, the body, complies with a force by a proportional

acceleration. Its compliance is proportional to the inverse of its mass m.
C m ∝ 1/m.

First the compliance model of a damped mass-spring system is determined,

by combining the compliance of the three separate elements in the frequency
domain. This helps to create a feeling for the real physics that determine
these dynamics. On this base the equations of motion are derived, that show
the real response of these damped mass-spring systems.
3 The term “admittance” is also sometimes used for this property. The admittance is more

known as the inverse of the impedance in electricity.

96 Chapter 3. Dynamics of motion systems

m x
x0
c F k

Staonary reference

Figure 3.5: A damped mass-spring system with an external force stimulus. The
reaction x on the force F is determined by the combined compliance of
the body (Cm ∝ 1/m), the spring (Cs ∝ 1/k) and the damper (Cd ∝ 1/c).

3.2.2 Combining dynamic elements

In the mass-spring system with damping from Figure 3.5 the applied force
is distributed over the three elements while these share the same position x
referenced to the position x0 , where the system is at rest when F = 0. Starting
with a spring, its compliance Cs is a simple expression assuming a position
independent spring constant:
x 1
Cs = = [m/N] (3.18)
F k
Note, that the sign is positive, as the displacement is in the same direction
as the externally applied force conform the Hooke – Newton law. The
compliance of a spring is a constant factor, so it is independent of time
and frequency. This means, that an amplitude Bode-plot would show a
straight horizontal line with magnitude Cs . This is called the spring-line. In
the phase plot the spring-line shows a phase of 0◦ , because the compliance
equation of the spring does not contain any (s) term in case one would
perform a Laplace transform on this constant.
For the damper the force equals in the time domain:
dx
F ( t) = c (3.19)
dt
After Laplace transform to bring the function to the frequency domain:

F ( s) = scx =⇒ F (ω) = jcω x (3.20)

This results in the following expression for the compliance of the damper as
function of the radial frequency (ω):
x 1
C d (ω ) = = [m/N] (3.21)
F jcω
3.2. Mass-spring systems with damping 97

Total force
x
Ft (Z )
Body Ct Spring
x
Fm (Z ) mZ 2 x x Fs
x
kx
Cm Cs

Damper
x
Fd (Z ) jcZ x
Cd

Figure 3.6: In a mass-spring-damper system the force is divided over the three
elements according to the inverse of their compliance while sharing the
same position.

For the magnitude, the absolute value of the compliance is taken, which
means, that the damper has a compliance, inversely proportional to the
frequency. In a magnitude Bode-plot this would show up as a straight
line with a down slope of −1, the damper-line. This also corresponds with
what is understandable by pure reasoning, because the force of a damper
is proportional to the velocity and with a harmonic movement x = A sin ω t
the related differentiation of the position introduces the additional ω term
in the amplitude. This means, that with an increase of the frequency the
velocity increases if the position amplitude is kept constant. This also means,
that a proportional higher force is needed at higher frequencies to move
the damper. Special attention should also be given to the j term in the
denominator, which means to say, that the compliance of the damper has
an imaginary value and as a consequence the damper-line in the phase plot
shows a phase shift of −90◦ .
The third element, the body reacts on a force according to the second law of
Newton in the time domain:
d2 x
F ( t) = m (3.22)
d t2
After Laplace transform this equation becomes in the frequency domain:

F ( s) = ms2 x =⇒ F (ω) = − mω2 x (3.23)

This gives the following expression for the compliance of a body as function
of the radial frequency (ω):
x 1
C m (ω) = =− [m/N] (3.24)
F m ω2
98 Chapter 3. Dynamics of motion systems

Table 3.1: Overview of the dynamic properties of body, spring and damper in the
time and frequency domain. The Laplace transform is indicated by a
 sign and the variable terms (t), (s) and (ω) are omitted for reasons of
simplification.

Item Spring Damper Body

Unit k [N/m] c [Ns/m] m [kg]

dx d2 x
External force Fs = kx Fd = c  csx = jcω x Fm = m  ms2 x = − mω2 x
dt dt2

x 1 x 1 1 x 1 1
Compliance Cs = = Cd = = = Cm = = =−
Fs k Fd cs jcω Fm ms2 m ω2
[m/N]

1 ◦ 1 1
Magnitude [m/N], ,0 , −90◦ , −180◦
k cω m ω2
Phase angle [◦ ]

The compliance of the body appears to be proportional to the inverse of the

frequency squared. In the amplitude Bode-plot this is represented by a
straight line with a slope of −2. This line should have been called the Body
line but, like with the naming of the mass-spring system this line is called
the mass-line. The −2 slope is also understandable by reasoning because
with a harmonic movement x = A sin ω t acceleration introduces a term ω2 in
the amplitude of the acceleration, due to the double derivative over time.
The minus sign is the result of j 2 = −1 in the denominator, which means,
that the phase plot of the mass-line shows a phase shift of −180◦ .
In Table 3.1 these relations are put together as a reference. When the
three elements are combined, such that they share the same position, the
total excitation force will be divided over the three elements, as shown
schematically in Figure 3.6. At first sight it is to be expected, that at any
frequency the total compliance Ct = x/F can never exceed the level of the
compliance of each of the elements. Although this assumption is not fully
correct, as will be shown later, it is useful to start with, in order to get a feel
for the relations.
With this assumption the following can be stated:
1 1 1 x
Ft = Fs + Fd + Fm = x( + + )= (3.25)
Cs Cd Cm Ct
3.2. Mass-spring systems with damping 99

1 1 k
Ÿ Z0
k mZ02 m
Log |x/F|
x 1 Mass- line (-2)
Spring-line Cs
Fs k Damper- line (-1) x 1
Cm
x 1 Fm mZ 2
Cd
Fd cZ
0q
Spring-line  0q 
Phase

Damper- line  90q 

Mass- line  180q 

180q
‰0 Log ‰

Figure 3.7: The response of body, damper and spring, each represented as separate
elements by a straight line in a Bode-plot with amplitude and phase.
The red dashed line shows the combined compliance plot. The natural
frequency ω0 where the response will roll-off is found at the intersection
of the spring and mass magnitude lines.

Which means, that the total compliance equals:

x 1
Ct = = [m/N] (3.26)
Ft 1 1 1
+ +
Cs Cd Cm

This corresponds with the statement, that at any frequency the element
with the least compliance determines the total compliance.
To illustrate the individual effect on the Bode-plot of each element, each
response is drawn separately in the amplitude and phase part of a Bode-plot
as shown in Figure 3.7. In this Plot the frequency is noted in the angular
frequency (ω), because of the clear relationship with the mathematical
expression for the compliance terms for each element.
The magnitude of the compliance of the combined mass-spring-damper
system can now be derived by taking the lowest value of the individual
elements depending on the frequency region. At low frequencies, the spring
determines the response of the system with the horizontal spring-line at
a compliance magnitude-level of 1/k. At high frequencies the behaviour is
fully determined by the body, according to its mass-line with a magnitude
level of 1/mω2 . The roll-off starts at the intersection of these lines at the
natural frequency ω0 . This can be concluded from the following reasoning.
100 Chapter 3. Dynamics of motion systems

Log |x/F|
Log |x/F|

0q 0q

Phase
Phase

180q 180q
‰0 ‰0
Log ‰ Log ‰
a: No damping b: Low damping
Log |x/F|

Log |x/F|
0q 0q

Phase
Phase

180q 180q
‰1 ‰0 ‰2 ‰1 ‰0
Log ‰ Log ‰
c: High damping d: Extreme damping; first order system

Figure 3.8: The Bode-plots of the response of a mass-spring system with different
levels of damping shows the effect on the resonance at the natural
frequency ω0 .

At intersection both magnitudes are equal, so:


1 1 2 k k
= =⇒ ω = =⇒ ω= = ω0 (3.27)
k m ω2 m m

In the mid frequency range, the damper determines the behaviour of the
system. Depending on the level of the magnitude (1/cω) of its corresponding
damper-line, the phase shows a gradual combination of the phase shift of
the different elements.
To get an initial idea about the effects of different levels of damping four
examples are shown in Figure 3.8.
Bode plot 3.8.a shows the response of mass-spring system without damping.
This is the situation where the assumption on the upper limitation of the com-
pliance is no longer true. At the natural frequency ω0 the amplitude of the
system increases by absorbing energy and loading it in alternating potential
energy (spring) or kinetic energy (mass), resulting in an infinite compliance.
At this natural frequency the phase jumps from 0◦ to −180◦ without gradual
transition. In the next section this effect will be better explained with the
equations of motion.
3.2. Mass-spring systems with damping 101

As soon as a small damper is added, as shown in Bode plot 3.8.b, the

peak in the compliance at ω0 is suppressed by the damper that absorbs
energy out of the dynamic system. Because the damping is still moderately
strong, the original infinite response is now only lowered to a finite value.
When a damper with a higher damping coefficient is applied, as shown in
Bode plot 3.8.c, the damping-line shifts further down in the amplitude plot.
If the damping is high enough, such that the damping-line is below the
intersection of the spring- and mass-line, the amplitude at the resonance
frequency is determined purely by the damping and the resonance peak
will not occur. This situation corresponds with a series of two low-pass
first-order systems, the first starting at a roll-off frequency ω1 < ω0 and
the second starting at a roll-off frequency ω2 > ω0 . If the damping is even
further increased, the mass-spring system is “over-”damped as shown in
Bode plot 3.8.d. The damper-line in the amplitude plot shifts almost com-
pletely below the mass-line. In this situation the dynamic response can be
approximated by a first-order system with a roll-off frequency of ω1 . Above
this frequency the damper dominates the system behaviour over the mass
while ω2 becomes a very high frequency where the magnitude of the response
is very low.

3.2.3 Transfer functions of the compliance

In this section the modelled frequency response of the damped mass-spring

system will be described in the form of a frequency dependent transfer
function (TF) with the Laplace variable s = j ω as variable. While transfer
functions work both in the time- and frequency domain, in the latter case
they are also often called the frequency response function (FRF) of the dy-
namic system. In this book the general term ”transfer function“ will be used
for both domains. These functions are derived from the reaction of a body
to a force with equations of motion that describe the position, velocity and
acceleration of the body. First the compliance behaviour will be derived, as
was schematically presented in the previous section. Although the angular
frequency is used in the mathematics of the transfer functions, the temporal
frequency will be shown in the graphical representations, because of its
clear relation with the engineering field of use. For a mechatronic designer
it is important to recognise and be able to work with this “dual” expression
with a 2π difference.
102 Chapter 3. Dynamics of motion systems

3.2.3.1 Damped mass-spring system.

To mathematically model the compliance of a generic mass-spring system

with a damper, the same configuration is used, as shown in Figure 3.5 in
the previous section. The system compliance equals the movement ( x) of the
position relative to the position x0 , where the system is at rest, in response
to a force (F ), exerted between the body and the stationary reference. The
object is guided with freedom to move in only one direction by a linear
guiding system, represented by the rollers. In the x direction, where the
force is acting, the spring with stiffness k and the damper with damping
coefficient c impact the motion of the body with mass m by their respective
compliances as described before.
The mathematical analysis begins by taking the balance of forces acting on
the body in the time domain:

d2 x dx
F ( t) = m +c + kx (3.28)
d t2 dt

In writing down this second-order differential equation it is necessary to

carefully look at the sign of the terms. With an external force in the upward
(positive) direction, the acceleration, velocity and position will all work in
that same direction.
Using the Laplace transform this differential equation can be written in the
frequency domain as:

F ( s) = ms2 x + csx + kx = x( ms2 + cs + k) (3.29)

From this equation the following relation for the total compliance Ct (s) can
be derived:
x 1
C t ( s) = =
F ms2 + cs + k
1
= m
k [m/N] (3.30)
2 cs
s + +1
k k

To qualify this expression, three entities are introduced, the damping ratio ζ,
the spring compliance Cs and the un-damped4 natural frequency ω0 . Of
these the damping ratio is a new term. It is a dimensionless number as
4 The un-damped natural frequency is mostly named just the natural frequency, which is

not completely correct as at higher damping values the peak in the compliance response will
shift slightly to a lower frequency as will be presented a bit further on in this chapter.
3.2. Mass-spring systems with damping 103

it gives the ratio between the actual damping coefficient c and the critical
damping coefficient c 0 defined by the level where the step response will
not show overshoot anymore, as will be shown in Figure 3.12 in the next
section. This is the situation when this second-order mass-spring system
just behaves like a combination of two equal first-order systems. It will be

shown later, that this critical damping coefficient equals 2 km.
To summarize the derived variables:
c
ζ=  (3.31)
2 km
1
Cs =
k

k
ω0 =
m
Using these terms in Equation (3.30) gives the following:
x Cs
C t ( s) = = 2 [m/N] (3.32)
F s s
+ 2ζ +1
ω20 ω0

To determine the amplitude of the signal the Laplace variable s is substituted

by j ω to give the equation as function of the radial frequency (ω):
x Cs Cs
C t (ω) = = 2 2 = [m/N] (3.33)
F j ω ω ω2 ω
+ j 2ζ +1 1− + j 2ζ
ω20 ω0 ω2 ω
  0   0
Real Imaginary

The amplitude is the absolute value of the vector, being the square root of
the sum of the squared vales of the real part and the imaginary part, as
shown in the Nyquist plot of Figure 3.9. This means, that the amplitude
equals:
x Cs
 
|C t | (ω) =   =  (3.34)
F  2 
 2 ω 2
 1− ω + 2ζ
ω20 ω0

As a first conclusion, in case of no damping (ζ = 0), the response becomes

infinite, when ω equals ω0 . This is the reason, that the frequency f 0 = ω0/2π
is called the natural or resonance frequency of the mass-spring system.
In presence of damping the maximum value for Ct at this frequency relative
to the spring-line becomes:
|C t |max 1
= (3.35)
Cs 2ζ
104 Chapter 3. Dynamics of motion systems

O Real axis

Phase angle φ ()

Imaginary axis
x
Ct
F

Figure 3.9: Polar (Nyquist) representation of the response of a damped mass-spring

system at one frequency.

In Chapter 2 it was shown, that the Laplace variable s introduces a phase

shift of +90◦ , when located in the numerator of the transfer function. Located
in the denominator of the transfer function the Laplace variable s introduces
a phase shift of −90◦ . This means, that the phase angle of any dynamic
transfer function equals the phase angle of the numerator part minus the
phase angle of the denominator part. While the numerator and denominator
generally consist of both real and imaginary terms the phase angle φ of both
numerator and denominator is determined by the arctangent of the ratio of
their respective imaginary and real terms where the quadrant of the angle
is given by the signs of the terms. For a positive real and imaginary value
the angle equals:

Imaginary
Phase = φ = arctan (3.36)
Real

The angle for a negative real value and a positive imaginary value equals
180◦ minus the above calculated arctangent value. A negative value for both
gives an angle of 180◦ plus the above calculated arctangent value while a
positive value for the real term and negative value for the imaginary term
gives a phase angle of 360◦ minus the calculated arctangent value.
For this damped mass-spring system the numerator of Equation (3.32) is
constant without imaginary part. As a consequence the phase angle of the
numerator is zero which means that the phase angle is only determined by
the denominator with a negative sign added to the above calculation. This
all is summarised in the following equation which works in two quadrants
because the imaginary term is always positive and only the real term has
two signs, positive at a frequency below ω0 and negative at a frequency
3.2. Mass-spring systems with damping 105

>^
10 †JQJJ\‹†JJ
†JQJ\‹†J
†JQ^\‹†^
>Z
10 †JQ\‹†
†‹†JQ\
Compliance [m/N]

>
†^‹†JQ^\
10

>\
10

>`
10

>{
10
0
Phase [deg]

>|J

>}J
1 ^ Z
10 10 10
Frequency [Hz]

Figure 3.10: Bode-plot of the compliance of a damped mass-spring system with

k = 1 · 104 N/m and m = 0.025 kg, giving a natural frequency f 0 of
100 Hz, at different values for ζ and Q .

above ω0 :
⎛ ⎞ ⎛ ⎞
ω ω
 ⎜ 2ζ ⎟ ⎜ 2ζ ⎟
0 ⎜ ω0 ⎟ ⎜ ω0 ⎟
φtot = φnum − φden = arctan − arctan ⎜
⎜ 2
⎟ = − arctan ⎜
⎟ ⎜
⎟ (3.37)
⎟
Ct ⎝ ω ⎠ ⎝ ω2 ⎠
1− 2 1− 2
ω0 ω0

To show the amplitude and the phase of this transfer graphically, an example
damped mass-spring system is defined, that will be used throughout most
of this section. The spring has a spring constant k of 1 · 104 N/m and the body
has a mass m of 0.025 kg, giving a natural frequency of:
 
1 k 1 1 · 104
f0 = = = 100 [Hz] (3.38)
2π m 2π 0.025

With the mathematical software MATLAB, Equation (3.32) results in the

Bode-plot of Figure 3.10. Clearly the same shape of the plot is shown as
derived by the graphical method of the previous section. Below ω0 the
graph starts with the horizontal spring-line, corresponding to a value of
C s = 1/ k = 1 · 10−4 . Above ω0 the graph follows the mass-line 1/mω2 with a
slope of −2. The compliance is calculated for different levels of damping.
106 Chapter 3. Dynamics of motion systems

The maximum value of the peak at for instance ζ = 0.25 is not at ω0 , but
at a slightly lower frequency, which behaviour will be explained in the
following section. It is also shown, that Equation (3.35) is correct, as for
instance the resonance peak at ζ = 0.005 has a magnitude of a factor hundred
above the magnitude of the spring-line. This factor hundred is equal to the
variable Q , the quality factor, that equals Q = 1/2ζ. This variable will be
further presented later, after the next section.

3.2.3.2 Critical damping and definition of ζ

The definition of ζ is based on the analysis of the poles of the compliance

transfer function. Poles are those values of s, that result in a zero in the
denominator of the transfer function, corresponding with an infinite value
of the transfer function. This is the most easily understood at the situation
of an un-damped mass-spring system where at s = ± j ω0 the magnitude
of the compliance becomes infinite. Poles can be shown in a geometric
complex plane representation that is called the Laplace plane because of
the relation with the Laplace variable s. The poles of a second-order system
appear as complex conjugate terms, because of the squared relation, and
in the undamped situation the poles are purely located on the imaginary
axis, symmetrical around the real axis. For the damped situation the full
definition of the Laplace variable s = σ + j ω must be used, so including the
real part σ, to determine the pole location. The denominator ( dc ) of the
compliance transfer function is a second-order differential equation with
variable s. To determine the poles this equation can be written in a more
generalised form as a multiplication of two terms:

d c ( s) = ( s − p 1 )( s − p 2 ) (3.39)

where p 1 and p 2 are the two poles. Generally these poles are complex
numbers and come as two conjugate complex terms, written as:

p 1 = −σ + j ωd and p 2 = −σ − j ωd (3.40)

where σ is the real part of the pole and the imaginary part contains ωd ,
the real resonance frequency in the damped situation. When these are
combined in the generalised description of the denominator, it becomes:

d c ( s) = ( s + σ − j ωd )( s + σ + j ωd ) = ( s + σ)2 + ω2d (3.41)

3.2. Mass-spring systems with damping 107

As a next step, equation (3.30) is written in a different form, in order to come

to a denominator notation, that can be compared with this pole notation.
1
x k
C t ( s) = = m cs (3.42)
F 2
s + +1
k k
#
Introducing the natural frequency ω0 = k/m and the spring compliance C s
results in:
x Cs
C t ( s) = = 2 (3.43)
F s cs
+ +1
ω20 k

Multiplication of both the numerator and the denominator of the equation

with ω20 gives:

x C s ω20
C t ( s) = = (3.44)
F csω20
s2 + + ω20
k
For the pole locations only the denominator part is relevant:
csω20
d c ( s) = s2 + + ω20 (3.45)
k
determine the poles the mid term is changed by replacing one factor ω0
To #
by k/m:
$
k
cs ω0
m csω0
d c ( s) = s2 + + ω20 = s2 +  + ω20 (3.46)
k km
When the middle and the last term of this equation are compared with the
corresponding terms in the expanded version of equation (3.41):

d c ( s) = ( s + σ)2 + ω2d = s2 + 2σ s + σ2 + ω2d (3.47)

then the following equations are obtained that relate the real and imaginary
value of the pole to ω0 :
c ω0
σ=  and ω20 = σ2 + ω2d (3.48)
2 km
The second relation is a circle equation that can be written as follows to
obtain the imaginary term ωd of the pole:

% 
 σ2
ωd = ω0 − σ = ω0 1 − 2
2 2 (3.49)
ω0
108 Chapter 3. Dynamics of motion systems

Im
]< 1

]= 1
]> 1

]> 1 Re

]< 1

Figure 3.11: Pole locations of a damped mass-spring system in the complex plane.
Depending on the damping ratio the poles are either real numbers,
that represent an overly damped system consisting of two first-order
systems, or complex conjugate numbers with less damping. The mag-
nitude of the real part relative to the imaginary part determines the
amount of damping in the system.

With Equation 3.48 for the relation of σ this results in:


c2
ωd = ω0 1− (3.50)
4 km
2
In the situation where the term c /4 km becomes equal to one, the poles
become two equal real negative poles. This means, that the second-order
transfer function of the dynamic system is reduced to a combination of two
equal first-order systems. This situation corresponds with the condition
 
c = 2 km, that is called the critical damping coefficient with ζ = c/(2 km) as
the related critical damping ratio.
When the damping is increased beyond that level the roll-off frequencies of
the two first-order systems become separated as was shown in Section 3.2.2.
The lower roll-off frequency corresponds with the pole that shifts to the right
in the complex plane and the higher roll-off frequency corresponds with
the pole that shifts to the left. The corresponding real poles of these two
first-order systems are equal to −1/τ = −ω0 where τ equals the time constant
of the system and ω0 the roll off frequency.
Figure 3.11 shows graphically the effect of the damping on the pole loca-
tions in the complex plane for a damped mass-spring system, corresponding
to the circle relation of the pole terms in Equation (3.48). The fact, that
damping corresponds with a dominant negative real term will be used also
3.2. Mass-spring systems with damping 109

>
x 10
1.6

Displacement (norm) [m]

 = 0.2
1.4  = 0.4
 = 0.6
1.2
 = 0.8
1  =1
 = 1.2
0.8

0.6

0.4

0.2

0
0 0.005 0.01 0.015 0.02 0.025 0.03
Time [s]

Figure 3.12: Position response of a mass-spring system in the time domain with
k = 1 · 104 N/m and m = 0.025 kg to a step Force of 1 N with different
levels of damping.

in Chapter 4 on motion control where a stable system requires the poles

to be located in the left half of the complex plane. It will also be shown
in that chapter with the presentation of state-space feedback control that
these poles correspond to the eigenvalues of an eigendynamics matrix A that
contains the terms of the transfer function in a vector-matrix notation.

From the above equations it follows that increasing the damping will shift
the real resonance frequency ωd below the un-damped natural frequency ω0 ,
until the real resonance frequency becomes equal to zero at ζ = 1. As a direct
consequence, the overshoot at a step response is zero, as can be seen in
Figure 3.12.
110 Chapter 3. Dynamics of motion systems

3.2.3.3 Quality-factor Q

In electrical engineering the “Quality-factor” Q is defined as a variable for the

measure in which a system resonates. The value of Q is determined by the
maximum value of the peak in the resonant system, relative to the level of the
spring-line. Electrical engineering is a typical frequency domain oriented
discipline and also Q belongs to that domain, because of the emphasis on
the behaviour at the natural frequency. Furthermore, resonance is often
positively valued in electronics for tuning filters, stabilising clocks and other
useful functions. For this reason Q = 1 is defined as the minimum level
where just no resonance occurs anymore.
In mechanical engineering, with its typical time domain orientation, the
damping ratio ζ is more common to use, because of its time response relation
and also because resonance is often seen as a negative effect. When possible,
a well behaved step response is aimed for which is the case at ζ = 1, as
was demonstrated in the previous section. In fact both terms can be used
together depending on the situation. In case of a resonator, a high Q value
is preferred, while a controlled, well damped system demands for a high ζ.
Fortunately there is a very simple and straightforward relation between Q
and ζ:

km 1
Q= = (3.51)
c 2ζ
The direct relation of Q with the resonance peak in the Bode-plot is shown
when combining Equation (3.51) with Equation (3.35):
|C t |max 1
= =Q (3.52)
Cs 2ζ
This result was also shown in Figure 3.10. In the example damped mass-
spring system with spring constant k of 1 · 104 N/m and a body with a mass m
of 0.025 kg, this means, that an excitation force amplitude of 1 N will result
in a 0.1 mm motion amplitude at the spring-line, a 1 mm peak amplitude at
ω0 , when Q = 10, and a 10 mm peak amplitude at ω0 , when Q = 100.

A special situation occurs when Q = ζ = 0.5 ≈ 0.7. In that case neither
the time domain response nor the frequency domain response shows any
periodicity for which reason this damping level is called aperiodic. It is an
optimal situation defined by the shortest response time of a dynamic system
without the occurrence of periodic movements due to the first resonance
frequency of the system.
Q and ζ also have a relation with the energy in the system. In case of a
resonating mechanical system the energy from the actuator is stored pe-
3.2. Mass-spring systems with damping 111

riodically in kinetic energy (mass) and potential energy (spring). When

the supply of energy is interrupted, the resonance will gradually decrease,
because in practice a part of the energy is lost in every cycle due to unavoid-
able damping effects in the material like hysteresis, air friction, and so on.
The mechanism causing this gradual loss of energy is based on the under-
standing, that the damping force and velocity are in phase and opposite to
each other. This means, that power is dissipated by the damper at a rate,
equal to the scalar multiplication of force times the velocity:

dx dx dx dx 2
P ( t) = F = −c = −c = − cv2 (3.53)
dt dt dt dt
This energy is dissipated into heat.
The exact relation of Q with energy is determined by looking at the ratio
between the stored energy in a resonating system and the dissipated or lost
energy per cycle. At the natural frequency ω0 the stored energy (E s ) is equal
to the kinetic energy of the body at its maximum speed (v̂) when the spring
is unloaded:
1
E s ( t) = m v̂2 (3.54)
2
The lost energy per cycle (E l ) at the natural frequency ω0 is calculated by us-
ing Equation 3.53 with the sinusoidal velocity v = vm sin ω0 t and integrating
the power over a time equal to the period (T ), which results in the following:


T
1 2 1 2π
E l ( t) = v̂2 sin2 ω0 t d t = c v̂ T = c v̂2 (3.55)
2 2 ω0
0
#
With ω0 = k/m and using above equations the energy ratio (R e ) equals:

1 
Es m v̂2 m ω0 km
Re = = 2 = = (3.56)
El 1 2 2π 2π c 2π c
c v̂
2 ω0
#
Because Q = 1/2ζ = km/c the following conclusion is valid:
Maximum energy stored
Q = 2πR e = 2π (3.57)
Energy lost per cycle

Note, that the mentioned “maximum energy stored” is the energy that
is present in the system at each specific cycle and this value decreases
continuously as a consequence of the dissipated energy in the damping per
cycle.
112 Chapter 3. Dynamics of motion systems

3.2.3.4 Behaviour around the natural frequency

As a thought experiment, at t = 0, a mass-spring system without damping is

excited at its un-damped natural frequency by means of a sinusoidal force
with a constant amplitude. As of t = 0 a continuous increase of the amplitude
will be observed, depending on the amplitude of the force in relation to the
mass of the body. This effect is fully comparable with a proportional linear
acceleration of a body in response to a constant force.
It was mentioned before, that at the the un-damped natural frequency
of a mass-spring system, the energy is continuously exchanged between
potential energy in the spring and kinetic energy in the body. Due to the
ongoing excitation, the captured amount of energy in the system is growing
constantly, causing ultimately an infinite amplitude of motion.
In practice the excitation frequency is hardly ever completely constant.
Further always some damping is present due to hysteresis and friction.
Finally springs with an infinite linear strain do not exist either. This all
means, that this infinite gain will never occur in reality.
Nevertheless it is extremely important to iterate that a Bode-plot is a sta-
tionary representation, based on the presence of a continuous stimulus at all
frequencies. This means, that if the excitation frequency is not sufficiently
long available at the resonance frequency, the resonating effect will have no
time to develop. It also means, that extremely heavy systems with a very
low resonance frequency, like for example maritime pontoons floating in
the sea, will take a long time before the effect becomes visible, when the
waves are exciting the pontoons at their resonance frequency. Because the
amount of energy per cycle added to the system by the waves is small, the
effect might be only significant or even disastrous after a long period.
This relation with energy is also clear, when observing the phase behaviour.
At the natural frequency the phase between force and position is − 90◦ ,
while also the velocity has a − 90◦ phase difference with the position in
case of sinusoidal movements. This means, that the force is in phase with
the velocity at the natural frequency, thus acting in the direction of the
movement and maximising the work. This leads to the conclusion, that
in a vibrating system the maximum efficiency in energy transfer from the
force of an actuator to a mechanical movement is realised at the natural
frequency.
In Figure 3.13 this phenomenon is illustrated by the Bode-plot of the velocity
response of the example mass-spring system with spring constant k of
1 · 104 N/m and a body with a mass m of 0.025 kg to an external force. This
velocity response transfer function is found simply by differentiation over
3.2. Mass-spring systems with damping 113

1
10

= 0.005 Q = 100

= 0.05 Q = 10

†JQ^\‹†^
0
Velocity response [m/Ns] 10
= 0.5 Q = 1

=1 Q = 0.5

†^‹†JQ^\
>
10

>^
10

>Z
10
Phase [deg]

>|J
1 ^ Z
10 10 10
Frequency [Hz]

Figure 3.13: Bode-plot of the response of the velocity of a damped mass-spring

system with k = 1 · 104 and m = 0.025 kg to a force with different values
for ζ and Q . It clearly shows, that the force is in phase with the velocity
at the natural frequency, achieving the maximum efficiency in energy
transfer.

time of the time domain transfer function of the position. In the frequency
domain this differentiation is equal to the multiplication of the system
compliance Ct (s) with s:
v sx( s) sC s
( s) = = sC t ( s) = 2 (3.58)
F F s s
2
+ 2ζ +1
ω0 ω0

The peak level at for instance Q = 100 can be checked by taking the earlier
found motion amplitude of 10 mm and multiply that value with ω to get
the maximum velocity. At 100 Hz = 628 rad/s this results in a maximum
velocity of 6.28m/s, that corresponds with the level of the modelled peak in
the Bode-plot.
As a closure of this part on damping and energy it is illustrative to draw a
plot of the velocity squared over a small area around the natural frequency.
Because the stored kinetic energy equals E = 0.5mv2 such a plot gives the
relative energy levels at different values of Q . In Figure 3.14 the relative
energy level is shown of the discussed system with f 0 = 100 Hz. The energy
114 Chapter 3. Dynamics of motion systems

4
10 10000
Q = 100
Q = 10
Q=1 9000

3 8000
10

7000

''‚
„
''‚„

6000
2
10

5000

4000
1
10
3000

2000

0
10 1000

75 100 125 75 100 125

€![Hz] €! [Hz]

Figure 3.14: Relative energy level as function of Q at a small band around the
natural frequency of 100 Hz. The narrow peak at high levels of Q is
especially prominent, when the phenomenon is presented on a linear
scale.

at Q = 1 is taken as reference level and it clearly shows, that the maximum

energy stored in the system is proportional to the Q level squared, corre-
sponding with the amplitude of the velocity squared. While it is shown
both with logarithmic and linear vertical scales this graphic representation
emphasises the very narrow frequency range, also called bandwidth at high
Q levels, a property, that is used to its full potential in timing devices like
Quartz oscillators, that exhibit Q factors up to a value of 107 .

3.2.4 Transmissibility of a damped mass-spring system

The term “transmissibility” refers to the capability of a system, to transmit

motion from one area to another, both inside a body as well as between
connected bodies. It is related to the elastic wave propagation, that was
shown in Chapter 2 and is relevant in controlled motion systems, where
the path between actuator and sensor always incorporates several flexible
elements and bodies with often only limited damping. This effect on control
will be further presented in the next section, but first the effect of a motion
from a vibrating support, transmitted through the spring and the damper to
3.2. Mass-spring systems with damping 115

m xm

c k
Table xf

Staonary reference

Figure 3.15: The transmissibility of a dynamic system reflects the sensitivity for
external movements of one body, in this case the table, to another
body.

a body, is examined. As application example one can think of the inspection

microscope on a vibrating table of the first part of this chapter. When the
sensitive instrument would would be connected to the table by means of
compliant springs, the vibrations from the table would be attenuated. This
basic vibration isolation principle is frequently applied in vibration-sensitive
instruments.
To calculate the transfer of movements from the table ( xf ) to the body with
mass (m), the simple model of Figure 3.15 is used. The distance xf − xm
between the two bodies is a constant value in the stationary situation without
excitation by floor movements. This constant distance has no influence on
the force equations as only the relative displacements will have an impact.
For that reason a constant value of zero is chosen for the derivation of the
equations of motion.
The total force Ft,m acting on the second body by the spring and the damper
is defined by the difference in position and velocity between the table and the
body and this force induces an acceleration. This force equation is written
in the time domain as follows:
d2 xm d( xf − xm )
Ft,m ( t) = m 2
=c + k ( xf − xm ) (3.59)
dt dt

After the Laplace transform this force equation equals in the frequency
domain:

Ft,m ( s) = ms2 xm = cs( xf − xm ) + k( xf − xm ) = xf ( cs + k) − xm ( cs + k) (3.60)

When shifting the terms the following expression is found:

xm ( ms2 + cs + k) = xf ( cs + k) (3.61)
116 Chapter 3. Dynamics of motion systems

which can be written into the following transfer of xf to xm :

cs
+1
xm ( s ) cs + k k
= = m 2 cs (3.62)
xf ( s ) ms2 + cs + k s + +1
k k
# 
After introduction of the variables ω0 = k/m and ζ = c/(2 km) this transfer
function becomes in the frequency domain as function of (s) and the radial
frequency (ω):
s
2ζ +1
xm ( s ) ω0
= 2 (3.63)
xf ( s ) s s
2
+ 2ζ +1
ω0 ω 0
ω
2 jζ +1
xm ω0
(ω) = (3.64)
xf ω2 ω
− 2 + 2 jζ +1
ω0 ω0

This transmissibility transfer function shows several differences, when

compared with the earlier discussed compliance. First of all the equation
is dimensionless because it represents the effect of one displacement on
another displacement. Furthermore there is no compliance term anymore
related to the spring, and at ω  ω0 the numerator equals the denominator.
This means, that the transmissibility at very low frequencies always starts
at a value of one. The last difference with the compliance transfer function is
the presence of an additional differentiating s term in the numerator that is
related to the damper. This term increases the transmissibility proportional
with the frequency depending on the damping ratio (ζ). Without this term
the transmissibility at higher frequencies would be determined only by the
denominator with a s2 term causing a −2 slope in the magnitude Bode-plot
of the transmissibility. For a sensitive instrument that would be placed on
the body this would be beneficial, as it reduces the transmission of vibration
of higher frequencies to the sensitive instrument. The related s term of the
damper in the numerator however decreases this beneficial effect and this
can be understood from the fact that a stronger damper gives a stronger
connection between the floor and the body.
These effects are all shown in the Bode-plot of Figure 3.16 where a system
is modelled with a natural frequency f 0 of 1 Hz with different damping
settings. Even when the damper is not very strong (the red line with Q = 10)
a rather normal resonance is shown at f 0 , with the −2 slope starting above
the natural frequency. For higher frequencies however, the transmissibility
3.2. Mass-spring systems with damping 117


10
 = 0.005 Q = 100
  = 0.05 Q = 10
10  = 0.25 Q=2
Transmissibility (abs)  = 0.5 Q=1
0
10


10


10


10


10
0
Phase [deg]

>|J

>}J
> 0 1 2 3
10 10 10 10 10
Frequency [Hz]

Figure 3.16: Bode-plot of the transmissibility of a damped mass-spring system in

response to external vibrations from the ground with a natural fre-
quency f 0 of 1 Hz. It clearly shows the negative effect on the reduction
of the higher frequencies.

is already much worse with an increase of about a factor ten at hundred

Hertz, when compared to the situation with Q = 0 (the blue line). Higher
levels of damping ultimately result in a first-order system as can be seen in
the phase plot with a 90◦ phase shift for Q = 1.
A vibration isolation system with low damping is often used in laboratory
equipment, when the natural frequency is sufficiently low to not disturb the
measurements. In case of an industrial application, this is often not suffi-
cient. Especially with optical systems that often have natural frequencies in
the 100 Hz range additional measures are necessary. In Chapter 9 on wafer
scanners, a method will be shown to solve the added transmissibility of the
damper by means of active controlled inertia based damping that does not
contribute negatively to the transmissibility at higher frequencies.
118 Chapter 3. Dynamics of motion systems

3.3 Multi-body dynamics and eigenmodes

The mechanics in real precision positioning systems are far more complex
than the one-dimensional single mass-spring-damper configurations, that
were presented in the previous section. When systems must be modelled
in multi dimensions, all equations need to be written in a six coordinate
direction vector and matrix notation, where also cross couplings between the
different directions can be taken into account. These equations often become
too complicated for straightforward analytical calculations and in practice
they are always done by using computer simulation software like MATLAB
and ANSYS. As explained before, this book focuses on the physical under-
standing of observed phenomena with their basic mathematical modelling,
rather than detailed multidimensional calculations. For this reason the use
of vectors is restricted to the bare minimum. That being said, even in one
direction realistic systems generally consist of a multitude of bodies and
springs and it is very important to be able to qualify an quantify the effects
of these coupled dynamics from a mechatronic design perspective. First of
all, elastically coupled bodies heavily influence the possibility to create a
stable controlled motion system, due to their related resonances. Secondly,
the measurement position in an active positioning system is hardly ever
at the same location as the actuator. For these reason this chapter will be
rounded off by presenting two methods to describe the behaviour of a higher
order mechanical system that consists of more than one body connected by
springs and dampers. The first method builds on the previous analytical
approach with equations of motion, that describe the movement of the bodies.
It is shown how these interact with each other. It also becomes clear, that
the methodology complicates quickly up to the level, that a real analytical
analysis without making errors is not realistic anymore. To cope with this
problem, the second method based on the addition of eigenmodes, with their
corresponding mode-shapes and eigenfrequencies is introduced. This has
proven to be a very powerful method to describe the dynamic properties
of real complex machines in such a way, that it more easily connects with
the design space of mechanical engineers. To illustrate the relation with
measurement even further also an example is shown, that works in two
dimensions.
3.3. Multi-body dynamics and eigenmodes 119

x
x1 x2

k
F m1 m2

Staonary reference

Figure 3.17: Mass-spring system with two bodies with mass m 1 and m 2 connected
by a spring with stiffness k. A force F excites the first body and the
resulting motion is measured at both bodies.

3.3.1 Dynamics of a two body mass-spring system

Mechanical structures can be modelled as a combination of solid bodies

connected with springs and dampers. Generally the passive damping of these
connections is quite limited and only determined by hysteresis properties
of the material or friction in the guiding. For this reason, in the following
the behaviour of the dynamic system will be examined without damping in
order to avoid long equations. The first example, as shown in Figure 3.17,
is a configuration where two bodies are coupled by a spring. The actuation
force is applied to the first body with its mass m 1 coupled to the second body
with mass m 2 by the spring with stiffness k. This example is representative
for a more realistic dynamic model of the optical pick-up unit of the CD
player, where the actuator can be seen as a body with mass m 1 and the lens
as a body with mass m 2 .

3.3.1.1 Analytical description

As a first step towards the analytical description of the system dynamics,

the balance of forces acting on both bodies is determined along the same
reasoning as with the equations of motion of the single body mass-spring
system. For the first body the force balance according to the second law of
Newton equals in the time domain:

d 2 x1 ( t )
m1 = F ( t) − k( x1 − x2 ) (3.65)
d t2
Note, that the force from the spring acts on both bodies in the opposite
direction and the external force only works on the first body. For the second
body the force balance according to the second law of Newton equals in the
120 Chapter 3. Dynamics of motion systems

time domain:
d2 x2 ( t)
m2 = k( x1 − x2 ) (3.66)
d t2
After Laplace transform these differential equations become in the frequency
domain respectively:

m 1 s2 x1 ( s) = F − k ( x1 − x2 ) (3.67)
2
m 2 s x2 ( s) = k ( x1 − x2 ) (3.68)

From these two equations the following transfer functions in the frequency
domain between x1 respectively x2 and F can be derived by first using
Equation (3.68) to write x2 as function of x1 and use that result to solve
Equation (3.67):

x1 m 2 s2 + k
( s) = (3.69)
F m 1 m 2 s4 + k ( m 1 + m 2 ) s2
x2 k
( s) = (3.70)
F m 1 m 2 s4 + k ( m 1 + m 2 ) s2

With these equations some qualitative conclusions can be derived, by looking

at different areas of the frequency spectrum. At very small values of s, the
s2 term in the numerator can be neglected with respect to 1 and in the
denominator the s4 can be neglected with respect to the s2 term. So for low
frequencies both responses become:
x1 x2 1
( s) = ( s) = 2 (3.71)
F F s (m1 + m2 )

This is a standard mass-line with slope of −2 in the Bode-plot. At high

frequencies, so at high values of s the following approximation is valid:
x1 1 x2 k
( s) = ( s) = (3.72)
F m 1 s2 F m 1 m 2 s4

This means, that at these high frequencies the first body will respond
according to the mass-line of only its own mass, while the second body will
respond with a slope of −4 corresponding with its inability to follow the
movement of the first mass. This is called the decoupling of the second body.
The transfer functions are written in a polynomial form, that is common in
control engineering environments using dedicated mathematical software
like MATLAB. To investigate the response at the intermediate frequency
range in an analytical way, it is better to write the equations in a different
form.
3.3. Multi-body dynamics and eigenmodes 121

3.3.1.2 Multiplicative expression

It is possible to re-arrange Equation (3.69) and (3.70) to obtain a form where

a multiplication of factors is used. The multiplicative expression for the
transfer function in the frequency domain x1/F (s) consists of three terms
and equals:
x1 1 1
( s) = ( m 2 s2 + k ) (3.73)
F ( m 1 + m 2 ) s2 Ms2 + k
With:
m1 m2
M=
m1 + m2
In this transfer function one can recognise the first term as being the compli-
ance of the free moving bodies together. For small values of s, corresponding
to a low excitation frequency, the two other terms combined are equal to one
and the system reacts as one free moving body as concluded in the previous
part. The second term (m 2 s2 + k) has a constant value equal to the stiffness
k at low frequencies. At high frequencies this factor increases with the
square of the frequency,
#
a +2 slope. At a frequency where s2 = ( j ω)2 = −k/m 2 ,
hence f = 1/2π k/m 2 Hz this term becomes zero. As a consequence, the total
transfer function will be equal to zero at this frequency. This phenomenon
is called an anti-resonance, although it is no real resonance, as it does not
store energy. The third term is similar to the transfer function of a single
mass-spring system.
Apparently there are three frequencies with a dynamic effect. At two fre-
quencies the transfer goes to infinity. These natural frequencies correspond
with the poles of the system and are called the eigenfrequencies.

1 k
f1 = 0 [Hz], and f2 = [Hz] (3.74)
2π M
The frequency, when the numerator equals zero, determines a zero of this
transfer function represented by the ’anti-resonance’.

1 k
fa = [Hz] (3.75)
2π m2
In a similar way as with the first body one can write the transfer function
in the frequency domain for the second body:
x2 1 k
( s) = (3.76)
F ( m 1 + m 2 ) s2 Ms2 + k
In this equation the same two poles are found, but no zero is found in this
transfer function.
122 Chapter 3. Dynamics of motion systems

3.3.1.3 Effect of different mass ratios

In the previous part several dynamic effects were demonstrated occurring

with two coupled masses. First the vertical shift of the −2 slope to a higher
level from low to high frequencies was shown due to the decoupling of the
second mass while the multiplicative notation showed three frequencies
with either a resonance or an anti-resonance. To illustrate these findings,
the Bode-plots of x1/F and x2/F will be shown for three different situations:

• m 1 < m 2 : This is for instance the situation of a body with a large mass
( m 2 ), actuated by a less heavy actuator (m 1 ), that is connected by
means of a flexible mount.

• m 1 = m 2 : This is the case when the mass of an actuator is equal to

the mass of the positioned object, for instance when the active mass is
optimised relative to the total moving mass of a positioning system.

• m 1 > m 2 , This is representative for the situation, where a large mass

(m 1 ) is positioned with elastically connected smaller masses (m 2 ), that
cause parasitic resonances.

Based on the fact, that M is always smaller than m 2 , the ’anti-resonance’

of the first body will occur at a lower frequency than the resonance at the
second eigenfrequency.
Depending on the ratio between m 1 and m 2 , the relation between these
frequencies will be different, as shown in Figure 3.18, 3.19 and 3.20.
At f = f a the transfer function of x1/F shows an ’anti-resonance’ according
to the zero in the transfer function and all force is directly transferred to
the second body. This is a special situation as one would expect a visible
resonance at the second body at its “own” natural frequency with the entire
spring. This thinking model is however wrong as on the graph of x2/F no
resonance is shown. This can be explained by the fact, that at this frequency
the amplitude x1 of the first body equals zero and no power is transferred
by the force to the second body, which would be necessary to increase the
amplitude.
The amplitude of the movement of the second body is such, that the cor-
responding strain of the spring creates a force that just compensates the
external force acting on the first body. This amplitude is equal to the po-
sition on the undisturbed line of the Bode-plot and even no reaction is
observed in the response of the second body, because at f a the amplitude
of x2/F equals the compliance of spring k (Cs = 1/k). This can be checked in
3.3. Multi-body dynamics and eigenmodes 123

a: Body 1 … ^

>^ >^
10 10

*^ *^
>Z >Z
Compliance [m/N] 10 10

> >
10 10

>\ >\
10 10

11 *
>` >`
10 10

>{ >{
10 10
0 >}J
Phase [deg]

>|J >^{J

>}J >Z`J
1 ^ Z 1 ^ Z
10 10 10 10 10 10
Frequency [Hz] Frequency [Hz]

Figure 3.18: Bode-plot of the response of a dual mass-spring system with m 1 =

2.5 · 10−3 kg, m 2 = 25 · 10−3 kg and k = 1 · 104 N/m. The values result in
a natural frequency of 333 Hz and an ’anti-resonance’ of 100 Hz for
the first body. After the “decoupling” of the second body the initial
compliance slope of −2 of the first body continues with the same slope
but at higher level, corresponding to the ratio between m 1 and m 1 + m 2
(=11). At the second body only the resonance is visible and its initial
compliance slope of −2 becomes −4 above the resonance, indicating the
inability of the second body to follow the movement of the first body.

Figure 3.18.b where the magnitude of the transfer function of the second
body at f a = 100 Hz equals 10−4 , which is equal to the compliance of the
connecting spring.
At f = f 2 both bodies will resonate while the movement of the first body
is 180◦ out of phase with respect to the movement of the second body as
they move in opposite directions. When f > f 2 the slope of the Bode-plot
of the first body will continue at −2, but at a higher level than found by
extrapolating the slope at f  f a . This is caused by the decoupling of the
second body, that does no longer join the movement of the first body. As a
consequence the slope of the response of the second body becomes twice as
steep (−2) as the slope of the first body. The compliance of the first body
124 Chapter 3. Dynamics of motion systems

a: Body 1 … ^

>^ >^
10 10

>Z >Z
10 10
Compliance [m/N]

> >
10 10

>\ >\
10 10

>` >`
10 10

>{ >{
10 10
0 >}J
Phase [deg]

>|J >^{J

>}J >Z`J
1 ^ Z 1 ^ Z
10 10 10 10 10 10
Frequency [Hz] Frequency [Hz]

Figure 3.19: Bode-plot of the response of a dual mass-spring system with bodies
having an equal mass m 1 = m 2 = 25 · 10−3 kg and k = 1 · 104 N/m. These
values result in a natural frequency of 141 Hz where both bodies move
in the opposite direction with equal amplitude. The ’anti-resonance’
measured at the first body is not changed compared with Figure 3.18.

is increased proportional to the ratio of the difference in mass, which for a

ratio of one to eleven means, that the compliance of the first body becomes
a factor eleven larger than was shown in Figure 3.18.a.

For the situation where m 1 = m 2 then f 2 = 2 · f a . In the Bode-plot of this
situation it is shown, that at f 2 both bodies move in counter phase with
the same amplitude. As then the middle of the spring does not move, both
bodies resonate with only one half of the spring. This double stiffness is the

cause of the 2 ratio between f 2 and f a .
For the situation where m 1 is much larger than m 2 , f 2 will approximate the
value of f a . Overall the second body will not have a large influence on the
movement of the first body. Only at the natural frequency it will show up as
a characteristic combination of a zero and a pole.
3.3. Multi-body dynamics and eigenmodes 125

a: Body 1 … ^

>^ >^
10 10

>Z >Z
Compliance [m/N] 10 10

> >
10 10

>\ >\
10 10

>` >`
10 10

>{ >{
10 10
0 >}J
Phase [deg]

>|J >^{J

>}J >Z`J
10 1 10 ^ 10 Z 10 1 10 ^ 10 Z
Frequency [Hz] Frequency [Hz]

Figure 3.20: Bode-plot of the response of a dual mass-spring system with the mass
of body one being ten times larger than the mass of body two. For this
example m 1 = 0.25 kg, m 2 = 25 · 10−3 kg and k = 1 · 104 N/m, resulting
in a natural frequency of 105 Hz, very close to the ’anti-resonance’.
Note, that the overall compliance is decreased due to the increase of
the total mass.

3.3.2 The additive method with eigenmodes

In the field of Structural Dynamics, the transfer functions from Equa-

tion (3.69) and Equation (3.70) are generally written in a different way,
derived from the multiplicative expression. The fourth-order differential
equation that consists of the two coupled second-order differential equations
is solved using an eigenvalue decomposition, that leads to two eigenmodes
with each a natural frequency called the eigenfrequency as presented in
Equation 3.74. Associated with eigenmodes also the mode-shapes are de-
rived, that visualise the related periodic deformations of these eigenmodes.
In Figure 3.21 the example of the previous section is shown, where the first
eigenfrequency f 1 is 0 Hz and the corresponding eigenmode has a mode-
shape represented by the joint motion of the two bodies. The eigenmode
of the other eigenfrequency f 2 has a mode-shape, that is defined by the
movement of both bodies in opposite directions, with an amplitude ratio,
126 Chapter 3. Dynamics of motion systems

k
F m1 m2

Staonary reference
a: Mode shape 1

F m1 m2

Staonary reference
b: Mode shape 2

Figure 3.21: The two eigenmodes with the corresponding mode-shapes of a dual
mass-spring system with a connecting spring. The first eigenmode is
the linear uniform motion in one direction at the first “eigenfrequency”
f 1 = 0 Hz. The second eigenmode is the elastic movement of the two
bodies opposite to each other at the second eigenfrequency f 2 . The
drawn mode direction is only one of the possible directions as the
movement is periodically reciprocating.

that depends on the mass ratio. One can imagine the movements of these
eigenmodes as stable oscillations without external forces to keep them going.
A constant movement will always continue at the absence of external forces
and the same goes for a once excited resonance without damping. It is this
straightforward imagination possibility, that makes their use so very valu-
able in practice. This is even more so because these modes are independent
as long as the system behaves linear. Under that condition their individual
response to a stimulus can be simply added to give to response of the total
system. In practical designs this linearisation is often allowed because of
the small deformations involved in precision mechatronic equipment.
As a last example of the analytic equations of motion, the related two transfer
functions are derived to illustrate this superposition principle. Starting
with a slightly different notation of Equation 3.73 the following equation is
obtained for the position of the first body in the frequency domain:

x1 1 1 m 2 s2 + k
( s) = (3.77)
F ( m 1 + m 2 ) s2 Ms2 + k

To convert this into two terms with 1/s2 , that can be added, as a first step Ms2
3.3. Multi-body dynamics and eigenmodes 127

will both be added and subtracted from the numerator of the last equation:

x1 1 1 ( Ms2 + k) + ( m 2 − M ) s2
( s) =
F ( m 1 + m 2 ) s2 Ms2 + k

1 1 m2 − M
= +
( m 1 + m 2 ) s2 Ms2 + k
⎛ ⎞
m2
1 ⎜1 − 1 ⎟
= ⎜ + M ⎟ (3.78)
(m1 + m2 ) s ⎝ 2 k ⎠
2
s +
M
With M = m 1 m 2/( m
#1 + m 2 ) and the natural angular frequency of the second
mode-shape ω2 = k/M , the total transfer function becomes written as the
addition of the transfer functions of two separate mass-spring systems:
x1 C 1 C2
= 2 + (3.79)
F s s + ω22
2

With:
1 1 1 1 m2
C1 = and C2 = − = (3.80)
m1 + m2 m1 m1 + m2 m1 + m2 m1

The same “transformation” steps, based on Equation 3.76, result in the

analytical transfer functions for the second body in the frequency domain.
x2 1 1 k
( s) =
F ( m 1 + m 2 ) s2 Ms2 + k

2 k
s + − s2
1 1 M
=
( m 1 + m 2 ) s2 k
s2 +
M
C1 C1
= 2 − (3.81)
s s2 + ω22

With this transformation Equation (3.79) and Equation (3.81) appear to

consist of two simple dynamic transfer functions, that are superimposed.
The first term is the typical compliance of a single mass and corresponds with
the first mode-shape. The second term corresponds with a second-order
mass spring system and becomes similar to Equation (3.32) by dividing
both numerator and denominator by ω22 . When looking at the difference
in both equations, it can be concluded, that for the first body, where the
external force is acting, the two transfer functions are added with different
compliance factors. On the other hand for the second body the second mode
is subtracted from the first mode with equal compliance factors. This
difference exists, because in the second mode both masses move in opposite
128 Chapter 3. Dynamics of motion systems

a: Body 1 b: Body 2
 >^
10 10

>Z >Z
10 10
Compliance [m/N]

> >
10 10

>\ >\
10 10

>` >`
10 10
Mode shape 1
Mode shape 2
>{ >{
Total system
10 10
0 0
Phase [deg]

>|J >}J

>}J
1 2 Z
>Z{J 1 2 Z
10 10 10 10 10 10
Frequency [Hz] Frequency [Hz]

Figure 3.22: Bode-plot of the response of the same dual mass-spring system of
Figure 3.18 with m 1 = 2.5 · 10−3 kg, m 2 = 25 · 10−3 kg and k = 1 · 104 N/m,
resulting from the combination of the two mode-shape responses. The
response of the first body (a:) is the result of the addition of both
mode-shape responses, while the response of the second body is the
result of the subtraction of the second mode-shape from the first mode-
shape. The difference between the responses of both bodies in the
spring-line of the second mode-shape is the direct consequence of the
difference in mass. The largest mass m 2 will get the smallest part
of the spring corresponding with a lower compliance. Also the ’anti-
resonance’ appears to be no resonance at all, as it is the result of the
combination of two equal amplitudes with 180◦ phase difference.

directions with an amplitude ratio, that is inversely proportional to the ratio

between the masses of the two bodies.
As a result of this different formulation of the transfer functions, the Bode-
plot of a complex system can be relatively simply be derived by combining
the Bode-plots of the separate eigenmodes. The result is shown in Fig-
ure 3.22, that is identical to Figure 3.18 in respect to the response of the
total system. For the response of the first body, at the frequency, where
both contributions have an equal amplitude, they have an opposite phase,
3.3. Multi-body dynamics and eigenmodes 129

and thus the two contributions will add to zero. This makes clear, that the
term ’anti-resonance’ is badly chosen as it is only the result of two equal
signals in counter phase. Similarly the transfer function x2/F of the second
body can be determined out of the two separate transfer functions. In the
second mode-shape the second body moves in the opposite direction from
the first body, which means, that the response of the second mode-shape
has to be subtracted from the response of the first mode-shape to achieve
the total response. Due to this phase inversion the anti-resonance does not
occur. At higher frequencies beyond f 2 the phase of the second eigenmode
is 180◦ delayed. Because then both components have approximately the
same magnitude and an opposing phase, the combined magnitude goes to
zero, leading to the −4 slope for high frequencies with the corresponding
−360◦ phase relation.

3.3.2.1 Multiple eigenmodes and modal analysis

The example of the previous section consisted only of a simple two body
mass-spring system, without a connection to the stationary world. Generally
the dynamics of a positioning system are more complex. When for instance a
spring is added in Figure 3.17, that connects the first body to the stationary
& '
reference, the transfer function of the first mode-shape becomes C1/ s2 + ω21 ,
similar to the second mode. When measuring the real dynamic transfer
function of a mechanical system, one could obtain a Bode-plot, like shown
in Figure 3.23. The first mode corresponds with the mass of 2.5 kg of
this example system, connected by a spring to the stationary world with
a stiffness k = 104 . At higher frequencies several other resonances are
visible, each corresponding to a different eigenmode, caused by several parts
connected to the main body by supports with a certain stiffness. All of these
eigenmodes add another two orders to the analytical differential equation
and this complicates the analysis considerably. Even without a thorough
higher-order analysis it is often possible to estimate the cause of these
eigenmodes, given the mass of the different parts in the structure. With
this information the mechanical designer can adapt the dynamic behaviour
to the required properties.
Unfortunately, frequently the resonances are due to three dimensional
deformations of the body itself, like shown for instance in Figure 3.24. In
such cases Finite Element Modelling of the dynamics, with software like
ANSYS, is necessary to calculate the eigenfrequencies and mode-shapes of
each eigenmode. Practical verification is done with a method called modal
analysis. In this method the response to a dynamic stimulus is measured
130 Chapter 3. Dynamics of motion systems

>^
10

>
10

Compliance [m/N]
>`
10

>}
10

>J
10

>^
10
0
Phase [deg]

>|J

>}J
0 1 ^ 3
10 10 10 10 10
Frequency [Hz]

Figure 3.23: Bode-plot of a realistic dynamic transfer function of a mechanical

system. The total mass of this example is 2.5 kg and the main body
is connected to the stationary world by a spring with a stiffness k of
104 N/m. The natural frequencies in the graph each correspond with
different eigenmodes, that are determined by parts that are attached
to the main body by means of a mount with a certain stiffness. The
measurement and the actuation are on the same location, resulting in
a phase between 0◦ and −180◦ . Measuring the position at one of the
decoupled masses would cause much more phase delay.

at different locations on the subject by means of accelerometers or other

sensitive sensors for measuring short range displacements. The stimulus
often consists of a force impulse, delivered by a calibrated “hammer”. An
impulse stimulus contains a wide range of frequencies as was presented in
Chapter 2. The different responses are compared with each other and with
the frequency spectrum of the stimulus by means of correlation algorithms,
that derive the related mode-shapes and amplitudes. Presently enhanced
correlation methods even enable this modal analysis without a separate
stimulus, by just using the random external vibrations, that are often
present in actual systems.
3.3. Multi-body dynamics and eigenmodes 131

Figure 3.24: One of several mode-shapes of a solid body as calculated with Finite
Element Modelling software. The red lines indicate the undeformed
shape. Excitation of this torsional eigenmode is often extremely diffi-
cult to avoid by other means than a careful lay out of the actuator and
sensor locations.

3.3.2.2 Location of actuators and sensors

In order to link this chapter with the following chapter on motion control, the
negative effect of eigenmodes on the dynamic performance of a controlled
motion system will be elaborated a little more from the aspect of phase
in relation to the position of actuators and sensors. These positions have
their influence on the observability and controllability of an active controlled
system which refers to the possibility to measure and control the eigenmodes
of a dynamic system.
In the previous sections on the two body mass-spring system it was shown,
that measuring the position at the second body results in a maximum phase
shift of −360◦ . This is far above the maximum of −180◦ , that is required
for stability in a feedback loop, as will be shown in the next chapter. Also
Figure 3.23 might look all right from a phase perspective but as soon as the
measurement system is connected to one of the decoupling masses instead
of directly at the actuator, the resulting phase delay would be significant.
This phase problem by eigenmodes is also observed in Figure 3.24, as the
shown torsional mode represents a movement, that is different for all parts
of the body both in phase and amplitude.
With the more simplified example of Figure 3.25 it will be illustrated, how
these eigenmodes and their corresponding frequency response measurement
are impacted by the location of the actuators and sensors. It emphasises the
132 Chapter 3. Dynamics of motion systems

F F

x x

Staonary reference Staonary reference

a: Mode shape 1 b: Mode shape 2

Figure 3.25: The mode-shapes of the two main, in plane, eigenmodes of a body,
connected to the stationary reference by means of a leaf-spring. The
first mode-shape consists of a swinging movement around a pole close
to the stationary reference and the second mode-shape is a reciprocat-
ing rotation around the centre of mass. Depending on the location of
actuation (F ) and sensing ( x), the observed dynamic behaviour of the
system is different, observed primarily with the second mode-shape.

importance of the right choice of this location in an actual design. This two
dimensional model consists of a rigid body connected to the stationary refer-
ence by means of a mass-less ideal leaf-spring with only two mode-shapes
of interest, a “swinging” movement around a pole close to the connection of
the leaf-spring with the stationary world and a rotation around the centre
of mass of the body. The method to combine the response of eigenmodes
of a single body is identical to the method of the combined eigenmodes of
a multiple body system, because in both cases the mode-shapes of the dif-
ferent eigenmodes are independent. This means, that they can be linearly
combined to give the resulting overall behaviour, as long as the coordinate
system of all mode-shapes is identical. This does not necessarily mean, that
the results are identical and it will be shown, that especially with different
locations of actuator and sensor the response becomes different.
A periodic force stimulus F with frequency f is exerted on the top side of the
body. To illustrate the difference in perceived behaviour, the measurement
of the movement is done at different positions between the bottom side and
the topside of the body.
For the calculation of the eigenfrequencies and the Bode-plot, some values
are chosen as example for the different parameters. The mass (m) of the
body equals 0.05 kg and the bending stiffness (k) of the leaf-spring for the
3.3. Multi-body dynamics and eigenmodes 133

first mode-shape is 5 · 103 N/m. These values result in an eigenfrequency of

50 Hz of the first eigenmode. The eigenfrequency of the second eigenmode is
determined by the moment of inertia around the centre of mass in combina-
tion with the rotation stiffness of the leaf-spring. As only the resulting effect
in the x direction is important, these values can be transformed into an
equivalent mass and stiffness, located at the point of contact of the excitation
force. When the equivalent mass is calculated at the corner of the square, a
value is found of approximately 0.033 kg. It is not without logic, that the
rotational stiffness of the second mode-shape is significantly higher than the
bending stiffness of the leaf-spring for the first mode-shape. For this reason
the equivalent stiffness at the point of exertion of the force is chosen to be
3 · 104 N/m in order to result in a eigenfrequency of approximately 150 Hz.
When the position measurement is done at the point of exertion of the force,
the observed behaviour is shown in the Bode-plot of Figure 3.26.a. The
responses of both eigenmodes are according to the response of a standard
mass-spring system and the combination results in a comparable response
as with a coupled dual mass-spring system, as shown before in Figure 3.22.
The only difference with the previous example is the eigenfrequency of the
first eigenmode, that is 50 Hz instead of zero. The same ’anti-resonance’ is
observed and it is shown, that a single flexible body can behave dynamically
comparable as a multi-body object connected with springs.
A totally different behaviour is however observed when the measurement
takes place at another location. When the motion is measured at the lower
side of the body, the resulting response is shown in Figure 3.26.b. For the
first eigenmode the magnitude of the response is reduced, when compared
with the measurement on top of the body, because the measurement takes
place closer to the mechanical pole of the swinging movement. The largest
and most relevant difference is however found in the second eigenmode,
as the magnitude is equal, but the phase is 180◦ shifted in respect to the
movement at the top. For this reason the combination of these responses
by addition becomes a subtraction. As a consequence at the frequency,
where both mode-shapes have an equal amplitude, the phase of both signals
is -180◦ and gives a higher value instead of the anti-resonance. At higher
frequencies above the second eigenfrequency the slope of this combined
response becomes −4 with a corresponding -360◦ phase. While this would
be comparable with the behaviour of the second body of the coupled dual
body mass-spring system, there is a difference. Instead of only the second
eigenfrequency, both eigenfrequencies are visible, which is due to the fact,
that the phenomenon is acting on one single body.
The conclusion is, that depending on the location of the position sensor,
134 Chapter 3. Dynamics of motion systems

a: Top side sensing …



>^ >^
10 10

>Z >Z
Compliance [m/N] 10 10

> >
10 10

>\ >\
10 10

>` >`
10 10
Mode shape 1 Mode shape 1
 ^  ^
>{
Ž^ >{
>^
10 10
0 0
Phase [deg]

>|J >}J

>}J
1 ^ Z
>Z{J 1 ^ Z
10 10 10 10 10 10
Frequency [Hz] Frequency [Hz]

Figure 3.26: Bode-plot of the response of the two mode-shapes of Figure 3.25. The
left graph (a:) shows the measured behaviour when sensing at the
same location as the force, while the right graph (b:) shows the mea-
sured response when sensing at the bottom of the body. In that situa-
tion the ’anti-resonance’ is disappeared but both resonances remain
present.

the observability of the mode-shape is changed. This is especially the case,

when the measurement sensor is located in the middle, where the move-
ment in the x direction due to the second mode-shape becomes zero. At this
point this eigenmode will not be observed anymore by the sensor. When
this configuration is applied in a closed loop feedback system, the dynamic
effects of the second eigenmode will not interfere with stability, which is a
positive property. Unfortunately this does not mean, that this eigenmode is
not excited, as unknown external forces, due to noise from the actuator or vi-
brations from outside, could contain frequencies around the eigenfrequency
of this not observed mode-shape.
For these reasons, in a real design it is always necessary to prevent these
unknown forces as much as possible, for instance by means of vibration
isolation. It is also necessary to make sure that the actuator is not capable
of exciting an un-observed mode-shape by following the same reasoning as
3.3. Multi-body dynamics and eigenmodes 135

above for the position of the actuator. As long as in the example this position
is in line with the centre of mass, the second eigenmode will not be excited
by the actuator. This optimal position from the point of view of avoiding
resonances in eigenmodes comes at a price in a reduced controllability
of the system. Even if the controller would have information about this
eigenmode it will not be able to control it because it is missing an actuator
at the right location. This can be solved with a separate actuator, only
intended to control this eigenmode, but more often very strict measures are
taken to prevent any other source of interference. This is one of the many
important attention points, when designing complex mechatronic systems
with extreme precision.

3.3.2.3 Summary

In this chapter some important lessons can be learnt, that are summarised
as follows:

• Stiffness, whether it is created mechanically or by means of a control

system, is determinative for precision.

• Every mechanical structure can be modelled as a combination of bodies,

springs and dampers, either as separate bodies or as finite elements
within a body.

• All body-spring combinations determine a mass-spring system with

damping with its related natural frequency and phase delay.

• Phase is a prominent factor regarding the possibility to control the

system.

• Compliance and transmissibility are similar but not equal.

• The quality factor Q and damping ratio ζ are interchangeable. Each

have their practical value.

• A damper is necessary to control the resonance at the natural frequency

but it affects the response for transmissibility at the higher frequencies.

• A complex system can be modelled as a simple combination of its

eigenmodes, each with its own mode-shape, eigenfrequency, damping
and phase behaviour.

• The position of the actuator and the sensor determine the observability
and controllability of the different eigenmodes.
136 Chapter 3. Dynamics of motion systems

• A precision design requires a careful lay-out of all elements, consider-

ing the eigenmodes.

Finally it can be concluded, that these insights help in designing actively

controlled dynamic systems with optimally located actuators and sensors,
that reduce the sensitivity for modal dynamic problems. With this conclusion
this chapter closes to continue with the active elements of a mechatronic
system.
Chapter 4

Motion Control

Introduction

As was presented in the previous chapters, most mechatronic systems are

actively controlled motion systems, which implies that these systems are of
a dynamic nature. If the mechatronic system to be controlled does not show
any dynamics within the required positioning bandwidth, the entire control
problem becomes quasi static and is therefore trivial.
In most motion control systems this is not the case and compensation of the
system dynamics via control is required to achieve the specified performance
in terms of precision, accuracy and frequency response. This chapter dis-
cusses the various approaches to guide and actively control motion systems.
As the name already indicates, motion control is all about the control of
a machine to follow a pre-defined trajectory in space and time, with var-
ious applications. Examples are precision position control with rejection
of disturbances due to vibrations from the environment or imperfections
of the mechanical system as well as path planning and velocity control for
scanning applications.
This chapter is divided in three parts. First the control loop will be examined
to give a global overview. The second part explains in more detail the more
”classical“ control design of feedforward and PID-feedback control with a
strong emphasis on the frequency response and the transfer functions. The
last part will give a short overview of the state-space representation that
is used in modern, model based control designs and allows to model the
system in the time domain..

137
138 Chapter 4. Motion Control

Informaon Energy
Guidance
Output disturbance
signal
Input
Feedforward disturbance Process
disturbance
Output
Pre-lter Feedback D/A Plant
-
Sensor disturbance

A/D sensor

Figure 4.1: Block diagram of a motion control system, including feedforward and
feedback control. The plant consists of the power amplifier, actuator
and the mechanical dynamics. The diagram clearly shows the different
places where interfering disturbances impact the system.

4.1 A walk around the control loop

Figure 4.1 shows a basic control loop of a positioning system where the
control part consists of a feedforward path and a feedback path. Control
systems could in principle consist only of a feedforward or feedback path, but
generally a combination is applied as each principle has its own advantages
and disadvantages, that are separately discussed in the following sections.
For simplicity this chapter is limited to linear dynamics which fortunately
holds for most practical precision motion control systems. In most cases also
a non-linear motion system can be linearised around a chosen operating
point.
The plant is a control engineering term for the uncontrolled physical system.
In a positioning system this consists of the power amplifier, actuator, and
the mechanical structure, as shown in Figure 4.2.
Each of these elements has its own inherent dynamic properties. They inter-
act in both directions in such a way that each element not only determines
the input of the next element but also influences the previous element by
its dynamic load. Often the plant is called just simply the “system”, which
can be confusing when it is not specified, as it can mean the controlled or
uncontrolled system, the control system or whether it includes the sensor
or not.
Also in this book the term ”system“ is more generally used and hence not
precisely defined. It can for instance be the mechanical mass-spring system
4.1. A walk around the control loop 139

Process
disturbance

v Mechanical x
Amplier Actuator
structure

Figure 4.2: In a mechatronic system the plant consist of the amplifier, the actuator
and the mechanical structure. Sometimes also the sensor is included
with the related electronics. The elements that transfer energy interact
with each other in two directions. For instance the amplifier determines
the input of the actuator, while the actuator determines a dynamic
“load”, that influences the behaviour of the amplifier. The same goes for
the actuator and the mechanical structure.

from the previous chapter, the control system or even the full controlled
mechatronic system including the sensor. Its meaning will however be clear
from the context.
As shown in Figure 4.1, the functional blocks that determine the plant with
the related disturbances are represented in red and the functional blocks
forming the control-system are represented in yellow.
It is necessary to clearly distinguish the domain of each functional block,
because everything that is located at the left side from the D/A and A/D-
converters, including the converters themselves, is only involved with in-
formation exchange. This means that these blocks can be interconnected
in serial or parallel configurations, without changing the properties of the
individual block. On the other hand however, the blocks that are shown
in red represent physical systems or sub-systems where energy exchange
is involved. This implies that changes in one sub-system (or block) may
also change the dynamics of the blocks that it is interacting with. A simple
example is a motor that behaves differently whether a mechanical load is
connected to it or not. These aspects of system thinking have to be consid-
ered carefully when interconnecting mechatronic sub-systems, including
the control system.
For analysis and discussion of the control part, first a short section on poles
and zeros will determine the red thread through the chapter, then feedfor-
ward and feedback control will be presented separately.
140 Chapter 4. Motion Control

4.1.1 Poles and zeros

In Section 3.2.3 of the previous chapter the pole of a transfer function was
introduced for those values of the Laplace parameter s = σ ± j ω where the
denominator of the transfer function becomes equal to zero. It was used to
explain the relation between the location of the poles in the Laplace plane
and the damping ratio ζ of the mass-spring system. It was shown that, when
the poles are located in the left-half of the Laplace plane, the ratio between
the real part and the imaginary part will determine the amount of damping
in the mass-spring system.
Many passive mechanical systems, like the damped mass-spring systems
from the previous chapter, are stable in the uncontrolled, open-loop situation.
In dynamic systems stability is defined by the ability of the system to be
insensitive to external stimuli or to return to a defined equilibrium situation
after the application of a stimulus.

4.1.1.1 Controlling unstable mechanical systems

There are also examples of passive mechanical systems that are not stable,
like an inverted pendulum or a piece of iron between two permanent magnets.
These systems can not be stably positioned at any location between two (or
more) extreme positions as they show a negative stiffness. To illustrate the
effect in the frequency domain the transfer function can be derived from the
general compliance transfer function of a standard mass spring system.
The compliance without damping was shown to be equal to:
x 1
C t ( s) = = 2
(4.1)
F ms + k

A spring with a negative stiffness implies that k has a negative value and as a
consequence
# the two poles of this transfer function are both real with a value
of ± k/m. One of these values is positive which means that the location of
this pole is on the real axis in the right-half of the Laplace plane, indicating
an instable pole. The related instability is manifested by a increasingly
accelerated movement away from the zero position as is shown in Figure 4.3.
With feedback control such an unstable system can be stabilised by shifting
the positive pole from the right to the left-half of the Laplace plane. This
effect can be explained with the example of Chapter 3 where proportional
negative feedback is shown to create a spring with a positive stiffness for the
positioning of an optical pick-up unit of a CD player. When this principle of
negative feedback is applied on a mechanical system with a negative stiffness
4.1. A walk around the control loop 141

like the magnetic bearing that will be presented later in this chapter, this
positive stiffness can compensate the negative stiffness when the total loop
gain is sufficiently high. When the positive stiffness by the feedback is even
higher than the negative stiffness of the mechanical system, the resulting
positive stiffness creates a normal mass-spring system with two conjugate
complex poles on the imaginary axis and this transformation implies that
the originally positive pole is shifted towards the left-half of the Laplace
plane by the negative feedback. The addition of damping by differentiating
the feedback signal, as will be explained later, will move the poles even
further to the left and such a magnetic bearing will behave just like any
other well-controlled positioning system.

4.1.1.2 Creating instability by active control

With active dynamic systems the presence of elements that add energy to the
plant like amplifiers and actuators can also induce instability in a normally
stable passive system, as can also be explained with a simple example.
In the previous chapter it was demonstrated that a negative proportional
feedback loop creates a system with the same behaviour as would occur with
a mechanical spring. This means that by only reversing the sign (= 180◦
phase) of this feedback loop, a negative stiffness would be created with the
inherent instable pole in the right-half of the Laplace plane.
In dynamic systems the phase in the loop is a function of the frequency and
in mechanical systems phase lag is present due to the influence of masses
and damping and higher order eigenmodes. Ultimately these effects result
in a large enough phase lag to turn the system into an unstable situation
even with negative feedback. For this reason the analysis of the poles is
of prime importance when dealing with active control of dynamic systems.
Figure 4.3 gives an overview of the impulse response of a dynamic system
for different locations of the poles in the Laplace plane. It clearly underlines
the observed phenomena. The location on the real axis corresponds with
non-oscillating gradually changing responses, either to a stable position for
poles in the left-half-plane or to infinity for poles in the right-half-plane. The
imaginary part of the pole determines the oscillatory behaviour in relation
to the real part. The magnitude of the imaginary part ( j ω) determines the
oscillation frequency and the ratio between the imaginary part and the real
part determines the damping ratio.
142 Chapter 4. Motion Control

Pole-Zero Map
10

4
Imaginary Axis

-2

-4

-6
-4 -3 -2 -1 0 1 2 3 4
Real Axis

Figure 4.3: The impulse response of a dynamic system as function of the pole
locations in the Laplace plane. In the left-half-plane the response
returns asymptotically to a zero deviation value while in the right-half-
plane the response grows exponentially to infinity. For simplicity only
the positive imaginary axis is shown but any pole with an imaginary
part also has a conjugate complex counterpart mirrored around the
real axis.

4.1.1.3 The zeros

The counterpart of the poles are the zeros. They are not detrimental to
stability as they correspond with a value of s where the numerator of the
transfer function becomes equal to zero. As an example, the anti-resonance
of the coupled mass-spring system is a pair of conjugate complex zeros of the
transfer function, as shown in Equation (3.73) in the previous chapter. The
drawback of a zero is the lack of reaction to a stimulus at that frequency.
It can, however, be beneficially used in feedforward control because zeros
can compensate poles as can be seen at the following generalised transfer
function written in pole-zero notation where p represent the poles and z
represent the zeros:
( s − z1 )( s − z2 )( s − z3 )( s − z4 )( s − z5 )....( s − z m )
G ( s) = (4.2)
( s − p 1 )( s − p 2 )( s − p 3 )( s − p 4 )( s − p 5 )....( s − p n )
4.1. A walk around the control loop 143

Input disturbance

r Feedforward u v Plant y
Cff (s) G(s)

Figure 4.4: Block diagram of a feedforward-controlled motion system with one input
and output (SISO). The transfer functions Cff (s) of the feedforward
controller and G(s) of the plant are frequency dependent, which is
denoted by the Laplace variable s.

where m and n are integers with m ≤ n.

An equal pole and zero cancel each other out and this can be used to improve
the dynamic behaviour of the system. Furthermore a single real zero in
the left-half plane gives a phase lead of 90◦ and a +1 slope in the Bode-plot,
corresponding to a differentiating behaviour.
In the following sections the existence and locations of poles and zeros will
be frequently addressed.

4.1.2 Properties of feedforward control

Figure 4.4 shows the typical basic configuration for feedforward control,
which sometimes is also called open-loop control. This example has only one
input and output variable which is in control terms a Single Input Single
Output (SISO) system.
It is important to note that in mechanical engineering a SISO positioning
system is also called a single degree of freedom system, relating to a specific
direction in an orthogonal three dimensional spatial coordinate system.
Unfortunately the term “degree of freedom” can lead to confusion in position
control of mechanical systems, because in control terms, feedforward and
feedback are considered as two degrees of freedom even though it relates to
one direction. For that reason the term is avoided further in this chapter.
The reference or guidance signal1 r is applied to the controller that has a
frequency dependent transfer function Cff (s). The output u of the controller
is connected to the input of the motion system that has a transfer function
1 According to the domain notation that was defined in Chapter 2 the ( t), ( s), (ω) and ( f ) terms,

that define whether the equation is in the time or the frequency domain, are only mentioned
once before the equal sign. One exception to this rule is applied: When transfer functions are
included without expanding them, these functions will also show an indicating term in order
to distinguish them from time invariable functions. The terms are never shown with variables
because these are variable by nature according to the domain that the function describes.
144 Chapter 4. Motion Control

G ( s) giving the output y, which is a position. In this configuration the

feedforward controller acts as a filter that modifies the reference signal in
such a way, that the motion of the controlled mechatronic system follows
the reference signal.
If one would like to achieve perfect control, which means that there is no
difference between the reference position and the actual position of the
system, the combined transfer function G t,ff (s) from r to y has to be equal to
one, hence show identity:
y
G t,ff ( s) = = C ff ( s)G ( s) = 1 (4.3)
r
In that case the feedforward controller has to be the exact inverse of the
plant

C ff ( s) = G ( s)−1 . (4.4)

If no dynamics are involved, the feedforward controller eventually would

only represent a gain that scales the reference signal. In reality positioning
systems include dynamics with a frequency dependent transfer function. In
that case also the dynamics of the positioning system have to be inverted,
which results in pole-zero cancellation between the controller poles and
system zeros as well as controller zeros and system poles.
Feedforward control is a very useful and often necessary first step in the
control of a complex dynamic motion system as it provides the following
advantages:

• No sensor required: No sensor information is fed back to the system

which means that a sensor can be left out, thus reducing the cost of
pure feedforward controlled systems.

• Predictable movement: If the reference signal (trajectory) is known

in advance, the phase lag and time delay in the system can be predicted
and therefore compensated.

• No introduction of instability: The poles of the controlled system

are not changed by feedforward control. Therefore no instability can be
introduced under the trivial condition that the feedforward controller
itself is stable.

• No feeding back of sensor noise: In precision motion systems,

positioning noise is a critical point that always has to be considered
in the control design. The lack of a sensor avoids insertion of the
measurement noise in the system.
4.1. A walk around the control loop 145

The feedforward problem can be more complicated as not always all plant
dynamics can easily be inverted. For example a system with a low pass
characteristic, a property of all positioning systems that are limited by the
mass-line, would require a feedforward controller with high-pass character-
istics, having a very high gain at high frequencies. From this reasoning it is
obvious that feedforward control of motion systems is applicable for a certain
frequency range only, depending on the system dynamics. A second problem
is that an unstable system cannot be stabilised with open-loop control by
cancelling the unstable pole of the system. Further also unstable transfer
zeros, non-minimum phase zeros with a not unique phase to magnitude
relationship, must not be inverted as they would lead to an unstable pole in
the feedforward controller.
This all is included in the following drawbacks and limitations of feedforward
control:

• Limitation to inverted low pass characteristic It is not possible

nor wise to create a controller with a very high gain at very high
frequencies.

• The plant has to be stable: Unstable systems cannot be controlled

with pure feedforward control. The smallest disturbance or noise,
which is always present, would cause the system to become unstable
even if an unstable pole would be cancelled by a corresponding zero.

• No compensation of model uncertainties: Variations in the sys-

tem dynamics, such as shifting of the resonance frequency or variation
of the damping, are not monitored and therefore not accounted for in
feedforward control.

• Can only compensate for known disturbances: Disturbances of

the motion system can only be compensated, if they can be measured.
Even with additional measurement these disturbances can only be
compensated to a certain limit, because for perfection the influence
of the control should be at least as immediate at the systems output
as the disturbance. This implies an infinitely fast reaction which is
practically impossible.

In motion control of mechatronic systems feedforward control is a very

important component mainly because it is faster than pure feedback control.
146 Chapter 4. Motion Control

Output disturbance

Input disturbance Process

disturbance

r Pre-Filter rf e Feedback u v Plant x y

F(s) _ Cfb (s) G(s)

Measurement
M(s)

Sensor disturbance

Figure 4.5: Block diagram of a SISO feedback controlled motion system. Each func-
tional part has its own, mostly frequency dependent transfer function.

4.1.3 Properties of feedback control

In feedback control the actual status of the motion system is monitored

by a sensor and the controller is generating a control action based on the
difference between the desired motion (reference signal) and the actual
system status (sensor signal).
Figure 4.5 shows the block diagram of a standard SISO feedback loop. The
output2 y is measured and compared with (subtracted from) r f which is the
the reference r after filtering. The result of this comparison is used as input
for the feedback controller.
Because the sensor signal is fed back in a closed-loop to the input of the
system, feedback control is also called closed-loop control.
The transfer function of a feedback loop is derived from the following equa-
tions in the frequency domain:
y G ( s)C fb ( s)
e = r f − M ( s) y, y = G ( s)C fb ( s) e, =⇒ T ( s) = = (4.5)
r f 1 + M ( s)G ( s)C fb ( s)
Including the input filter the total transfer function of the feedback loop
from the reference signal r to the output y as shown in Figure 4.5 is given
by:
y G ( s)C fb ( s)
T ( s) = = F ( s) (4.6)
r 1 + M ( s)G ( s)C fb ( s)
2 In many books on control theory and also a bit further in this book the output y is often

assumed to include the measurement sensor as part of the plant and in that case the output
disturbance and sensor disturbance are combined. In real systems with a not ideally linear
sensor it is sometimes better to show the dynamic properties and limitations of the sensor as a
separate transfer function.
4.1. A walk around the control loop 147

In control design one has the freedom to choose F ( s) and particularly Cfb (s)
such that the total transfer function fulfils the desired specifications. In the
pure error-feedback case, like with the positioning of the optical pick-up unit
of the CD player, the reference is constant and no pre-filter is needed. In
the next section, it is shown with this example, how the feedback controller
C fb ( s) can be designed to shape the dynamics of the feedback loop.
Just as the feedforward control approach has its advantages and disadvan-
tages, also feedback control offers some benefits and potential pitfalls:

• Stabilisation of unstable systems: Not all motion systems are in-

herently stable, some of them are marginally stable, some even un-
stable, like an inverted pendulum. As feedback control enables to
determine the place of the closed-loop poles of the controlled system,
unstable poles can be stabilised.

• Reduction of the effect of disturbances: Disturbances of the con-

trolled motion system are observed in the sensor signal, and therefore
the feedback controller can compensate for them.

• Handling of uncertainties: Feedback controlled systems can also

be designed for robustness which means that the stability and perfor-
mance requirements are guaranteed even for parameter variations of
the controlled mechatronic system.

Although feedback control provides some very good features, it has of course
also some pitfalls that have to be dealt with:

• A sensor is required: The feedback loop is closed, based on infor-

mation from a sensor. Therefore feedback control only can be as good
as the quality of the sensor signal allows. In precision positioning
systems accurate sensors are required with high resolution and band-
width, which are very costly. The measurement and sensing system
often takes a substantial part of the total systems budget.

• Limited reaction speed: A feedback controller only reacts on differ-

ences between the reference signal and the measured system status,
which means that the error has to occur first before the controller can
correct for it.

• Feedback of noise: By closing the loop, the positioning noise of the

motion system as well as sensor noise are also fed back, which has to
be considered at the system and control design.
148 Chapter 4. Motion Control

• Can introduce instability: Just as feedback control can stabilise an

unstable system, it can also make a system unstable that even would
be stable without control.

Feedback control is a very useful principle in motion systems as it allows

to stabilise marginally stable or unstable systems. Further it introduces
active damping of resonant modes, and it improves the robustness of the
controlled system.
4.2. Feedforward control 149

4.2 Feedforward control

For systems that are open-loop stable it is possible to apply feedforward

control to improve the system performance when following a predefined
trajectory like a reference signal or a repeating scanning motion. It was
indicated in the previous section that a feedforward controller basically con-
sists of a filter that is placed in series with the plant in order to compensate
its dynamics.

4.2.1 Model based open-loop control

In the following an example of a model-based feedforward controller is

introduced, a scanning unit with a piezoelectric actuator. This unit is applied
as a precision positioning stage with nanometre resolution at a small range
of movement. The measured frequency-response of this scanning unit is
shown in Figure 4.6.
A mathematical model of a second-order mass-spring system is fitted to this
measured response. The first eigenmode shows a weakly damped resonance
at the eigenfrequency of 10 kHz. The following data are obtained by this
system identification exercise. The natural angular frequency of the first
eigenmode ω1 = 2π f 1 = 6.28 · 104 rad/s, ζf as the fitted damping ratio and Cf
as the fitted compliance. With these data Equation (3.32) of the previous
chapter gives the transfer function of this scanning unit:

Cf C f ω21
G ( s) = = . (4.7)
s2 s s2 + 2ζf ω1 s + ω21
2
+ 2ζ f +1
ω1 ω1

When positioning at high speeds, the resonance at the first mode-shape

of this scanning unit causes oscillations that adversely affect the tracking
accuracy.

In order to solve this unfavourable behaviour a feedforward controller is

defined that compensates the dynamics of this scanner by first inverting the
transfer function, without changing the static gain of the positioning system.
This means that the transfer function of the controller ideally would become
equal to:

s2 + 2ζf ω1 s + ω21
C ff ( s) = (4.8)
ω21
150 Chapter 4. Motion Control

>
10

Magnitude (abs)
>`
10

>}
10
>J
10
3 4
10 10
Frequency [Hz]

0
Phase [deg]

>|J

>}J measured data

  
>^{J 3 4
10 10
Frequency [Hz]

Figure 4.6: Bode-plot of a piezoelectric-actuator based scanning unit for nanometre-

resolution positioning. It shows the measured response (solid line)
and the second-order model that is fitted for the low-frequency system
behaviour including the resonance peak, corresponding to the natural
frequency of the first mode-shape (dashed line).

This controller has a pair of zeros, corresponding with an anti-resonance

at the eigenfrequency of the first eigenmode of the scanner, with equal
damping. Unfortunately these zeros also imply an increasing magnitude of
the transfer function controller at higher frequencies with a +2 slope in the
Bode-plot. Such a behaviour is physically impossible so the controller needs
to be be modified in such a way that it becomes realisable. In this case it is
decided to create a resulting overall transfer function of the controller and
the plant that acts like a well damped mass-spring system with the same
natural frequency as the plant and an additional reduction of the excitation
of higher frequency eigenmodes.

In order to realise this controller first two poles have to be added, placed
at the same frequency as the resonance but with a higher damping ratio.
Typically a damping ratio between aperiodic and critical (0.7 < ζ < 1) is
applied to avoid oscillations. For ζ = 1 this results in the following transfer
function:
s2 + 2ζf ω1 s + ω21
C ff ( s) = . (4.9)
s2 + 2 · 1 · ω1 s + ω21
4.2. Feedforward control 151

This feedforward controller is basically a notch filter, because of the notch

shape of the anti-resonance in the Bode-plot.
In order to create an additional attenuation of higher frequency resonances
due to flexible-body mode-shapes, another first-order pole is added at the first
eigenfrequency which gives the transfer function of the complete feedforward
controller:
s2 + 2ζf ω1 s + ω21
C ff ( s) = ( ). (4.10)
( s + ω1 ) s2 + 2ω1 s + ω21

When this controller is connected in series with the scanning unit, the anti-
resonance of the controller and the resonance of the piezo-scanner cancel
each other out. This pole-zero cancellation is the only manipulation on poles
that can be achieved with feedforward control and results in this example
in a well damped third order transfer function:

G t,ff ( s) = G ( s)C ff ( s) (4.11)

((( (
Cf +(
(s (
( 2ζ(
2 2
f ω1 s + ω1 )
= ( ( ) (4.12)
2
+(
(s ( 2ζ( ((( 2 2 2
f ω1 s + ω1 ) ( s + ω1 ) s + 2ω1 s + ω1
(
Cf
= ( 2
) (4.13)
( s + ω1 ) 2
s + 2ω 1 s + ω1

The Bode-plot of the resulting mechatronic system is shown in Figure 4.7. It

demonstrates the compensation of the resonance at the first mode-shape of
the scanner. The controlled mechatronic system has low-pass characteristics
rolling off at the scanner’s first natural frequency.
The beneficial effect of such a feedforward controller on the performance of
this scanner is shown in Figure 4.8. By observing the difference between
the performance of the open-loop controlled and uncontrolled scanning unit
in a triangular scanning motion at 1000 lines per second, it is clear that the
scanner oscillations are suppressed and also the tracking of the triangular
scanning signal is improved significantly.
152 Chapter 4. Motion Control

>
10

Magnitude (abs)
>`
10

>}
10

>J
10
3 4
10 10
Frequency [Hz]

0
Phase [deg]

>|J

  
>}J 
 
>^{J
3 4
10 10
Frequency [Hz]

Figure 4.7: Bode-plot of a feedforward-controlled scanning unit for nanometre-

resolution positioning. It shows both the fitted dynamic model of the
scanning unit without control (solid line), the notch filter by the 3rd -
order feedforward controller (dashed line) and the resulting compen-
sated dynamic performance of the combined scanning unit and con-
troller (dashed-dotted line).

4.2.2 Input-shaping

Another open-loop method, that is often used in motion control, is called

input-shaping. With this method the reference signal is modified in a
different way than by the linear filtering and compensation as shown in the
previous section. As an example of this method, the comparable piezoelectric-
actuator driven scanning unit for nanometre-resolution positioning is used.
In this case however, the system dynamics of the positioning system are
dominated by a resonance frequency occurring at 22 kHz.
When applying a step signal (in simulation) to the scanning unit, it would
start to oscillate at its natural frequency where the oscillations would fade
away after the step according to the damping of this resonance. In a first
approximation, the scanner can be assumed to behave like a linear system
which means that a reduction of the the input step stimulus by a factor of
two would result in a reduction of the amplitude of the response by the same
factor two.
In case two of these steps are applied with only half the height of the full
4.2. Feedforward control 153


‘’
-2

-4

-2

-4

0.0 0.5 1.0 1.5 2.0 2.5 3.0

’

Figure 4.8: Laser-Vibrometer measured scanning motion of the piezoelectric-

actuator driven positioning system in response to a triangular scanning
signal at 1000 lines per second. Without control (upper graph) oscilla-
tions of the scanning unit are clearly observed at the sharp transitions
in the scanning signal. With pole-zero cancellation based open-loop
control (lower graph) these oscillations are suppressed and the tracking
accuracy is improved.

step, the same steady-state response would be obtained as with the full step
stimulus after all oscillations are damped out. If one of these half-height
steps is delayed by half the period of the scanner’s resonance frequency, the
oscillations that are caused by each individual step are 180◦ out of phase,
and cancel each other out.
This splitting of the reference signal into two (equal) signals and delaying
one of them by half the period of the system’s resonance is a typical example
of input-shaping. This method clearly is very different from pole-zero can-
cellation as it is time domain instead of frequency domain based filtering.
In the frequency domain these sampled adaptations to the input create a
frequency spectrum with a multiple of notch filters at the harmonics of the
frequency that these adaptations are applied. This effect is related to real
sampling that will be explained in Chapter 8 on measuring.
With the previous example of the feedforward controller with pole-zero can-
cellation it was shown that a triangular scanning signal to the scanning
unit “triggers” oscillations after each change of slope of the triangular signal.
These are caused by the discontinuity in the derivative and the correspond-
ing higher harmonics of the triangle waveform as presented by Fourier
154 Chapter 4. Motion Control

1 1

Scan signal (norm.)

?
‚Q„
0.5 0.5

0 0

-0.5 -0.5

-1 -1
0 0.3 0.6 0 0.3 0.6

1 1
Scan signal (norm.)

?
‚Q„
0.5 0.5

0 0

-0.5 -0.5

-1 -1
0 0.3 0.6 0 0.3 0.6
Time [ms] Time [ms]

Figure 4.9: Input-shaping control of the triangular scanning signal in a scanning-

probe microscope at 3900 lines per second. Due to the shaping of the
input signal by cutting the top of the triangular signal, oscillations of
the moving part are significantly reduced. The left column shows the
input signals without (top) and with (bottom) input-shaping while the
middle column shows the resulting motion profiles of the scanner. The
pictures on the right show a set of AFM data of a parallel grating that
has a pitch of 233 nm. The distortion and reduction of distortion of the
grating structure is clearly visible.

analysis in Chapter 2.
Applying input-shaping to this triangular scanning signal results in the
introduction of a plateau instead of the sharp peak, where the width of the
plateau corresponds to half the period of the scanner’s resonance as can be
seen in Figure 4.9. When at a positive slope the scanning signal reaches the
plateau and stops raising, the scanner starts to oscillate at its natural fre-
quency, so with a sinusoidal motion. After half the period of this oscillation,
the scanner is arrived at the same position as where the oscillation started,
but it is moving with almost the same speed in the opposite direction, due to
the low damping. When at that moment the scanning signal is changed into
a negative slope, a smooth transition is realised, without further exciting the
scanner’s resonance, and the actual motion follows the desired triangular
scanning motion.
4.2. Feedforward control 155

4.2.3 Adaptive feedforward control

It is important to emphasise that both examples of feedforward control,

the model-based pole-zero cancellation and the input-shaping, only work
reliably as long as the dynamic properties of the total plant are known
and remain constant. These dynamics include the transfer functions of
passive elements like the mechanics as well as active elements like the
amplifiers and actuators. In reality often external influences have an impact
on these dynamic properties, leading to an increasing deviation between
the parameters in the model and the reality. This deviation can be partly
solved by adaptive feedforward control, adapting the feedforward signal
by measuring its real behaviour. This method requires a sensor to obtain
information about the behaviour and for that reason it is often applied in
combination with feedback.
For repetitive processes such as a scanning motion the residual tracking
error can be learned from previous scans and can be used to optimise the
open-loop scanning signal that is applied to the motion system. This ver-
sion of adaptive feedforward control is called iterative learning control. By
measuring the output of a motion system over the known trajectory and
comparing the measured behaviour with the intended behaviour, informa-
tion is gathered about the amount of error in the model and the direction of
the necessary correction.
156 Chapter 4. Motion Control

4.3 PID feedback control

In the previous section it was shown how feedforward control can work
for an open-loop stable system with known properties and circumstances.
Striving for a maximum of predictability of the dynamic behaviour of a
mechatronic system is an important part of the design because of the ben-
efits of feedforward control. In spite of these efforts, even with the most
sophisticated models and by suppressing external disturbances to the bare
minimum, almost always some remaining errors need to be corrected by
feedback control. Next to these systems with remaining random errors,
motion systems with inherent instability, like a magnetic bearing, with
poles in the right-half-plane require feedback control for stabilisationn.
Feedback control is more complex and critical to design than feedforward
control. For that reason it receives generally more attention in the scientific
world and in the last decades several approaches have been introduced to
design an “optimal” feedback control system according to several optimi-
sation criteria. In most cases these innovations were driven by the new
possibilities of fast computer systems that enabled “real-time” modelling
and control of higher-order systems. Still the classical PID-control principle,
where PID stands for a proportional-integral-differential feedback loop is
predominantly used in industry because of its relatively straightforward
mode of operation combined with generally a sufficient performance. Knowl-
edge about the different properties of PID-control is crucial for the design of
high performance mechatronic systems. Without mastering this knowledge
all further refinements in control theory are of no practical value. For this
reason this section is fully devoted to PID-control with a full emphasis on
its representation in the frequency domain.
In spite of this dominance of PID-control in mechatronic systems it is ex-
pected that modern control methods will gradually be used more frequently.
For that reason some of the promising new elements of modern control
engineering will be introduced after this section, based on the state-space
representation in the time domain.
Before explaining the more generic methodology, PD-feedback control, shortly
named PD-control of an optical pick-up unit for a compact disk player is
presented to create an initial idea about the principle. This unit was intro-
duced in Chapter 3 on dynamics. In control terms this is a servo-system.
The word “servo” stems from the Latin word “servus”, which means “slave”
and refers to a control system that is designed to follow a moving target.
4.3. PID feedback control 157

Radial tracking
Optical disk
Focus

Hr Hf
Optical
kf Pick-up m
Unit
kr

Figure 4.10: The optical disk pick-up unit of Chapter 3 connected to the track by
means of a “virtual” spring, created by the position control system.

4.3.1 PD-control of a Compact-Disc player

In the introduction of the optical pick-up unit for a compact disk player the
relation between the control task and the maximum position error of the
lens in respect to the track of the CD was explained. The control system
creates a virtual connection between the track and the optical system as
indicated in Figure 4.10 by a mechanical spring. It was shown that the
required radial stiffness kr amounts to a value of 2.5 · 105 N/m for a moving
mass m = 10−2 kg and a corresponding natural frequency f 0 of 800 Hz. This
stiffness appeared to be equal to the total loop gain G t of the feedback loop
around the moving body that consisted of the combination of the gain of the
measurement system, the controller, the power amplifier and the actuator, as
shown schematically at the left side of Figure 4.11. It was further explained
that a spring and a mass need a damper to control the resonance at its
natural frequency. For this reason in this section also a virtual damper will
be added.
At the right side of Figure 4.11 the equivalent feedback loop for the position
control mechanism is shown. It is a simplified version of Figure 4.5 without
a reference signal and pre-filter. The plant consists of the amplifier, the
actuator, the mechanical moving body and the measurement sensor.
The loop will be traced “step by step” from within the plant. The input of the
mechanics is the force that is generated by the actuator of the positioning
system that holds the lens. The output of the (already perturbed) motion
system is the radial tracking position error εr between the lens position and
the centre of the track. In the equivalent feedback loop this radial error is
denoted by the control error e after the measurement.
158 Chapter 4. Motion Control

Io Vc Feedback e
Io GpVc Vc  GcVm Cfb (s)
Power amplier Controller

F Vm u Output
x
ms 2 Hr x disturbance
Moving body Vm G mH r
Measurement Plant Measurement
Actuator G(s) M(s)
F Ga I o Hr y

Figure 4.11: Mass-spring system with a virtual spring that is created by closed-loop
control. The plant consists of the power amplifier, actuator, moving
body and measurement system. In combination with the controller
the total system can be described by the simple feedback loop as shown
at the right side of the drawing. The disturbance, consisting of the
movement of the track due to eccentricity and external vibrations,
directly interferes with the output of the plant.

The positioning system with the lens is modelled as a single body with mass
m and its corresponding compliance transfer function C m ( s) is equal to:

x 1 1
C m ( s) = = =− (4.14)
F ms2 0.01ω2
The transfer function shows a −2 slope in the Bode-plot and a phase lag of
180◦ over the entire spectrum. The magnitude of the compliance shows a
gain of one at a unity-gain cross-over frequency of ≈ 1, 6 Hz, as shown in the
lower (blue) mass-line of Figure 4.12.

4.3.1.1 Proportional feedback

As a first step, the previously shown stiffness that is created by feedback

will be described in the common terms of control engineering. For this
example the gains of the amplifier, actuator and measurement sensor are
all assumed to be frequency independent constant factors with a value of
one for reason of simplicity. This means that the transfer function G (s) of
the total plant is equal to the compliance of the moving body Cm (s).
It was shown in Chapter 3 that the total loop gain G t , excluding the transfer
function of the body, has to be equal to the required radial stiffness kr =
2.5 · 105 . This means that the controller needs a proportional gain of k p = G t =
4.3. PID feedback control 159

^
10

Magnitude (abs)
1 *^

>^
10

*^
>
10
mass line, open-loop
>`
?>“*
10
0
Phase [deg]

>|J

>}J 0 1 ^ 3 
10 10 10 10 10
€!Xƒ’

Figure 4.12: Open-loop Bode-plot of the CD-player lens with a mass of 10−2 kg. P-
control with a gain kp = 2.5·105 proportionally shifts the total frequency
response of the combined system upwards, such that the unity-gain
cross-over frequency becomes equal to 800 Hz.

2.5 · 105 . The related control term of this proportional gain is proportional
control or P-control. In the loop this means that the feedback control transfer
function Cfb (s) becomes Cp = kp , without the Laplace variable as there is no
frequency dependency of the gain value.
Before the loop is closed, the open-loop Bode-plot of the P-control feedback
system is shown in the upper (black) mass-line of Figure 4.12. The open-
loop transfer function is the total transfer function of the system from the
input of the controller following all functional elements until the measured
sensor signal e, without connecting this signal with the feedback input of the
controller. This open-loop transfer function L p (s) in the frequency domain
is equal to:
kp 2.5 · 105
L p ( s) = G ( s)C p = C m ( s)C p = =⇒ L p (ω) = − . (4.15)
ms2 0.01ω2
P-control shifts the mass-line of the lens-system upwards in the Bode-plot
because it is only a simple gain factor. This shift in magnitude also causes
a shift of the unity-gain cross-over frequency (ωc = 2π f c ) to a higher level,
corresponding with the previously determined required value of the natural3
frequency f 0 = 800 Hz, which was directly related to the required maximum
3 The natural frequency ω = 2π f relates primarily to the first resonance frequency of a
0 0
passive dynamic mass-spring system. Later in this chapter an example is shown with a passive
160 Chapter 4. Motion Control

^
10

*^

Magnitude (abs)
1
10

0
10

>
10
?>“* *^

>^
?>“ *
10
0
Phase [deg]

>|J

>}J
^ 3 4
10 10 10
€!Xƒ’

Figure 4.13: Open-loop and closed-loop Bode-plot of the feedback controlled CD-
player lens with P-control at a gain of kp = 2.5 · 105 . The resulting
undamped resonance at the natural frequency of 800 Hz is clearly
visible.

position error.
When the feedback loop with only P-control is closed, the transfer function
becomes in the frequency domain:
L p ( s) 1 1
T p ( s) = = m =⇒ Tp (ω)  2 . (4.16)
1 + L p ( s) s2 + 1 ω
kp − +1
ωc

with the corresponding Bode-plot of the closed-loop system as shown in

Figure 4.13.
When the open-loop and the closed-loop response are compared, the first
difference is found at the frequency range below the unity-gain cross-over
frequency of the open-loop. The closed-loop response of the system shows a
constant gain of one, corresponding with a spring-line, and its phase shift
has become zero degrees. At the cross-over frequency a sharp resonance
peak is observed, and above this frequency the original −2 slope of the open-
loop response is followed.
When comparing the Bode-plot of the closed-loop system to the passive
dynamic systems discussed in the previous chapter, it is fully reconfirmed
that proportional feedback control creates a spring action.
dynamic system with resonance frequency ω0 , where the unity-gain cross-over frequency ωc is
different from ω0 . For that reason the different terms will be used throughout this chapter.
4.3. PID feedback control 161

The resonance peak is of course not acceptable for the closed-loop operated
system and in control engineering terms this effect is represented by the
location of the poles of the system in the Laplace plane. The closed-loop
poles are a conjugate complex pair on the imaginary axis which means that
the system is marginally stable as the phase-lag of the open-loop transfer
function at the cross-over frequency is 180◦ . In order to achieve an acceptable
control performance, some damping and phase-lead around the unity-gain
cross-over frequency needs to be added.

4.3.1.2 Proportional-differential feedback

In order to add damping to the positioning system of the lens, a force com-
ponent must be created, that is proportional to the velocity. This force can
be created by a signal that corresponds with the derivative of the position
error over time. With this derivative signal the feedback controller transfer
function Cfb (s) becomes equal to the Proportional Differential control or
PD-control transfer function Cpd ( s):

C fb ( s) = C pd ( s) = k p + k d s (4.17)

This PD-control transfer function is also often written in a different form:

kd
C pd ( s) = k p (1 + s) = k p (1 + τd s) (4.18)
kp

In this second notation τd represents the time constant of a first-order

differentiator that increases the magnitude of the transfer function with a
+1 slope above the radial frequency ωd = 1/τd . For radial frequencies below
ωd the derivative term ( s) is smaller than one and the feedback controller
behaves like a P-control system. For frequencies that are higher than this
ratio the derivative term will dominate and the controller shows mainly
differentiating behaviour.
In order to provide the desired damping effect, this change from proportional
characteristic to differentiating action has to take place at a sufficiently lower
frequency than the unity-gain cross-over frequency of the open-loop transfer
function L(s). In a mass-line based positioning system, as a “rule of thumb”
a good performance typically is achieved when the differentiating action
starts at one third of the unity-gain cross-over frequency. The combined
open-loop transfer function of the PD-control with the mass response of the
CD player lens equals:
kp + kd s
L pd ( s) = G ( s)C pd ( s) = . (4.19)
ms2
162 Chapter 4. Motion Control

^
10
*^

Magnitude (abs)
1
10

0
10
*
>
10 ?>“*
?”>“* *^

>^
?”>“ *
10
0
Phase [deg]

>|J

>}J ^ 3 4
10 10 10
€!Xƒ’

Figure 4.14: Bode-plot of the response of the feedback controlled CD player lens
with PD control without limitation of the differentiating action at high
frequencies. The unity-gain cross-over frequency is increased and the
closed-loop response has a −1 slope at high frequencies, which implies
a reduced attenuation of noise and resonances due to high-frequency
mode-shapes.

In Figure 4.14 the response of this transfer function is shown in red with a
−2 slope at the frequency range where k p is dominating. In the range where
the differentiation dominates, a −1 slope is visible in the amplitude-plot
with a corresponding less negative phase in the frequency range of the unity-
gain cross-over frequency. The resulting phase lag is approximately 120◦ as
compared to the original 180◦ . For this reason a differentiating action in the
control system is also often called a lead-network as it introduces a phase
lead in the open-loop response, reducing the original phase lag of the mass
response at the unity-gain cross-over frequency.
Closing the feedback loop for this system results in the following transfer
function of the closed-loop system:
L pd ( s) kp + kd s
Tpd ( s) = = (4.20)
1 + L pd ( s) ms2 + k p + k d s

This equation can be written in the frequency domain in the form as pre-
sented in the previous chapter for passive dynamic systems with the ratio
4.3. PID feedback control 163

ω/ωc as variable:

kd s
s+1 +1
kp ωd
Tpd ( s) = = (4.21)
ms2 k d s2 s
+ s+1 2
+ +1
kp kp ωc ωd
ω
2 jζ + 1
ω
Tpd (ω) =  2 c (4.22)
ω ω
− + 2 jζ +1
ωc ωc
#
with ωc = 2π f c = kp / m and ωd = kp/kd = 1/τd = ωc/2ζ.
This transfer function corresponds with the characteristic transfer function
of the mechanical damped transmissibility of a movement of the support to
a body, connected by a spring and a damper as expressed in Equation 3.64
in the previous chapter. The change of the −2 slope to a −1 slope at higher
frequencies is caused by the zero in the numerator of the transfer function.
This result is not surprising as the optical lens from the CD player is also
following the movement of the track, but in this case it is connected by a
virtual spring and a virtual damper. Likewise with the transmissibility
example, this configuration has a drawback due to the dominant coupling of
the damper at higher frequencies but in active feedback this effect can be
reduced by limiting the virtual damping at these frequencies.

4.3.1.3 Limiting the differentiating action

In a real positioning system many resonating mode-shapes occur at higher

frequencies and it is better to attenuate these to prevent instability. It
is also physically impossible and would not be wise to create a controller
with an infinite gain at infinite frequencies. For the active PD-controlled
system this means that the differentiating action must be limited at higher
frequencies, sufficiently above the unity-gain cross-over frequency in order
to retain the beneficial phase lead.
As a “rule of thumb”, typically a frequency of three times the cross-over
frequency is chosen as the frequency above which the differentiating action
is terminated. This tamed PD-control results in a characteristic behaviour
around the cross-over frequency that is sometimes also called lead-lag com-
pensation because the differentiation introduces a zero which gives a phase
lead and the termination is done by adding a pole which gives a phase lag,
compensating the lead of the differentiator zero. This compensator does not
really induce a lag as after the termination of the differentiation, the phase
164 Chapter 4. Motion Control

1
10

Magnitude (abs)
1 1
ft Zt
+1 2S 2SW t
1
fc Zc
2S
1 1
0 fd Zd
10
2S 2SW d

90
Phase [deg]

45

0
1 2 3 4
10 10 10 10
Frequency [Hz]

Figure 4.15: Bode-plot of limited or tamed D-control. The differentiating action

starts at f d and is terminated at f t . With the “rule of thumb” f d =
0.33 f 0 = 0.1 f t a phase lead of approximately 55◦ is obtained in the
controller.

of the combined controller becomes zero and the gain becomes proportional
again. This is shown in Figure 4.15 where f d = ωd/2π and f t = ωt/2π. It will
be shown later that an additional low-pass filter at f t is in practice even
more preferred in which case the slope becomes −1 and a real roll-off with
phase lag is obtained at higher frequencies.
Another issue of the PD-control action can also be observed in Figure 4.14
as the cross-over frequency is increased significantly due to the magnitude
increase by the differentiating term. With the typical “rule of thumb” value
ωd = 0.33ωc it is necessary to reduce k p with the same factor of 0.33 to achieve
an unchanged cross-over point. With that correction, the total open-loop
transfer function of the system becomes in the frequency domain:
1
s 1+
ωd ωd kp ω d k p 1 + τd s
L pdt ( s) = G ( s)C pdt ( s) = = . (4.23)
ωc ms 2 1 ωc ms2 1 + τt s
1+ s
ωt
with ωt = 1/τt being the radial frequency where the differentiator is “tamed”
at approximately three times ωc according to the “rule of thumb” which is
also approximately ten times ωd . The Bode-plot of the resulting response
with ωd = 0.33ωc = 0.1ωt or f d = 0.33 f 0 = 0.1 f t is shown in Figure 4.16 both
4.3. PID feedback control 165

^
10

*^

Magnitude (abs)
1
10 *^

0
*
10

?>“*
>
10 ?>“ * *^
 ?”>“*
 ?”>“ * *^
>^
10
0
Phase [deg]

>|J

>}J ^ 3 4
10 10 10
€!Xƒ’

Figure 4.16: Open-loop and closed-loop Bode-plot of the feedback controlled CD-
player lens for both P- and tamed PD-control with ωd = 0.33ωc = 0.1ωt .
It shows the reduction of the loop gain at lower frequencies and the
reduction of the attenuation at higher frequencies due to the necessary
phase compensation at the unity-gain cross-over frequency.

in open- and closed-loop, for comparison together with the response with
P-control only.
The unity-gain cross-over frequency of the open-loop transfer function has
become the roll-off frequency of the closed-loop system. The closed-loop
frequency response shows a nicely damped behaviour at the roll-off frequency
with a −2 slope, but at a sacrifice. First the total loop gain at low frequencies
is reduced by a factor 3, corresponding to a proportional reduction of the
virtual stiffness. As a direct consequence also the capability to suppress
errors at those frequencies is reduced. Secondly also at high frequencies
the attenuation of resonances and noise is reduced with a factor 3. It
is important to notice that this is an area of optimisation as ωd and ωt
can be placed closer together to achieve better error reduction at low and
high frequencies with a reduced damping around the unity gain cross-over
frequency ωc .
As another method to solve the reduction of the loop gain at lower frequencies,
an integrating control (I-control) can be introduced eventually forming the
well-known PID-control principle. First some more control theory will be
presented before further introducing this useful principle, because I-control
has no simple comparable element in the mechanical world.
166 Chapter 4. Motion Control

1.8

1.6

''
1.4

1.2

0.8

0.6

0.4

0.2 ?>
?”>
0
0 1 2 3 4 5
>Z
Time [s] x 10

Figure 4.17: Step response of the closed-loop controlled CD-player lens under PD-
control and P-control, clearly showing the beneficial damping effect on
the resonance at the unity-gain cross-over frequency ωc .

To finalise this example the simulated response of the closed-loop system to a

step in the reference signal is illustrated in Figure 4.17, clearly showing that
the PD-controlled system nicely follows the step in the reference signal with
only a small overshoot, whereas the pure P-controlled system continuously
oscillates around the final value due to the absence of damping.
It is worthwhile repeating here, that although simulated step responses may
be a useful tool to evaluate the control performance of a mechatronic system,
special care has to be taken, when eventually testing the controller on the
real hardware. In that situation step or even impulse like reference signals
may result in too aggressive peaks in the control signal and consequently in
the actuation force. Smoother signals like sinusoidal or low-pass filtered
steps and ramp-like reference signals may be more appropriate for testing
of the control performance on the real system. Of course care also has to be
taken when choosing the amplitude of the test signals in order not to get
into trouble with non-linearity or even damage to the hardware.
4.3. PID feedback control 167

Process disturbance Output disturbance

d n

r Pre-Filter rf e Feedback u v Plant x y

F(s) _ Cfb (s) G(s)

Figure 4.18: Simplified representation of a feedback loop in order to determine the

influence of the reference signal and the most important disturbance
sources on the outputs of the plant x, the feedback controller u and the
complete system y. The reference and feedback control input are in
the same position domain as the output of the complete system. The
process disturbance includes all disturbances from within the plant
while the output disturbance includes also the measurement errors.

4.3.2 Sensitivity functions of feedback control

As was demonstrated in the example of the tracking controller for the CD-
player lens, the dynamics of the controlled system can be directly modified
by closing the feedback loop. Feedback control allows to directly place the
system poles at values that are more useful for the operation of the motion
system than their natural locations. This enables a faster response of
the system with adequate damping. This property is in clear contrast to
feedforward control, where only pole-zero cancellation can be applied by
compensation with the inverse transfer function.
One of the other main advantages of feedback control was the possibility to
reduce the effect of disturbances. To illustrate this a simplified version of
the generic feedback loop is shown in Figure 4.18. The measurement system
is assumed to have a unity-gain transfer function. This means that for this
thinking model the reference and the input of the controller work in the
same position domain as the output of the system. The disturbance in the
measurement system is included in the output disturbance n. The process
disturbance can be anything occurring between the output of the feedback
controller and the output of the plant. For reason of simplicity it is inserted
between the plant and the controller. Disturbances from within the plant
can be calculated to this location by using the mathematical model of the
plant. With this simplified model, the transfer functions of the different
inputs of the system to three relevant output variables in the loop are written
down in a set of equations. The input variables include the reference and
the two sources of disturbances. The first output variable is x for the output
of the plant, being a subset of the system-states that will be introduced in
168 Chapter 4. Motion Control

Section 4.4. The second and third output variables are y for the output of
the total system and u for the output of the controller. In these transfer
functions for simplicity the Laplace variable (s) is omitted and C = Cfb :

G GC GCF
x = d− n+ r (4.24)
1 + GC 1 + GC 1 + GC
G 1 GCF
y = d+ n+ r (4.25)
1 + GC 1 + GC 1 + GC
GC C CF
u = − d− n+ r (4.26)
1 + GC 1 + GC 1 + GC

It should be noted that these transfer functions are written in a purely single
dimensional notation with scalar values. When more degrees of freedom are
controlled the different upper-case letters become matrices and the lower-
case letters become vectors. With multiplication of matrices the successive
order is important and the shown order is valid for the defined feedback
loop from Figure 4.18. When for instance disturbance d is neglected, the
vector notation would be x = Gv and v = Ce which gives x = GCe.
Although these transfer functions consist of a combination of nine transfer
functions, one for each input-output combination, some of them are the same.
When these doubles are omitted, six different transfer functions remain as
shown in Table 4.1. These functions are also called the gang of six, a term
introduced by the Swedish control scientist Karl Johan Åström from Lund
University, because this set of transfer functions gives an interesting insight
in how the feedback controlled system reacts to the different system inputs.
The transfer functions of the first column shows the influence of the reference
signal on the output of the plant and on the output of the feedback controller.
The first transfer function of the second column gives the influence of the
output disturbance on the output of the plant. while both transfer functions
in the second column give the influence of the two disturbances on the output
of the controller.
The first transfer function in the third column shows how the output of the
plant reacts to the process disturbance. The second transfer function of the
third column gives the influence of the output disturbance on the output of

Table 4.1: Gang of six.

x y GCF x u GC x y G
= = − =− = = =
r r 1 + GC n d 1 + GC d d 1 + GC

u CF u C y 1
= = =
r 1 + GC n 1 + GC n 1 + GC
4.3. PID feedback control 169

Table 4.2: Gang of four.

x y x u GC x y G
= =− =− = = =
r r n d 1 + GC d d 1 + GC

u u C y 1
= = =
r n 1 + GC n 1 + GC

the total system.

In case no input filter is applied F is equal to one and the first column
becomes equal to the second column, reducing the gang of six to the gang
of four as shown in Table 4.2. This short set of equations also corresponds
with the situation without a reference like the pure error feedback system
of the CD-player.
From these four, the two most important transfer functions are the com-
plementary sensitivity function T ( s), corresponding with the first transfer
function of the first column and the sensitivity function S ( s), corresponding
with the second transfer function of the second column.
Written in full Laplace notation the complementary sensitivity function
equals:
y G ( s)C fb ( s)
T ( s) = = , (4.27)
r 1 + G ( s)C fb ( s)

The complementary sensitivity function represents the system response

to the reference in case F = 1. It gives an indication how well the system
behaves as a servo-system, reliably following a target or a reference.
Written in full Laplace notation the sensitivity function equals:
y 1
S ( s) = = (4.28)
n 1 + G ( s)C fb ( s)

The sensitivity function represents the ability of the feedback controlled

system to reject disturbances.
The name ”complementary” in T (s) is based on the fact that T (s) and S (s)
add to unity, (T (s) + S (s) = 1).
170 Chapter 4. Motion Control

Loop Gain L(s) Closed loop

5
High gain Peak disturbance
1 

10 1 0

Magnitude (dB)
Magnitude (abs)

Magnitude (abs)
0 JQ{
T(s) Bandwidth S(s) -3
10
-5
> <
10
slope -10
>^
10
>Z -15
10 Low gain
10
>
0.1 1 *^J
^ 3 4
10 10 10 10
Low frequency Cross-over High frequency Frequency [Hz]
region region region
>|J

r e u y
Phase [deg]

C(s) G(s)
>}J -
Stability
L(s)

>^{J
1 ^ 3 4
10 10 10 10
Frequency [Hz]

Figure 4.19: Stability condition and robustness of a feedback controlled system.

The desired shape of these curves guide the control design by optimis-
ing the levels and slopes of the amplitude Bode-plot at low and high
frequencies for suppression of the disturbances and of the phase Bode-
plot in the cross-over frequency region. This is called loop shaping
design.

4.3.3 Stability and robustness in feedback control

In the CD player example it was demonstrated that the open-loop transfer

function of the PD-control feedback system needs to have certain properties.
After closing the loop, the controlled system should be stable and show good
performance regarding the accuracy in following the track. Based on these
requirements, certain conditions for stability and robustness to changing
conditions can be derived. These conditions for stability and robustness
also give insight in how the open-loop system should look like such that
the closed-loop system performs well. These insights can be used to select
the controller parameters for properly adjusting the transfer function of
the open-loop system. The optimal tuning of a feedback loop is called loop-
shaping design.
The most important and characteristic frequency area for the analysis of a
controlled mechatronic system is around the unity-gain cross-over frequency,
4.3. PID feedback control 171

as shown in Figure 4.19. In control engineering the term bandwidth is

often used in relation to this unity-gain cross-over frequency as above this
frequency the loop gain becomes smaller than one and consequently the
feedback controller becomes no longer effective. Usually the term bandwidth
is defined as the frequency band where the power of the output signal of a
system becomes less than half the desired power level. In terms of signal

amplitude the corresponding value is equal to 1/ 2 ≈ 0.7. In decibels this
value is equal to −3 dB and this value is a well-known definition for the
bandwidth of filters and other frequency dependent functional devices like
loudspeakers.
In the context of precision positioning systems it is better to define the
bandwidth as the frequency range, where the amplitude of the open-loop
transfer function exceeds a value of one, because it is the open-loop gain at
certain frequencies that is important in mechatronic systems.
The condition for closed-loop stability is that the total phase-lag of the
open-loop system, consisting of the feedback controller in series with the
mechatronic system, must be less than 180◦ in the frequency region of the
cross-over frequency. A system that has exactly 180◦ phase-lag at the cross-
over point is called marginally stable, like for example the CD player with
only P-control. In this situation the smallest additional time-delay or phase-
lag would make the closed-loop system unstable.
To determine the stability one can also analyse the poles of the closed-loop
system, where the stability condition is the same as for any linear system.
For asymptotic stability all system poles have to have a strictly negative
real part, which means that they have to be located in the left-half of the
Laplace plane.
The Nyquist plot, like shown in Figure 4.20, is the best frequency response
representation to analyse the robustness of a feedback system by the distance
and direction of the graph relative to the location of the −1 point on the real
axis. In this graph three values are shown that relate to the robustness of
the closed-loop feedback system, the gain margin, phase margin and vector
margin.

The gain margin determines by which factor the open-loop gain addition-
ally can increase before the closed-loop system goes unstable. It is
defined by the distance between the loop-gain L(s) and unity-gain at
the frequency where the phase-lag of L(s) becomes more negative than
−180◦ . The gain margin can have values between zero and infinite.
With first and second order transfer functions where the phase does
never become more negative than −180◦ the gain can be increased
172 Chapter 4. Motion Control

^ Q†
Q†Q^\
J …
^ …

 –


Q†Q`
 … Q†^ Q†ZQ^
` … J …

~
@•

0
? —
~
 



"€!
>

>^

>Z
>\ > >Z >^ > 0
@•


Figure 4.20: The phase, gain and vector margin can be easily derived in a Nyquist
plot. Stability is guaranteed when the −1 point on the real axis is kept
at the left hand side of the response-line upon passing with increasing
frequency. The dashed circles determine the magnitude peak (Q ) after
closing the loop at the frequencies where the response-line crosses the
circles. The blue response-line represents an example PID-controlled
mass-positioning system where the I-control action together with the
mass creates a phase of more than −180◦ at low frequencies.

theoretically to infinite, corresponding with an infinite gain margin.

The phase margin determines how much additional phase lag at the
unity-gain cross-over frequency is acceptable before the closed-loop
system becomes unstable. It is defined by the difference between the ac-
tual phase-lag of L(s) and -180◦ at the unity-gain cross-over frequency.

The Vector margin is defined by the closest distance in a Nyquist plot

between the graph and the −1-point on the real axis.

The following rules for stability can be derived from the gain- and phase-
margin in relation to the Nyquist plot:

• The system will become unstable upon closing the loop if the phase of
the open-loop transfer function passes the negative real axis (−180◦ ) at
an amplitude larger than one and does not return below the negative
real axis before the gain gets smaller than one.
4.3. PID feedback control 173

• A stable system after closing the loop is recognised in the Nyquist plot
when the −1 point on the real axis is kept at the left hand side upon
passing with increasing frequencies.

It is also interesting to observe in this example plot that the phase at low
frequencies is more than −180◦ shifted. This situation occurs when the
I-control action, that will be presented in the following subsection, is applied
with a purely mass based positioning system like the CD player pick-up unit,
so without a spring to the stationary reference. Even though such a system is
stable in the closed-loop situation, it is not readily known what the amount of
damping will be when the feedback loop is closed. To indicate the magnitude
of the closed-loop frequency response at different frequencies, circles are
shown that determine regions with a different maximum magnitude of the
closed-loop response above unity gain. The smaller the circle that the graph
will intersect, the higher this magnitude will be. These circles do not have
their centre at the −1 point on the real axis but on a shifting point to higher
real numbers for a lower magnitude. These circles are derived from the
magnitude of the complementary sensitivity function in the Nyquist plot
for a certain level of Q :
#
|GC | Re2 + Im2
|T ( s)| = =# =Q (4.29)
|1 + GC | (1 + Re)2 + Im2

With a bit of algebra this results in:

 2  2
Q2 Q
Re + + Im2 = (4.30)
Q2 − 1 Q2 − 1

Which is a circle with its centre on the real axis at a distance D from the
origin and with a diameter d :

Q2 Q
D=− and d= (4.31)
Q2 − 1 Q2 − 1

Modelling software like MATLAB automatically shows these circles and

although unfortunately often with only the dB level mentioned, they are use-
ful to get a first estimation of the expected closed-loop response of the system.
In further examples of the Nyquist plot only the Q level will be mentioned
to adhere to the preferred working with non-logarithmic numbers.
174 Chapter 4. Motion Control

0
10

Magnitude (abs)
>^
10

k x 10
>

100 N m 10
>`
Driven body
m 10
>}

Phase [deg]
0.25 kg 0

>|J

>}J
>^ 0 ^ 
10 10 10 10
Frequency [Hz]

Figure 4.21: The undamped mass-spring system and the Bode-plot of its transfer
function that is used as example to illustrate PID-control with some
real data.

4.3.4 PID-control of a mass-spring system

The previous example of the CD-player introduced P-control and D-control to

create a virtual spring and damper. To create a real PID-controller another
control action is added with its own beneficial effect, integrating control
or I-control. According to control theory a PID-controller is defined by the
following relation between the input error e and the output u of the controller
in the time and frequency domain:

t
d e ( t)
u ( t) = k p e ( t) + k i e(τ) dτ + k d (4.32)
dt
0

ki
u ( s) = e ( s) k p + + k d s =⇒
s

u ( s) ki
C pid ( s) = = kp + + kd s
e ( s) s

To show the effect of PID-control with practical values a PID-controller is

designed for a simple mass-spring system in a comparable setting as the
previous example of the PD-control of a CD player. In this case the body
is connected to the stationary reference with a very compliant spring with
a spring-constant k = 100 N/m in order to show the difference with a pure
inertial mass situation like with the CD player. The mass m of the body
equals 0.25 kg and very little passive damping is present in the system ( c ≈ 0)
which means that the mechanical damping is neglected for this analysis.
With these values and the gain of amplifier, actuator and sensor equal to
4.3. PID feedback control 175

one, the transfer function of the plant becomes as follows:

1
G ( s) = (4.33)
ms2 + k
1
= . (4.34)
0.25 s2 + 100

1
# frequencies the spring-line /k = 0.01 defines the transfer function. At
At low
ω0 = k/m = 20 rad/s the system has a resonance with an infinite high peak
due to the absence of damping. At ω  ω0 the transfer function is dominated
by the mass-line (1/ms2 ) with a −2 slope and a phase lag of 180◦ .
For this system the PID feedback controller will be designed with the fol-
lowing control objectives:

• The control bandwidth as defined by the unity-gain cross-over fre-

quency is 100 Hz.

• The feedback controlled system has to be asymptotically stable, which

means that all the poles are located in the left-half of the Laplace
plane.

• Sufficient damping of the feedback controlled system which implies no

or little oscillations.

• The feedback controlled system must have zero steady-state error in

response to the reference signal as well as disturbing forces acting on
the controlled position of the mass.

The desired unity-gain cross-over frequency of 100 Hz is equivalent to ωc =

628 rad/s, which is at a higher frequency than the resonance frequency of the
mass-spring oscillator. In that region the frequency response of the plant is
dominated by the mass-line (1/ms2 ). For this frequency the magnitude of
the transfer function of the plant G ( s) ≈ 1 · 10−5 .

4.3.4.1 P-control

With a pure P-control gain of kp = 1·105 the unity-gain cross-over frequency ωc

of the transfer function of the plant is raised to the required value of 628 rad/s
(100 Hz). For stability the total phase lag of the open-loop system has to
be smaller than 180◦ at the unity-gain cross-over frequency. For the system
discussed here a pure P-control gain results in a marginally stable system
that shows in closed-loop a sharp resonance peak at the cross-over frequency
with a sharply changing phase shift from 0◦ to −180◦ . To add stability
176 Chapter 4. Motion Control

and robustness to the feedback controlled system a differentiating action

(D-control) is added that generates a phase lead around the unity-gain
cross-over region, adding damping to the oscillatory system.
Necessary reduction of the P-control gain due to D-control: To add
sufficient phase lead (and damping) the differentiating action (D-control)
should start according to the “rule of thumb” at 0.33ωc . This means that
above this frequency the system shows a −1 slope instead of the original
−2 slope of the mass-line. As was shown at the CD player example this
requires that the gain kp is reduced by this factor three in order to retain a
loop-gain of one at 628 rad/s. As a result kp is chosen to be equal to 3.3 · 104 .

4.3.4.2 D-control

The differentiating action (D-control) is used to add phase lead and damping
to the system around the unity-gain cross-over frequency ωc . The differ-
entiating term of the PD-controller (kp + skd ) starts to dominate at ωc/3.
Rewriting this PD-controller as
kd
C pd ( s) = k p (1 + s ), (4.35)
kp

shows that this happens at

kp ωc
=s= . (4.36)
kd 3

This means that

kp 3 · 3.3 · 104
kd = 3· = ≈ 160. (4.37)
ωc 628

The differentiating action is ”tamed”, which means that it is limited to only

the area around 100Hz, in order to provide a steeper roll-off of the controller
at high frequencies and to limit the control effort at those frequencies. Ac-
cording to the “rule of thumb” the frequency above which the differentiation
is terminated should be a factor of 3.3 above the cross-over frequency. There-
fore a pole is added to the D-control gain at ωt = 3.3 · ωc ≈ 2000 rad/s.
As a result the transfer function of the D-control action becomes:
160
kd = (4.38)
5 · 10−4 s + 1
4.3. PID feedback control 177

6
10

Magnitude (abs)
-1
Tamed-D (=P)
I-control
5
+1 D-control
10

P-control

4
10
|J
Phase [deg]

45

>\

>|J
> 0 1 2 3 4
10 10 10 10 10 10
Frequency [Hz]

Figure 4.22: Bode-plot of the PID controller. From left to right first the −1 slope of
the I-control line is followed until it intersects with the horizontal P-
control line at 10 Hz. At the intersection of the P-control line with the
+1 slope of the D-control line at 33 Hz, the magnitude increases again
until the intersection at 330Hz with a second horizontal line, located
a factor 10 above the P-control line, determined by the ratio of the
differentiating and “taming” frequency where the controller transfer
function becomes proportional again. At 100Hz this controller has
the required gain of 1 · 105 with a phase lead of around 55◦ , giving an
equal phase margin for robustness.

4.3.4.3 I-control

An integrating action is added to the system in order to increase the loop-

gain at low frequencies and achieve a zero steady-state error in response to
the reference signal as well as in response to disturbances. In the frequency
domain, integration equals the addition of an s term in the denominator of
the transfer function and as a consequence it adds 90◦ of phase lag to the
system. In order to not affect the beneficial phase lead of the D-control too
much, the integrating action should stop another factor of 3.3 lower than
the starting point of the D-control action, so at ωi ≈ 0.1 · ωc ≈ 60 rad/s. At
this angular frequency the term ki / s has to become equal, and at higher
frequencies eventually smaller, than kp .
178 Chapter 4. Motion Control

4
-1
10

Magnitude (abs)
^
10
*^
-1
0
10

>^ *^
10

>\

>|J
Phase [deg]

>Z\
\\
>}J

>^^\

>^{J
> 0 1 ^ 3 4
10 10 10 10 10 10
Frequency [Hz]

Figure 4.23: Bode-plot of the open-loop transfer function of the total system with
the PID controller in series with the mass-spring system. Even though
the eigendynamics of the mechanical system show a clear resonance
at around 3 Hz the phase lag at the 100 Hz unity-gain cross-over
frequency remains well below 180◦ .

Therefore:

k i = k p · 60 ≈ 1.8 · 106 (4.39)

These controller gains for kp , kd , and ki result in the Bode-plot of the transfer
function of the controller shown in in Figure 4.22. The integrating action
up to 10 Hz and the differentiating action from 33 Hz until 330 Hz with the
corresponding phase behaviour are clearly visible.
Figure 4.23 shows the Bode-plot of the open PID-control loop, consisting
of the PID controller in series with the mass-spring system, as defined in
Equation 4.33. At low frequencies the integrating action is providing a
−1 slope for a high gain. At 100 Hz the gain of the open-loop system is
equal to one as required. Beyond the natural frequency of the uncontrolled
mass-spring system at ≈ 3.3 Hz the system shows after the sharp peak at the
resonance a −2 slope which is reduced to a −1 slope by the differentiating
action around the unity gain cross-over frequency. The phase margin of this
system is about 55◦ , being the distance between the phase lag of the system
(≈ 125◦ ) at the cross-over frequency of 100 Hz and the maximum limit of
180◦ phase lag. The amplitude margin of this system is infinite because in
4.3. PID feedback control 179

0
10

Magnitude (abs)
Ž^
>^ -1
10

+1
>
10
T(s) reference following
>`
S(s)
!˜
10
0
Phase [deg]

>\

>|J

>Z\

>}J
> 0 1 ^ 3 
10 10 10 10 10 10
Frequency [Hz]

Figure 4.24: Bode-plot of the closed-loop PID feedback controlled mass-spring sys-
tem. Both the complementary sensitivity function T(s), representing
the capability to follow the reference as the sensitivity function S(s),
representing the disturbance-rejection capability, are shown. The orig-
inal resonance has disappeared and is turned into a strong rejection
of disturbances at that frequency.

this ideally modelled example the open-loop system reaches the 180◦ phase
lag asymptotically only at infinitely high frequencies.
Figure 4.24 shows the Bode-plot of the closed-loop system. The phase plot is
generally not meaningful anymore in the closed-loop frequency response as
it does not give any further information on stability. For precision systems
that have to move synchronously, like with the wafer scanner of Chapter 9,
the closed-loop phase gives information on the time delay of the movement
and in that case it is still important information.
In this case the phase plot just shows the standard shape of a well damped
mass-spring system with a natural frequency at 100 Hz, corresponding with
the blue magnitude line of the complementary sensitivity function T (s). At
low frequencies the closed-loop gain is equal to one which means that the
system follows the reference signal with only a very small error. Beyond
the control bandwidth the feedback controlled system rolls off to higher
frequencies with a −2 slope. The fact that the feedback controlled system is
well damped can be observed by the absence of a resonance peak.
As mentioned with the previous example of the CD player, the roll off beyond
180 Chapter 4. Motion Control

Magnitude (abs)
1.5

0.5

0
0 100 200 300 400 500 600 700 800 900 1000

Frequency [Hz]

Figure 4.25: The sensitivity function S(s) of the PID-feedback controlled mass-
spring system plotted in a linear scale. This representation clearly
shows the increased sensitivity of this actively controlled system at
frequencies above the bandwidth of 100Hz. When the shaded area
below the magnitude of one is increased for better rejection of low
frequencies, the shaded area above a magnitude of one also increases.
This waterbed effect is inherent to feedback systems and can not be
avoided.

the control bandwidth with a −2 slope is because the differentiating action

is “tamed”. If that were not done, only a damped transmissibility behaviour
would be realised, with a −1 slope attenuation and an infinitely high control
gain at high frequencies. For this reason it is emphasised again that this
additional pole to stop the differentiator is important.
The sensitivity function S (s) shows small values and a +1 slope at low fre-
quencies. This corresponds with a good disturbance-rejection due to the
integrator that acts up to 10 Hz. At the natural frequency of the mass-spring
system a sharp dip is observed in S (s) with a steeper slope at higher fre-
quencies, which is due to the high gain by the resonance. It is very valuable
to be aware of this effect as this high gain can help to suppress external
disturbances at the resonance frequency of the mechanical system. This
effect is very much counter-intuitive and can only be applied when a system
suffers from a continuous disturbance at a fixed frequency. Examples are
not ideally rotating systems like the eccentricity in the track of a CD-player
or a hard-disk drive and the forces by mass unbalance in high speed spindles
of machining centres.
Beyond the unity-gain cross-over frequency S (s) is settling around a value
of one after a frequency range where the magnitude is larger than one, due
to the so called waterbed effect. This waterbed effect has got its name from
the property of any feedback control system that an increase in disturbance-
rejection at low frequencies, due to different controller settings, automati-
4.3. PID feedback control 181

cally causes an increase of the sensitivity at frequencies above the bandwidth.

This is indicated by the Bode Sensitivity Integral:

∞

ln |S (ω)| dω (4.40)
0

This integral is zero when the open-loop transfer function L(s) of the dynamic
system has at least two more poles than zeros and when all poles are located
in the left half of the Laplace plane. This condition is always true with
stable feedback controlled positioning systems.
The reason why the waterbed effect is not so clearly visible in a normal Bode-
plot is due to its logarithmic scales. Figure 4.25 shows the magnitude of S (s)
on a double linear scale. With this representation the frequency range above
100 Hz with an increased sensitivity appears to be much larger than the
frequency range below 100 Hz, while in a normal Bode-plot this difference
is not as pronounced. Also the magnitude is relatively reduced with double
logarithmic scales. In the linear representation the benefit of feedback seems
to have vanished but linear scales are also not representative for the real
value of feedback in the low-frequency area. A reduction of a factor 100 or a
factor 1000 of the disturbances at low frequencies can not be distinguished
on a linear scale. In most motion control systems the disturbances occur
more in the low-frequency area where the process disturbance d (s) has
typically a sort of 1/ f spectrum over the frequency range. This means that
in practice the beneficial effect of feedback in the low-frequency region by
far outweighs the disturbance amplification at higher frequencies due to
the waterbed effect. It is also for this reason that logarithmic scales are
preferred as long as the mechatronic designer is aware of the negative effect
on disturbance rejection just above the unity-gain cross-over frequency for
those cases that significant disturbances are also present in that frequency
region.
182 Chapter 4. Motion Control

4.3.5 PID-control of more complex systems

The previous sections presented rather simplified systems that in reality

hardly ever exist. In this section two examples will be presented to give a
flavour of the variations that can be encountered when designing controlled
motion systems. The first example deals with the control of a system with
negative stiffness like a magnetic bearing. The second example explains a
method how to deal with the resonances due to higher order eigenmodes
that are always present in any mechanical structure.

4.3.5.1 PID-control of a magnetic bearing

In Section 4.1.1 it was mentioned that a system with a negative stiffness

gives a pole in the right-half of the Laplace plane. It was also mentioned
that PID-control can be used to shift the pole to the stable left-half of the
Laplace plane by adding a positive stiffness and damping. Controlling of
such a system is important as some of the electromagnetic actuators that are
described in the next chapter show a negative stiffness. Especially magnetic
bearings are an example of such a system and as a consequence they can only
work with active feedback control. To illustrate how this control is achieved
Figure 4.26 shows the Bode-plot from the following transfer function of a full
plant in one direction (SISO), consisting of an actuator with a gain equal
to one and a (negative!) stiffness kn = −104 N/m, a moving mass m m = 0.1 kg
and a position sensor also with a gain of one:
x 1 1
C t ( s) = = = (4.41)
F m m s2 + k n 0.1 s2 − 104
1 1
C t (ω) = 2 4
=− (4.42)
−0.1ω − 10 0.1ω + 104
2

At first sight the magnitude plot looks like a well damped mass-spring
system but the phase plot is fully different with −180◦ phase over the entire
frequency band, caused by the negative stiffness value. This negative sign
implies that the position is always in the opposite direction of the force.
With increasing frequency the compliance of the mass-line will take over
the total compliance from the negative spring-line at their intersection
frequency4 f 0 :
 
1 1 kn 1 104
f0 = ω0 = = ≈ 50 [Hz] (4.43)
2π 2π m m 2π 0.1
4 This intersection frequency is not named ”natural-“ or ”eigenfrequency“ because no reso-

nance occurs and no eigenmode is observed.

4.3. PID feedback control 183

>Z
10

Magnitude (abs)
> ™'
*

10

>\ *

10
-2
>`
10

>{
10
0
Phase [deg]

>|J
>}J
10
0
10
1 f0 2
10
Z
10
Frequency [Hz]

Figure 4.26: The Bode-plot of the transfer function of a magnetic bearing shows a
frequency response without resonance and with a constant phase of
−180◦ due to the negative stiffness value.

The phase remains constant over the entire frequency band because the
mass-line also has a phase of −180◦ . Next to the negative sign a second ob-
served difference with a regular mass-spring system is the lack of resonance.
This is caused by the fact that the denominator does not become zero at
any value of ω and this corresponds
# again with the location of both poles on

k
the real axis at a value of ± /m = ± 105 ≈ 320. With this information the
design of a PID controller can be realised according to the same steps made
in the previous part of this section. Still some things will show to be quite
different because of the negative stiffness.
First P-control is applied in a magnitude that is at least larger than the
magnitude of the spring-line. In mechanical terms this means that after
closing the negative feedback loop the positive stiffness by the P-control
action will be larger than the negative stiffness of the actuator. For reasons
of robustness against variations in the negative stiffness the positive control
stiffness kp should be chosen minimally as large as the maximum value of
the negative stiffness that can occur. A practical value is at least a factor
two larger than the average value of the negative stiffness. This choice
results in a total positive stiffness that is always equal to or larger than the
magnitude of the negative stiffness.
The effect on the poles is illustrated in Figure 4.27 and it shows that with
only P-control, the closed-loop response would become identical to a regular
undamped mass-spring system. The equivalent natural frequency of this
system would correspond with the unity-gain cross-over frequency f c of
the combined negative-stiffness actuator with the P-control gain kp at a
184 Chapter 4. Motion Control

400
kp > kn
Stable damped
300
system Increase k d
200
Increase k p
100
Increase k p Increase k p
kp= kn
Imaginary Axis

0
™' k p < kn kp < kn ™'
-100 <kn <kn
Increase k p
-200
Increase k d
-300 Stable damped
system kp > kn
-400
-300 -200 -100 0 100 200 300
Real Axis

Figure 4.27: The poles of the instable magnetic bearing system with a moving mass
m m = 0.1 kg and a (negative!) stiffness of k n = −104 −N/m are located
at ±320 on the real axis. The poles are brought to the left-half of the
Laplace plane in two steps. First P-control with an increasing value
of is added until at kp = 2.3 · 104 an undamped system is created with
both conjugate complex poles at ±360 on the imaginary axis. In the
second step D-control is added to create damping by shifting the poles
over a circle to the left. Note that the circle is an ellipse because of
the difference in scales.

frequency that is equal to or slightly larger (square root of stiffness!) than f 0 .

The second step is the addition of D-control according to the rules of thumb
for D-control. In this case the proportional gain should however not be
reduced because then the total loop gain at the start of the differentiating
action would become lower than one with a phase of more than −180◦ , lead-
ing to instability. D-control will shift the poles to the left over a circle that
has been defined for a damped mass-spring system in Equation (3.48) of
Chapter 3. As a further consequence the D-control action will increase the
unity-gain cross-over frequency and in practice the open-loop unity-gain
cross-over frequency will become approximately twice the value of the in-
tersection frequency f 0 of the mass- and the spring-line of the magnetic
bearing.
This effect leads to an important conclusion regarding magnetic bearings:

The negative stiffness of a magnetic bearing sets a lower limit

to the bandwidth that has to be realised in the closed-loop system.
4.3. PID feedback control 185

^
10

Magnitude (abs)
0
10

>^
10 Open-loop
Closed-loop
PD - control
> PID - control
10
0
Phase [deg]

>|J

>}J

>^{J fc
> 0 1 ^ 3 
10 10 10 f0 10 10 10
Frequency [Hz]

Figure 4.28: The open-loop and closed-loop Bode-plot of the magnetic bearing with
negative stiffness with PD-control (dashed lines) and PID-control
(solid lines). In closed-loop the frequency response shows a magnitude
larger than one Q = 2 that stretches over a wide frequency range until
f c . With PD-control starting at 0 Hz and with PID-control starting
around 20 Hz.

When other dynamic problems like resonating mode shapes and

delays in the electronics prevent the realisation of such a high
bandwidth the magnetic bearing can not be made stable.

This statement is in principle valid for any system that is unstable in open-
loop, like an inverted pendulum or a rocket.
As a last addition I-control is applied in order to to increase the rather
limited open-loop gain below ωi = 0.1ωc . Even though it may seem that
a significantly higher proportional gain than the negative stiffness value
would create a very stiff system, this higher proportional gain only creates
a parallel stiffness after closing the loop. In the open-loop situation,
the multiplication results in a loop gain that is equal to the number of
times that the proportional gain is chosen larger than the negative stiffness,
which is often not sufficient for disturbance suppression at lower frequencies.
Furthermore a very high kp far above the negative stiffness would give a very
high unity-gain cross-over frequency with an increased risk on instability
to high-frequency resonances by eigenmodes of the mechanical system.
186 Chapter 4. Motion Control

JQ8

JQ6 Q†Q` Q†^

JQ4

Imaginary Axis
JQ^
Q†J
^JXƒ
J

>JQ^

>JQ4
JJXƒ
>JQ6

>JQ8 PD-control
PID-control
>
>^Q5 >^ >Q5 > >JQ5 J
Real Axis

Figure 4.29: The Nyquist plot of the open-loop frequency response of the mag-
netic bearing with PD- and PID-control explaining the magnitude
of the closed-loop frequency responses by the position relative to the
Q -circles.

Figure 4.28 shows the open-loop and closed-loop frequency response of the
combined PD- and PID-controller with the plant with negative stiffness. In
this case k p = 2.3 · 104 , which is 2.3 times the value of the negative stiffness.
In feedback this value adds to the negative stiffness to a total value of
4
k t = 1.3 · 10
# and the poles of the closed-loop system have an imaginary

value of ± kt/m = ± 1.3 · 105 ≈ 360, as shown in Figure 4.27. The D-control
action goes from 30 Hz until taming at 300 Hz giving a unity-gain cross-
over frequency of 100 Hz. I-control works below 10 Hz. The closed-loop
bandwidth is 100 Hz and the resonance peak is less damped (Q = 2) than
with the previous system with a positive stiffness value as the phase lead is
only around 30◦ . With PD-control this effect is even more prominent as it
stretches until 0 Hz. This effect is fully caused by the −180◦ of the negative
stiffness and can be better analysed with the Nyquist plot of Figure 4.29.
With PD-control the magnitude starts at kp × kn = 2.3 and runs around the
Q = 2 circle. Over the frequency band ranging from 0 Hz until just more
than 100 Hz the graph remains between the Q = 1.6 and the Q = 2 circle
which means that in closed-loop the magnitude will be around 1.8 in that
frequency range, which is observed in the Bode-plot. With PID-control the
graph runs just inside the Q = 2 circle from 20 Hz to 100 Hz, which results
in the observed magnitude of the closed-loop frequency response in that
frequency range in the Bode-plot, while the closed-loop response below 20 Hz
4.3. PID feedback control 187

measurement

k1 x1 x2
100 N m 2.5 ˜10 N m
5

Driven body Sensor

m1 m2
F 0.25 kg k2 0.025 kg

Figure 4.30: When the sensor is connected to the driven body with a compliant
mounting, the mass of the sensor will decouple at high frequencies
due to the second eigenmode, where the two bodies vibrate in opposite
directions by the elastic coupling of k2 .

approaches a value of one because of the high loop gain due to the integrator.
A reduction of the proportional gain below the example value k p = 2.5 · 104
will cause the graph on the Nyquist plot to approach the −1 point on the real
axis, resulting in a higher Q over a wider range of frequencies until at-half
the presently chosen value ( k p = 1.15 · 104 ), the system will start to resonate
at 20 Hz. Optimal tuning of such a magnetic bearing is clearly not trivial as
the optimisation depends on many parameters, like the expected frequency
spectrum of the disturbances and the frequency range of operation.

4.3.5.2 Eigenmodes above the desired bandwidth

At the presentation of D-control it was demonstrated that the D-action

should be terminated (tamed) above a frequency of approximately three
times the cross-over frequency. Next to the fact, that differentiation until
infinite frequencies is impossible anyway, this measure would be beneficial
in case of eigenfrequencies of the system at higher frequency levels. In
general it is indeed a good starting point, when designing the feedback
system, to try to keep the magnitude of the frequency response at these
higher eigenfrequencies below one, in order to guarantee stability.
In the following, this requirement will be examined in more detail and it
will be shown that it is possible to reduce this requirement to some extent
by manipulating the phase. For the example, the same mass-spring system
of the previous section is used, but now the position sensor is more real. It
is not mass-less co-located directly on the primary body, but it consists of
188 Chapter 4. Motion Control

6
10

Magnitude (abs)
4
10

^
10

0
10

>^
10
0
Phase [deg]

>|J

>}J

>^{J

>Z`J
> 0 1 ^ 3 4
10 10 10 10 10 10
Frequency [Hz]
5

3
Imaginary Axis

1
Q=1.6 Q†ZQ^
0

>

>^
> >^ 0 ^ 4
Real Axis

Figure 4.31: Bode and Nyquist plot of the open-loop transfer function L(s) of the
PID-controlled mass-spring system when the sensor is mounted on a
body that decouples dynamically at 500Hz with Q ≈ 20. The graph in
the Nyquist plot passes the −1 point on the real axis at the right hand
side and instability will occur after closing the loop. It illustrates the
necessity to keep the amplitude of the resonance below one in order to
maintain stability.

a body with a mass of 25 · 10−3 kg with a mounting stiffness of 2.5 · 105 N/m,
as shown in Figure 4.30. In Section 3.3.1 of Chapter 3 on the dynamics of
multiple coupled bodies the configuration appeared to result in a second
eigenmode with an eigenfrequency of ≈ 500 Hz. Because in practice such
situations show a very limited damping, a Q of ≈ 20 is used in the modelling.
4.3. PID feedback control 189

Figure 4.31 shows both the Bode-plot and the Nyquist plot of this situation.
The Nyquist plot is especially useful in this situation as it shows in clear
detail, what happens around the −1 point on the real axis with respect to
closed-loop stability in this more complex dynamic feedback system. The
phase at 500 Hz due to the second eigenmode changes from almost −150◦ ,
the mass-line combined with the D-control action, to below −180◦ . This
phase shift occurs in the frequency range where the magnitude becomes
larger than one due to the resonance peak of the second eigenmode and as
a consequence instability will occur.
The Nyquist plot clearly indicates that the −1 point on the real axis is kept at
the right hand side when passing. Only when the magnitude of the second
eigenmode is kept below one by a reduction of the loop gain this instability
condition would not occur, but then the resulting bandwidth of the controlled
system would be reduced to below ≈ 20 Hz.

Shifting the phase

Different approaches are possible to overcome this problem. Often a notch

filter is applied, with an inverse characteristic of the resonating eigenmode
to suppress this specific resonance. An example was shown with the model-
based feedforward controller of Section 4.2.1. This method requires however
an almost perfect tuning of the frequency of the filter to the resonance,
particularly with a very high Q -factor and in mass production every con-
troller will then be different due to the spread in the system dynamics by
production tolerances.
One might also consider adding damping to the decoupling spring k2 al-
though generally this spring is not compliant on purpose but the result of a
not infinite stiff mounting. Adding damping to a mounting is very hard to
do and sometimes even results in a reduced stiffness, shifting the second
mode to lower frequencies, like with rubber mounts.
Another method, more related to active control, is based on the understand-
ing that the problem occurs because the phase is almost −180◦ . When the
phase lag would be larger, a high magnitude above one would no longer
give a problem as long as the −1 point on the real axis is passed at the left
hand side. For the given example this can simply be arranged by inserting a
simple low-pass filter in the loop at a frequency around the eigenfrequency
of the decoupling mass. Combined with the first-order low-pass filter used
for the “taming” of the differentiator, this second low-pass filter forms a
second-order low-pass filter. This combination can also be replaced by a less
damped second-order low-pass filter in order to limit the phase shift at the
190 Chapter 4. Motion Control

1
10

Magnitude (abs)
0
10

@

o
> 
*}J
10 ™?
?`\JXƒ
?Z\JXƒ

0
Phase [deg]

>}J

>Z`J

>\\J 1 2 Z
10 10 10
€!Xƒ’
\

magnitude

~
@•


Z
?
$
2

1
Q†Q`
Q†ZQ^
0

>

>^
> >^ 0 2 
@•


Figure 4.32: Inserting an additional low-pass filter in a PID-controlled mass po-

sitioning system will only slightly decrease the phase margin, but
changes the phase at the decoupling mass resonance to below −180◦ .
The Nyquist plot shows more clearly the beneficial effect. When filtered
at 350 Hz the Q = 1.6 circle is just touched.

unity-gain cross-over frequency, but this is a detail that can be adapted when
a further optimisation of the system is required. For the understanding of
the principle it is useful to treat them first as two separate filters.
Figure 4.32 shows the effect of the additional low-pass filter on the open-loop
transfer function of the example PID-controlled mass-spring system. Next
to the situation without the additional low-pass filter, also the open-loop
response of the system with a low-pass filter at 350 or 650 Hz is shown. The
4.3. PID feedback control 191

1
10 ?`\JXƒ
?Z\JXƒ

Magnitude (abs) 10
0

>
10
-2
-3
š
'
T(s)
1
10
Magnitude (abs)

0
10

>

'
S(s)
10 2 3
10 10
Frequency [Hz]

Figure 4.33: The closed-loop Bode-plot of the sensitivity functions of the PID-
controlled mass positioning system with decoupling mass show the ef-
fect of the low pass filter frequency on the magnitude of these functions
at the cross-over frequency and the eigenfrequency of the decoupling
mass. At 350 Hz these magnitudes are equally balanced. This optimal
setting is also confirmed by the sensitivity plot at the lower side.

Bode-plot clearly demonstrates that such a filter has only a limited negative
effect on the phase margin at the cross-over frequency of 100 Hz, while the
phase at the resonance has definitively come below −180◦ . In this more
complex PID-control situation the Nyquist plot is an indispensable tool as
it more clearly shows the effects regarding stability. With the additional
first-order low-pass filter, stability is achieved as the −1 point on the real
axis of the Nyquist plot is passed at the left hand side, while without a low-
pass filter the system would be unstable. An optimal situation is created
when the filter frequency is at 350 Hz resulting in a symmetrical left hand
passage of the Q = 1.6 circle, indicating an equal magnitude of the closed-
loop response for the two frequencies that are closest to the circle. It results
in a maximum amplitude rise of 60 % both at the cross-over frequency and
at the second eigenmode. This optimum is confirmed in the complementary
sensitivity T ( s) and sensitivity S (s) Bode-plot of Figure 4.33.
192 Chapter 4. Motion Control

6
I-control
10

Magnitude (abs)
-1

+1
Phase
5
10 
$

D-control
-1
Noise
P-control
!
4
10
|J
Phase [deg]

45

>\

>|J
> 0 1 2 3 4
10 10 10 10 10 10
Frequency [Hz]

Figure 4.34: The PID-control transfer function with additional phase shifting low
pass filter can be seen as an optimal motion controller of a mechanical
system.

A clear benefit of this method is that the influence due to the resonance
of a second eigenmode with an amplitude larger than one is effectively
suppressed by the feedback. The original Q value from the uncontrolled
system is reduced from more than twenty to less than a factor two by tamed
PID-control with an additional first order low-pass filter. Another advantage
of the additional low-pass filter is the −3 slope of the sensitivity function T (s)
at higher frequencies which means that the resonances of higher eigenmodes
and the high-frequency sensor noise will be even better suppressed.

4.3.5.3 “Optimal” PID control

The described PID-control transfer function with a combination of the three

basic PID-control gains and a second-order low-pass filter for taming the
D-control action and shifting the phase at the first resonance is shown in
Figure 4.34. In practice this configuration has become a standard in low-
stiffness mass-positioning feedback systems with force actuation by Lorentz
actuators. It has proven to be an optimal choice in that application area
with respect to robustness against system variations and tolerances.
In control theory, the term optimal control generally has another though
closely related meaning, where a mathematical optimisation method is
4.3. PID feedback control 193

used to find a control algorithm for a given system, such that a certain
optimisation criterion is achieved. Optimality of this PID-control scheme
with second-order low-pass filter is understood as the best combination
of robustness and performance so corresponding with the “practicality”
criterion of dynamic positioning systems.

4.3.5.4 Open-loop and closed-loop

This section has clearly demonstrated the creation of a virtual spring and
damper with an additional integrating function by PID-feedback control.
It is unfortunate that a mechanical equivalent for the integral gain ki is
not readily available. The integrating action provides a high controller gain
C at low frequencies. For that reason the closed-loop frequency response
becomes equal to one in that low-frequency region, independent of the
transfer function of the plant. For that reason the integral action is acting
like a “super-spring” that increases the loop-gain at low frequencies and
gradually reduces its distortion to zero. As a result the steady-state error of
the feedback controlled system in response to a constant reference signal
also becomes equal to zero.
These virtual elements are virtually placed in parallel to the physical spring
and damper of the plant and the combination acts as if the virtual and
real stiffness values and damping coefficients are simply added. It is very
important however to realise that this effect only occurs upon closing the
loop. In the open-loop situation, when the output y is no longer connected
to the summing point with the reference signal r , the controller is placed
in series before the plant and in that case the transfer functions of the
controller and the plant are multiplied. This difference often leads to
confusion as can be illustrated by examining only the proportional branch
of the system as example, consisting of the stiffness of the spring and the
proportional gain.
From the previous chapter it is known that the transfer function of a physical
spring with stiffness k acts open-loop as a compliance with a value of k−1 .
In the open-loop situation the total compliance of this transfer function
can be increased with P-control with a value needed to bring the total
compliance to unity gain (⇒ kp = k). After closing the loop this value of
k p = k would result in an additional spring with a value of k giving a total
stiffness of 2 k in combination with the physical spring. At first sight there
appears no relation whatsoever between the unity gain open-loop value
and the double stiffness in closed-loop but when calculating the closed-loop
frequency response, the open-loop transfer function of G (s)C ( s) = 1 results in
194 Chapter 4. Motion Control

a feedback complementary sensitivity function T (s) = GC/1 + GC = 0.5. This

value indicates that the closed-loop compliance equals half the open-loop
compliance and implies that the stiffness of the closed-loop system has
become twice the stiffness of the physical spring, which corresponds with
the double stiffness that was determined by the addition.
This example shows that it is necessary to be always aware that the virtual
elements created by the feedback loop are only present when the feedback
loop is closed. In the open-loop situation these element-terms are mean-
ingless. The open-loop analysis is used for tuning the controller regarding
stability and loop gain as explained in this section and should not be mixed
with these element-terms.
4.4. State-space control representation 195

4.4 State-space control representation

Generally a single body positioning system needs to be positioned in six

orthogonal coordinate directions necessitating six actuators and six sensors.
In control system terms this positioning system is called a Multiple Input
Multiple Output system (MIMO) and complications can occur due to the
mutual interference or cross-coupling between each direction. When all
sensors and actuators are well aligned in six unique directions, the system
can be reduced to a combination of six independent SISO subsystems and
even when they are not aligned a SISO approach can often be achieved
by means of transformation matrices under the condition of a sufficiently
deterministic behaviour. An example of this principle will be presented in
Chapter 9 with the control of a wafer stage. In situations, when there are
strong (non-linear) cross-coupling terms between the several inputs and
outputs that can not be compensated in a straightforward deterministic way,
this approach is not sufficient especially when higher frequency eigenmodes
with a multitude of bodies also create cross-coupling terms.
The state-space representation has been introduced in control practice as a
time-domain related method to be used in digital controllers to calculate the
required MIMO control actions. The method is highly efficient because it
breaks down an nth -order differential equation into n first-order differential
equations that are written in a matrix notation. With the state-space model,
every successive control action can be calculated by straightforward vector-
matrix operations.
The state-space model has three different variables, the inputs u, the outputs
y and the state-variables x that describe the momentary state of the system.
Generally these variables are written as vectors because they contain a
number of elements. The state-variable vector defines a space of which the
axes are the state-variable elements in the vector. This has given the name
state-space to this modelling method. The total number of state-variables,
mostly just called “states”, is at least as large as the number of orders of the
differential equation that is modelled.
The standard state-space notation is given by the following two equations
that do not contain the ( t) term as the state space representation is uniquely
used in the time domain:

ẋ = Ax + Bu (4.44)
y = Cx + Du
196 Chapter 4. Motion Control

u x x y
B ³ dt C

Figure 4.35: Graphical representation of a standard multi dimensional state-space

system. The feedback section with matrix A represents the eigendy-
namics of the system. The input section of the system consists for
instance of the actuators with gain matrix B and the output section
of the system consists for instance of the sensors with gain matrix C.
The section with matrix D represents the direct feed-through of the
input to the output and is for instance caused by the cross-coupling
from the actuator current to the sensor signal.

These equations show how the state-variable x of a dynamic system and the
system output y evolve over time as a function of the matrices A, B, C, D, the
input u and the initial conditions x0 of the state vector x. The matrices are
the eigendynamics matrix A, the input matrix B, the output matrix C and a
feed-through gain D.
Figure 4.35 shows the structure of this standard state-space model in a
graphical way. Like described at the figure, the different elements of these
matrices all represent a certain physical property or function.

4.4.1 State-space in relation to motion control

State-space motion control uses a state-space model of the full mathemat-

ical description of the plant, including actuators and sensors, where the
controller actions are applied to the external connections of the uncontrolled
system, the inputs and outputs. To illustrate what this means, a few exam-
ples will be shown that were introduced before with the analytical approach
to motion control.
Chapter 3 presented a damped mass-spring system as a second-order system
and when combined with one or more spring-coupled bodies, it became a
higher-order system adding two orders for each coupled mass-spring system.
In general the order n of a dynamic system is equal to the number of energy
4.4. State-space control representation 197

containers in the system, like a spring (E = 0.5 kx2 ) and a body (E = 0.5 m ẋ2 ).
A damper does not contain energy. When the state-space model is used for
controlling the plant, the order of the differential equations that describe
the plant does not only include the mechanics but also the dynamics of the
applied sensors, amplifiers and actuators. A first-order filter at the sensor
introduces an additional state and the dynamics of the actuator also create
corresponding states.
First the different matrices will be examined from the viewpoint of motion
control:

• The eigendynamics matrix A includes the parameters that describe

the dynamics of the uncontrolled system, including the dynamics of the
power amplifier, the actuator, the mechanical system and the sensor.
Its dimension is square ( n × n).

• The input matrix B contains the simple gains or scaling factors of the
actuator or power amplifier, but not their dynamics as these have to
be included in matrix A, as stated above. The number of columns of
the input matrix is equal to the number of inputs ( i ) and the number
of rows equals the number of states n.

• The output matrix C contains the simple gains or scaling factors of

the measurement system, but also without the dynamic part as that
also has to be included in matrix A, as stated above. The number of
columns of the output matrix is equal to the number of states (n) and
the number of rows equals the number of outputs o.

• The feed-through matrix D is often not present in mechatronic system,

but can for instance be caused by actuator-sensor cross-talk, like when
using electromagnetic actuators and an Eddy-current sensor. Its di-
mension depends on the number of inputs and outputs with i columns
and o rows.

These matrices and the states in the state-space model are illustrated with
two examples, a damped mass-spring system, with and without a classical
PID-controller.
198 Chapter 4. Motion Control

u F 1 x2 
x x2 x x1 x y x
B ³ dt ³ dt 1C
m

c

m
k

m
A

Figure 4.36: Graphical representation of the scalar, single input and output (SISO),
state-space representation of a mass-spring system with damping.
The loops of the damper (velocity) and the spring (position) are shown
separately.

4.4.1.1 Damped mass-spring system

A simple single directional (SISO) damped mass-spring system with position

x of the body is described in state-space as follows:
The state x and its derivative ẋ equal:
* + * + * + * +
x1 x ẋ1 ẋ
x= = ẋ = = (4.45)
x2 ẋ ẋ2 ẍ

With this SISO example the input u would be the force with a scalar value
u = F . The output y is equal to the position x, so it is also a scalar. The
input matrix B in this case is the conversion from force to acceleration with
a mass factor and the eigendynamics matrix A contains the damper and
spring relations which is shown in the following expansion of the state-space
equations:
+ ⎡ * ⎤ * + ⎡ ⎤
ẋ 0 1 x 0
ẋ = Ax + B u =⇒ = ⎣ −k − c ⎦ · +⎣ 1 ⎦F
ẍ ẋ
m
* + m m
x
y = Cx + D u = [ 1 0 ] · + [ 0 ]u = x (4.46)
ẋ

When this is written out for the acceleration the following equation is ob-
tained:
−k c F
ẍ = x − ẋ + =⇒ F = m ẍ + c ẋ + kx (4.47)
m m m
which is the standard force equation that was used in Chapter 3 to derive
the frequency response of the system. Also with more elaborate systems
4.4. State-space control representation 199

the transfer function is derived by applying the Laplace transform on the

state-space model according to this example.
The state-space representation of this SISO mass-spring system is shown
in a graphical way in Figure 4.36 where x1 = x is the position, x2 = ẋ is the
velocity and ẋ2 = ẍ is the acceleration of the system. The feedback loops with
the stiffness and damping terms will show to be very illustrative when the
combination of a mass-spring system with feedback control is introduced in
the next section.
Another important aspect of the state-space model is the fact that the poles
of the transfer function are equal to the eigenvalues of the A matrix. These
eigenvalues are those values of s where the determinant det ( sI − A) equals
zero, with I being the unity matrix with the same dimension as A.
* + ⎡ ⎤ ⎡ ⎤
s 0 0 1 s −1
− ⎣ −k −c ⎦ = ⎣ k c ⎦=0 (4.48)
0 s +s
m m m m
When written out, this gives the following characteristic polynomial in s:
&c ' k cs k
s +s + = s2 + + =0 (4.49)
m m m m
The poles are equal to the eigenvalue solutions of s of this polynomial:

c k c2
p 1,2 = −σ ± j ωd = − ±j − (4.50)
2m m 4 m2
#
With ω0 = k/m these values of the poles become equal to the values that
were found in Section 3.2.3.2 where the critical damping of a mass-spring
system was derived:
$
k   
c
m k c2 c ω0 c2
p 1,2 = −σ ± j ωd = −  ±j 1− =−  ± j ω0 1 − (4.51)
2 km m 4 km 2 km 4 km

This means that by changing the terms in the A matrix, the eigenvalues
can be adapted such that the poles become located at a different position in
the Laplace plane. This is exactly what happens with feedback control and
the state-space model offers a very straightforward method to achieve an
optimal pole location, as will be shown in the following examples.
200 Chapter 4. Motion Control

y x r
1 x2 
x x2 x x1 x
m
B ³ dt ³ dt 1C
 ³ dt
e r-x
c

u F m
k

m
A
d
D - control
kd
dt
kp
P - control

I - control
ki

Figure 4.37: Graphical state-space representation of SISO PID-control, applied to

a damped mass-spring system. A feedback path, consisting of three
terms, calculates the output of the controller u from the error e after
differentiating and integrating. This representation shows that the
blue dash-dotted loop over k p has the same function as the internal
loop by the stiffness k and the red dash-dotted loop via k d has the same
function as the damping coefficient c as the series of an integral and a
differential term cancel each other out in the loop. The I-control action
has no corresponding equivalent inside the uncontrolled system and
is introduced to reduce the steady-state error. Note that the reference
is inserted between the output y and the controller.

4.4.1.2 PID-control feedback

To demonstrate the state-space approach with active feedback, the imple-

mentation in a PID-control setting is presented for a second-order mass-
spring system.
A graphical state-space representation of the second-order mass-spring
system under PID-feedback control is shown in Figure 4.37.
With this representation it is possible to trace the different feedback loops
that work on the system separately. The two internal loops of the uncon-
trolled mass-spring system determine its dynamics with the spring and
damper action. The three feedback control loops are added to these two
internal feedback loops.
The blue dash-dotted P-control feedback loop over kp runs parallel with
4.4. State-space control representation 201

the internal stiffness loop with element −k/m. In combination with the
input element 1/m by B, the value of kp is added to k. The P-control gain
therefore corresponds to a modification of the stiffness of the system, just as
presented in the example of the CD player. The similar relation exists for
the differential gain kd that corresponds to the damping coefficient c in the
mechanical system. The red dash-dotted loop over the differentiator block
( d/d t) converts the measured position into a velocity that serves as the input
to kd . In the D-control feedback loop this calculated velocity is equal to the
real velocity ẋ. In summary, kp and kd are used to change the stiffness and
damping of the system.
A special word of attention has to be given to the reference input and the I-
control loop with the additional integrator. In this PID-control example, the
reference is indicated at the same position as where it was located with the
analytical approach of the previous section on PID-control. With state-space
control this location is more free to choose. In theory the reference signal
can be inserted at any place in the control part because it is an externally
definable signal. Often the reference is inserted at the output of the feedback
controller. This is the same place where feedforward control is inserted, as
was indicated at the beginning of this chapter in Figure 4.1.
The integration of the error introduces a third feedback loop. There is
no equivalent element to the integrator in the uncontrolled second-order
mass-spring system. In closed-loop, the action of the integrator results
in a reduction of the steady-state error to zero and due to this effect it is
sometimes called a super-spring.
Regarding the placement of the poles it is clear that the feedback constants
directly determine the pole locations as kp + k determine the undamped nat-
ural frequency with two poles on the imaginary axis while kd + c shifts these
poles over a circle back to the real axis. The fact that kp and kd are added to
the terms in the A matrix indicates that another A matrix is obtained with
different eigenvalues and this effect is further used with full state feedback.
202 Chapter 4. Motion Control

4.4.2 State feedback

Even though PID controllers still determine the mainstream of practical

position control systems, it is useful to shortly introduce the more fundamen-
tal mathematical methodology, that enables to design control algorithms
for complex systems with a higher number of inputs and outputs as well as
system states. It is also a useful method when various positioning systems
suffer from cross-couplings. This fundamental approach starts with the
assumption that the best controller can be designed when the full model,
including the dynamics and all states is known. By applying proportional
feedback on the individual states, state feedback is realised. First this prin-
ciple will be better explained and in the following sections it will be shown,
how by means of system identification the state-space matrices can be de-
termined and how state estimators are used to approach the ideal situation
of all states being known as if they all are measured. Even in the case that
direct measurement is not possible it will be shown that the states can be
reconstructed (estimated) from a limited set of measured states.
The standard state-space notation of a dynamic system was shown to be
equal to:

ẋ = Ax + Bu
y = Cx + Du

In most systems the direct feed-through D is not present or very small and as
simplification D is further assumed to be equal to zero for this explanation
of state feedback. As explained, in state feedback it is assumed to have full
information of the state vector x. This means that all states of the system
are directly measured or calculated from a limited set of measurements.
The feedback-loop is closed with a feedback matrix K, with n columns and i
rows. The resulting output of the feedback controller equals u = −Kx. When
the reference is not taken into account, as that only adds an additional
setpoint term to u without changing the closed-loop system dynamics, the
state-space notation of the resulting dynamic system becomes:

ẋ = Ax + B(−Kx) = (A − BK)x (4.52)

y = Cx.

As a result a new system is created, looking the same as the open-loop system,
but with a new eigendynamics matrix A = A − BK. So to achieve the desired
performance of the closed-loop controlled system under state feedback, the
desired pole locations have to be defined for stable operation and the required
4.4. State-space control representation 203

Feedforward
r e u x2 x2 x1 y
k3 ³ dt B ³ dt ³ dt C
 
a2
a1

k2

k1
I - control loop K

Figure 4.38: State feedback on a SISO positioning system. The position x1 and
velocity x2 are both measured. The derivative of the velocity, ẋ2 equals
the acceleration and is in this case not measured nor required for full
state feedback as this second order system is fully described by two
states. By applying proportional feedback for the position and the
velocity, the poles can be placed at the appropriate locations in the
Laplace plane. The I-control loop is additional to reduce steady-state
errors.

feedback matrix K needs to be calculated. A state-space system is stable

if the eigenvalues of the A matrix have all a negative real part, directly
corresponding to the poles of the system’s transfer function. This means in a
closed-loop controlled system, that all eigenvalues of A = A − BK need to have
a negative real part and sufficient damping for the imaginary (oscillatory)
part in order to achieve a good control performance.
Figure 4.38 shows a graphic representation of a state feedback controller on
a one dimensional (SISO) mass-spring positioning system, where the two
states, the position and velocity, are measured and where the integral of
the position is calculated. It is shown that the individual parameters k#
of the feedback vector correspond to the individual parameters a # of the
system’s eigendynamics matrix. This is very similar to the case presented
before for the PID controller of the second-order mass-spring system. This
means that state feedback is equal to a set of P-control terms for a system
where all states are directly measured, and the individual P-control gains
correspond to the parameters of the feedback matrix K.
204 Chapter 4. Motion Control

In state-space control the integral action is a special case as one might say
that this creates another state by the controller itself without any direct
link to a similar state in the system. This state is not needed for controlling
the system dynamics as two states are in principle sufficient for a second
order system. As was previously explained, the I-control action is added
to a feedback control system in order to reduce the steady-state errors by
means of an increase of the loop-gain at low frequencies. For this reason
it is customary in state-space control to introduce the I-control action as a
separate negative feedback loop from the system output y to the input after
comparing with the reference r . The integration will only reduce errors in
the low-frequency area so high-frequency components of the reference signal
will not be controlled and even attenuated by the I-control action (1/ω). This
effect is solved by adding the high-frequency part of the reference to the
feedforward input of the system. In that case the state-feedback loops will
control this part of the frequency spectrum.
Side note: A critical designer of this control system also could say here:
“Why should I use two sensors. One position sensor is enough as I can
apply D-control to derive the velocity”. While this statement is fully correct,
one must not forget that it is always necessary to deal with sensor noise.
If the position sensor is noisy, the derivative action would amplify this
noise especially at high frequencies and this noise would be inserted in
the feedback loop. For higher order systems with for instance multiple
springs and bodies, where only the position of the last body is measured,
even multiple derivatives of the sensor signal would be needed to apply full
state feedback and this would amplify the noise even more. This is another
example of the often economic trade-off that a mechatronic designer needs
to make.

4.4.2.1 System Identification

For a successful controller design good knowledge of the dynamics of the

plant is crucial. Particularly for the design of model-based controllers a
mathematical model of the uncontrolled system dynamics is required. But
also for the tuning of classical controllers, such as a PID-controller, knowl-
edge of the transfer function of the plant is necessary. These models can be
obtained via physical modelling based on first principles, as described in sev-
eral chapters in this book for the mechanical structure, the amplifiers and
the actuators. An alternative to this physical modelling is to obtain a math-
ematical model by fitting a mathematical transfer function to the measured
frequency response of the dynamic system via the System Identification
4.4. State-space control representation 205

methodology.
With System Identification the model of the plant, including the sensor, is
obtained by fitting the parameters of the differential equation to a measured
set of input and output data. When the structure of the expected model is
known, only the model parameters are identified, which is called grey-box
system identification. If also the structure of the system is unknown, this is
called black-box system identification and the identification algorithm then
also has to estimate the model-order and structure. The system identifica-
tion directly gives the state-space form or the transfer function of the system,
but is always limited to the minimal realisation of the system, which means
that from the system dynamics of the plant only those eigenmodes are con-
sidered that can be observed as well as controlled. This is in general not a
problem when no other forces act on the system that excite the un-observed
and un-controlled modes and it can even be desired as it avoids excitations of
these modes by noise in the control loop. Nevertheless it is important to be
aware of this fact because the un-observable or un-controllable eigenmodes
might very well be a main cause for problems in a complex mechatronic
system that consists of many actuated subsystems.

4.4.2.2 State estimation

In the previous section it was shown how a suitable controller can be realised,
if all states are measured. It is however not always possible to measure all
positions, velocities and related derivatives in a mechatronic system, while
precision sensors are also quite expensive. As mentioned before, using the
derivative of the position signal in order to obtain the velocity signal, like in
a D-control feedback system is possible, but may be too sensitive to noise.
One way to get the full information of the state vector is to build a real-time
estimator that is based on the model of the plant including the eigendy-
namics. Such an estimator is also called an observer while it observes the
behaviour of a system by comparing it with the modelled behaviour. Such
an observer also allows a trade-off between the bandwidth (speed) of the es-
timation and the noise performance. An observer with an optimal trade-off
between these two important properties is called a Kalman-filter, named
after the Hungarian mathematician and electronic engineer Rudolph Emil
Kálmán.
Figure 4.39 shows the configuration of a state observer in combination with
state-feedback control of the observed system. The blocks in the dashed
box represent the real mechatronic system. The blocks in the dotted box
represent the mathematical model that is implemented on a computer to
206 Chapter 4. Motion Control

u x x y
B ³ dt C

Real
A system

e -
L
r u x obs xobs yo
B ³ dt C
-
A Real-me
model

K
Model-based controller

Figure 4.39: Model-based controller with an observer to estimate not measured

states in a state feedback control system. The real-time feedback
path is determined by the feedback matrix K based on estimated state
values from within the model. The model is updated by the difference
between the observer output yo and the real system output y via the
matrix L.

simulate the behaviour of the mechatronic system in real-time. When both

systems receive the same input signal u, and both systems are identical,
which means that a perfect model is available, both outputs y and yo should
be the same. However, in reality always modelling errors will occur while
also the mechatronic system can be disturbed by external forces that are not
taken into account and causing position and velocity errors. To compensate
for these deviations the observer-gain matrix L is introduced, that deter-
mines a feedback of the prediction error to the observer and is given by the
difference between the output of the model and the output of the real system
(e = yo − y). L has to be designed such that the closed-loop system for the
observer part is stable. This is the case if the feedback loop of the observer,
determined by A, L and C, only has poles in the left-half of the Laplace
plane. The choice of L influences the location of these poles, and determines
the speed at which the state observer follows changes in the actual states
of the physical system. When a fast response is chosen, the observed state
4.4. State-space control representation 207

vector will quickly follow changes in the system but will also respond more
to noise, originating for instance from the sensor. A slow response of the
state observer introduces some “smoothing” or masking of fast disturbing
signals, for instance when the measurement is corrupted by sensor noise.
Like with the name of an optimal observer, the observer gain that is provid-
ing the optimal trade-off between speed and noise performance for a known
co-variance of the sensor noise is called the Kalman-gain.

4.4.2.3 Additional remarks on state-space control

The analytical modelling of dynamic systems with transfer functions gets

most attention in this book because it is more intuitively connected with
the physical behaviour of a mechatronic system. Nevertheless the state-
space approach has several advantages over the analytical transfer function
approach especially with more complex systems. The matrix calculations
enable to directly create a MIMO controller where all cross-coupling between
different channels in taken into account. This second advantage entails
also a disadvantage as it becomes more difficult to distinguish problems
that are caused by malfunctioning of one element in the complex system.
This argument has long hampered the application of MIMO controllers in
practical mechatronic systems.
Furthermore the state-space model allows to more easily include any initial
conditions and additional disturbances that act on the same system states.
The observer based controller has another potential advantage. While
the standard implementation of an observer runs on all measurements in
real-time at the same sampling rate as the controller, it is also possible to
decouple these sampling rates. The time behaviour of the L matrix can be
chosen in relation to the level of fidelity of the model to reality. When the
model is almost ideal, the estimation error e will show very little deviations
and when these deviations are slow, L can be reduced to correct the model
only occasionally. This introduces the possibility to decouple the timing of
the measurement path from the timing of the feedback loop.
One can think of several situations where measurements can only be done
at certain long intervals. One example occurs when an object has to be
placed at another location for the measurement because of environmental
constraints. Another example is the situation where the process needs to be
terminated for a precise measurement because the process induces an excess
of noise and disturbances. In those cases, the observer provides information
to the feedback loop prior to the real measured value, so running rather as a
208 Chapter 4. Motion Control

simulation of the dynamics, thus preventing instability by the phase lag of

the slow measurement signal. As long as the measurement delay is known
and the model is sufficiently accurate, this method can be really effective
as it defines a process of frequent re-calibration of the real-time dynamic
model at regular intervals. One might also say that at regular intervals the
real feedback loop is temporarily broken and between these intervals, the
observer will act as a simulator or as an adaptive feedforward controller.
When the model is not perfect, a combination of real time local approximate
measurements with observer based correction at longer intervals on the
overall performance might still give the required result. The system would
behave like a ship sailing over the ocean in ancient times, where the captain
as controller determined his direction by real-time feedback corrections that
were based on the observation of the wind and the ocean waves, while he
corrects his model by infrequently ”shooting a star“ with his sextant.
4.5. Limitations of linear feedback control 209

4.5 Limitations of linear feedback control

As stated in the introduction of this chapter, for simplicity only linear (con-
trol) systems are discussed in this book. One of the main limitations of
linear control is the assumption that the system states of the plant are
possible and can be measured over the total range of interest. Further it is
assumed that the control action is not limited. In reality any linear system
will have its boundaries such as range limitations and other non-linearities.
For the sake of completeness this last section deals with some of the most
common non-linearities that occur in precision positioning systems. The
discussion on how to compensate or account for these possible limitations
by the design of an appropriate (non-linear) control system would result
in a book that is about as comprehensive as the one that you are reading
now. For that reason non-linear control is not included here even though
non-linear control carries a good potential to improve system stability and
push the performance-limits even higher. For further information on this
topic it is advised to read the control-related references that are listed in
the appendix.
When taking a detailed look again at the block diagram of the motion control
system shown in Figure 4.1 it is possible to identify some potential sources
of non-linearities.
On the side of the plant any physical system certainly will have its limita-
tions, such as a maximum measurement range of a sensor, called clipping
and the limitation of the actuation range of actuators by saturation. For
instance piezoelectric actuators that are based on a stiff design can only
move a few micrometres. Other examples of limitations are the maximum
force that a Lorentz actuator in a zero-stiffness design actuator can generate.
In the amplifiers any slew-rate limitation of the speed of change of the output
and maximum current or voltage of the power stage will have impact on
the system performance. Further to be mentioned are non-linearities like
backlash, friction, and non-linear stiffness in the mechanical components or
creep and hysteresis as they occur for instance in the piezoelectric transduc-
ers that are presented in Section 5.6 in the next chapter.
Also in the (digital) control system sources of non-linearities and limitations
can be identified, such as the quantisation (in value) and sampling (in time)
of the continuous sensor signal by the A/D-converter, the limited update
rate and resolution of the control signal due to the quantisation of the D/A
converter, but also timing uncertainties in the sampling of the converters,
the so-called jitter.
While the limitations due to the control system can easily be accounted for
210 Chapter 4. Motion Control

~*
1.8e6 ³ dt
@! 
”*

3.3e4 1/0.25 ³ dt ³ dt
 Limit
€! d
160 @! 0.001
dt "

› 100 /0.25

?


Figure 4.40: MATLAB-Simulink model of the PID controlled mass-spring system

of Equation. 4.33 with saturation of the actuation force.

by choosing the right components, limitations that are present in the plant
are a frequent source of errors and can even cause damage to a system when
being overlooked. One reason for such errors in practice is that standard
modelling software packages, such as Matlab and Simulink, are allowing
extremely large control action, error inputs, and system states, that may
not have any relation with capabilities of the real system. The limitations
(saturation) of actuators and sensors have to be taken into account and must
be integrated in the Simulink-models by adding appropriate function blocks,
like shown in Figure 4.40, in order to keep the model close to the actual
physical system. This means, that a practical insight in the limitations of
sensors, amplifiers and actuators is necessary to prevent costly mistakes in
modelling mechatronic systems.
As an example to visualise the effect of such limitations, Figure 4.41 shows
the behaviour of the PID-controlled system from the previous section (see
Equation 4.33) in case of a saturation of the actuation force, putting a
limit to the maximum possible control effort. The reference signal is a
trapezoidal signal with a fundamental frequency of 5 Hz and a peak-to-peak
amplitude of 0.2 m (dotted black line in the first panel of the figure). For
the unconstrained system the actuation force ranges up to about 1600 N
(blue solid line), while the control effort for the constrained system is limited
in the simulation to 1 kN (yellow dashed line) and 300 N (red solid line),
respectively.
While the control action exceeds the actuation force limit, the control-loop is
broken and the system does not react to changes in the control action as the
maximum actuation force is already applied (and with this the maximum
energy that can be supplied to the system per unit of time). This means
that due to this saturation the linear control system becomes non-linear
4.5. Limitations of linear feedback control 211

0.2

?
 0

-0.2
0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
@!"

1000
0
-1000

0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

0.1
›

-0.1
0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

1000
~*

-1000

0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

2000
”*

-2000
0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1

Figure 4.41: Behaviour of a mass positioning system under PID-control with vary-
ing limitation of the maximum actuation force: unconstrained (blue
line), limited to 1000 N (dashed yellow line), and limited to 300 N
(solid red line). The first plot shows the controlled position of the
mass in response to a trapezoidal reference signal (black dotted line).
The second plot shows the output of the PID-controller, equalling the
actuation force. The third plot denotes the error signal, which with
a scaling factor also equals the P-action. The fourth and fifth panel
show the I-action and D-action. The mismatch between the actuation
expected by the linear (unconstrained) PID-controller, given by the
blue lines, and the actual actuation causes a significant degradation
in control performance for the case where the actuation force is limited
to 300 N, and the system starts to fiercely oscillate.

and the system does no longer react as the PID-controller would expect.
For the case where the actuation force is limited to 1 kN, the saturation
level is reached only for a very short moment (dashed yellow line) due to the
D-gain when the reference signal changes. This is one of the reasons why
for practical systems one should use smooth reference signals rather than
steps and impulse-like signals. As this saturation is limiting the control
action only very little, the increase of the control error is also very small
212 Chapter 4. Motion Control

and the system remains in saturation only slightly longer for the time that
the unsaturated force would exceed this value of 1 kN. When the system
re-enters the linear regime, the PID-controller easily can compensate for
this slightly larger error during saturation.
For the case where the actuation force is limited to 300 N, however, the
energy per time unit that can be put into the system for control is much lower
than the PID-controller would demand for the given reference signal. The
mismatch between the linear system and the actual system is too large and
leads to a larger control error and a significant degradation of the system
performance. The main problem here is, that the I-gain of the controller
continues to accumulate the error during the entire time that the system
is in saturation, which takes significantly longer and consequently results
in much larger accumulation of the I-action as would be the case for the
unconstrained system. This effect is called integrator windup. This longer
integrated and larger error cannot easily be compensated when the system
re-enters the linear regime, as the I-action only starts to decrease once
the error changes its sign like when the system is moving in the opposite
direction. This results in strong oscillations of the system and even may
lead to limit-cycling as the too large accumulation of integral action can lead
to a bang-bang action where the control signal is just switching between the
positive and the negative saturation of the actuation force.
In real systems measures have to be taken in the controller to prevent these
phenomena. One of the possibilities is to reduce or even cancel the I-control
action until the system is in the linear range again within the saturation
limit of the actuation force. This is called anti-windup control and should be
present in all PID-control systems, and is often sufficient when long range
sensors are used.
This example demonstrates that linear control, as powerful as it is, also has
its limitations that have to be known and considered in the design of the
control system.
Another limitation, not shown in the figure, would be the limited measure-
ment range of a sensor. When the sensor is out-of-range, the control-loop is
not broken towards the actuator but the controller is "flying blind" once the
sensor signal clips. This means that the actual error may be much larger
than the controller observes, but does not get compensated. With very short
range sensors and long range actuators, like in a close tracking system, one
can reduce the P-gain and increase the D-gain until stability is achieved.
The auto focus of a photo camera is a good example of such a problem, where
often the capture range of the focal sensor is much smaller than the focal
4.5. Limitations of linear feedback control 213

setting range of the lens and outside the range no information is available
about the direction. As a result the camera keeps scanning past the sharp
spot without stopping in time.
As a last example for the limitation of linear control backlash needs to
get some attention as this non-linearity is often present in less advanced
mechatronic system where the cost of direct drive actuators and air bearings
is prohibitive. With a positioning system that consists of a rotary DC
motor with a gearwheel transmission, the backlash and friction of the
transmission disturbs the transfer between the motor and the positioned
mass. This non-linearity limits the possibility to achieve a fast and well
damped feedback controlled system with precise positioning of the mass
because during reversal the transmission is broken by the backlash. This
problem can be partly solved by mounting a tacho-generator directly at the
DC-motor shaft. This sensor provides reliable state information about the
velocity of the motor and can be used for stabilisation of the (high-bandwidth)
inner position/rotation feedback loop. Non-linear errors of the transmission
will however not be measured by the tacho-generator and as a consequence
they will not be reduced in the positioning of the load-mass.
This section of the book should for sure not be misunderstood as linear
control is extremely powerful and important in motion control systems. In
most applications the performance of linear control is however limited by
the boundaries of the physical plant and these have to be considered in the
controller design. As stated above we would like to refer to the references
on linear and non-linear control given in the appendix for further reading,
in order to get an insight how modern non-linear control methods may allow
to achieve a control performance beyond the limits of linear control.
214 Chapter 4. Motion Control

4.6 Conclusions on motion control

In this chapter it was shown that under suitable circumstances, feedback

and feedforward control enable to realise a significant improvement in the
dynamic performance of mechatronic motion systems. Feedback control
allows to modify the system properties by changing the pole locations of the
system, therefore offering to control unstable systems and add robustness
to the feedback controlled system. Feedforward control enables to improve
the performance of motion system for instance by zero-pole cancellation for
reference following, while not being limited by the conditions for stability
and, in general, being simpler and faster than feedback control.
With the combination of both, feedforward and feedback control, also called
two degree of freedom control, the system designer can optimise the control
design including the trade-off between performance and robustness.
It was also demonstrated that active control requires reliable information
of at least one of the system states. Also the behaviour of the plant has
to be known sufficiently well, either by means of system identification or
by modelling. In the following chapters the behaviour of the actuation and
sensing elements are presented to enable modelling of these important
elements of the mechatronic plant. The final chapter about wafer scanners
will further underline the learned theory by illustrating the importance of
both feedforward and feedback control with practical data.
Chapter 5

Electromechanic
actuators

The eigen dynamics of a mechatronic system are not only determined by its
mechanical dynamic performance. Like mentioned before, the incorporated
subsystems such as amplifiers, actuators and sensors have their own impact
on the full system dynamic performance. It requires a thorough understand-
ing of the interaction between the mechanical and electrical engineering
domains regarding energy conversion, in order to design a well balanced
system. This interaction is covered in this chapter on electromechanics.
Although the subject of electromechanics describes the transition from elec-
trical to mechanical engineering domains and vice versa, this chapter will
only cover the part that is related to actuation. The reversed phenomena
will be presented in Chapter 8 on measurement.
After a general introduction, the physics that are related to electromag-
netic forces is covered including the applied Maxwell equations and the
calculations on magnetic fields. Starting with Section 5.2 the behaviour of
linear actuators like moving-coil Lorentz actuators and the stronger but
non-linear reluctance actuators will be presented together with the optimal
combination of both in the hybrid actuator. These basic configurations are
sufficiently representative for the design of most electromagnetic actuators.
With the derived simplified equations, reasonable calculations can be made
on concept designs. Further attention is given to the electrical properties,
that are important for the integration in mechatronic systems, especially
related to the amplifier that has to deliver the current. Electromagnetic
actuators still determine the main-stream actuation principle in precision

215
216 Chapter 5. Electromechanic actuators

mechatronic systems, but piezoelectric actuators are increasingly important

especially in applications that need very large forces with small correc-
tive movements, like in machining tools. But also in very small systems
piezoelectric actuators are used to advantage, because they can create dis-
placements at very high frequencies. For this reason, Section 5.6 presents
an introduction on piezoelectric actuators.

Figure 5.1: Electric motor that was designed by Thomas Alva Edison to drive a
cooling fan. The motor contained no permanent magnets while the
magnetic flux was only created by current. A commutator with sliding
contacts allowed it to be used with direct current electric power that
was his preferred source of electrical energy.
(Courtesy of John Jenkins, www.sparkmuseum.com)
5.1. Electromagnetics 217

5.1 Electromagnetics

An electromagnetic actuator is a special type of electric motor. The term

actuator is reserved for those devices that directly convert electrical energy
into a movement over just a limited linear or rotational range. Before
presenting the theoretical background on electro magnetics first a bit of
history is presented.

5.1.0.4 History on magnetism

The first experiments regarding electricity by the American statesman and

scientist Benjamin Franklin (1706 – 1790) occurred already around the
year 1750. In spite of these early investigations the first real application,
the creation of artificial light, only emerged in the second half of the 19th
century. Also in this period the first electric motors were created based on
the experiments on electromagnetic forces by the Danish physicist Hans
Christian Oersted (1777 – 1851). An example of such an early electric
motor is the one shown in Figure 5.1 designed by Edison around 1898 for
use in a cooling fan. Thomas Alva Edison (1847 – 1931) was an American
inventor and scientist but above all he was a successful businessman who
did everything he could to increase interest for the use of electricity. Next
to his well-known contributions to the application of incandescent light, he
also made other electric appliances like the electric fan of which the electric
motor is shown here.
For a long time the rotating electric motor determined all electrical driven
motion in the world. These motors worked according to a variety of principles
ranging from asynchronous induction to synchronous permanent magnet
types, with or without commutation, that enabled them to run at speeds in
the order of several thousands revolutions per minute. The application of
gearboxes made it possible to convert their usually high rotation speed into
a slower and more controlled motion. Next to a rotating motion also linear
motion was possible with crown wheel and screw spindle mechanisms. In
the course of the 20th century already high levels of linear precision in the
order of ten to hundred micrometres could be achieved, with the use of pre
stressed (ball-screw) mechanisms. These were mainly used in machining
tools, but people started to look for alternatives that did not suffer from
the inherent backlash, friction and play of these mechanical drives, that
limited the use with the “servo” active position control systems that were just
introduced at that time. As discussed in the introduction of this book this
218 Chapter 5. Electromechanic actuators

has resulted in two different directions. The first originated in the machine
tool industry and mostly used the newly discovered piezoelectric properties
of certain materials to use them as actuators. The second direction was
based on positioning of systems without strong external forces and relied
mainly on electromagnetic energy conversion principles.

5.1.1 Maxwell equations

Electro magnetics is the physics area that describes the phenomena, associ-
ated with electric and magnetic fields and their interaction. Although this
book focuses on comprehension, rather than mathematics, it is not possible
to escape from the fact, that the physics in electro magnetics are governed
by stipulated laws and models. These laws in itself can not be understood
by definition, but have to be accepted for reason of their capability to predict
the behaviour of electromagnetic systems. Though this complicates the
matter for people, who want to know why things work as they do, these
laws are taken as a starting point and then the subject is presented as if its
understood what causes these laws to be true.
These basic laws are called the “Maxwell equations” and will be presented
in the following to be used in the further practical implementation.
The Scottish physicist and mathematician James Clerk Maxwell (1831-1879)
formulated his equations partially based on the work of previous scientists,
but the laws got his name because of the way he combined them. The
Maxwell equations are a set of four equations, with the status of physical
laws, stating the relationships between the electric and magnetic fields and
their sources being charge density and current density.
Table 5.1 defines the variables that play a role in this chapter. For a complete
overview also the electrical variables, that were defined in Chapter 2 are
mentioned. Like the previously defined electric field E, the electrical current
density J is a vector field representing the movement of free, or unbound
charges in space. The electric permittivity 0 has also been previously
defined and the electric charge density ρ q represents the amount of unbound
charges in a volume.
The magnetic field B is a vector field. Like with the electric field, it is
graphically represented by the density and direction of magnetic flux lines.
In the design of electromagnetic actuators, the magnitude of the magnetic
field in the direction of interest is called the flux density B. It directly
relates to the quantitative nature of the magnetic flux Φ in that direction,
calculated by integrating the magnetic field over a surface perpendicular to
5.1. Electromagnetics 219

Table 5.1: Physical variables in electromagnetism.

Physical quantity Symbol SI unit

 
Electric field E V/m
 
Electric current density J A/m2
Electric charge q [C]
 
Electric permittivity in vacuum 0 As/Vm
 
Electric charge density ρq C/m3
Magnetic field B [T]
Magnetic flux density B [T]
Magnetic flux Φ [Wb]
 
Magnetic permeability μ Vs/Am
 
Magnetising field H A/m
 
Magnetic field strength H A/m

that direction.
The magnetic permeability μ represents the possibility of a magnetic field
to pass through an object. The word permeability stems from the verb “to
permeate”. The permeability in vacuum is called μ0 and is a reference for
the permeability of other materials.
The magnetising field H and its related magnitude, the magnetic field
strength H , are directly connected to the magnetic field by the magnetic
permeability: B = μH and B = μ H .
The Maxwell equations can be written either in the integral or differential
form. The integral form is more easy for explaining the meaning and will
be used to derive the equations in this chapter. The differential form is
better suited for mathematical modelling. In order to keep this differential
form more easy to write down, the mathematical terms divergence (div) and
rotation (rot) are used. These are respectively the dot (div) and cross (rot)
product between the differential vector nabla (∇) and the vector of interest.
In the three dimensional space ∇ equals:

∂ ∂ ∂
∇= , , (5.1)
∂x ∂ y ∂z
220 Chapter 5. Electromechanic actuators

n̂
dS dl n̂ L
S ddSS S

Figure 5.2: Definitions of the vectors n̂ and dl, surface S and closed loop L that
are used for the Maxwell equations. The left drawing shows surface S ,
that encloses a volume and is used with both Gauss’s laws. The right
drawing shows the closed loop L with its enclosed surface S and is used
with Faraday’s and Ampère’s law.

With the previously defined variables and using the definitions of Figure 5.2,
the first Maxwell equation is written as follows:

• Gauss’s law (electric):

0
q enc
(E · n̂) dS = (5.2)
0
S
ρ
div E = ∇ · E = (5.3)
0

This first law was originally postulated by the German mathematician and
scientist Johann Carl Friedrich Gauss. It states that the surface-integral
of the electrical field over any closed three dimensional surface S , like for
instance a sphere, equals the charge qenc enclosed within the closed surface,
divided by the electric permittivity (0 ). It is directly related to Equation (2.2)
of Chapter 2. With electromagnetism this law is not used, because in
electromagnetic actuators all electrical charges are bound, which means
that they are not free and always in equilibrium with the positive charge of
protons of the wires that are used to carry the current. Nevertheless it is
mentioned for completeness and also to demonstrate the difference between
electric fields and magnetic fields as given in the second Maxwell equation.

• Gauss’s law (magnetic):

0
(B · n̂) dS = 0 (5.4)
S

div B = ∇ · B = 0 (5.5)
5.1. Electromagnetics 221

This second law of Gauss states that the surface-integral of the magnetic
field over a closed surface S is always zero. With any closed surface, the
magnetic flux entering the volume within the closed surface is equal to
the magnetic flux that exits that volume. When represented by flux lines
this can only be true when all flux lines form a closed loop. Gauss’s law
on magnetic fields is based on the observation that magnets always act as
dipoles, a north and south pole where the flux flows internally from south-
to north pole and externally back from north- to south pole. This is the main
difference with Gauss’s law on electric fields where electrical charges can
exist without a monopole counterpart. In case of a single charged particle,
the equivalent electrical field lines originate in the charge and just go to
infinity. Integrating over a surface surrounding that charge would result in a
finite value according to the first law of Maxwell. Insight and understanding
of Gauss’s law on magnetic fields is necessary for calculating the flux in
magnetic circuits.
The third Maxwell equation gives the relation between a change in the
magnetic field and the resulting induced electrical potential difference in a
wire that surrounds that field.

• Faraday’s law:
1 2
d
E · dl = − (B · n̂) dS (5.6)
dt
L S
∂
rot E = ∇ × E = − B (5.7)
∂t

This law was originally postulated by the English chemist and physicist
Michael Faraday (1979 – 1867). It states that the line-integral of the elec-
trical field over a closed loop L equals the change of the flux through the
open surface S bounded by the loop L. In Chapter 2 the electromotive force
Fe was introduced as this integral over the electric field inside a voltage
source. For this reason the third Maxwell equation is the foundation under
the theory that describes the creation of electricity by magnetism. A voltage
source is created inside the windings of a coil by a changing magnetic field
surrounded by that coil. It explains also the phenomenon of self-inductance,
as will be defined in Section 5.3. An important element in this law is the
minus sign that indicates, that the direction of the electric field is opposite
to the vector dl. It explains several properties of electromagnetic actuators
that will be presented in this chapter, including the damping effect that is
observed when the actuator is supplied from a low impedance source.
The conversion from mechanical into electric energy implicitly also leads
222 Chapter 5. Electromechanic actuators

to the reverse effect, which means that by inducing a current through an

electric wire within a magnetic field this wire will experience a force, the
Lorentz force, that will be presented further on in this chapter.
The fourth Maxwell equation gives the principle of the creation of a magnetic
field by an electric current.

• Ampère’s law:
1 2
d
B · dl = μ0 I + 0 μ0 (E · n̂) dS (5.8)
dt
L S
∂
rot B = ∇ × B = μ0 J + 0 μ0 E (5.9)
∂t

This law, originally postulated by Ampère, states that the line-integral of

the magnetic field over a closed loop L is equal to the sum of two terms.
The first term represents the current that flows through the opening of the
loop and the second term represents the change of the electric field over
the surface that is enclosed by the loop. This second term is in reality not
relevant for electromagnetic actuators, again due to the bound character of
the charges as mentioned with Gauss’s law on electric fields. This means
that in a reduced form, without the unbound charges, this law gives the
relation between the magnetic field and the current through a wire. It is
used in modelling magnetic fields induced by a current in a coil.

5.1.2 Magnetism caused by electric current

The Maxwell equations in their differential vectorial notation are applied in

many finite element modelling software packages. The resulting calcula-
tions reach near perfection, depending on the refinement of the chosen grid,
the exactness of the shape of the magnetic parts and the affordable amount
of computing time. For a mechatronic system designer however, it is often
necessary to quickly obtain a first estimation of the properties of an elec-
tromagnetic system, before entering into detailed analysis with dedicated
software. It is often also necessary to be able to recognise artifacts that are
created by the modelling software when the boundary conditions are badly
chosen. For that reason in the following sections the vectorial notations
are often approximated by scalar equations dealing with magnitude only,
assuming the direction of the fields and currents is known. This assumption
is only valid when the directional correlation between all parameters are
either parallel or orthogonal. This first order approximation is often allowed
5.1. Electromagnetics 223

+
B

I
_

Figure 5.3: The magnetic field, generated by a current through a wire is directed
clockwise, when looking in the direction of the current. The magnitude
of the magnetic field decreases proportional with the distance to the
wire.

in initial actuator designs, because of their orthogonal design. When nec-

essary an angular factor (sine/cosine) can be introduced as a second order
correction.
The steps toward usable equations start with the basic relation between
electricity and magnetism derived from Ampère’s law without the term for
the unbound charges:
1
B · dl = μ0 I (5.10)
L

When the loop is a circle with radius x around a wire with a current I as
shown in Figure 5.3, the magnetic field B has a constant flux density B over
the circle and directed along the circle. This means that the left integral
term becomes equal to 2π xB. As a result the flux density becomes equal to:

μ0 I
B ( x) = (5.11)
2π x

This equation is valid under the condition that the surrounding material
is vacuum, for which Ampère’s law was postulated. In words this equation
tells us, that the magnetic flux density is proportional to the current in the
wire and the magnetic permeability in vacuum and inversely proportional to
the distance. This dependence of the magnetic permeability automatically
224 Chapter 5. Electromechanic actuators

Figure 5.4: Three currents running in parallel wires, directed towards the observer,
each generate a magnetic field that is represented by the dashed field
lines. These fields add vectorial, according to the superposition principle
of magnetic fields, resulting in a larger total field that is represented
by the solid red field lines. It shows the straightening of the resulting
field lines at some distance parallel to the orientation of the wires.

results in a change of the flux density when the permeability would be

different.
More directly related to the electric current is the magnetising field H with
its scalar magnitude, the magnetic field strength H :

B ( x) I
H ( x) = = (5.12)
μ0 2π x

This means that the magnetic field strength is only determined by the
current and the distance to the wire. For this reason the vector field H,
corresponding with the magnetic field strength H was named “ magnetising
field” in order to indicate an electric current as the source of the magnetic
field B1 .
In practical electromagnetic systems wires are wound in coils with a certain
number of windings n. As can be seen in Figure 5.4 the magnetic fields
created by a multiple of electrical wires running in parallel, add vectorial,
because of the superposition principle of magnetic fields. As a result, by
winding the wire in a coil, the total field generated by the current increases
per winding added. After integration of the magnetic field of all windings
using Equation (5.11) over the cross section A c inside a coil with a winding
1 The real answer to the philosophical question of “what comes first” or “what causes what”,

the H or the B field, is beyond the needs of this chapter as both fields are equally present and
can be used where appropriate.
5.1. Electromagnetics 225

height hc , the following expression for the flux density (Bw ) and magnetic
field strength ( Hw ) inside the coil of Figure 5.5 would be obtained in vacuum
or air.
μ0 nI Bw,i nI
Bw ≈ =⇒ Hw = ≈ (5.13)
hc μ0 hc

This relation is a rough approximation because of the fact that the magnetic
field is not really homogeneous. The effects of the edges of the coil are
fully neglected, which limits this approximation to coils with a height to
diameter ratio of larger than one. Nevertheless it is useful in a qualitative
manner, as it shows the impact of the height of the coil. The reduction of
the magnetic field as function of the winding height hc , under further equal
conditions regarding current and number of windings, is caused by the
reduced addition of the separate flux levels of each winding at an increased
mutual distance. The diameter is cancelled out of the equation, again under
the condition, that it is approximately the same value or smaller than the
height. This can be reasoned from the fact, that in the mid position the
contributions of the different windings is equal, while closer to one side the
contribution of that side is proportionally increased, where the contribution
of the the other side is decreased. In approximation this means, that the
flux density is constant over the cross section.
The direction of the magnetic flux, relative to the current direction, directly
follows from the definitions for the Maxwell equations from Figure 5.2.
Following the direction of the current, the magnetic flux is rotating clockwise
around the wire as was shown in Figure 5.3. For a coil with multiple
windinngs this means that the direction of the flux inside the coil corresponds
with the current flowing around it in the clockwise direction. The place
where the flux leaves the magnet is defined as the “North pole” and the
other side as the “South pole”. Even though this seems to correspond with
their counterparts of the magnetic field of the planet Earth, the poles of the
earth are reversed as the definition of North pole in a magnet was made
to tell which side of a compass needle points towards the North pole of the
planet Earth. Because opposite poles attract each other this definition has
caused the reversal.

5.1.3 Hopkinson’s law

With the found relations between the current in a wire and the generated
magnetic field it is possible to derive a simplified model, that is useful for a
first order analysis of magnetic systems. Similar to Ohm’s law in electronics,
226 Chapter 5. Electromechanic actuators

Ao

hc  Ai Ai Ac

Southpole B Northpole

n windings

I V
_
+

Figure 5.5: The magnetic field from a current carrying coil consists of the vectorial
addition of the magnetic fields of all windings contributions. This
results in an approximately uniform field inside the coil that is shared
by all windings.

the British physicist and electrical engineer John Hopkinson (1849 – 1898)
stated that a magnetic system can be described as a combination of a source
of magnetism that, together with a magnetic load, determines the magnetic
“current”, the flux Φ.
It was shown in the previous section that an electric current acts as a source
of magnetism. In Hopkinson’s law of magnetics this source is defined as the
multiplication of the current with the number of windings. It is called the
magnetomotive force, familiar to the electromotive force for a voltage source
of electricity:

Fm = nI [A] (5.14)

For the magnetic resistance the term Reluctance (ℜ) is introduced because
of the English word “reluctant” in relation to the unwillingness of a material
to be permeated by a magnetic field. The magnetic reluctance is defined by
means of the following relation:
Φ
ℜ= (5.15)
μA

where Φ equals the length of the path, that the magnetic flux has to follow
and A equals the cross section of the path. This is logical, when considering
that the reluctance ℜ represents the amount of effort that is needed to
5.1. Electromagnetics 227

introduce a magnetic flux in a certain material. A larger surface and higher

values of the permeability will reduce this effort while a longer path length
will increase it.
With these definitions, Hopkinson’s law of magnetics, similar tot its electric
counterpart, is formulated as follows:
Fm nI μ A
Φ= = (5.16)
ℜ Φ

Generally neither the cross section A , the path Φ nor μ is constant over the
entire system, so in practice the reluctance is calculated as a summation
over the path of the flux of the reluctance contributions of each region with
different properties. This means that each reluctance element adds to the
total reluctance just like with their electrical counterparts.
To check this law the induced magnetic field by a current in the coil from
Figure 5.5 in vacuum is determined, using Hopkinson’s law of magnetics.
This results in a flux density Bw inside the windings of the coil of:
Φw Fm nI
Bw = = = (5.17)
Ac Acℜ Acℜ

The reluctance can be modelled to consist of the part ℜ i with length i inside
the coil and the part ℜ o with length o outside the coil:
i o
ℜ = ℜi + ℜ o = + (5.18)
A i μ0 A o μ0

where A i and A o equal the cross sections of the flux path inside and outside
of the coil. The first term of the reluctance inside the coil can be determined
quite straightforward, because of the constant cross section A i = A c and the
height of the coil (i = hc ). The second term of the reluctance outside the
coil is more complicated, as the length o for each flux line will be different
ranging from hc at the exit of the coil to infinite. Fortunately the cross
section A o is also infinite, as it consists of all space outside the coil. As a
result the reluctance of the outside part becomes small in respect to the
inner part. This means, that the reluctance can be approximated by only
taking the inner part:
i
ℜ≈ (5.19)
A c μ0

The flux density and related field strength become:

μ0 nI μ0 nI nI nI
Bw ≈ ≈ =⇒ Hw ≈ =≈ (5.20)
i hc i hc
228 Chapter 5. Electromechanic actuators

Figure 5.6: A living frog floating inside a coil with 20T magnetic flux density that
was demonstrated by Nobel prize winner Andre Geim. The diamagnetic
properties of the water in the frog stabilises its position inside the coil.
(Courtesy of Radboud University Nijmegen)

Which is equal to the previously found relations.

Further it also shows clearly the relation between the magnetomotive force
Fm = nI and the magnetic field strength Hw of the magnetising field Hw .

Fm
Hw ≈ =⇒ Fm ≈ H w  i (5.21)
i

5.1.3.1 Ferromagnetic materials

One method to increase the magnetic flux, without increasing the current, is
the use of a material with a higher permeability. Because μ0 is the reference,
materials other than vacuum have a relative permeability μr in respect to the
permeability in vacuum. This value equals around 1 for non-ferromagnetic
materials like air, glass, many metals and plastics but with ferromagnetic
materials μr can reach values of about 100 for iron and nickel to as large
as 100,000 for materials like Superpermalloy. Most non-ferromagnetic
materials show some very limited effects and are called paramagnetic if
they show a μr of just above 1 or diamagnetic if they show a μr of just
smaller than 1. The latter have the “strange” property of being repelled by
a magnetic field. A famous example is the levitated living frog shown in
Figure 5.6. It was demonstrated by the Dutch/Russian Nobel prize winner
Andre Konstantinovitsj Geim (1965) at the High Field Magnetic laboratory
of the Radboud University in Nijmegen, the Netherlands. The frog is stably
5.1. Electromagnetics 229

ferromagnec yoke

I
I V
+ V +
_ Ay
n windings
_ Ay

Figure 5.7: Magnetic flux generated by a current carrying coil wound around a
ferromagnetic yoke is much larger than without the yoke, because of
the high magnetic permeability of the ferromagnetic material.

floating inside a very strong magnetic field of about 20T, because of the very
small diamagnetic properties of water, which is the main constituent of a
frog.
The μr values of diamagnetic materials are extremely small. For this reason
they are still of little practical use in electro magnetic systems and will
therefore not be further presented in this chapter. On the other hand real
ferromagnetic materials are frequently used for creating strong magnetic
fields.

5.1.4 Coil with ferromagnetic yoke

When a yoke of ferromagnetic material is added to a coil, the flux and flux
density will increase significantly as the ferromagnetic material determines
a low reluctance path for the flux, both inside and outside the coil, as shown
in Figure 5.7. By using Hopkinson’s law of magnetics the flux in the yoke
can be calculated.
The magnetomotive force of the current equals:

Fm = nI (5.22)

With a sufficiently high value of μr the reluctance is mainly determined by

the path through the material with the high permeability. This is mostly
the case with the usual ferromagnetic materials in mechatronic systems.
230 Chapter 5. Electromechanic actuators

Bw Saturaon

Bw P0 P r H w Bw P0 H w

In vacuum or air H w (I )

With ferromagnec core

Saturaon

Figure 5.8: Relation between the flux density Bw and the magnetic field strength
Hw inside a current carrying coil, with and without a ferromagnetic
yoke. Because Hw is proportional to the current in the coil, an increase
in the current will result in an increase of the flux density at a rate
determined by the magnetic permeability. With a ferromagnetic yoke,
the initially strong effect of μr will reduce above a certain level of the
flux density, due to saturation. At even higher levels the curve will
asymptotically continue parallel to the line with μ0 only.

With this condition the reluctance becomes:

y
ℜ= (5.23)
μ0 μr A y

with y and A y being respectively the length and the cross section of the flux
path inside the yoke. Note that in spite of the two dimensional case theses
values are approximated one dimensional where A c is assumed orthogonal
to the flux and  c is the average path length.
The flux Φy in the yoke is equal to the flux Φw in the coil:

Fm A y μ0 μr nI
Φy = Φw = = (5.24)
ℜ y

The corresponding flux density and magnetic field strength inside the coil
equal:
Φy μ0 μr nI Bw nI
Bw = = =⇒ Hw = = (5.25)
Ay y μ0 μr y

5.1.4.1 Magnetisation curve

From the previous equations it can be concluded, that an increase of the

relative permeability μr of the material of the yoke results in a proportional
5.1. Electromagnetics 231

increase of the flux and the flux density. In ferromagnetic materials the
relative permeability μr is not constant but reduces at higher levels of
flux density. This is visualised in the magnetisation curve, a graphical
representation of the flux density Bw as function of the field strength Hw . The
horizontal axis is also proportional with the magnetomotive force and the
current in the magnetising coils, because the field strength is independent
of the permeability.
In Figure 5.8 the magnetisation curve of a ferromagnetic material in the
configuration of Figure 5.7 is compared with the situation without a ferro-
magnetic yoke.
The current is increased from zero starting at the origin. This current causes
a proportional magnetising field with a magnitude equal to the magnetic
field strength Hw . This also corresponds with an increasing magnetic field
of which the magnitude, the flux density, depends on the level of the relative
magnetic permeability of the material.
Above a certain value of the flux density, the material becomes saturated.
This is caused by small magnetic areas in the materials micro-structure,
that are responsible for the relative permeability and have to direct them-
selves to the magnetic field. These areas are called Weiss domains after
the French Physicist Pierre-Ernest Weiss (1865 – 1940), who discovered the
magnetic orientation of these elementary “building blocks”. A Weiss domain
has a typical size of 10−6 to 10−8 m and contains approximately 106 to 109
atoms. The orientation to the external magnetic field takes place inside
the Weiss domain and can be imagined as an elastic effect. Initially the
domains only partly orient themselves but with higher levels of magnetisa-
tion they gradually become completely oriented. As soon as this happens,
the beneficial effect to the magnetic field is reduced. When the current is
increased beyond this level the flux density increases further proportional
to μ0 only so parallel to the line corresponding with the situation without a
ferromagnetic yoke.

5.1.5 Permanent magnets

Ferromagnetic materials also show a certain amount of hysteresis. This

means that after applying and removing a magnetic field, the material
will retain a remnant magnetic property. Like the saturation effect, this is
caused in the material itself, where the Weiss domains need some energy to
change their magnetic orientation. This can be imagined as a friction effect
that limits the possibility for the Weiss domains to change its orientation.
232 Chapter 5. Electromechanic actuators

Bw

Br

Hc

H w (I )
Virgin
curve

Figure 5.9: Hysteresis of a permanent magnet. After the first magnetisation cycle
the material will retain some magnetic energy, resulting in a magnetic
field that is present, even when the magnetizing field caused by the
current is zero. The corresponding flux density Br is called the remnant
flux density and the negative field strength that would be required to
cancel the magnetic field due to the hysteresis, is called the coercive
force Hc .

For many applications this hysteresis is a drawback, as it causes energy loss,

due to the need to supply the energy to change the orientation of the Weiss
domains. On the other hand however, this special property can help creating
magnetic fields, without electrical energy. This is the case in permanent
magnets that are utilised in most electromagnetic actuators for precision
mechatronics.
A permanent magnet is a ferromagnetic material with a high hysteresis
as shown in Figure 5.9 for the same magnetic configuration of Figure 5.7.
To illustrate the hysteresis effect, a not yet magnetised material is used
at the start of the cycle. In that case the magnetisation by a coil with
an increasing current will begin at the origin of the graph and ends until
saturation is achieved, like in the previous example. This initial curve
on the graph follows the virgin curve of the material, because of the not
yet magnetised material. When, after saturation, the current is reduced
back to zero, the magnetic field strength will by definition also be zero.
Due to the hysteresis, however, the magnetic flux density remains positive
until a current is applied in the opposite, negative direction. When this
negative current is increased, the flux density will follow the upper line of
the graph according to the arrows, until the material is saturated in the
opposite magnetisation direction. When the current is reversed again, the
flux density will follow the lower line until saturation is reached in the other
5.1. Electromagnetics 233

Br,n | 1.4 T
NdFeB

AlNiCo Br,a | 1.2 T

Stabilized Bm
D

Br,f | 0.35 T
Ferrites
Bm P0 H m
 H c,f  H c,a
Hm

Figure 5.10: Demagnetisation graph of three different kinds of permanent magnets.

The Alnico material exhibits a curved graph with a very high remnant
flux density, but needs stabilisation. Magnetic material based on
Ferrite (Iron oxides) and the modern very strong material NeFeB
have a demagnetisation line parallel to Bm = μ0 Hm and retain their
magnetic properties also with occasional strong demagnetisation fields
to a level of Hc .

magnetisation direction again. The virgin curve will only be followed once.
For a permanent magnet only the upper left (or lower right) quadrant of
the magnetisation curve is used. Without an external magnetising field,
the flux density of a magnetised permanent magnet will be equal to the
remnant flux density Br . The field strength of the internal magnetising field
is equal to the external field strength, that would be necessary to cancel
the magnetic field. The corresponding magnitude of this field strength is
called the coercive force Hc . In search for ever stronger magnets, material
science has delivered a wealth of combinations of several, sometimes quite
exotic, permanent magnetic materials with very different characteristics.
Examples of the demagnetisation graph of three different permanent magnet
materials are shown in Figure 5.10.
One of the graphs belongs to a magnet material that has been rather popular
in the past. It was an alloy of Cobalt, Nickel and Aluminium with some
other additives, carrying trade names like Alnico and Ticonal. They showed
a large remnant flux density but also a curved graph. This is caused by the
Weiss domains that already start to re-orient themselves to the negative
magnetic field, giving a relative permeability μr > 1. Because of the high
hysteresis in this material this re-orientation of the Weiss domains becomes
234 Chapter 5. Electromechanic actuators

permanent with as a result a reduced remnant flux density. This property

is not preferred when these materials are used in electromagnetic actuators
that work with current carrying coils. An unintended demagnetisation by
the magnetic field of the actuator coil can rather easily lead to a permanent
demagnetisation, resulting in a less strong actuator. In the figure this is
shown with the dashed line starting at point D. As soon as the magnet is
demagnetised until point D, its flux density will follow the dashed line, when
the external demagnetising field is reduced again. As a consequence these
types of magnets need to be brought to this level on purpose, when external
demagnetising fields are expected. This stabilising action is especially
necessary with actuators as used in mechatronic positioning systems, and
it reduces the potential of these materials to a large extent. It also requires
that the magnets are magnetised after mounting in the magnetic circuit,
because the high reluctance without a closing magnetic circuit is equivalent
to the presence of a high external demagnetising field.
For this reason, ideal permanent magnet materials would need to contain
Weiss domains that orient themselves only at magnetic fields with a higher
magnitude than those that are present in the magnetic circuit that they
are applied in. Fortunately these materials exist, with examples like the
frequently used ferrites with iron oxides as base component and the more
advanced composites of Samarium-Cobalt and Neodymium-Iron-Boron (Nd-
FeB). In these materials the turning of the Weiss domains starts at a field
strength larger than Hc , which means that their relative permeability is
equal to one at a field strength below Hc . As a consequence, the demagneti-
sation curve runs parallel to Bm = μ0 Hm .
These magnets behave like an air coil without the need to supply electric
energy, because the relative permeability of these materials is equal to one.
From the fact that an external field strength Hc is required to compensate
the internal field strength, the following relation can be written for the
equivalent magnetomotive force of an ideal permanent magnet:
B r m
Fm = Hc m = (5.26)
μ0

In this equation the length m of the permanent magnet is equal to the length
i of the flux path inside the equivalent coil as noted in Equation (5.21).

To illustrate the strength of permanent magnets it is interesting to add some

numbers to this equation. For example a permanent magnet with a B r of 1
T and a length of 10 mm is equivalent to a coil with a magnetomotive force
of Fm = 8 · 103 , which is 80 A in 100 windings. This clearly underlines the
5.1. Electromagnetics 235

value of permanent magnets as this current would require a lot of electrical

power to generate this magnetic field by a small coil only.

5.1.5.1 Thermal behaviour and Curie temperature

The most important drawback of a permanent magnet, when applied in

reliable and predictable mechatronic systems, is the temperature depen-
dency of its demagnetisation graph. This behaviour is caused by the Weiss
domains that lose their hysteresis at higher temperatures. The effect is di-
rectly related to the Curie temperature Tc , named after the French physicist
and Nobel prize winner Pierre Curie (1859 – 1906), for his investigations
on magnetism and piezoelectricity. Above the Curie temperature all Weiss
domains become completely disoriented and the magnetic properties are
lost. While some permanent magnet materials, like the expensive SmCo
alloy, have a rather high Curie temperature with ≈ 700 ◦ C, more affordable
modern materials like NdFeB alloy have a relatively low Curie temperature
of around 300 ◦ C, which limits their use to temperatures below approxi-
mately 100 ◦ C.
Even at lower temperatures the magnetic properties of NdFeB alloy decrease
with a factor of around 1 · 10−3 per ◦ C. Fortunately at these moderate levels
the demagnetisation is still reversible and by measuring the temperature,
the effect on the magnetism can be compensated.

5.1.6 Creating a magnetic field in an air-gap

In the previous section, the method to create a permanent magnet was

explained, starting with a continuous yoke of high-hysteresis ferromagnetic
material that was inserted in a coil, like shown in Figure 5.7. After mag-
netising that yoke, the obtained permanent magnet material would still
be rather useless as the magnetic field would remain inside the yoke with
a value B r of the flux density. When the permanent magnet is to be used
to create a magnetic field in an air-gap, it is necessary to create an empty
space, an air-gap inside the magnet that the magnetic flux has to cross.
This can be imagined by cutting the yoke open and aligning it to a straight
magnet. The flux generated inside the permanent magnet will exit at the
cutting edges, the pole pieces to find a return path through the open space
as shown in Figure 5.11. It is obvious that this external return path for
the flux represents a considerable reluctance and as a consequence the flux
will be lower than was the case when the magnet was still enclosed in the
236 Chapter 5. Electromechanic actuators

Southpole S B N Northpole

Figure 5.11: A piece of permanent magnet material creates an external magnetic

field comparable with the field of a current carrying coil, as shown in
Figure 5.5.

continuous yoke. This means that a de-magnetising field in the opposite

direction is created by the external path that forces the flux density down,
according to the demagnetisation graph.
This is easier understood by using Hopkinson’s law of magnetic fields again,
where the magnetomotive force of the permanent magnet creates a flux that
is proportional to the sum of the internal and external reluctance. When the
external reluctance is zero, like when enclosed in the continuous yoke, the
flux is maximum at B r and only determined by the internal reluctance. As
soon as the magnet is opened with two pole pieces, the external reluctance
reduces the flux as mentioned before.
This all implies, that by balancing the reluctance of the external flux path
with the internal reluctance of the permanent magnet an optimal use of the
permanent magnet material can be achieved with a suitable flux density
outside the permanent magnet.
As an example to create an idea about the principle, the magnetic field in an
air-gap will be calculated in a roughly approximating way to be able to use it
in a Lorentz actuator. This quite standard configuration to create a magnetic
field in an air-gap consists of a permanent magnet and a ferromagnetic yoke
that is used to concentrate the magnetic field to the air-gap as schematically
shown in Figure 5.12.
The equivalent magnetomotive force of the permanent magnet is used to
5.1. Electromagnetics 237

Ferromagnec
Ferromagnec (iron) yoke
core/yoke

Ay
Am Ag lg Ay
S N

Am Ag
Figure 5.12: Magnetic field in the air-gap caused by a permanent magnet when
neglecting flux loss. The reluctance of the magnetic flux path is a
combination of the reluctance of the permanent magnet, the air-gap
and the ferromagnetic yoke.

calculate the flux in the circuit:

F B r m
Φm = = (5.27)
ℜt μ0 ℜt

In this configuration the magnetic reluctance of the complete magnetic

path is given by a series of three reluctances that have to be added, the
internal reluctance of the permanent magnet itself, the reluctance of the
ferromagnetic yoke and the reluctance of the air-gap.
m y g
ℜt = ℜm + ℜy + ℜg = + + (5.28)
A m μ0 A y μ0 μr A g μ0

By combining this with Equation (5.27) the following expression for the
magnetic flux of the permanent magnet is obtained:
Br A m
Φm = (5.29)
A m y A m g
1+ +
A y m μr A g m

In many practical situations μr  c A m/A y m which means that the equation

can be reduced to:
Br A m
Φm = (5.30)
A m g
1+
A g m

The flux in the air-gap equals only a part of the total flux of the magnet.
The other part is the leakage flux or stray flux, that follows a path outside
238 Chapter 5. Electromechanic actuators

Air -
S N gap

Figure 5.13: Stray flux reduces the useful flux in the air-gap. With FEM modelling
software this is clearly made visible in a comparable configuration, as
used in Figure 5.12. By counting the flux lines inside and outside the
gap the loss factor λ appears to be about 0.45. More than half of the
total flux is lost.

the useful area of the air-gap. This stray flux with its corresponding loss
factor λ is caused by the fact, that all materials conduct magnetism, even
vacuum and air. Magnetic insulation is not possible and this means that
the magnetic field will seek the lowest energy situation by using all space
available. The magnitude of this effect is shown in Figure 5.13 where in the
example configuration a magnet with a Br of 1 T creates a flux density in the
air-gap of only around 0.3 T and the loss factor λ appears to be approximately
0.45 as can be checked by counting the field lines.
For this reason in this roughly approximating calculation the following
relation for the flux Φ g in the air-gap is allowed:

Φg = Bg A g = λΦm = λBm A m (5.31)

In practice the value for λ will be between 0.25 and 0.75 depending on the
configuration. Next to the shown example with λ ≈ 0.45 two examples with
a lower and higher loss factor will be shown as soon as the mathematical
analysis is finished.
When Equation (5.30) is combined with Equation (5.31) the magnetic flux
density in the air-gap can be calculated:
Φg λΦM Am λB r λB r
Bg = = = = (5.32)
Ag Ag Ag A m g Ag g
1+ +
A g m A m m

This relation reads as follows, by first noting that the flux density in the air-
5.1. Electromagnetics 239

gap is proportional to the maximum flux density of the permanent magnet.

The flux density in the air-gap is reduced by the loss factor λ and further
determined by the ratio between the length and cross section of the magnet
and the air-gap. The longer the air-gap with respect to the magnet, the less
flux density will be achieved and a larger surface of the air-gap in respect
to the magnet will also give a lower flux density. In the situation that the
cross sections of the magnet and air-gap are equal, like in the example, this
equation reduces to:
λB r
Bg = (5.33)
g
1+
m
In the example m = 3g and with Br = 1 the resulting flux density inside the
air-gap would be equal to approximately 0.3 Tesla. More than one decimal
is not significant in this very approximating calculation.

5.1.6.1 Optimal use of a permanent magnet

In Figure 5.14 it is shown that a permanent magnet with a straight demag-

netisation graph is most efficiently used when the external field strength
equals Hc /2 and hence the flux density equals Br /2. This is based on the
understanding that the specific magnetic energy per unit of volume is pro-
portional to the product of flux density and field strength. This can be
checked by using the units:
A2 Vs energy
BH = μ H 2 =⇒ 2
· = (5.34)
m Am volume
In the past it was customary, when designing a magnetic circuit with perma-
nent magnets, to work with these values, as it would result in an optimum
use of the expensive magnet materials. The discovery of the relatively af-
fordable Neodymium based magnets created the possibility to realise very
strong magnetic fields in less optimal designs from a magnet material usage
point of view. These configurations are applied in extreme high performance
positioning systems where the cost of the magnets is less important. In the
following part the more optimal configurations will be presented, as they
are a good starting point for gaining further knowledge in this field.

5.1.6.2 Flat magnets to reduce stray flux

In order to reduce the stray flux and optimise the use of expensive magnet
materials it is often preferred to use a different configuration than the
240 Chapter 5. Electromechanic actuators

B·H = energy per volume

Br

Bm

1
2 Br

Hc  12 Hc
Hm

Figure 5.14: The magnetic energy of B · H of a permanent magnet with a straight

demagnetisation graph is maximum at half the value of Br and Hc .
When the magnetic circuit is dimensioned such that the permanent
magnet is utilised at its maximum magnetic energy, the volume and
cost of the magnetic material is minimised.

one shown in Figure 5.12 by choosing the cross section of the magnets
large in respect to their length. It also appears to be useful to not include a
ferromagnetic part between the magnet and the air-gap as the ferromagnetic
part would determine a low reluctance path to the sides of the air-gap even
for the flux coming from the inner part of the magnet. This is shown in
Figure 5.15, where at the right side the effect of ferromagnetic pole pieces
is shown and at the left side the optimal configuration with permanent
magnets directly adjacent to the air-gap.
In this configuration the path length though air is often equal to the path
length through the magnets and the flux density in the air-gap using Equa-
tion (5.33) becomes:
λB r
Bg = (5.35)
2
When applying magnets with a Br of 1 T this second example shows a flux
density in the air-gap of approximately 0.4 T. The flux density in the perma-
nent magnet can be determined by using Equation (5.31) with a loss factor
λ = 0.75, as found by counting the magnetic field lines. Because A m = A g
this calculation results in a flux density of the magnet Bm of approximately
Br /2, which means that the permanent magnet is used at its optimal energy
point.
5.1. Electromagnetics 241

N N
S S

Air-gap Air-gap
N N
S S

Figure 5.15: With flat permanent magnets, directly interfacing the air-gap, the
stray flux is kept at a minimum. Often ferromagnetic pole pieces
are suggested to spread the magnetic field more uniform, however,
as shown in the right figure, this pole piece significantly increases
the stray flux, due to the low reluctance path to the edges of the
magnet. Even with a smaller air-gap the loss factor λ reduces from
approximately 0.75 to 0.6 as can be checked by counting the flux lines.

Air-gap

N N
S S

Figure 5.16: Concentration of flux by ferromagnetic pole pieces. This rotation sym-
metric structure with a large but inexpensive magnet, as commonly
applied in loudspeakers, creates a strong magnetic field directed in
the radial direction in the air-gap. A large part of the magnetic flux is
however lost outside the air-gap.

5.1.6.3 Low cost loudspeaker magnet configuration

When a very high flux density has to be created in a relatively small air-gap,
it is possible to concentrate the flux by using ferromagnetic pole pieces that
have a smaller surface at the air-gap than at the surface of the magnet. Even
though this creates an increased loss it is often used with very inexpensive
242 Chapter 5. Electromechanic actuators

magnets like the ferrites that are usually applied in loudspeakers. The
effect on the flux is shown in Figure 5.16.
With Equation (5.32) this concentrating effect can be calculated, although
due to the high amount of stray flux (λ < 0.3), the magnet needs to be consid-
erably larger than would be the case with the flat magnets of Figure 5.15.
Still in this example an inexpensive magnet with a Br of 0,4 T creates a
flux density in the air-gap of approximately 0.7 T , which is fully acceptable
when low cost is the primary specification and size is no issue.
In the next section the magnetic field in the air-gap is used to create a
Lorentz actuator by inserting a current conducting wire in the magnetic
field.
5.2. Lorentz actuator 243

5.2 Lorentz actuator

Lorentz actuators are predominantly applied in high precision positioning

systems because of their inherent low mechanical stiffness between the
stationary and the moving part. Also the linear relation between current
and force combined with the favourable dynamic properties are important
factors. The low stiffness reduces the amount of external motion that is
transferred from the support structure through the actuator to the moving
part (transmissibility!). These movements can be caused by vibrations
of surrounding machines but also by the reaction forces of the actuator
itself, exciting resonances in the support structure. As will be shown later,
Lorentz actuators have also some drawbacks like the relatively modest
force to current ratio which limits the maximum acceleration levels and the
achievable range of motion or “stroke”. As the name implies, the Lorentz
actuator is based on the Lorentz force only.

5.2.1 Lorentz force

The Dutch physicist and Nobel prize winner Hendrik Antoon Lorentz (1853 –
1928) formulated the Lorentz force as a completion to the Maxwell equations.
The law of Faraday describes the effect of a changing magnetic field on
electrical charges hence generating electricity from kinetic energy. Based
on energy conservation laws creating electrical energy from motion is fully
complementary to creating motion energy from electrical energy so the laws
of Lorentz and Faraday are strongly related.
In vectorial notation the formulation of Lorentz describes the force on a
moving charged particle as:

F = q(E + v × B) (5.36)

with v [m/s] equals the instantaneous velocity of the particle. The first
part of the Equation qE is the electrostatic force and the second part is the
electromagnetic force. This second term is used in electromagnetic actuators.
Next to the force on a moving particle it equally represents the force on a
current flowing through a wire with length w [m], inserted in the magnetic
field. For this situation the moving charge equals the current times the
length, qv = w I, and with this relation the electromagnetic Lorentz force is
equal to:

F = w I × B (5.37)
244 Chapter 5. Electromechanic actuators

&
I
& &
F B

Figure 5.17: Determining the direction of the Lorentz force with the corkscrew rule.
When the corkscrew is rotated right handed, from the direction of
the positive current to the direction of the magnetic field (arrow), the
movement of the point of the corkscrew determines the direction of
the force.

For the magnetic force on a wire at an angle α relative to the direction of a

magnetic field with flux density B, carrying a current I , this relation leads
to the scalar notation of the Lorentz force of electromagnetic actuators of
which the magnitude is given by:

F = BI w sin α (5.38)

The direction of this force is orthogonal to the plane that is determined

by the direction of the magnetic field and the current, due to the “cross
product” in the vectorial Lorentz equation. This rule can be remembered as
the right hand or corkscrew rule that states that the positive force direction
is found when rotating a corkscrew from the positive current direction onto
the direction of the magnetic field as shown in Figure 5.17. Of course for a
real mechanical engineer any normal right turning screw will also suit the
purpose, but the corkscrew is more easy to remember.
In most practical cases the Lorentz force must be maximised which means
that sin α is kept as much as possible equal to one. This means that the
simplified equation becomes equal to:

F = BI w (5.39)

And with multiple windings the Lorentz force becomes:

F = BI nw = BI w,t (5.40)

where w,t equals the total length of the wire inserted in the magnetic field.
This equation is widely used as a general equation for linear actuators based
5.2. Lorentz actuator 245

Permanent Magnets
Ferromagnec yoke
F

Magnec Field

Flat wound coil

F
A

Detail A

Figure 5.18: Basic flat type Lorentz actuator. The Force F is proportional to the
current and the amount of winding length  of the wire in the coil
inserted in the magnetic field B. Due to the large amount of coil
outside the air-gap this is not a very efficient configuration. Detail
A shows the forces acting on a wire segment at the centre of the coil,
clearly indicating the inefficiency.

on the Lorentz principle. It clearly shows that the force only depends on
the current, the total wire length and the magnetic flux density. Because
there is no direct relation with the position, the actuator ideally has zero
stiffness. This is the main reasons why a Lorentz actuator is preferred in
precision positioning systems as it avoids transmissibility of vibrations from
the stationary part to the moving part. Later it will be shown that this
is the ideal situation. In reality at higher current levels some stiffness is
observed due to non-linearity and position dependency of the B value.
Figure 5.18 shows a basic flat type Lorentz actuator configuration for il-
lustrating the principle only as it is not the best design possible. It uses
the flat high efficiency permanent magnet configuration as described in the
previous section and the coil is flat wound and inserted partly in the mag-
netic air-gap. The force of this actuator can in principle be calculated using
Equation (5.40). The permanent magnet flux is directed perpendicular to
the current. When the curvature of the wires is neglected, the force equals
approximately BI w,t , where w,t is the part of the total length of the wire
inside the air-gap. The inner windings contribute little to the force, because
of the curvature. This is shown in Detail A that indicates the spread force
contributions along a wire segment at the centre. Only the component of
246 Chapter 5. Electromechanic actuators

Permanent Magnets
Ferromagnec yoke
F=0

Magnec Field

Flat wound coil

Figure 5.19: Non operating actuator with mechanically connected magnets and
coils. The force is created between the magnetic flux and the current
and not between the magnet-coil combination and the iron part of the
system.

the force pointing to the right contributes to the total force.

A drawback of calculating the force with the simplified equation is however
that it sometimes leads to mistakes. To illustrate this, a real life example of
a non-operating Lorentz actuator is shown in Figure 5.19. Originally the
idea was meant to increase the stroke by extending the iron parts and move
the magnets and the coil together. Unfortunately, this idea does not work
at all, even though the current carrying coil is inserted in an area with a
significant magnetic field. The reason of the lack of observed force is, that
the force acts between the current and the permanent magnet flux. As a
consequence, no external force will occur when the coil and the magnets
are mechanically connected. To avoid these errors, it is preferable to work
with another more generally suitable relation that is based on the change
of flux over the windings as function of the movement of the coil relative
to the magnetic field. When one side of a closed winding is inserted in
a magnetic field and it moves with a value d x relative to the permanent
magnetic system in the direction of the force, the flux Φw inside the winding
will change according to the flux density and the length w of the inserted
part of the winding:
dΦ w
dΦw = Bw d x =⇒ = B w (5.41)
dx
The Lorentz force for each winding can now also be written as:
dΦw
Fw = I (5.42)
dx
And for multiple windings the relation becomes:
dΦw
F = nI (5.43)
dx
5.2. Lorentz actuator 247

Permanent Magnets Coil

Ferromagnec yoke

Useless windings

Round coil

Race track

Figure 5.20: Flat type Lorentz actuator with improved Force to current ratio. By
increasing the width and adding two additional magnets like shown
in the upper drawing, the force is doubled at the same power loss in
the windings. Using an oval shape of the coil like a “race track” as
shown in the lower drawing, will improve the force to power ratio even
more.

Using this equation it is obvious that the actuator of Figure 5.19 is not
working, because with that configuration there is no change of the flux
through the coil as function of the displacement ( dΦw / d x = 0).

5.2.2 Improving the force of a Lorentz actuator

In the previous section it was demonstrated that the windings of the Lorentz
actuator of Figure 5.18 are not very well utilised. While the resistive power
248 Chapter 5. Electromechanic actuators

loss (Pl = I 2 R ) of an actuator is directly proportional with the length of the

windings, the actuator should be designed with a maximum coverage of
the coil by magnetic flux density and a maximum contribution of the force
of each part of the windings in the right direction. A first improvement
can be achieved by using both sides of the coil. At first sight it might be
useful, to extend the permanent magnet, in order to cover the entire coil.
Unfortunately however all forces would direct to the centre of the coil and
cancel each other out. This corresponds with the fact that in that situation
dΦw/d x = 0. This problem can be solved by reversing the magnetic field of
the right side of the coil where the current runs in the opposite direction of
the current in the left side. By adding two permanent magnets, magnetised
in the opposite direction of the first set of magnets and widening the magnet
system, as shown in the upper drawing in Figure 5.20 the coil is almost
completely used. As a result the useful force is doubled in respect to the
first example, with the same power loss in the coil.
A further improvement can be achieved by changing the coil into an oval
shape as shown in the lower drawing in Figure 5.20. This is known as a race
track coil and this configuration has become the “de facto” standard in flat
Lorentz actuators for fast and high precision positioning systems, because
of the optimal use of the windings.

5.2.3 The moving-coil loudspeaker actuator

The most widely applied version of the Lorentz actuator is the moving-coil
loudspeaker actuator. This type is a fully rotation symmetric configuration
with a round coil that is inserted in a magnetic field that is directed radial
towards the centre of the structure, like shown in Figure 5.16. Even though
the magnetic field in the air-gap is three dimensional the orientation is
orthogonal to the current of the windings. For this reason the approximating
scalar expression for the force can be used also in this example. A version
with permanent magnets directly adjacent to the air-gap with the coil is
shown in Figure 5.21. In this configuration the coil is completely surrounded
by the permanent magnet, resulting in a maximum efficiency.

5.2.4 Position dependency of the Lorentz force

A Lorentz actuator, like the moving-coil loudspeaker type, shows a force

to current ratio (F/I ) that is dependent on the position of the coil in the
air-gap. This effect is shown in the red line of the graph in Figure 5.22
5.2. Lorentz actuator 249

A Permanent Magnets Ferromagnec yoke

Coil
A A-A

Figure 5.21: moving-coil loudspeaker motor with permanent magnets directly ad-
jacent to the air-gap. In the middle position the coil is fully inserted
in the magnetic field giving a maximum force to power ratio.

and is caused by the fact, that off the centre, the coil is not completely
surrounded by the same magnitude of the magnetic field. When used in
a closed loop positioning system this implies a change in the control loop
amplification (gain) and a non-linear positive or negative stiffness depending
on the current level and the position relative to the centre position x0 .
This can be explained as follows. Assume the coil carries a current with
a direction that corresponds to a force in the positive x direction. When
the coil is pushed by an external force F x in the negative x direction from
outside into the air-gap, at first the force needed to move the coil increases.
This is due to the increased part of the height hc,g of the coil inside the
air-gap and continues until at x0 the maximum force is reached. As a
consequence, this trajectory showed a positive stiffness, dF/d x ≥ 0. When
the coil is moved further in the negative x direction the force decreases
again, which corresponds with a negative stiffness, dF/d x ≤ 0.
250 Chapter 5. Electromechanic actuators

hc (u) hg  hc hc (u)
hc hg (o) hg (o)

F/I(x)
hc,g(x)
hg

x0 x

x x
hc,g

a: under-hung (u) b: over-hung (o)

Figure 5.22: A Lorentz actuator has a limited stroke, determined by the dimensions
of the air-gap and the coil. When the coil is only partly inserted with
an effective height hc,g that is inserted in the air-gap, the Force to
current ratio (F/I ) is reduced. By choosing different values for the
height hc of the coil and the height hg of the air-gap, the Force to
current ratio as function of the position x can be made more constant
over a certain range of x. The gradual decrease of the magnetic field
(stray flux) at the edges of the air-gap softens the transitions.

5.2.4.1 Over-hung and under-hung coil

In order to reduce the position dependency of a Lorentz actuator, in practice

the height hg of the air-gap is chosen different from the height hc of the
coil. When the coil height exceeds, hangs over, the height of the gap, this
configuration is called an over-hung Lorentz actuator.
Its advantage is the optimal use of most of the permanent magnetic flux
while also the position dependency is more evenly smoothed’out. The draw-
back of an over-hung actuator is the large number of coil windings, that
is not utilised effectively. In the situation, where the coil is the moving
part, this means that the moving mass is larger than without an over-hung
situation. It is a typical choice for loudspeakers where the cost of the magnet
5.2. Lorentz actuator 251

outweighs the value of a high efficiency of the loudspeaker. The amount of

overhang is chosen depending on the allowed non-linearity related distortion.
For non- critical applications like loudspeakers for cars, “Public-address”
in stadiums and amplification of instruments in pop-music, no overhang is
applied, giving a maximum efficiency at increased distortion. For high-end
sound reproduction, often a large over hang is chosen to reduce distortion
at a sacrifice on efficiency. This is the reason why amplifiers for high-end
home equipment needs to have a relatively high output power.
The configuration where the height of the gap exceeds the height of the coil
is called an under-hung Lorentz actuator. The benefits and drawbacks are
just reversed. It makes best use of the coil with a resulting reduced mass
but at a relative high cost of the permanent magnetic part. The position
dependency of the force to current ratio is better than in the over-hung
configuration, when the coil is still completely inside the air-gap, but it
worsens more rapidly as soon as the coil reaches the outer range of the
air-gap, due to its small size. An under-hung Lorentz actuator is chosen
mainly in loudspeakers, when a long stroke is not needed and low moving
mass and distortion is the primary goal. This is especially the case for mid-
to high-frequency loudspeakers, the squawkers and tweeters.

5.2.5 Electronic commutation

The only method to increase the range of a Lorentz actuator, while pre-
serving a constant force to current ratio, is by combining several coils and
magnets into a electronically commutated actuator. The principle is shown
in Figure 5.23. A set of an even number of alternately magnetised perma-
nent magnets creates an alternating field in the air-gap. The coil is divided
in three equal separate sections. In Figure 5.22 it was shown that the force
factor B of each coil segment is dependent on the position. By choosing an
optimal height relative to the permanent magnets, the force factor of each
coil segment becomes an almost ideal sinusoidal function of the position.
When the centres of the segments have a distance dc equal to one third
of the total height h4 of four magnets the spatial sinusoidal forces have a
spatial phase difference of 120◦ .
This is a three-phase actuator configuration. The relation between the sinu-
soidal functions is completely comparable to three-phase power distribution
networks where three temporal sinusoidal voltages, called R, S and T with a
phase difference of 120◦ , are used to transport electrical energy. One reason
to use this method is the fact that the three sinusoidal functions add to zero
252 Chapter 5. Electromechanic actuators

h4

R S T
F

dc
2

Q5
 [N]

JQ5

>JQ5

>
R S T
?


Figure 5.23: The range of a Lorentz actuator can be extended by electronic com-
mutation, using more coils and magnets. Each coil segment has a
force factor that changes sinusoidal as function of the position with a
120◦ spatial phase difference with the other coil segments. By chang-
ing the direction of the current at the zero force positions of each coil,
an almost constant force is obtained. The values are modelled for a
current per coil of 1 A and a value of (B)max = 1.

at any position in the period, as can be proven by applying trigonometry on

the following function.
 
2 4
sin( x) + sin x − π + sin x − π = (5.44)
3 3
 
2 2
sin( x) + sin( x) cos π − sin π cos( x)+
3 3
 
4 4
+ sin( x) cos π − sin π cos( x) = 0
3 3

With power distribution this means that the average voltage of three wires
carrying these voltages is zero at any time. This prevents electromagnetic
5.2. Lorentz actuator 253

radiation from the power lines. Another important value is the possibility
to directly drive rotating AC inductance motors for traction and machining
centres. In mechatronic positioning systems it appears to be also very useful
in long range linear motors.
When all coil segments in the configuration of Figure 5.23 would get the
same DC current the resulting force would be zero. This might seem useless,
but when the current of each coil is reversed at the exact position ,where
the force in that segment is zero, the forces of all three segments add to
an almost constant value of two times the maximum value of BI  per coil
section, with a variation as indicated in the figure.
In rotating DC motors this current commutation process is mostly done by
means of mechanical sliding contacts, although presently this commutation
is increasingly achieved electronically in order to avoid sparking at the
commutation points and wear of the sliding contacts.
The application of electronic commutation can also help solving the problem
of the not perfect continuous force by exchanging the hard-switching by a
more continuous current change.

5.2.5.1 three-phase electronic control

By using a special three-phase amplifier as will be presented in Section 6.3

of Chapter 6 the current to each coil segment can be controlled in such
a way that the magnitude changes with the same sinusoidal function of
the position as the force factor. The resulting force then becomes equal
to the multiplication of the amplitude Î of the current, the amplitude of
the force factor (B)max and the sine function squared. Figure 5.24 shows
the signals and forces when the current and (B)max -value are equal to the
hard-switched example of Figure 5.23.
The squared sinuses produce a three-phase combination at double the spa-
tial frequency of the original sine wave and these sinuses add to three times
their average value. For the example with 1 A current and a maximum
(B)max value of one, the resulting force would be constant at a level of
3 · 0.5 = 1.5.
Even though this force level is slightly below the level with the hard-switching
commutation of Figure 5.23, the method of three-phase electronic commu-
tation is preferred, because of the constant force and the absence of fast
transients in the currents and forces that might introduce electromagnetic
interference and excite uncontrollable eigenmodes in the mechanism.
254 Chapter 5. Electromechanic actuators

 R S T
JQ8
JQ6
JQ4
JQ2

 [N]
J
>JQ2
>JQ4
>JQ6
>JQ8
>

?


JQ8
JQ6
JQ4
JQ2
I [A]
B[N\A]

J
>JQ2
>JQ4
>JQ6
>JQ8 R S T
>

?


Figure 5.24: three-phase commutation with a sinusoidal control of the currents in

each coil segment in phase with their (B) factor results in a force per
segment with a spatial frequency that is double the original spatial
frequency of the coils around an average value that equals half the
(BI )max -value. The resulting total force of the three coil segments
is the sum of the average values of the force in each segment and is
independent of the position.

5.2.6 Figure of merit of a Lorentz actuator

Before presenting the non-linear behaviour of electromagnetic actuators a

short story is given on how at Philips Electronics, the design of a Lorentz
actuator was continuously improved by working with a figure of merit.
During the early years of developing actuators for the CD drive like presented
in Chapter 1 many different design concepts were investigated. To arrive at
a constant current to force ratio [N/A] over a stroke of about one millimetre
one could use a under-hung “short coil - long magnet” combination or an
over-hung “long coil - short magnet”. Also designs using the coil as mover
and designs with moving magnets were made. When actuating in only one
or two directions the remaining degrees of freedom must be restrained by
guiding elements. Both sliding elements and flexible designs are possible.
Similarly one could use translation or rotation as primary kinematic solution
for the motion. Looking at all these options, which have all been tried, it
became clear that decision making was not easy. To settle this a criterion
5.2. Lorentz actuator 255

was derived that was successfully used for that purpose, the figure of merit.
The basic function of the actuator is to generate a certain acceleration of the
optical element. Driving with higher currents can deliver more acceleration
but the penalty is a higher dissipation. The final “Figure of Merit” was
called: “G’s per square root of Watt’s”. How many G’s one would get per watt
of power. The higher the figure the better. If the figure of merit is called Q m
the relation is:
G
Qm =  (5.45)
W att
Although a bit unconventional, unscientific and at least not according to SI
units this number proved to be quite suitable and has been in use for about
10 years. When analysing the figure, using the actuator design aspects, the
reason becomes clear. As a first step, the figure of merit is written as follows
according to SI rules:
F
a m F
Qm =  =  =  (5.46)
P P m P
With a as acceleration and P as power. In such a mechanism this power
almost completely consists of the resistive power loss (P = Pl ), as the move-
ments that are made are very small. This relation already makes clear
that a small mass is favourable. A large force at low dissipation also leads
to a higher score. One step further F can be eliminated, using the simple
Lorentz actuator formula F = BI w,t with w,t being the total length of the
wires inside the magnetic field of the permanent magnet.
BI w,t
Qm = # (5.47)
m Pl

The second step is related to geometric aspects. When the part of the length
of each winding in the air-gap (active part) is called w,a and the length
of the part of each winding outside the air-gap (passive part) w,p and the
number of windings is n, it is allowed to say:
BI nw,a
Qm = # (5.48)
m Pl

Next the dissipation can be eliminated by using Pl = I 2 R . Upon substitu-

tion it is shown, that the current is present both in the nominator and
denominator so it cancels out:
Bnw,a
Qm =  (5.49)
m R
256 Chapter 5. Electromechanic actuators

The resistance R is the total resistance, including the active part and the
passive part of each winding. With n windings of a resistivity ρ r and a cross
section per wire A w , the resistance of the coil becomes:
( )
nρ r w,a + w,p
R= (5.50)
Aw

With γ being the fill factor of the windings and A c,w the cross section of the
coil windings ( bc hc ), the cross section per wire becomes:
γ A c,w
Aw = (5.51)
n
With this value the resistance equals:
( )
n2 ρ r w,a + w,p
R= (5.52)
γ A c,w

The active part of the volume of the coil within the magnetic field Vc,a
is equal to A c,w w,a . When substituted together with Equation (5.52) in
Equation (5.49), this leads with a bit algebra to the following expression for
the figure of merit:
 
B γVc,a w,a
Qm = ( ) (5.53)
m ρr w,a + w,p

It is demonstrated, that the actuator figure, “G’s per square root of Watt’s”
has been changed into clear understandable and far more logical engineering
choices. By maximising B, γ, Vc,a and minimising m, ρ r and w,p relative to
w,a , it became possible to realise an optimal design at such a low price,
that this part of the CD player changed from a costly part as in the first CD
players to the mass produced cheap commodity it is today.
5.3. Reluctance actuator 257

5.3 Reluctance actuator

The force of an electromagnetic actuator is not always linearly related to the

current like in an ideal Lorentz actuator. In general, at increased current
levels, the force will change more disproportional due to different causes.
First of all the magnetic field of the coil adds to the magnetic field of the
permanent magnets. As a result the iron parts in the magnetic assembly
can get saturated, with as a consequence an increased reluctance with a
corresponding reduction of the magnetic flux by the permanent magnet.
This means that at higher current levels the force will be smaller than
expected from the linear relation F = BI w . An other source of non-linearity
is the direct interaction of the iron part and the coil. The coil will act as an
electromagnet, attracting the iron in a direction that is independent of the
direction of the current. This attractive force is called the reluctance force,
because it is related to the change of reluctance when the coil approaches
the iron, and its magnitude is proportional to the current squared.
The reluctance force is used to create many actuators as will be presented
later in this chapter but its impact on linearity is a drawback when applied
in a precision positioning system. As a first step in the treatise of the
reluctance force, it will be examined as an unwanted effect in a Lorentz
actuator with possible means to reduce it. After this analysis, the analytical
force equations of electromagnetic actuators of any type will be derived from
the law of conservation of energy. It will be followed by a presentation of the
variable reluctance actuator and the permanent magnet biased reluctance
actuator.

5.3.1 Reluctance force in Lorentz actuator

In precision positioning systems, the Lorentz actuator is chosen mainly in

situations where no transmissibility due to motor stiffness is allowed. As
was demonstrated, Lorentz actuators are not completely free of stiffness.
The first cause for residual stiffness is based on the relation between force
and position and was discussed in the previous section. It represented a non-
linear stiffness, that can be kept at a low level by limitation of the movement
at the centre or by using electrical commutation. In case of a short stroke
actuator, the reluctance force can cause another stiffness factor. Figure 5.25
shows the induced flux by the current in the coil with an actuator where
the coil is located slightly outside the optimal position in the middle of the
air-gap. The permanent magnets are not shown, because the reluctance
258 Chapter 5. Electromechanic actuators

Ferromagnec yoke

Flux from the coil

Coil 'y
Ring
Ring
F
d' y
I e r Ef r 
dt

d' y
'e r I e r
dt

Figure 5.25: The reluctance force in a Lorentz actuator is a force that acts indepen-
dent of the permanent magnet flux. It is caused by the attraction of
the iron part by the magnetic field of the coil. For high frequencies
this force can be reduced by a conductive ring, that counteracts the
change of the magnetic flux in the iron part, as shown at the right.

force is only determined by the ferromagnetic part.

It is clear, that the reluctance of the flux caused by the current in the coil
is minimal when the coil is located completely inside the ferromagnetic
part. The reluctance force is related to a maximum in the magnetic flux as
function of the position. This means that the reluctance force on the coil is
always in the direction of the ferromagnetic part.
One way of reducing this reluctance force would be to extend the ferromag-
netic part to the right to beyond the magnets but that would increase the
mass and requires the coil holder to be longer.

5.3.1.1 Eddy-current ring

A second method of reducing the reluctance force is by using a conductive

ring, connected to the ferromagnetic part as shown in Figure 5.25. The
effectiveness of this method is based on the variability of the flux from
the coil. The change of the flux Φy inside the ring, by the current in the
coil, induces an inner Electric field Ef over the ring according to Faraday’s
law. Because the ring is closed, this electric field causes the electrons to
move, resulting in a current I e in the ring in the direction of the electric
field, as explained in Chapter 2 on electricity. This current is also called an
eddy-current, because in principle it is induced in any conductive material
5.3. Reluctance actuator 259

Non-ferromagnec material

Figure 5.26: Lorentz actuator without a ferromagnetic part to cancel the reluctance
force. The return path of the permanent magnet flux goes only through
air.

inserted in a changing magnetic field and it behaves like circular running

currents inside the material like the “eddy” currents or swirl in a fast flowing
river. In its turn, this induced eddy-current in the ring causes a magnetic
field Φe according to Ampère’s law in the opposite direction of the magnetic
flux of the coil. This means that the change of the magnetic flux of the coil
inside the ring is suppressed.
The total flux inside the ring equals the integral of the change of the flux
over time. As the change of flux increases with frequency, the reducing
effect of the ring on the flux, and correspondingly also on the reluctance
force, is most effective at higher frequencies and it is completely absent at
steady state currents. Because of the limitation to higher frequencies this
method is frequently applied in loudspeakers. For precision mechatronic
systems it is however not suitable, as those require a reliable operation, also
with non-varying, steady state forces.

5.3.1.2 Ironless stator

For those more critical applications where the linear performance is far
more important than the cost of a bit more magnet material, one can decide
to entirely leave the ferromagnetic part away. Figure 5.26 shows such a con-
figuration, where the magnets are connected by means of non-ferromagnetic
material. A second ring of magnets has been added to compensate for the
higher reluctance at the inside of the magnetic system. The increase of the
reluctance for the permanent magnet flux at the outside is less dramatic for
260 Chapter 5. Electromechanic actuators

the same reason as was presented with the air coil from Figure 5.4, where
the reluctance of the path outside the coil could be neglected in respect to
the reluctance inside the coil.
This configuration is used in the wafer stage that is presented in Chapter 9
on wafer scanners.

5.3.2 Analytical derivation of the reluctance force

As will be explained with the Reluctance actuator in the next section, Equa-
tion (5.43) is not valid under all circumstances as it is limited to the situation
where dΦw/d x is only depending on the permanent magnet and not on the
current. This means that it is necessary to use a different approach to
determine the total force that acts on an electromagnetic actuator including
the reluctance force. A suitable method is based on the law of conservation
of energy in a closed system.
In an electromagnetic actuator, electrical energy is converted in three differ-
ent kinds of energy:

1. Useful mechanical energy (work).

2. Stored magnetic energy.
3. Heat loss.

In Figure 5.27 the electrical equivalent circuit diagram of an electromag-

netic actuator is shown, with a current source as the electric power input,
a magnetic potential energy storage, represented by the self-inductance
L, and a resistive element that represents the thermal power output and
the motor voltage that is related to the velocity. The electric element that
represents the self-inductance is called an inductor because it works on the
induced magnetic field by a current.
The self-inductance is a very important dynamic property of an electro-
magnetic actuator. It has been given the unit Henry [H], named after the
American physicist Joseph Henry, who approximately at the same time as
Faraday, determined the relation between the induced voltage over a coil
as function of a change of enclosed flux, following Faraday’s third Maxwell
equation.
The self-inductance is defined as the ratio between the total flux Φw,t of all
windings summed, and the current in a coil.

Φw,t
L= [H] (5.54)
I
5.3. Reluctance actuator 261

I
+ 'w I
VR R +
+ _ Ef r 
d' w
VL  L
dI
Current
source + Va
dt dt
_ VL L L
n2
}
_ _
+
Vm Inductor L with n windings

Figure 5.27: Electrical equivalent circuit diagram of an electromagnetic actuator.

The resistor R represents the resistivity of the windings and induced
eddy-current losses, the inductor L represents the stored magnetic
energy and the voltage source Vm is the induced voltage by the velocity.
The right drawing shows the working principle of the inductor. It
transforms a current change into an electromotive force that opposes
to the external voltage, that causes the current change.

The total flux Φw,t is equal to nΦw , because a coil consists of n windings,
and each winding encloses the same flux Φw . The logic behind this relation
of the self-inductance is based on Faraday’s law, stating that the induced
Electric field Ef in a winding is equivalent to a change in flux over the closed
surface inside the winding. This creates a proportional electromotive force
over the winding. When each of the windings in a coil gets the same change
of flux, the total coil would show an induced electromotive forced, that is
the sum of the electromotive force of each individual winding as they are
placed in series. Due to the minus sign in Faraday’s law, the electromotive
force within the coil is directed from the negative to the positive electrode,
as explained in Chapter 2. This direction is opposite to the current direction
so, when the electrodes are defined as in Figure 5.27, the induced voltage
VL at the electrodes of the inductor is equal to this electromotive force:
dΦ w dΦw,t
VL = Fe = n = (5.55)
dt dt
With the defined relation of L, the total current induced flux equals Φw,t = LI ,
resulting in:
dI
VL = L (5.56)
dt
262 Chapter 5. Electromechanic actuators

It is to be expected that the self-inductance is determined by the dimensions

and magnetic properties of the coil. In terms of number of windings and
reluctance, this relation can be derived with Hopkinson’s law of magnetics
and is written as follows:
Φw,t Φw nI n2
L= =n =n = (5.57)
I I Iℜ ℜ
The implicit consequences of the self-inductance of a coil in the dynamic
domain will be presented further in Section 5.4.1. In the present section it
will be used to calculate the reluctance force in a five step approach based
on the law of conservation of energy.

Step 1: Power input Using the equivalent circuit diagram of Figure 5.27,
at any moment in time the electrical power is equal to:

Pin ( t) = Va ( t) I ( t) (5.58)

For simplification the time dependency term ( t) is omitted in the further

derived equations.
The total voltage Va consists of three parts determined by the different
elements of the equivalent circuit diagram:

Va = VR + VL + Vm (5.59)

The voltage over the resistor equals:

VR = IR (5.60)

The induced voltage over the self-inductance due to the change of the current
equals:
dI
VL = L (5.61)
dt
The motion voltage is the induced voltage due to the change of the flux by
the movement only. In fact the change of the flux Φw consists of two parts.
One part is related to a change in the current in the coil and matches with
the self-inductance. The other part is related to a change in the position:
∂Φw d x
Vm = n (5.62)
∂x dt
so that:
dI ∂Φ w d x
Pin = I 2 R + IL + nI (5.63)
dt ∂x dt
5.3. Reluctance actuator 263

Step 2: Power storage Energy can be stored in only one element, the
self-inductance. The stored magnetic energy is equal to the electric energy
needed to create it, so it is calculated by integrating the power that was
necessary to insert a current in the self-inductance. This power is equal to
the current times the voltage caused by the change of current:
dI
PL = I ( t)VL = I ( t)L (5.64)
dt
If the current at t0 equals 0 and at t1 equals I 1 the stored magnetic energy
E L at current level I 1 for a certain value of L is calculated by integrating
the power over the time from t0 to t1 :

t1 
t1 
I1
dI 1 2
EL = P L ( t) d t = I ( t) L dt = L I ( t) d I = LI (5.65)
dt 2 1
t0 t0 0

In case the self-inductance is depending on the position x( t) and the level of

I 1 is again the variable I , this can be written more generic as:
1
EL = L( x( t)) I 2 (5.66)
2
The total power flowing into this storage element at any moment is the time
derivative of this energy:
d
Pstorage = (E L )
dt
d I 1 2 dL ( x( t))
= IL + I (5.67)
dt 2 dt

Step 3: Power output Power can exit the system in only two ways: useful
as mechanical power or useless as dissipated heat.
Pout = Pdiss + Pmech (5.68)
The dissipated heat equals:
Pdiss = I 2 R (5.69)
The mechanical power is determined in the next step.

Step 4: Power balance The total power balance becomes:

Pstorage = Pin − Pout

   
d I 1 2 dL ( x( t)) dI ∂Φ w d x 3 4
IL + I = I 2 R + IL + nI − I 2 R + Pmech
dt 2 dt dt ∂x dt
(5.70)
264 Chapter 5. Electromechanic actuators

After some elimination of terms, the mechanical output power becomes:

∂Φ w d x 1 2 dL ( x( t))
Pmech = nI − I (5.71)
∂x dt 2 dt

Step 5: Force Mechanical power is delivered when a force is exerted

during motion:
dx
Pmech = Fv = F (5.72)
dt
or
dt
F = Pmech
dx
∂Φ w 1 2 d L ( x)
= nI − I (5.73)
∂x 2 dx

In this equation the first term is the linear relation of the force to the cur-
rent, that is always present in any actuator. The second term is the squared
relation of the force to the current and is caused by the magnetic energy that
is stored in the self-inductance. In the Lorentz actuator of Section 5.3.1.2
the self-inductance is independent of the position because there is no sur-
rounding ferromagnetic material that can influence the reluctance of the
magnetic field from the coil. In that configuration the second term is zero
and the general equation reduces to the previously derived Equation (5.43).

dΦw
F = nI (5.74)
dx
In the following it will be shown that for a reluctance actuator without any
permanent magnets the situation is quite different, because in that case the
second term becomes exactly half of the first term which means that half
the useful electrical energy (excluding resistive losses) is used to “charge”
the coil with magnetic energy and only the other half is used for mechanical
energy.

5.3.3 Variable reluctance actuator.

An example of an actuator working only on the reluctance force is the

variable reluctance actuator of Figure 5.28. It is also just called a “reluctance
actuator”. The principle is most widely applied as a basic electromagnet,
that is created with a current-carrying coil around a ferromagnetic yoke.
This electromagnet attracts other pieces of ferromagnetic material and most
5.3. Reluctance actuator 265

mover
A y,m x
dx
Ag Ag Ag
F

A y,s
stator
'w

I
n
Figure 5.28: The variable reluctance actuator only works by the principle of the
reluctance force. The moving part is attracted with a force proportional
to the current squared and inversely proportional to the distance
squared.

people know it from the magnet that pulls-up a car with a helicopter in
a crime movie. To calculate the force, Equation (5.73) is used, where the
relative movement of the mover to the stator equals d x:
∂Φ w 1 2 d L ( x)
F = nI − I (5.75)
∂x 2 dx
In this case, without a permanent magnet, ∂Φw /∂ x is only determined by
the current and directly related to the self-inductance, that in its turn is
only a function of the position:
∂Φw d LI
= (5.76)
∂x dx n
Now these two equations combine to:
d L ( x) 1 2 d L ( x) 1 2 d L ( x)
F = I2 − I = I (5.77)
dx 2 dx 2 dx
The self-inductance is determined by the series reluctance of the two ferro-
magnetic parts, the stator and the mover, with a total length y = y,s + y,m
and the two air-gaps with a total length of 2g , because the magnetic flux
has to cross the air-gap twice.

Φt n2 n2
L= = = (5.78)
I ℜc + ℜ g y 2 g
+
A y μ0 μr A g μ0
266 Chapter 5. Electromechanic actuators

With μr  1 and relative large values of g , this can be approximated2 into:

A g μ0
L ≈ n2 (5.79)
2 g

With this value for the self-inductance, Equation (5.77) with x = g gives
after a bit of algebra:
 2
nI μ0 A g
F ≈− (5.80)
g 4

The minus sign indicates that the force on the mover is directed in the
negative x direction. This corresponds with the known “pulling” direction
of an electromagnet. It is also clear from this equation that the reluctance
actuator is extremely non-linear. First of all the force only works in one
direction and the “gain”, that equals the force to current ratio, increases with
higher currents. But also the dependence on a movement in the x direction
is high and non-linear which implies a significant stiffness. The magnitude
of the force decreases when the displacement increases. This means, that
the actuator has a negative non-linear stiffness. When saturation effects are
neglected, an infinitely small gap could create an infinite force. In reality
the flux density is limited to the saturation of the applied ferromagnetic
material, which corresponds for ferromagnetic (soft) iron to a value in the
order of 2 T. The resulting force can then be calculated with the equations
of the next section.

5.3.3.1 Electromagnetic relay

An electromagnetic relay, as schematically shown in Figure 5.29, is a well-

known application of a reluctance actuator. The squared force relation of
the reluctance force is optimally utilised in this electrically activated switch.
A relay consists of a stator from a ferromagnetic material provided with a
coil and a mover which is connected via an elastic hinge to the stator. The
elastic hinge is pre-stressed so that the mover is pushed to its limit with a
certain force. An electrical contact is made when the mover is attracted to
the stator. When a current starts flowing in the coil, at a certain current
level the reluctance force equals the pre-stress force and the mover will start
moving towards the stator. This movement decreases the air-gap so the force
will increase further. This results in a non-linear exponential avalanche
effect that creates a very strong force and a fast closing of the contacts.
2 Later it will be shown that this approximation can cause large errors with high levels of

flux density and small values of g !

5.3. Reluctance actuator 267

Elasc hinge

Figure 5.29: An electromagnetic relay is a bistable system. the switch is pulled

towards the end stop by a pre-stressed elastic hinge. When the current
surpasses a certain threshold the magnetic force is larger than the
pre-stress and the switch will move. This movement reduces the
reluctance of the magnetic field with a resulting increase in force. The
“avalanche” effect will close the switch at a high speed.

5.3.3.2 Force exerted by a magnetic field

With the force Equation (5.80) of a reluctance actuator also the magnetic
flux density B g in an air-gap by a permanent magnet, or any other source,
can be related to the force that is exerted between the two sides of the air-gap.
In the reluctance actuator, the relation between the magnetic flux density
in the air-gap and the current is derived from the total flux that passes the
air-gap. It has a clear relation with the self-inductance L:
Φw LI
Bg = = (5.81)
A g nA g
The self-inductance value L of the reluctance actuator was obtained with
Equation (5.79):
A g μ0
L ≈ n2 (5.82)
2g
This gives with Equation (5.81):
nI μ0 2 B g g
Bg ≈ ⇒ nI ≈ (5.83)
2g μ0
This can be combined with Equation (5.80):
 2
nI μ0 A g
F ≈− (5.84)
g 4
268 Chapter 5. Electromechanic actuators

giving with a bit algebra:

B2g A g
F ≈− (5.85)
μ0

The total gap surface is 2 A g so a magnetic flux density B causes an equivalent

virtual pulling “pressure” Pm of:

F B2g
Pm ≈ ≈− (5.86)
2 Ag 2μ 0

This means for example that a magnetic flux density of 1 T creates a pulling
force per unit of surface (negative pressure) of:
1
P1T ≈ ≈ 0.4 [MPa] (5.87)
8π · 10−7
This equals a negative pressure of 4 times vacuum, which immediately
give the reason why permanent magnets are often used for a contact-less
pre-load of air-bearings as applied in precision positioning systems.

5.3.4 Hybrid actuator

It is valuable to investigate how a reluctance actuator can be made suitable

for controlled positioning systems, because of the high force-to-current ratio
in configurations where a very small air-gap is allowed. The first problem
to solve is the unidirectional force.

5.3.4.1 Double variable reluctance actuator

The limitation by the unidirectional force can either be solved by providing

a passive force in the other direction by means of a spring or by combining
two reluctance actuators into one actuator, like shown in Figure 5.30. This
configuration proves to be a very interesting system as will follow from the
calculations. First of all the force on the central mover equals the difference
of the forces of each actuator. In the previous example of the single reluctance
actuator the positive x direction was defined from the stator upwards to the
mover resulting in a minus sign for the force. In this example the positive
x direction is defined in the right direction as shown in the picture. This
means that the force that acts on the mover by the left actuator half (F1 ) has
a negative sign and the force by the right actuator half (F2 ) has a positive
sign. Using Equation (5.80) twice, for each half of the actuator, the total
5.3. Reluctance actuator 269

x dx

I1
F1 I2
F2

C1 C2

A g,1 A g,2 F

Figure 5.30: A double reluctance actuator with a central mover shows a linear
current to force relationship in the middle position. This is caused by
the balancing of the force of the two non-linear actuators.

force becomes:
 2  2  
nI 2 μ0 A g nI 1 μ0 A g n 2 μ0 A g I 22 I 12
F = F2 − F1 = − = − (5.88)
g,2 4 g,1 4 4 2g,2 2g,1

This shows as a first conclusion, that in the mid position, where g,1 = g,2 ,
the force is zero when the currents are equal. As a next step it is interesting
to see what happens if the currents in both halves are modulated, while
keeping the sum of the currents constant. This means that current I 1
becomes I a − Δ I and current I 2 becomes I a + Δ I , while I a is the average
current.
The calculations start with Equation (5.88), with the mover in the mid
position, g,1 = g,2 = g .
 
n2 μ0 A g I 22 − I 12
F= (5.89)
4 2g

With I 1 = I a − Δ I and I 2 = I a + Δ I the force becomes:

   
n2 μ0 A g ( I a + Δ I )2 − ( I a − Δ I )2 n 2 μ0 A g 4 I a Δ I
F= = (5.90)
4 2g 4 2g

With this result, the force to current ratio of this actuator for the current
change Δ I can be written as:
F I a n 2 A g μ0
= (5.91)
ΔI 2g
270 Chapter 5. Electromechanic actuators

This means that the force to current ratio is depending on the level of the
current level that is common for both halves. It gives the possibility to
control the differential gain by this common current level! The second
conclusion is that the force is linearly related to the differential current
at the same mid-position! This means that this actuator can be used in a
linear control system with a separately controllable gain. Unfortunately
this is only the case at the mid position. At other positions the actuator
half with the smallest gap will dominate the other half, and the linearity
gradually disappears. Still for small displacements with a high force this
configuration can be useful.
Another drawback of this configuration is the negative stiffness that depends
on the common current. This effect can be explained from the fact that at
any position away from the mid-position the mover will mainly be attracted
to the position where it is closest to the stator. This makes this actuator less
suitable for situations, where a direct coupling between stator and mover
is undesired like in a vibration isolation system. In principle this negative
stiffness can be compensated with the positive stiffness of a mechanical
spring, but in practice it is difficult to achieve a reliable reduction of more
than a factor 10 due to tolerances.
With the reluctance actuator it was shown that by combining two reluctance
actuators the actuator can be linearised by providing it with a current
that is common for both halves. This principle has as drawback that a
DC current will contribute to the power losses in the actuator and also
the negative stiffness is quite large. It is a logical step to replace the flux
created by this common mode current by the flux of a permanent magnet
and thus combine the best of both worlds: the high force of the reluctance
actuator with the linearity of the permanent magnet actuator. To introduce
a permanent magnet in the double reluctance actuator it is necessary to
carefully consider the working principle as explained previously. First
of all the force is generated by the flux in the air-gap. Because of the
squared relation between flux and force, a force is obtained that is equal
to the multiplication of the flux related to the common current and the
flux related to the differential current. This means that a perfectly linear
very strong actuator is obtained under the following two conditions. First it
must be possible to create a common flux must in the air-gap by means of a
permanent magnet that is independent of the position of the mover. Secondly
the permanent magnet is not allowed to introduce a significant additional
reluctance for the flux from the current in the coil. In the following section
it is shown how this can be achieved.
5.3. Reluctance actuator 271

dx dx

I1 I1 I2

C1 C2 I2 C1 C2

F F

Figure 5.31: The flux in the double reluctance actuator can be changed by inverting
the current direction in one of the coils. Because of the squared relation
in the reluctance force this inversion has no influence on the forces.
Nevertheless the flux path is completely changed. This creates the
possibility to change the material properties of the mover without
impacting the reluctance force.

5.3.4.2 Combining two sources of magnetic flux

The first step is to be aware that the double reluctance actuator of Figure 5.30
works independent of the current direction in the coils due to the squared
relation between current and force. Though it has no impact on the force, the
flux is different if the current in one coil is reversed as shown in Figure 5.31.
If the flux in both coils is in the same direction (upwards or downwards)
the flux of both coils has to flow back vertically through the mover. When
the current in one of the coils is reversed, as shown in the right picture,
the flux of both coils share the same path which only crosses the gap. The
vertical path in the mover is avoided which means that the middle part
of the mover can be replaced by another non-ferromagnetic element like
a permanent magnet. This is shown in Figure 5.32 where the flux of the
permanent magnet perfectly combines with the induced flux of the current
without adding additional reluctance to the latter. In the right picture a
simplified version is shown where one side of the mover is connected to the
stator by means of an elastic hinge that enables rotation in combination
with a low reluctance path for the permanent magnet flux. This elastic
hinge can be used to partly compensate the still not completely avoidable
inherent negative stiffness of the actuator as will be explained later.
It is clear that the permanent magnet creates an unstable situation because
272 Chapter 5. Electromechanic actuators

x F x F
dx dx

I1 N I2 I1 N I2
PM PM S
S
C1 C2 C1 C2

Elasc hinge

Figure 5.32: By inserting a permanent magnet in the mover of the double reluctance
actuator, an additional flux is created, that can replace the flux that
was generated by the average current of the original double reluctance
actuator. This replacement strongly reduces the power consumption
of the actuator.

the magnetic circuit has its lowest reluctance at one of the two positions
as shown schematically in Figure 5.33. With the mover in the middle
position the flux is equally divided over the two half’s of the actuator and the
attraction force in both air-gaps cancel each other out. As soon as the mover
moves in one of the two directions the flux will increase in that half and
decrease in the other which will cause the negative stiffness. Figure 5.34 is
used to calculate the flux of the permanent magnet. While the reluctance
for each gap is proportional to their length x, the total reluctance of both
gaps combined is calculated by taking the reluctance of both gaps in parallel
according to the following relation.
1 1 1
= + (5.92)
ℜ ℜ1 ℜ 2
With d x the displacement of the mover, g,1 = g + d x and g,2 = g − d x being
the length of both gaps, A g the cross section of the gaps perpendicular to
the flux and the ferromagnetic material having an infinite μr , the following
equation is obtained:

1 2g − ( d x)2
ℜ=  = (5.93)
1 1 2μ0 A g g
μ0 A g +
g + d x  g − d x

Around the mid position with d x  g , this reluctance is approximately con-

stant but at the outer positions this causes the total reluctance to decrease.
This leads to a higher flux and stronger negative stiffness, however to a
5.3. Reluctance actuator 273

Figure 5.33: The permanent magnet flux division in a hybrid actuator depends on
the position of the mover.

lesser extent than with the double reluctance actuator from the previous
section. This is caused by the fact that the reluctance of the permanent
magnet field is dominated by the permanent magnet material itself which
limits the flux density to below Br . Remember that a modern permanent
magnet acts like an air coil with a very large current. With the reluctance
actuator a small gap will lead to a steep increase of flux density that would
only be limited by the saturation. Because of this lower and more linear
negative stiffness of the hybrid actuator it can more easily be compensated
by a positive spring, which in this configuration is determined by the elastic
hinge. In Figure 5.34 the effect of a current through the coils is shown. The
flux from each coil will follow the path of lowest reluctance, according to the
yellow arrows, showing that it will not pass the permanent magnet but go
round in the ferromagnetic part. As mentioned the winding direction of the
coils is such that their fluxes positively add together.
Depending on the current direction, the combination of the flux of the coils
with the flux of the permanent magnets results in a lower flux density in one
air-gap and a higher flux density in the other air-gap, causing an attracting
force in the direction of the increased flux density. As this happens in two
directions, the force is depending on the current direction and the effect is
linearised according to the same reasoning as with Equation (5.91) for the
double reluctance actuator.

5.3.4.3 Hybrid force calculation

In order to approximate the magnitude of the force, first Equation (5.73),

the general equation of electromagnetic actuators is applied.
dΦw 1 2 dL( x)
F = nI − I (5.94)
dx 2 dx
274 Chapter 5. Electromechanic actuators

x
Ag F
Flux increase dx Flux decrease

A g,1 N A g,2
I2
I1
n n
Coil 1 S Coil 2

Figure 5.34: The flux in a hybrid actuator is a combination of the flux of the per-
manent magnet and the flux of the coils. In one air-gap the flux of
the coils has the same direction as the flux of the permanent magnet,
increasing the total magnetic field, while at the other air-gap it is
just opposite. This results in a strong net force in the direction of the
strongest magnetic field.

The self-inductance of the coils is hardly influenced by the position of the

mover because the reluctance of the coils consists of the reluctance of the
two air-gaps in series.

ℜ = ℜ1 + ℜ 2 (5.95)

When one gap gets smaller the other gets bigger so the second term can be
neglected and the known relation from the Lorentz actuator is used:
dΦw
F = nI (5.96)
dx
In this equation it is necessary to replace n by 2n because the flux of the
windings of both coils are added together as they are working in the same
direction. The flux through the windings Φw consists of the flux of the per-
manent magnet and the flux caused by the current through the coils. Under
condition of a non-saturated iron part, as mentioned in Equation (5.95), the
flux caused by the current is determined by the reluctance of the air-gaps
in series. The total reluctance of these gaps is equal to:
g,1 + g,2
ℜ= (5.97)
μ0 A g
5.3. Reluctance actuator 275

This is a constant because the sum of both gap lengths is constant. This
means dΦw / d x is only determined by the permanent magnet flux. The
force can be approximated by taking the two extreme positions when the
mover is just hitting the stator. In both situations the iron determines a
low reluctance path for the permanent magnet, which means that the flux
density Bm of the permanent magnet will approach Br . So by changing the
position from one side to the other the average flux changes approximately
from zero to Br A m . This gives:
dΦ w Bm A m Br A m
= ≈ (5.98)
dx g,1 + g,2 g,1 + g,2

With the two equal coils (2nI ) this results in an estimated force to current3
ratio of:
F dΦw Br A m
≈ 2n ≈ 2n (5.99)
I dx g,1 + g,2

Like all actuators, also the hybrid actuator has a position dependency of the
force, not only related to the mentioned negative stiffness, but also due to the
position dependent force to current ratio. In the hybrid actuator this second
effect is caused by the changing total reluctance of the permanent magnet
circuit as function of the position. As discussed earlier, the reluctance for
the permanent magnet at the mid position equals the the reluctance of the
two air-gaps in parallel. In the outer positions one of the air-gaps is very
small, resulting in a lower total reluctance. This means that the flux of the
permanent magnet increases at the outer positions, giving a higher value
of dΦw / d x than in the mid position.
It is clear that these calculations are very approximative due to the complex
nature of the system, with potential saturation of the iron part and the
approximation of the permanent magnet flux in the extreme positions.
To check this approximation, the force in the mid position can be calculated
in another way, by means of Equation (5.86) with the cross section of the
air-gaps and an estimation of the flux of the permanent magnet.
Like with the double reluctance actuator two counteracting forces are com-
bined. The force in air-gap 1 works in the negative x direction and the force
in air-gap 2 in the positive x direction. The flux density in the air-gaps (B g )
equals the combination of the flux density by the permanent magnet (Bg,m )
and the flux density caused by the coil (Bg,c ). With the current direction as
defined in Figure 5.34 the total flux density in air-gap 1 equals Bg,m + Bg,c
3 Here the current ( I ) represents the variable current to control the force in a position

controller. It is comparable with the Δ I from Equation (5.91)

276 Chapter 5. Electromechanic actuators

and in air-gap 2 it equals Bg,m − Bg,c and the following equation is obtained:
Ag Ag 2 A g Bg,m Bg,c
F = F2 − F1 = (Bg,m − Bg,c )2 − (Bg,m + Bg,c )2 = − (5.100)
2μ 0 2μ0 μ0
The minus term is due to the fact that a positive current in the shown
direction results in a force on the mover to the left which is the negative x
direction.
In case the air-gap cross section is large with respect to the length of the
air-gap, the leakage flux will be moderate (λ > 0.80). With a well designed
system, the flux density of the magnet is chosen to be Br /2 and the flux is
divided over two halves which means that a flux density in each gap is equal
to:
0. 8 · B r A m
Bg,m = (5.101)
2 2 Ag
When the reluctance of the air-gaps is large in comparison with the reluc-
tance of the ferromagnetic parts, the flux induced by the current equals
(two coils with n windings!):
Φw 2 nI 2 nI μ0
Bg,c = = = (5.102)
Ag A g ℜ g,1 + g,2
Combining these equations for Bg,m and Bg,c with Equation (5.100) gives the
following force to current ratio of the hybrid actuator:
F Br A m
= 0.8 n (5.103)
I g,1 + g,2

When this result is compared with Equation (5.99) it clearly shows more
than a factor two decrease of the force in the mid position which is caused by
the reluctance of the air-gaps for the permanent magnet flux and the choice
to use as little as possible magnet material (Bm = Br /2). In practical designs
often a larger magnet is chosen which brings the force to a higher value and
reduces the position dependency of the force to current ratio because the
flux density will be more constant closer to Br .
It can also be concluded that with increasing complexity of the magnetic
circuit the need for more exact calculations using FEM computer simulations
is more strong. Nevertheless the presented calculations provide a good sense
for the order of magnitude the real forces in a practical actuator.

5.3.4.4 Magnetic bearings

A nice application example of a hybrid actuator is its use in a magnetic

bearing. Figure 5.35 shows a fully integrated magnetic bearing that supports
5.3. Reluctance actuator 277

Figure 5.35: Five degrees of freedom homo-polar magnetic bearing. The flux by the
current in the coils (red arrows) is guided via a different route than the
flux from the permanent magnets (blue arrows). In the air-gaps they
combine to higher or lower values depending on the current direction,
thereby enabling a change in the force in different directions.

a fast-rotating shaft in five directions, three orthogonal linear directions

and two rotations. Only the rotation around the central axis is left free. This
concept is designed for a high speed micro-milling centre that has to operate
at speeds above 300.000 rpm. Also in this system the permanent magnets
create a bias flux which is modulated by the coils. By tracing the flux of the
different coil sections as indicated with the red arrows and combine these
with the flux from the permanent magnet as indicated with the blue arrows
it can be imagined how all forces are created. It is worthwhile to notice,
that also in this case the permanent magnets do not increase the reluctance
of the flux from the coils. Still the configuration is essentially different from
the example in the previous section as it uses the low reluctance path in
the third dimension for the flux induced by the current in the coils. This
is just another illustration of the large amount of configurations, that are
possible in electromagnetic actuators. This example of a magnetic bearing
is called homo-polar because the permanent magnet flux flows in the axial
direction through the rotating shaft, which means that the flux in the shaft
does not change due to the rotation. As a consequence of this configuration,
power consuming eddy-currents are prevented that would otherwise be
induced by the changing flux in the rotating shaft. An important property
of a magnetic bearing is its inherent negative stiffness. The shaft needs to
rotate freely which means that this stiffness can not be compensated by a
mechanical spring like the elastic hinge of the previously presented linear
278 Chapter 5. Electromechanic actuators

hybrid actuator. In Chapter 4 it is shown how negative feedback control

enables to stabilise the position of the shaft inside the magnetic bearing by
providing a virtual spring with positive stiffness and damping. This active
control of magnetic bearings has enabled their use in critical applications
that do not allow mechanical contact and require strict control of deviations
from the ideal position of the shaft at often extremely high rotation speeds.
5.4. Application of electromagnetic actuators 279

5.4 Application of electromagnetic actuators

The application of the three presented electromagnetic actuators in mecha-

tronic systems is determined by their properties in relation to the other
parts of the system. This section will first present the electrical interface
with the amplifier followed by a comparison of the three types with some
realistic data to illustrate their different characteristics.

5.4.1 Electrical interface properties

In mechatronic positioning systems the actuator is always used in combi-

nation with an amplifier. This amplifier has also its limitations as will be
presented in Chapter 6. The impedance of the actuator strongly influences
the total transfer function and stability of the amplifier-actuator combina-
tion and also the amplifier output characteristics strongly influences the
behaviour of the actuator. in this section these effects will be elaborated.
First the dynamic effect of the self-inductance will be investigated in the
frequency domain, when the amplifier can be approximated as an ideal
current source. This means that the behaviour of the actuator does not
influence the current of the amplifier. It will be shown that these dynamic
properties will only depend on the dimensions of the coil and not on the
number of windings and the wire thickness.
Directly related to the self-inductance is the ability of the actuator to change
the acceleration rapidly. This is represented by the jerk., that is the deriva-
tive of acceleration over time.
As a last step the amplifier will be modelled in a more realistic way, where
the behaviour of the actuator influences the current. It will become clear,
that the actuator shows damping properties, when the amplifier is not an
ideal current source.

5.4.1.1 Dynamic effects of self-inductance

In Figure 5.27 it was shown that an electromagnetic actuator can be repre-

sented in the electrical domain as a series of a resistance, a self-inductance
and a voltage source. When the actuator is supplied by a current, the voltage
over the actuator will be equal to:
dI
Va = Vm + IR + L (5.104)
dt
280 Chapter 5. Electromechanic actuators

d
b
h

b
Figure 5.36: Two different coil configurations with their basic dimensions shown
as an example of the wide amount of possibilities. Next to these
rotation symmetrical shapes, also square, race-track and even three
dimensional shapes are used.

In order to get an idea of real values, the resistance from the windings is
calculated with the following relation:
n w
R = ρr (5.105)
Aw
with

ρ r = resistivity of the wiring material. [Ω/m]

w = average length per winding. [m]
n = number of windings.
A w = cross section of the wire. [m2 ]

When designing a coil of a certain size, it mostly starts with the dimensions,
giving the average length of each winding (w = π d ) and its cross section
( A c,w = bh) as shown for two typical coil configurations in Figure 5.36.
A c,w relates to A w in the following manner:
γ A c,w
Aw = (5.106)
n
With γ being the fill factor which reduces the useful volume of the coil due
to round windings and the insulation. Practical values for γ range between
approximately 0.5 for thin round wires to 0.9 for flat wires. With these
factors the resistance becomes:
ρ r n 2 w
R= (5.107)
γ A c,w
5.4. Application of electromagnetic actuators 281

Assuming γ and ρ r are constant for a given coil size, the resistance is pro-
portional to the number of windings squared.
To estimate the effect of the self-inductance, Equation (5.104) is used in the
mechanical stationary situation, which means that Vm = 0
dI
Va ( t) = IR + L (5.108)
dt
After applying the Laplace transform to bring the equation to the frequency
domain the following relation is found:

j ωL
Va ( s) = I (R + sL) =⇒ Va (ω) = IR 1 + (5.109)
R

To give a value to the ability to change the force rapidly, the electrical time
constant τe is introduced, defined as:
1 L
τe = = (5.110)
ω0 R

With Equation (5.109) the impedance Z becomes:

Va
Z (ω) = = R (1 + j ωτe ) (5.111)
I
This makes clear, that above ω0 = 1/τe , the impedance, and consequently also
the voltage that is necessary to drive a current to the actuator, increases
proportional with a slope of +1 as function of the frequency. This entails a
heavy requirement for the amplifier that has to deliver both a high voltage
to cope with the self-inductance and a high current to deliver the force at
elevated frequencies.
For this reason it is important to keep τe as low as possible. When the
resistance R is replaced by Equation (5.107) and Equation (5.57) is used to
2
replace the self-inductance L by n /ℜ, the following relation for the electrical
time constant is obtained:
γ A c,w
τe = (5.112)
ℜρ r  w

This shows that τe can only be tuned by the dimensions of the coil, the
resistivity and the reluctance.

5.4.1.2 Limitation of the “jerk”

In high speed precision controlled positioning systems, the voltage that is

required to change the current in the self-inductance of the actuator, needs
282 Chapter 5. Electromechanic actuators

to be actively limited in order to stay within the maximum voltage range

of the power amplifier. When exceeding this range, the amplifier does not
reliably transmit the control information to the actuator. This in fact means
that the system is (temporarily) out of control.
In most actuators the current is proportional to the force and as a con-
sequence it is proportional to the acceleration of the driven mass. This
means that by limiting the change of current, the change of acceleration, the
“jerk” is limited. Because of this limitation, high speed precision controlled
positioning systems need to control both position, speed, acceleration and
jerk. Sometimes even the derivative of the jerk over time is controlled. This
entity is called snap.
In case of fast moving reciprocating systems like wafer scanners it is clear
that a limitation of the jerk automatically limits the maximum attainable
acceleration as being the integral of the jerk over time. It is also for this
reason that a Lorentz actuator is often the preferred choice in these systems
as it possesses the lowest value of τe from all actuator types, especially when
the ferromagnetic part is completely omitted, like the example shown in
Figure 5.26. It is however also true that a more efficient actuator based
on the reluctance or hybrid principle would needs less current and this
could compensate for the higher voltage that is necessary to realise the high
d I/d t through the self-inductance. This is another area for optimisation in
mechatronic systems.

5.4.1.3 Damping caused by source impedance

In the previous modelling of the behaviour of an electromagnetic actuator

it was always assumed that the current was given. In practice the current
is supplied by a power amplifier. In the following chapter on electronics it
will be explained that a power amplifier is not capable to deliver a current
that is fully independent of the load. It will be shown that the output of the
amplifier possesses a certain source impedance that influences the behaviour
of the connected actuator, the load. As a result all electromagnetic actuators
show a certain amount of damping, when connected to a power amplifier.
This damping is caused by the motion voltage Vm that is induced in the coil,
due to the change of flux that is related to the movement. This voltage in its
turn will cause a current through the circuit that is further determined by
the series of the impedance of the actuator and the source impedance of the
amplifier as shown in Figure 5.37. In the following the effect of this current
will be explained by means of the Lorentz actuator, because of its constant
permanent magnet flux that simplifies the example.
5.4. Application of electromagnetic actuators 283

It Is Im
Rs Rs Rs

Ra Ra Ra
+ +
_
Vs
L = _
Vs
L + L

+ +
Amplier Actuator Vm Amplier Actuator Amplier Actuator Vm
(source) (load) _ (source) (load) (source) (load) _

Figure 5.37: Electrical circuit of an amplifier with a resistive output impedance

R s , that acts as a source for an electromagnetic actuator, represented
by its electrical equivalent. The total current in the circuit I t is the
combination of the current I s , caused by the amplifier and the current
I m , caused by the motion voltage Vm , I t = I s − I m . The damping is
caused by I m only.

In principle the effect of the two voltage sources can be combined linearly
and the total current in the circuit is equal to the sum of the currents caused
by both voltage sources.
Vs Vm Vs − Vm
It = Is − Im = − = (5.113)
R s + R a + j ωL R s + R a + j ωL R s + R a + j ωL

This means that it is allowed to analyse the behaviour of an actuator for each
voltage source separately, while the other source is replaced by a conducting
wire. The current caused by the amplifier voltage creates the force that
was described in the previous sections. The current caused by the motion
voltage is the new factor. It creates the damping effect by the resulting force
in the actuator, counteracting the velocity. The related damping coefficient
c can be derived as follows for a Lorentz actuator.
Using Figure 5.38 for the directions, it starts with the motion voltage:

Vm = (Bw,t )v [V] (5.114)

with w,t being the total length of the windings inside the magnetic field.
To determine the resulting current, the impedance of the self-inductance is
neglected in respect to Ra + Rs . This is mostly allowed with Lorentz actuators
working at relatively low frequencies.

Vm (Bw,t )v
Im = = [A] (5.115)
Rs + Ra Rs + Ra
284 Chapter 5. Electromechanic actuators

dx
v
dt

BA w,t

2
v
Fd  BA w,t v
Ra Im 
Ra

Figure 5.38: A Lorentz actuator shows a velocity dependent damping force when the
two terminals of the coil are connected by means of a low impedance
circuit (Rs = 0). Following Faraday’s law, the induced electric field by
the movement-related flux-change creates a current I m in the direction
as shown, due to the minus sign. This current generates a Lorentz
Force Fd that is directed in the opposite direction of the movement
according to the corkscrew rule.

As the current will flow in the same magnetic field, a Lorentz force will
occur:

Fd = (Bw,t ) I m [N] (5.116)

(Bw,t )v
Filling in for I m gives:
Rs + Ra

(Bw,t )2 v
Fd = [N] (5.117)
Rs + Ra

So the damping coefficient c is:

Fd (Bw,t )2
c= = [Ns/m] (5.118)
v Rs + Ra

The conclusion of this equation is, that the damping is proportional to the
B-factor squared. This factor is also related to the force to current ratio.
This means, that it is not preferred to reduce this B-factor when less damp-
ing is needed. In that case it is better to control the damping by adapting
the resistive value of the circuit.
In loudspeaker systems it is necessary to create damping, in order to sup-
press the resonance of the loudspeaker cone with the stiffness of its support
and the air cabinet. For that reason amplifiers for music need to have a
low output impedance. For precision mechatronic systems it is more often
required to avoid the motion related damping as much as possible, because
5.4. Application of electromagnetic actuators 285

of the transmissibility of external vibrations through the electromagnetic

damper. For that reason in those cases the output impedance of the amplifier
should be as high as possible. This means that the amplifier will act like a
current source with a current level, that is almost independent of the load.
As an additional effect at first sight the self-inductance will not cause any
trouble in the dynamic performance of the system, but as will be shown in
the next chapter the impedance of the self-inductance can interfere with
the stability of the amplifier. It remains important to keep the electrical
time constant τ e as low as possible.

5.4.2 Comparison of the actuation principles

In designing mechatronic systems, making choices is often rather difficult

because of the many, sometimes contradictory, requirements. In practice,
many decisions are made based on experience and personal preference,
which can lead to the solutions like the mousetraps in the first chapter.
This is often also the case with the choice of an electromagnetic actuator.
Many people, especially those working with high velocity and extremely
accurate positioning systems, almost without thinking choose a Lorentz
actuator, because of its linearity and low values of mechanical stiffness,
that could transfer vibrations from outside. In the application of the wafer
scanners that are presented in Chapter 9, the inherent drawback of the
limited force in relation to the electrical power has resulted in an immense
power consumption up to several kilowatts per stage. In fact, the maximum
acceleration of a Lorentz actuator is limited to a physical maximum that
depend on the allowable thermal dissipation. Above a certain specific force
per unit of mass, the necessary increase of mass to cope with an increased
current will more and more reduce the benefit of the additional force that was
needed for acceleration. This has led to clever combinations with different
actuator types, like the reluctance actuator that were not considered suitable
before. With the help of fast modern control algorithms and adequate sensors
it is possible to compensate partly the negative properties. In the machine
tool industry, the actively controlled cutting systems have to handle large
forces at moderate speeds. Initially piezoelectric actuators were applied
because of their high stiffness, as will be discussed in the last section of this
chapter. More recently however, also the use of electromagnetic actuators of
the hybrid type has been investigated, because these combine a high force
relative to the consumed electric power, a well controllable linear behaviour
and a moderate stiffness. Although the actuation range of a hybrid actuator
is in the order of one millimetre or less, this is still always much larger than
286 Chapter 5. Electromechanic actuators

dc=30 mm

hc=15 mm

bc=3,3 mm

Figure 5.39: Dimensions of the coil, used for the comparison of the three different
actuator types.

the stroke of a piezoelectric actuator, with its maximum range of about 1

μm per mm actuator length.

5.4.2.1 Standard coil dimension for the comparison

To illustrate the different properties of the three discussed actuator types,

an example with practical values will be presented to close off this sec-
tion on electromagnetic actuators. As said, several parameters have to be
determined, of which the most important are:

• Current to force ratio in relation to power dissipation

• Electrical time constant

• Moving mass

The main force properties of the Lorentz, reluctance and hybrid actuator
types, will be compared by means of a small calculation example, all with
the same electrical power. For comparison, a coil is chosen with a fixed
configuration and total cross section of the windings as shown in Figure 5.39.

As a next step the number of windings for this design needs to be determined.
It was already shown in Equation (5.112) that the number of windings has no
influence on the dynamic properties of the actuator. The winding volume is
fixed, which means that also the mass is constant. Also the power dissipation
I 2 R is no factor in the design decision as the current is inversely proportional
to n and the resistance is proportional to n2 . This means that the only item
that determines the number of windings is the electrical power source. More
windings imply a higher voltage and a lower current, which means that the
number of windings are adapted to the amplifier. In motion systems with
very high power also the maximum wire thickness of the supply wires is a
5.4. Application of electromagnetic actuators 287

constraint, but for this comparison study this plays no role. This coil has
an average length per winding w of ∼ 100 mm and a cross section A c of 50
mm2 . In order to simplify the example, a look up table as shown in Table 5.2
is used for the wiring properties. A practical value of wire is chosen of
0.75 mm, including insulation, enabling easy calculation for this example
case. With this wire diameter, the volume can be filled with approximately
100 windings, giving a total length w,t of ten metres, which results in a
resistance of around 0.5 Ohm. One Ampère of current will then give half a
Watt of Power. With this coil the performance of the three different actuators
will be evaluated.

Table 5.2: Look up table for electrical coil windings of copper.

Wirediameter Wirediameter Windings per Ohm per 100 m

insulated (mm) cm2
(mm)
0.259 0.2 1890 55.8
0.282 0.22 1540 64.1
0.316 0.25 1230 35.7
0.342 0.27 1060 30.6
0.35 0.28 1000 28.5
0.374 0.3 890 24.8
0.396 0.32 750 21.8
0.43 0.35 640 18.2
0.46 0.38 560 15.5
0.487 0.4 510 13.9
0.54 0.45 400 11.2
0.595 0.5 310 8.9
0.65 0.55 270 7.38
0.7 0.6 230 6.21
0.75 0.65 199 5.29
0.81 0.7 174 4.56
0.86 0.75 132 3.97
0.92 0.8 118 3.49
0.97 0.85 106 3.11
1.03 0.9 96 2.76
288 Chapter 5. Electromechanic actuators

5.4.2.2 Force of the Lorentz actuator

With the Lorentz actuator, the diameter of the chosen coil limits the diameter
of the yoke. This also limits the capability of this yoke to transfer a high
amount of magnetic flux. For this reason the chosen coil dimensions are
more wide than long. In order to avoid saturation of the yoke, the flux
density should remain below 2 T. With this value the surface of the air-gap
times the flux density in the air-gap should be lower than the maximum flux
density times the surface of the yoke. With a small calculation this means
that the uniform magnetic field in the air-gap can be maximum 0.7 T. The
height of the air-gap is equal to the coil and in the mid position F = BI t
results in a current to force ratio of 7 N/A. In reality this value will be lower
because of stray flux.

5.4.2.3 Force of the reluctance actuator

When the same coil is applied in a reluctance actuator Equation (5.80) can
be used:
 2
nI μ0 A g
F =− (5.119)
g 4

If the length of the air-gap equals g = 10−3 m, which is quite large, and
A g = 6 · 10−4 m2 , then the resulting force to current ratio would amount up to
only 3.75 N/A, which is not yet impressive. When the air-gap is reduced to
0.3 mm, the force with 1 A current already becomes 37.5 N. At full closure of
the gap the iron might ultimately saturate with a flux density of around 2
Tesla. With Equation (5.85) the maximum attainable force would be equal
to:
B2g A g 4 · 6 · 10−4
Fm = − =− ≈ 2000 [N] (5.120)
μ0 4π · 10−7

This clearly shows the value of a reluctance actuator but it is necessary to ad-
mit that a large error is made due to the approximation on the permeability
of the iron (μr = ∞). This is far too optimistic when the flux density ap-
proaches saturation. By using Hopkinson’s law of magnetics, the minimum
permeability to realise this value can be determined with the stated flux
density of 2 T, using the length y of the flux path through the ferromagnetic
part and assuming a fully closed airgap g = 0:

Φ nI μ0 μr nI
B= = = =2 [T] (5.121)
Ag Agℜ y
5.4. Application of electromagnetic actuators 289

With the known values for this example, nI = 100 and y ≈ 0.1 m, the required
minimum relative permeability becomes:
2y
μr ≥ ≈ 1.6 · 103 (5.122)
μ0 nI

This high value is not realistic with normal ferromagnetic materials. A

value for μr of 500 is more normal, giving a maximum flux density of about
0.7 T, with a corresponding force of ≈ 230 N. Even with this limitation the
reluctance actuator still is capable to deliver the largest force of the three
electromagnetic actuators.

5.4.2.4 Force of the hybrid actuator

For the force of the hybrid actuator Equation (5.103) is used:

Br A m
F ≈ 0.8 nI (5.123)
g,1 + g,2

The same key dimensions as with the other actuators are chosen:

A m = A g = 6 · 10−4 [m2 ] (fits inside the coil)

Br = 1.2 [T]
n = 50 (The coil is split in two to keep the power identical)
g,1 + g,2 = 1 · 10−3 [m] (same stroke as reluctance)

These values result in a linear! current to force ratio of about 30 N/A, which
clearly shows the combined performance of “the best of both worlds”, a high
force with a linear force to current relation over a moderate range.

5.4.2.5 Dynamic differences

The next criterion for the right actuator choice is the ability to change the
force rapidly. This is reflected by the electrical time constant and the related
necessary limitation of the jerk. For the same three actuators with the
example coil as shown in Figure 5.40, the electrical time constant is purely
determined by the reluctance of the flux resulting from the current in the
coil. The higher the reluctance, the better it is.
When drawing the field lines of the magnetic field, caused by the current
through the coil, in the three actuators, it is clear that the reluctance is
quite high for the Lorentz actuator because it includes a large air path. This
would even be better when no ferromagnetic part is applied. The reluctance
290 Chapter 5. Electromechanic actuators

Lorentz Reluctance Hybrid

ƒ high ƒ low ƒ low
L low L high L high
m high (yoke) m moderate m moderate

Figure 5.40: Difference in reluctance, self-inductance and mass of the three actua-
tor types scaled to the same coil dimensions.

of the Lorentz actuator is in any case considerably smaller than with the
Reluctance and hybrid actuator because of their small air-gap. For this
reason, a Lorentz actuator is better suited for high speed precision actua-
tion with fast changing currents, while the other actuators can be applied
where high forces in a semi static situation are required. Nevertheless,
as mentioned in the previous section, the better force to current ratio of
the reluctance and hybrid actuator compensates the negative aspect of the
higher self-inductance to some extent because of the lower required current
and especially the hybrid actuator is very promising for future applications.
A detailed FEM analysis should determine the real optimum as a smaller
gap influences the dynamic properties of the hybrid and reluctance actuator
in a clearly interrelated way.

5.4.2.6 Moving mass

The last difference is the moving mass. As Figure 5.40 has been scaled to
the same coil size, it might be concluded, that the Lorentz actuator is at
an advantage when the coil would be the moving part. This however is not
always the case. In a precision system the wires to transport the current
to the coil determine a relatively high stiffness, that is not preferred for
the reason of transmissibility of external vibrations. This can be avoided
by choosing the magnetic circuit as the moving part. Furthermore, due to
the lower force to power ratio, more current is needed for the same force,
which implies a higher power loss. In high power precision systems, the
resulting heat has to be taken away, sometimes even by water cooling, before
5.5. Intermezzo: electric transformers 291

it radiates to other sensitive parts. The necessary cables and hoses and
cooling plates considerably increase the stiffness of connection to the coil
section and necessitate to take the permanent magnet part as mover.
In fact the Lorentz actuator has the largest moving mass of all, when looking
only to the force related to the power. Furthermore, when considering that
the variable reluctance- and hybrid actuator both have the possibility to add
more coil windings, it is clear that these actuators are absolutely preferred
from a high force to power ratio point of view in applications where the
dynamic properties and linearity are less important.

5.5 Intermezzo: electric transformers

As a transition marker between the electromagnetic actuators and Chapter 6

on electronics, the operation principle of an electric transformer is a very
valid candidate. This important component is mostly called just shortly a
transformer, because it is in most cases clear what is meant based on the
context. An electric transformer is a passive component that transforms an
alternating input voltage and current into an alternating output voltage
and current, often with a different magnitude but without changing the
power. This means that a high input voltage with a low input current will
be transformed into a low output voltage with a high output current or the
other way around. A transformer only changes the ratio between the voltage
and current of an alternating signal and this transformation takes place in
two steps. First the electric power is transformed into magnetic power in a
similar manner as with reluctance actuators, but then, instead of transform-
ing the magnetic power into motion, the transformation is reversed again
into electricity at another set of terminals, without a conductive connection
with the source. This principle gives an inherent galvanic insulation, which
is one of the main reasons of the application of a transformer. In mecha-
tronic systems the transformer is mainly used in power supplies, but it is
also used for measuring signals at long distances, where galvanic insulation
prevents interference by ground loops, as will be presented in Chapter 8.
Figure 5.41 shows the layout of a basic transformer with two coils wound
on a shared ferromagnetic yoke, called the transformer core. Because of the
low reluctance of the ferromagnetic material both coils share the same flux.
In principle a multitude of different coils could be applied, either connected
in series, like in the autotransformer or all kept separately to create many
different voltages that are mutually galvanic insulated. To explain the
principle, the shown configuration with only two separate coils is sufficient.
292 Chapter 5. Electromechanic actuators

Transformer core )w
Source Load
Ii Io
+ +
I V
_
+ +
Vi Vo
n1 n2
_

_ _

Figure 5.41: In an electric transformer two coils are wound around the same ferro-
magnetic yoke, called the transformer core. Because they share the
same flux, the voltage of both coils relate to the ratio between the
windings of each coil.

In order to transform the electric current and voltage to another value, the
number of windings per coil is different and a higher number of windings
corresponds to a higher voltage with a lower current level. The primary
windings are connected to the electrical source and the secondary windings
are connected to the load. In principle a transformer can transform electric
energy in two directions and as such the term primary and secondary
windings is not logical, but this distinction has been made for practical
reasons. It is for instance not advisable to reverse the connections of a
transformer that is designed to transform the dangerous mains voltage
into a safe lower voltage, as the result would be an even higher voltage
and eventually the transformer will burn out when the fuses are not blown
before.
The figure also illustrates the galvanic insulation between the primary and
secondary windings as they each consist of insulated wires.

5.5.1 Ideal transformer

In an ideal transformer, the windings are assumed to have no resistance and

the ferromagnetic material is assumed to have infinite permeability and no
magnetic limitations nor energy losses with changing magnetic fields.
To explain the working principle the relation between the magnetic field
and the alternating input voltage Vi over the primary winding n1 of an ideal
5.5. Intermezzo: electric transformers 293

transformer is determined by the same relation as with the self-inductance

of a coil:
dΦw
Vi (ω) = n 1 = V̂i sin ω t (5.124)
dt
where Φw is the flux in the core.
The secondary winding shares the same flux which means that the output
voltage Vo equals:
dΦw
Vo (ω) = n 2 = V̂o sin ω t (5.125)
dt
With this equation the voltages from the primary and secondary windings
appear to relate as follows:
Vo n 2
= (5.126)
Vi n 1

To determine the primary and secondary currents, the flux in the core is
first calculated. From Equation (5.124) the change of flux in the core should
equal:

dΦw V̂i
= sin ω t (5.127)
dt n1

By integrating over time, the flux becomes:

V̂i
Φ w (ω ) = − cos ω t (5.128)
ωn1

The primary current can be calculated from the relation between flux and
current, corresponding with Hopkinson’s law of magnetics:
n1 I i
Φw = (5.129)
ℜ
In the situation without an external load the secondary current is zero and
the unloaded primary current I i,0 equals

Φw (ω)ℜ V̂i ℜ
I i,0 (ω) = =− cos ω t (5.130)
n1 ω n21

A low reluctance of the core results in a low primary current in the unloaded
situation. In an ideal transformer with μr is infinite, I i,0 will be equal to
zero. As soon as a load is present at the secondary windings, the resulting
secondary current I o creates an magnetic flux ΦI,o in the core that adds to
Φw . A primary current will start to flow with a corresponding flux ΦI,i that
compensates the additional flux caused by the secondary current, because
294 Chapter 5. Electromechanic actuators

the total flux has to remain unchanged, corresponding with the primary
voltage. In simplified form without ω t this reads as follows for an ideal
transformer:

ΔΦw = ΦI,i + ΦI,o = 0 (5.131)

The directions of the voltages and currents as defined in Figure 5.41 follow
from Ampère’s and Faraday’s laws. The flux from the output current should
point into the other direction of the flux by the primary current. With the
defined directions the equation becomes:
n1 I i − n2 I 0
ΦI,i + ΦI,o = =0 =⇒ n1 I i = n2 I o (5.132)
ℜ
As a consequence the currents from the primary and secondary windings
relate as follows:
I o n1
= (5.133)
I i n2
The result can be verified by a sanity check, based on the fact that an ideal
transformer does not dissipate power. This means:

P = Pi = I i Vi = Po = I o Vo (5.134)

Which coincides with the found relations.

5.5.2 Real transformer

In reality a transformer is not ideal as the windings have a certain resistance

and the core is not ideally magnetically conducting. This means in the first
place, that the coupling of the flux of the primary and secondary windings
is not 100 %. To improve this coupling, the primary and secondary windings
are preferably wound as close as possible together. To illustrate the effect of
this, the electrical equivalent of a real transformer is shown in Figure 5.42
with the series resistance and uncoupled self-inductance in the primary and
secondary windings. The resistance results in power loss and a decreased
output voltage in case of a loading current. The uncoupled self-inductance
will influence the dynamic behaviour of the system.
The most important limitation of a transformer is however determined by
the magnetic saturation of the core.
With Equation (5.128) the maximum flux and flux density in a transformer
can be derived:
 
 V̂ 
 p 
|Φ̂w | = |B̂A c | =   (5.135)
 ωn1 
5.5. Intermezzo: electric transformers 295

L1 Transformer L
Source 2
Load

+
Coupled
Vi Vo
windings
R1 R2

Figure 5.42: Electrical equivalent of a real transformer.

Because the flux density has to remain below the saturation level Bs , the
cross section of the core needs to be larger than:

V̂p
Ac ≥ (5.136)
Bs ω n 1

The conclusion that can be drawn from this equation is that a transformer
can not be used to transform DC voltages as that would require an infinitely
large core. For AC voltages of low frequencies the number of the primary
windings could be increased, but that will automatically lead to an increase
in the resistive losses. This means that the design of a transformer is a
compromise between power, efficiency and size.
Because of the relation with the frequency it is unavoidable that high power
transformers that have to operate at 50 Hz are very heavy. To solve that
problem, the switched-mode power supply has been created. In this principle
first the low-frequency mains voltage is converted electronically into a very
high frequency of several kHz, before it is transformed to a lower voltage by
means of a small high-frequency transformer. The operation principle is
comparable with the switched-mode power amplifiers, that will be presented
in Section 6.3.3 in Chapter 6.
296 Chapter 5. Electromechanic actuators

5.6 Piezoelectric actuators

5.6.1 Piezoelectricity

The word “piezo” is derived from the Greek word “πιζ o”, which means
“to press”. The piezoelectric effect was first observed by the Pierre Curie
together with his brother Paul-Jacques Curie (1856 – 1941) who also was
a physicist. They discovered that under compression a quartz crystal is
generating an electric potential, which is called the direct piezoelectric effect.
It later has been discovered that this effect also is reversible, which is called
the converse piezoelectric effect or inverse piezoelectric effect: when an electric
field is applied to the piezoelectric material by applying a voltage to the
electrodes at the piezo-surface, the material expands or contracts depending
on the direction of the electric field.
Besides quartz, several other materials are found to show a piezoelectric
behaviour:

• Some natural crystals like Rochelle salt.

• Some natural tissue such as bones and wood.

• Synthetic piezo-ceramics such as Lead Zirconate Titanate (PZT) and

Lead Lanthanum Zirconate Titanate (PLZT).

• Synthetic polymers such as Polyvinylidene Fluoride (PVDF).

The reason for the piezoelectric behaviour is an asymmetry in the molecular

structure of the piezoelectric material which gives an asymmetry in the
charge distribution inside the materials as shown in Figure 5.43. This
charge asymmetry is called polarisation and is caused by large atoms that
are located inside a crystal structure formed by other atoms. The PZT
material that is shown in the figure as example consists of a Lead oxide
(PbO3 ) crystalline structure where a certain amount of Titanium (Ti) and
Zirconium (Zr) atoms are located according to a Perovskite structure. At
elevated temperatures the structure is large enough to accommodate the
Ti/Zr atoms. In that case the material is isotropic and does not possess
piezoelectric properties. When cooling down below the Curie temperature
the shrinking structure is deformed into a tetragonal ferroelectric structure
where the Ti/Zr atoms are “pushed” aside with a simultaneous displacement
of the oxygen atoms. The oxygen atoms are negative charged as they received
electrons from the other atoms by the chemical reaction that created the
5.6. Piezoelectric actuators 297

Pb (+)

+ O (-)

Ti,Zr (+)

Above the Curie temperature Below the Curie temperature

Figure 5.43: The piezoelectric effect in PZT material is caused by an asymmetry in

the crystal structure of the material where large atoms like Titanium
or Zirconium are “frozen” into the structure when cooling down from
above the Curie temperature.

structure which means that a charge displacement occurs, creating a dipole

in the crystal.

5.6.1.1 Poling

Similar to a permanent magnet material, piezoelectric material with the

same polarisation direction is grouped in Weiss domains. In a multi-
crystalline material these domains are normally randomly oriented and the
piezoelectric effect is not observed on a macroscopic level. To utilise the
piezoelectric properties on the macroscopic scale, the Weiss domains have
to be aligned. In some materials like quartz this happens naturally, but
like in PZT a poling process is necessary for this alignment, as is shown in
Figure 5.44. During poling a strong electric field (>2 kV/mm) is applied to
the PZT material while its temperature gets elevated close to the Curie tem-
perature which enables the Weiss domains to align to the external electric
field. When cooling down the PZT material and switching off the external
field, the domains rotate slightly such that the material becomes neutrally
charged again, but the main orientation is maintained and the material has
obtained a remnant polarisation.
After poling, the piezoelectric properties are also observed on the macro-
scopic scale. An external mechanical stress or electric field can disturb the
macroscopic balance of the piezo-material with observable effects on the
outside of the material. In Figure 5.45 on the left the direct piezoelectric
effect is shown, where an external force on the piezo-material generates a
298 Chapter 5. Electromechanic actuators

…"
”!

@$


- - - - -
+ + + + +
- - - -

+
- + + + +
- - - -
+

+
-
+ + + +
- - - -
+ + + +
+

›•
 E

Figure 5.44: Poling process of a piezoelectric material. Without poling the Weiss
domains are not aligned. During poling a strong external electric field
is applied and also the temperature is elevated to the Curie tempera-
ture. After cooling down the material, the electric field is switched off
and the material is “frozen” into a state where the piezoelectric effect
is present on the macroscopic scale as the Weiss domains stay almost
perfectly aligned.

voltage at the electrodes of the piezo. The right side of the figure illustrates
the converse piezoelectric effect where an external voltage, generating an
electric field across the piezo-material, causes the piezo-material to expand
or contract, depending on the direction of the voltage.

5.6.2 Transducer models

Piezoelectricity is the combination of the electric and mechanical behav-

ior of the material, which is given by the equation for the spatial electric
displacement D:

D = εE, (5.137)

with permittivity ε, and the electric field E,

The Hooke – Newton law, as was defined in Chapter 3, relates the strain S
of a material to its compliance C and the externally applied stress T, when
written vectorial in multi dimensions:

S = CT (5.138)
5.6. Piezoelectric actuators 299

A0 + 'A
A0
V
polarizaon
n ?


d0 + 'd
F d0

”

ƒ
<š'
ƒ
<

Figure 5.45: Schematic of the piezoelectric effect. Under compression a voltage

can be measured at the electrodes (direct piezoelectric effect). When
applying a voltage to the electrodes, depending on the sign, the piezo-
material expands or contracts (converse piezoelectric effect).

Combining these for all directions of the material into the so-called cou-
pled equations, according to the IEEE standard on piezoelectricity, the
constitutional laws for a piezo4 -material in strain charge format are:

S = CE T + dT E (5.139)
D = dT + εT E, (5.140)

with the following variables, listed with the corresponding units and dimen-
sions:

S in [-] 6 × 1 strain vector

T in [N/m2 ] 6 × 1 stress vector

E in [V/m] 3 × 1 electric field vector

D in [C/m2 ] 3 × 1 electric displacement vector,

and the following constants:

2
CE in [m /N] 6 × 6 compliance matrix with constant electrical field

d in [m/V] or [C/N] 3 × 6 piezoelectric coefficient matrix

εT in [F/m] 3 × 3 dielectric coefficient matrix with constant stress.

4 Often the term piezoelectric is just shortened into “piezo”.
300 Chapter 5. Electromechanic actuators

In data sheets of piezo-actuators, often the piezoelectric coefficient di j is

given, where the indices i and j indicate the directions of polarisation and
strain. So d33 indicates the strain parallel to the polarisation and is referred
to as the piezo-gain for stack actuators, while d31 indicates that the strain
is orthogonal to the polarisation giving the piezo-gain for instance for tube
actuators.
The constitutional laws describe the physical behaviour of a piezoelectric
element. For a better understanding of how this material works as a macro-
scopic actuator, such as a piezoelectric stack-actuator, these equations can
be converted from the stress and electric field strength into a format that
used the external force F in [N] and applied voltage V in [V], and from strain
and electric displacement into displacement Δ in [m] and charge q in [C].
For simplicity and also because it is close to most practical applications for
piezoelectric actuators, a homogenous material, electric field and stress is
assumed, as well as their respective alignment with the polarisation and
strain. Under these conditions, this conversion gives the following set of
simple scalar multiplications according to the geometry of the actuator with
cross section A and thickness 0 , resulting in:

F = T·A (5.141)
V = E · 0
Δ = S · 0
q = D · A,

The material equations for the stiffness kpz in [N/m] and capacitance C in [F]
are as follows:
A
k pz = (5.142)
s E · 0
εT · A
C = .
0

When using these results in the constitutional laws as defined in Equation

(5.139) the following simple equations are derived:

Δ( t) = k−1
pz · F ( t) + d i j · V ( t) (5.143)
q ( t) = d i j · F ( t) + C · V ( t).

From these equations it can be concluded that an external voltage, applied

to the piezo-actuator, charges the capacitance of the piezo with a charge q
5.6. Piezoelectric actuators 301

and that it causes a displacement Δ. Simultaneously an external force on

the piezo-actuator is manifested by a displacement Δ due to the high but
finite stiffness of the piezo-material and that it generates a charge q as well.
These properties of piezoelectric materials enable these electro-mechanical
actuators to be used for both actuation as well as sensing applications, as
described before.
The transducer model, that is given in this section, describes a linear math-
ematical model of the interplay between the electrical and the mechanical
domain. Although this model is very important to describe piezoelectric ac-
tuators, for analysis as well as system design, it does not cover the non-linear
properties of piezoelectric materials, which are discussed in the following
section.

5.6.3 Nonlinearity of piezoelectric transducers

Although piezoelectric actuators have no moving parts where friction and

backlash can occur, which might otherwise compromise the precision of the
actuator such as in a DC-motor with a gear box, these actuators show other
non-linear behavior such as creep and hysteresis.

5.6.3.1 Creep

A piezoelectric material shows creep after changes of the applied voltage.

This effect is observed as a slow drift, caused by the effect of the actuation
voltage on the remnant polarisation of the piezoelectric material. This creep
effect is logarithmic as function of time according to the following equation:
t
Δ( t) = Δ( t=0.1) · (1 + γ · lg ), (5.144)
0. 1
where Δ( t=0.1) is the displacement 0.1 seconds after application of the change
in the actuation voltage when the fast dynamic transients have settled and
γ is the creep constant, which typically is in the order of 1 – 2 %.
After a few hours, the displacement due to creep can be as large as 10 % of the
total displacement, which may lead to significant positioning errors in open
loop operated positioning systems that are based on piezoelectric actuation.
If the coefficients connected to piezo-creep are known well enough, it can
be compensated (or at least significantly reduced) in an open loop manner
without the use of position sensors. Even with this compensation already
a small uncertainty in these parameters may accumulate in a positioning
error that is not acceptable in precision positioning applications that need
302 Chapter 5. Electromechanic actuators

a long term stability in the order of nanometres or even less. A position

sensor in combination with the piezoelectric actuator can be an alternative
solution while a very small integrating action will fully compensate the
positioning error due to the slow creep of the piezoelectric material.

5.6.3.2 Hysteresis

Hysteresis is a non-linearity that occurs in piezoelectric material and is

manifested by a difference between the motion path for a raising actuation
voltage as compared to a declining actuation voltage, and can be as large as
15 % of the desired elongation of the piezo-actuator. The source for this hys-
teresis is molecular friction, non-ideal material properties and polarisation.
Hysteresis can be described in a way that some Weiss domains align easier
to the external field than others, which causes them to align earlier on the
raising actuation voltage as well as on the declining actuation voltage as
compared to those domains who align not as easy, resulting in the difference
between the raising and falling actuator elongation as a function of the
actuation voltage. This means that the actual elongation of a piezo-actuator
depends both on the currently applied voltage as well as on the last turning
point where the time derivative of the actuation voltage changed its sign. A
demonstration of various hysteresis loops is shown in Figure 5.46, where
a triangular actuation voltage with its amplitude declining over time has
been applied to a piezoelectric stack actuator, while the elongation of the
actuator has been measured by a strain gage sensor.

It is worthwhile to mention here that the hysteresis effect is occurring mainly

in voltage driven piezo-actuation. When the driving signal is controlled
with respect to the charge q that is applied to the piezoelectric actuator,
hysteresis does not occur. This can be explained from the following. The
converse piezoelectric effect corresponds with a charge displacement that
creates the deformation of the crystal while this charge displacement relates
to the voltage by the capacitance of the actuator. The main part of the
hysteresis occurs in the dielectric properties of the material that determine
the capacitance value. When the charge is directly controlled, this dielectric
value has no influence anymore. In Chapter 8 a similar approach will be
presented to measure acceleration by charge amplification when using the
direct piezoelectric effect.
For voltage controlled systems, piezo-hysteresis can be modelled by so-called
hysteresis operators, such as a Preisach model, and can be compensated
for in an open-loop manner. However, just as in the case of creep, several
5.6. Piezoelectric actuators 303

8
6

Strain gage signal [V]

4
2

-2

-4

-6

-8

-10
0 25 50 75 100 125 150
@!
—’

Figure 5.46: Measured hysteresis loops of a piezoelectric stack actuator for various
amplitudes of the input signal over a 150 V range. The strain gauge
signal is proportional to the measured elongation of the actuator.

parameters of the hysteresis model have to be known perfectly, including

the actuation history, in order to achieve a precise compensation of the
hysteresis.
For applications where accuracy is not as important as precise positioning
in a repetitive fashion, such as the scanning motion in an Atomic Force
Microscope, hysteresis is not much of a problem and calibration and pre-
shaping of the driving voltage allows to compensate for most of the hysteresis.
In these applications the repeatability of the position (precision) can easily
be made better than 1 nm if only the rising or the falling voltage branch is
used, even though the difference between these two branches may well be
in the order of a micrometre.
Another way to compensate for piezo-hysteresis is feedback control, just in
the same way as described for the compensation of creep.

5.6.3.3 Aging

Another effect that occurs in piezoelectric material is aging, where due to

de-poling the piezo-gain reduces over time. This phenomenon happens on
the scale of several months or years. When the piezoelectric material is
304 Chapter 5. Electromechanic actuators

used as an actuator, the material gets re-polarised whenever a high field

gets applied, which means close to the maximum drive voltage. In this case
material aging can be neglected. When the piezoelectric material is used
as a sensor or for the generation of charge, where no external voltage is
applied, reduction of the piezo-gain due to aging may have to be considered.

5.6.4 Mechanical considerations

In contrast to Lorentz actuators, which have zero stiffness properties, most

piezoelectric actuators have a high stiffness due to the fact that they are
made of ceramic materials. The mechanical properties of the piezo-actuator
have to be taken into account when designing a mechatronic positioning
system.
The high stiffness properties of piezo-actuators enable to realise very fast
positioning systems with high actuation forces in the range of several kN
and very high resonance frequencies f 0 , which is next to the sub-nanometre
resolution of piezo-actuators one of the main advantages to use these types
of actuators. One disadvantage of these actuators that also is linked to the
property of high stiffness is the limited actuation range, which as a rule of
thumb is that a piezo-ceramic can expand by about 0.1 % of its total length
at the maximum applied voltage.
Another possible disadvantage of the high stiffness can be the high transmis-
sion of vibrations from the support. This makes this actuator less suitable
for applications like the optical pick-up unit of a CD player or stages in a
wafer scanner where disturbances from the vibrating environment have to
be avoided. Piezo-electric actuators are best applied when these vibrations
from the surrounding can be sufficiently reduced by additional methods or
when the total system is so small and integrated that accelerations by the
overall movement of the total system can be neglected.

5.6.4.1 Piezo-stiffness

In the data sheet of each actuator some typical parameters are specified.
Next to the geometric properties and maximum actuation voltage, important
parameters for the mechanical design are the blocking force F max and the
displacement range for the unloaded and unrestrained actuator Δ0 . Often
also the actuator stiffness is given, denoting the small signal stiffness kpz ,
relating to the other parameters via

Fmax ≈ k pz · Δ0 . (5.145)

5.6. Piezoelectric actuators 305

The blocking force Fmax denotes the force that a piezo-actuator generates
when it is confined on both sides by an infinite stiff support. This means
that the entire force generated by the converse piezoelectric effect presses
against the clamping support structure as no elongation of the piezo occurs.
When the piezo-actuator is not confined or restrained, the force generated
by the actuator is available for elongation of the piezo, which means that
the actuation force is in balance with the intrinsic spring stiffness of the
piezo-actuator, resulting in a displacement Δ0 as shown in Figure 5.45.
The given value of kpz is only valid for small signal conditions. For larger
signal amplitudes, another term is superimposed on kpz that is linked to
polarisation effects of the material, where the large signal stiffness can be
up to a factor of two smaller than the small signal stiffness. Furthermore
the output impedance of the power amplifier driving the piezo-actuator has
an influence on the actuator stiffness, as the charge that is built up in the
material due to external forces, if not drained out of the piezo-material,
generates a counter-force. This is the reason why a piezo-actuator with open
electrodes appears stiffer than a piezo-actuator with shortened electrodes .

5.6.4.2 Actuator types

Many different types and shapes of piezoelectric actuators exist, such as

piezoelectric bending elements, tubes, cones and plates, shear-actuators
and stack-actuators. Only two of the most important versions are presented
in some more detail because the presented design considerations are equally
valid for the other actuator types.
Piezoelectric tube actuators enable positioning in all three linear spatial
directions by means of only one single actuator with multiple electrodes
on its barrel as shown in Figure 5.47. The lateral motion in the x− and
y−directions is generated by elongating the piezo-material at one electrode
of the tube while contracting it at the electrode on the opposite side of the
tube where the same signal gets applied but with a reversed sign of the
actuation voltage . This causes a bending of the tube in x− or y−direction
where the displacement Δ x amounts to

2 2 · d 31 · 2 · V
Δx = , (5.146)
π · Di · d

with the tube geometry of length , wall thickness d , and inner diameter
D i , and actuation voltage v.
Actuation in z−direction is done by applying a voltage on all electrodes
on the entire circumference causing a uniform elongation with a vertical
306 Chapter 5. Electromechanic actuators

Figure 5.47: A tube-scanner consists of one tubular piezoelectric element with

different electrodes at the inside and outside. Different electrode
configurations can be used depending on the application. The x− and
y−actuators of the examples on the right each consist of two oppositely
working active areas at each side of the tube. The z− actuator is a
uniform linear expanding element.

displacement Δ z of
V
Δ z = d 31 ·  · . (5.147)
d

Piezoelectric tubes are very popular for 3D-positioning applications with

high precision and resolution, because of their simple design, low cost and
low capacitance, which makes the design of the power amplifier to drive
the actuator easier. A disadvantage of tube actuators, however, is that the
x − y−motion is defining a spherical segment that causes a cross-coupling
also into the vertical z−direction, which can be a compromising factor when
very precise positioning is required.

Piezo-stack actuators as used in the example in Figure 5.48 are multi-layer

actuators, typically build from sheets of d33 piezo-material. The individual
about 50μm-thick sheets have electrodes on both sides to apply the actuation
voltage and are stacked and glued together to form a longer actuator. This
stacking of multiple layers allows to generate larger strokes with smaller
actuation voltages, in the order of 150 V, as the strength of the electric field
is given by the voltage divided by the layer thickness. Without stacking of
the piezo-layers, the actuation voltage for the same actuator length would
5.6. Piezoelectric actuators 307

be in the order of several kV, which is more difficult to achieve with standard
electronics and also poses a larger safety risk. As the poling and actuation
direction is in the stacking direction, the number of layers n chosen will
define the unloaded length 0 and the positioning range of the stack-actuator.
0 eventually is n times the unloaded and unactuated thickness of the indi-
vidual layer h plus the thickness of the electrodes, glue lines and potentially
a layer of unactuated ceramic material at both ends of the stack-actuator
for isolation. With the actuation voltage V the actuator displacement Δ
amounts to:
V
Δ = n · h · d 33 · . (5.148)
h

5.6.4.3 Actuator integration

An appropriate integration of the piezo-actuator in the mechanical structure

of the mechatronic system allows to prevent failure of the actuator as well
as to reduce or amplify the positioning range for actuation.
Pre-loading the actuators by an external compressive force enables the
prevention of fatal tensional loading forces. Due to their ceramic, nature
piezo-actuators are very stiff but also very brittle. Furthermore in case of
stack-actuators, the individual piezo-sheets are glued together generating a
potential weakness at the bond-line. This means that although piezoelectric
material can withstand very high compressive forces, they should not be
loaded in tension, torsion, or shear if not explicitly specified for this purpose.
For applications where the piezoelectric actuator has to position a mass
in both directions with pushing and pulling, it must be assured that the
acceleration at pulling does not generate too large tension forces. To prevent
accidental damage by these tensional forces, piezoelectric actuators are
often pre-loaded with springs. This can either be done with a simple helical
spring that works in parallel with the piezo-actuator. A better alternative
are flexure springs that at the same time also confine the piezo-motion to the
direction of actuation and suppress parasitic rocking and torsional motion
of the actuator. In both cases the stiffness k f of the pre-loading (flexure)
spring increases the stiffness of the total system ktot to

k tot = k pz + k f . (5.149)

While the loading mass has no influence on the (static) positioning range
of the piezo-actuator, the loading spring reduces the maximum actuation
range as the actuation force now has to work against the piezo-stiffness
as well as the loading spring. So a loading spring that is as stiff as the
308 Chapter 5. Electromechanic actuators

x x0 + x

A0 + 'A
b
a

Figure 5.48: Amplification of the elongation of a piezo-actuator by a parallel flexure

mechanism with amplification factor r = b/a.

piezo itself kf = kpz will reduce the actuation range in half. What may be
desirable in some cases is to have a constant pre-loading force that is not a
function of the position but ensures that a constant pre-loading force of the
actuator is present. This can be achieved by an elongated spring with a low
stiffness. Without pre-loading, piezo-actuators can de-laminate when being
contracted too quickly, where the high tensile forces can cause micro-cracks
or even total fracture of the piezo-material.

5.6.4.4 Mechanical amplification

Mechanical amplification by means of leverage enables a larger positioning

range than the maximum stroke of the actuator would allow. Figure 5.48
shows a amplification scheme based on parallel flexures that allows to am-
plify the extension of the piezo-actuator stack Δ to the desired displacement
Δ x with the ratio r of the total lever length b to the position a at the flexure
where the piezo-actuator is attached.

b
r = (5.150)
a
Δx = r · Δ (5.151)
(5.152)

Due to the mechanical amplification r the effective stiffness k tot on the load
side as well as the resonance frequency f res , which depends on the total
positioned mass and the effective system stiffness, are reduced according to
5.6. Piezoelectric actuators 309

the equations

k pz
k tot = (5.153)
r2
f0
f res = . (5.154)
r

5.6.4.5 Multiple directions by stacking

Except for piezoelectric tubes, piezo-actuators are unidirectional actuators.

When positioning in more than one direction is required, individual actua-
tors can be be combined by means of mechanical design, including flexure
mechanisms and stacking of multiple actuators. Figure 5.49 shows three
different implementations of a three degree of freedom piezoelectric scanner
that are based on the stacking of three separate actuators. The typical posi-
tioning range of these scanners can vary from a few hundred nanometres
up to about 150 micrometres, depending on the length of the applied stack.
The first design on top shows a tripod-scanner where three stack-actuators
are glued together orthogonally, forming a tripod. The advantage of this
design is its simplicity and that all three directions are implemented in
parallel. A disadvantage of the tripod design is, similar to the piezo-tube,
that this design also shows a cross-coupling, called scanner bow, between
the x− and y− direction into the z−direction, which is caused by the fact that
for each actuator the mounting point of the two other respective actuators
act as pivoting points.
Two designs that do not suffer from scanner bow are the serial scanner on the
right side of the figure and the flexure scanner on the bottom. In the serial
scanner the individual actuation directions are implemented in a serial way,
stacked on top of each other. This means that the x−actuator moves the
y− and the z−actuator, the y−actuator moves the z−actuator, and only the
z−actuator is moving only the object of interest, shown in the schematic
by the small sample disc on top of the z−actuator. The disadvantage of
the serial design is that the speed performance for the lower positioning
directions is significantly less than for the ones that are stacked on top, due
to the much larger mass that has to be moved.
This difference in performance for the different positioning axes is avoided
by implementing the actuation for the various directions in parallel. This is
achieved in the high speed flexure-scanner for a scanning probe microscope
(SPM) as shown at the bottom of the picture. This design shows a parallel
implementation of the x− and y−directions by combining the individual
310 Chapter 5. Electromechanic actuators

Sliding surface z
x •!

y x
y z
Mechanical ground
Mechanical ground

Tripod-scanner Serial scanner

Figure 5.49: Different designs of piezoelectric scanners, positioning systems for

translation in all three spatial directions. The three degrees of freedom
are each determined by separate piezoelectric actuators. The high-
speed scanner of the picture is able to avoid tensional stresses in the
x− and y−direction by actuating from two sides.

stack actuators that work in a push-pull configuration by means of a flexure

mechanism, resulting in the same performance and resonance frequencies
for the x− and y−directions. The z−actuator is located on top of the flexure
structure and is implemented in a serial way. In the particular application
of the SPM, the highest positioning bandwidth is required in the z−direction
and therefore the total mass moved by the z−actuator has been minimised.
5.6. Piezoelectric actuators 311

5.6.5 Electrical considerations

As explained in the section on piezoelectricity, poling of the piezo-material

is achieved by aligning the Weiss domains with the application of a strong
external electric field as shown in Figure 5.44. Actuation of the piezoelectric
material is done by applying an electric field according to Equation 5.139.
As can be seen from the hysteresis loop in Figure 5.46 only positive actuation
voltages are applied. Piezoelectric material is able to accept weak electric
fields in the opposite direction to about 10 % of the maximum field strength
in the positive actuation direction. A stronger electric field in the opposite
direction can cause de-polarisation of the material and the remnant macro-
scopic polarisation in the material due the poling process would be lost.
For that reason piezoelectric materials that are not specified for symmetric
actuation voltages should only be driven with positive voltages in order to
avoid de-polarisation of the material.

5.6.5.1 Charge vs. voltage control

As shortly mentioned before when presenting hysteresis, there are two

methods to control a piezoelectric actuator. One way is to control the applied
voltage and the second is to control the charge that is applied to the piezo-
actuator. Both have their advantages and disadvantages.
In case of charge control the hysteresis is avoided as the elongation of the
actuator is a linear function of the applied charge. To control the charge, the
piezo-actuator needs to be driven by a charge source. This means that the
current is measured and integrated in the amplifier and actively controlled
to the demanded value by means of feedback. With this method the voltage
at the output of the amplifier is not observed nor controlled which means
that the amplifier has an infinite output impedance. A consequence of this
infinite impedance is that charge that is generated in the piezo-actuator
due to external forces, is not removed from the actuator so it will add to the
charge that was supplied by the amplifier. This results in a higher stiffness
of the actuator in combination with the amplifier, as mentioned also in the
previous section for an open connection. This increased stiffness gives a
higher resonance frequency of the positioning system. The disadvantage of
charge control, however, is that it is not DC-stable, showing more drift, and
in general has more low frequency positioning noise than voltage control
as the charge control involves integrated current-feedback operation of the
amplifier.
In voltage controlled piezo-actuators the design of the power amplifier is
312 Chapter 5. Electromechanic actuators

simpler, a better noise performance can be achieved, and the system shows
less drift. The design of the power amplifier, which in this case has a low
output impedance, and the better noise performance may give an advantage
involving the system costs. A disadvantage of a voltage controlled piezo-
actuator, however, is certainly the hysteresis which in this case may have to
be compensated by other means.
Another interesting design aspect in voltage controlled piezo-actuators is
that the output impedance of the power amplifier forms together with the
capacitance of the piezo-actuator a first-order low-pass filter. On the one
hand this means that a potential bandwidth limitation may have to be
considered in the system design, on the other hand this also gives a design
freedom as adding a resistor in series with the piezo-element not only limits
the maximum current, protecting the power amplifier and the actuator, but
it also allows to shape the frequency response and introduce a roll-off before
the resonance frequency of the actuator, which may be desirable particular
in some open-loop positioning applications.

5.6.5.2 Self-sensing actuation

The transducer properties of piezoelectric material, as given by the constitu-

tional laws and Equation 5.143, allows its simultaneous use as an actuator
and a sensor.
Figure 5.50 shows the electric equivalent diagram of piezoelectric transducer,
in this case the x−direction of the tube scanner depicted in Figure 5.47, which
is integrated in a capacitive bridge circuit. The ratio of the capacitances
in both vertical branches of the capacitive bridge is equal (C1/C2 = Cpz/Cs ).
If the piezo-transducer is considered for a moment only as a capacitor Cpz ,
this implies that the charge distribution in the bridge circuit is in balance
and no differential voltage Vm is measured across the bridge.
The piezo-transducer is in reality not a pure capacitor, because a force
acting on the piezo-element will generate an additional charge q = di j F ( t),
according to Equation 5.143. This changes the voltage across the piezo and
leads to a charge imbalance in the bridge circuit, which can be observed by
the measurement voltage Vm across the bridge. This means that external
and reaction forces, as they occur at the load mass of the piezo, can be
observed at the capacitive bridge circuit and eventually can be used for
force feedback operation by feeding back the measurement voltage Vm to
the power amplifier via a feedback controller, as shown in Figure 5.51. This
feedback is based on the self-sensing capability of the piezo-transducer and
allows for active damping of the resonances without the use of an explicit
5.6. Piezoelectric actuators 313

Piezoelectric
transducer
+ +

Cpz
C1 Vpz
q di j F (t )

Vi -
+
+ - Vm Vs  Vr +

Vr C2 Cs
Vs
- - -

Figure 5.50: Electrical model of a piezoelectric actuator and capacitive bridge cir-
cuit to measure the charge that is generated by external forces on the
actuator.

+ ud (t) Gx(s) x(t)

ui (t) A
- us (t)
Gs (s)
+
um(t)
-
1 ur (t)

1+ r
Gm(s)

LMS

C(s)

Figure 5.51: Schematic of the feedback control for active damping of the piezo-
actuator via self-sensing actuation. The gain γ and the LMS-block
form an adaptive balancing circuit to cancel gain variations due to the
piezo-hysteresis, but have no direct effect on the active damping.

position sensor to measure the piezo-elongation, which is a very cost efficient

implementation to improve the system performance.
A demonstration of active damping to reduce the resonance peak of the
piezoelectric tube-actuator is shown in the Bode plot of Figure 5.52. The
red solid line shows a measured frequency response G x (s) of the piezo-tube
without active damping. The blue line shows the frequency response of the
314 Chapter 5. Electromechanic actuators

Magn
de (abs)
0

-1
10

-2
10
102 103 104

0
Phase [deg]

-90 Modelled undamped

Measured undamped
-180 Modelled damped
Measured damped

102 103 104

Frequency [Hz]

Figure 5.52: Bode plot of the measured and modelled frequency response of both
the undamped and the actively damped piezoelectric tube-actuator.

second-order model that has been fitted to the measured data. Based on
the fitted model a feedback controller C (s) has been designed to dampen the
resonance peak of the tube scanner, resulting in the simulated frequency
response of the controlled model as given by the black line in Figure 5.52.
The green line of the measured frequency response of the transfer G x (s) with
active damping via the self-sensing actuation circuit confirms the reduction
of the resonance peak by almost a factor ten, as predicted in the simulation.
Chapter 6

Analogue electronics in
mechatronic systems

Analogue electronic circuits are indispensable in the realisation of a mecha-

tronic system. Both measurement and actuation need control of electric
signals and knowledge about the important behavioural aspects of electron-
ics is necessary to be able to design a balanced mechatronic system. This
chapter will present low power electronics as applied in measuring, filtering
and control and power electronics that are used for driving electromechanic
actuators.
Generally electronics are divided in two different fields, digital and analogue.
This distinction has developed over the years into two completely separated
fields. Digital electronics consist of electronic switches that are designed to
perform logical functions with binary numbers having only two states, “off”
and “on”. These logical devices range from simple “AND” and “OR” operands
to fully programmable digital signal processors and micro controllers. Be-
cause of this binary simplification it is hardly recognized anymore that all
electronic switches show analogue behaviour. More so, it is the analogue
behaviour, that determines the maximum frequency and switching speed
of these devices. For this reason modern high speed electronic circuits like
processors could only be designed with a clear knowledge of the analogue
properties.
In Figure 6.1 it is shown, that next to power electronics also the control
interface between the controller, the hardware and the observer part from
the measurement sensor to the controller belong to the electronic domain in
a mechatronic system. It will be demonstrated that even fast controllers are

315
316 Chapter 6. Analogue electronics in mechatronic systems

Informaon Energy
Electrical
Feedforward Heat loss
Power
Mechanical
Error
DÎA Power
Posion + ž
setpoint Controller Amplier Actuator Mechanism
-

Measured posion
Sensor
Control electronics + so$ware DÍA Hardware

Electronics

Figure 6.1: Main locations of electronics in a mechatronic system.

best realised directly by analogue electronics. With the focus on these mainly
analogue processes within a mechatronic system, analogue electronics de-
termine the proportional behaviour of a dynamic system by amplification
and filtering of signals.
Also another factor justifies the focus on analogue electronics in this book.
The overwhelming amount of electronic products on the market, from serious
systems to gadgets, work mostly with digital electronics. This abundance
has resulted in a conviction with many people, that electronics are “ready”
and that it is not necessary to investigate this technology anymore in order
to be able to use it. Furthermore, the availability of computer simulation
software has notably resulted in a reduction of the basic understanding
of analogue electronics, because of lack of critical attitude. This has often
resulted in mistakes when designing mechatronic systems. This observation
is underlined by the fact, that even with the mentioned ample availability of
electronics in our society, only very few people, outside the confined field of
electronic expert, are able to combine the knowledge of electronics with the
less abstract mechanical dynamics. It appears to be a real different domain
for many mechatronic designers.
This large chapter consists of three sections. Section 6.1 on passive electron-
ics concentrates on signal manipulation by filtering without amplification.
The section includes impedances, sources, complex elements and passive
filters. Also a parallel is drawn between the dynamics of an electronic filter
and the dynamics of a mechanical system to show the similarity, in order to
enlarge the mutual understanding of electronic and mechanical engineers.
Section 6.2 consists of the important non-linear and active building blocks
of an electronic circuit, followed by the design of low power amplifiers, both
for small signal amplification and active filtering. Section 6.3 presents both
6.1. Passive electronics 317

linear and switched-mode power amplifiers that serve as interface between

the actuator and the controller. As is true for most of the chapters in this
book, the entire field is large enough to fill many books without overlap,
so necessary limitations are applied with a focus on basic understanding,
rather than in depth analysis. Nevertheless the material as presented is suf-
ficiently suitable for the basic design of electronics in mechatronic systems
of average complexity.

6.1 Passive electronics

Analogue electronics consist of an active part and a passive part. The

difference is, whether energy is added to the signal or not. An amplifier
is a typical active component, while a resistor is a typical passive element.
It is important to first understand passive circuits, before presenting the
complexity of active electronics. It will show that even with only passive
parts, several useful functions can be realised. This section starts with
some considerations on electric networks.

6.1.1 Network theory and laws

Electric networks consist of a combination of electric sources and different

passive components. The behaviour of the complete network is defined by
several laws. This section describes these laws, starting with the abstract
definition of the voltage and current source, as they are used as the signal
input for the passive circuits that are presented next.

6.1.1.1 Voltage source

In Chapter 2 the voltage source was introduced. An ideal voltage source has
an internal electromotive force Fe that is equal to the external voltage, even
if a current is delivered to the load. As an example from real life, the mains
power supply socket at home in most countries in Europe behaves almost
like an ideal voltage source of approximately 230 V alternating current (AC)
with a frequency of 50 Hz. This voltage remains almost constant even when
more lighting or stronger appliances are switched on.
The word “almost” refers to the fact that “ideal” does not exist in the real
world. Just imagine a voltage source of any arbitrary voltage that could
deliver less than 1 mA or more than 1 GA without any effect on its voltage.
318 Chapter 6. Analogue electronics in mechatronic systems

_ _ _
+ + +

Figure 6.2: The different symbols of a voltage source, that are used in schematic
diagrams of electronic circuits, show the sign of the voltage by means
of a plus and minus. Sometimes the minus is left away. The most right
version of these symbols is used in this book, as it clearly emphasises
the zero source impedance of an ideal voltage source.

This would require the source to have the capability of delivering infinite
power and that is unfortunately impossible. Nevertheless the example of
the mains supply is a very close approximation of an ideal voltage source as
long as the currents are kept within reasonable limits.
In a real voltage source, the external voltage is not fully equal to the elec-
tromotive force, due to an internal series impedance, originating from its
physical properties. One can think of the internal wiring, the size of chemi-
cal electrodes and storage components and aspects related to the chemical
separation of charge. This series impedance shows an internal electric field,
opposite to the electromotive force, just like inside an external load. This
field depends on the current and results in a change of the voltage differ-
ence over the electrodes of the source. In case of a resistive impedance, the
voltage will be lower at an increased delivered current.
An ideal voltage source would have no internal series impedance and it can
be approximated as a conducting wire when the influence of another electric
source in the circuit has to be examined. This was shown in Chapter 5 about
the interaction between an actuator and an amplifier, where the current
from the two voltage sources could be derived separately by short-circuiting
the other voltage source.
For this reason the version of the symbol that is located at the right in
Figure 6.2 is used in this book when drawing a schematic diagram of an
electronic circuit.

6.1.1.2 Current source

Most electronic circuits can be modelled with a voltage source, but sometimes
it is more practical to work with the other extreme source of electricity, the
current source. An ideal current source delivers a positive or negative
current that is independent of the impedance of the conductive load and as a
consequence the current is independent of the voltage. This functionality is
less easy to understand from reasoning, when using the model of an internal
6.1. Passive electronics 319

Figure 6.3: The different symbols of a current source show the direction of the
current by means of an arrow. The most right version of these symbols
is used in this book because of the clear semantics of the “blocking” line,
representing the infinite internal impedance.

electrical field. Instead one might use the thinking model of a voltage source,
that adapts its electromotive force actively to the load in order to keep the
current at the same level. It will be shown that this indeed is a method to
create a current source by active feedback control. When a current source
is applied in a circuit, another external voltage source, like the movement
induced voltage of an actuator, will not influence the current anymore. This
implies that an ideal current source acts like an open wire for the current
that is induced by other electric sources. In other words, a current source
has an infinite output impedance. It is indeed true that the basic model
of a current source is a voltage source with an infinite electromotive force
and an infinite series resistance. This very artificial model is the reason for
the preferred symbol as shown at the right side of Figure 6.3, as a current
source can be replaced by an open connection in a circuit when the influence
of another electric source has to be examined.
It is also obvious, that an ideal current source can not exist either. An ideal
current source would imply for example that a continuous lightning flash
over an infinite distance could be created. Nevertheless an almost ideal
current source can be created that operates over a limited voltage range and
it is a very important element in mechatronic systems in two ways. First of
all, several sensors in measurement systems behave like a current source
when the voltage is kept within a certain range. But even more important
is the need for a current source when supplying an actuator with electrical
power. In Chapter 5 it was shown, that a low impedance of the source results
in damping and unwanted transmissibility of external vibrations through
the actuator. With the application of a current-source power amplifier as
will be presented in Section 6.3 that effect can be prevented.

6.1.1.3 Theorem of Norton and Thevenin

A real electric source can be modelled in two ways. The first consists of
a voltage source in series with a finite impedance and is postulated by
320 Chapter 6. Analogue electronics in mechatronic systems

Zs Il Il

+
+
+
Zl
VTh Vo Zl IN Zs
Vo
_

_ _

Figure 6.4: Equivalent representation of real electric sources. The Thevenin model
at the left consists of a voltage source with a series impedance. The
Norton model at the right consists of a current source with a parallel
impedance with the same value Zs as with the Thevenin model. The
relation between VTh and I N follows from Ohm’s law: VTh = I N Zs .

the French telegraph engineer Leon Charles Thevenin (1857-1926). The

second method of a current source with that same impedance in parallel
was postulated by the American electrotechnical engineer Edward Lawry
Norton (1898-1983). Both versions are shown in Figure 6.4.
The Thevenin model is again the most straightforward to imagine. An
external current I l through a load Zl will cause a voltage drop over the
internal series impedance Zs , resulting in a lower voltage over the electrodes
of the source. The Norton model can be explained by defining the current
I N to be equal to the maximum current I l,max , that can be delivered by the
source in the Thevenin model. This maximum current occurs when the
electrodes are short-circuited and equals with Ohm’s law:
VTh
I N = I l,max = (6.1)
Z
With this value, the voltage at the electrodes of both systems would be equal
to VTh , when no load is connected to the electrodes. As soon as a load Zl
starts to pull a current I l , the voltage at the electrodes of the Thevenin
model will drop with Zs I l , due to the series impedance Zs . With the Norton
model, the same current I l will reduce the current through the parallel
impedance Zs , giving the same voltage drop over the electrodes.

6.1.1.4 Kirchhoff’s laws

Besides Ohm’s law, as described in Chapter 2, two other important laws

regarding electronic circuits need to be mentioned. They were named after
Gustav Robert Kirchhoff (1824-1887), the German physicist who postulated
them. Kirchhoff’s laws are related to the currents and voltages in a network
of electronic impedances. The first of Kirchhoff’s laws deals with the currents
6.1. Passive electronics 321

I1 I3
node
Z1 _

+
_ V2

+
V1 V3 Z2 I2 Z3
_ loop

+
Figure 6.5: The laws of Kirchhoff. The currents at a node in a circuit add to zero
( I 1 + I 2 + I 3 = 0). Also the voltages over a closed-loop add to zero (V1 + V2 +
V3 = 0). As long as the voltage and current directions are well defined,
these laws help to calculate voltages and currents in complex electrical
networks.

at a connecting point in a network and is based on the understanding that

charge can not be stored at a connection point.

I node = 0 (6.2)

The second of Kirchhoff’s laws states that the sum of voltages over any
closed-loop in an electronic circuit can only equal zero.

Vloop = 0 (6.3)

This is shown as an example in Figure 6.5.

6.1.1.5 Impedances in series or parallel

With the presented laws the currents and voltages in several electric circuits
can be calculated. One example is the situation where different impedances
are combined in series or parallel as shown in Figure 6.6. In the situation
with two impedances in series, the current is shared and the impedance of
the combination becomes:
Vt V1 + V2 I t Z1 + I t Z2
Zserial = = = = Z1 + Z2 (6.4)
It It It
With two parallel impedances, the voltage is shared and the impedance of
the combination becomes:
Vt Vt Vt 1
Zparallel = = = = (6.5)
It I1 + I2 Vt Vt 1 1
+ +
Z1 Z2 Z 1 Z 2
322 Chapter 6. Analogue electronics in mechatronic systems

Vt
V1 V2 Vt

Z1 Z2 Î Zs
It It
serial

Vt Vt
I1
Z1 Î Zp
It It
Z2
parallel I2

Figure 6.6: Impedances in series simply add to the total impedance, as they share
the same current, while the total voltage equals the sum of the voltage
over each impedance. Parallel impedances share the same voltage,
while the currents are divided over the impedances.

6.1.1.6 Voltage divider

Figure 6.7 shows a voltage divider that consists of two impedances.

To calculate the output voltage, first the total current I from the source Vin
into the series circuit Z1 + Z2 is calculated and multiplied with Z2 to get the
voltage over Z2 .

Vi Z2 Vi
I= Vo = I Z2 = (6.6)
Z1 + Z2 Z1 + Z2

The relation between the output voltage Vo and the input voltage Vi equals:

Vo Z2
= (6.7)
Vi Z1 + Z2

The equivalent circuit according to Thevenin is shown in the middle drawing

of Figure 6.7.
The Thevenin voltage equals:
Z2
VTh = Vin (6.8)
Z1 + Z2

To find the Thevenin impedance a special insight is used, stating that this
impedance is equal to the impedance of all paths inside the electric circuit
6.1. Passive electronics 323

I
I
Z1 ZTh

+
+
R1

+
+

+
Vi I VTh Vo I
Z2 Vo Vi
_

+
_ _ Vo
R2
_ _ _

a: Voltage divider b: Thevenin equivalent c: Potenometer

Figure 6.7: A voltage divider consists of two serial impedances and is based on
the principle of the shared current. The voltage Vi gives a current I
determined by the sum of both impedances, while the output voltage
Vo is determined by the current and Z2 only. A special version of the
voltage divider is the potentiometer, that can vary the ratio between
its resistive impedances R1 and R2 to control the output voltage by a
slider.

as observed from the outside. To determine that impedance a voltage source

can be replaced by a conducting wire and a current source can be replaced
by an open wire. In the situation of the voltage divider this means that ZTh
is equal to Z1 and Z2 in parallel as the impedance of the voltage source is
zero:
1
ZTh = (6.9)
1 1
+
Z1 Z2

With these values also the Norton equivalent of the voltage divider can be
modelled according to Figure 6.4. A special execution of a voltage divider
is the potentiometer as shown at the right drawing of Figure 6.7. A poten-
tiometer consists of a resistor from the input to ground with a conductive
slider that contacts the resistor on a location that is determined by an exter-
nal rotating or translating element. The potentiometer creates a variable
voltage at the output as function of the position of the slider.
324 Chapter 6. Analogue electronics in mechatronic systems

Real 
'
0.8
Normalised power '! load
! Ps
Ri
0.6 Output power Po

+
~Pi Io
›Ÿ


+
0.4 ›Ÿ

Vi Io Ro Vo
0.2 _

0
0 2 4 6 8 10

 Ro R
i

Figure 6.8: The maximum power from a real voltage source is 25 % of the intrinsic
power capability of the source and this maximum occurs when the
external resistive load is equal to the resistive value of the source
impedance. The efficiency is then only 50 %.

6.1.1.7 Maximum power of a real voltage source

A real voltage source consists of an ideal voltage source with an internal

source impedance. It is interesting to investigate the maximum power
that the source can deliver, because it is to be expected that the internal
impedance will limit this power. In Chapter 2 it was explained that only
resistive impedances dissipate power, which means that only the resistor
value of the source impedance needs to be taken to calculate the internal
power loss that is dissipated when a current is delivered to an electric load.
This external load is for this simplified example approximated by a simple
resistor and in that case the combined circuit behaves like a resistive voltage
divider as shown at the right circuit diagram of Figure 6.8.
The output power in the load resistor equals the resistor value of the load
times the current squared:
Vi
Po = I o2 R o , Io = =⇒
Ro + Ri
Ro Vi2 Vi2
Po = Vi2 = = (6.10)
(R o + R i )2 R o2 + 2R o R i + R i2 R i2
R o + 2R i +
Ro

This expression goes to zero for Ro = 0, which corresponds with the fact
that then the output voltage is zero, and when Ro = ∞ because then the
output current is zero. The output shows a maximum when Ro = Ri which
can be proven by finding the value of Ro that equals to a zero value of the
6.1. Passive electronics 325

differentiation of the denominator of Equation (6.10) to Ro :

 
R i2
d R o + 2R i +
Ro R i2
=− +1 (6.11)
dR o R o2

which is zero for Ro ± Ri corresponding with a maximum for the output

power.
At the left side of the figure this useful output power is shown graphically
together with the graph of the power delivered by the internal ideal voltage
source (Ps = Vi I o ), the power loss in the internal resistor (Pi = I o2 Ri ) and the
efficiency being the output power divided by the source power. This graph
shows that the source will deliver the maximum power with a very small
value of the load resistance but most of this power is dissipated internally
into heat. This is a phenomenon one can observe when short-circuiting a
battery with sometimes hazardous effects when this is done with a Lithium
based battery! It is also shown that the maximum useful output power
equals only 25 % of the maximum power that the source can deliver.

The above reasoning should not be misunderstood. For a given real source
the maximum power is obtained when the connected load impedance equals
the internal impedance of the source. The efficiency is then however only
50 % and for a higher efficiency the load impedance needs to be significantly
smaller than the source impedance. This means that from an efficiency point
of view it is better to not use a power supply at its theoretical maximum
power capability. With for example a battery it implies that the total energy
should be drained over an extended period of time, which is mostly the case.
When the battery is loaded at is maximum output power the output voltage
would be half and then half the energy would be wasted in heat inside the
battery.
An example where the maximum transfer of power is aimed for is in trans-
port of signals over a long cable that is terminated with its characteristic
impedance, a term that will be explained in more detail in Chapter 8. With
a cable of for instance a characteristic impedance of 50 Ω, the source before
the cable and the load after the cable should both have an impedance of the
same 50 Ω and in that case the maximum signal power is transferred that
the source could deliver. This maximum power level guarantees a maximum
signal to noise ratio.
326 Chapter 6. Analogue electronics in mechatronic systems

Resistor (R)

+
Capacitor (C)

Inductor (L)

Figure 6.9: Symbols of the passive elements or “building blocks” in an electronic

circuit. The two symbols for a resistor are equivalent. The left symbol
of the capacitor is a normal capacitor, where the right symbol is a
polarised electrolytic capacitor. This type needs a unidirectional voltage
difference to keep the high capacitance and avoid current leakage. The
two symbols of the inductor are also equivalent.

6.1.2 Impedances in electronic networks

The complex impedance Z can have aspects of three different properties, a

frequency independent part, corresponding with a resistor, a part where
the impedance decreases with the frequency, the capacitor and a part where
the impedance increases with the frequency, the inductor. Figure 6.9 shows
the symbols of these elements as used in an electronic circuit diagram.
Mostly the elements are indicated as separate items and as a consequence
it looks like they are physically separable. This is not always the case.
In Chapter 5 the electrical equivalent of an electromagnetic actuator was
presented, consisting of a voltage source, an inductor and a resistor. These
elements all were a property of the same element, the coil, so by definition
they are not separable. In electronic circuits mostly the elements are indeed
treated as separate functional building blocks and for most less-critical
applications this is quite sufficient.

6.1.2.1 Resistors

The simplest passive element is the resistor (R in Ohm (Ω))with a frequency

independent impedance, because of its non-complex, real character. The
relation between voltage and current is according to Ohm’s law and as a
consequence of the real impedance, the voltage and current have a propor-
tional relation without time dependency so they are always in phase. For
that reason, the power dissipated in a resistor equals the scalar product of
the momentary voltage and the current value.

Pl = IV (6.12)
6.1. Passive electronics 327

Figure 6.10: An overview of different resistors.

To get an idea about the variety of resistors, Figure 6.10 shows an overview
of several types of resistors that are used in practical electronic circuits.
Resistors exist in an enormous range of resistance values from less than
1 Ω to more than 1 GΩ. This range is available in almost the same size,
which makes them suitable in almost any application. Resistors are realised
with different technologies, ranging from winding a metal wire with a high
resistivity (wire-wound resistor) to the application of a thin carbon- or
metal layer on a non-conductive ceramic round-or-rectangular body. In the
latter case, the correct resistance value is obtained by trimming part of the
resistive layer away with a laser. These carbon or metal film resistors are
for instance recognisable in the figure by the colour code rings that indicate
the value.
Over the years, the electronic industry has made their best effort to achieve
almost ideal properties of all electronic elements. Nevertheless always some
parasitic effect is present, that needs to be considered in critical applica-
tions. A resistor always has some capacitance and self-inductance, while
requirements on these properties differ per situation. Next to the parasitic
effects, the maximum dissipated power and the allowable temperature have
given a boost to the enormous diversity in resistors. High power resistors
are mainly wire wound, because of the insensitivity of the metal to high
temperatures. These high power resistors are recognisable by their size, the
often visible winding structure and the possible mounting to a heat sink, like
shown with the green, the white and gold resistors on the photograph. As a
328 Chapter 6. Analogue electronics in mechatronic systems

+
+
q

+
+

+
E

+
d _
V _
_
_ _ q
_ _ _
_
_
_ _A _ _ _
_ _ _ _

Figure 6.11: Basic principle of a capacitor.

consequence of the windings, wire wound resistors show a relatively high

parasitic self-inductance, that makes them less suitable for high-frequency
applications. With small high-frequency signals, the Surface Mount De-
vices (SMDs) are preferred as they can directly be mounted on a printed
circuit board (PCB) without lead wires, that otherwise could introduce some
self-inductance. SMD components are most frequently used in low power
modern electronics and recognisable in the figure as the little white and
blue blocks in the middle.
Although resistors have also some parasitic capacitance, depending on the
size and the type, that capacitance is only important when really high
frequencies are used like in precision measurement systems where often
low-frequency signals are modulated with a high-frequency signal.

6.1.2.2 Capacitors

A capacitor is a complex impedance as its impedance is frequency dependent.

In electric circuits it is characterised by the symbol C and its value has the
unit Farad (F), named after Faraday. A capacitor is a reactive impedance
because it can store energy in an internal electrical field between a pair of
parallel plates, the electrodes, as shown in Figure 6.11.
In Chapter 2 on electricity, the relation between an electric field and its
potential difference was shown. When two parallel plates with a different
electric charge are positioned very close to each other, like in a capacitor, the
field, corresponding with a charge difference between these plates, will be
approximately homogeneous. This can be reasoned from the superposition
of fields by drawing the field of every infinitesimal charge element and
adding the field arrows as vectors. Field arrows of all charge elements that
6.1. Passive electronics 329

point parallel to the plates will cancel each other out and as a result the
remaining field lines all run orthogonal to the surface of the plates.
With a given electric field between the two electrodes, the potential difference
increases with the distance of the electrodes. This implies, that with a given
potential difference, the magnitude of the corresponding electric field is
inversely proportional to the distance of the electrodes. This also implies,
that the amount of charge difference between one electrode to the other is
inversely proportional to the distance. In a capacitor, the charge difference
is the result of the external voltage and it is not difficult to imagine that
with a larger surface ( A ) proportionally more charge needs to be transported
between the electrodes to retain the same electric field over the entire
surface.
As a last factor that determines the behaviour of a capacitor, the insulating
material between the electrodes has its influence, due to the polarisability
of the material. This polarisability is related to the property of certain
materials, to create an electric dipole. By directing itself along the electric
field, with the positive side directed towards the negative electrode, these
dipoles counteract the electric field and as a consequence they increase the
amount of charge displacement needed to sustain the electric field. This
is in some way comparable with magnetism, where the Weiss domains of
a material with a high relative permeability μr direct to the magnetising
field H, reducing that field relative to the magnetic field B = μ o μr H1 . The
electric property, that corresponds with μ is called the electric permittivity
(ε) with units As/Vm. In vacuum the value is called ε0 and it has a direct
relation with the magnetic permeability of vacuum (μ0 ), that was presented
in Chapter 5 and the speed of light c in vacuum:
1
ε0 μ0 = (6.13)
c2
From this relation, when using the known values of c and μ0 it follows that:
1
ε0 = ≈ 8.8541878176 · 10−12 [As/Vm] (6.14)
c 2 μ0

When the material between the plate is different from vacuum the total
permittivity equals ε0 εr with the relative permittivity or dielectric constant
εr ranging from ≈ 1 for air to < 100 for most solid materials and even higher
for some ceramics and certain special polymers.
1 Due to the different definitions this is somewhat confusing, as the Weiss domains help to

increase the magnetic field B relative to the magnetising field H. It becomes comparable when
the magnetic field is taken as the source of the magnetising field. As mentioned in Section 5.1.2,
the cause and effect between both fields is a philosophical discussion as they simply co-exist.
330 Chapter 6. Analogue electronics in mechatronic systems

1
Sine
Cosine
JQ8 Power
JQ6

JQ4

JQ2

>JQ2

>JQ4

>JQ6

>JQ8

>

0 1 2
Period

Figure 6.12: The power of two multiplied sinusoidal functions with 90◦ phase differ-
ence averages to zero over time. The momentary power is alternating
positive and negative. It represents the energy storage in the reactive
impedance.

With this information it is possible to define the capacitance of a capaci-

tor, which is its capability to store energy in an electric field by a charge
displacement qd at a given voltage:
qd ε A
C= = [F] (6.15)
V d

From a physical point of view it is important to note, that no charge is stored

in a capacitor as a whole, but only at its electrodes. Charging a capacitor
means storage of energy and displacement of charge. Due to the charge
displacement a positive charge q+d is present at the positive electrode with an
equal negative charge q−d at the negative electrode. The mentioned charge
displacement qd equals the integral of the current in the capacitor over time
and because of this time relationship it is interesting to see how a capacitor
behaves, when it is supplied with a changing (alternating, AC) voltage.
The charge displacement is proportional to the voltage ( qd = CV ) and the
current is equal to the rate of the change of charge so it is allowed to say:
d qd dV ( t)
I ( t) = =C (6.16)
dt dt
With this equation the voltage over a capacitor can be calculated at any time
t 1 when the starting voltage V0 at t 0 and the current is given:


t1
1
V ( t) = V0 + I ( t) d t (6.17)
C
0
6.1. Passive electronics 331

The energy stored in a capacitor is calculated by taking the integral of the

power PC needed to achieve the voltage over the time from t0 , where the
voltage is zero, to t1 , where the voltage is V1 , and using Equation (6.16):


t1 
t1 
t1 
V1
dV 1
EC = PC ( t)d t = V ( t) I ( t)d t = V ( t)C dt = C V ( t)dV = CV12 (6.18)
dt 2
t0 t0 t0 0

To determine the frequency domain behaviour of a capacitor, Equation (6.16)

gives the complex impedance, by applying the Laplace transform and defin-
ing V0 = 0:
dV ( t ) V ( s) 1 1
I ( t) = C ⇒ I ( s) = C · sV ( s) ⇒ Z ( s) = = ⇒ Z (ω) = (6.19)
dt I ( s) sC j ωC

This means that the impedance of a capacitor is a complex number, where

the voltage shows a phase lag of 90◦ relative to the current and the impedance
decreases proportional with increasing frequency, a +1 slope! When supplied
from an AC source, the average power over time is zero because of this phase
relationship, which follows from multiplying a sine and a cosine:

A 1 sin(ω( t)) A 2 cos(ω( t)) = 0.5 A 1 A 2 sin(2ω( t) (6.20)

This is illustrated in Figure 6.12 for A 1 = A 2 = 1. The power used to charge

the capacitor alternates with an equal period with a negative power where
the capacitor is discharged.
The capacitance value of a capacitor covers an extremely wide range from
several Farad to smaller than one pico Farad. Initially large capacitors were
very difficult to obtain. Just think of a metal plate of 1 m2 at a distance of
100 μm in air. This results in a capacitance of only ≈ 1 · 10−7 F. For this reason
most practical capacitor values are given in either pF (10−12 ), nF (10−9 ) or
μF (10−6 ) Figure 6.13 shows the large variety of types of capacitors. The
largest capacitance values are achieved with electrolytic capacitors ranging
from a few hundred nF to around 10.000 μF, the big blue Philips capacitor,
the green one and the tube with the nut. Electrolytic capacitors consist of
two thin foils with a large surface of mostly aluminium, covered with an
insulating oxide layer that acts as a high permittivity layer between the
conductive foils. These foils are wound with a thin paper layer in between
that is soaked with an electrolyte that is conductive and keeps the oxide
layer in good order at the occasion of a damaged part in the oxide-layer.
This self-healing effect improves their initially low reliability, but after some
time the capacitance will inevitably reduce as a side effect of the renewed
oxidation. The thin oxide layer with a high relative permeability is the cause
332 Chapter 6. Analogue electronics in mechatronic systems

Figure 6.13: An overview of different capacitors.

for the initially high capacitance and for that reason electrolytic capacitors
are mostly applied in power supplies to store energy.
Recent developments on a variation on electrolytic capacitors with an elec-
trochemical double layer, have increased the maximum capacitance to over
one Farad. Although their maximum voltage per cell is limited, the high en-
ergy density of these super-capacitors makes them applicable in high power
applications as soon as all technological issues, regarding the combination
of multiple capacitors for higher voltages, are solved.
Other smaller capacitors are made by winding thin foils of conductive ma-
terial with intermediate non-conductive foils of a synthetic material with
favourable dielectric properties, (the other round capacitors in the figure).
Their capacitance ranges from around 100 pF to 100 μF. The smallest ca-
pacitor values are realised with single- or multilayer ceramic bodies with a
metal layer as plate material. They can range between 1 pF to 1 μF.
Due to the winding and the thin foil, the electrolytic and foil capacitors both
suffer from a parasitic resistance and some self-inductance. The parasitic
resistance is represented by its equivalent series resistance (ESR), that
limits the minimum impedance at high frequencies and induces a loss of
power. For high-frequency applications, the ceramic capacitor is preferred as
it shows the least parasitic inductance and resistance and it is also available
in an SMD shape.
6.1. Passive electronics 333

6.1.2.3 Inductors

The inductor consists of a coil with a certain self-inductance (L in Henry (H))

and was introduced in Chapter 5 as a property of electromagnetic actuators.
An inductor is often named just shortly a coil or in some older literature a
solenoid. In electronic circuits it acts as the second reactive element with a
frequency dependent behaviour as it can store energy in a magnetic field.
As was presented in Chapter 5, the voltage over an inductor equals:

dΦw dΦw,t
V ( t) = n = (6.21)
dt dt
Where Φw is the flux per winding and Φw,t is the flux integrated over all
windings (nΦ). The self-inductance was defined as the total amount of flux
generated per Ampère:

Φw,t nΦw n2
L= = = ⇒ Φw,t = LI (6.22)
I I ℜ

From this equation, combined with Equation (6.19), the Voltage as function
of time becomes:
dΦw,t d I ( t)
V ( t) = =L (6.23)
dt dt
Also previously presented was the stored magnetic energy E L in an inductor.
Like with the capacitor, it can be calculated by integrating the power over
the time from t0 , where the current is zero until t1 , where the current equals
I 1 for a certain value of L:


t1 
t1 
I1
dI 1 2
EL = P L ( t) d t = I ( t) L dt = L I ( t) d I = LI (6.24)
dt 2 1
t0 t0 0

In the frequency domain for AC currents the impedance of an inductor is

obtained by the Laplace transform of Equation (6.23):
V ( s) L · sI ( s)
Z ( s) = = = sL ⇒ Z (ω) = j ωL (6.25)
I ( s) I ( s)

This means that just like with the capacitor, also the impedance of an induc-
tor is a complex number. In this case however, the phase of the voltage leads
90◦ to the current and the impedance increases with increasing frequency.
Like with the capacitor, the average power over an inductor is zero, when
supplied from an AC source, as was illustrated in Figure 6.12. In every
period, the energy stored in the magnetic field in the inductor alternately
334 Chapter 6. Analogue electronics in mechatronic systems

Figure 6.14: An overview of different inductors and ferromagnetic cores.

increases in the positive power cycle and decreases in the negative power
cycle.
In reality, inductors are all made by winding wires around a core as can be
seen in Figure 6.14. This core can consist of a material with a high relative
permeability to achieve a high self-inductance or a low permeability for
reason of linearity. For very high frequencies even a core can be omitted
and only a section with a spiral track on a printed circuit board can suit the
purpose. With some numbers in a real coil with n = 100 windings, a core
diameter ( A ) of 10*10 mm, a core length ( l ) of 100 mm and a μr of 100, the
following relatively low value of the self-inductance is obtained:

n2 n2 μ0 μr A 104 · 4π · 10−7 · 10−4

L= = = = 1.25 · 10−3 [H] (6.26)
ℜ l 0. 1
For this reason the values of inductors are often given either in mH or μH.
Practical inductors suffer from many different parasitic properties. The
intimately wound windings show a significant parasitic capacitance and
resistance. When high currents are involved and a large self-inductance is
needed, the size of the inductor will be large. Also the core gives many prob-
lems. Generally the relative permeability is depending on the momentary
flux density by saturation. This means in practice, that the self-inductance
is reduced with larger currents. And last but not least, the changing flux
inside the core will induce eddy-currents within the core. This phenomenon
was explained in Chapter 5 with the eddy-current ring in a Lorentz actuator.
The eddy-currents in the core of an inductor will reduce the effective self-
inductance and will result in dissipated energy, because the eddy-currents
run in a resistive material. The eddy-currents can be limited by using a
6.1. Passive electronics 335

non-conductive core material. An ideal group of materials in that respect are

the ferrites, sintered material consisting of insulated, partly oxidised iron
particles. Also a plastic compound of Iron particles with a thin oxide layer
can be used. For low frequencies the eddy-current losses can be sufficiently
limited by laminating the ferromagnetic core in thin sheets. Because the
eddy-currents are running perpendicular to the flux lines the lamination
should be in the direction of the flux to have the maximum effect.
Another loss in the core is caused by the magnetic hysteresis. Contrary to
permanent magnet materials, the ferromagnetic core of an inductor should
ideally show no hysteresis. It is interesting to know that ferrites both can
be designed to have a high hysteresis to create a permanent magnet, like
mentioned in Chapter 5 as for having an extremely low hysteresis for in-
ductor cores, all at acceptable cost levels. For laminated cores an alloy of
iron with 2 – 5 % of silicon is a well-known material with a low hysteresis
and is widely used in power transformers, rotating electrical machines and
electromagnetic actuators.

6.1.3 Passive filters

Electronic filters are frequently applied in mechatronic systems for selecting

frequency areas of interest, the pass-band, for rejecting frequency areas
that introduce noise and other disturbances, the attenuation-band or for
correcting phase properties in feedback systems. The field of electronic filters
is very large and consists of passive filters with only resistors, capacitors
and inductors and active filters with amplifiers. Passive filters are mainly
used, when it is not possible to use an amplifier. This is for instance the case
between an amplifier and an actuator like with the switched-mode power
amplifiers of Section 6.3. But also with sensors at remote places at high
temperatures passive filters can be useful.

6.1.3.1 Passive first-order RC-filters

The most basic filter configuration is the RC-filter that is derived from the
voltage divider of Figure 6.7, where one of the impedances is a frequency
dependent component, like for instance a capacitor. The first filter to explain
this principle is a first-order low-pass filter, that can be used for instance to
reject noise above the cut-off frequency or corner-frequency ω0 of the filter.
This filter consists of a capacitor and a resistor as shown in Figure 6.15.
As the impedance of the capacitor decreases with increasing frequency it can
336 Chapter 6. Analogue electronics in mechatronic systems

 J

Magnitude (abs)

Magnitude [dB]
>J
*
J >^J

>ZJ
*^
Z1=R J >J

+
>\J
J
+

Phase [deg]
Z2
Vi jZC Vo >\
_
>|J
_ J
J
J

J
^
J
Z
J


Frequency [Hz]

Figure 6.15: An RC low-pass filter network and the Bode plot of its transfer function
with R = 15 kΩ and C = 0.1 μF, giving a time constant τ = 1.5 ms. The
increasing impedance Z2 of the capacitor reduces the magnitude of the
output signal at frequencies above the corner-frequency f 0 = ω0/2π ≈
100 Hz, where the magnitude of Z2 becomes larger than Z1 .

be reasoned, that this filter reduces the amplitude of the higher frequencies
above the frequency, where the impedance of the capacitor becomes smaller
than the impedance of the resistor. To derive the transfer function F (ω) of
this filter in the frequency domain Equation (6.7) of the voltage divider is
used:
Vo Z2
F (ω ) = = (6.7)
V i Z1 + Z2
With Z1 = R , Z2 = 1/ j ωC , the transfer function becomes:
1
Vo j ωC 1
F (ω ) = = = (6.27)
Vi 1 j ωRC + 1
R+
j ωC
At this point, the time constant τ = RC also called the RC -time can be intro-
duced. Because the transfer function of Equation (6.27) is dimensionless,
RC is inversely proportional to the frequency ω, so it has the unit of time.
It is also known from control theory, that the time constant τ defines the
step response of a system starting at zero:
& t
'
V = Vst 1 − e− τ (6.28)

With the time constant τ, Equation (6.27) becomes:

1
F (ω ) = (6.29)
j ωτ + 1
6.1. Passive electronics 337

 J

Magnitude (abs)
>J

Magnitude [dB]
* >^J
J
>ZJ

>J
*^
1

+
J
>\J
Z1 |J
jZC
+

Phase [deg]
Vi Z2=R Vo \

_
_ J
J
J
J

J
^
J
Z
J


Frequency [Hz]

Figure 6.16: An RC high-pass filter network and the Bode plot of its transfer func-
tion with R = 15 kΩ and C = 0.1 μF, giving a time constant τ = 1.5 ms.
The increasing impedance Z2 of the capacitor reduces the magni-
tude of the output signal at frequencies below the corner-frequency
f 0 = ω0/2π ≈ 100 Hz, where the magnitude of Z2 becomes larger than
Z1 .

To determine the amplitude at the corner-frequency ω0 = 1/τ, the absolute

value of the transfer function is calculated by taking the root of the sum of
the squared real and imaginary term of the numerator and denominator:
 $
12 1
|F (ω0 )| = = = 0.707 . . . (in dB: -3 dB) (6.30)
1 + 12
2 2

The Bode plot of this filter is shown in Figure 6.15 for an example with
R = 15 kΩ and C = 0.1 μF. This gives a time constant of τ = 1.5 ms and a
corresponding corner-frequency of ω0 = 1/τ ≈ 660 rad/s and f 0 = ω0/2π ≈ 100 Hz.
The −1 slope of the magnitude with -20 dB per decade is as expected with a
first-order system.
It is important to note that the characteristics of a passive filter are influ-
enced by the load after the filter. When for example the filter of Figure 6.15
would be loaded with a resistive load, equal to Z1 , two effects would be ob-
served. Firstly the gain of this filter in the pass-band would have a value of
0.5, because the load and Z1 would act like a frequency independent voltage
divider. Secondly the cross-over frequency would be a factor 2 higher. Both
effects can be calculated by first determining the equivalent Thevenin source
voltage and impedance of Z1 , combined with the load and then combine that
system with the capacitor.
To prevent this effect, these filters are commonly combined with buffer am-
338 Chapter 6. Analogue electronics in mechatronic systems

plifiers with a very high input impedance and low output (source) impedance.
These will be presented later.
The second filter of this kind is the high-pass filter, as used for instance to
reject the DC value of a signal. Its basic design consists also on a capacitor
and a resistor, but their location is exchanged as shown in Figure 6.16. It
is clear, that at low frequencies the high impedance of the capacitor will
prevent low-frequency current to flow into the output resistor.
Also with the high-pass filter Equation (6.7) is used to derive the transfer
function F (ω):
Vo Z2
F (ω ) = = (6.31)
Vi Z1 + Z2

With Z1 = 1/ j ωC , Z2 = R and τ = RC the transfer function becomes:

Vo R j ωRC j ωτ
F (ω ) = = = = (6.32)
Vi 1 j ωRC + 1 j ωτ + 1
+R
j ωC

The difference with the low-pass filter is the j ω term in the numerator,
corresponding with the differentiating Laplace variable s, which gives +1
slope with 90◦ phase lead and a 0 dB corner-frequency at ω0 = 1/τ. This term,
combined with the transfer function of the low-pass filter, results in the
Bode plot of Figure 6.16 for the same values of the resistor and the capacitor.
Also with this filter a loading impedance over V0 will influence its character-
istics. Even though the high-frequency pass-band gain will not be changed,
the load impedance comes parallel to Z2 resulting in an increase of the
cross-over frequency with more attenuation at lower frequencies.

6.1.3.2 Passive higher-order RC-filters

Often the attenuation of unwanted frequencies is required to be stronger

than what is obtained with simple first-order filters. To create filters with a
higher attenuation, a higher order filter needs to be applied. In principle
this can be done by cascading several first-order filters. Figure 6.17 shows a
third order configuration with a special measure to prevent mutual interfer-
ence between the three filter sections. Normally the simple combination of
identical passive filter sections introduces several problems, because each fil-
ter section would determine a load for the preceding section. To reduce these
effects, the value for the resistor of each following segment can be chosen for
instance a factor 10 higher than the resistor value in the preceding section,
as shown in the figure. This helps reducing the problem, but it means that
6.1. Passive electronics 339

Z1=R1 Z2=R2 Z3=R3

+
1 1 1

+
Z2 Z4 Z6
jZC1 jZC2 jZC3
Vi Vo
_

R3 10 ˜ R2 100 ˜ R1 C1 10 ˜ C2 100 ˜ C3

Figure 6.17: Creating a third order low-pass filter by cascading three first-order
filters is possible with minimal mutual load. It is however better
to introduce buffer amplifiers, as the output impedance of the filter
becomes very high.

the effective output impedance of the combined filter would become very
high. For that reason it is preferred to insert intermediate buffer amplifiers
with a gain of one, like shown in Figure 6.18. How these amplifiers are made
will be presented later in this chapter. For the explanation of the principle,
it is sufficient to accept that these amplifiers enable to combine the separate
sections without mutual influence, because they combine a very high input
impedance with a very low output impedance. In that case the combined
transfer function becomes a simple multiplication of the transfer functions
of the separate filters:
1
F ( s) = (6.33)
( j ωτ1 + 1)( j ωτ2 + 1)( j ωτ3 + 1)

For an RC-time τ1 = τ2 = τ3 = 0.01 s, the result is shown in the Bode plot of

Figure 6.18. The figure clearly shows that the slope in the attenuation band
is increased, but the transition to the pass-band is far from ideal. At the
corner-frequency f 0 = ω0/2π ≈ 100 Hz, the attenuation is already 9 dB and
the pass band bandwidth at -3 dB is at ≈ 0.5× the corner-frequency, with a
very gradual transition to a slope of −3.
As will be presented in the Section 6.2.5 on active filters, this trade-off
between the properties in the attenuation-band and the pass-band can be
optimised by the application of positive feedback, which is equal to increasing
the imaginary term of the poles of the transfer function. When combining
only first-order filters all poles are located on the real axis which implies a
very high damping of the higher order filter. By relocating the poles, the gain
at the cross-over frequency can be increased while keeping an acceptable
steep slope in the attenuation band. It is worthwhile to add to this statement
that it is not possible to create an ideal filter with an infinitely steep slope,
340 Chapter 6. Analogue electronics in mechatronic systems

Z1=R G=1 Z3=R G=1 Z5=R G=1

+
+
1 1 1
Vi Z2 Z4 Z6
jZC jZC jZC Vo
_

1 J
Magnitude (abs)

Magnitude [dB]
-1
J >^J
*^
J >J
-3
J >`J
*
J >}J
J
Phase [deg]

>|J

>}J

>^{J
1 ^ 3
J J J
Frequency [Hz]

Figure 6.18: Cascading three first-order low-pass filters with intermediate buffer
amplifiers results in a very gradual transition between the pass band
below the corner-frequency and the attenuation band above f 0 . At the
corner-frequency the attenuation is already 9 dB. The Bode plot is
drawn with R = 15 kΩ and C = 0.1 μF, corresponding with f 0 ≈ 100 Hz .

without phase and magnitude effects in the pass band and it will be shown
that these electronic dynamics have a close relationship in behaviour with
mechanical dynamics.

6.1.3.3 Passive LCR-filters

When it is not possible to use amplifiers, higher order filters can also be
created by applying the other complex impedance, the inductor in an LCR-
filter. When the resistor in the previous filters is replaced by an inductor,
the filter effect will be enlarged, because the impedance of the inductor
increases as function of the frequency.

Parallel resonant LCR-filters

In Figure 6.19 this second-order filtering effect is shown for a low-pass filter
configuration, where the capacitor and inductor have a value of respectively
100 μF and 25 mH.
6.1. Passive electronics 341

The same approach as with the first-order filters is used to derive the transfer
function:
Vo Z2
F (ω) = = (6.34)
Vi Z1 + Z2

With Z1 = j ωL, Z2 = 1/ j ωC the transfer function becomes:

1
Vo j ωC 1
F (ω) = = = (6.35)
Vi 1 ω2 LC + 1
j ωL +
j ωC

The corner-frequency of this filter is defined by:

1
ω0 =  (6.36)
LC
For the example of Figure 6.19 with a capacitor of 100 μF and an inductor

of 25 mH, the corner-frequency becomes ω0 = 4 · 105 = 630 rad/s and f 0 =
ω0/2π = 100 Hz.

With the defined value of ω0 the transfer function in the frequency domain
is written as:
1
F (ω) = 2
(6.37)
ω
− +1
ω20

Similar to the mechanical mass-spring system in Chapter 3, this transfer

function shows a resonance with infinite amplitude at ω = ω0 , while the
corner-frequency of the filter is equal to the undamped natural frequency
of the electronic resonator. In the mechanical situation, the height of this
resonance is tuned by adding a damper. In electronic filters a resistor can
act as a damping element as it dissipates energy. This resistive damping
can be applied in two different ways. The first possibility is to apply the
resistor at the output of the filter as a load. In that case the resistor will
absorb more energy, when its value is low. The other possibility is to apply
the resistor in series with the source, and in that case the resistor will take
more energy when the resistance is high. This interesting contradictory
effect will become more clear when the two configurations are modelled.
The first configuration is the previously shown LC low-pass filter with a
damping load resistor at the output, as shown in Figure 6.20. This configu-
ration is also called a parallel-resonant filter, because from the point of view
of the impedances, the voltage source is a short-circuit. This means that all
impedances are essentially connected in parallel.
342 Chapter 6. Analogue electronics in mechatronic systems

3
J `J
^

Magnitude (abs)
J

Magnitude [dB]
J
1
J ^J
1 J
-1
J >^J
*^

+
J >J
Z1 jZ L J
+

>\

Phase [deg]
1
Vi Z2 >|J
jZC Vo
_ >Z\

_ >}J
J
1
J
^
J
3

Frequency [Hz]

Figure 6.19: A second-order LC low-pass filter network and the Bode plot of its
transfer function with C = 100 μF and L = 25 mH, giving a corner-
frequency f 0 =100 Hz. Because of the lack of damping, the magnitude
at the corner-frequency becomes infinite.

Again the same approach can be taken as with the previous filters to derive
the transfer function, but in this case the impedance to ground consists of
two parallel impedances Z2 and Z3 . The parallel configuration is noted with
symbol (∥):
Vo Z2 ∥ Z3 1 1
F (ω ) = = = =  (6.38)
Vi Z1 + Z2 ∥ Z3 Z1 1 1
+ 1 Z1 + +1
Z2 ∥ Z3 Z2 Z3

With Z1 = sL, Z2 = 1/sC , Z3 = R , the transfer function becomes:

Vo 1 1
F (ω ) = =  = (6.39)
Vi 1 L
j ω L j ωC + + 1 −ω2 LC + j ω + 1
R R

The same relation for the corner-frequency ω0 = 1/LC is used and the
following relation for the quality factor Q and damping ratio ζ is introduced:


1 C
Q= =R (6.40)
2ζ L

With these defined values the the transfer function in the frequency domain
becomes:
1 1
F (ω ) = 2
= 2
(6.41)
ω 2ζω ω ω
− +j +1 − +j +1
ω20 ω0 ω20 ω0 Q
6.1. Passive electronics 343

J

Magnitude (abs)
1

Magnitude [dB]
J
>J
*
J >^J

>ZJ

+
*^
Z1 jZ L J
0
>J
+

>\

Phase [deg]
1 Z3
Vi Z2
jZC R Vo >|J

_ >Z\

_ >}J
 ^ Z
J J J
Frequency [Hz]

Figure 6.20: A second-order damped LC low-pass filter network and the Bode plot
of its transfer function with 100 μF, L = 25 mH and R = 16 Ω. These
values result in a corner-frequency f 0 = 100 Hz and a damping with Q =
1, ζ = 0.5. A lower value of the resistor corresponds with an increased
level of damping.

This represents a well controlled dynamic performance, when the damp-

ing ratio is chosen somewhere between 0.5 and 1. The resulting transfer
function is flat in the pass-band and shows an almost constant slope in the
attenuation-band. The Bode plot of this function has a clear resemblance
with the compliance presentation in Chapter 3 of a mechanical damped
mass-spring system. This mechanical versus electrical dynamic analogy
will be shortly summarised a bit later in this chapter after completion of
the presentation on the passive filters.
The parallel resonant configuration can also be used in a high-pass filter,
by exchanging the location of the capacitor and the inductor, like shown in
Figure 6.21.
Again the same approach is applied to derive the transfer function:
Vo Z2 ∥ Z3 1 1
F (ω) = = = =  (6.42)
Vi Z1 + Z2 ∥ Z3 Z1 1 1
+ 1 Z1 + +1
Z2 ∥ Z3 Z2 Z3

With Z1 = 1/sC , Z2 = sL, Z3 = R , the transfer function becomes:

Vo 1 −ω2 LC
F (ω) = =  = (6.43)
Vi 1 1 1 L
+ +1 −ω2 LC + j ω +1
j ωC j ω L R R

Also the same relation
# for the corner-frequency ω0 = 1/ LC and damping
1
ratio Q = /(2ζ) = R C/L is used and the transfer function in the frequency
344 Chapter 6. Analogue electronics in mechatronic systems

10

Magnitude (abs)
1 0

Magnitude (dB)
>J
-1
10 >^J

>ZJ
-2
10

+
>J
1 180
Z1
jZC
+

135

Phase (deg)
Z3
Vi R Vo 90

_ Z2 jZ L 45

0
_ 10
1
10
2
10
3

Frequency (Hz)

Figure 6.21: A second-order damped LC high-pass filter network and the Bode
plot of its transfer function with 100 μF, L = 25 mH and R = 16 Ω.
These values result in a corner-frequency f 0 = 100 Hz and a damping
with Q = 1, ζ = 0.5. A lower value of the resistor corresponds with an
increased level of damping.

domain becomes:
ω2 ω2
− −
ω20 ω20
F (ω ) = = (6.44)
ω2 2ζω ω2 ω
− +j +1 − +j +1
ω20 ω0 ω20 ω0 Q

Like with the first-order filters, the only difference with the low-pass version
is the numerator, where in this case the frequency terms are squared,
corresponding with s2 . This results in a 180◦ phase shift (the minus term).
The resulting Bode plot is in a dynamic sense completely comparable to the
low-pass version with the same component values.

Series-resonant LCR-filters

When a series resistor is applied to create the damping, the series-resonant

circuit of Figure 6.22 is obtained.
The first observation in this configuration is that simultaneously a second-
order low-pass and a high-pass filter2 is obtained. All impedances are in
series so they share the same current and the voltages add together to V0 .
With these findings, the low-pass filter transfer function in the frequency
2 The voltage over the resistor behaves like a first-order band-pass filter around the corner-

frequency but it is left to the reader to examine this further.

6.1. Passive electronics 345

Z1=R

+
+
Vi Z2 jZ L
Vo,HP
_
_

+
1
Z3
jZC
Vo,LP
_

Figure 6.22: A second-order damped LC low-pass and high-pass filter network in

series configuration. A higher value of the resistor corresponds with
an increased level of damping.

domain becomes:
1
Vo,LP Z3 j ωC 1
FLP (ω) = = = = (6.45)
Vi Z1 + Z2 + Z3 1 −ω2 LC + j ωRC + 1
R+ + j ωL
j ωC

Likewise the transfer function of the high-pass filter function is:

Vo,HP Z2 j ωL −ω2 LC
FHP (ω) = = = = (6.46)
Vi Z1 + Z2 + Z3 1 −ω2 LC + j ωRC + 1
R+ + j ωL
j ωC

When comparing these equations with Equation (6.39) and Equation (6.43),
the difference in the damping term RC versus L/R in the previous filters is
clear. The quality factor for the series resonant filter can be determined by
replacing the term in the parallel filter by the following steps.
The parallel quality factor equals

C R
Q=R = CL (6.47)
L L

When exchanging L/R by RC the quality factor for the series resonant filter
becomes:
 
1  1 L 1 R C
Q= CL = with the damping ratio ζ = = (6.48)
RC R C 2Q 2 L
346 Chapter 6. Analogue electronics in mechatronic systems


When these terms are applied, together with the unchanged ω0 = 1/ LC
the same equations in the frequency domain are obtained as previously
determined in Equation (6.41) and Equation (6.44).

6.1.4 Mechanical-electrical dynamic analogy

The previous presentation of passive second-order filters with an inductor,

a capacitor and a resistor is so much similar to the presentation of the
dynamics of a damped mass spring system that one wonders how these
elements from electronics can be linked to their counterpart in the body,
spring and damper of a mechanical dynamic system. Indeed this connection
is possible and even more than one analogy can be defined of which one is
presented here. In this equivalent dynamic system analogy, a parallel L-C-R-
filter, driven by a periodically changing current source, is compared with the
velocity response of a damped mass spring system driven by a periodically
changing force, acting between the reference plane and the mass as shown in
Figure 6.23. In this comparison first the current I is assumed to correspond
with the force F . The logic of this assumption is based on the fact that at
a connecting node of a mechanical system all forces add to zero, like in an
electrical node all currents add to zero according to Kirchhoff’s law. Also in
a series configuration of mechanical elements the Force is shared, while in
a series configuration of electric elements the current is shared. The next
assumption is that the voltage V corresponds with the relative velocity v.
This assumption is based on power as voltage times current and velocity
times force both are equal to power. As a next step the impedances can
be compared. Ohm’s law in electricity can be replaced by a mechanical
counterpart: ẋ = F Zm , where Zm equals the mechanical impedance. Like
with electric impedances the three mechanical elements have their own
typical impedance. The most straightforward is the damper with ZD = 1/c,
corresponding with the resistor ZR = R . The frequency dependent impedance
of a Capacitor ZC = 1/ j ωC corresponds with the mechanical impedance of
the body ZB = 1/ j ω m. As the last of the elements, the spring corresponds
with the inductor, not only because of their almost equivalent symbols but
also by their behaviour. The mechanical impedance of the spring equals
ZS = j ω/k corresponding with the impedance of the inductor, ZI = j ωL. These
impedances show, that the capacitance (C ) corresponds with the mass ( m),
the self-inductance (L) with the compliance of the spring (Cs = 1/k) and the
Resistor (R ) with the inverse of the damping coefficient ( c) of the damper.
To check this all in combination, the behaviour of the electrical circuit is
first determined. According to Ohm’s law the output voltage of the circuit
6.1. Passive electronics 347

+
Z1 jZ L m
I Z3 Vo x
1
Z2
jZC c F k
_

x V
Zm œ Ze
F I

Body œ Capacitor Spring œ Inductor

x 1 V 1 x jZ V
ZB œ ZC
x ZS œ ZL jZ L
FB jZ m I jZC FS k I

Damper œ Resistor
x 1 V
ZD œR
FD c I

Figure 6.23: The analogy between a mechanical and electronic dynamic system,
where voltage corresponds with velocity and current with force.

equals:

Vo = I ( Z1 ∥ Z2 ∥ Z3 ) (6.49)

From this follows:

Vo 1 1 j ωL
= = = (6.50)
I 1 1 1 1 1 L
+ + + j ωC + −ω2 LC + j ω + 1
Z1 Z2 Z3 j ωL R R

As a next step, the values for ω0 and the damping ratio ζ for both the electrical
parallel resonant filter and the mechanical configuration are determined by
mutually interchanging the mentioned terms:
 
1 k 1 1 L c
ω0 =  = and ζ = = =  (6.51)
LC m 2Q 2R C 2 km

With these terms, Equation (6.50) gives the following transfer functions for
both the electrical and mechanical configuration:
1
jω
Vo j ωL ẋ k
(ω) = and (ω) = (6.52)
I ω2 2 j ωζ F ω2 2 j ωζ
− 2+ +1 − 2+ +1
ω0 ω0 ω0 ω0
348 Chapter 6. Analogue electronics in mechatronic systems

^
10

= 0.005 Q = 100 R = 1/c = 60

= 0.05 Q = 10 R = 1/c = 6

= 0.5 Q=1 R = 1/c = 0.6
1
10
Magnitude Zm , Z e

0
10

>
10

>^
10
Phase [deg]

>|J
^ 3 4
10 10 10
Frequency [Hz]

Figure 6.24: Bode plot of the equivalent electrical impedance Ze = V/I and the
mechanical impedance Zm = ẋ/F for 1/k = L = 10−4 and m = C =
2.5 · 10−4 . Below the corner-frequency of 1 kHz the line follows the
spring/inductor line with a +1 slope and 90◦ phase lead. At the corner-
frequency the impedance is real, which means that the level is only de-
termined by the dissipative element, the resistor or the damper. Above
1 kHz the line follows a −1 slope, according to the mass/capacitance.

As an illustration, a Bode plot of the impedance of both systems is shown in

Figure 6.24 with some values for the example. In the electrical domain the
plot shows a resonance at the corner-frequency f 0 = ω0/2π = 1000 Hz, with a
resistive maximum depending on the resistor value. An application of such
a circuit can be in a filter to reject a small frequency area, a band-reject
filter, when the voltage divider of Figure 6.7 is used and Z1 is replaced by
this circuit. When Z2 is replaced by the circuit a band-pass filter will result,
that attenuates all frequencies outside of the corner-frequency of the circuit.
The equivalent transfer function of the mechanical impedance corresponds
with the velocity response of Equation (3.58) from in Chapter 3. The +1
slope with a 90◦ phase lead below the corner-frequency corresponds with
the spring impedance in the mechanical system and the impedance of the
inductor in the electronic circuit. The −1 slope with a 90◦ phase lag above the
corner-frequency corresponds with the mass impedance in the mechanical
system and the impedance of the capacitor in the electronic circuit.
6.1. Passive electronics 349

In Table 6.1 these findings are summarised.

Table 6.1: Electrical and mechanical analogy.

Mechanical Electrical

Force (F ) Electric Current ( I )

Speed (v) Electric Voltage (V )
Mass ( m) Capacitance (C )
Compliance (1/k) Self-inductance (L)
Damping coefficient ( c) Conductivity (1/R )
Energy (E = 0.5 · mv2 ) Energy (E = 0.5 · CV 2 )
Power (P = Fv) Power (P = IV )
350 Chapter 6. Analogue electronics in mechatronic systems

6.2 Active electronics

Next to linear passive components, electronic circuits consist of different

non-linear and active electronic building blocks. At the beginning of the age
of electronics, the first of these active elements were based on the interaction
between free electrons and charged metal elements in vacuum tubes. The
discovery of the semiconductor diode effect in 1897 by the German physicist
Karl Ferdinand Braun (1850 – 1918) started a cycle that would change the
face of the world. The next significant event was the patenting in 1930 of
a field effect transistor by the American physicist Julius Edgar Lilienfeld
(1882 – 1963) who based his invention on a theoretical analysis without
practical verification. A real working transistor was created more than 15
years later at Bell labs in 1947 by the American physicists William Bradford
Shockley (1910 – 1989), John Bardeen (1908 – 1991) and Walter Houser
Brattain (1902 – 1987) who won the Nobel prize for this important achieve-
ment. The real breakthrough in electronics came after the independent
invention in 1958 of the integrated circuit by Jack Kilby (1923 – 2005), an
American physicist, working at Texas Instruments, and by the American
physicist Robert Noyce (1927 – 1990), co-founder of Fairchild Semiconductor.
With these inventions the vacuum tubes became rapidly obsolete for pro-
fessional applications. Presently the entire electronic field is divided in
discrete components, with individual functional elements and the integrated
circuits (ICs), that contain a multitude of different elements integrated on

Figure 6.25: Integrated circuits with passive components on a surface mounted

printed circuit board.
6.2. Active electronics 351

Diode

Bipolar transistor

Figure 6.26: Two different symbols of a diode and a bipolar NPN (left) and PNP
transistor. The arrows point towards the current conducting direction.

one single piece of Silicon, called a chip. These chips are packaged in mostly
plastic housings with terminals, like shown in Figure 6.25. In the last few
decades many versions of semiconductor-based discrete components have
been created, ranging from the semiconductor diode and the bipolar tran-
sistor to special functional parts, like avalanche and zener diodes, the diac
and triac, the field effect transistor (FET), the Metal Oxide Semiconductor
FET (MOSFET) and manymore.
Integrated circuits are also diversified over many different applications,
with a clear distinction between analogue and digital ICs. The best example
of an analogue IC is the operational amplifier because of its applicability in
a wide variation of electronic functions. Digital ICs are used for logical oper-
ations, digital processing and computer memory. In the past decades, their
complexity has grown tremendously and presently they sometimes contain
hundred million transistors or more, all connected by a huge network of
integrated wiring. It would be impossible to give a complete overview of all
elements and diversity of electronics in the context of this book. For that
reason only the three most common examples of the discrete components
are explained more in depth, the Silicon diode, which is a basic non-linear
element, the Silicon bipolar transistor, that is used to create active amplify-
ing electronics and the MOSFET that is used in digital electronics and in
switched-mode power amplifiers.

6.2.1 Basic discrete semiconductors

The electronic circuit symbols for a diode and a bipolar transistor are shown
in Figure 6.26. A diode allows an electric current to flow in one direction,
but blocks it in the opposite direction. It can be seen as an electronic version
of a check valve. Like a check valve that needs a pressure to open and shows
a potential leakage in the blocking direction, also electric diodes do not
352 Chapter 6. Analogue electronics in mechatronic systems

Si Free electron Si Hole

Si Si
Si Si
Si Si Si
Si Si Si
Si Si Si Si
Si Si
Si Si Si Si
Si P B
Si Si Si
Si Si
Si Si Si Si
Si Si Si
Si Si Si
Si Si
Si Si Si
Si Si Si
P Si B Si
Si Si
Si Si Si
Si Si Si
Si Si Si Si
Si Si
Si Si
Si Si
Si Si
N doped Silicon P doped Silicon

Figure 6.27: N- and P-material, created in a Silicon crystal structure, where one
of the atoms is exchanged by Phosphor, to create N-material with
one free electron or by Boron, to create P-material with one missing
electron, a hole.

display a perfect on-off directionality but have a more complex non-linear

electrical characteristic. In principle a diode is a passive semiconductor as
it does not add energy.
The bipolar transistor is a real active component, that can be used to amplify
signals by converting energy from a power supply into a signal with more
power. Next to amplification a transistor can also be used as an electrically
controlled switch. In this section first the working principle of a bipolar diode
and transistor are explained based on the physics behind semiconductors in
order to understand their main properties.

Semiconductor physics

Semiconductors are chemically situated in the middle of the periodic system,

which means that they can attain the ‘ideal’ state either by receiving n
electrons, or by giving away the same amount of electrons. Examples of
semiconductor materials are Silicon and Germanium. The most widely used
material is Silicon, because of its excellent electronic properties, combined
with its ample availability as the main constituent of sand.
Pure Silicon has exactly 4 electrons per atom in its outer shell. Silicon would
need to either receive 4 additional electrons or give these 4 electrons away
in order to reach the “semi ideal” state. In the bulk material the 4 electrons
of each atom are shared with their neighbours to achieve a true chemical
bond. For this reason these electrons are firmly bound in the material
structure. This immobility of the electrons means that pure Silicon behaves
like an insulator. In order to achieve a semi-conducting state, it is necessary
6.2. Active electronics 353

to exchange some of the Silicon atoms by near neighbours in the periodic

system, that either have three electrons in their outer shell, like Boron (B),
or five electrons, like Phosphor (P). This exchange of atoms is called ‘doping’
and the resulting structure is shown in Figure 6.27.
Depending on the doping material, the doped Silicon shows a different
behaviour. In case of doping with Phosphor atoms, additional not-bound,
free electrons are made available. For that reason the silicon becomes an
N-material. When the doping is done with Boron, so called holes are created,
missing electron places. These electrons and holes are called charge carriers,
as the doping in the material composition creates the possibility for the semi-
conductor to conduct current by means of these carriers. This conductivity
is caused by the mobility of the free electrons and the holes. The mobility
of the electrons is easy to understand, as they are not rigidly connected to
an atom core, while the mobility of the holes is a bit strange and can be
explained as follows. The presence of the hole disturbs the energy states in
the material in such a way, that the normally bound electrons of the adjacent
Silicon atoms tend to jump onto that hole, when an internal electric field
drives it in that direction. As a result this electron leaves an empty place,
a new hole, where the electron came from. This new hole can be seen as a
displacement of the original hole in the opposite direction of the “jumped”
electron. For this reason, charge transport by holes takes place in the same
direction as the current. Note that due to this mechanism, also the charge
transport by holes is in fact achieved by electrons. The difference is, that the
charge transport by the holes in the P-material is done by normally bound
electrons from the Silicon, while the charge transport by the electrons in
the N-material is done by the free electrons, coming from the doping atoms.
Before entering the next section, where N- and P-material will be combined
to create a diode, it is important to add some additional remarks:

• Although the P-material has free holes and the N-material free elec-
trons, those materials are not charged as they originate from a not
charged doping atom added to the Silicon crystal. This is fully identical
to an uncharged metal that contains many free electrons.

• The concentration of the doping elements is very low. This concen-

tration varies between approximately 10−6 and 10−8 depending on the
purpose.

• The speed of the conduction in P-material is lower than the speed of

conduction in N-material, due to the indirect conduction mode of holes.
This difference is observed in high-frequency applications.
354 Chapter 6. Analogue electronics in mechatronic systems

Figure 6.28: Combining N- and P-material creates a depletion layer, because elec-
trons of the N-material recombine with the holes of the P-material.
As a result, the N-material becomes positively charged and the P-
material becomes negatively charged, according to the large plus and
minus sign. This results in a measurable potential difference between
both materials. The shown small plus and minus signs represent the
remaining free moving charges.

6.2.1.1 Semiconductor diode

When N- and P-material are brought into electrical contact, the electrons of
the N-material and the holes of the P-material will diffuse into each other
and the electrons will fill the holes at the interface area. As a result an
intermediate layer is created without free electrons and holes. This process
is called the recombination of the electrons and the holes and this layer is
called the depletion layer, as shown in Figure 6.28. This layer behaves like
an insulator due to the lack of free movable charge carriers.
Because the P-material loses positively charged holes and the N-material
loses negatively charged electrons, the P-material becomes negatively charged
and the N-material becomes positively charged, creating an electromotive
force, that can be detected by its external potential difference.
In Figure 6.29 it is shown what happens, if an external voltage source is
applied between the combined N- and P-material. At the left the external
voltage is in the same direction as the electromotive force, caused by the
recombination of the charge carriers. The external source adds electrons
to the P-material and takes electrons from the N-material. This results
in even less free charge carriers, and because of this enlarged depletion
layer no current can continue to flow. The thickness of the depletion layer
becomes proportional to the external voltage and the internal electromotive
force that corresponds to this large depletion layer is in equilibrium with
the externally applied voltage.
At the situation of the right of Figure 6.29, the external voltage is applied in
a direction opposite to the electromotive force of the recombined depletion
6.2. Active electronics 355

Figure 6.29: A semiconductor diode conducts current in one direction, when the
external voltage is directed opposite to the internal electromotive force
and blocks it in the other direction. In the conductive mode both the
electrons and the holes contribute to the current.

layer. In that situation electrons are added to the N-material and taken
from the P-material, compensating the effect of the recombination. As a
result the depletion layer disappears and a conductive path is created where
both the holes and the free electrons contribute to the current.
It is obvious that these effects don’t happen instantaneous without any
transition phenomena between conducting and blocking. This is illustrated
in Figure 6.30, that shows the real characteristics of a typical Silicon semi-
conductor diode. In the conductive direction it is to be expected, that the
internal electromotive force of the depletion layer determines a certain
threshold level, that the external voltage has to surpass in order to make
the diode conducting. Furthermore, due to non-ideal material properties,
always some leakage current is present in the blocking direction. This
leakage current also reduces the external voltage that can be measured
at the terminals of a not connected diode below the electromotive force
caused by the depletion layer. For a silicon based diode the threshold voltage
Vth is around 0.6 – 0.7 V in practical situations. This value is strongly
temperature dependent, because it is determined by the mobility of the
charge carriers, that cause the depletion layer. This mobility is increased at
elevated temperatures.
Another side effect is caused by the fact, that the depletion layer is a non-
charged (insulating) layer between two parallel plates, which acts like a
parasitic capacitor. The need to charge this capacitor determines the current
needed to change the diode from the conductive state into the blocking
356 Chapter 6. Analogue electronics in mechatronic systems

8 ˜103 2 ˜109

6 ˜103 1.5 ˜109

4 ˜103 1 ˜109
Id [A]

Id [A]
2 ˜103 5 ˜1010

0 0 Ideal
Real
2 ˜103 5 ˜1010

-0.2 0 0.2 0.4 0.6 0.8 -0.2 -0.1 0 0.1 0.2 0.3
Vd [V] Vd [V]

Figure 6.30: Real versus ideal characteristics of a silicon diode. A threshold voltage
Vth of approximately 0.6 V is present in the conductive direction at
moderate current levels above the mA range, as shown at the left.
At extremely low current levels in the nA range also some leakage
current occurs with a voltage in the reverse direction.

state and vice versa, when the current is reversed. This is an important
characteristic, because diodes are frequently used to rectify alternating
currents into direct currents in power supplies and high-frequency signal
demodulators.
Figure 6.31 shows an example of the most frequently used rectifier circuit,
the bridge-rectifier, that consists of 4 diodes and a storage capacitor. As can
be seen in the drawing both the positive and the negative voltage from the
input results in a unidirectional positive voltage at the output. The large
storage capacitor will flatten the pulsating voltage into a DC voltage with a
ripple that depends on the current in the load and the capacitance value of
the storage capacitor. This can be filtered further with a passive low-pass
filter, as was presented in the previous part.
In the previous chapter the electric transformer was presented. One of the
conclusions was that a high-frequency AC voltage is necessary to be able to
reduce the size. The rectification of a high-frequency AC voltage requires
very fast switching diodes in order to reduce the power dissipation, related to
the continuous charging and discharging of the intrinsic parasitic capacitor
from the depletion layer. This dissipation is caused by the unavoidable
parasitic source resistance in the circuit, that causes the power loss. Another
application of fast diodes is in the switched-mode and resonant amplifiers
that will be presented in the section on power amplifiers.
Apparently even this single component can be optimised for many different
applications.
6.2. Active electronics 357

+
+ R Vo _ R Vo
_ _
Vi Vi
_

+
posive cycle negave cycle

with capacitor

+
+ without capacitor
R Vo
+

C
_
Vi
_

storage capacitor output voltage ripple

Figure 6.31: Diode bridge rectifier for AC voltage. Both the flow of current at
the positive and the negative cycle of the input voltage result in a
unidirectional current in the load, because of the diode configuration.
By adding a storage capacitor a DC voltage is obtained with only a
limited AC ripple. The capacitor is charged in the positive slope of the
rectified input signal and discharged by the load in the negative slope.

6.2.1.2 Bipolar transistors

A bipolar transistor is a combination of three layers of successive N- and

P-material. Depending on the order of the layers with N- and P-material
between the emitter and the collector, the transistor is an NPN or a PNP
type.
To explain the working principle, a simplified thinking model of an NPN
transistor is shown in Figure 6.32. When a positive voltage is applied
between the collector and the emitter, the interface between the collector
and the base will be non-conductive and the interface between the base and
the emitter is in the conductive mode. The net result is that the transistor
is not conducting as the layers are connected in series.
As soon as a positive current is applied between the base and the emitter,
electrons start to move from the emitter to the base in the conductive
direction of the base-emitter PN-junction. The base is extremely thin and the
358 Chapter 6. Analogue electronics in mechatronic systems

Figure 6.32: Combining N-, P- and N-material gives an NPN transistor. Without
base current I b the transistor is not conducting. A base current in
the conductive direction of the base-emitter junction causes a large
number of electrons to pass the collector-base barrier, resulting in a
collector current that is larger than the base current.

high voltage at the collector determines a strong electric field, accelerating

the electrons that move from the emitter towards the base to fill the holes
that were created by the base current. Because of the accelerated state of
these hot electrons, not all of them recombine with holes in the base, but the
majority breaks through the barrier between the base and the collector. In
this way an electron flow is generated from emitter to collector, equivalent
to a current flow from collector to emitter. This current is larger than the
current in the base that caused it. Increasing the current on the base ( I b )
will trigger more emitter electrons, that break through the collector-base
junction.
With this principle it is explained that a transistor acts as a current amplifier,
with a current-amplification ratio β, also called hfe , between the collector
and base current:

Ic
β = h fe = (6.53)
Ib

The example was given for an NPN transistor, where the electrons play the
main role. For that reason in that configuration the electrons are called the
majority carriers and the holes are the minority carriers as they only play a
role in the base current.
With a PNP transistor it is just the other way around. The holes are the
majority carriers and all currents run in the opposite direction of the NPN
6.2. Active electronics 359

Ic Ic

N: Collector P: Collector
P: Base N: Base

Ib Vc-e Ib Vc-e
N: Emier P: Emier

Figure 6.33: Current directions in an NPN and PNP transistor.

example as shown in Figure 6.33. Because of the different character of

the majority carriers, free electrons or holes, both types of transistors are
never fully identical, aside of the current direction. It is easier to achieve
a high β with an NPN than with a PNP transistor and holes are slower
than electrons. Still this choice enables the electronic designer to apply
these different transistors in almost symmetrical circuits that control both
positive and negative currents.

6.2.1.3 MOSFET

The name MOSFET means Metal Oxide Semiconductor Field Effect Tran-
sistor. It is a transistor that operates because of a externally controllable
electric field, while it uses a metal oxide to insulate the input where the
controlling signal is applied. The electrodes of a MOSFET have different
names than with a bipolar transistor. The input electrode is called the Gate
and is equivalent in function to the base in a bipolar transistor. The Drain
is equivalent in function to the collector as it drains the electrons in an
N-channel MOSFET, which is the most common version and the Source is
equivalent in function to the emitter as it is the source of the electrons in
an N-channel MOSFET. in order to distinguish it from a voltage or current
source, the Source, Gate and Drain of a MOSFET will be written as a name
starting with a capital.
MOSFETs are the most common semiconductors in integrated logic circuits,
like microprocessors and memory, because of their speed, size and possibility
to work with very low power supply voltage levels. The speed is high because
the current flow takes place by majority carriers only, which are electrons
in N-channel MOSFETs and as mentioned, charge transport by electrons is
360 Chapter 6. Analogue electronics in mechatronic systems

Drain Field Gate Gate Source

Contact Oxide Oxide 
ƒ Contact
ID ID
D
Gate
Channel n-Source Channel,
n-Drain
Substrate
G V
GS

p-Substrate 0 S
0 VT VGS

Gate + + + + +
Drain N + Channel + N Source N N
++ + + + + P
P + + + P P
+ + + +
+ + + +
+ + + +

O< On

Figure 6.34: A MOSFET operates by an electric field. With an N-channel MOSFET

the channel is located in the substrate, made of P-type Silicon, between
the Drain D and the Source S, both made of N-type Silicon. In an
enhancement-mode MOSFET the channel is normally not conducting,
but when a positive voltage is applied between the insulated Gate and
the Source the corresponding electric field will drive away the holes
from the P-type channel and attract electrons from the Source, that as
a result will flow towards the Drain. The relation between the Gate to
Source voltage and the Drain current is less steep than with a bipolar
transistor and the behaviour is more like a variable resistor.
(Courtesy of International Rectifier)

significantly faster than the indirect charge transport by holes. Next to the
small signal type MOSFETs, also high power versions have been developed,
that consist of millions of separate small MOSFETs connected in parallel.
They are applied in high power switching applications and will be further
presented in Section 6.3.
The structure of one type of N-channel MOSFET, the horizontal enhancement-
mode type is shown in Figure 6.34. Many different types of MOSFETs exist,
the depletion-mode MOSFETs are normally conducting, but the normally
non-conducting enhancement-mode version is most frequently used. The
horizontal type is the most easy to use for explaining the working principle,
while the vertical type is used to create power MOSFETs.
The substrate of a horizontal N-channel enhancement-mode MOSFET con-
sists of P-doped Silicon. The Drain and the Source are made of N-doped
Silicon, separated by the substrate. The area in the substrate between
the Drain and the Source is called the channel. Unlike with the bipolar
6.2. Active electronics 361

transistor, the depletion layer between the P-and N-material plays no other
role than to insulate the Drain from the substrate when the voltage on the
Drain is positive relative to the substrate. Normally the junction between
the Source and the substrate is in the conductive direction during operation
and for that reason they are mostly internally connected.
The real functionality takes place in the channel area around the Gate. The
Gate is made of metal and is electrically insulated from the substrate, the
channel and the Source by a metal-oxide layer. This gave the name “MOS”
to the device, although often the metal of the Gate is replaced by high doped
poly-crystalline Silicon and the oxide by Silicon oxide, because of the better
compatibility with the Silicon manufacturing process. This makes however
no difference for the working principle.
The oxide layer between the Gate and the channel is very thin and when a
positive voltage is applied to the Gate, relative to the Source, the holes in the
channel are pushed away due to the electric field by the Gate voltage, while
the electrons are attracted. As a result, a large number of free electrons
become available in the channel. Effectively the original P-doped substrate
becomes a virtual N-material, with equal properties as the material of the
Drain and the Source. This way, the channel becomes a conductive path
between the Drain and the Source, with a width that is determined by the
voltage level on the Gate. The resistance of the channel depends on the
width and with this principle a MOSFET behaves like a controllable resistor.
The depletion mode MOSFET works in the opposite way. In that case the
channel is in principle conducting by a different doping, and with N-material
of the channel a negative voltage on the Gate will repel the electrons, creat-
ing a depleted area with an increased resistance. Because this type is not
used in amplifiers for mechatronic systems it will not further be treated.
In spite of the quite simple functionality of MOSFETs, bipolar transistors
are still more often used in analogue amplifiers and for that reason first
amplifiers with bipolar transistors will be presented. The power MOSFETs
will return when switched-mode amplifiers are presented because of their
excellent switching performance.

6.2.2 One transistor amplifiers

With both bipolar transistor types, several electronic functional building

blocks can be realised. Many years ago all analogue electronics consisted
of a combination of these configurations that were only built with discrete
components. Presently most analogue electronics are realised with ICs
called Operational Amplifiers. In order to better understand the working
362 Chapter 6. Analogue electronics in mechatronic systems

principle of these integrated amplifiers, first the basic building blocks are
explained in this section.

6.2.2.1 Emitter follower

The first circuit is the emitter follower from Figure 6.35. It is also called
a common collector configuration as the collector is directly connected to
the common power supply. The principle of the emitter follower is based on
the current-amplification of the applied NPN transistor. It can for instance
be used as the buffer amplifier in the cascaded passive first-order filter of
Figure 6.18 to match a high impedance input with a low impedance output.
In principle its operation is quite straightforward. The collector is connected
to a positive voltage with in this example a value of +15 V. The emitter is
connected to a more negative voltage via an emitter resistor. For the negative
voltage a value of −15 V is chosen to be able to amplify alternating signals.
The input voltage is applied on the base. When this input voltage has a
value in between the positive and negative supply voltage, the base emitter
junction is in the conductive mode. Like with a Silicon diode, the conducting
base-emitter junction has a threshold voltage Vb−e,th of ≈ 0.6 V. For that
reason, the emitter voltage will be approximately 0.6 V more negative than
the input voltage.
The difference between the emitter voltage and the negative supply voltage
determines the emitter current I e through the emitter resistor.
Ve − Vp− Vi − 0.6 + 15
Ie = = (6.54)
Re Re
This current is equal to the sum of the collector current and the base current:


1
Ie = Ib + Ic = Ic +1 (6.55)
β

For high values of β, the collector current becomes almost equal to the emit-
ter current and for this section the base current will further be neglected.
The current in the emitter resistor is necessary to be able to deliver both
positive and negative currents to a load, as shown in Figure 6.35.b, and is
called the idle current. When the input voltage at the base is an alternating
sinusoidal signal, like shown in Figure 6.35.c, then the emitter voltage
follows the input voltage with a constant difference of ≈ 0.6 V. The external
load will pull either a positive current or a negative current from the node
between the emitter and the emitter resistor. The positive current is deliv-
ered by the transistor, on top of the current flowing into the emitter resistor.
6.2. Active electronics 363

Vp 15 V V p 15 V

Ib
5 V Vth | 0.6 V
Vi
Ie | Ic Io ! 0
+
Vb-e,th
0.6 V Vo

+
+

_
Vi Vi 0V
Ie | Ic
+

+
_ Ve _
Vi  0.6_

Rload
Io  0 Vo
Re

Re
Vclip
_ 5 V

Vp 15 V Vp 15 V

a: no load b: with load c: signals

Figure 6.35: The NPN transistor emitter follower is a current amplifier used to
achieve a low output impedance. The emitter voltage follows the
input voltage that is applied to the base with a difference of ≈ 0.6 V.
When a load is connected, the positive output current is delivered by
increasing the emitter current of the transistor. The negative output
current comes from the emitter resistor Re by a reduction of the emitter
current of the transistor. Ultimately the emitter current is zero and
the maximum negative voltage is reached, showing a phenomenon
called “clipping” at a level that is determined by the load in relation
to the emitter resistor and the negative power supply voltage.

The negative current is delivered by the resistor by reducing the current

from the transistor. This can be done until the emitter current is zero. The
transistor is not capable of delivering a negative current and as a result the
amplifier will show a clipping effect at a maximum negative level that can
be calculated from the voltage divider that is determined by Re and Rload
with the negative supply voltage:

R load
Vclip = Vp− (6.56)
R e + R load

This single transistor emitter follower is only suitable in situations, where

no large output currents are required. This limitation is both due to the
mentioned clipping at the negative cycle and also because of the heat dissi-
pation in the emitter resistor and transistor by the idle current, even when
no signal is amplified. This single-ended Class A configuration is especially
inefficient with high power amplifiers. For that reason a more symmetrical
solution is often applied, where the negative current is delivered by a PNP
transistor. This configuration is shown in Figure 6.36. The input voltage
364 Chapter 6. Analogue electronics in mechatronic systems

Vp 15 V

5 V Vi Vo with Vbias
Io ! 0
Vi

+
+

0V
Vbias

+
_ _
| 1.2 V Vo without Vbias

Rload
Io  0 Vo
_ 5 V

Vp 15 V

Figure 6.36: A symmetric push-pull emitter follower. The positive current is de-
livered by the NPN transistor and the negative current by the PNP
transistor. A bias voltage is added to avoid non-linearity at the transi-
tion between a positive and a negative current, that would otherwise
be caused by to the threshold voltage of the base-emitter junction of
the transistors.

is applied simultaneously to the base of both transistors. In the positive

cycle of the signal the NPN transistor operates almost identical as in the
single-ended configuration, but it only delivers current to the load, not into
an emitter resistor. In the negative cycle the PNP transistor takes over
the task to deliver the output current. This so called Push-Pull class B
configuration has one little problem that is caused by the threshold voltage
of the base emitter junction of ≈ 0.6 V per transistor. This gives a strong
non-linearity at the transition between the positive and negative current,
called cross-over distortion. Because the gain becomes almost zero at 0 V
input, feedback can not solve that non-linearity and it is necessary to reduce
it by applying a bias voltage between the bases of both transistors. The
tuning of this bias voltage is quite critical, due to the large temperature
dependency of the base-emitter threshold voltage. In most cases a certain
limited idle current is set running through both transistors and the bias
voltage is temperature compensated to add robustness.
This configuration is called Push Pull class AB.

6.2.2.2 Voltage amplifier

Based on the findings with the emitter follower, two other configurations
are shown in Figure 6.37, the variable current source and the voltage ampli-
6.2. Active electronics 365

Rload

Rload

Rc
+ Io Io
_
_
Vp

+
Ic
+
Vb-e,th Ie
0.6 V Vo
+

+
_
Io
Vi Vi

+
+

_ _

Re
Re

Vi  0.6 Vi  0.6

_ _ _

a: variable current source b: voltage amplier

Figure 6.37: Two configurations where the load is applied to the collector.
a: With the variable current source, the collector current is only
determined by the input voltage and the emitter resistor. It has a very
high output impedance.
b: A voltage amplifier is created from a variable current source, by
inserting a resistor between the positive power supply and the collector.
The voltage over the resistor is proportional to the current and the
resistor value. With a high value for Rc , voltage amplification is
obtained at the collector output.

fier. Both circuits belong to the same type and are called collector follower
or common emitter configurations, because the output is taken from the
collector and the emitter is either directly or almost directly connected to
the common ground.
Like with the example of the emitter follower, the NPN transistor is used as
the representative example. For the PNP transistor only the signs of the
voltages and currents need to be reversed to get the same results.
The circuit of Figure 6.37.a. acts as a variable current source with a negative
current and works as follows: As long as the voltage of the collector is kept at
any value more positive than the voltage at the base, the emitter current is
only determined by the input voltage and the emitter resistance like with the
emitter follower. With a high value of the current-amplification ratio β, the
base current can be neglected and the collector current becomes equal to the
emitter current. As a consequence of this reasoning, the collector current is
not dependent on the collector voltage, which is the defined behaviour of a
current source.
The current source will prove to be very useful in the design of a differential
366 Chapter 6. Analogue electronics in mechatronic systems

amplifier, but first it is used to create a single transistor voltage amplifier by

including a collector resistor between the collector and the positive supply
rail, as shown in Figure 6.37.b.
The current in the collector determines the voltage over the resistor resulting
in an output collector voltage Vo equal to:

(Vi − Vth )R c
Vo = Vp − I c R c = Vp − I e R c = Vp − (6.57)
Re

For the AC part of the input signal the constant values of Vth and Vp don’t
matter, which means that this configuration gives an AC voltage amplifica-
tion of:
Vo Rc
Ga = =− (6.58)
Vi Re

With an ideal transistor this gain is independent of the values of the resistors.
The minus term means that the sign of the signal is inverted. Although this
looks equal to a 180◦ phase shift this effect is not related to a phase delay
but only with a sign reversal.
Like all components that were presented, also the transistor is not ideal.
Its current-amplification ratio is not infinite nor constant and the character-
istics are non-linear because of the base-emitter junction, as was shown for
a diode in Figure 6.30. Furthermore, like the diode, also the transistor has
its dynamic limitations, that will be dealt with after the section about the
operational amplifier.
Finally, the emitter voltage does not follow the base voltage perfectly, but it
shows a certain output resistance Rs,e with a value of approximately:

0.025
R s,e = [Ω] (6.59)
IC

while this value is also temperature dependent.

Due to this finite emitter impedance the amplification ratio of Equation (6.58)
can not be made infinite by choosing Re equal to zero. In that case the inter-
nal emitter output resistance Rs,e determines the maximum amplification.
With for instance a idle current of 1 mA, this output resistance is 25 Ω and
with a supply voltage of 20 V and a 10 kΩ collector resistance, keeping the
collector voltage at about 10 V, a gain of 400 is achieved. This number is
reduced by the not infinite current-amplification ratio and, because of the
non-linear base-emitter threshold voltage to current ratio, this amplifier is
very non-linear. In fact a larger emitter resistor linearises the amplifier at
6.2. Active electronics 367

the expense of a reduced gain. In the past a lot of attention of electronic cir-
cuit designers was spent in optimising the settings of these single transistor
amplifiers in order to avoid the use of multiple costly transistors.

6.2.2.3 Differential amplifier

As the last configuration, before integrating them all in an operational

amplifier, the differential amplifier is presented in Figure 6.38. In essence
this configuration consists of two single transistor voltage amplifiers that
are connected via their emitters. It was explained that a voltage amplifier
consists of a variable current source that creates a voltage over the collector
resistor. This current is determined by the emitter resistor and the difference
between the emitter voltage and the negative supply voltage. With the
differential amplifier however, the collector current is not only determined
by this difference but also by the difference between both base voltages.
To explain the functionality, the collector resistor values of both transistors
are assumed identical, Rc,1 = Rc,2 = Rc . Also both emitter resistors are
assumed identical, Re,1 = Re,2 = Re , while a third emitter resistor Rc,e is
added that carries the current of both transistors.
The input signals Vi,1 and Vi,2 are modelled to consist of a common part and
a differential part.

Vi,1 = Vi,c + Vi,d and Vi,2 = Vi,c − Vi,d (6.60)

The common part of the signal is called the common-mode signal and this
part results in an equal emitter current value I e,1 = I e,2 for both transistors at
a value of half of the total current I t . This total current is determined by total
resistance to the negative supply voltage Re,t = Re,c + Re,1 ∥ Re,2 = Re,c + 0.5Re ,
because both emitter voltages are equal.
The part of the current in both transistors that is caused by the common-
mode input signal can than be calculated as:
Vi,c − 0.6
I e,1 = I e,2 = 0.5 I e,t = 0.5 (6.61)
R e,c + 0.5R e

In this equation the threshold voltage Vth ≈ 0.6 V has been taken into account.
The output voltage that is caused by the common-mode part of the input
voltage is equal for both collectors, because of the equal collector resistances:

Vi,c − 0.6
Vo,1 = Vo,2 = 0.5R c (6.62)
R e,c + 0.5R e
368 Chapter 6. Analogue electronics in mechatronic systems

Vp Vp

Rc,2

Rc,2
Rc,1

Rc,1
Vo,1 Vo,2 Vo,1 Vo,2

T1 T2 T1 T2
+ Vi,1 Vi,2 Vi,1 Vi,2

+
+

+
_ _ _ _

Re,1

Re,2
T3

+
_
Vc,set
Re,c

Rcs
Vp Vp
di<erenal pair long tailed pair

Figure 6.38: The differential amplifier consists of a pair of equal transistors with
the emitters connected either directly or via a pair of emitter resistors.
a: The differential amplification is determined by the collector re-
sistances and the total resistance between the emitters while the
common-mode amplification is determined by the collector resistors
and the total resistance from the emitters to the negative supply volt-
age.
b: With the high impedance current source T3 , the common-mode
amplification is minimised.

The amplification of the differential part, the differential-mode signal, is only

determined by the collector resistors and the series value of the resistors
R e,1 and R e,2 . This is true, because for a differential signal the voltage at
the connection point between both equal resistors does not change when
one input voltage gets a positive dVi , while the other input gets an equal
negative dVi . As a result, the common-mode current through Re,c will be
divided differently over both transistors in a ratio depending on the value of
R e,1 = R e,2 = R e . The current difference will then be:

Vi,1 − Vi,2 dVi

d I e,1 = − d I e,2 = = (6.63)
2R e Re

A positive difference between Vi,1 and Vi,2 gives a corresponding higher

current in T1 and a decreased current in T2 . As a consequence, with the
equal collector resistances, the part of the output voltage, caused by the
6.2. Active electronics 369

differential-mode part of the input voltage, becomes differential:

dVi
−Vo,1 = Vo,2 = R c (6.64)
Re

With these relations it is possible to optimise the amplifier for a certain

application. Generally the goal is to realise an ideal differential amplifier
without any common-mode amplification. This is for instance useful in
measurement systems, to avoid errors due to common-mode interference
signals. This means, that the value of Re,c needs to be maximised and
the value for Re needs to be minimised. This ultimately results in the
configuration of Figure 6.38.b, where the common emitter resistor is replaced
by a third transistor in the current source mode, with a fixed input voltage
Vc,set , that determines the common current through R cs . This configuration
is called a long tailed pair because of the additional current source in the
tail of the differential amplifier. In this example the two emitter resistors
are omitted, in order to increase the gain to the maximum value possible.
When linearity is needed they can be added also in this configuration like
previously explained with the single transistor voltage amplifier.
This differential amplifier configuration gives two output voltages that can
either be used directly by choosing one of the outputs, but mostly it is used
integrated in an operational amplifier.
370 Chapter 6. Analogue electronics in mechatronic systems

—
 ”~?
OPA
627
< 
 1 8 NC
–In 2 – 7 +VS
OPA
627
+In 3 + 6 !!
–VS 4 5 < 
 ~š

Figure 6.39: The symbol, pinning and housing of an operational amplifier.

(Courtesy of Texas Instruments)

6.2.3 Operational amplifier

The operational amplifier is possibly the most important and successful

example of an analogue IC. It is frequently used in measurement and sensor
systems, power amplifiers and many other applications, both in consumer
and professional equipment. Its success is largely based on its versatility,
ease of use and reliable results. While it is an amplifier with a very high
gain, it is always used with feedback and because of that situation, it in-
volves the same dynamic issues on stability, that are known from any other
feedback system. For that reason it is presented here quite thoroughly as
in mechatronic systems all performance issues are related to dynamics and
stability.

6.2.3.1 Basic operational amplifier design

A basic operational amplifier consists of three functional stages as shown3

in Figure 6.40.
The first stage, the differential input stage, is designed as a long tailed pair
with current source T3 , that keeps the sum of the currents through T1 and
T2 constant. This configuration maximises the ratio between the differential-
mode gain and the common-mode gain, the common-mode rejection ratio
(CMRR).
The second stage, the gain stage around T4 ,T5 and T6 , is a combination of
a differential pair and a voltage amplifier. Without the resistors between
the emitters, it creates a high differential gain with low linearity, which is
3 To avoid confusion the voltage source for the power supply, input and output voltage is not

drawn in the figure but all mentioned values are noted in respect to the reference which in
general is ground or 0V from the power supply. This is quite common in electronic circuits.
6.2. Active electronics 371

Vp+

T4 T5
T7
T1 T2 Vo
V i+

T8
T3
V i- T6

Vp-
Di<erenal amplicaon Voltage gain Current gain

Figure 6.40: Basic Operational amplifier with bipolar transistors, diodes and re-
sistors. Three functional stages can be distinguished, a first stage
that amplifies only the differential input voltage, a second stage with
a maximum voltage gain and a third stage with a maximum current
gain.

not a big issue4 as an operational amplifier is uniquely used with feedback.

With this gain the voltage difference from the first stage is multiplied.
The last stage, the power output stage, has to be able to deliver the required
current to the load. This is accomplished by a symmetrical power output
stage in Class AB push-pull configuration. The necessary bias voltage is
realised by the two diodes between the bases of T7 and T8 and solve the
problem at the crossover between a positive and negative output current
while they keep the transistors in just a conductive mode around a few mA
of idle current.
4 Neglecting the non-linearity is only allowed in non-critical situations. In reality, one should

design any component in a control-loop as linear as possible in order to avoid residual errors.
372 Chapter 6. Analogue electronics in mechatronic systems

6.2.3.2 Operational amplifier with feedback

An operational amplifier is always used with negative or positive feedback.

Negative feedback is applied when a linear behaviour is needed and positive
feedback is applied for special functions like the Schmitt trigger that will be
used in Section 6.3 to design a pulse-width modulator for a switched-mode
amplifier. When the term feedback is used in a linear system it generally
means negative feedback, but in the section on active filters positive feedback
is applied to create a suitable the transfer function.
For a non-inverting amplifier, applying negative feedback results in the
representation as shown5 in Figure 6.41. The circuit is drawn such that the
classic feedback control loop is recognisable with a set point, an output and
a differential stage, that compares the set point with the value fed back from
the output. For an amplifier with an open-loop gain G o , the closed-loop gain
(G a ) equals the complementary sensitivity function of the feedback circuit,
as explained in Chapter 4 on motion control, and is given by:
Vo Go 1
Ga = = = (6.65)
Vi 1 + G o G f 1
+ Gf
Go

Where the feedback gain G f is given by the voltage divider:

R1
Gf = (6.66)
R1 + R2

In an ideal operational amplifier G o is infinite. This results in a closed-loop

gain of:
R1 + R2
Ga = (6.67)
R1

This means that the amplification is only depending on the ratio of the
resistors R1 and R2 . It also means that the difference in input voltages
becomes zero at infinite G o . In other words: “The output will do anything to
make the voltage at the minus terminal, the feedback point, equal to the
voltage on the plus terminal, the reference point”.
This understanding is used in most simplified designs of electronic circuits,
like presented in the following section, where different amplifier and filter
combinations are shown based on operational amplifiers.
5 The power supply is omitted for simplicity in all further circuit diagrams with operational

amplifiers. Unless otherwise mentioned a symmetrical power supply with an equal positive
and negative supply voltage of sufficient magnitude is assumed to be present.
6.2. Active electronics 373

Input + Gf

+
 Output
+ Go

+
_ _
R2

Vi Vo

R1
_ _

Figure 6.41: Operational amplifier is used with feedback.

6.2.4 Linear amplifiers with operational amplifiers

In this section several basic operational amplifier configurations will be

presented using ideal operational amplifier characteristics:

• The open-loop gain is infinite.

• The input current is zero with an infinite input impedance.

• The output current is unlimited, with a zero output impedance.

• The common-mode amplification is zero.

• No dynamic-, offset-, power supply- or other limitations and deviations

are present.

Even though in reality all these statements are not true, in many less
critical applications general operational amplifiers behave sufficiently ideal
to justify the use of the configurations in this section.

6.2.4.1 Design rules

The following rules for designing linear circuits with operational amplifiers
directly follow from the above mentioned understanding of the working
principle of a feedback controlled operational amplifier and is used through-
out all electronic circuits in this book, even when non-ideal amplifiers are
considered:
374 Chapter 6. Analogue electronics in mechatronic systems

1. When negative feedback is applied, the output does whatever it can to

keep the negative input equal to the positive input (V + = V − ).

2. The current in the circuit is not influenced by the input currents as

these are zero.

These rules are applied in the following steps to determine the function of
most operational amplifier circuits:

1. Check if negative feedback is present.

2. Calculate the voltage on the positive input as this is in most cases only
determined by the input voltage and not by the feedback voltage.

3. Calculate the currents in the resistors and remaining voltages by

applying rule 1 and 2 .

6.2.4.2 Non-inverting amplifier

The non-inverting amplifier configuration is equal to the previously pre-

sented operational amplifier with feedback of Figure 6.41. It is drawn in its
simplified form in Figure 6.42 with a special version in the second picture,
which is the unity-gain non-inverting amplifier or buffer amplifier, that
was used in the previous part about passive filters, to reduce the mutual
interaction of the different filter sections.
Although the closed-loop gain is already determined previously, the validity
of the rules and steps is verified by applying them also in this configuration.
First it is checked that indeed negative feedback is applied, which is true
via the voltage divider by R1 and R2 .
Then rule one is applied and the amplifier will do whatever it can to make
voltage V − at the negative input equal to the voltage at the positive input,
V + = Vi .
With these conditions the current I 1 in R1 should be:
Vi
I1 = (6.68)
R1

The current I 2 in R2 is equal to I 1 because of rule two, the input current of

the amplifier is zero. This means that the output voltage as function of the
input voltage becomes equal to:
Vo I1 R1 + R2
= G a = (R 1 + R 2 ) = (6.69)
Vi Vi R1
6.2. Active electronics 375

+
+
G ()

+
_

I2 R2
_
Vi Vo
G ()

+
+

+
I1 R1
Vi Vo
_ _ _ _

Figure 6.42: The non-inverting amplifier has a gain larger than one, determined by
the voltage divider in the feedback path. When the output is directly
connected to the minus input, giving a feedback gain of one, the unity-
gain buffer amplifier is obtained.

This equation is equal to the previously obtained Equation (6.67) and the
rules are verified.
When the resistors in the feedback path are chosen such, that R1 = ∞ and
R 2 = 0, the gain G a becomes equal to one. As a result, this configuration acts
as a unity-gain buffer amplifier. A very important property of a non-inverting
amplifier is the high input resistance of the circuit, which is only determined
by the common-mode input resistance of the operational amplifier at the +
input multiplied by the open-loop gain of the amplifier as the negative input
is made to follow the positive input. This open-loop value of this resistance
is already high and with feedback it is almost infinite even with a real
operational amplifier. The differential input impedance does not play a role
as the voltage on the − input is equal to the voltage on the + input. This
configuration is especially useful when very weak sensor signals with a low
voltage and high source impedance need to be amplified.
On the other hand, the output resistance of this amplifier is very low, because
of the ideal operational amplifier characteristics. Even if there was a small
output resistance, the feedback will reduce the internal output resistance
by the total loop gain, according to the sensitivity function of the feedback
system.
376 Chapter 6. Analogue electronics in mechatronic systems

R2

_
I2
R1

+
I1 G ()

+
+
Vi Vo
_ _

Figure 6.43: In an inverting amplifier the output has a different sign as the input.

6.2.4.3 Inverting amplifier

The inverting amplifier is shown in Figure 6.43. The amplification is calcu-

lated as follows.
According to the rules first the presence of negative feedback is checked,
which is true via R2 . Then the positive input voltage is zero as it is connected
to ground.
With rule one the output will take care that V + = V − = 0 and with this infor-
mation the current I 1 through R1 can be calculated by applying Ohms law:

Vi − V − Vi
I1 = = (6.70)
R1 R1
I 2 relates to the output voltage as follows:
V − − Vo Vo
I2 = =− (6.71)
R2 R2
The current I 2 through R2 is equal to I 1 , because of rule two, resulting in
the following closed-loop gain of the circuit:
Vi Vo Vo R2
I1 = I2 = =− =⇒ Ga = =− (6.72)
R1 R2 Vi R1
The minus sign indicates, that the output signal is both amplified and
inverted, as it has a different sign than the input signal. Contrary to the
non-inverting amplifier, the closed-loop feedback gain can also be lower than
one by choosing R2 < R1 .
The main drawback of the inverting amplifier is its relatively low input
resistance, that equals R1 . For that reason this configuration is applied in
those situations, where the low input resistance gives no problems and the
inverted functionality is needed. It is also useful, when signals have to be
added or subtracted.
6.2. Active electronics 377

R4
I1 I4
R1 _

+
I2 G ()

+
+
R2

Vi,1 + Vi,2 I3 Vo
R3
+
_ _ _ Vi,3 _

Figure 6.44: Adding signals is achieved by adding the currents of different inputs
at the negative input, that acts like a virtual ground.
When R1 = R2 = R3 = R the output signal is equal to the inverted sum
of all input signals, amplified with a factor R4/R .

6.2.4.4 Adding and subtracting signals

A very powerful application of operational amplifiers is based on their capa-

bility to add and subtract signals in a configuration, as shown in Figure 6.44.

The calculation of the amplification is as follows. Again the rules are applied.
There is negative feedback and the positive input voltage is zero and equal
to the negative input voltage, V + = V − = 0.
Because of rule two, no current can flow into the amplifier inputs and the
currents through the resistors relate according to Kirchhoff’s first law:

I1 + I2 + I3 − I4 = 0 =⇒ I1 + I2 + I3 = I4 (6.73)

With Ohm’s law these currents are calculated from the different voltage and
resistor values:

Vi,1 Vi,2 Vi,3 Vo

+ + =− (6.74)
R1 R2 R3 R4

This results in the following output voltage:

R4 R4 R4
Vo = G a,1 Vi,1 + G a,2 Vi,2 + G a,3 Vi,3 = − Vi,1 − Vi,2 − Vi,3 (6.75)
R1 R2 R3

In the inverting and the adding configuration, the negative input acts like a
virtual ground, because the output keeps the negative input at the same
378 Chapter 6. Analogue electronics in mechatronic systems

I2
R2
I1
Inverng R1 _

+
G ()

+
Non +
R1
inverng

+
Vi,1 Vi,2 R2 Vo

_ _ _

Figure 6.45: Subtracting signals is achieved by combining a non-inverting and an

inverting configuration and using both the positive and negative input
of an operational amplifier.

level as the grounded positive input. This is the reason, that the signal at
one input terminal is not influenced by the signal at another input terminal.
This is especially useful, when signals are combined from sources that are
so sensitive that other voltages on its output can cause interference by non-
linearity.
This is different from the non-inverting amplifier.
When signals need to be subtracted, the circuit as shown in Figure 6.45 can
be used. It is a combination of a non-inverting and an inverting amplifier
configuration within one operational amplifier. The standard rules are
applied also for this calculation of the amplification.
As the first step it is checked that negative feedback is applied via R2 .
The next step is to calculate the voltage at the positive input and directly
apply rule one:

R2
V − = V + = Vi,2 (6.76)
R1 + R2

Using rule two, the relation between the output voltage Vo , the first input
voltage Vi,1 and V − can be determined by using Ohm’s law and Kirchhoff’s
first law:
Vi,1 − V − V − − Vo
I1 − I2 = − =0 (6.77)
R1 R2

Note the minus sign in I 2 that is caused by the defined direction of the arrow
for I 2 in Figure 6.45.
6.2. Active electronics 379

With a little bit of algebra, this equation gives the relation for V − :
R1 R2
V − = Vo + Vi,1 (6.78)
R1 + R2 R1 + R2

When replacing V− by the result of Equation (6.76), the following relation

between the three voltages is obtained:

R2 R1 R2 ( )
Vi,2 = Vo + Vi,1 =⇒ R 1 V0 = R 2 V1,2 − Vi,1 (6.79)
R1 + R2 R1 + R2 R1 + R2

Resulting in:
R2 ( )
Vo = Vi,2 − Vi,1 (6.80)
R1

It is important to be aware that the voltage divider with R1 and R2 after

Vi,2 with the same values as the resistors in the feedback circuit is neces-
sary, because of the difference of amplification of a non-inverting and an
inverting amplifier configuration. Sometimes the mistake is made to leave
the resistors at the positive input away for simplification. This results in a
higher gain for the non-inverted input signal, relative to the inverted input
signal with a corresponding high common-mode amplification. In the idea
situation this common-mode amplification is zero when all resistances have
an equal value.
A disadvantage of this single operational amplifier subtracting configuration
is the difference in loading of the two input signals. The positive input has
an input impedance equal to R1 + R2 , while the negative input has an input
impedance of R1 only. Further the current I 1 is determined by both Vi(1)
and Vi(2) , which means that Vi(2) influences the current from Vi(1) . This can
cause problems when the source of Vi(1) has a voltage dependent impedance.
In those cases, this has to be solved, for instance by inserting a separate
unity-gain buffer amplifier between this source and the inverting input. In
Chapter 8 the instrumentation amplifier will be introduced, a differential
amplifier that uses two special buffer amplifiers, one for each input, that
solves this issue in the most elegant way.

6.2.4.5 Transimpedance amplifier

Several sensors that are applied in measurement systems give a current

signal as function of the measurement parameter. The output impedance
of these sensors is very high with a current source characteristic and the
measurement current often needs to be converted into a voltage. Such a
380 Chapter 6. Analogue electronics in mechatronic systems

_
I2

+
G ()

+
I1 R I1 +
Vo Vo
_ _

Passive I to Vš' 

 


Figure 6.46: A current to voltage converter creates an output voltage from the input
current. A load resistor to ground acts like a passive I to V converter.
With an operational amplifier one can create a converter that both
adds energy to the signal and avoids a voltage over the current source.

converter is called a current-to-voltage ( I to V ) converter, also named a

current controlled voltage source or transimpedance amplifier.
In its simplest form a current to voltage converter is just a load resistor to
ground as shown in the left drawing of Figure 6.46. The voltage over the
resistor equals Vo = I 1 R according to Ohm’s law.
This simple configuration has two drawbacks. First of all there is no energy
amplification of the often very weak current signal and secondly the result-
ing voltage is also present over the input current source. When this current
source is not ideal with a non-infinite output impedance, the voltage will
influence the current.
To solve these problems, an operational amplifier can be used for the con-
version as shown at the right drawing in Figure 6.46. The configuration is
similar to the inverting amplifier but the input resistance is left away. This
can be done because the impedance of the current source is in most cases
much larger than the input resistor from a normal inverting amplifier and
as they are connected in series the smaller value can be omitted.
Following the rules of an operational amplifier, the output of the amplifier
will get a value such that the minus input voltage will become equal to the
plus input voltage, being equal to 0 V ground level. With a zero current
into the input of the operational amplifier the output voltage needs to be
such that it creates a current I 2 in the feedback resistor equal to the input
current I 1 .
The main differences with the simple resistor are clear. The virtual ground-
ing of the input current at the minus input keeps the voltage over the input
current source at zero Volt so even with a less perfect current source there
is no problem. Furthermore the amplifier adds energy to the signal. Only
6.2. Active electronics 381

the sign is inverted and a positive current will result in a negative output
voltage Vo = − I 1 R .
This active configuration acts like an impedance that the current has to
pass through (trans) from the input to the output, for which reason this
amplifier is called a transimpedance amplifier.

6.2.4.6 Transconductance amplifier

In Chapter 5 it was shown that electromagnetic actuators require a con-

trolled current in order to control the force. In such a case a converter is
needed that converts the control voltage into a current with a current source
characteristic.
Such a converter is called a voltage-to-current (V to I ) converter, also named
a voltage-controlled current source or transconductance amplifier. The name
transconductance is based on its inverted property when compared with the
transimpedance amplifier where conductance is the inverse of impedance.
Although in theory it is possible to create such a converter with a high ohmic
series resistor between the voltage source and the load such a resistor would
reduce the energy of the signal and for that reason a series resistor is not a
realistic option when controlling actuators.
A transconductance amplifier can be created by introducing current mea-
surement in the feedback loop of an operational amplifier as shown in
Figure 6.47.
The left drawing shows a configuration where a small measurement resistor
R 3 is connected in series with the load that is connected to the output ter-
minal of the amplifier. The current I o creates a current sensing voltage Vcs
over this resistor equal to Vcs = I o R3 . This voltage is fed back to the negative
input of the amplifier.
The amplifier obeys to rule one and does whatever it can to keep the negative
input voltage equal to the positive input voltage, which is zero because the
configuration is designed as an inverting amplifier. As a consequence of this
design, the current sensing voltage Vcs over R3 will be kept at:

R2
Vcs = −Vi (6.81)
R1
To realise this voltage, the amplifier needs to supply an output current
through the actuator and R3 at a value of:

Vcs R2
Io = = −Vi (6.82)
R3 R1 R3
This means that the current through the load is only determined by the
input voltage.
382 Chapter 6. Analogue electronics in mechatronic systems

+
G ()
R1 _

R3
Io
Load

Vi

R2

Ra
+ Load Io
G () _ Vcs I o R3
R1 _
+

Ra
Io R

R2 + R
Vi
G ()
Vcs I o R3 _ R
R3

Io
_
R
 –!  

Figure 6.47: A voltage to current converter creates an output current that is deter-
mined by the input voltage. By measuring the output current with
a small resistor and using that signal in the feedback loop a current
source characteristic is realised. With a floating load this is simply
achieved as shown in the left drawing. A grounded load requires a
floating measurement resistor with a separate differential amplifier
to create the current sensing voltage Vcs for current-feedback.

This simple configuration has one drawback in the fact that the load itself
is not grounded at one of its terminals. This is called a floating load with
a common-mode voltage relative to ground equal to the voltage over the
current sensing resistor.
In case the load needs to be grounded the configuration at the right side
of Figure 6.47 can be applied where the load and current sensing resistor
are reversed. With this reversal the current sensing resistor is floating
with a common-mode voltage that is equal to the voltage over the load. To
cancel this common-mode voltage an additional subtracting amplifier is
added that gives an output voltage to ground equal to the voltage over the
current sensing resistor. The remaining part is equal to the floating-load
configuration.
6.2. Active electronics 383

6.2.5 Active electronic filters

The presented amplifier configurations all used pure resistors to determine

their closed-loop transfer function. In principle these resistors can be ex-
changed by complex impedances, in order to achieve a frequency dependent
behaviour. Similar with passive filters, this could be done with capacitors
and inductors, but the active nature of operational amplifiers creates the
possibility, to avoid the large and expensive inductors and create filters of
any kind and order with only capacitors and resistors. Also this field is
extremely wide and requires a strict selection for this book. The examples
chosen include simple first-order filters, an analogue PID controller and
active second-order filters, as these are fully representative for the most
frequently used configurations in mechatronic systems.

6.2.5.1 Integrator and first-order low-pass filter

For most of the filters, the inverting amplifier of Figure 6.43 is used as a
starting point. The integrator and the related first-order low-pass filter are
both shown in Figure 6.48, because they differ with only one resistor.
Ohm’s law is also valid for complex impedances, so the transfer function of
an inverting amplifier according to Equation (6.72) can be written as follows:

Z2
Ga = − (6.83)
Z1

When adding a capacitor to one of these impedances, this transfer function

becomes frequency dependent. If for instance R2 is infinite and a capacitor is
placed parallel to it, the inverting amplifier becomes an inverting integrator
with transfer function:
Z2 1 1
I − ( s) = − =− , I − (ω) = − (6.84)
Z1 sR 1 C j ωR 1 C

In order to compensate the minus sign, a second inverting amplifier has

been added. This avoids the confusion of the phase of the integrator with
the 180◦ phase of the inversion6 . This results in the following integrator
without inversion:
1
I (ω) = (6.85)
j ωR 1 C
6 This additional inversion is often not necessary in complex filters with a multitude of

inverting filter and amplification steps. In those situations eventually one additional inverter
can be added if the resulting signal is still inverted.
384 Chapter 6. Analogue electronics in mechatronic systems

Z2

R2 ( or15k)
Z1
R (10 k)
C (0.1F)
R1 (15k)
_ R (10 k)
_
+

G () G ()

+
+ +
Vi Vo
_ _

^ J
J
Low pass
Magnitude (abs)

Magnitude [dB]
J
1 Integrator ^J

1 J

-1
J >^J

*^
J >J

J
Phase [deg]

>\

>|J
J 1 ^ 3 
J J J J J
Frequency [Hz]

Figure 6.48: A capacitor in the feedback path of an inverting amplifier creates an

integrator, when the resistance of R2 is infinite. A finite value of R2
creates a first-order low-pass filter. The Bode plot shows both transfer
functions using the indicated values of the passive elements. The
second inverter with a gain of −1 corrects the inversion by the first
inverter.

In the Bode plot this non-inverting integrator gives a line with a slope
of -1, that intersects the 0 dB level at the unity gain cross-over frequency
ω0 = 2π f 0 = 1/R 1 C and a continuous phase of −90◦ over the entire frequency
band.
When the resistor R2 over C becomes smaller than infinite, it will limit
the amplification at low frequencies and as a result a first-order low-pass
filter is created. The transfer function is determined by calculating Z2 as
the impedances of R2 and C in parallel and including the gain of −1 by the
6.2. Active electronics 385

second inverter:
Vo Z2 1 R2 1
F (ω) = = =  = (6.86)
Vi Z1 1 R 1 (1 + j ωR 2 C )
R1 + j ωC
R2

This represents a first-order filter with an RC-time constant τ = R2 C and

a gain of R2/R1 in the pass band below the corner-frequency ω0 = 2π f 0 = 1/τ.
At f 0 the gain is −3 dB and at higher frequencies the slope is −1, equal to
the integrator. This is shown in the Bode plot for the indicated values of the
passive elements, resulting in a pass band gain of one and f 0 = 100 Hz.
Note that the unity-gain cross-over frequency ω0 for the integrator and the
corner-frequency ω0 for the low-pass filter are equal in the Bode plot of
Figure 6.48 because R1 and R2 are chosen to be equal for this example in
order to fit into the same Bode Plot and enable comparing of the results.
Generally, even in normal inverting and non-inverting amplifiers, a small
capacitor is almost always placed parallel to the feedback resistor over an
operational amplifier. This is done for two reasons. First of all, it is never
really useful in mechatronic systems, to amplify high frequencies above
the range of interest, as at these frequencies many undesired dynamic
eigenmodes are present in the mechanics. The second reason is related to
the non-ideal properties of any electronic amplifier regarding its dynamic
performance. More details will be presented after this section on filters,
but it is already useful to recognise that a capacitor in the feedback path
determines a differentiating action within the loop. This differentiation
introduces a phase lead, that improves stability of the closed-loop system.
This differentiating action of the capacitor is the result of the fact, that the
current in a capacitor advances on the voltage and it is even better explained
with the following example, the differentiator.

6.2.5.2 Differentiator and first-order high-pass filter

The differentiator and the almost identical first-order high-pass filter are
created with the same inverting amplifier configuration, but in this case a
capacitor is placed in the forward path as shown in Figure 6.49.
When the value of R1 is zero, a pure differentiator is created. When the gain
of −1 of the second inverter stage is included, the transfer function is:
Z2
D ( s) = = sR 2 C , D (ω) = j ωR 2 C (6.87)
Z1

In the Bode plot this results in a line with a slope of +1, that intersects the
0 dB level at the unity-gain cross-over frequency ω0 = 1/R2 C and a constant
386 Chapter 6. Analogue electronics in mechatronic systems

Z2

Z1
R2 (10k) R (10k)

R1 (0 or 10k) _ R (10k)
_
+

C (1F) G () G ()

+
+ +
Vi Vo
_ _

^
J J
Magnitude (abs)

High pass

Magnitude [dB]
1
J ”
< ^J

1 J

-1
J >^J

*^
J >J

|J
Phase [deg]

\

J
J 1 ^ 3 
J J J J J
Frequency [Hz]

Figure 6.49: A capacitor in the forward path of an inverting amplifier creates a

differentiator, when the resistance of R1 is zero. A higher value of
R 1 creates a first-order high-pass filter. The Bode plot shows both
transfer functions, using the indicated values of the passive elements.

phase lead of 90◦ over the entire frequency band.

Generally a pure electronic differentiator is impossible by definition, as the
gain at infinitely high frequencies would need to become infinite. It is also
far from practical in reality and measures should be taken to limit the gain
of a differentiator at high frequencies. This is accomplished by adding a
series resistor to R1 and this results in a first-order high-pass filter. The
related transfer function, including the gain of −1 of the second inverter,
becomes:
Vo Z2 R2 R2 1 R 2 j ωR 1 C
F (ω ) = = = = = (6.88)
Vi Z1 1 R1 1 R 1 1 + j ωR 1 C
+ R1 1+
j ωC j ωR 1 C
6.2. Active electronics 387

This represents a first-order filter with an RC-time constant τ = R1 C and

a gain of R2/R1 in the pass band above the corner-frequency ω0 = 1/τ. At
ω0 the gain is -3 dB and at lower frequencies the slope is +1, equal to the
differentiator. This is shown in the Bode plot for the indicated values of the
passive elements, resulting in a pass band gain of one and ω0 = 100 rad/s.

6.2.6 Analogue PID controller

Before the creation of fast digital controllers, operational amplifiers were

used to realise the PID-control function. Figure 6.50 shows an example
of such a circuit, that consists of a combination of the differentiator and
integrator, that were presented before.
Before deriving the transfer function, first the three elements of control are
examined using the values of the passive elements in the circuit. They will
appear to be chosen according to the rules of thumb for a PID-controller,
as described in Chapter 4 with a first order taming of the D-control action.
Because this controller is meant to be used in the domain of mechanical
engineering the frequency will be presented in Hertz and the magnitude in
the Bode plot is absolute instead of in dB.
Starting at zero Hertz the impedance of the capacitors is infinite and the
gain of the circuit will be infinite too, just as expected from an I-control
system. At an increasing frequency the impedance of C2 will dominate the
gain of the controller, because the impedance of C1 can still be neglected
compared to the impedance of the parallel resistor R3 . As a result, the gain
will show a −1 slope in the Bode plot, conform the I-control action. At the
integrator corner-frequency f i = ωi/2π = 1/(2πR2 C2 ) ≈ 20 Hz, the impedance of
C 2 becomes smaller than the impedance of R 2 and the negative slope will
flatten out. At the differentiator corner-frequency f d = ωd/2π = 1/(2πR3 C1 ) ≈
60 Hz, the impedance of C 1 becomes smaller than the impedance of R 3 and
the corresponding differentiating action will show a +1 slope in the Bode
plot, that is terminated (tamed PID-control) as soon as the impedance of
C 1 becomes smaller than R 1 . This happens at the taming corner-frequency
f t = ωt/2π = 1/(2πR 1 C 3 ) ≈ 600 Hz.
As was presented in Chapter 4 on the PID controller, a differentiating action
gives an additional gain at the cross-over frequency. In this example the
second inverter is used to correct the proportional gain for this additional
gain. The maximum phase lead of the controller occurs at the cross-over
frequency of 200 Hz, where the gain is one. This value is chosen for this
example. In reality in most cases a higher proportional gain is needed, to
388 Chapter 6. Analogue electronics in mechatronic systems

compensate the low gain of the plant. The circuit can be adapted to this
gain by changing the resistors of the second inverting amplifier.

6.2.6.1 Transfer function

The transfer function of the electronic PID-controller is derived in the fre-

quency domain for reason of simplicity and starts with the same general
transfer function for the inverting amplifier as with the previous filters,
including the gain − g i of second inverting amplifier:
1
+ R2
Vo Z2 sC 2
C pid ( s) = = gi = gi (6.89)
Vi Z1 1
1 1
+
1 R3
+ R1
sC 1

With shifting of the different terms the following equation is obtained:

 
1 sC 1 1
C pid ( s) = g i + R2 + (6.90)
sC 2 1 + sR 1 C 1 R 3

At this point the relevant time constants of the controller are chosen accord-
ing to the corner-frequencies, defined in the previous part. The first one, τi
defines the integrator corner-frequency f i . The second one, τd defines the
differentiator corner-frequency f d and the third one τt defines the taming
corner-frequency f t .
1 1
τi = = = R2 C2 (6.91)
ωi 2π f i
1 1
τd = = = (R 1 + R 3 ) C 1 ≈ R 3 C 1
ω d 2π f d
1 1
τt = = = R1 C1
ωt 2π f t

The approximation in τd is in most cases allowed as R1  R3 .

By using these terms the following result can be obtained:
   
R2 1 sC 1 R 2 R 2 τi s + 1 τ d s + 1
C pid ( s) = g i +1 + 1 = gi (6.92)
R 3 τi s 1 + sR 1 C 1 R3 τi s τt s + 1

The first term is a proportional term, the second term the integrator and
the third term the tamed differentiator. As will be shown further on, the
proportional term needs a correction factor, to become equal to the P-control
gain kp
6.2. Active electronics 389

Z1 Z2

R3 (8k2) R2 (8k2) R (1k)

C2 (1F)
R1 (820) _ R (3k3)
_
+

C1 (0.33F) G () G ()

+
+ +
Vi Vo
_ _

2
10
Magnitude (abs)

1
10

0
10

>
fi fd ft
10
|J
Phase [deg]

\

>\

>|J
> 0 1 2 3 4
10 10 10 10 10 10
Frequency [Hz]

Figure 6.50: Analogue PID controller and its Bode plot. The indicated values of
the resistors and capacitors are chosen for a PID controller, according
to the rules of thumb for a mass control system, with a targeted
unity-gain of the controller at the 200 Hz bandwidth. In reality the
proportional gain should be adapted to the real gain of the plant at
200Hz.

In the Bode plot the frequency response of this controller is shown with the
values as derived from the given components, τi = 8.2 · 10−3 , τd = 2.7 · 10−3 and
τt = 2.7 · 10−4 .
As presented in Chapter 4, the positive phase in the frequency band where
the lead network is working, creates an effective damping in a motion control
feedback configuration, like the servo system in a CD player. The integration
at low frequencies reduces steady state errors, due to constant disturbing
forces like gravity.
By tuning the values for the capacitors and resistors, the controller can be
adapted to the plant by means of loop shaping.
390 Chapter 6. Analogue electronics in mechatronic systems

6.2.6.2 Control gains

The control gains kp for P-control , kd for D-control and ki for I-control,
can be related to the time constants and passive component values of this
controller. For this derivation the taming factor 1/(τt s + 1) is not taken into
account as this represents just an additional pole, that is applied without
impact on the standard control gains.
To determine the control gains, Equation (6.92) is written in the additive
way by multiplication of the terms. This gives the following result, that
corresponds with Equation (4.32) from Chapter 4:
 
R2 τd R2 1 R2 ki
C pid ( s) = g i 1+ + gi + gi τd s = k p + + kd s (6.93)
R3 τi R 3 τi s R3 s

With this relation and applying the approximation in τd , the proportional

gain kp is equal to:
  
R2 τd R2 (R 1 + R 3 ) C 1 R2 C1
kp = g i 1+ = gi 1+ ≈ gi + (6.94)
R3 τi R3 R2 C2 R3 C2

The I-control gain is equal to:

R2 1 R2 1 1
ki = g i = gi = gi (6.95)
R 3 τi R3 R2 C2 R3 C2

The D-control gain is equal to:

R2 R2
kd = g i τd = g i (R 1 + R 3 ) C 1 ≈ g i R 2 C 1 (6.96)
R3 R3

Also in these expressions the approximation is based on the small value of

R 1 relative to R 3 .

6.2.6.3 High speed PID control

Even though digital controllers have taken over many control actions in
modern equipment, still one important advantage of analogue controllers
over digital ones remains and that is the lack of delay due to sampling.
In the digital world differentiating is done at any sampling moment by
taking at least two samples, the present and the previous one, and dividing
the difference over the sampling time. The result of this operation can
only be supplied to the plant at the next sampling moment so at least one
sample period later and as a result, part of the beneficial phase lead will be
lost. One of the solutions is to use a model based controller, to predict the
states, as explained in Chapter 4, but in unpredictable circumstances it is
6.2. Active electronics 391

Z3

I3
Z1 Z2 +

+
I1 I2 G ()

+
_

Z4
Vi Vx V+ Vo

_ _ _ _

Figure 6.51: Sallen-Key active second-order filter configuration with positive feed-
back.

always better to avoid delays in the D-control part. This requirement has
pushed sampling periods to ever smaller values, but in extremely fast control
situations, like with a piezoelectric scanner for a video-rate Atomic Force
Microscope, with bandwidths that exceed the Mega Hertz level, analogue
controllers might still represent the best solution possible.

6.2.7 Higher-order electronic filters

In Section 6.1.3 on passive filters it was shown, that inductors are required,
to achieve higher order electronic filters with an adequate transfer function.
Especially in the low-frequency area, the required high values of the self-
inductance lead to large sizes of the components and inductors in general
are rather expensive to manufacture. In this section on electronic filters
it will be shown with one example how, by using an operational amplifier,
these filters can be created with only small size capacitors and resistors. In
principle, these filters are designed to be applied with low power signals
and not between amplifiers and actuators, so they can be used only in the
path before the power amplifier.
The example that is presented here is a Sallen-Key filter named after R.P
Sallen and E.L. Key, two engineers of MIT Lincoln laboratory in 1955. It is
built around a non-inverting unity-gain amplifier, as shown in Figure 6.51,
where the output signal is positively fed back to the input to create complex
poles in the transfer function with imaginary parts.
The rules are applied again to calculate the transfer function. The positive
input is equal to the voltage at the negative input and also equal to the
392 Chapter 6. Analogue electronics in mechatronic systems

output voltage, as the operational amplifier acts as a unity-gain voltage

follower:

V + = Vo (6.97)

With this information current I 2 is determined:

V + Vo
I2 = = (6.98)
Z4 Z4

The following step is to determine the voltage Vx :


Z2
Vx = Vo + I 2 Z2 = Vo 1 + (6.99)
Z4

With this voltage I 1 and I 3 can be determined:

 
Z2 Z2
Vi − Vo 1 + Vo 1 + −1
Vi − Vx Z4 Vx − Vo Z4 Z2
I1 = = I3 = = = Vo (6.100)
Z1 Z1 Z3 Z1 Z3 Z4

With Kirchhoff’s first law on currents, the sum of all currents at the common
node of Z1 , Z2 and Z3 equals zero. This means with the defined directions
according to the arrows:

I1 − I2 − I3 = 0 (6.101)

Using the previously found equations for the different currents, with some al-
gebra the following generic transfer function of this configuration is derived:

Vo 1
=F= (6.102)
Vi Z1 Z2 Z1 + Z2
+1
Z3 Z4 Z4

With this equation the transfer function of the second-order low-pass filter
of Figure 6.52 will be determined as the first example.

6.2.7.1 Second-order low-pass filter

The following values for the impedances are chosen for this example:
1 1
Z1 = Z2 = R 1 = R 2 = R , Z3 = , Z4 = (6.103)
sC 1 sC 2

This gives as transfer function of the filter:

Vo 1
= F ( s) = 2 2
(6.104)
Vi R C 1 C 2 s + 2RC 2 s + 1
6.2. Active electronics 393

C1
R1 R2 +

+
G ()

+
_

C2

Vi Vo
_ _

Figure 6.52: Sallen-Key active second-order low-pass filter.

This is the transfer function of a second-order low-pass filter, with the

following values for the corner-frequency and damping:
 
1 C2 1 1 C1
ω 0 = 2π f 0 = # , ζ = RC 2 ω0 = , Q= = (6.105)
R C1 C2 C1 2ζ 2 C2

It appears that the quality-factor Q can be made very high by relatively

increasing the value of the positive feedback capacitor and a high damping
can be made with a low value of the positive feedback capacitor. This clearly
shows the effect on the imaginary part of the poles by the positive feedback.

6.2.7.2 Second-order high-pass filter

By exchanging the capacitors and resistors, R1 with C1 and R2 with C2 ,

a high-pass filter is obtained and it is left up to the reader to check the
derivation of the following corresponding transfer function:

Vo R 2 C 1 C 2 s2
= F ( s) = 2 (6.106)
Vi R C 1 C 2 s2 + 2RC 2 s + 1

In this case the same values for ω0 , f 0 , ζ and Q apply.

6.2.7.3 Different types of active filters

By choosing suitable values of the resistors and capacitors, electronic filters

can be created with almost any frequency and level of damping, without the
need for large inductors. They can be combined at will, to realise higher
394 Chapter 6. Analogue electronics in mechatronic systems

J

 J


!  …’
Magnitude (abs)
>J

*
J >^J

‹†QZ\š' >ZJ

‹†JQ{…!
*^ ‹†JQZ…
J >J
}J
? ’

|J

J
 ^ Z
J J J
€!Xƒ’

Figure 6.53: Three typical characteristics of a second-order high-pass filter with the
same corner-frequency but different levels of damping. For the But-
terworth characteristics the −3 dB bandwidth is equal to the corner-
frequency of 100Hz of this example. The Chebyshev characteristics
result in a −3 dB bandwidth at 70 Hz with the steepest slope below
the pass band. The Bessel characteristics result in a −3 dB bandwidth
at 200 Hz and a very gradual slope below the pass band.

order filters of any slope in the attenuation band.

To illustrate this with an example, Figure 6.53 shows three typical, often
used characteristics of a second-order high-pass filter with the same corner-
frequency of 100 Hz, but with a different amount of damping.
When the damping has a Q of 0.7, the resulting lack of resonance and
a magnitude of -3dB at the corner-frequency gives the filter Butterworth
characteristics, named after the British physicist Stephen Butterworth
(1885 –1958), who invented this filter type. This characteristic is often seen
as the most ideal filter type, although the slope in the attenuation band, just
below the −3 dB frequency determining the bandwidth, is not very steep.
This steepness of the slope at the −3 dB frequency can be improved with
a higher Q level. For instance a value of Q = 1.35 results in the Chebyshev
characteristics, named after the Russian mathematician Pafnuty Lvovich
Chebyshev (1821 – 1894), because the characteristics are derived from his
polynomials. The Chebyshev characteristic shows a ripple of +3 dB in the
pass band, but also a more steep slope below the −3 dB frequency.
On the other hand, when phase characteristics need to be as gradual as
possible, like in loudspeaker crossover filters, a lower value of Q is often
preferred, like with the shown Bessel filter with Q = 0.3, named after the
German astronomer and mathematician Friedrich Wilhelm Bessel (1784 –
6.2. Active electronics 395

1846). The Bessel characteristic shows a constant group delay in the pass
band, representing a linear phase to frequency relationship dφ/d f .
The same characteristics can be shown for low-pass filters and these exam-
ples are only a limited set of the entire range of different filter configurations,
that are designed for very specific purposes. Their design belongs to the do-
main of the specialist and with digital control even more transfer functions
can be realised, including variable delay and other tricks. Nevertheless the
mentioned configurations are frequently applied and already very suitable
for the design of general mechatronic systems.

6.2.8 Ideal and real properties of operational amplifiers

All the previously shown configurations assumed ideal characteristics of the

applied operational amplifier. One could wonder, why thousands of different
types of these universal building blocks are designed and manufactured,
when one would be sufficient. The reason is that like all components, also
the operational amplifier is not ideal and the more it needs to approach
the ideal behaviour, the higher the cost will be. In this section, different
characteristics will be presented that limit the performance of the circuits
with operational amplifiers. In most cases these limitations can be neglected,
but it is important to be aware of them for the more critical applications
with requirements, that are impaired by these limitations.
In Figure 6.54 an overview is given of the real characteristics of a typical high
performance operational amplifier. The parameters, with their influence on
the performance, will be explained separately. It will become clear that most
limitations result from the basic building blocks of an operational amplifier,
the transistors, diodes, resistors and capacitors.
First the dynamic parameters will be presented in a more thorough way,
because of their relatively dominant influence on the functionality and
stability of mechatronic systems. This is followed by a reduced presentation
of the less impairing limitations.

6.2.8.1 Dynamic limitations

The most important limitation of all is determined by the finite gain and
speed of an operational amplifier. Already at the presentation of a differ-
entiator it was mentioned, that an infinite gain at infinite frequencies is
impossible. This speed limitation is caused mainly by the parasitic capacitor
over the PN-junctions of the applied transistors. Especially the parasitic
396 Chapter 6. Analogue electronics in mechatronic systems

At TA = +25°C, and VS = ±15V, unless otherwise noted.

OPA627
PARAMETER CONDITIONS MIN TYP MAX UNITS
OFFSET VOLTAGE (1)
Input Offset Voltage 40 10 μV
AP, BP, AU Grades 100 250 μV
Average Drift 0.4 0.8 μV/° C
AP, BP, AU Grades 0.8 2 μV/° C
Power Supply Rejection VS = ±4.5 to ±18V 106 120 dB
INPUT BIAS CURRENT (2)
Input Bias Current VCM = 0V 1 5 pA
Over Specified Temperature VCM = 0V 1 nA
SM Grade VCM = 0V 50 nA
Over Common-Mode Voltage VCM = ±10V 1 pA
Input Offset Current VCM = 0V 0.5 5 pA
Over Specified Temperature VCM = 0V 1 nA
SM Grade 50 nA
NOISE
Input Voltage Noise
Noise Density: f = 10Hz 15 40 nV/√Hz
f = 100Hz 8 20 nV/√Hz
f = 1kHz 5.2 8 nV/√Hz
f = 10kHz 4.5 6 nV/√Hz
Voltage Noise, BW = 0.1Hz to 10Hz 0.6 1.6 μVp-p
Input Bias Current Noise
Noise Density, f = 100Hz 1.6 2.5 fA/√Hz
Current Noise, BW = 0.1Hz to 10Hz 30 60 fAp-p
INPUT IMPEDANCE
Differential 1013 || 8 Ω || pF
Common-Mode 1013 || 7 Ω || pF
INPUT VOLTAGE RANGE
Common-Mode Input Range ±11 ±11.5 V
Over Specified Temperature ±10.5 ±11 V
Common-Mode Rejection VCM = ±10.5V 106 116 dB
OPEN-LOOP GAIN
Open-Loop Voltage Gain VO = ±10V, RL = 1kΩ 112 120 dB
Over Specified Temperature VO = ±10V, RL = 1kΩ 106 117 dB
SM Grade VO = ±10V, RL = 1kΩ 100 114 dB
FREQUENCY RESPONSE
Slew Rate: OPA627 G =
–1, 10V Step 40 55 V/μs
OPA637 G =
–4, 10V Step 100 135 V/μs
Settling Time: OPA627 0.01% G –1, 10V Step
= 550 ns
0.1% G =
–1, 10V Step 450 ns
OPA637 0.01% G =
–4, 10V Step 450 ns
0.1% G –4, 10V Step
= 300 ns
Gain-Bandwidth Product: OPA627 G=1 16 MHz
OPA637 G = 10 80 MHz
Total Harmonic Distortion + Noise G = +1, f = 1kHz 0.00003 %
POWER SUPPLY
Specified Operating Voltage ±15 V
Operating Voltage Range ±4.5 ±18 V
Current ±7 ±7.5 mA
OUTPUT
Voltage Output RL = 1kΩ ±11.5 ±12.3 V
Over Specified Temperature ±11 ±11.5 V
Current Output VO = ±10V ±45 mA
Short-Circuit Current ±35 +70/–55 ±100 mA
Output Impedance, Open-Loop 1MHz 55 Ω
TEMPERATURE RANGE
Specification: AP, BP, AM, BM, AU –25 +85 °C
SM –55 +125 °C
Storage: AM, BM, SM –60 +150 °C
AP, BP, AU –40 +125 °C
θJ-A: AM, BM, SM 200 ° C/W
AP, BP 100 ° C/W
AU 160 ° C/W

Figure 6.54: Characteristics of a typical high performance operational amplifier.

An extracted version of the original data sheet of the the OPA 627 of
Burr Brown. (Courtesy of Texas Instruments)
6.2. Active electronics 397

R
dVcb Ic
I cb Ccb
dt t Vcb
Ccb

Ib Ii  I M
Vp
Ii
t

Figure 6.55: Speed limitation due to the collector base capacitor Ccb . A voltage
swing at the collector causes a current in Ccb , that has to be supplied
by the input voltage. With a voltage amplifier, this is perceived at the
input as a Miller capacitor with a larger value than Ccb .

capacitor Ccb between the collector and base is important, as it reduces the
current-amplification at higher frequencies. The effect of this capacitor on
a stepwise change in the input current is shown in Figure 6.55).
This input current ( I i ) would normally be equal to the base current ( I b )
and cause a proportional change in the collector current ( I c ). The collector
current will cause a proportional drop of the collector voltage over resistor
R , that induces a current ( I cb ) in C cb , that in its turn reduces the current in
the base:
dVcb
I b = I i − I cb = I i − C cb (6.107)
dt
To achieve a high speed, additional base current is necessary to compensate
this effect depending on the capacitor value. Because of the relation of the
current in Ccb with the voltage swing at the collector, this parasitic capacitor
is perceived at the input as a much larger capacitor, the Miller capacitor
CM .

C M = C cb (G v + 1) (6.108)

where G v is equal to the voltage amplification ratio of the transistor. With an

emitter follower this capacitor can be neglected but with a voltage follower
with an amplification of 100 even a value of 4 pF of a normal signal transistor
becomes significant. The resulting Miller capacitor of 400 pF, combined
with a practical input resistor of 10 kΩ, already limits the bandwidth of the
single transistor voltage amplifier to f = 1/(2πRC )=40 kHz.
398 Chapter 6. Analogue electronics in mechatronic systems

VCC+
36 pF

IN+ T5 T7
T4 37 pF

14 pF
15 Ω

T1 T2 OUT
7 pF
IN–
15 Ω

T3
T6 T8
460 Ω

VCC–

Figure 6.56: Schematic circuit drawing of the NE5532, a general purpose audio am-
plifier with bipolar transistor inputs. The transistors from Figure 6.40
are made recognisable and several more transistors are added for con-
trolling the internal currents and voltages and for protection like the
current protection transistors between T7 and T8 . The dominant pole
is created by a feedback capacitor of 37 pF around the high gain stage
by T4 and T5 . The other capacitors are added for further fine-tuning
the open-loop frequency response.
(Courtesy of Texas Instruments)

Open-loop gain and frequency response

A real operational amplifier contains many transistors to control the in-

ternal currents and voltages and to protect the operational amplifier from
overload. As examples the schematic circuit diagram of two real amplifiers,
optimised for different purposes, are shown in Figure 6.56 and Figure 6.57.
The NE5532 is a general purpose audio operational amplifier, designed for
stability and low cost. The OPA627 is designed as a precision amplifier with
Field Effect Transistors (FET) at the input to reduce the input currents.
Like a MOSFET, a FET uses an electric field to control the current but the
Gate is not insulated by an oxide but by another PN-junction.
The integration of many transistors in an operational amplifier results
by definition in several parasitic capacitances. These capacitances cause
problems in the dynamic performance, because they determine RC-times
(poles) in the transfer function with the associated phase lag of 90◦ at ω =
6.2. Active electronics 399

OPA627
®

7
Trim +VS
Trim
1 5
C1

C2
Output
6

+In –In
3 2

–VS
4

Figure 6.57: Schematic circuit drawing of the OPA627, a high precision amplifier
with Field Effect Transistor (FET) inputs to reduce the input current.
The dominant pole is created by increasing the pole of the parasitic
capacitors around the large gain amplifier stage between C1 and C2 .
(Courtesy of Texas Instruments)

2π f = 1/RC for each capacitor with its source resistance. This effect is even
increased with the collector-base capacitance of each transistor as explained
with the Miller capacitor. When two or more of these first-order poles are
located in the frequency range where the gain is larger than one, the closed-
loop transfer function becomes marginally stable or even unstable when
some additional phase lag is introduced by a third pole.
To illustrate this dynamic effect, an example open-loop frequency response
with marginal stability is shown in Figure 6.58. The red line shows the
frequency response of an amplifier with an open-loop gain of 106 and two
poles below 1 MHz. The first pole is located at 100 Hz and the second pole at
10 kHz, where the gain is still 104 . The phase margin approaches zero even
below 1MHz, where still some gain is present. This can create problems
when the closed-loop gain is < 100 as a phase margin of approximately 45◦ is
preferred for a well damped behaviour.
This problem can be solved by limiting the open-loop gain and creating one
dominant pole at a lower frequency. This causes the amplification to go down
with a slope of −1 from that frequency. The pole can be realised by placing
an internal larger capacitor in the part of the operational amplifier, that is
most prone to parasitic poles, like those related to the Miller capacitances
in the high voltage gain stage. This capacitor can be placed parallel to
an existing parasitic collector-base capacitance to move the main pole to
400 Chapter 6. Analogue electronics in mechatronic systems

6
10 120

10
5 -1 100

Magnitude (abs)

Magnitude [dB]
4
10 -1 }J
3
10 60
2 -2
10 40
1
10 20
10
0 Uncompensated 0
Compensated
0
Phase [deg]

>\

>|J

>Z\

>}J
0 2 4 6
10 10 10 10
Frequency [Hz]

Figure 6.58: Poles in the transfer function of an operational amplifier can be can-
celled by one dominant pole. This example shows two poles in the
uncompensated situation, one at 100 Hz and one at 10 kHz. It is clear
that the phase margin past 100 kHz is very small. With the dominant
pole at 1 Hz in the compensated situation the phase margin becomes
45◦ .

a lower frequency, like C2 in the OPA627. But also a feedback capacitor

can be used around the total high gain amplifier stage, like in the NE5532.
The indicated feedback capacitor of 37 pF cancels one of the parasitic poles
inside the loop and creates a new one, because it gives a differentiating
action inside the loop (a zero).
By placing this pole at a sufficiently low frequency, the total gain of the
amplifier crosses the unity-gain level before other poles have a significant
impact on the phase. With this method, the circuit with an operational
amplifier is stable at any closed-loop feedback configuration with a phase
margin of 45◦ at the unity-gain cross-over frequency of 1 MHz for this ex-
ample, as shown with the blue line in Figure 6.58. Because of the effect,
that the first pole is moved to the left and the second pole to the right in the
Bode plot, this method is called pole-splitting.

Gain-bandwidth product

The creation of only one dominant pole in the frequency range where the
gain is larger than one is reflected by a continuous −1 slope in the frequency
response above the corner-frequency that is determined by this pole. As a
consequence the product of the gain and the frequency is constant at any
frequency between the corner-frequency and the unity-gain cross-over fre-
6.2. Active electronics 401

Figure 6.59: The slew-rate of an operational amplifier limits the speed of change of
the output voltage. At high frequencies the output can not follow the
sinusoidal waveform anymore and tracks the signal with its maximum
dV/dt.

quency. This constant value is called the gain-bandwidth product and with
this value a designer can immediately determine the achieved bandwidth
at a certain closed-loop amplification. Ideally the gain-bandwidth product
is infinite.

Slew rate

Another consequence of the internal capacitances is the limited speed of

change of the output voltage. This slew-rate is determined by the maximum
current, that is available inside the operational amplifier to change the
voltage over the internal capacitors, when amplifying high-frequency AC
signals. The source of this current is limited by the nature of the design,
which means that the voltage over the capacitors can not change faster, than
a certain value of dV/d t. In practice this means, that at high frequencies the
maximum amplitude of the signal is limited. Above these levels a sinusoidal
waveform will change into a phase shifted triangular waveform with fixed
slopes, because of the maximum dV/d t. Ideally the slew-rate is infinite.

6.2.8.2 Limitations on the inputs

Input offset voltage and stability

The input offset voltage is the voltage needed between the + and - inputs, to
achieve 0 V at the output. This is caused by small differences between the
transistors in the differential gain stage of the operational amplifier. The
presence of an offset voltage means in a feedback circuit, that the first rule
in reality is a bit different. It should have been said that the amplifier will
402 Chapter 6. Analogue electronics in mechatronic systems

R2

I bias
_

+
G ()

+
+

+
V- V+ R1  Vo
I bias
_ _ _

Figure 6.60: The bias current in both inputs of an operational amplifier causes a
voltage drop over the source resistors R1 and R2 . When both resistors
are equal the effect on the output voltage is zero.

do whatever it can to achieve a difference between the input voltages, that

is equal to the offset voltage. The offset voltage is only important in case
of amplifying very small DC voltages, because in practice this voltage is
smaller than a few mV. In some operational amplifiers it can be adjusted to
near zero Volt. The long-term stability of the offset voltage, also called drift,
is important to know, when very small DC voltages need to be amplified
accurately over a long period of time. This drift is mainly caused by the
temperature sensitivity of the transistors.
Ideally the offset voltage is zero without any drift.

Input bias and offset current

The input bias and offset current represent the currents that flow into or
out of the input terminals and that are required to operate the amplifier. In
electronics biasing refers to adding a voltage or current to the useful signal
in order to make the system work. The presence of this current implies,
that also the second rule about zero currents in the inputs is not completely
true. In a bipolar transistor input stage, like with the NE5532, the bias
current is necessary to drive the first transistor with a current level in the
order of several nA. With special transistors like the FETs in the OPA627,
the input current can be reduced to several pA. A low value of the bias
current is important when the path from the inputs to ground has a high
resistive value. This current would otherwise influence the output voltage to
compensate the related voltage drop over the resistors. The influence of the
6.2. Active electronics 403

bias current can be reduced by designing the amplifier in such a way that
the source (Thevenin) impedance of the circuit around the + input equals the
impedance of the circuit around the − input. This is shown in Figure 6.60
where the output voltage is zero in case of an equal bias current on both
inputs and when both resistors are equal:

V + = 0 − R 1 I bias
+
(6.109)
− −
Vo = V + R 2 I bias
V+ = V− =⇒ −
Vo = R 2 I bias +
− R 1 I bias =0

It is clear that an offset in the bias current will result in an offset voltage at
the output even when the resistors R1 and R2 are equal.
It is especially important to be aware that a capacitor as is applied in the
feedback path of an integrator, is not able to conduct DC-currents like the
bias current.
Ideally the bias current is zero.

Noise voltage and current

Due to temperature, stochastic signals are generated in resistances and all

other components of the operational amplifier. This noise is represented as
a voltage over the inputs and as a current into the input. The noise voltage
is directly translated into a noise voltage to the output, determined by the
feedback circuit and the noise currents are translated to the output voltage,
through the source impedance of the circuits around the inputs. The noise
can be represented by a cumulative value, but is mostly represented by a
noise density value as a function of the frequency area.
Ideally the noise is zero.

Input resistance

The input resistance is the value that corresponds with the part of the input
current that is dependent on the input voltage so excluding the bias cur-
rent. This property is due to the transistors and resistors around the input
circuits. Differential-mode is the resistance between both input terminals
and common-mode is the resistance of both inputs to ground. In case of
a feedback system, the differential-mode resistance plays a very limited
role as the voltage between both inputs is kept close to zero. In reality the
differential input resistance is reduced with a factor equal to the loop gain.
With an inverting amplifier also the common-mode resistance is of little
404 Chapter 6. Analogue electronics in mechatronic systems

importance as both input voltages are kept at ground level. Only with a non-
inverting configuration, the common-mode input resistance has influence
on the input currents.
Ideally the input resistance is infinite.

Input voltage range

In most operational amplifiers, the inputs are not allowed to exceed the
power supply voltage level and normal operation is only guaranteed over
an even smaller range. Often the inputs are protected against occasional
higher voltages by diodes but these will also reduce the input resistance
and for that reason, protective input diodes are not used in very critical
designs. Exceeding these values results in clipping of the output voltage to
the maximum level and it can even cause damage to the device.
Ideally the input voltage range is infinite.

Common-mode rejection ratio

The purpose of an operational amplifier is to amplify only the voltage differ-

ence between the two inputs, the differential-mode voltage. This means that
the output voltage should be completely independent of the common-mode
voltage of the input terminals. The ratio between the gain of the differential-
mode to the gain of the common-mode is called the common-mode rejection
ratio (CMRR).
Ideally the CMRR is infinite.

6.2.8.3 Power supply and output limitations

Power supply rejection ratio

Noise and level variations such as ripple of the power supply does preferably
not have have influence on the output voltage. The power supply rejection
ratio (PSRR) is defined as the ratio between a power supply voltage change
and an equivalent input voltage giving the same effect on the output.
Ideally the PSRR is infinite.

Open-loop output resistance

This determines the change of the output voltage as function of the output
current. In a feedback system this value is reduced by a factor equal to
6.2. Active electronics 405

the loop gain. Because the loop gain is frequency dependent, due to the
main pole, also the effective closed-loop output impedance will be frequency
dependent. As it increases with frequency, this can be modelled as a very
small series inductor.

Power consumption

The power consumption is determined by many factors. The first factor

is related to the idle current that is needed in the transistors and these
are modulated by the signal. The power consumption is related to the
slew-rate, as a high slew-rate requires a high internal current to charge
the capacitors. Further also the maximum output current determines the
power consumption.
The power consumption should ideally be zero and for portable devices
special operational amplifiers are developed, that show a very low idle
current level below 1 mA.

6.2.9 Closing remarks on low-power electronics

It is demonstrated that low power analogue electronics can be applied in

many different ways, that all deal with the manipulation of electric signals
by amplification and filtering. The operational amplifier has developed
over time into an extremely versatile building block, optimised for different
applications. With the basic knowledge of this chapter, it is possible to
select suitable components in an analogue electronic system without help of
specialists, but still too many things have remained untouched and need
attention to achieve a real solid design. A few of them are summarised
below:

• The properties of the power supply.

Especially at high frequencies the power supply will not be a very
good voltage source anymore. This requires the addition of ceramic
capacitors between the power supply and ground, as close as possi-
ble to the operational amplifier, to short high-frequency interference
to ground, combined with a small electrolytic capacitor to dampen
any high-frequency oscillations by its internal parasitic resistance.
Sometimes even local linear power supply stabilisation is required. A
stabilised power supply is a special version of a power amplifier, that
will be presented in the following chapter. For moderate power levels,
406 Chapter 6. Analogue electronics in mechatronic systems

voltage stabilising circuits are available as IC, that can directly be

mounted on a printed circuit board.

• Mains supply issues.

The power socket of the mains is not really a nice 230 V AC voltage
source. Sometimes the signal hardly looks like a sine wave at all. The
worst are spikes of several kV, due to circuit breakers at other places
or even caused by a lightning stroke. The electronic system needs
additional safety components to lead away this excessive voltage from
the sensitive parts.

• Precautions for Electromagnetic compatibility (EMC).

Electric current runs in loops and magnetic fields induce voltages in
these loops. These voltages are added to the signal and can cause
errors. This requires a careful lay out of the wiring of an electronic
circuit. This subject will return in Chapter 8 on measurement.

• Precautions against Electrostatic discharge (ESD).

Especially ICs with MOSFET inputs are sensitive for electrostatic
charge build up, that can damage the thin layer between the Gate and
the channel. Besides careful handling in a discharged environment,
this also means that the electronic circuit needs protective components
to prevent electrostatic charge build up.

• The effect of safety grounding.

Most professional electronic equipment is provided with a grounded
housing for human safety. This is not always the most preferred
situation from an interference point of view. Also this subject will
return in Chapter 8.

These and several other issues make the electronic field to an important, ex-
citing and rather difficult specialisation, where a vast amount of knowledge
is gained in the last decades.
6.3. Power amplifiers 407

6.3 Power amplifiers

Combined with the actuators and the mechanics, the power amplifier deter-
mines the dynamic behaviour of the mechatronic plant. Power amplifiers
differ from the amplifiers in the previous section in their focus on reliable
electrical power delivery and efficiency. As will be demonstrated, this fo-
cus has invoked several special technologies, that are only partly or hardly
required in low power signal amplification. With the example of a modern
tube amplifier according to ancient technology as shown in Figure 6.61 it
is already clear that power is equal to size and heat but even with more
efficient semiconductor devices like the ones shown in Figure 6.62 a power
amplifier will remain always recognisable by the large components that
have to sustain elevated voltages, currents and heat. It is also true to say
that the design of power amplifiers will appear to be quite similar in certain
ways, in spite of these differences. Especially the presented topology of the
operational amplifier can be directly used for the design of linear power

Figure 6.61: Only 50 years ago proportional electronic control of high power could
only be achieved by means of electron tubes working in vacuum at very
high voltages with glowing wires to create free electrons. The shown
audio power amplifier can deliver two times 80 Watt while producing
about the same amount of heat. Fortunately semiconductors can do
the job with less effort although also with less beauty.
408 Chapter 6. Analogue electronics in mechatronic systems

Figure 6.62: The high power semiconductors as shown on the right can be distin-
guished from signal transistors as shown left by their mechanical
means to connect them with a heat sink in order to keep the tem-
perature of the semiconductor material below the level where the
semiconductor stops working (≈ 150 − 200◦ ).

amplifiers.
This section starts with an overview of the requirements of power amplifiers
with an emphasis on the application with electromagnetic actuators. This
will be followed by a subsection on linear amplifiers for moderate power
levels with both a voltage and a current source output. The last subsection
will present pulse-width Modulated switching output stages, that are used
to reduce the dissipation at high power levels.

6.3.1 General properties of power amplifiers

The following main specification elements determine the design of a power

amplifier.

• Power delivery capability.

• Dynamic properties.

• Output impedance.

• Efficiency.

• Linearity.
6.3. Power amplifiers 409

Power delivery capability

A large power implies automatically a high value of the current and volt-
age. As it appears to be difficult to integrate high voltage electronics, the
power stages in these amplifiers are often built with active and passive
discrete components. These large coils, resistors, capacitors and high power
semiconductors generally produce a lot of heat. Silicon semiconductors can
withstand temperatures on the chip of 150◦ – 200◦ , but due to the heat
transfer resistance of the packaging, mostly the temperatures at the outside
of the device have to be kept at a more reduced temperature level.
These thermal issues automatically require efficient methods to remove
the heat from the sensitive electronics. The mechanical design of signal
amplification electronics with thin printed circuit boards is a rather two
dimensional exercise but the lay-out of a power amplifier is a real three
dimensional mechanical challenge. The thermal design with cooling plates
and the use of large parts requires a design, that takes the mass and the
environmental mechanical dynamics under serious consideration. Solder
joints for instance are not capable of sustaining high mechanical loads and,
while this is a frequent cause of malfunctioning electronics, this is just an
example of the special skills needed when designing power amplifiers.
Another issue of the high power is the potential interference by magnetic and
electric fields, the Electro Magnetic Compatibility (EMC), caused by rapidly
changing high voltages and currents. Also this aspect requires a careful
three dimensional lay-out with shielding plates and conscious routing of the
wiring.

Dynamic properties

When a power amplifier is used in a mechatronic system, it determines a

part of the feedback loop. A signal delay caused by the amplifier adds one or
more poles with their related phase lag to the open-loop transfer function
and reduces the stability margins. Especially with a digitally controlled
amplifier (sample time) this is an aspect to pay attention to. In an analogue
amplifier this delay is only related to the frequency response. Even when
the amplifier is fast enough under normal resistive loads, the complex load
of an electromagnetic actuator (L, R, C ) will influence the internal stability
of the amplifier. For that reason, power amplifiers for high performance
mechatronics need to be designed specifically for a certain application.
410 Chapter 6. Analogue electronics in mechatronic systems

Output impedance

A low output impedance of an amplifier is required when damping in the

system is necessary, as was presented in Section 5.4.1.3 of Chapter 5. This
is for instance the case if the amplifier is used in combination with loud-
speakers, where the amplitude at the lower resonance frequency needs to
be limited.
A high output impedance is useful when the current and the force needs
to be independent of the movement and the self-inductance of the actuator
windings. This is mostly the case in servo-controlled positioning systems
with Lorentz actuators. With a high output impedance there is no force
change caused by a relative velocity of the mover and stator, resulting in a
reduced transmissibility between these parts when compared with a low out-
put impedance amplifier. In Section 5.4.1 of Chapter 5 it was further shown
that the inherent compensation of the self-inductance by a high output
impedance has a certain danger, because too often mechatronic designers
tend to think that the self-inductance is not important anymore, because of
the availability of powerful electronics. In a high speed positioning servo-
system this perception can have as consequence that the amplifier has to
deliver a significantly higher voltage to overcome the high values of L( d I/d t)
than is required to deliver the power in the system, because these maximum
values mostly occur at different moments in a periodic cycle due to their
reactive 90◦ phase relation. In that case the maximum voltage times the
maximum current of the amplifier needs to be far higher than its power ca-
pability. This requirement can have a significant impact in the total system
cost and reliability and is elaborated more in Section 6.3.2.2.

Efficiency

An amplifier transforms a supply voltage and current into an actuator

voltage and current. When this transformation is done by means of a linear
element like a transistor, the actuator current has to run through this
transistor. This actuator current multiplied with the voltage difference
between the power supply and the output determines the power that is
dissipated inside the amplifier. In case of the above mentioned high self-
inductance, this inherently leads to a high power dissipation. As an ultimate
solution a switched-mode pulse-width modulated amplifier is necessary to
reduce or overcome the dissipation.
6.3. Power amplifiers 411

Linearity, freedom of distortion

Non-linearity in an amplifier has several effects. In audio appliances it is

the source of audible distortion, but in professional mechatronic systems
the non-linearity impairs the behaviour in two other ways. First of all it
changes the gain of the amplifier as function of the actual value of the output
voltage or current. As a consequence the loop-gain in a feedback system
becomes dependent on the actual output voltage with ultimately a chance
to become unstable. In practice it is not difficult to limit the non-linearity
to such a low level, that this instability issue hardly ever occurs, with one
exception: The ultimate non-linearity happens when the output voltage has
reached its maximum level, close to the supply voltage. At that moment the
output voltage “clips” and all control is lost. This effect was introduced in
Section 4.5 of Chapter 4, where it was explained how such extreme clipping
can result in an uncontrolled “limit-cycling” of the motion system.
But even without this extreme example, a small non-linearity can cause
errors with feed forward control. As long as the non-linearity is reproducible
and deterministic it can be compensated, but non-linearity of an amplifier
is a sign of a non-optimal design and that almost automatically implies the
occurrence of thermal drift.

6.3.2 Linear power amplifiers

Figure 6.63 shows a basic design of a linear power amplifier with a volt-
age source output stage. At a first glance this design looks very familiar
to a low power operational amplifier. There are however at least two sig-
nificant differences. First of all the gain stage is designed to be able to
deliver a high output voltage, significantly above the regular power supply
voltage of ≈ +/ − 15 V of the differential input stage. This is achieved with
a level-shifter that consists of a discrete pair of high voltage transistors
in voltage amplification mode. The second difference is the high-current
power-output stage, that consists of a set of transistors with a very high
current-amplification ratio, like the Darlington pairs that are named af-
ter the American electro technician Sidney Darlington (1906 – 1997), who
invented the configuration. These pairs are in fact two cascaded emitter
followers and can deliver currents up to several Ampères with a current-
amplification of more than 1000. Also sometimes high-power MOSFETS
are used as these do not require a continuous input current.
The function of the level-shifter is as follows. The input of transistor T3 is at
a low voltage level and gets its input voltage from a low-power operational
412 Chapter 6. Analogue electronics in mechatronic systems

Vp 50 V

68 
Da1

T1
27 k

10 k
50 pF

0.1 

I idle | 10 mA
_

+
Vbias

+
0.1 
G () T3 _ | 2.7 V Vo

Rload
1k

+
_
+

Vi T2
_
10 k
Da2
Vp 50 V

Figure 6.63: Basic design of a linear power amplifier with a voltage source output
stage and a closed-loop gain of ≈ 30 dB, showing the main components.
The transistors T1 , T2 and T3 determine a level-shifter to achieve a
large output voltage range. The push-pull output stage consists of two
Darlington pairs Da1 and Da2 , that can handle a large output current
up to 10 A. The operational amplifier serves as differential input and
gain stage and the capacitor C creates the dominant pole for stability.

amplifier, acting as a combination of a differential input stage and a high-

gain stage. T3 is a voltage amplifier with a collector resistor of 10 kΩ. The
collector voltage of T3 is used as input for the base of T2 , that also acts as a
voltage amplifier with a current source as collector resistor.
T1 with its emitter resistor of 60 Ω and the two diodes, that give a constant
base voltage for T1 define a current source with an idle current level of
≈ 10 mA, taking the base-emitter threshold voltage of 0.6 V of T1 into account.
This configuration guarantees, that the voltage swing is not limited as with
a resistor instead of a current source, the collector current of T2 would be
dependent on the output voltage. Furthermore, a current level of 10 mA is
sufficient to drive the Darlington pairs up to an output current level of 10 A,
when the current-amplification of the Darlington pairs is ≥1000.
The feedback circuit is arranged as a non-inverting amplifier with a gain of
28 which is equivalent to ≈ 30 dB. A separate feedback capacitor is added to
create the dominant pole, while reducing the non-linearity of the high gain
level-shifter. The resulting closed-loop bandwidth is 100 kHz, determined
by the RC-time of the capacitor with the feedback resistor of 30 kΩ.
6.3. Power amplifiers 413

+
G ()

+
+ R1 _
Ra

Actuator L
a
Io -1
Vm

+
1 Ra

Loopgain
Z0
Vi _ Vo We La
R2 -2

Vcs I o R3 Z0 Z

R3
_ _

Figure 6.64: Current-feedback is achieved by measuring the current with the series
resistor R3 and using this voltage as feedback signal. Because of
rule one the current through the actuator is only dependent on the
input voltage and not on the output voltage. Unfortunately the self-
inductance introduces an additional pole that can cause instability.

6.3.2.1 High output impedance amplifiers

The electromagnetic actuators that are used in most precision mechatronic

motion systems require a power amplifier with a high output impedance. The
previously presented high power output stage has a low output impedance,
due to the emitter follower configuration and the voltage-feedback loop.

Current-feedback

This problem can be solved by creating a transconductance amplifier where

the high output impedance is created by means of current- instead of voltage-
feedback. This principle was introduced in the previous section and is re-
peated here in Figure 6.64 to explain the issues that are related to this
configuration when used with a reactive load. In this case the power ampli-
fier is symbolised as a standard operational amplifier and can be assumed
identical to the amplifier of Figure 6.63 without the voltage-feedback circuit.
The current is measured with a small series resistor R3 and the voltage over
this resistor is fed back to the negative input of the power amplifier.
As was shown in Section 6.2.4.6 this configuration creates an output current
that is only depending on the input voltage and independent of the actuator
414 Chapter 6. Analogue electronics in mechatronic systems

+
G ()

+
R1 _
+
Ra

Actuator L
a
Io -1
Vm

+
1 Ra

Loopgain
Z0 R2C
Vi _ Vo We La
R2 -2

Z0 Z

R3
C
_ _

Figure 6.65: Pole compensation by a parallel capacitor over the feedback resis-
tor cancels the effect of the self-inductance but impairs the current
measurement and reduces the closed-loop gain at higher frequencies.
This can in principle be compensated by an additional differentiating
lead-network in the controller.

voltage. Unfortunately this is only true when the feedback loop has sufficient
closed-loop gain and phase margin. With a real electromagnetic actuator
the self-inductance of the actuator-windings causes problems, because above
a certain frequency the impedance of the self-inductance will determine
the impedance of the actuator as was explained in Section 5.4.1.1. This
frequency equals:
1 Ra
ωe = = (6.110)
τe La

Above this frequency, the feedback path from the output of the amplifier
to Vcs will have a low pass characteristic with one pole and a −1 slope that
adds to the internal −1 slope of the operational amplifier. This results in a
marginal stability of the feedback system as is visible at the combined −2
slope in the simplified amplitude Bode plot of Figure 6.64. This effect could
be compensated by creating a differentiating action, a lead-network in the
loop. This is possible by adding a parallel capacitor over R2 with the value:
La
τe = R 2 C = (6.111)
Ra

This would indeed increase the phase margin as shown in Figure 6.65 but
it would also change the closed-loop current source characteristics of the
6.3. Power amplifiers 415

Vp 50 V

+
Vbias

1
| 1.4 V _

100 
9k1

_
Da1
G () T1
1k

+
+

100 
Vbias _

+
Vi
_ | 1.4 V

Rload
Vo
T2
Da2 _

100 
+
Vbias

1
| 1.4 V _
Vp 50 V

Figure 6.66: By reversing the Darlington pairs of Figure 6.63 a current source
output is obtained. This amplifier delivers 10 A output current at 1 V
input voltage. For the positive half of the signal, transistor T1 acts
as a level-shifting voltage amplifier with a gain of one, Da1 converts
the voltage of the collector of T1 into an output current. The negative
cycle is taken care of by T2 and Da2 with their surrounding resistors.
The operational amplifier only acts as a non-inverting amplifier with
a gain of 10, without feedback over the total amplifier.

system, that in its turn could be compensated by an additional lead-network

in the controller. It is up to the reader to work this further out as an exercise.
Generally it is better to avoid these compensation circuits as they are never
perfectly tuned. The solution is that a power output stage is designed with
a high output-impedance in open-loop as shown in Figure 6.66.

Current-source output stage

In this extreme example the current-feedback loop is completely avoided and

the voltage amplifier level-shifter transistors and output stage are linearised
with large emitter resistors. Furthermore the output stage of the amplifier
is designed directly with a high impedance.
In this example, the operational amplifier acts only as a non-inverting
416 Chapter 6. Analogue electronics in mechatronic systems

voltage amplifier with a gain of ten and a low output impedance and its
output voltage is used as the input for the two level-shifter voltage amplifying
transistors T1 and T2 . The current in these transistors is determined by
the resistor of 100 Ω, that is shared to linearise the transition between
the positive and the negative cycle. In the positive cycle of the signal, the
current in T1 will increase, resulting in an equal decrease of the voltage at
the base of the Darlington pair Da1 , because of the 100 Ω collector resistor.
This results in an increased voltage over the 1 Ω resistor in the emitter of
Da1 , with a corresponding increase of the positive output current. For the
negative cycle T2 and Da2 act the same.
A special aspect in this configuration is the equal value for the shared
emitter resistor and the collector resistors of the level-shifters. This is done
to linearise the amplification of the level-shifters, because there is no further
feedback that could correct any non-linearity. This is also the reason for the
relatively high value of the emitter resistors of the Darlington pairs. The
low value of 100 Ω around the level-shifters is chosen such that the current
in the level-shifters at the maximum output current of 10 A is more than
a factor ten higher than the input current of the Darlington pairs, again
for linearisation reasons. At 10 A output current, the corresponding 10 V
voltage over the 100 Ω resistors requires a current in the level-shifters of
100 mA. The input current of the Darlington pairs is then 10 mA because of
the current-amplification ratio and that is only 10 % of the current in the
level-shifters.
It is clear from this example that all these measures come at a price in power
dissipation and performance. The relatively high currents and resistor
values both limit the output voltage and require significant measures to
keep the temperatures of the semiconductors and the other parts at an
acceptable level.
For that reason a more optimal solution will be to combine the current-
feedback configuration with grounded load of Section 6.2.4.6 with the output
stage from Figure 6.66, while using smaller emitter resistors to limit the
power dissipation. The current-feedback will reduce any non-linearity and
the collector follower configuration will reduce the problem of the additional
pole by the self-inductance of the actuator.
At the end of this chapter, when presenting three-phase resonant-mode
amplifiers, an alternative loss-less current sensing method will be shown
which gives an almost ideal amplifier configuration and can be used also
with linear power amplifiers.
6.3. Power amplifiers 417

Vp+
Current
I,V Voltage
V1
0

Vp- t1 t

Figure 6.67: With a complex load, the current from the amplifier will have a phase
difference with the voltage. This means, that the amplifier needs to
have a four quadrant capability by delivering positive and negative
currents at any sign of the output voltage. This requirement implies
a very high power dissipation.

6.3.2.2 Dynamic loads, four-quadrant operation

When an inductive or capacitive load is applied, the power dissipation in a

linear amplifier is even more severe, because in that situation the current
is out of phase with the voltage. As was explained in Chapter 5, electro-
magnetic actuators in mechatronic positioning systems show a sometimes
large self-inductance and with piezoelectric actuators the load is mainly
capacitive. In the extreme case of a purely reactive load the maximum cur-
rent needs to be delivered at zero voltage, while at a quarter of the period a
positive current is delivered with a negative voltage and at another quarter
it is just the other way around, like shown in Figure 6.67. The capability
to deliver currents with a different sign from the voltage at any moment
is called a four quadrant operation. In linear amplifiers, the requirement
to drive complex loads implies that the voltage difference over the power
transistors at certain current levels can become even larger than the supply
voltage. For instance in Figure 6.67 at time t1 the still high positive cur-
rent has to be delivered by the power transistor connected to the positive
power supply, while the output voltage is still negative. The voltage over the
transistor that conducts the current is then as large as V1 , which is more
than Vp+ .
Fortunately, reality is not always this extreme, and when the resistive part
of the impedance of the actuator is large, the negative effect is reduced.
This is the case in most audio applications where the reactive currents in
loudspeakers are quite limited.
418 Chapter 6. Analogue electronics in mechatronic systems

Motion induced voltage

In mechatronic positioning systems the real problem is caused by the motion

voltage of the actuator when high velocities are applied. At acceleration,
the motion voltage increases in phase with the current and electric power is
inserted in the system and converted into kinetic energy. The deceleration
phase is however completely the opposite. While the motion voltage still has
the same sign as with the acceleration, the current needs to be reversed and
the full amount of kinetic energy has to be absorbed by the amplifier.
Especially for this reason, the use of linear amplifiers in mechatronic systems
is restricted to moderately low levels of power, with limited velocities, like
in short stroke Lorentz actuators.
Because of these thermal issues and an increasing need for higher out-
put power, the switched-mode power amplifiers are developed, that base
their operation principle on alternately exchanging energy in inductors and
capacitors.
6.3. Power amplifiers 419

6.3.3 Switched-mode power amplifiers

At the start of Chapter 6, the capacitor and inductor were introduced as

reactive components, capable to store electric energy. It was also shown that
filters can be designed with capacitors and inductors that have resonating
properties. At resonance the energy flows alternately between the capacitor
and the inductor. Furthermore a voltage over a capacitor can not be changed
infinitely fast, but needs a current to change. The same is true for the current
in an inductor, that needs a voltage to change. These effects are applied
in the following section where electronic switches are used to direct the
currents and voltages in reactive elements in order to create an amplifier.
Two typical switched-mode amplifiers will get an illustrating role in this
section, both applied in the wafer stages of ASML. The first example as
shown in Figure 6.68 is an early prototype, designed around 1980 at Philips
“NatLab” and was used to drive the first electric wafer stage in the Silicon
Repeater as was presented in Chapter 1. This amplifier delivered moderate
output voltages with ±30 V and currents up to 5 A. The second amplifier is
a three-phase high-power amplifier, designed around 1995 at the Philips
Centre For Technology and its basic principle is still used to drive the 5 kW
long-stroke actuators in the Twinscan wafer scanners.
After an introduction of the first example amplifier, this section continues
with the the required fast electronic switches with the power MOSFET as
the main element. This is followed by an explanation of the principle of
pulse-width modulation in a switched-mode voltage-source amplifier with
the first amplifier as example. With this basic understanding, a deeper
analysis of current-flows in switching power-output stages is presented,
finalised by an optimal design method, using the resonant-mode of the
reactive components in the three-phase ultra-high power amplifier of the
second example.

6.3.3.1 First example amplifier

The first example amplifier is a proven design of a pulse-width modulated

switched-mode amplifier. A copy of the original hand drawing of the circuit
diagram from 1980 is shown in Figure 6.69. This amplifier was designed
for driving the linear electric motors of the wafer stage, shown on the cover
of this book for the Silicon Repeater Mark 2. The circuit diagram gives an
indication of the relative simplicity of the design, with some digital switch-
ing components, like C-MOS integrated circuits (the little triangles), two
operational amplifiers, six power MOSFETs, three bipolar transistors and
420 Chapter 6. Analogue electronics in mechatronic systems

Figure 6.68: The switched-mode power amplifier from Figure 6.69 that was de-
signed for the first electric wafer stage for the Silicon Repeater wafer
stepper.

several passive components. Because of its straightforward functionality,

this amplifier is used as example when explaining the working principle of
switched-mode amplifiers and the functionality of every main building block
will be illustrated by using the actual component values and other system
properties, that will become clear in this chapter.

• The switching frequency f s of this amplifier is 150 kHz at 0 V output

voltage.

• The switching output stage consists of two single-ended switching

power units in counter phase using only a single supply voltage of 30 V.
This is equivalent to one single-ended output stage with two supply
voltages of plus and minus 30 V.

• The output filter consists of an inductor with a self-inductance of 1 mH

and a capacitor of 0.33 μF, giving a corner-frequency f 0 of 8.7 kHz with
a −2 slope in the attenuation band.

• The impedance of the actuator consists of a resistor of 5 Ω in series

with a self-inductance of 8 mH, giving an electric time constant τe =
L/R = 1.6 ms
6.3. Power amplifiers 421

Figure 6.69: An early example of a pulse-width modulated switched-mode power

amplifier, designed for the linear motors of the first electrical wafer
stepper of Philips Electronics and ASML. (Original hand drawing of
the circuit diagram from 1980.)
422 Chapter 6. Analogue electronics in mechatronic systems

6.3.3.2 Power MOSFET, a fast high-power switch

The currents and voltages in switched-mode amplifiers need to be switched

at a very high frequency in the order of 100 kHz or more. An ideal electronic
switch is able to conduct current in both directions, has zero resistance
at closure, can withstand very high currents and voltages and has zero
switching time. And last but not least an ideal electronic switch does not
require energy to change its state.
It is clear that also in this case “ideal” does not exist, although the continuous
improvements in electronic technology have brought us near ideal devices
named power MOSFETs.
In precision mechatronic positioning systems, bipolar transistors are seldom
used as switching devices. The reason is, that they require a relatively high
base current. Further they are relatively slow, due to the low speed of the
holes as charge carrier and they suffer from a phenomenon called secondary
breakdown, a destructive mode that limits their power capability. On the
other hand their ruggedness has made them popular in power-conversion
systems, that do not need a very high switching frequency. The high base
current has been solved by integrating a MOSFET as input device before
the bipolar transistor. This combined device is called an Insulated Gate
Bipolar Transistor (IGBT) and can switch several kilo Volt with several kilo
Ampère current. Because of their relatively high switching times in the
order of several tens of microseconds they are not preferred in high precision
mechatronic positioning systems but found their application in electrical
transport systems like trains and hybrid cars.
For the more high-frequency applications always N-channel power MOS-
FETs are used, as they offer switching speeds in the order of a few nanosec-
onds, because they use only electrons as charge carriers. Power MOSFETs
are created by placing millions of small MOSFETs in parallel. This is
allowed as each separate MOSFET behaves like a variable resistor and
resistors can be placed parallel to create a smaller resistor value. This
parallelism requires a special configuration of the MOSFET elements, dif-
ferent from the horizontal configuration that was shown in Figure 6.34. The
vertical structure from Figure 6.70, with the Drain at the bottom side of the
chip and the Source at the top side, enables to directly connect all Drains
and all Sources, while the Gates are connected via an insulated embedded
wiring network spread over the chip.
One special property of a MOSFET has not been mentioned yet. Most
certainly the reader will have recognised, that a MOSFET is an essential
6.3. Power amplifiers 423

!rce Gate Pysilicn

Oxide Gate
S!rce

ƒ

p+ Channels p+ D
n+ p n+

n- Epi Layer
G
n+ Substrate S

Drain

ƒ Drain

Figure 6.70: A power MOSFET consists of millions of small MOSFETs connected

in parallel. This is accomplished by a vertical design where the Source
and the Drain are at two sides of the chip and the Gates are embedded
in an insulating oxide layer. The symbol shows that the substrate is
connected to the Source. As a direct consequence of this connection,
an additional diode is present between the Drain and the Source,
determined by the junction between the Drain and the substrate. This
diode is used effectively in switched-mode amplifiers. (Courtesy of
International Rectifier)

symmetric structure, where the current can flow in both directions. Indeed
that is the case and although these devices are optimised to work with
currents running in only one direction, especially in power MOSFETs this
phenomenon is used to advantage in switched-mode amplifiers. This op-
timisation to work in only one direction has a direct connection with the
practice to connect the Source directly to the substrate, as they will have the
same potential, when the current runs from the Drain to the Source in an
N-channel MOSFET. As a consequence, the junction between the N-material
of the Drain and the P-material of the substrate determines a Silicon diode
between the Source and the Drain. This diode helps conducting current
from the Source to the Drain, opposite to the normal current direction and
it will be demonstrated, that this is a very useful property in switching
inductive loads.
Like all other elements in electronics, also power MOSFETs suffer from
several parasitic capacitances and for that reason they are not really ideal.
The most important is the capacitance of the Gate, that amounts in the
order of 10 nF for a power MOSFET that can switch more than 200 A. Also
the capacitance between the Drain and the Gate plays a similar negative
424 Chapter 6. Analogue electronics in mechatronic systems

role as the collector-base capacitance of a bipolar transistor. Its Miller

capacitance as experienced at the Gate is fortunately less than the input
capacitance, but still the rather large total capacitance requires a charging
and discharging current each time the MOSFET switch needs to change its
state. As a result a net current is flowing in and out of the Gate. In high
power applications the switching voltage at the Gate needs to be at least
10 V and with a switching frequency f s of 100 kHz, an input capacitance of
10 nF gives an effective current I g to the Gate of:
1
I g = f s CV 2 = 105 · 10−8 · 100 = 0.1 [A] (6.112)
2
This value needs to be compared with the maximum current of 200 A that
can be switched. It is still below the value of a comparable Darlington pair in
a continuous current situation, with a current-amplification factor β = 1000,
and this is without the additional base current necessary to charge the
Miller capacitance. In fact this current level to the Gate is quite acceptable.
It should be noted however, that this current is directly proportional to the
switching speed. It is one of the sources of power dissipation in a switched-
mode power amplifier.

6.3.3.3 Pulse-width modulation

switched-mode power amplifiers operate according to the principle of pulse-

width modulation. This principle is based on the understanding from Fourier
analysis, that a periodic signal can be decomposed into one constant value
a 0 , representing the average DC value of the signal and a set of higher
harmonics, of which the lowest frequency equals the temporal frequency of
the periodic signal. By filtering these harmonics with a low-pass filter, only
the average value remains.
For a square-wave signal the average value is zero, when the negative cycle
is equal in time to the positive cycle. The ratio between the positive cycle
and the full period is called the duty-cycle. Because of this definition, the
average DC value VDC of a square-wave signal is equal to the following
expression, where V − is the negative voltage of the square-wave signal and
V + the positive value:
( )
VDC = V − + V + − V− · duty − cycle. (6.113)

With V + = −V − a duty-cycle of 50 % gives an average voltage of zero Volt.

By modulating the duty-cycle of a square-wave signal with a constant am-
plitude, the average value of the square-wave signal can be changed. This
6.3. Power amplifiers 425

process is called pulse-width modulation. When the square-wave signal

is created by means of switches, alternately connecting the output of an
amplifier to either a positive or an equal negative high power supply voltage,
the average value of the output voltage is determined only by the duty-cycle,
independent of a load connected to this output. This average voltage can
be obtained by filtering the higher Fourier terms out of the signal with a
passive low-pass filter, consisting of an inductor and a capacitor.

Pulse-width modulator

The first step to create a switched-mode amplifier is to convert a low-

frequency signal into a square-wave signal with an average value that
is equal to the momentary value of the low-frequency signal. In the example
amplifier this process is based on a triangle – square-wave pulse-width
modulator and is explained with the help of Figure 6.71.
The heart of the modulator consists of an inverting integrator and a Schmitt
trigger, named after the American scientist Otto Herbert Schmitt (1913 –
1998) who invented it. The output of the Schmitt trigger can only have
two states. In this amplifier one state is a positive voltage and the other
state is a negative voltage, both with an equal magnitude. The output can
only change its state from negative to positive, when the input surpasses
a certain positive threshold and it can change its state from positive to
negative only when the input comes below a certain negative threshold.
In this amplifier also both threshold levels are equal. The output of the
comparator is used as input for the high power switching stage and the
high-frequency part of the output of that stage is filtered out by a passive
LC-filter, transferring only the average DC value of the output signal. The
switching output signal before the filter is fed back to the integrator. First
the principle will be explained when assuming that both outputs are equal,
because this switched output is an almost equal square-waveform as the
output of the comparator.
Starting with the comparator in the positive output state, the output of
the inverting integrator will show a negative slope over time at a speed
depending on the voltage of the comparator, the input voltage, the capac-
itor value and the resistor values of R1 and R2 . As soon as the output of
the integrator has reached the negative threshold of the Schmitt trigger,
the latter will switch its state to a negative voltage and the output of the
integrator will start to rise again, until the positive threshold is reached.
At that moment the Schmitt trigger changes its state back to the positive
426 Chapter 6. Analogue electronics in mechatronic systems




+ 1 Vo

+
G ()

+
_ Vi …
R1
~ 

 LC- Vo
+

 

_
Vi 2
!!
_
C
R2
 _

Vi

2
Vo

Figure 6.71: pulse-width modulator as used in the switched-mode power amplifier

of Figure 6.69. The sum of Vi and the output of the balanced switching
output stage is integrated by the inverting integrator. In combination
with the Schmitt trigger in a closed feedback loop, this creates an
oscillator with an average output voltage proportional to the input
voltage.

output voltage. This process continues, resulting in a square-wave signal

at the output of the Schmitt trigger with a corresponding triangle wave
signal at the output of the integrator. When the input voltage equals zero,
the resulting square-wave and triangle wave will be fully symmetrical with
both an average value of zero Volt. A soon as a positive input voltage is
applied, the resulting current in R1 in the direction of the minus input of
the integrator will add to the current in R2 . As a result the down slope of
the integrator will become steeper relative to the up slope, resulting in a
tilted triangle wave at the output of the integrator with a non-symmetrical
square-wave at the output of the Schmitt trigger. The triangle wave output
of the integrator will always remain zero in average as the gain of the inte-
grator is infinite for zero Hertz. This means that the average value of its
inputs also needs to be zero to fulfil rule one of the operational amplifier.
This implies that the average value of the square-wave output of the Schmitt
trigger is proportional to the input voltage in a ratio that is determined
by R1 and R2 . By filtering the square-wave by means of the LC-filter, this
average value is obtained at the output and the system acts like a normal
inverting amplifier.
6.3. Power amplifiers 427

In reality the balanced switching output stage creates a high power version
of the square-wave from the output of the Schmitt trigger. By measuring the
output before the filtering, a signal is obtained that in case of an ideal power
stage would be proportional to the output of the Schmitt trigger. Because
nothing is ideal, this feedback of the real switched output signal will help to
correct errors in this stage because of the high gain of the integrator.
The Schmitt trigger in the example amplifier of Figure 6.69 is realised with
positive feedback around the second operational amplifier (IC2).
428 Chapter 6. Analogue electronics in mechatronic systems

6.3.3.4 High-power output stage

To investigate the relation between current, voltage and power, Figure 6.72
gives a detailed insight in the behaviour of a switching power output stage,
including the reactive components.
To illustrate the effects with real numbers this example shows the signals
with the actual component values of the example amplifier, a 150 kHz
switching frequency with an output filter consisting of an inductance of
1 mH and a capacitor of 0.33 μF. An equivalent single-ended output stage is
chosen with equal positive and negative supply voltages of 30 V. A duty-cycle
of 50 % results in an average output voltage of 0 V.
One cycle of 6.66 μs is shown in the figure, equally divided in four steps.
The starting point of the first step is taken when the current in the inductor
is zero and the first power MOSFET M1 is just switched on, transferring
the full positive power supply voltage of 30 V to the voltage Vb at the bridge
between the switches. As will become clear after the last step, the voltage Vo
at the output of the filter is then at its maximum negative value of ≈ −0.1V
and never grows bigger than an alternating value of this same magnitude.
Because of this low value of the output voltage at this duty-cycle, the total
voltage difference over the inductor is only determined by the bridge-voltage
Vb = 30 V and as a result the current in the inductor of 1 mH will increase
in the first 1.66 μs to a maximum level Î L determined by:
τ
dI Vb ( t) Vb
Vb = L =⇒ Î L = dτ = d t ≈ 0.05 [A] (6.114)
dt L L
0

This current flows in the capacitor of 0.33 μF and the voltage change is
calculated by integrating the current over the 1.66 μs:
τ
dVo I L ( t) Î L
IL = C =⇒ δVo = dτ = 0.5 d t ≈ 0.1 [V] (6.115)
dt C C
0

As a result the voltage Vo is zero after the first step.

In the second step M1 is switched off and M2 is switched on shortly after
but not immediately, as that could cause a short-circuit current as will be
explained later. However, in spite of this short delay, the voltage Vb will
quite rapidly jump to the negative power supply voltage of −30 V after M1
is switched off. This is caused by the current in the inductor, that needs
a voltage to change and will remain running until a negative voltage has
reduced the current level to zero again. Because of the delay in switching of
6.3. Power amplifiers 429

Vp+ Vp
+

Io Io

M1 Bridge L M1 L
+ + +
IL Io Io IL Io
Vb IC Vo IC Vo
C C
- - -
M2 M2
- -
Vp Vp
Step 1: M1 switched on. Step 2: M1 
 <“2 switching on.
IL
š!
Q   IL
š!
! 2 and decreases.
Vo
—
'Q  Vo
—
'Q
+ +
Vp Vp

Io Io
M1 L M1 L
+ +
Io IL Io IL Io
IC Vo IC Vo
C C
- -
M2 M2
- -
Vp Vp
Step 3: M2 switched on. Step 4: M2
 <“1 switching on.
IL
š!
 
Q IL
š!
! 1 and decreases
Vo
—'Q  — Vo 'Q

Vb + 30 V Vb Vb Vb
0 0 0 0

+ 0.05 A - 30 V
IL IL IL IL
0 0 0 0
3.33 s

- 0.05 A
Vo Vo + 0.1 V Vo Vo
0 0 0 0

t0 - 0.1 V t t0 t t0 t t0 t
Step 1: Step 2: Step 3: Step 4:

Figure 6.72: Current flow in a switching power stage of a switched-mode power

amplifier with power MOSFET switches and an LC-filter. The red ar-
rows correspond with the momentary direction of the high-frequency
current within a cycle and the black arrows correspond with the di-
rection of a positive low-frequency current. One cycle consists of four
equal steps of 1.66 μs for each step, corresponding with an average
output voltage of 0 V.
430 Chapter 6. Analogue electronics in mechatronic systems

+30 V +30 V
Vb 0V Vb 0V

+3.05 A

IL +3A

+0.05 A
IL 0A +15.1 V ; +3.02 A
Vo+15 V
Io +3 A

+0.1 V ; +0.02 A
Io 0A
Vo 0V
t0 t1
Transion from 50 % to 75 % duty cycle at t1 A$er stabilisao‚
$ '„

Figure 6.73: When the amplifier is loaded with a resistive load of 5 Ω a change in
the duty-cycle from 50 % to 75 % will cause an average output voltage
of +15 V with a corresponding average output current of 3 A. At the
transition moment t0 first the current and the voltage will increase
until an equilibrium is reached. The ripple remains almost constant
with only a change of waveform.

M2, the current will first flow through the diode of M2 before M2 is switched
on completely. This proves the usefulness of this diode as without the diode
the voltage Vb could be forced to more negative values than Vp− , leading to
damage of the MOSFET.
As a result of the negative voltage over L the current I L will decrease again
until it is zero at the end of the second step.
The third and the fourth step are identical to the first and second step. Only
the signs are different and as a result the end of the fourth step is identical
to the beginning of the first step.
In this still quite ideal situation it is clear than no power is dissipated in any
element. The switching MOSFETs either conduct current without a voltage
between their terminals and the reactive elements only store energy.
The reasoning from Figure 6.72 is true for the situation with 50 % duty-cycle
and zero average output voltage. In that case also the average output current
I o to a resistive load is zero. This amplifier was designed for a 5 Ω load
and Figure 6.73 shows what happens with the current, when the duty-cycle
is changed to 75 % and the load of 5 Ω is applied. At that duty-cycle the
6.3. Power amplifiers 431

average output voltage will be 15 V and the corresponding average output

current is 3 A. As soon as the duty-cycle is changed, the current I L in the
inductor L c will increase, with a corresponding more continuous increase
of the output voltage Vo and output current I o . A higher level of Vo has as
consequence that the voltage difference over the inductor at the upward
current slope is reduced and the voltage difference at the downward current
slope is increased. This causes these slopes to change until at equilibrium
the slopes just correspond with the difference in timing. For instance at
75 % duty-cycle, the voltage difference in the upward current slope in the
inductor is 15 V and the voltage difference at the downward current-slope
is 45 V, while also the timing is 1 : 3. This means that the peak to peak AC
level of the current is unchanged and this also means that the peak-to-peak
level of the voltage is not changed. In fact only the shape of the signals is
changed.

Magnitude of the ripple voltage

The calculated ripple voltage of Equation (6.115) can be verified by using

the attenuation of the LC-filter on the original square-wave signal. Because
of the −2 slope in the attenuation band of this second-order filter, the peak
ripple voltage V̂ripple due to the peak output voltage V̂b at the switching
output equals approximately:
 2  2
f0 8.6
V̂ripple = V̂b = 30 ≈ 100 [mV] (6.116)
fs 150

This corresponds with the value found in the previous analysis and is about
-50 dB below the maximum voltage level of the switching output. It is
important to note that this ripple voltage is always present, even without a
low-frequency output signal. The only way to reduce this ripple would be by
either increasing the switching frequency, by taking a higher order filter,
or by lowering the corner-frequency of the filter. In one way this illustrates
the need for very fast switches when these amplifiers are applied in high
precision positioning systems. On the other hand a signal of 100 mV at
150 kHz is not expected to give much acceleration in a heavy positioning
system and it is also not audible too.
Like with all decisions in mechatronic design, for every specific situation it
has to be examined, whether these artefacts would give a problem or not.
432 Chapter 6. Analogue electronics in mechatronic systems

6.3.3.5 Preliminary conclusions and other issues

With these results it is shown that in principle an amplifier can be created

that gives a low-frequency output signal proportional to the input signal
by modulating a high-frequency switching signal and filtering the high-
frequency components. No power dissipation takes place, because there are
only switches and reactive components and all would be well when there
were not some issues that still need attention.

• The Gates of the N-channel MOSFETs need to be driven relative to

the Source.

• charge-pumping occurs with single-ended output stages.

• The interaction between the output filter and the load impedance.

• Speed limitations of the power MOSFETs.

• The need for a current source output impedance.

• The need for three-phase amplification

These issues are all addressed in the following subsections to complete this
chapter on power amplifiers.

6.3.3.6 Driving the power MOSFETs

A specific requirement of power MOSFETs is that they need a voltage of

around 10 V between the Gate and the Source to be switched on. Also the use
of only N-channel MOSFETs poses special requirements on the electronic
circuitry to drive the Gates, while the high input capacitance at the Gate
requires the driver to be capable of delivering a very high current in an
extremely short time. For the MOSFET at the negative supply side, this
means that the switching voltage has to be supplied between the Source
at the negative supply voltage Vp− and the Gate. For the MOSFET at the
positive supply side, the Source is even not at a constant voltage as the
bridge-voltage Vb of the output switches from VP− to Vp+ . The controller
electronics always operate at a low voltage (5 –10 V) around ground level.
As a consequence of these requirements a floating Gate-drive circuit needs
to be used as shown in Figure 6.74. A floating Gate-drive circuit has its own
power supply, galvanic insulated from the power supply of the amplifier and
the switching commands are transferred by an insulating opto-coupler. A
6.3. Power amplifiers 433

Po
' MOSFET switching branch Vb +12V Linear
Vb +15V Vp+
regulator

Linear Enable
regulator circuit D
Opto-coupler +5V

Po
'
1 4 G
Gate-driver
2 3
MOSFET
S
Vb LF output
Bridge Output lter
™'FET switching branch Vp- +12V Linear
Vp-+15V
regulator

Linear Enable
regulator circuit D
Opto-coupler +5V

Neg'
1 4 G
Gate-driver
2 3
MOSFET
S
Vp-

Figure 6.74: Gate drivers for the switching power MOSFETs need to relate the
Gate voltage to the Source voltage, that is either at the negative supply
voltage Vp− or at the bridge-voltage Vb . These voltages are different
from ground and Vb is even fast switching between both supply voltages
which means that the Gate driving circuits must be insulated from
the control part. This is done by means of opto-couplers.
(Courtesy of ASML)

photo coupler consists of an LED with a photo-transistor, which is a bipolar

transistor where the base is exposed to the light that has to be measured.
The photons that fall on the exposed base will excite electrons that act like
the regular base current in a normal transistor, resulting in a current at the
collector when the transistor is supplied with a suitable voltage. One can
also use a photo-diode for this purpose where the photons excite electrons in
the depletion layer between the N- and P-doted silicon but the sensitivity of
a photo-transistor is higher because of the current amplification ratio which
is often preferred.
The insulation of the power supply is not shown in the figure. In general this
can be realised in the main power supply by using an electric transformer
that is designed with a low capacitive coupling between the windings to
prevent high-frequency currents by the switching bridge-voltage.
The linear regulators are devices that keep their output voltage at a constant
level and reduce interference from other electronics that are connected to
the same power supply.
434 Chapter 6. Analogue electronics in mechatronic systems

6.3.3.7 Charge-pumping

The presented single configuration with two power supplies appears to give
a problem with these energy sources, because the load is connected between
its output and ground. The problem occurs when the amplifier delivers a
low-frequency or DC current to the load and is explained with the help of
the black arrows in Figure 6.72.
When a DC current is delivered to the load, this DC current is superimposed
on the high-frequency AC currents as shown with the red arrows. When this
positive current is larger than the amplitude of the high-frequency current,
the total current in the inductor has become unidirectional in all steps of
the switching process. For step one and four that is no problem because
this positive current is supplied by the positive power supply. But in step
two and three this positive current will be supplied from the negative power
supply to ground. This means that the negative power supply delivers a
negative power. Or in other words, the negative power supply receives power.
This is also imaginable when recollecting the fact that the positive current
equals a flow of negatively charged electrons and in step two and three these
electrons move from ground towards the negative power supply. As a result
of this process, the negative power supply gets more charge.
The energy involved in this process originally came from the positive power
supply in step one where it was stored in the inductor and for that reason
this process is called charge-pumping. With a positive DC output current
of the amplifier, the negative supply should be capable to accept all this
negative charge by delivering this positive current. Like with a linear power
amplifier, most standard power supplies do not have this four-quadrant
capability where both the positive and negative supply voltage can handle a
positive and negative current. Take for instance the power supply that is
made with the diode-bridge rectifier and storage capacitor from Figure 6.31
of the previous section. A positive current from the negative power supply,
that is caused by charge-pumping, will result in a continuous increase
of the negative supply voltage, because the diodes of the rectifier can not
conduct current in the reverse direction. This increase of the negative supply
voltage will continue until damage occurs in the storage capacitor or in other
components.
The same effect but in the opposite direction happens with a negative DC
current, in which case a continuous charge is transported from the negative
to the positive power supply.
When only alternating currents are to be delivered to the load like in audio
6.3. Power amplifiers 435

power amplifiers, the storage capacitor can handle the short excess of charge
in each signal period. Mechatronic systems often require a more extended
period with a DC current component and in that case a solution for this
charge-pumping phenomenon has to be found.
One solution can be to use a power supply with four-quadrant capability.
In that case the resulting energy can for instance be dissipated somewhere
inside the power supply in a resistor, although then a large part of the benefit
of the switched-mode amplifier is lost. The best solution is to transfer the
energy back to the mains supply. This is achieved in special switched-mode
power supplies, that operate with the same principles as a switched-mode
amplifier. By switching currents and voltages over reactive elements and
the application of high-frequency transformers to change the voltage levels,
they are optimised for DC voltages only, with as much as possible a voltage
source output impedance. With additional switches the conversion from the
mains power to the DC output power can be reversed but power supplies
with that capability are more complex than power supplies that work in only
one direction.
Another good example of a power supply with four-quadrant capability is a
rechargeable battery, as long as the maximum charge level is not exceeded.
When the power supply is not able to handle these currents that run contin-
uously in the opposite direction of the voltage, the problem can be solved in
the amplifier itself, by using the H-bridge or dual-ended configuration as is
applied in the example amplifier.

6.3.3.8 Dual-ended configuration

In the dual-ended H-bridge configuration, one terminal of the load is con-

nected to one single-ended amplifier, while the other terminal is connected
to a second single-ended amplifier. The second amplifier has to deliver the
same voltage and current as the first amplifier but with a different sign. This
means that both amplifiers share the same current at half of the required
voltage.
One benefit of this configuration is the possibility to use a lower supply
voltage or even a single supply voltage like in the example amplifier from
Section 6.3.3.1. Using a single power supply reduces the complexity of
the total system as one single-ended output stage is less complex than a
complete second power supply. Human safety is also less affected because
the maximum voltage difference at any location in the amplifier is only one
time the power supply voltage. Furthermore, the possibility to ground the
436 Chapter 6. Analogue electronics in mechatronic systems

Vp+ Vp+

M1 M3
Load

M2 M4
Vp- Vp-

Vp+ Vp-
Step 1: M1 and M3
 Q !"Q

Vp+ Vp+

M1 M3
Load

M2 M4
Vp- Vp-

Vp- Vp+
Step 2: M1 and M3
 <Q !"Q

Figure 6.75: In a dual-ended H-bridge configuration the load is connected between

two single-ended amplifiers that work in counter phase. The magnetic
energy in the inductors, build up in step one is delivered by both power
supplies and gained back in step two.

Source of the power MOSFET at the negative side avoids the need for an
insulated Gate drive circuit for that power MOSFET.
The second advantage of a dual-ended configuration in a switched-mode
amplifier is the absence of charge-pumping. In Figure 6.75 it is shown that,
in the event of a positive DC current to the load at step one, the current
flows from the positive supply voltage to the negative supply voltage. Both
supplies are loaded with the same current and both deliver energy to the
inductors and the load. At step 2 the current flows from the negative supply
to the positive supply, while both supplies regain energy from the stored
magnetic energy of the inductors.
6.3. Power amplifiers 437

Vp+

Filter Load
1 mH
M1


0.33 1 F 8 mH

F

M2
Vp-

Figure 6.76: An inductive load requires an additional RC network to correct the

damping of the LC-filter in a switched-mode amplifier.

In this configuration each power supply has to absorb at maximum the

magnetic energy that it inserted in the same switching cycle. This means
that charge-pumping is fully avoided.
Because of these benefits the dual-ended configuration has become the de-
facto standard in the industry for single-phase amplifiers, that drive only
one load.

6.3.3.9 Output filter

When calculating the damping at the corner-frequency of the applied output

filter in the example amplifier, one could have concluded that a 5 Ω load to
the LC-filter with an inductor of 1 mH and a capacitor of 0.33μ F results in
a quality factor of:

C
Q=R ≈ 0. 1 (6.117)
L

This is a very low value and if this load was purely resistive in reality, a filter
with an inductor of 0.1 mH and a capacitor of 3.3 μF would have been more
appropriate. The load is however an electromagnetic actuator with a series
inductance of 8 mH. This gives an electrical time constant of τe = 1.6 ms and
as a result the actual impedance rises above ≈ 100 Hz with a slope of +1 to
become approximately 430 Ω around the corner-frequency of 8.6 kHz of the
LC-filter. With that almost purely inductive value the LC-filter would have
almost no damping anymore and noise on the input of the amplifier would
be strongly amplified at that frequency.
438 Chapter 6. Analogue electronics in mechatronic systems

To correct this side-effect of the complex load impedance, an additional

RC-network needs to be added that determines the load impedance above
approximately 1 kHz with a resistive value of 47 Ω, giving a Q -factor of
almost one. In principle this resistor will dissipate power at higher frequen-
cies, so it should not be chosen with a too low value. For that reason the
inductor and capacitor values of the LC-filter were not chosen optimal for a
resistor of 5 Ω but for this higher resistive value.
With a piezoelectric actuator this problem is quite different, as this kind of
actuator gives a capacitive load, connected parallel to the capacitor of the
LC-filter. This does not introduce additional stability problems but it also
does not reduce the Q -factor. In that case, next to a parallel RC-network
for damping, it might even be necessary to introduce a series inductor to
the piezoelectric actuator in order to avoid high-frequency currents in the
actuator.
From these short considerations it is clear that also the output filter section
of a switched-mode amplifier is subject to dynamic optimisation in respect
to the application.

6.3.4 Resonant-mode power amplifiers

The most important limitation of any component in a mechatronic system

is determined by its dynamic properties. The power MOSFETs provide
no exception to that rule. In this section a method will be presented, the
resonant-mode output stage, that optimises the performance of a switched-
mode output stage for these dynamic limitations by using the resonance of
the inductor of the LC-filter with an additional capacitor.

Hard-switching

A major problem of the switched-mode power conversion as described above is

the hard-switching transients that occur over the MOSFETs. At the moment
that a MOSFET is switched off, the bridge-voltage changes very fast, forced
by the current in the inductor that needs a voltage step to change. This
voltage-transient causes a current through the parasitic capacitor between
the Drain and the Gate of the switching MOSFET, that keeps the MOSFET
conducting for a short time, resulting in a less steep voltage-transient.
During this short moment, where the bridge-voltage is not equal to one of
the supply voltages, the current through the MOSFET will cause power
dissipation.
6.3. Power amplifiers 439

A second hard-switching effect is related to the conducting diode in step

two and four, when the current flows in the opposite direction through the
MOSFET. Even when the MOSFET is switched on, some current will still
run through the diode. A semiconductor diode needs time to build up a new
depletion layer to become non-conducting when the current reverses. This
reverse-recovery time of the integrated diode in a MOSFET switching power
transistor amounts up to approximately 100 – 500 ns. During this time a
real short-circuit occurs, when the opposing power MOSFET is switched on.
This problem can be alleviated by paying attention to three aspects:

1. Reduction of the current through the intrinsic diodes of the MOSFET.

This can be achieved by adding very fast diodes with a low threshold
voltage parallel to the power MOSFETs. For moderate power levels
the Schottky diodes can serve that purpose. They are different from
normal Silicon diodes as they are based on a combination of a metal and
N-doted silicon, to achieve a diode functionality. Unfortunately these
diodes have a rather high leakage current in the reverse direction and
the reverse voltage is limited to below 100 V. For that reason the use of
Schottky diodes is limited to low voltage amplifiers. Fortunately also
other special switching diodes have been developed for this purpose
based on regular N- and P-material.

2. The MOSFETs should switch on with a little delay.

Preferably the MOSFET is switched on, when its intrinsic diode is
conducting, at least after the moment that the diode over the other
MOSFET switch has stopped conducting.

3. The steepness of the transient slope of the bridge-voltage has to be

reduced.

The second condition was already fulfilled in the 50 % duty-cycle situation

of Figure 6.72 but as soon as the duty-cycle is changed to a value, where the
current in the inductor becomes unidirectional, this condition is not fulfilled
anymore.
The entire problem can be solved by tackling all items simultaneously and
this is accomplished with a resonant version of a power output stage as
shown in Figure 6.77. In this configuration three measures are taken. First
the transient slope of the voltage is controlled by two additional capacitors,
secondly the MOSFET is always switched on at zero current and thirdly
fast parallel diodes are added to transfer the current in the reactive phase.
440 Chapter 6. Analogue electronics in mechatronic systems

+
Vp

C r1
Gate-driver 1
M1 L
Bridge
+
IL Io
IC Vo
C r2 C
Gate-driver 2 -
M2

-
Vp

Figure 6.77: A resonant switching amplifier uses two additional capacitors Cr1 and
C r2 parallel to M1 and M2 that reduce the transient speed in a
non-dissipative way to a sufficiently low level that the dissipa-
tion in the MOSFETs caused by their Drain-Gate capacitors
is largely avoided. The additional parallel diodes are faster
and have a slightly lower threshold voltage than the intrinsic
diodes of the MOSFETs.
(Courtesy of ASML)

6.3.4.1 Switching sequence of the output stage

In Figure 6.78 the basic principle of the control of the resonant-mode output
stage is shown for a single-ended output stage with two supply voltages.
Because the problem originates in the current, this controller is essentially
a current-feedback system. As a direct consequence of this controller, the
amplifier becomes a current source amplifier, like is preferred in a mecha-
tronic positioning system.
The controller continuously monitors the current I L through the inductor
and switches the MOSFETs when I L has reached certain predetermined
reference levels. These levels are always positive for the MOSFET that
switches the positive voltage and negative for the MOSFET that switches
the negative voltage. In case of a positive output current, the positive ref-
erence level is increased and with a negative output current, the negative
reference level is increased. The MOSFETs are switched on again as soon
as their current is reversed at the moment of zero current.
6.3. Power amplifiers 441

Current Controller +

-
Iset +

IL
-
+

Io
- +
IL Ilim
I

Io -
+ Ilim t
Ilim

- Iset
Ilim

Figure 6.78: The controller of a resonant power output stage measures the current
through the inductor. When this current has reached a certain limit,
that is defined by the output current plus an offset, the MOSFETs are
switched off. When the current is zero, the MOSFETs are switched
on. This results in a controlled output current.
(Courtesy of ASML)

The switching behaviour with this controller is shown in Figure 6.79. In this
figure some simplifications are applied. First the two capacitors from the
bridge point to the power supplies are replaced by one capacitor to ground.
This is allowed as power supplies have a voltage source characteristic with
an equivalent impedance to ground of 0 Ω. The second simplification is the
replacement of the capacitor of the output filter by a voltage source. This is
allowed as the voltage over the capacitor will change only very little within
one switching cycle of the high-frequent current, as was demonstrated with
the normal switching output stage in Figure 6.72.
The main difference with the switching cycle of Figure 6.72 is the moment
that the MOSFETs are switched off. This is determined uniquely by a pos-
itive current level for M1 and a negative current level for M2. After the
MOSFETs are switched off, the voltage Vb of the bridge will not change ex-
tremely fast, because of the additional capacitor Cr . The slope is determined
by the current level and the capacitor value. During this slope the energy
in the inductor flows partly into the capacitor and the total behaviour is
determined by the capacitor and the inductor only.
In fact it is this short moment in the natural resonating mode of the LC-
combination that gave the name of “resonant-mode” to this configuration.
442 Chapter 6. Analogue electronics in mechatronic systems

Vp+ Vp
+

Vp Vp
L Vb L Vb
M1 Bridge M1
+
IL + IL +
Vo Vp Cr Vo Vp
Cr
-  - 
Vb IL I lim IL I lim
-
 M2 
M2 I lim I lim
- -
Vp Vp

Step 1: M1 switched on, Step 2: M1 

 <“
!' 

'Q
IL IL
!!
Q

+ +
Vp Vp
Vp
Vb Vp
M1 L L Vb
M1
Vp
IL + IL +
Vo I

I lim Cr Vo Vp
Cr L
- - 

I lim IL I lim
M2 
M2 I lim
- -
Vp Vp
Step 3: M1<“ Step 4: M2 switched on,
IL
!!
   ƒQ !' 
'Q
IL

Vp+ Vp
+

Vp
Vb Vp
L L Vb
M1 Bridge M1

IL + V p IL +
Cr Vo  Cr Vo Vp
IL I lim
- - 
 IL I lim
I lim
M2 M2 
I lim
- -
Vp Vp
Step 5: M2
 <“ Step 6: M2<“
!!
Q
IL !!
   ƒQ
IL

Figure 6.79: In a resonant switching power stage after switching one MOSFET off,
the current is taken over by the capacitor and without any dissipation
the bridge-voltage changes in a controlled way. The MOSFETS are
switched on as soon as the current reverses.
6.3. Power amplifiers 443

IS
+ IC
_

VH 0 IIPP
VH
B

IC IC
+

_ VH CH B
IS
+

Figure 6.80: A Hall sensor and a current sensing module that measures the current
indirectly by measuring its related magnetic field with a Hall sensor.
(courtesy of LEM)

The second difference is that the MOSFETs will only be switched on, when
the current reverses in the branch, consisting of the MOSFET and his par-
allel diode. That switching moment effectively cancels all dynamic issues in
either the diodes and the parasitic capacitors. In principle this alleviates
the need for the additional parallel diodes but imperfections in timing will
always remain so they are retained for additional robustness of the design.
One additional interesting observation can be done with this current-controlled
power stage. Like with a normal switching output stage the duty-cycle of
the switching needs to relate to the output voltage. This is true, because
the output voltage of the LC-filter equals the average of the high-frequency
signal before the filter and this average is determined by the duty-cycle.
As a consequence of the current source behaviour of this output stage, the
duty-cycle becomes a function of the load. In Figure 6.79 the duty-cycle
is 50 % and with a short-circuited output of 0 V, this duty-cycle remains
unchanged, even when the average current is not zero. The situation with a
resistive load was shown in Figure 6.78. In that case both the duty-cycle
and the frequency change.
The detailed description of all properties and signals in such an amplifier
would go far beyond the goal of this book, unfortunately, but it is interesting
to see how the current sensing can be done in such an amplifier.
444 Chapter 6. Analogue electronics in mechatronic systems

6.3.4.2 Lossless current sensing

In very high power positioning systems, where these resonant power am-
plifiers are applied, the voltage levels are very high up to almost 1000 V,
with current levels of several tens of Ampères. The simple inclusion of
a series resistor to measure the current, would become a bit of a hazard.
Furthermore, the three-phase configuration, that will be presented in the
next section, requires the current to be measured at the high voltage output
of the amplifier. For this reason a lossless current sensor is used that works
with magnetism and active control.

The working principle of this current sensor is shown in Figure 6.80. It

consists of a ferromagnetic ring with a small air gap, that is placed around
the wire of which the current has to be measured. A Hall sensor measures
the magnetic field in the air gap, that is generated by the current I p in the
wire.
The Hall sensor is named after the American physicist Edwin Robert Hall
(1855 – 1938), who discovered the Hall effect that determines the working of
this sensor. A current I c running through a conductive plate would normally
run straight from one terminal to the other. The Hall effect is related to
the influence of a magnetic field orthogonal to the surface of the plate as
the Lorentz force on the current drives the electrons to one side of the plate.
This results in an electromotive force that is measured as a voltage between
two terminals, perpendicular to the current terminals.
The voltage of a Hall sensor is approximately proportional to the magnetic
flux density B.
Because of the non-linear behaviour of the Hall sensor, a compensation
circuit is applied, that consists of an auxiliary winding on the ring and an
amplifier with a push-pull output stage. The current I s through the auxiliary
winding is controlled in such a way, that the magnetic field through the
sensor remains zero. Because of the magnetic coupling of the auxiliary
winding and the current conducting wire by the ferromagnetic ring, I s is
proportional to I p .

6.3.5 Three-phase amplifiers

The need for ever higher accelerations and velocities of advanced mecha-
tronic equipment has resulted in linear motors based on the Lorentz prin-
ciple with three-phase electronic commutation, that have a large range of
6.3. Power amplifiers 445

24 V supply 700 V supply

Three phase

 magnitude set point R current R

transformaon
-

Phase
- current S 5 kW

?
 set point S


controller actuator
serial link
(data & control) current T

enable / reset / error front

switching control panel
info

Figure 6.81: Functional diagram of a three-phase power amplifier. The information

from the position controller is transferred with a digital link to the
phase transformation unit, that translates the information in a two
phase (R,S) set point for the amplifiers. The third phase (T) is derived
by subtracting the R and S from one. Switching current amplifiers
deliver the current to the high power actuator.
(Courtesy of ASML)

motion. An example of such an actuator was presented in Section 5.2.5 of

Chapter 5 on electronic commutation.
Generally, actuators in mechatronic positioning systems need to deliver
a force that is only determined by the position controller and not by the
position. It was demonstrated that a position independent force can be
realised, when the three coil segments of a three-phase actuator are driven
by an amplifier with three current outputs of which the magnitude of the
current at a certain force set point changes sinusoidal as function of the
position, in phase with the B force factor of the related coil segment.

6.3.5.1 Concept of three-phase amplifier

The basic scheme of the second amplifier example of this chapter, that
was designed for driving a 3 phase motor with 5 kW power, is shown in
Figure 6.81. The digital controller provides a force setpoint via a serial
data link to the amplifier, while status information is transferred back to
the controller for diagnostic purposes. The force setpoint consists of the
magnitude and the position of the actuator as this is necessary to guarantee
that the currents are in phase with the actual B force value of each actuator
446 Chapter 6. Analogue electronics in mechatronic systems

coil segment.
With this information a mathematical phase transformation is executed that
creates two 120◦ shifted set point signals R and S with a sinusoidal relation to
the position. Because R+S+T=0, the third setpoint signal can be determined
by subtracting R and S from one. The three setpoint values are used as
inputs for a current amplifier where, after digital to analogue conversion,
each phase of the actuator is driven by a single-ended current-controlled
power stage with a single power supply.

6.3.5.2 Three-phase switching power stages

Figure 6.82 shows how the three-phase actuator is connected to the three
high-power resonant-mode switching output stages. In this configuration the
three phases of the actuator coils are connected in a star configuration, with
one central connecting point and the other terminals of the coil sections are
connected to their corresponding phase terminal of the amplifier. Because
of the application of only one power supply, this amplifier becomes a special
version of the dual-ended amplifier from Section 6.3.3.8. The central point
will have a voltage of an almost constant level, equal to approximately half
the supply voltage. This voltage is constant as all currents at this node add
to zero and the first of Kirchhoff’s laws is fulfilled.

Charge-pumping with three-phase amplifiers

One additional advantage of using three-phase amplifiers is, that the same
rule of R+S+T=0 means that at any moment in time a positive low-frequency
current in one phase is compensated by the currents in the other phases.
This means also that the charge-pumping from one phase is compensated
by the charge-pumping of the other phases and as a result the net value of
charge-pumping is zero.
6.3. Power amplifiers 447

Vp+

M3
Phase R Vp+


Actuator
L
L Phase T
M5
Vp+ M4
L 

M1
M6
Phase S

M2

1
JQ8
JQ6
JQ4
JQ2
š!

J
>JQ2
>JQ4
>JQ6
>JQ8 R S T
>

?


Figure 6.82: A three-phase actuator only needs three single-ended output stages.
charge-pumping is avoided because all currents flow between the
positive and negative power supplies, just like with the dual-ended
configuration.

6.3.6 Some last remarks on electronics

The presented overview of power amplifiers is only so far complete, that it

covers the most important effects of the main principles in combination with
their application. For moderate power delivery and situations where power
dissipation is not a big issue, linear amplifiers are preferred, because they
avoid the related phase delay, caused by the output filter. For high power
and situations where four-quadrant operation is necessary, a switched-mode
amplifier is in fact the only solution.
448 Chapter 6. Analogue electronics in mechatronic systems

Although not specifically mentioned anymore, a switched-mode amplifier

has no problems with four-quadrant usage, as the power dissipation, due
to the difference between the output voltage and the supply voltages, is
avoided by the switching process and reactive components. Nevertheless the
kinetic energy of a decelerating fast moving actuator needs to be dumped
somewhere. When that energy is not converted into heat by dissipation
in the amplifier, this has to be done in the power supply. The only way to
achieve true high efficient four-quadrant power delivery to an actuator is by
a combination of a four-quadrant power supply and a switched-mode power
amplifier.
This methodology is one of the factors in the high efficiency of hybrid cars,
where rotary actuators are driven by four-quadrant amplifiers and the
batteries serve as four-quadrant power supplies to recuperate the kinetic
energy from the braking actions. A modern hybrid car is indeed a real full
mechatronic positioning system, although the precision is not extremely
high.
Chapter 7

Optics in mechatronic
systems

From the moment that mechatronics became a real precision engineering

discipline, the application of optics was indispensable for achieving its
ultimate performance. When controlling positioning accuracies in the order
of a micrometre or less, reliable measuring tools are a prerequisite and
only optical sensors can give the required combination of resolution, range
and accuracy. But this is not the only relevant field of optics in respect of
mechatronics. Modern imaging systems like telescopes, wafer lithography
exposure systems and modern photographic equipment all need advanced
mechatronic systems to control the imaging properties in both static and
dynamic circumstances. This means that optics represent both a supporting
technology and an application area for mechatronics.
Within its limited boundary conditions, this chapter will mainly focus on
the understanding of several important terms and properties, related with
the use of optics in mechatronic systems. It should help to enable the
mechatronic designer to integrate optical systems in their designs in close
cooperation with optical specialists. Where deemed necessary, based on our
experience, the material is treated more in depth.
The chapter starts in Section 7.1 with the general properties of light and
light sources. It will introduce the different properties of light sources like
wavelength, coherence and radiance.
Section 7.2 introduces the physical model for reflection and refraction ac-
cording to “Fermat’s principle”.

449
450 Chapter 7. Optics in mechatronic systems

Section 7.3 deals with geometric optics. This is a practical method used to
calculate the imaging properties of an optical system, where the light is
approximated by ideal rays (ray tracing) and only the properties of materials
on reflection and refraction are taken under consideration. This method is
useful for a mechatronic engineer when designing simple optical systems
that might for instance be used in laboratory test setups, made from “off
the shelve” components.
For precision mechatronics the physical optics from Section 7.4 are the most
important. Physical optics deals with the effects that are based on the wave
character of light, including interference and diffraction. These effects limit
the ultimate imaging properties of an optical system and are also used for
position measurement with interferometers and encoders, as presented in
the next chapter.
Section 7.5 gives a short introduction on adaptive optics, where active con-
trol is used to correct imaging errors, enabling near-to-perfect imaging
properties of for instance terrestrial telescopes.
7.1. Properties of light and light sources 451

7.1 Properties of light and light sources

In physics the properties of light are described in two ways, as electromag-

netic waves and as particles. The still mostly applied theory states that
light acts like an electromagnetic wave, as was presented in Chapter 2. This
theory explains the occurrence of interference and most of the observations
that can be done with light.
The more recent theory on quantum optics is based on observations with
extremely low energy levels of the light. The invention of the photomultiplier,
working with a very high voltage, made it possible to detect light at these
very low energy levels, by creating an avalanche of electrons as soon as one
electron is excited by the light. It appeared that below a certain energy
level, light starts to behave like separate particles called photons, each with
a fixed energy that appeared to be equal to:
hc
E = h fp = [J] (7.1)
λ

with:

h = Planck’s constant( ≈ 6.6 · 10−34 ) [Js]

8
c = The speed of light (≈ 3 · 10 ) [m/s]
f p = The frequency of the photon [Hz]
λ = The wavelength of the photon [m]

When filling in some real numbers, the extremely low number of roughly
E = 3.5 · 10−19 J is obtained for 1 photon of visible light (λ ≈ 600 nm). A
low level of one μW (μJ/s) of visible light power would hence be equal to
about 3 · 1012 photons per second. This still immensely high number implies
that it is not always necessary to take this quantum character into account.
Nevertheless, with very sensitive measurements, the noise level is limited
by the photon character and it partly explains why a larger image sensor in
a photographic camera shows a better noise performance.
At several occasions it will appear to be practical to use the photons and
electrons for the explanation of the physical phenomena that are related
with light.
452 Chapter 7. Optics in mechatronic systems

Visible light

400 500 600 700

›
 
12,000 K
5,500 K
2,000 K

0 1000 2000 3000 4000

Wavelength (nm)

Figure 7.1: Thermal radiation spectra of a black body at different temperatures.

The visible light spectrum shows to cover only a small area of the wide
spectrum of this kind of light source, explaining the low efficiency of
incandescent lamps.

7.1.1 Light generation by thermal radiation

Many sources of light exist, ranging from thermal radiation by heated bodies
to secondary emission of photons by electrons that change momentum or
change state within an atom. Thermal radiation is often related to the
visible light spectrum and is caused by the increasingly fast vibrations of
charged particles inside atoms at increasing temperature. These vibrations
of charge create electromagnetic waves, following the Maxwell equations.
Due to this thermal nature of origin, the induced light consists of a wide
range of wavelength values with a peak at a wavelength, which is determined
by the absolute temperature in ◦ K according to “Wien’s displacement law”,
named after the German Physicist Wilhelm Carl Werner Otto Fritz Franz
Wien (1864 – 1928). He stated that the shape of the wavelength distribution
of a black body, defined as having a surface with a frequency independent
emissivity, is independent of the temperature, while the maximum emission
occurs at a wavelength equal to:
b
λ= [m] (7.2)
T
7.1. Properties of light and light sources 453

With b being the Wien’s displacement constant of ≈ 2.9 · 10−3 [mK]. With this
relation a pure black body starts radiating visible red light at a temperature
of about 2000 K. This corresponds with a peak radiation in the wavelength
area of about 1500 nm, so most of the light is still in the invisible infra
red region. At higher temperatures the peak emission will move to shorter
wavelengths, while a black body temperature of 5500K has its peak emission
around 530 nm, which equals the mid of the visible spectrum, like the
daylight as emitted by the sun. A higher temperature will shift the peak to
even shorter wavelengths into the invisible ultraviolet region. Figure 7.1
shows the spectra of these three examples.

7.1.2 Photons by electron energy state variation

The aforementioned spectra from a thermal radiator show an inherently

very wide range of frequencies. Often this is not preferable for technical
applications, because many properties of optical elements are wavelength
dependent. This leads for instance to the separation of sunlight into differ-
ent colours in the raindrops of a rainbow that is caused by the wavelength
dependent refractive index of water.
For that reason monochromatic light sources are often preferred. Fortu-
nately several alternative light sources exist, mostly based on the emittance
of photons by electrons as these change their energy state in a well defined
way. Electrons in an atom can possess different energy levels, also called
excitation states. Normally these electrons are in the “ground state”, the
lowest energy level. When excited by some external energy source, they
move to different higher energy excitation states, depending on the energy
that caused the excitation. These higher excitation states are generally not
stable and the electron will ultimately fall back to a lower and more stable
state or to the ground state, by emitting a photon of which the frequency is
determined by the difference in energy levels of the electron states.
Electrons can be excited by temperature like in plasma gas discharge lamps,
by fast moving other electrons like in fluorescent tubes or by means of other
photons like in a laser.
The gas in a plasma gas discharge lamp, like used in the modern Xenon
headlights of a car, is ionised by a very high voltage. As a result of this
ionisation, part of the electrons will be able to move freely, thereby making
the gas conductive and a current will flow. The corresponding electric power
will heat the gas and it becomes a plasma with many free electrons that
form a relatively low resistance path for a high current to run at a lower
voltage of ≈ 100 V, than the value of ≈ 1000 V that was necessary to start
454 Chapter 7. Optics in mechatronic systems

365 436
405
546
Relave Spectral Power level

579

300 400 500 600 700 800

Wavelength of the light

Figure 7.2: The light power spectrum of a mercury arc gas discharge lamp shows
peaks corresponding with different energy states of the electrons around
the mercury atoms in the plasma.

the process. These plasma gas discharge lamps radiate light at very specific
wavelengths, determined by the gas atoms inside the plasma. An industrial
mercury gas discharge lamp radiates with a spectrum as shown in Figure 7.2
with different narrow peaks of which the peak at 365 nm (I-line), at 405 nm
(H-line) and at 436 nm (G-line) were used in the illumination of the early
versions of wafer steppers.
The well-known fluorescent tube is also a Mercury type gas discharge lamp
that works a bit differently. It does not work with a real plasma but the
electrons are emitted by heated electrodes at the start and move to the
other electrode through vacuum that is slightly doted with mercury atoms
of which the outer electrons will be excited by the moving electrons. The
230 V AC voltage of the mains is not sufficiently high to start the lamp. For
this reason a coil with a large self-inductance is used to both create the high
starting voltage by interrupting the current through the coil by a “starter”
switch and to heat the two electrodes by this starting current, in order to
emit the first electrons that have to start the process. The peaks at 546 and
579 nm in the visible light region of Mercury are not sufficient to give an
acceptable light spectrum for domestic use and most of the light is emitted
in the invisible ultraviolet frequency range. For this reason a fluorescent
7.1. Properties of light and light sources 455

tube is coated at the inside with special fluorescent powder that converts the
UV light into a wider spectrum in the visible light range, increasing both
the efficiency and improving the colour. Although widely used in general
lighting in buildings, these fluorescent tubes are of limited importance as
light sources for precision mechatronic purposes due to their low radiance,
a term that will be defined later and is related to the large surface that
radiates the light with these lamps.

7.1.2.1 Light emitting diodes

Much more interesting for the mechatronic engineering field is the laser,
which is the subject of the next section, but also the Light Emitting Diode
(LED) is shown to be of increasing importance. LEDs are already used
for many years as indicator in electronic equipment and only in the last
few years they became known for illumination purposes. The principle of
working of an LED is also based on electrons that change their excitation
state, but now not within a gas but in a solid material. For that reason
LEDs are also called solid-state lamps.
An LED is a special kind of semiconductor diode, of which the working prin-
ciple was presented in Chapter 6. P and N material are combined causing
either a barrier or a conductive path for the electric current, depending on
the direction of the voltage.
In case of conduction, the electrons recombine with the holes at the transi-
tion zone from n to p material. This recombination implies the filling of an
empty place around an atom, which corresponds to a change of the energy
level. This would result in emission of a photon, were it not that Si or Ge
are indirect band-gap materials, showing no radiation at recombination.
This means that a normal diode will not emit light nor require the power
that corresponds with that light emission. While this is not a problem for
“normal” semiconductor diodes it does not create light. Fortunately other
materials, composed from materials of adjacent valence bands (III and V) of
the periodic system like Ga,As,In,P and N, have a direct band-gap and they
do result in diodes that emit photons with a wavelength depending on the
material, at the cost of additional electric energy. For instance GaAs gives
infra red light, AlGaAs gives red light and GaAsP gives yellow light.
It has long been quite difficult to create shorter wavelengths, because that
requires a wider band gap and not all material combinations are easily
to manufacture. Eventually GaN appeared to be the right combination to
create blue light.
The single colour of an LED and its rather low power of initially only a few
456 Chapter 7. Optics in mechatronic systems

mW has long prevented the application in high power illumination systems,

like domestic lighting. The breakthrough in creating blue LEDs made it
possible to either combine red green and blue LEDs to create white light or
use a phosphorescent material, to convert the short wavelength of the blue
light to longer wavelengths. Much research has been spent on improving
the efficiency and power output of both the LEDs themselves and on the
phosphorescence, resulting in the recently introduced LED lamps for illumi-
nation purposes.
The main advantage of a LED as a light source in engineering applications
is its almost monochromatic spectrum and its relatively small source di-
mensions. This will prove to be important for high resolution imaging as
will be shown later when presenting the term radiance. LEDs are used for
instance in the incremental optical measurement systems (encoders) that
will be presented in Chapter 8.

7.1.2.2 Laser as an ideal light source

The ultimate light source is the laser. Laser is an acronym for Light Amplifi-
cation by Stimulated Emission of Radiation and is based on the phenomenon
that an electron at its excited state can be stimulated to drop to its lower
state by a photon. This photon needs to have the same frequency as the
frequency of the photon that the electron would emit by its change of energy
state. In other words, one photon will create another identical photon by
“triggering” an excited electron to drop to a lower energy state.
Like shown in Figure 7.3 one photon could stimulate two or more electrons
resulting in three or more photons with the same frequency and phase!.
This phase relationship is called coherence.
This physical process would of course come to an end, when all electrons
have reached their lower state E 0 . The process of emitting photons, when
lowering the energy level of an electron, is fortunately reversible. This
means that another light source, not necessarily coherent, like a normal gas
discharge lamp, can be applied to excite the electrons to an energy level E 2
above the excitation state E 1 needed for the stimulated emission.
This higher energy level E 2 is even less stable than E 1 , so the electrons will
drop to E 1 , while sending a photon of a frequency that is not used in the
lasing and can be neglected in the considerations. The additional external
light source for exciting the electrons is called an “Optical Pump”.
It can be concluded that the complete system amplifies the incoming photons
with a gain that depends on the frequency of the photons. This frequency
is not infinitely sharply fixed, but has some tolerance due to uncertainty
7.1. Properties of light and light sources 457

E2
E1
Electron - - Photon

Photon

Photon

E0

Figure 7.3: Light Amplification by Stimulated Emission of Radiation (LASER). An

external source of energy excites the electrons to a higher unstable
state E 2 , where they first drop to a more stable state E 1 until a photon
with a suitable wavelength passes and triggers the electron to fall back
to its original ground state E 0 , under emission of a photon with the
same wavelength as that of the trigger photon.

considerations on the energy levels of the electrons of which the deep treatise
exceeds the scope of this book.
To accommodate this effect, the term “gain bandwidth” is used, which is
quite different from the gain-bandwidth product that was defined for an
operational amplifier. Optics designers often use the term Full-width at
half-maximum (FWHM), the width of the frequency band that covers half
the optical output power, as shown in Figure 7.4. This corresponds with the
well-known -3 dB point on a logarithmic scale of electronic engineers.
The gain bandwidth tells something about the capability of the laser to
amplify photons of a specific frequency within the bandwidth range and it
automatically determines the bandwidth of the emitted light spectrum from
a laser. Other words in literature for this quality are “line width” of a single
frequency laser and “optical bandwidth”.
The bandwidth of a laser is one of its determining factors in the suitability
as a light source for technical applications. For imaging purposes a high
bandwidth can result in chromatic aberrations due to the frequency depen-
dent properties of many lens materials. On the other hand a very small
bandwidth can give problems due to the occurrence of scattered speckle,
caused by interference. This phenomenon is known from the light of a laser
458 Chapter 7. Optics in mechatronic systems

Opcal power (lin)

FWHM

Frequency

Figure 7.4: Definition of the full width at half maximum power gain bandwidth of
a laser light source. The smaller the bandwidth is, the better is the
temporal coherence of the laser.

pointer which shows tiny spots over the illuminated area that change as a
function of displacement and angle of the incident light. These spots could
impair the image of an object that is illuminated by a laser.
An other important property of a laser, partly related to this bandwidth
and the occurrence of speckle, is the coherence. Two kinds of coherence
exist namely spatial coherence and temporal coherence. The spatial coher-
ence relates to the correlation between the electromagnetic field inside the
laser beam at a certain moment in time as function of the location in
the beam. It is a measure for the perfection of the shape of the emitted
wavefront, a term that is explained later. The spatial coherence determines
the divergence of the laser beam and hence the capability to keep a fixed
diameter over a long range.
The temporal coherence of a laser relates to the correlation between the
electromagnetic field on a certain location inside the laser beam as func-
tion of time. It is related to the bandwidth of the light from a laser and
decreases as a function of the distance from the laser source. The correla-
tion between the temporal coherence and the bandwidth is illustrated in
Figure 7.5 where 7 graphs are shown of frequencies with a little difference
between each frequency. In case all frequencies are in phase at the light
source, which means that all values of the field are equally zero at t = 0, then
the coherence disappears at some distance from the light source, at a time
t = t c that relates to the distance with the propagation speed of light. This
time depends on the bandwidth of the source. With a wide bandwidth, the
time is very short and with a single frequency source, this time is very long.
The time that passes, until coherence is lost, is called the coherence time.
Because light propagates at a constant speed in vacuum, this coherence time
7.1. Properties of light and light sources 459

In phase Out of phase

value

t= 0
Time

Figure 7.5: Electromagnetic field waves as function of time for light of different
frequencies, starting all in phase. It shows the loss of phase relationship
after a certain propagation time that corresponds to the distance from
the light source.

directly relates to the more well-known and for technical applications more
useful term coherence length, being the coherence time times the velocity of
light.
In laser interferometry measurement systems as presented in the next chap-
ter, the required coherence length is often several metres. As an example to
get a feel for the required bandwidth of the light, a coherence length of ten
metres implies that the lowest frequency in the bandwidth will need to be
less than one period different from the highest frequency in the bandwidth
over this length of ten metres. One period of for instance 500 nm relates
to 10 m as 5 · 10−8 , so the bandwidth can not be more than 5 · 10−8 times the
frequency.
This kind of precision is not achieved by the electron-photon physics in the
laser only. Lasers, as used for high precision measurement systems, often
consist of a combination of a laser tube with an optical resonance chamber,
according to the Fabry-Perot principle as will be presented in Section 7.4.2.1.
This resonator consists of two parallel mirrors at a very stable distance,
such that a standing wave occurs between the mirrors. An example of a
laser utilising this principle is the Helium Neon (HeNe) laser as shown
in Figure 7.6 which radiates red light at 632.8 nm. This type of laser is
already for many years used for distance measurement purposes, because
the wavelength can be additionally stabilised by means of Iodine gas that
shows a very narrow natural frequency and enables active control of the
460 Chapter 7. Optics in mechatronic systems

Glass envelope
- High Voltage + High Voltage
Output

Reector Laser cavity with Helium Neon Gas High reector

~1% transmission

Figure 7.6: Cross section of a Helium Neon Laser. A high voltage creates a plasma
where the electrons of the gas are continuously excited to their higher
energy state and emit photons of different wavelengths, while falling
back from that state. Lasing takes place induced by photons of only
one wavelength, leading to a maximum emission at 632.8 nm. The
two reflectors create an optical resonator that further enhances the
frequency stability of the laser.

distance between the mirrors. It enables the use of this laser as a reference
for the standard metre. Next to this precision HeNe-laser, a large amount of
different laser types are introduced of which the semiconductor diode-laser
has been instrumental to the optical registration of signals on a disk like
the red laser diode for the CD-player and the blue lase diode for the Blu-Ray
player. These diode-lasers work according to the same principle as an LED
but with additional internal lasing and small mirror facets on the exit points
of the semiconductor material to create a Fabry-Perot resonance chamber.

7.1.3 Useful power from a light source

Besides the frequency and phase of a light source, also the radiated energy,
the size of the source and the direction of radiation are of importance. These
all influence the amount of light that can be captured by an optical system
and transmitted to the area of interest.
The illumination power Pi [W] of the radiated light, sometimes also called
the radiant flux [Φ], is not sufficient to define the usefulness of a light
source. First of all it contains all frequencies. But even when a light source
is monochromatic, the pointing direction, spread and the radiating surface
are undefined. Many terms are introduced to narrow down these undefined
parameters of a light source in relation to the optical system, of which the
following are explained here:

• The radiant emittance and irradiance.

• The radiance.

• The optical throughput or etendue.

7.1. Properties of light and light sources 461

7.1.3.1 Radiant emittance and irradiance

The radiant emittance M from a light source equals the illumination power,
 
radiated by the source, divided by its radiating surface W/m2 .
The radiant emittance closely relates to the term irradiance that was defined
in Chapter 2. The irradiance I r equals the power density of the received,
 
incident light on a defined surface W/m2 . Often the term intensity is
also used for this term but the definition is not very clear as in optics and
radiology the term intensity is also used for the energy flow per unit of solid
angle. For this reason the term irradiance is preferred in optics.
The irradiance determines all observations in optics as it is the only property
of the light that can be measured with a sensor or the human eye. A photo
detector with a given surface converts the light power, being the irradiance
times the surface, into an electric current.
With a given value for the radiant emittance of a source, the irradiance,
received at a surface in its neighbourhood, decreases proportional to the
squared distance from the source. Contrary to the terms from the next part,
the irradiance can be increased again by concentrating light on a spot with
an optical system.

7.1.3.2 Radiance

The size of a light source relates to its spatial coherence and for that reason
a point source, with an infinitely small radiating surface, pointing into the
right direction, appears to be the most ideal in respect to the performance of
an optical system. In that respect, a perfect parallel laser beam fulfils that
demand, as it can be modelled as a point source located at infinity, pointing
in only one direction.
The term radiance is used to describe the extent to which a light source
behaves like such an ideal point source. The radiance is proportional to the
radiant emittance and for that reason it is also often called the brightness
of a light source. As will be shown, the radiance also includes the radiated
direction of the light. With a real point source, the radiance would by
definition be infinite.
In physical terms the radiance L, see Figure 7.7, is defined within the next
expression for the illumination power, radiated from an infinitesimal part
of a light source in a certain direction:

dPi = L( x, y, α, β) dΩ d A cos ϑ [W ] (7.3)

With:
462 Chapter 7. Optics in mechatronic systems

Electrode

Plasma arch
dA

Normal on dA
ϑ



Glass bulb

Electrode

Figure 7.7: Definition of radiance of a light source. As example a plasma gas

discharge lamp is taken, where the plasma arch is approximated as a
light emitting cylinder. (Courtesy of ASML)

• A : The radiating surface of the light source. [m]

• dΩ: The solid cone angle that defines the radiating direction of an
infinitesimal radiating surface element d A . [sr]

• dPi : The illumination power, radiated from the infinitesimal surface

element into its related the solid cone. [W]

• L( x, y, α, β) The radiance of the source at the infinitesimal surface ele-

ment located at coordinate x, y in the direction α, β into its solid cone.
[W/m2 sr]

• ϑ: The angle between the surface normal and the direction of the
radiating solid cone. [rad]

When the radiance is constant over the total surface A and the solid cone
angle is pointing in the direction of the normal on the surface, the expression
of the total illumination power of a light source into a certain direction,
defined by the solid cone angle Ω, can be simplified into:

Pi,t = LA Ω [W ] (7.4)

As previously mentioned, a small surface of the source appears to be better

for the performance of an optical system. This is due to the physical law of
7.1. Properties of light and light sources 463

conservation of radiance stating that the radiance, as received by an

optical system, divided by the refractive index squared, can never increase.
At best the radiance remains the same over the optical path of a system1 .
The importance of this conservation law of radiance is illustrated with an
example from photography. The energy of the light that is needed to activate
one pixel at a digital camera appears to determine the minimum radiance
level of the light at the source when a certain maximum exposure time is
allowed. To understand this, it is first necessary to realise that an image
in an optical system acts as a source of light for the next element, in this
case the photo sensitive pixel in the sensor. That means that the total
power of the image spot at that pixel can be calculated with Equation (7.4).
The dimension A is defined by the imaged object and the solid angle Ω is
defined by the aperture of the photographic lens. The illumination power
is determined by the necessary energy to activate the pixel, divided by the
exposure time. These terms together determine the required radiance of
the source:
Pi,t
L≥ (7.5)
AΩ
Another example to underline the effect of this conservation law is observed,
when imaging a large object like the sun by demagnifying optics like a
burning glass. The image of the sun becomes a very bright spot with an
increased irradiance. On the other hand however, the surface of the image
is much smaller than the surface of the object. Corresponding with that
demagnification, the solid angle of the light at the image is proportionally
larger than the solid angle of the incident light rays, as these are almost
parallel. As a result the radiance is not changed. A third example is based
on the idea that the radiance of a light source might be increased by shield-
ing a part of its surface. That method can be useful in case one needs a
smaller radiating surface, but the radiated illumination power will decrease
proportionally and again no gain in radiance is achieved. It can even be
proven that concentrating the light from all surface elements of a light
source onto a small hole, will not increase the radiance of the resulting spot,
even though the irradiance is high.
1 Normally the refractive index is the same at the source and the image. Only with immersion

optics in lithographic exposure systems, the radiance can be increased at the image because of
the increased refractive index.
464 Chapter 7. Optics in mechatronic systems

7.1.3.3 Etendue

From the previous it can be concluded that the radiance is maximum when
the light is radiated from a very small surface at a very small angle. Unfor-
tunately only an ideal laser is capable of creating such a perfection and in
reality the optical system can only collect a part of the light.
For this reason another quality term is introduced, the optical throughput,
also often named with the French word etendue. The etendue is a measure
for an optical system of its ability to collect as much as possible light from a
source and retain as much as possible of the original radiance.
The full calculation of the etendue requires an integration of the light sent
by all infinitesimal surface elements of the source and received by all in-
finitesimal surface elements of the entry-pupil2 of the optical system. In
the practical case of a light source with a small surface, relative to the
entry-pupil of the optical system, where also both surfaces of the radiating
source and the entry-pupil are essentially parallel (ϑ = 0), the etendue G
can be approximated as:

G = Ω A [m2 sr] (7.6)

With Equation (7.4) this results in the following approximating expression

for the illumination power, captured from the source by an optical system:

Pi,c = LG [W] (7.7)

With a given amount of necessary illumination power and a given radiance

of the source, this means that the etendue should be maximised. Corre-
sponding to the law of conservation of radiance, this means that the etendue
should be maximum directly at the entry-pupil, as it can not be increased
by optical means after the entry-pupil. This effect is illustrated with the
last example from above, about shielding a part of the surface of the source.
This would decrease the etendue, resulting in a decrease of the illumination
power.

2 As will be explained later, the entry-pupil is the cross section of an optical system that

defines which incident light rays are captured.

7.2. Reflection and refraction 465

7.2 Reflection and refraction

To introduce the subject of this section it is necessary to remain aware once

more that physical theory is based on observations that have to fit in a valid
model. One of the first and most striking observations that people have
made on light is that it travels along straight lines as long as the material,
where it works in, has constant properties.
It was the French lawyer! Pierre the Fermat (≈ 1601 - 1665), who postulated
that light follows the path that requires the least time to reach a certain
place. This Fermat’s principle of least time is still used today and it will
be used in this section to explain phenomena like reflection and refraction.
Before entering in these phenomena it must be mentioned that this again is
a strange theory, as one could wonder how light as a series of photons can
know which path to follow to be first in time. Well like Richard Feynman
states in his famous “Feynman lectures on physics”:

But what does it do, how does it find out? Does it smell the nearby
paths and check them against each other? The answer is yes, it
does, in a way.

To understand this a bit better it is necessary to revert to the theory of

light, where a photon can also be represented as a continuous wave and this
wave behaviour gives the photons their “information” about the path and
the obstructions in the path. This is sufficient for the scope of this book to
be able to use it in the next part to describe wavefronts.
Like stated in Chapter 2, reading Richard Feynman’s work is highly recom-
mended for those readers, who would like to find out more about this duality
of light and the related theory that is based on a probability analysis on the
path of photons that in principle can take any path but most probably will
only follow the path of least time.
When presenting phase gratings, this theory will be used to explain mirrors
that reflect in other directions than as presented in the next section.

7.2.1 Reflection and refraction according to the least

time

In Figure 7.8 a light source A is shown that emits photons.

When determining the path of the photons to place B, according to Fermat
they should follow the path of least time. As photons only move in straight
466 Chapter 7. Optics in mechatronic systems

B
O
A

 i r
C D E Mirror

B’

Figure 7.8: The path of the photons, reflecting from the surface of a mirror derived
by the principle of least time, results in equal angles of incidence ϑi and
reflection ϑr .

lines as long as they remain in the same medium, the shortest time cor-
responds with the shortest distance. This shortest distance is a straight
line that can be drawn directly from A to B, but that path is obstructed by
screen O and the photons can only arrive at B via the mirror.
Evidently the path via C is not taking the least time and also via D it is
not optimal yet. To determine the exact right place to reflect, a little trick
is used and a mirrored place B’ is drawn. When drawing a straight line
between A and B’ it is known that this is the shortest path so, when the
photon is reflected at the intersection point E, this is the shortest path to B.
With a little bit of trigonometry it becomes clear that the angle of incidence
ϑi equals the angle of reflection ϑr , which corresponds with what was taught
at high school.
This relation is not too difficult to derive for simple reflection, but the real
value of Fermat’s model is in predicting the path, when the photons go from
one medium to another. For this derivation Figure 7.9 is drawn, where
photons from the same source A located in a low density medium (air) now
have to reach position B located in a high density medium like glass.
First of all it was stated by Fermat that a more dense medium will slow
down the light. The speed of light ( c m ) in any medium compares with the
7.2. Reflection and refraction 467

i r
F
Air n1 C E
Glass n2 D
G t

Figure 7.9: The path of photons refracting at the interface between two different
media, derived by the principle of least time, results in unequal angles
of incidence ϑi and refraction ϑt . The shortest path is not the fastest!
Note that also some light is reflected.

speed of light in vacuum ( c) with a factor equal to its refractive index (n).
c
cm = (7.8)
n
This means that it takes more time to propagate in a high density medium,
than it would take in a low density medium for the same distance. To
accommodate this difference in speed, the term Optical path length (OPL)
is often used, where OPL equals n times the real geometric path length.
In terms of time the method of Fermat is equal to stating that photons
search for the shortest optical path, hence taking the refractive index into
account3 .
The shortest path from A would be a straight line intersecting the interface
between the media at C. This geometric path does however not correspond
with the shortest optical path with the least time, as the part from C to B
would add considerable to the time.
One might say that the photons recognise that it is better to stay a bit longer
in the air and find a more optimal path, the shortest optical path.
When moving the intersection point from E to the right, first the gain
3 One might conclude that photons slow down in a high index medium, but photons always

travel with the same constant speed. The observed delay is caused by interaction with the
electrons in the medium, where the original photons are exchanged for new photons with a
small delay.
468 Chapter 7. Optics in mechatronic systems

in time from shortening the path from the intersection to B outweighs

the loss of time from lengthening the path from A to the intersection. At
a certain moment however this gain becomes zero and the total time will
increase again. This can be simply understood, when considering an extreme
intersection point, fully to the right in the figure.
Somewhere in between, the optimal point is found and as a first assumption
point E is taken. An optimum means that the change of the optical path
length as function of the location on the intersection becomes zero. This
means that the time for following the optimal path, compared with the time
for a nearby path, through for instance intersection point D need to be
almost equal. By drawing two perpendicular lines, one from D to F and
one from E to G, the difference in path length though air, when comparing
intersection point D with E, is approximately the length between F and E
〈EF〉. The difference in path length through the glass is approximately the
length between D and G 〈DG〉. This approximation is valid when distance
between D and E 〈DE〉 is very small. If E is the optimal point then:

OPLair = n 1 〈EF〉 = OPLglass = n 2 〈DG〉 (7.9)

〈EF〉 equals 〈DE〉 times the sine of the angle between points F, D and E.
From trigonometry it follows that this angle is equal to the angle of incidence
ϑi .
Likewise 〈DG〉 equals 〈DE〉 times the sine of the angle between points E,D
and G, being equal to the angle of the transmitted and refracted light ϑt .
From this all follows:

n 1 〈DE〉 sin ϑi = n 2 〈DE〉 sin ϑt (7.10)

After cancelling 〈DE〉, this equation gives the law of Snell, named after the
Dutch mathematician and astronomer Willebrord Snel van Royen(1580 –
1626), also known with the Latin name “Snellius”. When light goes from
a medium with refractive index n1 to a medium with refractive index n2
Snell’s law becomes as follows:

n 1 sin ϑi = n 2 sin ϑt (7.11)

When calculating the refracted angle from a known incident angle, Snell’s
law is written as:
n1
sin ϑt = sin ϑi (7.12)
n2
7.2. Reflection and refraction 469

7.2.1.1 Partial reflection and refraction

The above reasoning from the least time also shows that reversing the
direction of the photons, so going from glass to air will result in the same
shortest path, followed in the reverse direction. In this reversed situation,
the angle of incidence in the original law of Snell becomes the angle of
refraction and vice versa, resulting for the refracted angle as function of the
incident angle:
n2
sin ϑt = sin ϑi (7.13)
n1

Because sin ϑt can never be larger than one, above the critical angle ϑi =
arcsin n 1/n 2 no light will refract any more and total reflection occurs. This
effect is known from light guiding in fibres and flat glass or acrylic plates,
when illuminated from the side.
This effect does of course not occur only at or above the critical angle and
in reality at any optical surface a combination of reflection and refraction
occurs. For reflected light the irradiance ratio to the incident light is given
in the following equation:

I r,r n 1 cos ϑi − n 2 cos ϑt

= (7.14)
I r,i n 1 cos ϑi + n 2 cos ϑt

with n1 respectively n2 equalling the refractive index of the medium of

incidence respectively the other medium.
Without the need to be able to derive this formula it shows for instance
that with very small angles of incidence this ratio will be very small, which
means that most of the light will be transmitted. At larger angles this ratio
will become larger until most of the light is reflected.
The equation can be checked, because at a certain angle 100 % reflection
can be achieved, when going from a higher refractive index to a lower value.
This corresponds with the equation as ϑ t will become 90◦ at the critical angle
ϑ i = arcsin n 1 / n 2 and cos ϑ t will then be zero resulting in one as the outcome
of the equation.
In practical optics mostly a controlled amount of either reflected or refracted
light is aimed for. This is achieved by coatings on the surface that consist
of successive thin layers of materials with a different refractive index that
form an optical resonator, like mentioned with the laser. Such a coating can
be tuned to either reflect or transmit light with a certain wavelength or a
range of wavelengths.
470 Chapter 7. Optics in mechatronic systems

Source

Figure 7.10: Wavefront model of the propagation of light. Concentric spheres rep-
resent photons that simultaneously originated at the source in the
centre of the sphere. The distance between the spheres represents
either the wavelength of the light or a proportional value thereof.

7.2.2 Concept of wavefront

Before applying these rules on reflection and refraction, first the concept of
a wavefront is introduced. When observing a number of photons in vacuum,
all originating from a light source at t = 0, these photons will move with the
same constant speed of light in different directions away from the source.
At any moment in time, they will all be at the same distance from the source
and their position is defined by a sphere with its centre at the source.
The wave equivalent of this reasoning is as follows. At any moment in time,
light from a monochromatic coherent light source in vacuum will be in phase
on any concentric sphere around the source. For that reason these spheres
are called a wavefront, which is a mixing of the motion character of the
photons and the continuity character of the waves.
When working with this thinking concept of a wavefront, it is necessary to
remain aware that it is a very artificial concept as in reality light never is
completely monochromatic nor completely coherent over long ranges, like
was explained with the laser. A coherence length of metres is large, so
light from a star is never coherent. Nevertheless it is a very useful concept,
because it visualises different optical effects.
It is used to illustrate phase relations and the local direction of the light is
always orthogonal to the wavefront. A curved wavefront corresponds with
diverging or converging light rays and a flat wavefront corresponds with
parallel rays of light. This is illustrated in two dimensions in Figure 7.10,
7.2. Reflection and refraction 471

Figure 7.11: Distortion of a flat wavefront caused by the reduced propagation speed
of part of the wave, when passing through a medium with a higher
refractive index.

where the spheres are approximated by circles. The waves originating from
the point source are depicted, as if they have only one frequency with a
wavelength equal to the distance of the solid or dashed lines, while the
solid lines represent the positive maxima and the dashed lines the negative
maxima of the sine wave.
Often the dashed lines are shown only when the phase relationship is im-
portant. When only the direction is investigated, they often are omitted.

7.2.2.1 A wavefront is not real

In reality the spheres would need to have a very small distance, equal to the
wavelength of the light and they would need to move with the speed of light.
This means by definition that a drawing of a wavefront only shows a snapshot
at a distinct moment of a propagating travelling wave. It does by no means
imply that a wavefront drawing represents a standing wave.
When examining interference at a very small scale, like in Section 7.4, the
distance between the spheres is equal to the wavelength. At large scale
optics however, when only the direction of the light matters, often drawings
on wavefronts are used with much larger distances of the lines, where their
distance is only proportional to the real wavelength.
From Figure 7.10 it is clearly observable that the wavefront becomes more
and more flattened out with increasing distance from the source, which
explains the almost parallel rays of light coming from the sun.
Several qualitative statements can be derived with the wavefront model
472 Chapter 7. Optics in mechatronic systems

regarding the effects of changes in the refractive index of the medium where
the waves pass through. In case these media are not homogeneous, the
waves will partly slow down resulting in a distorted wavefront, like shown
in Figure 7.11. In fact the medium in blue acts like a kind of lens and it
illustrates the observed effects, when we look through non homogeneous
gas or fluid, like the air above a heated surface.
In Figure 7.12 the previously described reflection and refraction is shown
with flat wavefronts and a shorter wavelength (lower propagation speed) in
the glass. From the fact that at the connection points both reflected and
refracted waves should have a fixed phase relationship, the law of Snell
can also be derived in a graphical way. On purpose, the term “fixed phase
relationship” is used, instead of “the same phase”, as at reflection from
a less dense medium to a more dense medium a 180◦ phase shift occurs,
comparable with the rigidly connected chain of mass spring systems in
Chapter 2. More details on this aspect can be found in specialised books,
dealing with the Fresnel equations.

Air

Glass

Figure 7.12: Reflection and refraction at the interface between two different me-
dia derived by using wavefronts. A higher refractive index causes a
shorter wavelength that at the interface is in phase with the incom-
ing wavefronts. The reflected wavefront shows 180◦ phase reversal,
because of the reflection from a low density to a high density medium.
7.3. Geometric Optics 473

7.3 Geometric Optics

Geometric optics is an important area of optics used for predicting imaging

properties of optical systems by approximating the light as an independent
set of separate rays and by neglecting effects like interference and diffraction
that relate to the wave character of light. This approximation is only valid,
when the important details of the object to be imaged are essentially larger
than the wavelength of the light, but even in more critical designs geometric
optics can give very useful predictive results.
Optical imaging is based on the two main principles that were introduced
in the previous section, reflection and refraction. In case of refraction, the
transparent optical elements are called lenses and in case of reflection, the
non-transparent optical elements are called mirrors. Reflective systems are
also named catoptric systems and refractive systems dioptric, a term that
relates to the term dioptre for the “strength” of a lens. The dioptre is the
reciprocal of the focal length and mainly used within ophthalmology.
A combination of lenses and mirrors in an imaging system is called a cata-
dioptric system.
As most smaller optics from everyday life consist of refractive elements, this
section will be limited to dioptric optics only. The treatise is in principle
more complicated than with the catoptric counterpart, due to the refraction
involved.
Two essential differences however need to be mentioned. First of all, re-
fraction in general is wavelength dependent and as will be shown, this
dependency introduces chromatic aberrations that do not occur with re-
flection. Secondly an imaging system with only refractive lens elements
is less dependent on small rotations (tilting) of each element around the
axes perpendicular to the optical axis, while any reflective surface is ex-
tremely sensitive for tilting. This property poses severe requirements on
the mounting stability of these reflective elements.

7.3.1 Imaging with refractive lens elements

Parallel rays of light approaching a piece of glass with two parallel planar
surfaces will be refracted and continue their path with the same parallelism
and distance. This is the case, because the refraction at the point of insertion
is compensated by the refraction at the exit point of the glass. This can be
seen in Figure 7.13 that also shows what would happen with a piece of glass
with two curved parallel surfaces with the same curvature. Even in that
474 Chapter 7. Optics in mechatronic systems

A
B

Figure 7.13: Refraction of two pieces of glass with parallel surfaces, one flat and one
curved, shows that only a shift of the lines will occur. The refraction
at point A is reversed at point B
.

case the rays would propagate approximately parallel after the passage and
only their distance will be changed.
In order to achieve an optical image it is necessary to use a piece of glass with
a varying thickness, slowing down the rays depending on their incidence
angle and position.
The invention of a lens is already more than three thousand years old. By
coincidence of nature it is far more easy to achieve a curved surface than
a flat surface with optical quality. When polishing a piece of glass by a
circular motion against a relatively soft counter surface with abrasive paste,
automatically a curved surface will result. The abrasion takes place most
at the outer part of the piece of glass, because the outer part makes a larger
movement in this process and the force exerted near the edges is also larger,
resulting in an essentially spherical surface.
As a first step into the modelling of the imaging properties of a lens, the
refraction at a spherical surface with radius R is shown in Figure 7.14.
First the optical axis is defined as being the axis of rotational symmetry of
the optical system. A ray parallel to the optical axis, entering the surface at
point A, is refracted in the direction of the normal towards the centre of the
surface, according to the law of Snell. A ray drawn at the optical axis would
have an angle of incidence of 90◦ in the Vertex. As a consequence this ray
would continue its path with the same direction and intersect the first ray
at F.
It can be reasoned that all parallel rays at the same distance from the optical
axis as the first ray at A will after refraction cross the optical axis at the
7.3. Geometric Optics 475

ϑi A

ϑr
Opcal axis Vertex F
Focal point
R

Figure 7.14: Refraction at a spherical surface results in converging rays that inter-
sect at the focal point F .

same point F, because of the rotational symmetry and the point F is called
the focal point.
As a an approximation also rays entering closer to the optical axis will bend
towards the same focal point F with only a small error causing the “spherical
aberration” of a lens. This aberration is small, when only those rays are
considered that are very close and almost parallel to the optical axis, the so
called paraxial rays.

7.3.1.1 Sign conventions

It is important to introduce some sign conventions to achieve the right sign

for the image, positive for the same side of the optical axis as the object
and negative for the opposite side. The formal sign convention in imaging
systems starts with the light source or object positioned at the left side of
the system and is a bit counter intuitive, as it does not correspond with a
standard coordinate system. The following is commonly used:

• F o , the distance of first focal point to lens is positive when located at

the left side of the lens.

• F i , the distance of second focal point to the lens is positive when located
at the right side of the lens.

• S o , the distance of object to lens is positive when located at the left

side of the lens.
476 Chapter 7. Optics in mechatronic systems

Biconvex Biconcave Posive Negave Plano- Plano-

meniscus meniscus concave convex

Figure 7.15: Different lenses exist with names that are determined by the curvature
of their two surfaces.

• S i , the distance of image to the lens is positive when located at the

right side of the lens.

• y, the vertical axis is positive above the optical axis.

• R , the radius of the surface of the lens is positive if the centre is located
at the right side of the lens.

As a consequence of these conventions a biconvex lens has for instance one

positive radius (left) and one negative (right).

7.3.1.2 Real lens elements

A real lens consists of two surfaces with different curvatures where the
same law of Snell can be applied.
A wide variety of lenses exist, as shown in Figure 7.15. The different names
are derived from the curvature shape. A rounded surface is either convex,
flat (plano) or concave and all combinations can be applied.
A biconvex lens is called a positive lens as it will bend the rays towards
the optical axis (converging). Consequently a biconcave lens is a negative
(diverging) lens, and any other combination is positive or negative depending
on whether the optical path through the lens is largest near the optical axis
or at the outer part of the lens.
A combination of a convex and a concave surface is called a meniscus because
of the similarity with the shape of the human meniscus and it is the standard
configuration for eye-glasses.
The first example of imaging optics that will be shown in this section, is
the positive biconvex lens from Figure 7.16. This lens refracts the rays in
7.3. Geometric Optics 477

Opcal axis Focal point

R2 R1

f
?
'
d

Virtual focal point Opcal axis

R2
R1

f
™'

Figure 7.16: A positive lens, consisting of two convex surfaces that both converge
the refracted rays to the optical axis, gives a shorter focal length, than
with one convex surface only.
A negative lens, consisting of two concave surfaces that both diverge
the refracted rays, creates a “virtual” focal point at the same side as
the object.

a converging direction towards the optical axis. It can be seen that both
surfaces refract the rays that come from the left in the same direction,
resulting in a focal point closer to the lens than would be the case with one
surface only. With a negative biconcave lens the opposite result is obtained.
In this case, the rays are refracted away from the optical axis and seem to be
originating from a point located at the left side of the lens. This non-existing
point is called a virtual focal-point, because it is not the real origin of the
rays that are observed at the right side of the lens.
The same method of “the least time” by Fermat can be used to derive the
equation for the focal length of a lens but it is also possible to use the law of
Snell directly.
478 Chapter 7. Optics in mechatronic systems

So Si

Object
Fo Fi

Image

f f

Figure 7.17: A positive thin lens gives a real image, when the object is further away
than the focal length f .

Without showing the total derivation, in both cases the following equation
can be derived for the focal length of the lens:

1 1 1 ( n − 1) d
= ( n − 1) − + (7.15)
f R 1 R 2 nR 1 R 2

with:

f = the focal length of the lens [m]

n = the refractive index of the lens material
R 1 = the radius of curvature of the incident lens surface [m]
R 2 = the radius of curvature of the exit lens surface [m]
d = the thickness of the lens measured at the optical axis
(7.16)

With a very thin lens, this equation simplifies into the thin-lens equation
also called the lensmakers equation:

1 1 1
= ( n − 1) − (7.17)
f R1 R2

In Figure 7.17 the imaging properties of a positive lens are shown4 .

The object that is located at a distance S1 from the left side of the lens is
imaged at a distance S2 from the right side. The rays really pass through
the image, which means it is a real image. As will be shown in the next
4 As of this example, the light rays are represented by lines instead of arrows and the light is

assumed always to come in from the left en exit from the right, according to the sign convention.
Further the optical axis is always represented by the centre line like in the previous figures.
7.3. Geometric Optics 479

figure, a real image can only be realised under the condition that S1 is larger
than f . With thin lenses the location of the image can be approximately
determined by using two rays of which the path or trace is followed. This
process is called ray tracing. The first ray is drawn through the middle of
the lens and does not change direction nor position. This is allowed as long
as the lens is very thin. The middle of the lens will then behave like two
parallel surfaces.
The second ray is drawn parallel to the optical axis and will be refracted to
the focal point of the lens. The image is found at the intersection of the first
and second ray.
The relation between S1 and S2 is determined by the focal distance f , fol-
lowing a simple relation that can be derived by trigonometry. This relation
is called the Gaussian lens formula:
1 1 1
+ = (7.18)
S1 S2 f

A simple sanity check proves this formula, as an object at infinity will result
in an image in the focal point and vice versa. It also is imaginable that a
distance of the object from the lens at twice the focal length results in full
symmetry of object and image.

7.3.1.3 Magnification

In the previous example it was shown that the size of the image differs
in most cases from the size of the object, resulting in a magnification M
that depends on the position of the object relative to the focal point of the
lens. This magnification can also be derived by means of straightforward
trigonometry, showing that the magnification is proportional to the ratio
between the distances. As a consequence of the sign conventions that were
defined previously, the magnification M of a lens has to be written as follows:

S2
M=− (7.19)
S1

The minus sign represents the image location at the opposite side of the
optical axis as the location of the object.
As a next step, it is interesting to see what happens when the object comes
closer to the lens. At the image location this will result in an increasing
distance to the right until the object is located at the focal point, resulting
in an image at infinity. When the object approaches the lens even further,
480 Chapter 7. Optics in mechatronic systems

So

Si

Virtual Object
Image Fo Fi

?
' f f

So

Si

Object
Fo Fi
Virtual
Image

™' f f

Figure 7.18: A positive lens will give a virtual image, when the object is closer to
the lens than the focal length. A negative lens always gives a virtual
image, independent of the position of the object.

first the image jumps from +∞ to −∞ and then the image will approach the
lens again, but now from left as a virtual image. This is illustrated in the
upper drawing of Figure 7.18, where the resulting diverging set of rays at
the right side of the lens seem to originate from the virtual image.
The rules for magnification and location of the object and image are still
valid in this situation and because of the negative value of S2 , the sign of
the magnification becomes positive. This corresponds with the position of
the image at the same side of the optical axis as the object. Also the position
of the image can be derived from the Gaussian Lens formula, where S2
becomes negative if S1 < f .
Because of the virtual focal point, it is to be expected that a negative lens
7.3. Geometric Optics 481

will give a virtual image. This is indeed confirmed in the lower drawing of
Figure 7.18, where the image is derived with the same rays, one parallel to
the optical axis and one through the centre of the lens.
It is however not possible to create a real image with a negative lens by
shifting the object closer to the lens. This can be concluded by examining
the two rays in the figure, when the object would shift to the right. Even
past the focal point, the rays will remain diverging at the right side of the
lens, while the virtual image always remains closer to the lens then the
object.

7.3.2 Aberrations

The quality of the imaging by optical systems is determined by many fac-

tors. Aside from the influence of the wave character of the light as will be
presented in Section 7.4, these errors can consist of aberrations and stray
light, leading to a reduction of contrast and sharpness.
Stray light is caused by reflections on edges and the housing of the lens.
Also insufficient polishing of the optical surface causes the occurrence of
stray light, resulting in flare at the image.
All deviations of the trace of each ray in respect to the ideal trace to the
image location are called aberrations. The three most typical aberrations„
the Spherical Aberration, Astigmatism and Coma will be presented first,
followed by the geometric and chromatic aberrations.

7.3.2.1 Spherical aberration

When presenting refraction at spherical surfaces, it was mentioned that

parallel rays at different distances of the optical axis will refract to almost
the same focal point with only a small error. This error is called the
Spherical Aberration as it is caused by the pure spherical surface shape.
Figure 7.19 illustrates this effect with the most extreme example of a positive
lens, a full sphere.
In case of a purely spherical surface, the rays at a larger distance are
refracted stronger than the rays more close to the optical axis. At a larger
distance from the optical axis, the angle of incidence both at the entry
surface and at the exit surface become increasingly large. The exiting
point is closer to the optical axis where the normal to the surface is more
horizontal, because of the refraction towards the optical axis. As a result, a
stronger refraction occurs at the exit point of the lens than would be needed
482 Chapter 7. Optics in mechatronic systems

Figure 7.19: Spherical aberration at a lens with spherical surfaces. Rays at a

different distance from the optical axis have a different focal length.
The pure sphere at the left shows the most extreme example of this
aberration, which makes a sphere unsuitable for normal imaging.
With normal lenses spherical aberration can be solved by means of an
aspheric departure d from the ideal sphere shape.

to reach the same focal point as where a beam closer to the optical axis is
directed. Ultimately, with the most outer rays, the angles become so large
that the angle of incidence at the exit surface becomes larger than the angle
of total reflection and the light can not escape anymore.
Although the pure sphere is an extreme example, all lenses with spherical
surfaces show spherical aberration, because the radius of curvature at a
larger distance is too small to refract the outer rays to the same unique focal
point. This reasoning implies that an increase of that radius as function of
the distance can solve the problem. This deviation of the spherical surface
is called asphericity with a departure value d as shown in the figure.
An aspheric surface can not be produced by the standard production meth-
ods for polished optical surfaces. Computer controlled milling, grinding
and polishing machines are used to create these aspherical surfaces. An
example of such a machine is the magneto-rheological polishing machine of
the company QED shown in Figure 7.20. It uses the property of fluids that
stiffen by a magnetic field to concentrate the polishing force very precisely
on the surface of the lens where material has to be removed. The fluid is
continuously refreshed and filtered, which means that the process parame-
ters are not influenced by wear or floating glass particles.
Due to the high cost level of these machines the departure magnitude is a
7.3. Geometric Optics 483

Spindle
Rotaon

Lens Lens
sweeps
through
MR Fluid Nozzle Sucon
Sheared Pressure
uid distribuon Wheel

Pump Electromagnet Pump

Magnec eld gradient

MR uid condioner

Figure 7.20: Magneto-rheological polishing process used to create aspherical op-

tical surfaces. The abrasive particles are suspended in a magneto-
rheological fluid that stiffens in the magnetic field, applied near the
optical surface to be polished. (courtesy of QED)

significant factor in the total cost of these optics.

7.3.2.2 Astigmatism

A second typical aberration is the astigmatism, as shown in Figure 7.21.

This aberration is observed when rays that propagate in one plane through
the optical axis have a different focal distance than rays propagating in an
orthogonal plane through the optical axis.
This aberration occurs, when the radius of curvature of the lens surface
is different for different planes through the lens. A lens surface having
a combination of a spherical and a cylindrical profile is an example of a
shape that causes spherical aberration. This causes the lens to become non
rotation symmetric. With eye glasses, astigmatism is for this reason also
called a cylinder. It is the most frequently observed aberration of the human
eye.
The example shown in the drawing image of the figure illustrates that
astigmatism can even occur for paraxial rays to an image location on the
optical axis. With non-paraxial rays, the imaging of objects at some distance
of the optical axis is analysed by tracing rays from different points on the
object in two important planes as shown in the lower drawing of the figure.
The first plane, the tangential plane which is also called the meridional
484 Chapter 7. Optics in mechatronic systems

F2
F1

IS


IT



Figure 7.21: Astigmatism is the effect that is observed, when the rays in one plane
have a different focal point than rays in a perpendicular plane. This
is caused by a cylinder shape superimposed on the spherical shape of
the lens. Astigmatism in the image of an object that is not located on
the optical axis is recognised by different focal distance in two planes,
the sagittal plane with image (IS ) and tangential plane with image
(IT ) .

plane, is defined by the point on the object and the optical axis. The second
plane, the sagittal plane, is orthogonal to this plane and also includes the
point on the object that is analysed. This means that the sagittal plane does
not include the optical axis. With a single lens element it intersects the
optical axis at the centre of the lens.
Rays in the tangential plane are called tangential rays and consequently
rays in the sagittal plane are called sagittal rays. Astigmatism occurs when
the sagittal and tangential rays have a different focal point and this can even
occur with a rotation symmetric lens. In that case however the astigmatism
for an object point on the optical axis would be zero.
A lens with a partly cylinder shape causes even astigmatism on the optical
axis as was shown in the upper drawing of Figure 7.21 because of the non
rotation symmetric shape.
7.3. Geometric Optics 485

Image

Figure 7.22: Coma only occurs for non paraxial rays and is caused by a difference
in focal length for rays that enter the lens at different positions. As a
result the image looks like a “comet-tail”.

7.3.2.3 Coma

While spherical aberration and astigmatism can both occur even with parax-
ial rays and objects at the optical axis, other aberrations only occur with
rays that correspond with an image that is not located on the optical axis.
An example of this kind of aberrations is the coma as shown in Figure 7.22.
Coma is caused by the same phenomenon as spherical aberration. A larger
angle of incidence at the entry surface of a lens results in an excessively
refracted beam at the exit surface. This means that the rays that enter the
lower part of the lens, where they are almost perpendicular to the surface,
will be directed to a different focal point at a larger distance than the rays
entering the upper part of the lens.
From the figure it can be seen that in this case also no real sharp focal point
can be found. The image will look like a comet shape, depending of the
position of the chosen image plane. Coma is recognised by a rather sharp
and intense spot with a flare of decreasing intensity but increasing size in
the direction of the optical axis.

7.3.2.4 Geometric and chromatic aberrations

Other aberrations include geometric aberrations like field curvature and

distortion, while chromatic aberration is observed when light with different
wavelengths is imaged.
Field curvature means that a flat subject in the object plane is not imaged
sharply flat at the image location but at a curved focal plane. This is shown
in Figure 7.23. The object and image plane are often called a field plane
to distinguish them from the aperture plane or the pupil plane that will be
486 Chapter 7. Optics in mechatronic systems

Figure 7.23: Field curvature is the effect that the part of the object that is further
away from the optical axis is imaged at another distance from the lens
than the part that is close to the optical axis. Although again caused
by the same mechanism as spherical aberration and coma it is not the
same phenomenon.

Figure 7.24: Barrel and Pincushion distortion are caused by a difference in mag-
nification between parts of the object at a different distance from the
optical axis.

defined later.
Distortion is an error in the magnification of a lens. When the magnification
of the lens is depending on the distance of the object to the optical axis, the
shape of the image of a large subject becomes distorted. Figure 7.24 shows
the two most frequently observed distortions in for instance photo cameras,
the barrel distortion where the magnification decreases for objects further
away from the optical axis and the Pincushion distortion with the opposite
effect. These two examples are only first-order linear types of distortion and
in reality also higher-order distortion effects can occur.
Chromatic aberration is caused by the difference in refractive index of a
transparent material for different wavelengths. This property is called
7.3. Geometric Optics 487

Figure 7.25: Chromatic aberration is caused by the wavelength dependent refrac-

tive index of the lens material. Different wavelengths (colours) will
have a different focal length.

chromatic dispersion because it causes the well-known separation of colours

in for instance a rainbow and a glass prism.
In Figure 7.25 an example with a normal dispersion is shown, where the
refractive index increases with decreasing wavelength. This causes the focal
point for blue light to be closer to the positive lens than the focal point for
red light. Often this aberration also results in a wavelength dependent
magnification, which is visible in photography as purple fringing at the
outer parts of the image.

7.3.3 Combining multiple optical elements

The mentioned aberrations all have impact on the quality of the image
and hence need to be reduced to an acceptable level, depending on the
application.
By introducing an aspheric surface it was demonstrated how spherical
aberration can be avoided and asphericity can also be used to reduce other
errors like coma.
With low cost optics and in case of severe size limitations, like with the
photo camera in a cell phone, this is the only way to solve these problems.
Another example of such a limited optical system is the lens for the optical
pick-up unit of a CD-player. These lenses are often made from plastic,
molded in the desired aspheric shape.
In high performance imaging systems, this aspheric single-lens solution
is not sufficient and a multitude of lens elements, both spherical and as-
pherical, are often combined to create an imaging system that is minimally
488 Chapter 7. Optics in mechatronic systems

Crown Flint

Figure 7.26: By combining a positive lens made of Crown glass and a negative lens
made from Flint glass an achromatic doublet is obtained, where the
chromatic aberration of a single lens element is compensated.

affected by the mentioned aberrations.

The first example of a multiple lens element is a combination of two lens
elements, that are made of materials with different refractive properties, to
correct chromatic aberration. When a positive lens, made of Crown glass is
combined with a negative lens, made of Flint glass it results in the achro-
matic doublet of Figure 7.26.
Crown glass is a material with a very high refractive index and a corre-
sponding high dispersion, while Flint glass has a low refractive index with
a low dispersion. The resulting reduction of the overall dispersion comes at
a price, because the focal length of the combination is larger than the focal
length of the positive lens alone.
Although this combination gives an improvement, more measures are needed
to correct a lens over a wide range of wavelengths. Nevertheless it is a build-
ing block that can be used in combination with more elements in advanced
optical systems.

7.3.3.1 Combining two positive lenses

From the wide variety of combinations with positive and negative lenses,
only the combination of two positive lenses will be explained in this section
as representative example.
Figure 7.27 shows this configuration, where the distance between the positive
lenses is smaller than the sum of their focal lengths.
The determination of the image location is done in two steps. First the
image I1 is determined that would be created by the first lens L1 only.
As explained previously, this is achieved by drawing two rays. The first ray
7.3. Geometric Optics 489

f2 f2

L1 d L2
1
3

O 2 F1,i
I2
F1,o 4 F2,o F2,i
I1

f1 f1

Figure 7.27: Optical system, consisting of two positive lenses positioned close to-
gether. The image is determined by first determining the image by
L1 via ray 1 and 2 and use that intermediate image to construct the
image after L2 by means of an additional ray 3 drawn from I1 through
the centre of L2 combined with ray 4 through both focal points.

is drawn parallel to the optical axis and will be refracted through the focal
point at the image side F1,i . The second ray propagates through the optical
centre of L1 so it is not refracted. At the cross section of these rays, image I1
is drawn in a dashed shape, because in reality the rays will be intercepted
by lens L2 .
With the help of image I1 , the third ray can be drawn, which is the one that
would pass through the optical centre of L2 and image I1 .
As a last step to determine the image, the fourth ray propagates from the
object through the focal point F1,o , resulting in a refracted beam after L1
parallel to the optical axis, that will in its turn be refracted by L2 through
its focal point F2,i .
The image of the combination of the two lenses is found at the intersection
of these last two rays.
The two lenses appear to behave as one single lens and with a bit of trigonom-
etry the following approximative relation for the focal length of the combi-
nation can be derived:
1 1 1 d
= + − (7.20)
f f1 f2 f1 f2
Which for very small values of d relative to f 1 and f 2 becomes:
1 1 1
= + (7.21)
f f1 f2
It is also clear that the image is inverted, like with a single positive lens.
490 Chapter 7. Optics in mechatronic systems

L1 L2
I2
O
F1,o F1,i F2,o
I1 F2,i Ft,i

f1 f1 f2 f2

Figure 7.28: Optical system, consisting of two positive lenses positioned far apart.
The image is also determined by first determining the intermediate
image by L1 . The final image is determined straightforward, because
the intermediate image is positioned between the focal points of both
lenses.

A very different situation is shown in Figure 7.28, where the lenses are
positioned further apart such that their distance is larger than the sum of
their focal lengths.
With this configuration, the image of the object by L1 is positioned between
both lenses and is called the intermediate image. The second image is simply
determined by taking the intermediate image as object for L2 and use the
same rules as with a single lens system.
From the figure it appears that the inverting effect of L1 is reversed by L2 ,
resulting in an image with the same orientation as the object.
By tracing the ray from the object, that was parallel to the optical axis, the
focal point of the combination Ft,i is found. When L2 is positioned closer to
L1 in Figure 7.28, it appears that Ft,i will move more to the right. This can
be concluded from the position of the image when the object is kept at a fixed
distance from L1 . The three rays that converge at I2 will rapidly diverge.
As a result the magnification of the combined two lenses is increased and
the focal point is shifted. This effect indicates that with two positive lens
elements a zoom lens can be created.
A special situation occurs when L2 is shifted so close to L1 that F2,o becomes
located at the same position as I1 . In that situation the second image I2 will
be located at infinity as I1 is at the focal point of L2 .
When L2 is moved even further towards L1 , I1 will enter the area between
F2,o and L2 resulting in a virtual image at the left side of the system, similar
7.3. Geometric Optics 491

Entry-pupil Exit-pupil

Stop

O I

L3
L1 L2

Figure 7.29: An optical system with an aperture stop that limits the outer rays in
the system. The entry-pupil is the image of the aperture stop observed
at the entry of the system, while the exit-pupil equals the image of
the aperture stop as observed at the exit of the optical system.

to the effect that was demonstrated a few pages back in Figure ??. It is true
that at the very moment of crossing the focal point of L1 , the image swept
from positive infinity to negative infinity and will again come closer from
the left, when L2 is moved even further.
Ultimately both focal point will coincide which results in the double-telecentric
lens that will be presented after the introduction of the important aperture
stop and pupil.

7.3.4 Aperture stop and pupil

All optical imaging systems have one important limitation that is caused
by their size. In Section 7.4 on physical optics it will be shown, how the
ultimate imaging properties of an optical system are determined by the
maximum capture angle of the rays that can pass the system. This quality
of capturing as much as possible light is also directly related to the law of
conservation of radiance, as introduced in Section 7.1.3.
A geometrical measure for this dimensional property is the pupil that defines
which rays are captured and transmitted by the optical system and which
rays are blocked.
With a single lens element, this pupil is determined by the diameter of the
element itself. With a more complicated system, consisting of a multitude
of lens elements, the outer rays are in principle limited in a not well defined
manner, depending on the direction of the rays and the position and sizes of
the different elements.
492 Chapter 7. Optics in mechatronic systems

L1 L2
Aperture stop

O Chie
f ra
lo y
F1,o F2,o F2,i
F1,i I
li

f1 f1 f2 f2

Figure 7.30: When combining two positive lenses with coinciding focal points, a
double-telecentric optical system is obtained, where an object at F1,o
will be imaged at F2,i with a magnification equal to the ratio of the
focal lengths. The chief ray through the joint focal centre is parallel to
the optical axis both at the object and at the image site. The position of
the object or the image sensor has no influence on the magnification. A
symmetric cone of light from the object around the chief ray, indicated
by the dashed rays, will also create a symmetric cone of light at the
image.

It is not preferable to determine the pupil by the mountings of the lens

elements by means of metal constructions with either springs, screws and/or
glue, because this would lead to scattering of light in undefined directions.
To solve these issues, preferably a separate element is added that determines
this pupil. This element is called an aperture stop, often just shortly named
“the stop”. Other names are iris diaphragm or just iris.
Figure 7.29 shows a schematic example of an optical system that consists
of three lens elements, where an aperture stop is inserted between lens L2
and L3 . The first observation that can be made is that the most outer rays
that pass through the system are no longer determined by the size of the
lenses but only by the aperture stop. The second observation is that only
the image of the stop by the lenses between the stop and the observer can
be seen from outside. This image is called the entry-pupil at the object side
and the exit-pupil at the image side.
7.3. Geometric Optics 493

7.3.5 Telecentricity

When the focal points of two positive lenses coincide, the configuration
obtains a special property, called double-telecentricity and is explained with
the dual lens-element system of Figure 7.30.
A paraxial ray from O will be refracted by L1 through the joint focal point
and continue its path towards L2 , where it will be refracted parallel to the
optical axis again.
The location of the image will be at F2,i , because all rays of any point of the
object will be refracted into a set of parallel rays in the space between the
lenses. As a consequence they themselves will be refracted to the focal point
of the second lens element L2 .
The aperture stop in a telecentric system is located at the joint focal point.
The ray through the joint focal point is called the chief ray. Officially a chief
ray is defined as any ray under an angle with the optical axis that passes
through the centre of the aperture stop.
With a double-telecentric system a paraxial ray from any point on the object
will be a chief ray and become also a paraxial ray at the image side of
the system. As a consequence, the magnification of the system is only
determined by the ratio of the focal lengths:
li f2
M=− =− (7.22)
lo f1

This means that a displacement of the object in any direction left or right
from the drawn position in Figure 7.30 will not change the size of the image,
as the path of the drawn paraxial ray will not be influenced by this displace-
ment.
It should be noted that even with this constant magnification, the focal
position of the image will change with a displacement of the object along the
optical axis. This means that these kind of lenses are specially chosen when
the requirements on magnification are more severe than the requirements
on focal distance. This is the case with for instance the projection lens of
the wafer stepper that was introduced in Chapter 1. When the wafer with
photo-sensitive resist is placed at the image location with a slight error in
its distance to the lens, this error will only result in a less than perfectly
sharp image but not in a change of the dimensions.
In the described configuration, this magnification is maintained both for a
position change of the image sensor and for a position change of the object,
which makes it a double-telecentric optical system. Also examples with
single-telecentricity exist as will be presented later.
494 Chapter 7. Optics in mechatronic systems

L1 A B L2

2
C
O Chief r
ay
F1,o F2,o F2,i
F1,i
3 I
C

A B
1

Figure 7.31: Different positions of the aperture stop have different effects. Locating
the aperture stop at A instead of the preferred position B at the joint
focal point will result in ray 1 to be blocked at an undefined location.
A smaller aperture stop at C clearly defines the cone of light bounded
by ray 2 and 3.

7.3.5.1 Pupil in a telecentric system

Figure 7.31 shows a double-telecentric lens configuration with the aperture

stop at different locations to illustrate the different effects.
At first sight it seems not important where the aperture stop is exactly
located, because the rays between the lenses are propagating in parallel.
Still it is preferred to position this stop at the coinciding focal points, as
there the impact of the size of the object and image is minimised.
When the stop would be located near lens L1 as shown in the figure at point
A, a ray coming from the top of the object going just past the lower part of
the stop, would end-up outside lens L2 . This means that the surroundings
of L2 would determine another undefined aperture stop for these lower rays.
It can be concluded that the maximum allowable diameter of the stop is
determined both by the size of the lens elements, the size of the object and/or
image and by the distance of the aperture stop to the joint focal point.
It is also shown in the figure that the aperture stop defines a cone of light
around the chief ray. With the double-telecentric lens this cone is always
pointing in a parallel direction to the optical axis.
The image of the stop of a double-telecentric lens is at infinity because of
the position at the joint focal point. In practice this has given cause to the
name pupil plane for the orthogonal plane to the optical axis that is located
at the aperture stop.
7.3. Geometric Optics 495

Also with non-telecentric optics it is preferred that the rays from any spot
on the object propagate parallel at the plane where the aperture stop is
located. This is the case with all optics that need a controllable aperture
stop, like photographic camera lenses. This can be seen from Figure 7.31
where for instance a smaller aperture stop at position C will only influence
the amount of light. As each point on the object will still be imaged by
means of light rays through the smaller aperture the overall image shape is
not altered by shadowing a part of the image. For that reason also with non
telecentric lenses this plane where the aperture is located is called a pupil
plane.
With the physical optics of the following section, this pupil plane will further
prove to be a very important area to determine the quality of an imaging
system.

7.3.5.2 Practical applications and constraints

The shown example with only two positive lenses represents an almost ideal
situation. When a different amount of elements are used like in Figure 7.29,
there is not always a joint focal point defined. In those cases still single-
telecentricity can be created, depending on the location of the aperture stop
inside the lens system.
The first possibility is an object-space telecentric system, where the aperture
stop is located at the first focal point inside the system when entering from
the object side. In that case the entry-pupil is located at infinity.
An example of such an object-space telecentric lens is the measurement
microscope from Chapter 3. A small displacement in the vertical direction
is not allowed to cause an error in the measurement of the object size and
object-space telecentricity prevents this error.
The second possibility is an image-space telecentric system, where the aper-
ture stop is located at the first focal point inside the system when entering
from the image side. This causes the exit-pupil to be located at infinity.
Image-space telecentric lenses are preferred in digital photography cameras.
Non-telecentricity at the image sensor has as consequence, that the cone of
light to the image is not symmetric around a paraxial ray but tilted under
an angle with the optical axis. This can lead to image artifacts, due to the
stack of a colour filter and multi-lens array before the image sensor. Only
an image-space telecentric lens guarantees that for instance no red light
ends up at a blue sensor element.
Due to size constraints, practical photographic lenses can only approximate
496 Chapter 7. Optics in mechatronic systems

telecentricity. Perfect telecentricity would require that at any position on

the image sensor a full symmetric cone of light could be delivered by the
objective. With the shown example of Figure 7.31, this requirement would
imply that L2 would need to be much larger than the image sensor and this
size would certainly not fit in a normal camera with a full frame sensor of
24 × 36 mm. As a result vignetting would occur, a darkening of the outer part
of the image. With a smaller sensor like the APS-C type of 23 × 15 mm, the
telecentricity is more easily to approach without increasing vignetting. The
telecentricity of a camera lens can be checked by estimating the distance of
the exit-pupil when looking at the image side through the lens. It should be
as far away as possible but will in most cases not be located much further
away than half-way the lens.
7.4. Physical Optics 497

7.4 Physical Optics

One of the models to describe the properties of light was shown to be based
on the theory of electromagnetic wave propagation.
In Chapter 2 it was explained that electromagnetic waves consist of a com-
bination of two oscillating orthogonal vector fields, an electric E and a
magnetic B field. In Chapter 5, the interrelation between magnetic and
electric fields was defined by Faraday’s and Ampére’s law of the Maxwell
equations and these laws are used in dedicated modelling software for the
calculation of the real properties of light in optical systems. In this section
the wave character of the light is used to explain the performance of optical
systems without extensive calculations. Instead, graphical representations
will be used of only one of the two interrelated vector fields. The electric
E-field is often chosen for this purpose and is presented graphical as a sine
wave.
An unfortunate fact that is encountered in analysing the physical properties
of optics, is caused by the extreme frequency of the related electromagnetic
waves. Electric or magnetic oscillating fields with frequencies in the order
of 1015 Hz are impossible to measure directly. As mentioned before, only the
average optical irradiance I r can be detected by means of photo detectors
that convert the related flow of photons into an electric current.
The irradiance of an electromagnetic wave is proportional to the power of
the sinusoidal electric field, given by the following expression:
cn0 2
I r = cn0 E2rms = Ê [W/m2 ] (7.23)
2
In the following subsections, first the concept of polarisation is introduced,
because this concept is widely used in optical measurement systems. This
is followed by the explanation of the interference effects of the interaction
between different waves and the theory on diffraction and gratings. The
section will be completed by deriving the diffraction limited resolution of an
optical system, where the influence of the capture angle on the quality of
the image will be demonstrated.

7.4.1 Polarisation

The polarisation of an electromagnetic field is based on its transversal

character and defines in which direction the electric field is oscillating. To
see what this means, the electric field of the light is represented in the 3D
vectorial graph of Figure 7.32. It applies an orthogonal coordinate system,
498 Chapter 7. Optics in mechatronic systems

z
y

z
Figure 7.32: The electric field of linear polarised light, represented in an orthogonal
coordinate system, defined by x in the direction of propagation of the
wave and y, z under 45◦ with the polarisation direction of the electric
field. The projection on the y – z plane is a line. in the upper part
the y and z components are in phase while in the lower part they are
180◦ out of phase. Shifting the phase of one of the components with
180◦ results in a rotation of the polarisation direction of 90◦ on the
y – z plane.

with x being the direction of propagation.

In this example, the field oscillates in one direction and the coordinate
system is chosen such that the oscillation direction is under 45◦ with the y
and z axis.
7.4. Physical Optics 499

The components of the wave in the x – y , y – z and x – z plane are deduced by

simple projection. Projection of the wave in the y – z plane gives a straight
line, for which reason this oscillating electric field is called linear polarised.
The amplitude of each component in the x – y and x – z plane depends on
the polarisation direction in respect to the chosen coordinate system, which
in this example results in an equal amplitude.
The polarisation direction of the light emitted from a light source is de-
termined by its origin and generally normal light sources like a light bulb
or the sun are “randomly polarised”. This means that there is hardly any
phase relationship. In order to create a linearly polarised light source, a po-
lariser can be used, an optical element that only transmits the polarisation
component of the light that corresponds to the orientation of the polariser.
Polarisation is created in nature at reflecting surfaces, because the reflection
of an electromagnetic wave is depending on the polarisation direction rela-
tive to the orientation of the reflecting surface. Light polarised in the plane
of incidence, defined by the normal and the propagation direction, is called
p-polarised light and reflects less than light polarised in the orthogonal
direction, the s-polarised light. This effect is maximum at the Brewster’s
angle ϑB = arctan(n2 / n1 ), which for glass amounts to ≈ 56◦ . For this reason
sunglasses with polarising glass reduce the effect of reflected light from the
sea or other reflecting surfaces that has become polarised by the reflection.
A laser can be made to emit only one polarisation direction by inserting
a quartz plate at the brewster angle inside the resonating chamber. The
stimulated excitation will result in photons that behave in the frequency
domain as having almost the same frequency, phase and polarisation as the
photon that caused the excitation. Especially a well designed Helium Neon
laser can create an almost ideal polarisation that can be used in position
measurement systems as will be presented in the Chapter 8.

7.4.1.1 Birefringence

It is often observed that the propagation speed of light inside a transparent

material depends on the polarisation direction. This optical anisotropy is
called birefringence and is caused by the atomic structure of the material.
Especially crystalline materials show this property. Birefringent materials
can be used to change the polarisation direction of light by inserting a cer-
tain length of this material in the beam. Starting with linear polarised light
the anisotropic directions of this birefringent material need to be orientated
under 45◦ with the polarisation direction of the light according to the coordi-
500 Chapter 7. Optics in mechatronic systems

z
Figure 7.33: Elliptical polarised light. When the y component relative to the z
component of the electric field of an electromagnetic wave is slightly
different from 180 or 0◦ the projection of the electric field in the y – z
plane becomes an ellipse.

nate system from Figure 7.32. As a result both polarisation components of

the wave will no longer have the same phase relationship because one of
them is delayed in respect to the other. What this means is illustrated in
Figure 7.33 where the phase difference between both components is slightly
less than 180◦ . As a result the projection of the combined wave on the y − z
plane will no longer be a straight line but elliptical. For this reason this
electric field is called elliptical polarised. The ellipticity is depending on the
phase difference and in case the phase difference is +/- 90◦ the projection
on the y − z plane becomes a circle, as shown in Figure 7.34. This circular
polarised light will give an equal amplitude of the projection in any direction
independent of the chosen coordinate system. One can visualise the propa-
gation of a circular polarised vector field as a rotating vector v originating
at a line O in the propagation direction pointing to a perfectly symmetrical
spiral around the line of origin. The rotation direction of the spiraling vector
depends on the sign of the phase difference which means that two different
versions of elliptical or circular polarised light exist.
7.4. Physical Optics 501

O
x
v

z
Figure 7.34: Circular polarised light. When the phase relationship between the y
and z component of the electric field of an electromagnetic wave equals
90◦ , the projection of the electric field in the y − z plane becomes a
circle. The spiral can rotate left or right handed depending whether
the phase difference is positive or negative.

Polarisation as such is more complicated than explained in this summary

but this basic understanding is sufficient to model optical systems for en-
gineering purposes. In Chapter 8 the manipulation of the polarisation
direction by using thin plates of birefringent material will be applied to
enable the realisation of very accurate incremental optical measurement
systems.

7.4.2 Interference

In Chapter 2 it was shown that the momentary values of different fields

at a certain location can be added together. When two beams of light
with an equal wavelength are combined, the resulting combination will
show interference effects. These effects are determined by the electric field
amplitude Ê and phase difference ϕ of both beams of light.
For two beams with an equal field amplitude, the effect of the phase dif-
ference is expressed mathematically as follows: When the electric field
magnitude of both beams is given by E1 = Ê sin(ω t) and E2 = Ê sin(ω t + ϕ), the
502 Chapter 7. Optics in mechatronic systems

φ=0
I r,c= 4 I r

φ = 0.5 π
I r,c= 2 I r

φ = 0.75 π
Ir,c= 0.6 I r

φ= π
Ir,c= 0

Figure 7.35: Interference between two waves with the same wavelength and ampli-
tude but different phase. The resulting electric field amplitude can
range between zero and twice the amplitude of each wave giving a
combined irradiance I r,c of maximum four times the irradiance I r of
each wave.

combined electric field becomes:

  & −ϕ '
( ) 2ω t + ϕ
Ec = Ê sin(ω t) + sin(ω t + ϕ) = 2Ê sin cos
2 2
&ϕ' & ϕ'
= 2Ê cos sin ω t + (7.24)
2 2
First of all it can be concluded that the phase of the resulting wave is the
average of the phase of both waves.
Secondly the amplitude becomes twice the value of the amplitude of one
beam, when ϕ = 0, and zero, when ϕ = π or 180◦ . This is shown in Figure 7.35,
where the upper graph shows the situation, when ϕ = 0. This situation where
both amplitudes add to a double amplitude is called constructive interference.
The lower graph shows the effect of destructive interference with ϕ = 180◦ ,
where both amplitudes cancel each other.
7.4. Physical Optics 503

When measuring the resulting electric field, only the irradiance can be
observed. The irradiance can be determined by using the amplitude term of
Equation 7.24 in Equation (7.23):
cn0 2 cn0 & & ϕ ''2 &ϕ'
I r,c = Êc = 2Ê cos = 2 cn0 Ê2 cos2 (7.25)
2 2 2 2
After expanding the squared cosine term with its trigonometric identity,
this equation becomes:

1 + cos ϕ ( )
I r,c = 2 cn0 Ê2 = cn0 Ê2 1 + cos ϕ (7.26)
2

The resulting irradiance of the combined electric field varies between 2 cn0 Ê2 ,
when ϕ = 0 + n · 2π, and zero, when ϕ = π + n · 2π, with the integer n ≥ 1.
The irradiance of the separate fields is equal to:

cn0 Ê2
I r,1 = I r,2 = I r = (7.27)
2
This irradiance level is a factor four below the irradiance level of the com-
bined fields at full constructive interference. This is caused by the squared
relation between the double amplitude of the combined fields and the ir-
radiance. Although this result seems to be contradicting to the law of
conservation of energy, in reality an optical system that shows constructive
interference, also shows an equal amount of destructive interference at
another location. The energy from the destructive interference area will
add to the energy at the other area and the total energy within the system
is not changed. This effect will be mentioned at different occasions in the
following pages.

7.4.2.1 Fabry-Perot interferometer

Interference is frequently used in optical measurement systems to deter-

mine distances and shapes of objects. As an example two beams of light
that originate from the same coherent light source are guided over two
different paths, one fixed and one changing, to one sensor location where
they interfere with each other. The irradiance of the re-combined beams
becomes determined by their path difference, which induces a timing or
phase difference. By measuring the combined irradiance, the change in the
length of the second path can be determined.
This measurement principle is called an interferometer. Many versions of an
interferometer exist and in the next chapter the Michelson interferometer
504 Chapter 7. Optics in mechatronic systems

Mirror 1 Mirror 2

rO O
d
2

100 %
Transmission

Wavelength

Figure 7.36: Fabry-Perot interferometer consisting of two parallel mirrors at a fixed

distance. The upper graph shows the situation where the distance of
the mirrors equals an integer amount of half the wavelength of the light
resulting in a high transmission because of optical resonance. In the
middle graph part the wavelength does not “fit” between the mirrors
so the transmission is zero. the lower graph shows the transmission
as function of the wavelength where the black trace corresponds with
a low finesse and the red trace with a higher finesse.

will be introduced to measure distances. Here the Fabry-Perot interferome-

ter is further investigated as an example as this principle was previously
presented as the optical resonator for increasing the coherence of a laser.
Figure 7.36 shows the principle of a Fabry-Perot interferometer. Two mirrors
are positioned perfectly parallel at a fixed distance. Mirrors have always
some transmission. A full 100 % reflection is not possible so some photons
will pass the first mirror and enter the cavity between the mirrors, where
most of them will be reflected at the second mirror back to the first mirror,
reflected again and so on.
Due to their wave behaviour something special occurs, when the distance d
of the mirrors equals an integer (r ) amount of half the wavelength of the
light ( d = r λ/2). In that case the reflected light at the left entry-mirror will
7.4. Physical Optics 505

show constructive interference with the light just entering the cavity and
as a result a standing wave occurs.
This standing wave manifests itself as if the interferometer becomes fully
transparent for that wavelength as shown in the upper half of the figure.
Other frequencies, that do not fulfil the requirement λ = 2 d /r , will result
in less constructive or even destructive interference and can not pass the
interferometer, depending on the amount of destructive interference. This
system acts like an optical notch filter.
The capability of this filter to distinguish different wavelengths is called
finesse and is determined by the ratio between reflection and transmission
of the mirrors of which the interferometer is made. When the transmission
is large also many other frequencies will pass, because there is hardly any
internal reflection, which would cause destructive interference. This is for
instance the case with glass in a normal window. On the other hand, the
more the mirrors become ideal reflectors the less other frequencies will pass.
A special version of a Fabry-Perot interferometer is the dichroic coating that
is used to prevent reflections on a lens or increase reflections on a mirror.
It consists of many thin layers of material with a different refractive index
that each determine a very small cavity tuned to a specific frequency. By
combining a multitude of layers it is possible to either create an optical
filter with a very high finesse or a more wide-band behaviour like with
photographic lenses.

7.4.3 Diffraction

The famous Dutch scientist Christiaan Huygens (1929 – 1695) postulated

in his work Trait é de la lumi ère, that wave propagation takes place because
every point of incidence of a wave will act as a source of a new spherical
wave with the same frequency and in phase as the original wave.
With a multitude of these kind of sources he could construct the resulting
wavefront of light by connecting the wavefronts of each individual source,
while having by definition the same phase.
A flat wavefront would then consist of an infinite amount of separate wave-
fronts, where only the parallel plane to the original wavefront would satisfy
this phase relationship.
His theory was later rejected for reason of not sufficient validity in all cases
and the particle based theory with photons was found more suitable. Never-
theless the theory of Huygens was not so wrong and it can be used to model
the effect of diffraction of light at a grating in a graphical way.
506 Chapter 7. Optics in mechatronic systems

0th order
-1st order 1st order
-2 nd order 2nd order

ϑ1

λ
p p

Figure 7.37: Diffraction of light at a transmissive amplitude grating. The holes or

slits behave like separate light sources with a fixed phase relationship,
determined by the incoming wavefront. By drawing lines tangential
to the circles with equal phase of all slits, the wavefront of a diffrac-
tion order is found. In the intermediate directions, wavefronts with
an opposite sign result in destructive interference The angles of the
resulting diffraction orders depend on the ratio between the distance
of the holes ( p) and the wavelength of the light. Only for the 2nd order
all tangential lines, connecting the wavefronts of the three holes, are
shown.

7.4.3.1 Amplitude gratings

A first example of a diffraction grating is the transmissive amplitude grating

as shown in Figure 7.37. The name is based on the property of this grating
that light from an area with a high electric field amplitude is combined
with light from a low electric field amplitude. The grating in this example
consists of three very small parallel slits that are defined orthogonal to the
plane of sight in a further completely opaque plate.
A flat wavefront of light with only one wavelength is drawn parallel to the
grating, approaching this grating from below5 .
According to Huygens this incoming light will create separate sources of
5 Always remember that a wavefront drawing is a snapshot of a normally travelling wave

of the electromagnetic field of light. It is used in all examples of diffraction only to show the
phase relationship between the different light waves
7.4. Physical Optics 507

light at the slits with equal amplitude and phase. Orthogonal to the plane
of sight these sources create a line source so the resulting wavefront will be
like a cylinder of which only the cross section is shown in the figure.
The resulting wavefronts are found by connecting the wavefronts of the
different line-sources that are mutually in phase. Several favourable ver-
sions of these tangential lines can be found as is shown in the figure. The
orthogonal directions to each of these tangential lines are directions where
the light from all sources interfere in a constructive way and light will be ob-
served in those directions. The other directions in between show destructive
interference as both light with a positive and a negative phase is combined.
As a result, the diffraction at this grating manifests itself as a series of
separate light beams under an angle ϑ that is determined by the wavelength
of the light and the distance of the slits in the grating. These separate
beams are called diffraction orders with increasing number when starting
from the optical axis. In principle the diffraction is mainly observed at some
distance from the slits, the far field. At a closer distance the amplitude of
the field of each slit at the location of observation will be different and at a
distance in the order of the period of the grating, the light of only one slit
will be observed to be propagating in all directions.
From the figure it can be concluded that the angle between the orders will
be larger with a larger wavelength and/or a shorter periodic distance of the
slits. This relation can be derived by trigonometry which results in:
Nλ
ϑN = arcsin (7.28)
p

where λ is the wavelength of the light, p is the periodicity of the grating

(distance of the centres of the slits) and N is an integer, equal to the order
number. N can be positive and negative, which means that the diffraction
occurs at two sides. It can be noted that the maximum amount of orders
equals the ratio between p and λ as the value of a sine is limited to one.
When using light of different wavelengths, diffraction would result in spatial
separation of the different wavelengths, giving the colouring effects with
visible light as for instance can be observed when looking to a CD-disk.
It is also true, that a larger irregular grating that consists of a number
of sub-gratings with different periodic distances would result in spatial
separation of the diffracted parts that belong to the different sub-gratings,
even with one wavelength of light. The shorter periodic distances will have
a larger diffraction angle than larger distances. A shorter periodic distance
corresponds with a higher spatial frequency of the grating lines. The fact
508 Chapter 7. Optics in mechatronic systems

that diffraction distinguishes areas with different spatial frequencies will

be used later when analysing the imaging quality of an optical system.
The optical effect of the transmissive amplitude grating with transparent
slits in an opaque plate is also observed with a reflective amplitude grating.
In that case the light approaches the grating from above, while the slits
are replaced by reflecting stripes on a non reflecting surface. The reflective
stripes will each act as a source of light like the slits of a transmission grating
and the resulting diffraction orders can be determined in an identical way.

7.4.3.2 Phase gratings

Amplitude gratings have one disadvantage, as they reduce the total amount
of light by the opaque or absorbing part of the surface. This property limits
the maximum reflectivity or transmissivity to approximately 50 %.
One might decrease the loss of light by making the slit wider than the
opaque part, but then the interference will be less. Ultimately the object
becomes just transparent or reflective, which was not the aim of the grating.
Fortunately the 50 % loss of light can be avoided by using a phase grating.
In its reflective form, this kind of grating consists of a row of connected
reflective stripes alternating at two different heights.

Reflective phase grating

A phase grating is a bit more complicated to model than an amplitude

grating. The main reason for the complexity is the fact that the height
in relation to the wavelength determines the irradiance ratio between the
different orders. The principle will be explained in two ways. First the same
wavefront and phase relations are used as with the amplitude gratings. The
second method uses the complete wavefront over the entire surface of the
grating.
The example configuration for the explanation, as shown in Figure 7.38, is a
special version of a phase grating, where the height difference of the steps in
the grating equals exactly a quarter of the wavelength of the perpendicular
incoming light. In that case, the orthogonally reflected light coming from
a high area will have a 180◦ phase difference to the reflected light from a
low area, as the latter has to travel two times the height of the grating.
This optical path difference results in a phase difference of λ/2 between the
light reflected by both areas. When these areas have an equal surface, the
resulting light at the far field in the 0th order direction will be cancelled out
7.4. Physical Optics 509

-1st order 0th order 1st order

Incoming light

0.25λ p ϑ1

Figure 7.38: Diffraction at a reflective phase grating, illuminated perpendicular

to the surface. Light, reflecting straight back from the bottom of the
grating, has to pass two times λ/4 more distance than light reflecting
from the top. These two will then have λ/2 = 180◦ phase relationship,
so destructive interference occurs at the far field and no light will be
reflected in the 0th order direction. Under an angle the path differences
cancel out, so constructive interference occurs, as can be observed from
the tangential lines of equal phase.

by destructive interference.
Knowing that energy can not get lost and no dissipation takes place, the
light should show an equal constructive interference in another direction
and indeed, at a certain angle, the travelled distances will be equal again
for all light coming in orthogonal and parting in that diffraction direction
as can be seen with the tangential lines of equal phase in the figure.
Identical to the previously explained transmission grating this diffraction
angle is equal to Equation (7.28):
Nλ
ϑN = arcsin (7.29)
p

This means that also with a phase grating higher order diffraction angles
occur, but in practice the geometry of the grating is often chosen such that
these higher orders are limited to N = ±1 or only have a very low irradiance.
Although only the effect of one point in the middle of each surface is shown,
this point is representative for the entire grating as for every point on the
low area a corresponding point at the high area can be found that combines
in the same way. By integration the total effect can be determined and
510 Chapter 7. Optics in mechatronic systems

B
O
A

1 2 3

? Mirror

Figure 7.39: With a reflective phase grating, light can be reflected by a different
location than determined by the law of reflection, as long as the phase
is equal for the different paths (1,2,3) of the grating.

with the second method to explain the working principle of the transmissive
version of a phase grating, the total surface is used.
Because of the fact that with a phase grating no absorbing surface is present,
the total amount of light is retained.

Violating the law of reflection?

Several readers might wonder if reflective phase gratings can work at all
with a real mirroring surface. With real measurements on these gratings
it indeed appears that the reflected light will act different than would be
expected by applying the law of reflection that was derived in Section 7.2.
Figure 7.39 shows the reflection at a mirror and a grating. At first sight only
the middle part of a large mirror is needed for reflection and the photons
emitted by the source in other directions are not useful and can never reach
the point of observation.
With a grating surface however, this is no longer true as the photons can
be reflected in the diffraction orders even with a mirrored surface. This
is due to the fact that the simple wave theory is not sufficient to explain
everything. Although the method of least time is useful, another method is
needed to explain what really happens at a mirrored grating. Again, like
also mentioned in Section 7.2, the theory of light acting as particles helps
for this explanation.
In a nutshell this theory states that photons in principle can follow all kind
of different paths to a target point of observation. The probability of a
certain path to this location is however determined by the question whether
other photons from the same source, that arrive at that location via different
7.4. Physical Optics 511

0th order 1st order Air

”!' š!' n=1
interference interference

n = 1.5 p Glass

Figure 7.40: Diffraction at a transmission phase grating, illuminated from below

perpendicular to the surface. The step size is such that in air (n = 1) it
equals one wavelength and in the high index material like glass with
n = 1.5 it equals 1.5 times the wavelength. In the 0th order direction
this results in two wavefronts with 180◦ phase difference that cancel
each other out while simultaneously the 1st order direction does not
show any phase difference.

paths, all have an equal time relation to their temporal periodicity (read,
the same phase). For a reflective grating that means that a photon that
is reflected at one high step should have the same phase at the point of
observation as any other photon from the same source reflected by another
high area.
This again is an example of a theory that as such is not understandable,
because how can a photon know anything about its colleagues, before it is
arrived at the point of observation?.
Nevertheless the theory is useful as it perfectly predicts their behaviour.

Transmission phase grating

A transmissive version of the phase grating can be realised in a transparent

plate by a similar surface profile as shown with the reflective phase grating.
In this case however the height of the steps is different from the reflective
phase grating, in order to obtain the same effect.
When no light in the 1st order is allowed, a larger step size is needed in order
512 Chapter 7. Optics in mechatronic systems

to create the necessary λ/2 phase difference between the photons that pass
through the area with a shorter optical path length and the photons that
pass through the area with a longer optical path length. This means that
the height is depending on the refractive index of the transparent material.
Figure 7.40 shows such a grating and its principle is explained by using the
second method with the full surface. With this method, the transmitted
wavefronts are drawn as if they remain flat, instead of a point source with a
circular wavefront. This is the extreme other side of the approximation and
shows to give the same result. In this example, the step size is chosen such
that its height corresponds with one wavelength in air and 1.5 wavelength
in the high index material, like glass with n = 1.5. When observing the
0th order the wavefronts from the high and low areas have a 180◦ phase
difference and at the far field these wavefronts will cancel each other out
by destructive interference. This is fully comparable with the effect of the
reflective phase grating.

Phase shifting

The transmission phase grating is more easy to visualise in a drawing than

the reflective phase grating, because in transmission the incoming light
and the diffracted light are at different sides of the grating. For that reason
the important phase behaviour of a phase grating at different step sizes
is better explained with transparent phase gratings, while the following
reasoning is also valid for reflective phase gratings.
It was shown in Equation (7.24) that the phase of two interfering waves with
equal magnitude will equal the average of the phase of the two waves. For
the phase grating of Figure 7.40 this means that the phase of the resulting
wave of the 0th order would be 90◦ different from each interfering wave, even
though the amplitude would be zero. The phase of the 1st order wave would
be in phase with both interfering waves.
It is even more interesting to see what happens, when the step size is
different from the above example. This is shown in Figure 7.41. The upper
graph shows the previous situation where the 0th order is cancelled with
90◦ phase difference and the 1st order is in phase with both waves.
7.4. Physical Optics 513

Average wavefront Average wavefront

0th order 1st order Air
n=1
0.25·λ = 90o
{

n = 1.5 Glass
a: Stepsize 1·λ in air, 1.5·λ in glass.
Air
?
'
$ n=1
< 90 o

n = 1.5 Glass
b: Stepsize 0.66·λ in air, , 1·λ in glass.

?
'
$ Air
n=1

n = 1.5 Glass
c: Stepsize 0.33·λ in air, 0.5·λ in glass.

Figure 7.41: By varying the step size of the phase grating, the magnitude and
phase of the 0th and 1st order is also varied. A reduced step height (b:
and c:) results in a phase advancement of both orders. The irradiance
of the orders shifts from 100 % in the 1st order at a: to ultimately
100 % in the 0th order, when the step size would be zero.
514 Chapter 7. Optics in mechatronic systems

When the step size is changed, as shown in Figure 7.41 b: and c:, the phase of
the wave through the high area is less delayed in respect to the wave through
the low area. As a result both orders get a partial positive phase shift.
With a smaller step size, the phase difference between the interfering waves
in the 0th order direction will be less than 180◦ . Consequently the magnitude
of the irradiance of the resulting wave will no longer be zero, while the phase
delay will be less then 90◦ . At the same time also the waves in the 1st order
direction will be less perfectly in phase, resulting in a phase difference of the
resulting wave and a smaller magnitude of the irradiation. Ultimately when
the step size is zero the phase grating will be just a transparent plate again,
only transmitting light in the same direction as the incoming light. With
the reflective version of the phase grating this ultimate situation would
simply be a flat mirror.
As a result of this reasoning, the phase of the 0th order will always remain
delayed relative to the phase of the 1st order diffracted wave. While the delay
of 90◦ from the 0th order relative to the incoming light gradually reduces to
zero, the phase of the 1st order will gradually shift to a positive 90◦ difference
with the incoming light. Regarding the magnitude, following the law of
conservation of energy, the sum of the irradiance of the different orders
always remains equal to the irradiance of the incoming light as the size of
the beams is not affected by the grating. By choosing the right height it is
for instance possible to create a grating with the same irradiance for the 0th
and the two 1st order beams. In that situation the phase shift of both orders
should be equal. A small approximating calculation shows the effect using
Equation (7.26) in a simplified form:

1 + cos ϕ
I r,o = I r,i (7.30)
2

with I r,o equaling the irradiance of an output order and I r,i being the irra-
diance of the incoming light. When the light going into higher orders is
estimated to be about 25 % of the total irradiance, which is a realistic figure,
then 25 % is available for each of the three main orders. This level requires
the cosine term to be equal to 0.25 for all three orders, resulting in a phase
angle of:

ϕ = arccos(2 · 0.25 − 1) ≈ 120◦ (7.31)

This means for the 0th order that the phase difference between the interfering
waves equals 120◦ and as a consequence the (negative!) phase shift of the
0th order after interference becomes half of that value equaling −60◦ relative
to phase of the incoming light. For the 1st order the phase is shifted in the
7.4. Physical Optics 515

positive direction with the same value of 60◦ relative to 0◦ .

The phase difference of 120◦ in this special phase grating proves to be very
useful in the advanced interferometry encoders that will be presented in
Chapter 8.

7.4.3.3 Direction of the incoming light

The shown examples all had a strictly defined direction of the incoming
beam of light, orthogonal to the surface and one may wonder what happens,
when the light approaches the grating from another direction.
In principle, with a reflective grating, a beam with an angle of incidence ϑ
results in a 0th order directed with a same angle ϑ opposite to the normal
fully according to the laws of reflection. In fact as a first-order approxima-
tion the geometric effect just linearly combines with the diffraction effect.
In reality a large angle of incidence of the incoming light will cancel out
certain diffraction orders that would otherwise be directed into the reflecting
material while other orders appear at the opposite side of the normal. The
interested reader can derive this when drawing the different graphs for
other light directions.
One consequence of this reasoning is that the effect works in two directions.
With for instance a reflective amplitude grating, light that enters under an
angle corresponding with the 1st order will be reflected mainly in the direc-
tion of the 0th order. This reverse effect can also be explained by reasoning
that the optical path lengths for photons going in the opposite direction are
equal, resulting in equal travelling time and an equal phase for their wave
property.
With transmission versions of the different gratings in principle the same
reasoning is valid al long as the transparent part is planar, without optical
strength or wedge form. Otherwise the angle should be corrected for these
effects.

7.4.4 Imaging quality based on diffraction

In a purely physical manner, an optical object can be approximated as an

accumulation of spatially separated area’s with a different irradiance. This
irradiance can either originate from local light sources like is the case with
a TV-screen or by transmitted or reflected light from a separate source.
It was shown in the previous section on physical optics that separate illu-
minated lines at equal distance with dark lines in between behave like a
516 Chapter 7. Optics in mechatronic systems

3rd Pupil plane

Irradiance 2nd Irradiance

1st

p ϑ 0th order

Illuminaon -1st

-2nd

-3rd
Object Image

Figure 7.42: Imaging of a transparent grating with a square wave irradiance profile
illuminated perpendicular to the grating from the left. It shows that
the higher diffraction orders that represent the spatial high frequency
content of the image are located at the pupil plane further away from
the optical axis.

grating and create separate beams of light under an angle depending on

the wavelength of the light and the distance of the lines. A small distance
between the lines is equivalent to a high density of the lines. This density is
expressed in the spatial frequency f of the lines, defined as a number of lines
per unit of length (1/m). Similar to electrical signals, the spatial frequency
distribution of an optical object can be analysed by means of a Fourier series
expansion. Application of the Fourier series expansion to a periodic optical
signal with a spatial frequency f results in its sinusoidal harmonics, each
with a spatial frequency n · f . Figure 7.42 shows what this means, when
imaging an object that consists of opaque and transparent stripes at equal
distance. It is a transparent amplitude grating that is illuminated from the
left perpendicular to the grating.
For this example a double-telecentric imaging system is taken for reason of
simplicity but the following reasoning is applicable for all imaging systems.
Starting at the left side, diffraction orders are generated at the grating
that are captured by the optical system. In the pupil plane these orders
all cause parallel beams of light, the higher order beam further away from
the optical axis than the lower order beam. After the second lens these
beams recombine to the image. This image has the same shape as the
object at a different size, because of the demagnification in this example.
The higher irradiance at the image is caused by the law of conservation of
energy, smaller details will show a higher irradiance.
With diffraction theory it was shown that the angle ϑ between the diffrac-
tion orders equals arcsin nλ/d which means that a higher spatial frequency
7.4. Physical Optics 517

Pupil plane
3rd
Irradiance 2nd Irradiance

1st

0th order

Illuminaon -1st

-2nd

-3rd
Object Image

Figure 7.43: Imaging of a transparent grating with a square wave irradiance profile
with a small aperture stop at the pupil plane showing the spatial low
pass filtering of an optical system when blocking the higher diffrac-
tion orders.

(smaller d ) should correspond with a larger angle. This is exactly what

is observed when the image side is compared with the object side of the
system.
What is not shown, because the figure is erroneous to that respect, is the
effect of the missing higher orders that are not captured by the lens. With
a real optical system a decrease in steepness and amplitude of the irradi-
ance profile of the image would be observed. This effect is illustrated in
Figure 7.43 where a smaller aperture is inserted at the pupil plane in the
optical system, cutting off all but the 0th and 1st order. The diffraction orders
created by the square wave irradiance profile each represent a different
harmonic of the spatial frequency. The 0th order represents the average
light irradiance so the DC value of the image, a 0 from the Fourier series
expansion. The 1st order represents the sinusoidal spatial frequency of the
image and the higher orders represent the higher frequency components.
Because optical signals are a bit different from electrical signals it is useful
to highlight some differences. First of all, negative signal values do not exist
for the irradiance as being the magnitude of the electric field squared. This
means that when there is an image, there is always an average irradiance
value and consequently the 0th order is never zero. Secondly the higher
optical orders correspond with the odd harmonics from the Fourier series
expansion as a square wave consists only of the odd harmonics. In Chapter 2
it was shown that this is true for all signals with a certain symmetry.
The shown examples had a regular square wave pattern but in reality objects
never consist of only one spatial frequency. From Fourier analysis it is also
518 Chapter 7. Optics in mechatronic systems

0.8

0.6

0.4

0.2

0
1
0.5 1
0 0.5
0
-0.5
-0.5
-1 -1

Figure 7.44: Airy-disk irradiance profile of the image of an ideal point-source cre-
ated by a diffraction limited optical system. The fringes around the
centre peak are caused by the sharp filtering of the optical system
which is equivalent to a higher order dynamic system.

known that a random signal can be approximated by an infinite amount of

sinusoidal frequency components.
This leads to the following important statements for optical objects in the
spatial frequency domain:

• Any optical object or image can be represented by an infinite amount

of spatial frequencies in three dimensions.

• The high spatial frequencies represent the details of the object and
they are “encoded” in the area in the pupil plane that is the most
distant from the optical axis.

• High spatial frequencies require a large capture angle of the optical

system.

In other words, an optical imaging system acts like a spatial low pass filter
with a cut off frequency that is determined by the capture angle. A small
capture angle results in a lower optical resolution and less sharp images.
An optical system of which the performance is only determined by this
diffraction related limitation is called diffraction limited. In a diffraction
limited optical system all aberrations are smaller than the smallest details
in the image. An example of a diffraction limited image is the airy-disk of
7.4. Physical Optics 519

Figure 7.44 which is an image of an ideal point-source. The point source with
its infinitely small size and infinitely high radiance is the spatial equivalent
of an impulse, having an infinite amount of frequencies. In the airy-disk the
higher order response of the spatial low pass filtering by the optical system
is visible by the shown oscillations around the irradiance peak in the centre,
a spatial “dynamic” phenomenon. As soon as the performance of an optical
system is not ideally diffraction limited, the airy-disk of the image of a point
source will show both a wider and lower middle irradiance peak. The ratio
between the peak height of a real optical system and the peak height of a
diffraction limited version of this optical system is called the Strehl ratio,
named after the German physicist Karl Strehl (1864 – 1940). The Strehl
ratio is a measure for the quality of an optical system.

7.4.4.1 Numerical aperture and f-number

Two closely related terms are used in optics to relate the above presented
capture angle with the resolution of the imaging system, the Numerical
Aperture (NA) in professional imaging systems and the f-number 〈 f /#〉 in
photography. With the help of Figure 7.45 both terms will be explained.
The f-number is the simplest term of the two as it does not take the refractive
index into account and equals just the ratio between the diameter of the
entry-pupil and the focal length.
f
〈 f /#〉 = (7.32)
d
This term is perfectly usable in photographic systems that work in air with
objects that are mostly far away and with an image location near the focal
plane. The maximum entry-pupil of a photographic lens is almost equal
to the diameter of the front lens. Usually the units, also called stops differ

a factor 2 because the captured light from the object is a function of the
surface area of entry-pupil, being squared proportional to its diameter. This
means that a stop is a factor two in light irradiance.
Regular photography lenses have an f-number between f /2.8 and f /22. Often
a practical lens design is optimised for aberrations at an f-number around
f /8. This already indicates that these lenses are generally not diffraction
limited with the exception of very expensive professional photographic lenses
that can be used wide open and sometimes even show f-numbers up to f /1. It
is obvious that these lenses need to be vary large to achieve such an opening
angle, like those used by sports photographers around football stadiums. It
also points convincingly to the inherent limitations in resolution of the small
520 Chapter 7. Optics in mechatronic systems

f
Entry-pupil

d 

Figure 7.45: Definition of the Numerical Aperture (NA= n sin ϑ) and the f-number
( 〈 f /#〉 = f /d ) in an optical system. A large NA or a low 〈 f /#〉 correspond
to more captured diffraction orders giving a better resolution. For the
illustration the entry-pupil as used for the f-number with photographic
lenses is drawn just before the lens. With a telecentric lens the pupil
is located at infinity for which reason the capture angle is taken as
reference for the NA.

lenses in pocket-size cameras and cell phones. Contrary to the photographic

application, professional imaging systems like wafer scanners, that are used
for defining the structures in lithography for IC manufacturing, often have
both the object and the image close by and show significantly higher capture
angles. In this case the Numerical Aperture (NA) is a more suitable quality
value and it is defined as follows:

NA = n sin ϑ (7.33)

This term includes the refractive index n which is related to the fact that the
wavelength of light with a specific frequency is shorter at a higher refractive
index. As a consequence the angle between the diffraction orders is smaller.
In other words an optical system can capture more orders in a high refractive
index medium like water or oil than in air.
This is the reason for oil-immersion in microscopes and water-immersion in
wafer scanners6
The use of the sine instead of the tangent as with the f-number is more
practical as it clearly limits the value in air to one as the unattainable
6 The application of immersion fluids in an existing air-based optical design does not au-

tomatically increase the NA but only enables the designer to achieve a higher NA with an
adapted design that could otherwise not be realised in air.
7.4. Physical Optics 521

goal. With an NA in air that is equal to one, the object or image should be
positioned inside the lens element which is not possible. With small angles
and an refractive index of n = 1 the relation between f-number and the NA
can be approximated as follows:
d 1 1
NA = n sin ϑ = n sin arctan ≈ or 〈 f /#〉 ≈ (7.34)
2f 2〈 f /#〉 2NA

The numerical aperture determines the maximum resolution of a lens, also

called Critical Dimension (CD)7 , with the following relation:
λ
CD = k 1 (7.35)
NA

with λ the wavelength of the light in vacuum because the refractive index
has already been taken into account in the NA value and λ is mostly known
from the light source. The term k1 is named just like that, the k-one factor
and equals around 0.5 for a not optimised system.
With special “tricks” that mainly consist of illumination methods at the
object side, like shown in Figure 7.46, the system is able to image at a higher
resolution. The figure shows as an example the effect of illumination from
a direction under an angle with the optical axis, causing a tilting of all
the diffraction orders. With a perpendicular illumination the higher orders
would be blocked at the aperture stop (the dashed lines) but with this angular
illumination both the 0th and one of the higher orders is able to pass the
aperture, giving enough optical information to create an image.
When the object would be illuminated from two directions under the same
angle with the optical axis or even from all directions in three dimensions (a
cone) under the same angle, called annular illumination even both 1st orders
would be captured. Further also contrast enhancement methods at the
image side help to lower this value. The real minimum value for k1 is 0.25
as this is the situation where not even a fraction of the 1st order is captured
any more with whatever illumination method.
The mentioned methods for achieving a low k1 are applied in waferscanners
as these are not aimed at reproducing a nice representative image of the
object like in photography. Instead, they are designed to produce an image
of a certain structure by optimising all related elements. In that case the
7 The term (NA) and (CD are not written as a variable in one italic symbol and strictly it

should be given a different real variable name but the used notation is quite common in those
industries where it is relevant. For this reason the NA and the CD are noted with straight
capitals.
522 Chapter 7. Optics in mechatronic systems

Pupil plane
Irradiance Irradiance

0th order

Illuminaon
-1st

Object Image

Figure 7.46: Illumination under an angle with the optical axis enables imaging
in case the aperture is too small for capturing both 1st order angles
(dashed lines). The changed angles enable the capturing of the 0th
and one of the 1st orders.

object, also called the mask or reticle, can be adapted to make optimal use
of the possibilities of the entire optical system. The contrast enhancement
techniques at the image are realised in the photosensitive resist that sharply
distinguishes between different levels of irradiance within only a few percent.
It is needless to say that this requires a tight control of the average light
level, called Dose-control.

7.4.4.2 Depth of focus

The tolerance in positioning of an object or an image sensor along the optical

axis in respect to the optical system has many names and depth of focus is
one of them. Other names like depth of field relate often to the tolerance
in position of the object while depth of focus is used at the image side. The
use of the word “field” can be confusing as it also relates to a force field
and in imaging the object and image plane are also often named a field.
Furthermore both positions have a simple relation depending on the focal
length of the optical system. For these reasons the name depth of focus
(DOF) is used in this book.
When the position of the sensor to the lens is fixated, the object is only
imaged sharply around a small area that depends on several parameters.
This tolerance depends primarily on the aperture angle. This means that a
high NA or low f-number, as would be necessary to achieve the maximum
resolution possible, will result in a low depth of focus. With a large aperture
angle only a little displacement away from the focal point will rapidly result
7.4. Physical Optics 523

Image
sensor

c
d

|f
dS So

Figure 7.47: The depth of focus (dS ) of a photographic camera depends on different
parameters: the distance of the subject to the lens (S o ), the minimal
resolution ( c) and the f-number ( 〈 f /#〉 = f /d ).

in a larger luminous spot proportional to this displacement. In non critical

optical systems like in photography where the minimum resolution ( c) is
higher than the diffraction limited value, the allowable error ( dS ) in the
focal displacement of the object (S0 ) in relation to the diameter ( d ) of the
entry-pupil can be derived from simple geometry as shown in Figure 7.47:
dS c
=± (7.36)
So d

With the f-number ( 〈 f /#〉 = f /d ) the following relation for the maximum
allowable focal error is derived:
c S 0 〈 f /#〉
dS = ± (7.37)
f

A large distance, a high f-number, a large minimum resolution value and a

small focal distance all result in a high depth of focus. This corresponds with
the experience of photographers that a telephoto-lens shows a smaller depth
of focus than a wide-angle lens that a high f-number is required for landscape
pictures with objects both nearby and far off that have to be imaged sharp.
It is also the reason that the integrated cameras in inexpensive cell-phones
have often a wide angle lens because of they fail a focusing motor.
The situation at the image side of an optical system is shown in Figure 7.48.
Although the reasoning is identical as for the object side, the distance from
the image to the lens is in most cases much smaller so also the allowable
error in the positioning of the image sensor of a photographic camera is far
smaller. Fortunately this sensor is flat and with automatic focusing it is
possible to correct for misalignment.
More interesting is the shown effect at the focal point when observing a
524 Chapter 7. Optics in mechatronic systems

ϑ
dS F

Figure 7.48: Depth of focus with diffraction limited optics. The enlarged picture
at the right shows the irradiance at the focal point. The profile of an
airy-disk can be observed at the cross section of the focal point, with
dark and light areas next to the central spot. Further it also shows
that the size of the spot remains about the same over a larger range
along the optical axis than expected from the crossing lines, resulting
in a depth of focus ( dS ) of about 2 – 3 times the spot size.

diffraction limited image. In that case it is not possible to simply derive the
tolerance from geometry of the two crossing most outer rays as around the
focal point the wave character shows a very interesting effect.
It was shown in the previous section that the size of the image at the focal
point can never have an infinitely small size because of the size of the airy-
disk that is determined by the numerical aperture.
In spite of this minimum size of the spot in the centre, the maximum
irradiance area spreads over a larger area in the direction of the optical axis
than would be expected from purely geometrical analysis. This results in a
larger focal range according to the following relation:
nλ
dS = ± (7.38)
NA2
In Chapter 9 it will be shown that because of this phenomenon an optical
resolution in the order of 50 nm can be achieved with a depth of focus in the
order of 100 nm, when using immersion optics with a very high numerical
aperture of ≈ 1, 35.
7.5. Adaptive optics 525

7.5 Adaptive optics

In the past decades, the principles of active control are increasingly applied
in complex imaging systems to achieve the ultimate performance. This
mechatronic field of adaptive optics has proven its value in astronomy and
high precision optics for IC lithography but also in medical instrumentation
it became a method to achieve results that were hitherto deemed impossible.
In this section one of the main causes of a less than optimal optical per-
formance of an imaging system is presented which is the generally strong
temperature dependency of the optical path of the light.
It will be shown that the resulting problems can be reduced by inserting
active optical elements such as deformable mirrors with a suitable measure-
ment and control system.

7.5.1 Thermal effects in optical imaging systems

Optical systems operate by virtue of differences in the velocity of light, rep-

resented by the refractive index of the transparent material. Unfortunately
the refractive index is largely influenced by temperature. The impact on the
refractive index by temperature is mainly caused by the changing density
of most materials as function of temperature. In most cases an increasing
temperature will cause a decreasing density. With a lower density there are
fewer atoms to slow down the light over the trajectory which is equivalent
to a lower refractive index.
This however is not always the case. Especially in glass, the refractive index
can increase or decrease as function of rising temperature due to the com-
position with special elements. Next to this direct effect on the refractive
index, the size of an object depends on the temperature. For a lens this
means that also its curvature is not constant.
To examine these effects in a qualitative manner the previously introduced
lens-maker’s equation can be used:

1 1 1
= ( n − 1) − (7.39)
f R1 R2

At first sight it might seem possible to select a glass type with such a positive
d n/ dT that it compensates the effect of the increasing radius by the larger
lens size at a higher temperature.
In reality however often this choice is mostly not available while the effect
on geometry and refractive index of otherwise preferred materials can be
526 Chapter 7. Optics in mechatronic systems

Figure 7.49: The car looks as if standing in water while being in the middle of the
Mojave desert at 40 ◦ C. This mirage is caused by a lens effect that
refracts light from the sky towards the observer. This lens effect is
created by the hot air with a low density close to the ground. Also the
turbulence caused by the thermally induced motion of the air is clearly
visible at the telephone poles while the image of the approaching car
at the horizon is even hardly distinguishable.

quite different.
With for instance the professional lenses for IC lithographic exposure equip-
ment working at the Deep UV wavelength of 193 nm, it is necessary to use
Fused Silica as lens material that shows an large temperature dependency
of the refractive index dn/ dT = 15 · 10−6 while the thermal expansion coeffi-
cient is only ≈ 0.5 · 10−6 .
This means that one ◦ K of temperature could shift the focal plane of a lens
with a focal length of 10 mm with 150 nm which is not acceptable for such a
precision optical system. Even worse is the situation where the temperature
is not constant over the optical system due to absorption of light causing
additional aberrations.
Another well-known example of thermally induced optical errors is the effect
of temperature differences in air like shown in Figure 7.11 in Section 7.2.2. It
is the cause for the famous phenomenon of a mirage as shown in Figure 7.49.
7.5. Adaptive optics 527

7.5.2 Correcting the wavefront

The pupil plane inside an optical system is shown to contain all the spatial
information about the object that is imaged. This information is stored in
the different diffraction orders that all propagate parallel for any point on
the subject.
The wavefront at the pupil plane should be flat because of these parallel
rays. This means that at the pupil plane aberrations can be distinguished
as deviations from this flat wavefront. It was further also shown that the
information on the details is spread over the entire pupil plane with the finer
structures more distant from the optical axis. Adaptive Optics is based on
the principle to introduce a controllable optical element in the optical path
that corrects these deviations from the ideal wavefront after measurement
by a wavefront sensor.
The first application of adaptive optics has been in terrestrial telescopes
where the turbulence and temperature effects in the atmosphere determine
changes in the refractive index leading to wavefront errors.Especially with
the ever larger size of these telescopes, necessary to capture more light
and increase the resolution, the influence of the air becomes dominant. As
an example Figure 7.50 shows the E-ELT, the Extreme Large Telescope of
the European Southern observatory that is planned to be finished around
2018. The largest mirror has a diameter of ≈ 40 metres. It is impossible
to manufacture such large mirrors in one piece and it became customary
for telescopes to compose such a large mirror from a multitude of smaller
elements. The main mirror of the E-ELT consists of around 1000 smaller
mirror elements of 1.4 metres wide. Each of these mirror-elements is pre-
cisely positioned relative to the other mirrors, such that the light from each
element is in phase with the light of the other elements. Only under that
condition the combined mirrors will act like one large mirror with one con-
sistent wavefront.
The precision of this matching needs to be better than λ/20 which is about
15 nm, because light with wavelengths as short as 300 nm has to be imaged.
It is well understandable that a wavefront that passes more than 50 km of
atmosphere can easily become distorted by more than this 15 nm and for
that reason two of the four other mirrors of the telescope have an actively
controlled surface shape.
In the following two sections, first the Zernike modes are introduced as a
mathematically description of the wavefront aberrations, while in the second
section the principle of operation of adaptive optics is further explained.
528 Chapter 7. Optics in mechatronic systems

Figure 7.50: The Extremely Large Telescope of the European Southern Observatory
(E-ELT) has a main mirror with a diameter of just less than 40 metres
that consists of ≈ 1000 segments with a diameter of ≈ 1.4 metres that
all have to be aligned with an accuracy of λ/20 ≈15 – 50 nm. Two
adaptive secondary mirrors are used to correct for wavefront errors
by atmospheric disturbance. (courtesy of ESO)

7.5.2.1 Zernike modes

A deviation from an ideal wavefront can mathematically be described in

orthogonal polynomials, called Zernike polynomials, that are defined in a
polar coordinate system over a circular plane normalised to radius 1 (unit
disk) around the optical axis. The full waveform deviation is then approx-
imated by the summation of a plurality of singular shapes (modes) of a
different order and magnitude. This approximation is comparable with the
Fourier analysis of signal waveforms and modal shapes of a dynamic system.
The different Zernike polynomials or Zernike modes that define these sin-
gular shapes are named after the Dutch scientist and Nobel prize winner
Fritz Zernike (1888-1966), who defined them. Next to wavefronts, also the
surface of optical elements and the image shape in the field plane could
be described by these polynomials, but mostly the use of Zernike modes is
limited to the pupil plane where the ideal wavefront should be flat.
There are even and odd Zernike polynomials, where the even ones are defined
7.5. Adaptive optics 529

as:
q q
Z p (ρ , ϕ) = R p (ρ ) cos( qϕ) (7.40)

and the uneven ones as:

−q q
Z p (ρ , ϕ) = R p (ρ ) sin( qϕ) (7.41)

where both p and q are non negative integer numbers, p − q are even and
p ≥ q.
The angle ϕ is the azimuthal angle, between a reference vector and another
vector that originates in the centre and points towards the location of interest.
The value ρ is the normalised radial distance on the unit disk with diameter
of 1.
The radial polynomials R qp (ρ ) are given by:

( p−
 q)/2
q (−1)k ( p − k)!
R p (ρ ) = ρ p−2k (7.42)
k=0 k!(( p + q)/2 − k)!(( p − q)/2 − k)!

This equation results in the following list of radial polynomials for the first
21 Zernike modes, where p + q is the order of the Zernike mode8 :

R 00 (ρ ) = 1 (7.43)
R 11 (ρ ) = r
R 20 (ρ ) = 2ρ 2 − 1
R 22 (ρ ) = ρ 2
R 31 (ρ ) = 3ρ 3 − 2ρ
R 33 (ρ ) = ρ 3
R 40 (ρ ) = 6ρ 4 − 6ρ 2 + 1
R 42 (ρ ) = 4ρ 4 − 3ρ 2
R 44 (ρ ) = ρ 4
R 51 (ρ ) = 10ρ 5 − 12ρ 3 + 3ρ
R 53 (ρ ) = 5ρ 5 − 4ρ 3
R 55 (ρ ) = ρ 5
8 Several other definitions of these polynomials exist that differ in details, for instance by

normalising on maximal values at the edge of the circle, like 1 as used here. Also the units
can relate to metric values like [nm] or wavelengths. Further the order is often mentioned
as ” p“ equalling the radial order and ” q“ equalling the azimuthal order and the sequence
of numbering is either simply based on an increasing radial order or on a more complicated
expansion with spherical terms for all quadratic numbers. All these definitions are correct as
long as maintained consequently in all communication.
530 Chapter 7. Optics in mechatronic systems

Is important to iterate another part of the definitions of a wavefront for

the explanation of the meaning of these modes in a descriptive form. A
wavefront is a representation of the light coming from an infinitely small
spot while the photons over the wavefront have a fixed phase relation or
equal timing when starting at the same moment. Delaying part of the
photons will distort the wavefront and as a consequence the direction of the
photons is changed. This all means that each point on the object creates a
wavefront that is flat at the pupil plane and can de distorted by errors in
the imaging system.
In Figure 7.51 a graphical overview of the first 21 Zernike modes is given to
visualise the different shapes. The upper image corresponding with Z00 = 1 is
the plane wave term, equal to the mean value of the wavefront. It represents
a constant delay of all photons in the wavefront, caused by for instance a flat
piece of glass. One step down, where p = 1, the Zernikes represent the effect
of tilting around the vertical or horizontal axis. These effects can be caused
by a wedge shape piece of glass, called a prism and means that the main
direction of the wavefront is changed over a tilting angle. The third row
with p = 2 shows in the middle the effect of defocus which causes a parabolic
wavefront in the pupil plane and is represented by Z20 . As a consequence of
the defocussing, the inner rays will be either delayed in respect to the outer
rays or vice versa and they will no longer coincide on the same focal point
on the image. The image becomes unsharp.
In the section on geometric optics three aberrations were introduced, spher-
ical aberration, astigmatism and coma. The examples were shown with
positive single lens elements where the pupil is determined by the size of the
positive lens itself. For linking the aberrations with the images in the figure
one can imagine the pupil plane with parallel rays to be somewhere inside
the lens element. The primary versions of the three mentioned aberrations
are represented by the 4 th order Zernike terms.
Spherical aberration is represented by Z40 and a cross section through the
middle looks like a double sine wave. When comparing with Figure 7.19
the outer and inner rays are delayed in respect to the rays in between. It
was shown in that figure that it should be corrected by adding more glass,
meaning more delay, in that intermediate area.
Astigmatism is represented by Z2−2 and Z22 and can be seen as a saddle shape.
When comparing with Figure 7.21 it is shown that rays in one plane are
delayed in respect to rays on the optical axis while rays in an orthogonal
plane are in advance to the rays at the optical axis.
Coma is represented by Z3−1 and Z31 and its cross section through the low and
7.5. Adaptive optics 531

Piston

Tip Tilt
Z00

@ @
Z1-1 Z11
Defocus
Trefoil Trefoil
Z2-2 Z20 Z22
š
‹! "
 ‹! "

Z3-3 Z3-1 Z31 Z33



Z4-4 Z4-2 Z40 Z42 Z44

Z5-5 Z5-3 Z5-1 Z51 Z53 Z55

Figure 7.51: Height map of the first 21 Zernike modes describing deviations of an
ideal flat surface in a systematic manner. Generally deviations of
wavefronts can be modelled as a combination of several Zernike modes
with different magnitude. In this chart blue corresponds with 1, green
with zero and red with -1.
(courtesy of Claudio Rocchini)

high area looks like a single sine wave. When comparing with Figure 7.22
this is less easy to imagine and with the higher order terms it is not useful
to even try, but one thing is certain, the higher the Zernike terms the more
refined the correction mechanism should be.
Several of the higher order terms can be seen as high spatial frequency
versions of the described primary aberrations. For instance Z42 and Z4−2
represent secondary astigmatism and Z51 and Z5−1 is secondary coma.
The other Zernike modes also have their own name based on their shape
like trefoil for Z33 and Z3−3 , quadrafoil for Z44 and Z4−4 , pentafoil for Z55 and
Z5−5 and consequently Z53 and Z5−3 are called secondary trefoil.
532 Chapter 7. Optics in mechatronic systems

Distorted wavefront Beam splier Mul - lens array CCD -

image sensor

Wavefront
analyzer

Shack -Hartmann
wavefront sensor

Actuator array
Image

Controller
Deformable mirror

Figure 7.52: Principle of adaptive optics with feedforward compensation. By in-

serting a beam splitter in the pupil area the distorted wavefront can
be measured and corrected by a deformable mirror. Notice that the
deformation magnitude of the deformable mirror has to be less than
the distortion magnitude of the wavefront depending on the angle of
incidence.

7.5.2.2 Adaptive optics as correction mechanism

Based on these mathematically described deviations of an optical system,

adaptive optics is developed as a mechatronic method where the wavefront
errors in the pupil plane are measured and corrected by means of an actively
controlled optical element and a suitably tuned control system.
In principle one could consider automatic focusing of a photo camera as a
kind of adaptive optics and indeed it shows all elements of an active con-
trolled system including sensor, actuator and controller. Generally however
the term adaptive optics is reserved for those applications where the curva-
ture of optical elements themselves is actively controlled in order to reduce
the wavefront errors.
7.5. Adaptive optics 533

7.5.3 Principle of operation

The principle of operation of adaptive optics is shown in Figure 7.52 which

shows a feed forward compensation method. The light with the distorted
wavefront enters the system at the upper left side, as shown by the dashed
line in the pupil area where the light is running parallel.
As a first step a beam splitter splits the light in two paths. The irradiance
ratio between these paths is not important for the working principle but one
can imagine that with an expensive telescope it would be a waste to send
too many photons to the wavefront sensor.
The wavefront is analysed with a wavefront sensor that is based on the prin-
ciple that light propagates in a direction orthogonal to the wavefront. The
most simple version of a wavefront sensor consists of a plate with tiny holes
where the light through the holes will illuminate a CCD image sensor like
used in a digital camera that is positioned behind the plate. The position of
the illuminated areas is determined by the direction of the light orthogonal
to the local wavefront. With a suitable algorithm the wavefront can be recon-
structed from the position of the spots on the CCD sensor. This wavefront
sensor is called the Hartmann sensor after the German physicist Johannes
Franz Hartmann (1865 – 1936) who invented the principle. It was later
improved by The American physicist Roland Shack to the Shack-Hartmann
wavefront sensor by replacing the tiny holes by an array of small lenses
that are shown in the figure. In the figure this principle is indicated in an
exaggerated way, where the distorted waveform results into the mentioned
spatially shifted pattern of spots. Although only five lenslets are drawn,
in reality many more lenslets are used depending on the required spatial
resolution. This refined multi-lens array covers two degrees of freedom in
order to precisely analyse the wavefront over the entire pupil plane.
The obtained wavefront information is used in a controller to adapt a de-
formable mirror in such a way that it compensates the waveform distortion
by introducing Zernike terms with an equal magnitude as the distorted
wavefront but with an opposite sign.
The deformable mirror is equipped with a multitude of actuators that are
capable of covering all relevant Zernike terms, comparable to the multi-lens
array of the wavefront sensor. At the end of this section an example of such
an actuated mirror will be shown in more detail.
It should be noted that the deformation of this correcting mirror has to be
less than the magnitude of the distortion of the wavefront. This is caused
by the angle of incidence of the light in respect to the surface of the mirror.
When the mirror surface is orthogonal to the incident light the optical path
534 Chapter 7. Optics in mechatronic systems

Acve mirror
Distorted wavefront
Controllerr

Image

Wavefront sensor

Figure 7.53: Principle of adaptive optics with feedback control. The mirror is con-
trolled such that the error on the wavefront sensor is kept as close as
possible to zero.

difference of light reflecting on the mirror is twice the displacement of the

surface. When the surface is tilted under an angle of 45◦ as in the shown

examples the timing difference of the light 2 times the displacement. A
larger tilt angle will require a larger deformation magnitude.
Although this kind of feedforward compensation is used in practice, it re-
quires a precisely predictable behaviour of the actuated mirror and a flawless
sensing of the wavefront over a large range. This is only possible with fre-
quent calibration.
For that reason more often a second option is applied by “reshuffling” the
parts and creating a real-time feedback loop as is shown in Figure 7.53.
With feedback control the light is reflected first at the active mirror followed
by the beam splitter with the sensor and the imaging part.
The mirror is controlled in such a way that the deviations as measured in
the sensor are as close as possible to zero. This zeroing tracking-control
approach has as benefit that the sensor only needs to be calibrated for a
plane wavefront and that the actuators on the mirror only have to be linear
enough to not impair the stability of the feedback loop.
In principle all controlled channels from one measured wavefront location to
the corresponding actuation part are interconnected and mutually interfere
with each other both mechanically and optically so the control algorithm is
not obvious. Especially when the errors in large space telescopes need to
7.5. Adaptive optics 535

|{J

Membrane
suspension c material

?
'!™'!
|J
Permanent magnet Planar coil

Figure 7.54: Actively controlled mirror with “Hybrid” permanent magnet biased
actuators. The actuator at the left has no current, in the middle
the current direction increases the permanent magnet flux and at
the right the current direction decreases the flux of the permanent
magnet. (courtesy of Roger Hamelinck, TU/e)

be corrected it appears that for a 40 metre telescope around 105 actuators

would be needed. For real-time control this amount of channels requires
a multitude of more than 104 FPGA processors when using classic SISO
control. For this reason a real MIMO control system with distributed con-
trol algorithms is preferred which takes the actions of the neighbouring
elements into account.
Many possibilities have been investigated to realise controllable mirrors,
often with piezoelectric or electrostatic actuation of the thin plate that acts
as the deformable mirror.
As an example, Figure 7.54 shows a design that was realised at the Eind-
hoven University of Technology. This configuration uses permanent magnet
biased reluctance actuators to determine the shape of the mirror.
536 Chapter 7. Optics in mechatronic systems

This adaptive-optics mirror-system consists of several layers. The upper

layer is a thin sheet of glass with a suitable reflection coating that acts as
the mirror. It is connected by thin strands of steel wire to the middle layer,
the moving part of the actuator, consisting of ferromagnetic pieces of metal
that are suspended by a membrane. The lower layer, the stationary part,
consists of a base plate with coaxial permanent magnets and flat wound
coils that create a magnetic field that can be modulated by the current in the
coils. Without current the permanent magnet secures a certain pretension.
Depending on the current direction, the attraction force to the ferromagnetic
piece of metal is reduced or increased. Due to the coupling with the thin
wires, the mirror can adapt its shape according to the average position of
the actuator without introducing discontinuities at the connection points
while also the heat transfer from the actuators to the mirror is minimised.
Only a small part of a larger mirror is shown. In the real system the
actuator part will be assembled from a multitude of the shown hexagonal
substructures but the mirror itself will be one larger surface.
Side note: This version of the “hybrid” actuator has the permanent magnet
inserted in series with the flux path of the coil. According to the theory, this
design increases the reluctance of the magnetic flux induced by the current,
resulting in a reduced energy efficiency.
A more preferred hybrid actuator design with the magnet outside the flux
path of the coil would require a symmetric configuration with an additional
part on the other side of the membrane suspension as described in Sec-
tion 5.3.4. Such a design would complicate the layout excessively and also
the benefits of a more optimal magnetic configuration are limited because
of the small size per actuator with its relatively larger reluctance of the air
gaps. For this reason the shown design is more optimal when all factors are
taken into account, including manufacturability.
Chapter 8

Measurement in
mechatronic systems

The precision of mechatronic systems can not be better than the measure-
ment accuracy of all relevant parameters that are needed to control and
adapt the behaviour of the system. The other elements in the system are
never capable of correcting unknown measurement errors.
The field of metrology deals with measurement in general and is in itself
quite immense, even when only dealing with mechanical quantities. One of
the reasons for this wide scope is the economic need for undisputed quan-
tities when trading goods. Metrology has been practised from the very
moment that people started to exchange valuable items and this has re-
sulted in agreements and rules on a global scale regarding traceability, the
possibility to relate a measurement value to an agreed standard like the
kilogram and the metre. For the same reason as with the other chapters,
this chapter has to remain limited to the most relevant measurement items
for precision motion systems, being the measurement of forces, position and
motion. This includes sensors for stress, velocity and acceleration, based on
physics principles like piezoelectricity, electromagnetism and optics.
This chapter is divided in seven sections.
Section 8.1 is a more general introduction of the basic principles and mod-
elling of measurement systems. Also some definitions from the traceable
metrology field will be introduced, especially regarding measurement uncer-
tainty.
Section 8.2 presents the statistics around random errors by introducing
an important statistical method, called Dynamic Error Budgeting. This

537
538 Chapter 8. Measurement in mechatronic systems

Physical Signal Signal Technical

Sensing
quanty Condioning processing value

Figure 8.1: Measurement systems can be modelled as a series of three elements.

The sensing part translates the physical signal into an electrical signal.
The signal conditioning adds robustness to the sensor signal and the
signal processing translates the signal into a quantitative value.

method enables a mechatronic designer to determine the total error in a

dynamic system from different error source contributions. While statistics
are very important for the analysis of measurement errors, this method
is also applicable for the estimation of positioning errors in a mechatronic
motion system.
Section 8.3 concentrates on the electronic signals in the most sensitive
elements of a measurement system. Different sources of errors will be dis-
tinguished and methods are presented to reduce the negative effects.
Section 8.4 introduces dedicated electronic circuits that are used to add
robustness to weak signals. It strongly builds on the basic theory of opera-
tional amplifier circuits that was presented in Chapter 6.
Section 8.5 presents the methods to create digital data from the analogue
signals in a reliable way.
Section 8.6 is the first of three application sections and concentrates on
sensors in short range position measurement systems.
Section 8.7 introduces sensors for the dynamic measurement of mechanical
quantities like force, velocity and acceleration.
Section 8.8 presents the important long range optical position measurement
systems with encoders and laser interferometry.
8.1. Introduction to measurement systems 539

8.1 Introduction to measurement systems

The function of a measurement system is to translate a physical quantity,

the measurand, into a representative engineering quantity. Depending
on the application, this can be an analogue or a digital representation.
For example the pointer of a speedometer in a car is a typical analogue
representation, while the numerical reading of a fuel pump in a gas station
is an example of a digital representation.
A basic measurement system for controlling a process or function can be mod-
elled to consist of 3 successive functional elements1 as shown in Figure 8.1
that each have very specific properties.

• A sensing element that converts a variable quantity of a physical

phenomenon (physical signal) into an analogue electrical signal.

• A signal conditioning element that adds robustness to the analogue

signal by means of amplification such that it can be transferred to
another location without risk of disturbances by interfering signals.

• A signal processing element that converts the analogue signal into a

different analogue or a digital value that is adapted to the requirements
for active control of a larger system and process.

Examples of sensors are a thermocouple that creates a voltage as function of

temperature, a light sensitive diode that converts a stream of photons into
an electric current and a moving coil in a magnetic field that gives a voltage
as function of the velocity. Low-power signal amplifiers are typical examples
of signal conditioning elements and an analogue-to-digital converter is a
typical signal processing element.

8.1.1 Errors in measurement systems, uncertainty

All elements of a measurement system show deviations in their intended

performance. These deviations lead to an overall difference between the
measured quantity of the measurand end the real value that would have been
measured with a perfect instrument. This difference, called measurement
error is the subject of research of many scientists.
1 In general literature on measuring a 4 th element is often added that describes the data

representation part by means of a display, an analogue pointer, printer or plotter. Because of

the limitation to measurements for controlling a dynamic process without an external display,
this element is omitted here.
540 Chapter 8. Measurement in mechatronic systems

In official traceable metrology for economic purposes several terms are

defined to classify and deal with measurement errors. These terms are
laid down in the ”Guide to the Expression of Uncertainty in Measurement“
(GUM) and the ”International vocabulary of metrology, basic and general
concepts and associated terms“ (VIM) under responsibility of the ”Joint
Committee for Guides in Metrology“ (JCGM) where the ”Bureau Interna-
tional des Poids et Mesures“ (BIPM) is participating as international stan-
dards organisation. This part of metrology is less relevant for mechatronic
systems, because most controlled motion systems rather rely on internal
references than on external traceable, often also called absolute standards.
Nevertheless it is useful to be aware of these conventions when using the
different terms in communication with others, for which reason some of
these definitions will be explained in this section.
Errors in measurement systems can be divided in two parts, systematic
errors and random errors. Systematic errors are those measurement errors
that in replicate measurements remain constant or vary in a predictable,
deterministic manner. Random errors are those measurement errors that
in replicate measurements vary in an unpredictable manner.
As an example of a predictable systematic error one can think of the influence
of temperature on the measurement of another physical quantity. It is
often possible to measure the temperature and compensate for its influence,
when the physical model of the system regarding temperature is known. An
example of a random error is for instance the momentary voltage value of the
thermal noise in a resistor. In order to decrease the error in measurement
systems it is obvious that it is necessary to especially reduce the amount of
random errors to a minimum.
The measurement uncertainty of a measurement system is quite strictly
defined as a non-negative parameter, characterising the dispersion of the
quantity values, being attributed to a measurand, based on the information
used. It includes all random and non-compensated systematic errors. It case
of compensation of predictable systematic errors it also includes random
errors in the measurement that is used for the compensation, like the
temperature measurement as mentioned in the example before.
The term measurement accuracy is a relative, mainly qualitative term,
mostly used when indicating the difference between two measurements
of which one is more accurate than the other. When quantised the given
numbers are often only relating to the order of magnitude and as such this
term is less strict than measurement uncertainty.
The term measurement precision is a relative, quantitative term, usually
8.1. Introduction to measurement systems 541

expressed numerically by measures of imprecision relative to the maximum

value of the measurand. One can think of the standard deviation, variance,
or coefficient of variation under the specified conditions of measurement. In
fact, measurement precision relates measurement uncertainty to the real
value.
Last but not least, the term measurement resolution refers to the smallest
change in a quantity being measured that causes a perceptible change in
the corresponding indication. It is very important to recognise that the
maximum resolution of a measurement system is quite something different
than uncertainty. In fact the resolution has always the smallest value of the
two, for which reason several suppliers of measuring systems like to only
mention the resolution in their marketing communication.

8.1.1.1 The ultimate in uncertainty

Many years ago scientists like Laplace were convinced that somewhere in
the future all physical processes would be exactly predictable.
Quantum mechanics however, has given us other insights with for instance
the Heisenberg uncertainty principle, postulated by the German physicist
and Nobel prize winner Werner Karl Heisenberg (1901 – 1976). His prin-
ciple states that the standard deviation in the measurement errors of two
parameters, like place (σx ) and impulse (σp ) or energy (σE ) and time (σt ),
can not be smaller in combination, than given by the following expression:
h h
σx σp ≥ and σx σp ≥
4π 4π

with the Planck constant h = (6.6260693 ± 0.0000011) · 10−34 [Js].

Fortunately in the practical world of mechatronic reality this number is
so small that it is not necessary to take this uncertainty limit into ac-
count. There is still ample room for scientists and engineers to work on
improvements in the precision of measurement systems and keep consis-
tently shifting the limits in measurement uncertainty.
542 Chapter 8. Measurement in mechatronic systems

8.1.2 Functional model of a measurement system ele-

ment

Every element of a measurement system can be modelled as a separate

subsystem with a primary input for the useful information, secondary in-
puts of external error sources and an output. This model is represented in
Figure 8.2.
Ideally an element would only show a proportional linear transfer function
L( i ) from the measurand input i to the output o, with a transfer gain of K  .
In most sensors however a certain level of non-linearity causes a not fully
proportional output value. This non-linearity can be expressed mathemati-
cally as a series-expanded polynomial that contains the higher-order terms
of the total transfer function of the element.

∞
N ( i) = cn i n (8.1)
n=2

Mostly the used terms are limited to only those that contribute to the
significance of the output value o, as determined by the uncertainty level of
the measurement.
Next to the non-linearity, external influences can impair the functionality of
the element in two ways. Firstly, the external disturbance input can result
in an additional value of the output of the element irrespective of the mea-
surand input value. This disturbance input is called the interfering input,
with a transfer gain K i . The second way of impairing the measurement
output is by changing the proportional gain of the measurand input transfer
function and is called the modifying input, with a transfer gain K m . In prin-
ciple, the interfering input can have a similar non-linear transfer function
as the measurand input, but often the higher-order terms of the related
polynomial can be neglected, as the sensitivity of the interfering input is
mostly much less than the sensitivity of the measurand. In high precision
systems, however, this linearisation is not always allowed, especially when
these effects need to be compensated.
Many external influences show a combination of both effects. Sometimes
even the gain factors K m and K i depend on the actual value of the input or
output signal, which makes it even more awkward to model. After compen-
sation of the systematic errors, a precision system generally mainly shows
random interfering errors.
Two other factors that influence the output of the element are shown in the
model, the setpoint and the dynamics. The setpoint (a) is an intentional
offset-value to define the range of output values that belongs to a certain
8.1. Introduction to measurement systems 543

Error source, disturbance

Modifying Input (em) Interfering Input (ei)

Mulplied Added
M (i, em ) K m  i ˜ em  I (ei ) K i ˜ f  ein   n t 1

Element Linear + + Dynamics Element

Input (i) L(i ) KA ˜ i TF f t,Z  Output (o)

Non linearity
N (i ) f in   n t 2 Setpoint (a)

Figure 8.2: Functional model of an element in a measurement system. In case of

a sensor the element input is the measurand. Two ways of disturbing
the output signal are shown. The modifying input changes the gain of
the transfer function of the element input, while the interfering input
just adds a signal to the output. Also non-linearity and dynamics can
influence the functionality of the element.

range of input values. For instance with a thermal sensor, a temperature

range of 5 – 25◦ C could correspond with an output voltage range of 0 – 10 V.
This example would require a linear transfer gain K  = 0.5 and a setpoint of
a = 5 V.
The Dynamics box represents the time or frequency dependent behaviour
of the element and can be modelled with dynamic transfer functions. An
typical example of an element with a first-order dynamic transfer function
is a thermal sensor. The dynamic behaviour is determined by the heat
capacity of the sensor and the heat conductivity between the sensor and the
body to be measured. Second- and higher-order transfer functions are most
often present in mechatronic positioning systems when the position sensor
is located at a different place than the actuator.
544 Chapter 8. Measurement in mechatronic systems

8.2 Dynamic error budgeting

Random errors can only be quantified by means of statistics because of

their unpredictability. Precision measurements are often repeated several
times to reduce the impact of random measurement errors by applying
statistical methods. Measurements in mechatronic systems are mostly
continuous and the errors are a combination of a multitude of continuous
interfering random signals. Also these can be treated in a comparable
statistical way. After a short introduction in the statistics of errors with
repeated single measurements, this section concentrates on the statistics
of signals by introducing the concept2 of Dynamic Error Budgeting (DEB).
This methodology to derive the total error of a system from its different
dynamic contributors, has proven to be extremely valuable in determining
and solving error sources that act on real measurement and positioning
systems.

8.2.1 Error statistics in repeated measurements

In official traceable metrology, errors are investigated by means of the analy-

sis of many measurements. The observed distribution of the different values
gives information about the origin of possible error sources, their character
and methods of improvement. The related field of probability statistics is
extremely wide with many different distributions that are represented with
a Probability Density Function p( x) (PDF), giving the probability p x1 ,x2 that
a measurement value is observed within a certain range of values from x1
to x2 .

x2

p x1 ,x2 = p ( x) d x (8.2)
x1

The probability that the measurement can have any value, is by definition
equal to one. This implies that the integral from −∞ to +∞ of p( x) always
equals one. The related Cumulative Probability Function (CPF) is defined
by the integration of p( x) from −∞ to x. Often the PDF and the CPF are
represented in a graphical way, like shown in the example of the following
section. This representation clearly illustrates these statistical properties
of the system. On the horizontal axis the range of values is shown while the
2 This section is derived from the Phd thesis of Leon Jabben from our laboratory in Delft.

His thesis can be downloaded for more details.

8.2. Dynamic error budgeting 545

vertical axis denotes the probability density or the cumulative probability.

Generally both axes are linear scaled.

8.2.2 The normal distribution

In precision positioning systems, errors are observed as deviations in sig-

nals with corresponding frequencies and amplitudes. For a precision mea-
surement system in general, all systematic errors are compensated and
calibrated. This means that only uncorrelated random signals remain as
error source. As mentioned these errors have are mainly an interfering
character which means that they are independent of the measurand.
Random errors in signals can almost always be represented by means of a
special probability density function, the normal distribution or Gaussian
distribution, named after the same Johann Carl Friedrich Gauss, who postu-
lated two of the Maxwell equations. The normal distribution is characterised
by its mean value x and its variance σ2x , while the square root of the variance
is called the standard deviation σx . These terms are defined in the following
way for multiple measurements:
1 n
mean : x= xi
n i=1
(8.3)
1  n
2
variance : σ2x = ( xi − x̄) ,
n − 1 i=1

For random signals also a third term is introduced, the signal power Ps that
proves to be useful in combining several error signals. For random signals
the three terms are calculated as follows:

T
1
mean : x = lim x( t) d t
T →∞ 2T
−T

T
1
power : Ps = lim x( t)2 d t (8.4)
T →∞ 2T
−T

T
1
variance : σ2x = lim ( x ( t ) − x )2 d t
T →∞ 2T
−T

These definitions are somewhat extreme as in practice the “lim” term is

never really necessary. Most disturbance signals are quite continuous and
in practice a period T is chosen that sufficiently represents these interfering
signals. A practical “rule of thumb” is to take T equal to the period of the
lowest relevant frequency of the spectrum in the disturbance signal. For
546 Chapter 8. Measurement in mechatronic systems

p(x) 0.339 cum p(x) 1

Vx

0.5

x 3Vx x 2Vx x Vx x x Vx x 2Vx x3Vx x 3Vx x 2Vx x Vx x x Vx x 2Vx x3Vx
x x
?

 
š!!'



Figure 8.3: The probability density function and cumulative probability function of
a normal distributed random signal with a mean value x̄ and a standard
deviation σx . The probability that a measurement value is located in a
certain range is equal to the surface enclosed by the graph of the PDF
and the horizontal lines that correspond with the range boundaries.

this reason in practice the “lim” term can be omitted.

It is also useful to remark that the variance equals the power and the stan-
dard deviation equals the root of the power for signals with a zero mean
value. This standard deviation is then equal to the Root Mean Square value
(RMS) as defined in Section 2.1.3.3 of Chapter 2.
Figure 8.3 shows a graphical representation of the probability density func-
tion and the cumulative probability function of a normal distributed random
signal.
With the above definitions the mathematical expression of the probability
density function is as follows:
⎛ ⎞
( x − x̄)2
⎝− ⎠
1 2σ2x
p ( x) =  e (8.5)
σx 2π

The calculation of the probability that the error is between x̄ ± σx , x̄ ± 2σx or

x̄ ± 3σx , results in the following values:
σx
p −σx ,σx = p( x) d x = 0.683
−σx
2σx
p −2σx ,2σx = p( x) d x = 0.955
−2σx
3σx
p −3σx ,3σx = p( x) d x = 0.997 (8.6)
−3σx

These calculations imply that only 68 % of the measured values can be found
in a range of ±1σ around the mean value. A range of ±2σ already contains
8.2. Dynamic error budgeting 547

more than 95 % of all measurements, while at ±3σ, less than 0.3 % is found
outside the range. For this reason measurement systems are often specified
with either 2σ or 3σ values for the measurement error.

8.2.3 Combining different error sources

Under certain conditions, different error signals can be combined by quite

simple relations. In case of several mutually independent variables, it is
allowed to combine the mean value and standard deviation of each variable
by means of the method of the root of the sum of squares. When αn equals
the proportion of the specific variable in the total measurement value, the
following expressions are used to determine the mean value and the standard
variation of the combined measurement:

x̄tot = α1 x̄1 + α2 x̄2 + α3 x̄3 + ......

%
σtot = (α1 σ1 )2 + (α2 σ2 )2 + (α3 σ3 )2 + ..... (8.7)

In practice it is often allowed to linearise the different effects because of the

small variations of the inputs around a certain average value. This allows
the use of simplified calculations with systems of which the errors are given
in terms of their mean value and standard deviation.
The linearisation starts with the behaviour of an element according to the
model written as a combination of the different inputs:

o = K  i + a + N ( i ) + K m ie m + K i e i (8.8)

For small deviations the non-linear term N ( i ) can be combined with the
linear term L( i ) and the equation becomes:
∂o ∂o ∂o
δo = δi + δem + δei (8.9)
∂i ∂em ∂ei
where:
∂o ∂o ∂o
≈ K, ≈ Km i and ≈ Ki (8.10)
∂i ∂em ∂ei
In most cases the different inputs are independent and their impact can be
statistically combined using the rule of “the root of the sum of squares”. The
combined standard deviation of the element becomes:

 2  2  2
∂o ∂o ∂o
σo = σ2i + σ2e m + σ2e i (8.11)
∂i ∂em ∂ei
And the median can be derived by simple addition:

ō = K  ī + a + N ( i ) + K m ie m + K i e i (8.12)
548 Chapter 8. Measurement in mechatronic systems

8.2.4 Power spectral density and cumulative power

Dynamic error budgeting is a method to derive the total error in a dynamic

system by using the described statistical methods to combine the contri-
butions of all disturbance sources into one error value. The method starts
with “the root of the sum of squares” that directly relates to the power of
the different disturbance signals that act on a measurement system. When
the total error of a measuring instrument by different error sources needs
to be determined with this method, it is most useful to use two important
related functions, the Power Spectral Density (PSD) and the Cumulative
Power Spectrum (CPS). The power spectral density P ( f ) gives the power of
a signal within a certain frequency range:

f2

P f1 , f2 = P( f ) (8.13)
f1

The cumulative power spectrum is calculated by integrating P ( f ) starting

from 0 Hz with increasing frequency until no further increase is observed.
Also these spectra are shown in a graph with the frequency on the horizontal
axis and the magnitude of the power density or the cumulative power on
the vertical axis. The axes of the power spectral density are preferably
double logarithmic, because the error signal characteristics often show a
strong similarity with the Bode-plot of the dynamic transfer functions. The
magnitude scale of the cumulative power spectrum is noted on a linear
scale in order to avoid confusion with the contributions of the different error
sources.
To illustrate this method of dynamic error budgeting, Figure 8.4 shows as
an example the effect of three error sources on the inspection microscope
that was introduced in Chapter 3.
The first error source is due to the transmission of floor vibrations to the
sensitive part of the instrument. To prevent this transmission, a vibration
isolation system is applied, consisting of springs that connect the mass of
the table and the microscope with the floor. This vibration isolation system
has a first eigenfrequency at 2 Hz. The shown power spectral density of
this first error source is the combination of the unfiltered vibrations of the
floor, the transmissibility transfer function of the vibration isolation system
and the sensitivity of the instrument to this interfering error source. At low
frequencies the microscope acts as a rigid body and the impact is low. At
the resonance of 2 Hz a maximum is reached in the spectral density, with
a negative slope at higher frequencies due to the transmissibility trans-
fer function of the vibration isolation system. At 160 Hz the resonance of
8.2. Dynamic error budgeting 549

@!
!
M!
\JXƒ

!

Object

Table to

2 Hz

Floor vibraon
4
10 100
3 ~90 nm
8 .10 80
0
CPS [nm ]
P( f ) [nm 2 /Hz]

10
2

CAS [nm]
3
6 .10 60
3
4 .10 40
total
>\
10 '
 3
2 .10 20
@!
! CPS ~30 nm
”

! CAS
> 0 1 2 3 0 > 0
0 1 2 3
10 10 10 10 10 10 10 10 10 10
Frequency [Hz] Frequency [Hz]
? 
š!!' 
! !

Figure 8.4: The power spectral density, the cumulative power spectrum and the
cumulative amplitude spectrum of an inspection microscope with dif-
ferent error sources, the floor vibrations, acoustic disturbance forces
acting on the measurement head and the dynamic disturbance from
the measurement head itself. The resonance of the vibration isolation
system at 2 Hz and the resonance of the microscope head at 160 Hz
show to be the main error contributors.
(Courtesy of Leon Jabben)

the measurement head shows the typical dynamic decoupling effect of this
eigenmode.
The second error source originates from acoustical disturbances by for in-
stance the air conditioning unit of the room, where the microscope is used.
The power spectral density of the combination of these disturbances with the
sensitivity of the instrument shows a maximum at the resonance frequency
of 160 Hz.
The third error source is caused by the dynamic properties of a feedback
controlled positioning system that moves the measurement head within
550 Chapter 8. Measurement in mechatronic systems

the microscope. By a bad design of the controller and a noisy sensor it

increases the internal resonance of the microscope head which means that
these dynamics are also most disturbing at the 160 Hz resonance frequency.
The total power spectral density is the sum of all PSD graphs and due to
the logarithmic scale, it is equal to the enveloping curve of the three error
contributors. In this example the total error spectrum appears to be only
determined by the floor vibrations and the dynamic disturbances by the
positioning system of the microscope head. The acoustic disturbance effect
remains below the error floor over the full frequency range.
The next step that is necessary to derive the total error is the determination
of the cumulative power spectrum. Starting at low frequencies, a first rise
in the cumulative spectrum is observed in the graph at 2 Hz and a much
larger second rise at 160 Hz with a final power level of ≈ 8.1 · 103 nm2 . This
difference in magnitude is confusing at first sight, as the surface of the
peak in the PSD at 2 Hz is much larger than the surface of the peak at
160 Hz. This is however caused by the logarithmic frequency scale of the
PSD. Integrating the PSD over that low-frequency area only results in a
relatively small cumulative value.
To determine the standard deviation of the final error, the root of the total
CPS value should be taken, which is 90 nm in the above example. Due to
the two eigenmodes in the system, this value is much higher than the 10 nm
error that was calculated in Chapter 3. This difference is caused by the two
undamped resonances and the largest contributor to this error is the badly
designed controller of the positioning system of the microscope head.
As was shown in Chapter 4 on motion control, an undamped mass-spring
system can be controlled with a well tuned PD-control setting such that
the resonance is completely suppressed by shifting the poles to the left in
the complex plane. In that case no resonance will occur in the dynamic
disturbance and the step in the CPS at 160 Hz will become negligible. The
remaining error of around 30 nm is then only determined by the floor vibra-
tions. As a consequence the next improvement measures should be focused
on the vibration isolation system where some damping might be sufficient
to sufficiently reduce the peak at 2 Hz. Too much damping will however
increase the disturbance by the floor vibrations at higher frequencies so this
measure should be taken with precaution.

8.2.5 Cumulative amplitude

When trying to achieve the minimum error level in a complete measuring

system, it is frequently observed that designers directly convert the CPS
8.2. Dynamic error budgeting 551

graph into a Cumulative Amplitude Spectrum (CAS)3 graph by deriving the

root of the CPS values. Although the resulting fully cumulated end value
remains the same, this representation is not advisable for problem solving.
By taking the root, the impact of the disturbances at the low-frequency side
is visually enlarged, relative to the impact of the disturbances at higher
frequencies.
This is clearly illustrated by the red line of the CAS graph as compared
with the blue line of the CPS graph. When observing the CAS graph only, a
designer could erroneously conclude that the total error can be reduced with
approximately 30 nm by simply adding damping to the vibration isolation
system. Unfortunately however, the PSD would then only be reduced from
≈ 8.1 · 103 nm2 to ≈ 7 · 103 nm2 , resulting in an error amplitude of 83 nm, only
an almost negligible 9 nm below the previous value. This counter intuitive
effect is fully caused by "the root of the sum of squares".

8.2.5.1 Variations on dynamic error budgeting

The cumulative power spectrum can in principle also be determined by

starting the integration at higher frequencies and integrating towards lower
frequencies. This would result in a cumulative power spectrum with the
same rise magnitudes at critical frequencies as when starting integration
at low frequencies, so it would lead to the same conclusions. Based on
such a reversed PSD a cumulative amplitude spectrum would however look
completely different. In the example case of the inspection microscope the
resonance at 160 Hz would be even more emphasised and the effect at
2 Hz would almost disappear. Also this other approach shows that only the
cumulative power spectrum should be used for evaluating the impact of
different error sources.

8.2.6 Sources of noise and disturbances

The sources of disturbing signals in mechatronic systems all have either a

mechanical or an electronic nature while many of the observed phenomena
have a thermal root cause. Large scale deformations by heating can partly
be avoided by a careful design, that takes all expansion effects into account,
but ultimately always some thermal effects will remain, especially when
3 It should be noted that the word “Amplitude” in the CAS is erroneous as not the amplitude

of a signal is calculated, but the standard deviation or the RMS value in the case that the
mean value is zero.
552 Chapter 8. Measurement in mechatronic systems

the heat flow is not constant. Most probably this direct thermal disturbance
on system accuracy will always remain a limiting factor in the precision of
high performance mechatronics.
Thermal effects have their dynamic impact also on the small scale as they
are the root cause for the noise in electronic components like resistors.
The large scale deformations by thermal effects mainly occur in the low-
frequency area. In the following a short overview is given of some important
sources of higher frequency dynamic noise in the mechanical and electronic
domain:

8.2.6.1 Mechanical noise

The most important source of mechanical dynamic disturbances are vibra-

tions that either are caused by movement of the support, the floor vibrations,
or by forces from vibrating parts in the machine itself. Floor vibrations can
be caused by traffic, wind, earthquakes and other machines or moving people
in the neighbourhood. The power spectral density of floor vibrations can
consist of real uncorrelated random signal and systematic signals caused
for instance by rotating equipment. When the ultimate of precision is aimed
for, it is always necessary to analyse the vibrations on a certain location
with measurements over a representative time span, at least over a full
twenty-four hours period on a busy day. For production equipment like wafer
scanners it is not possible to adapt the machine to all different locations
and it is better to work with a certain standard specification. Figure 8.5.
shows the specification of an ASML wafer scanner in comparison with the
BBN criteria that were defined by the American company Raytheon BBN
Technologies.

8.2.6.2 Electronic noise

Electronic noise has many origins, which are named after their behaviour,
like thermal, shot, excess, burst and avalanche noise. The most relevant
sources of electric noise that impair precision mechatronic systems is pre-
sented in the following short overview:

Thermal noise

Any resistor will have a fluctuating potential difference across its ends that
is superimposed on the voltage caused by the current through the resistor.
8.2. Dynamic error budgeting 553

>\
10
ASML
BBN C
BBN D

>`

PSD [(m/s 2 )2 /Hz]

10

>{
10

>}
10 0 1 2
10 10 10
frequency [Hz]

Figure 8.5: Power spectral density of floor vibration specifications from ASML and
BBN.
(courtesy of Leon Jabben)

The fluctuating voltage is caused by the thermally induced random motion

of charge carriers like electrons. Thermal noise has a normal probability
density function and it has a flat power (white noise) spectral density, called.
In electric systems the energy is dissipated in the resistors. The noise from
a resistor can be described as a voltage source in series with the resistor,
with a power spectral density of:
2
NT = 4 kTR [V /Hz], (8.14)

with k the Boltzmann’s constant (1.38 · 10−23 J/K), T the temperature and
R the resistance. To give an example, a resistor of 1 kΩ, at 20◦ C will show
noise with a RMS value of 0.13 μV from 0 Hz up to 1 kHz.

Shot noise

Shot noise results from the random passage of individual charge carriers
across a potential barrier. This is often seen with junctions in a transistor.
The noise has a normal probability density function and has a white spectral
density:
2
NS = 2 qI DC [A /Hz], (8.15)
554 Chapter 8. Measurement in mechatronic systems

with qe the charge of an electron (1.6 · 10−19 [C]), I DC the average current
[A]. An average current of 1 A will introduce noise with an RMS value of
18 nA from zero up to one kHz.

Excess noise

The noise in excess of the thermal and shot noise when a current passes
through a resistor or a semiconductor, is called excess noise. Other names
are flicker noise or one-over-f (1/ f ) noise. This noise source results from
fluctuating conductivity due to imperfect contact between two materials.
This is the reason why carbon composition resistors, which are made up
of many tiny particles molded together, show more excess noise than wire
wound resistors. The power spectral density of excess noise increases when
the frequency decreases:
Kf 2
NE = [V /Hz], (8.16)
fα

where K f is dependent on the average (DC) voltage drop over the resistor
and the index α is usually between 0.8 and 1.4, and often set to unity for an
approximate calculation. For resistors the excess noise is proportional to
the average voltage drop V over the resistor, which is why manufacturers
typically specify the excess noise as a noise index CR for one frequency
decade:
σ∗V · 106
CR = [μV/V] (8.17)
V
with σ∗V the being the standard deviation over one decade frequency range
of the voltage. For standard resistors the noise index CR typically ranges
from 1 to 10. For example, if the noise index equals 10, an average voltage
drop of one volt introduces noise with an RMS value of 17 μV in a frequency
range from 1 up to 1000 Hz. Note that a frequency range 1 mHz up to 1 Hz
introduces an equal amount of noise because of the low-frequency range!

8.2.6.3 Using noise data from data-sheets

The given data on noise in different data-sheets of electronic components


are most frequently given in terms of [V/ Hz] as an RMS density instead
of the power density. This is done to be able to easily calculate the noise-
voltage contribution of one single electronic component. It is important to
emphasise here that these values should first be squared and then combined
8.2. Dynamic error budgeting 555

to determine the total power spectral density of the system.

Many errors have been made in the past by designers who either used the
RMS data in a PSD and got a very low disturbance level after taking the
root of the CPS. But also a non-statistical worst-case addition of the noise
voltages per electronic element gives erroneous results as the noise voltages
are mostly uncorrelated and as a consequence the worst-case addition gives
a too high value.
These errors either result in an unobserved problem that will pop-up in
the realisation of the mechatronic system when the real measurements will
unveil the noise or it will give rise to a too expensive design, because the
calculated worst-case noise value had to remain below the specification of
the total system.
In any case these errors can be avoided by using the method of Dynamic
Error Budgeting in the right way.
556 Chapter 8. Measurement in mechatronic systems

8.3 Sensitive signals in measurement systems

Most of the information exchange inside and between all elements in a

measurement system takes place in the electronic domain. Figure 8.6
shows a simplified electronic model of the measurement system. The input
impedance of every successive element determines the load for the output
of the preceding element. Depending on the different properties of each
element the requirements for the successive element can be quite extreme.
In most cases the output impedance of the sensing element is reasonably
high. A large output current with a high voltage would correspond with a
high power level that has to originate from the physical phenomenon that is
measured. This means that in general the input impedance of the successive
element should be either be as high as possible with a very small input
current or as low as possible with a very small input voltage. In both cases
the input power will be limited.
Also the power supply connections are shown. The power supply of the sens-
ing element is dashed because not all sensing principles need an external
power input. Still all these connections can and will cause errors of a mainly
interfering character, when not designed well.
This section will deal with the most sensitive part of the measurement chain,
the sensing element and its interconnections.

+ + +
Vp Vp Vp

Sensing 


 Signal processing
Zs Zo,a
+

+
+
+

A-D Data
Vs Vi,a Z Va = GVi Vi,p Z Converter
_ i,a _ i,p

_ _

Vp
- Vp
- Vp
-

Figure 8.6: The electronic model of a measurement system shows the connection
between the three elements and the power supplies. Every successive
element determines a load for the preceding element. The output of
the first two elements is shown in the Thevenin model with a voltage
source. Depending on the properties of the system also a current source
Norton equivalent model can be applied.
8.3. Sensitive signals in measurement systems 557

8.3.1 Sensing element

The sensitivity of a sensing element for a certain physical quantity is based

on the effect of that quantity on an electrical property of that sensor. Almost
by definition this does not exclude sensitivity for other physical phenomena
that will act either as an interfering input, as a modifying input or as a
combination of both. This is the first problem that a mechatronic designer
encounters, when choosing a suitable sensor. The most prominent example
of this phenomenon is the influence of temperature on almost every electri-
cal property of any sensor. Even when measuring the temperature itself,
the measurement is often influenced by temperature values elsewhere in
the measurement system.
The following non exhaustive list shows some representative sensing prin-
ciples. They can be distinguished into two different operation principles.
The first principle relates to the capability of a sensor to directly generate
an electrical signal by converting energy from the physical phenomenon
into electric energy. The second principle is based on the variability of an
electrical impedance by the physical phenomenon.
Examples of direct generation of an electrical signal are:

• The voltage difference between different metals to measure tempera-

ture differences.

• The voltage induced in a coil by a changing magnetic field, proportional

to the relative velocity.

• The voltage induced over a piezoelectric crystal by deformation due to

a force.

• Electrons in the depletion layer of a diode that are excited by photons

of the incident light and create an electric current.

Examples of a variable electrical impedance are:

• The change of resistance of a resistor by temperature or strain.

• The change of the capacitance value of a capacitor by a displacement

of the electrodes.

• The change of the self inductance of an inductor by a change in the

magnetic geometry.

Not all sensors are capable of directly converting a physical quantity into
electricity. They first need to create one of the primary effects from the list
558 Chapter 8. Measurement in mechatronic systems

above by additional preceding steps. Take for example the measurement

of Force. Force is a physical quantity that can not be directly measured as
it manifests itself either as the acceleration of a mass or by deformation of
material. This means that for measuring a force, it is necessary to measure
either the corresponding acceleration or the deformation. Later in this
chapter it will be shown that acceleration is in most cases measured by
measuring the deformation of a material that is caused by the force that is
needed to accelerate a known seismic mass.

8.3.2 Converting an impedance into an electric signal

In the previous section it was shown that sensing principles can be based on
a changing electrical impedance. Before this information can be treated in
the same way as a directly generated electrical signal, it is first necessary
to convert this impedance into an electrical signal4 .
The most simple way to achieve this conversion is by applying Ohms Law
and either supply the impedance with a well-known current and measure
the resulting voltage over the impedance or do just the opposite by applying
a voltage source and measuring the current. An AC source must be used
with a complex impedance like a capacitor or an inductor, but apart from
that the reasoning is the same.
Although modern electronics can be made rather precise, often special mea-
sures are needed to prevent errors that are related to the the low sensitivity
of many sensing impedances. When a potentiometer is used to measure
the displacement of the slider, the output voltage ranges from the supply
voltage Vi to zero. This is a sufficiently large signal, when compared with
interfering signals, and generally no further measures are needed. Another
example of a very sensitive resistive sensor is the NTC, a resistor with a
negative temperature coefficient based on semiconductor properties. Its
resistance can for instance range between 50 and 1000 Ω for a temperature
change from 0 – 100 ◦ C. With 1 mA current this results in a sensitivity of
approximately 10 mV/◦ C, which is also sufficient for most practical measure-
ments.
Unfortunately most variable impedance sensors have a much lower sensitiv-
ity and the noise in the supply current or voltage will play a much larger role
with those sensors. Take for instance a strain gage, a resistor to measure the
strain in a material by an increase of its resistance, when elongated. With
4 Often this conversion is considered as a part of the signal conditioning element. This is

however not logical with the aforementioned definition of the sensing element
8.3. Sensitive signals in measurement systems 559

Z3

+
Z1

3
_
+

Vp Vp

+
Vo V1  V2

+
Z2 V1 Z4

+
V2

+
_ _

4
2
V1 V2
_ _ _ _

Figure 8.7: A Wheatstone bridge has two branches each consisting of a voltage di-
vider. Only the differential voltage between the outputs of both branches
is used as measurement signal. Frequently the Wheatstone bridge is
drawn in a tilted square configuration as shown right to emphasise
the differential output voltage Vo . When all resistors are equal the
differential voltage is zero and the common-mode voltage is half the
supply voltage.

those elements, the practical resistance change upon load δR/R is often less
than 1 %. When this strain gage is supplied with a constant current source
I p , it is very difficult to keep the noise in this source below 10−6 × I p . With
an average value of R = 100 Ω, this noise causes an error with a constant
RMS value in the measurement of 10−6 × 100 Ω. With δR/R ≤ 0.01, the signal
to noise ratio is less than 104 and because the error signal has a constant
RMS value this is unacceptable for any precision measurement system.
Also the temperature has a large effect on most resistive sensors and this
influence directly interferes with the resistance change of the measurement.

8.3.2.1 Wheatstone bridge

To reduce the negative effects associated by the noise in the voltage source
of impedance measurement electronics, sensors like strain gages are mostly
connected in a Wheatstone bridge, named after the British scientist Sir
Charles Wheatstone (1802 - 1875), who popularised its use after the inven-
tion by the British scientist Samuel Hunter Christie (1784 - 1865).
The thinking model behind the usefulness of the Wheatstone bridge is best
explained by means of Figure 8.7. In principle the bridge consists of two
voltage dividers that share the same voltage source. These voltage dividers
560 Chapter 8. Measurement in mechatronic systems

are also called the two branches of the Wheatstone bridge. The first branch
consists of Z1 and Z2 with a corresponding output voltage V1 . Similarly
the second branch consists of Z3 and Z4 with output voltage V2 . The total
measurement signal is obtained by using an ideal differential amplifier to
amplify only the difference voltage Vo = V1 − V2 without loading the bridge.
The benefit of this configuration becomes clear when starting in the situa-
tion that all four resistors are equal. This would result in equal voltages
V1 = V2 = 1/2Vp and the voltage difference Vo would then be zero. This is
called an ideally balanced bridge and in that case the influence of the power
supply voltage Vp would be cancelled in the differential voltage. It would
only create a common-mode noise voltage and this will be rejected by the
differential amplifier.
As soon as one of the resistors, for instance Z2 , changes its value, it will
cause V1 to change, resulting in a non-zero voltage difference Vo . In this
new situation the noise of the voltage source will also be observed in the
differential voltage signal but at a much smaller value, proportional to
the measurement signal. With the same sensor and source voltage of the
previous example, this would result in a relative RMS noise level of 10−6 of
the signal value, which is far better than the previously obtained constant
RMS noise level of 10−4 of the maximum signal value.
In theory it is not really necessary that all resistors are equal to realise a
balanced bridge, as long as V1 remains almost equal to V2 . This condition
is always met, when the ratios of both branches of the Wheatstone bridge
are equal ( Z1 : Z2 = Z3 : Z4 ). For example one of the branches can consist
of resistors with much higher values than those in the other branch, like
Z1 = Z2 = 2 kΩ and Z3 = Z4 = 200 Ω.
In principle also Z1 can be different from Z2 as long as their ratio is equal
to the ratio between Z3 and Z4 but that is in most cases not a preferred
situation. With most sensors the impedance variation is only a small fraction
of the nominal impedance of the sensor, proportional to the measurand i ,
so δ Z/Z ∝ i  1. For those sensors a maximum sensitivity is obtained when
Z1 = Z2 and Z3 = Z4 .
This statement can be proven by calculating the sensitivity of for instance
V1 as function of its measuring impedance Z2 = R + δR for different values of
the other impedance Z1 = xR . With δR/R = c m i , where c m is a constant, δR
can then also be written as R c m i and the voltage V1 becomes:
R + δR R (1 + c m i ) 1 + cm i
V1 = Vp = Vp = Vp (8.18)
xR + R + δR R ( x + 1 + cm i) x + 1 + cm i

The sensitivity can be approximated as mainly linear because c m i  1 and

8.3. Sensitive signals in measurement systems 561

0.25


'
"‚cs )
0.2

0.15

0.1

0.05

0
0 1 2 3 4 5 6
~ x

Figure 8.8: The sensitivity factor of a wheatstone bridge is maximum when the
impedance values in each branch are equal ( x = 1), resulting in a working
point of half the supply voltage.

the corresponding value of the linear gain K  becomes equal to the derivative
of V1 over i :
dV1 ( x + 1 + c m i )( c m ) − (1 + c m i )( c m )
K = = Vp
di ( x + 1 + c m i )2
xc m x
= Vp ≈ Vp c m = Vp c m c s (8.19)
( x + 1 + c m i )2 ( x + 1)2
with c s being the sensitivity factor of the Wheatstone bridge. The mentioned
approximation is allowed when c m i  1. The maximum sensitivity is found
when the derivative of c s over x becomes zero:
d c s ( x + 1)2 − x(2 x + 2) ( x + 1)2 − 2 x( x + 1) 1− x
= 4
= 4
= (8.20)
dx ( x + 1) ( x + 1) ( x + 1)3
This derivative is zero for x = 1 and x = ∞, but the second value of x also
gives a zero value for the sensitivity K  . This means that only x = 1 is a
useful answer, proving the statement that the values of the impedances in
each branch of the Wheatstone bridge should be approximately equal for a
maximum sensitivity.
This conclusion is further illustrated in Figure 8.8, emphasizing the non-
linear relation between the sensitivity and the ratio of the impedances. Only
in a small area around x = 1 the sensitivity is almost constant.
As a last variation on the Wheatstone bridge, the two branches might differ
in the kind of impedance that is applied. For instance one branch can
consist of two resistors, while the other branch can consist of two complex
562 Chapter 8. Measurement in mechatronic systems

+
Vp Vp Vp Vp
R R R R  R RR R RR R  R

1
3

3
Vo _ Vo _ Vo _ Vo _
+

+
4

4
2

2
R RR R RR R RR R  R RR

a: Single b: Double, opposite sign c: Double, same sign d: Two double, opposite sign

Figure 8.9: Four configurations of a Wheatstone bridge with maximum sensitivity,

showing the different configurations that can be chosen when a mul-
tiple of variable impedances are available. Versions a: and c: suffer
from a small non-linearity and both are not temperature compensated.
The other two versions are both linear around the working point and
temperature compensated.

impedances of the same kind, like two capacitors or two inductors. Other
combinations with for instance only one capacitor are not allowed as then the
output voltages would show a relative phase shift for AC voltages because
of the filtering. This would result in a measurable voltage difference with a
corresponding noticeable noise level, even without measurement signal.
The application of complex impedances will be demonstrated more in detail
in Section 8.6.2 and thereafter with the presentation of capacitive and
inductive proximity sensors.

Temperature compensation and linearisation

When all resistors in a balanced resistive Wheatstone bridge share the same
temperature and the same temperature sensitivity, the output voltage is
also no longer affected by the temperature. This seems attractive at first
sight, but these conditions are not easily met, because the temperature
of all elements are hardly ever equal. The not-sensing impedances of the
bridge are in most cases located inside the measurement instrument, at
some distance from the measurement location. In principle the requirement
for an equal temperature can be alleviated a bit as only each branch needs
to be isothermal. This is based on the reasoning that in that case the ratio
of both branches is not affected by the temperature. This means that one
of the branches can remain inside the measuring instrument at an equal
temperature, while the other branch is fully located at the measurement
site.
8.3. Sensitive signals in measurement systems 563

When possible, this second element in the sensor branch of a Wheatstone

bridge should also be used for active measuring the phenomenon of interest.
As will be demonstrated in the following this second sensor would be opti-
mally applied, when it gives an opposite change of its impedance with an
equal magnitude as the first sensor impedance. To investigate this effect,
first the sensitivity of the full Wheatstone bridge is written down. The
bridge is assumed optimal in respect to sensitivity so all impedances are
as much as possible equal. Using the notation for the impedances from
Figure 8.9, the output voltage Vo is given by the following generic equation:

Z2 Z4
Vo = Vp − (8.21)
Z1 + Z2 Z3 + Z4

With the given values, the output voltage of the single sensor version with
Z4 = R + δR and Z1 = Z2 = Z3 = R becomes:

1 R + δR
Vo = Vp − (8.22)
2 2R + δR

After combining the terms with a common denominator, this equation can
be written as:
δR δR
Vo = −Vp ≈ −Vp (8.23)
4R + 2δR 4R
The approximation is allowed as long as δR  2R . In the not approximated
equation the δR term in the denominator causes a small non-linearity in
the sensitivity. Also the thermal effects in Z4 are not compensated.
The situation becomes however quite different with two sensing elements
with an opposite sign of the impedance change. This is shown in the second
drawing of Figure 8.9, where the second sensing impedance is located at
the other impedance of the same branch, so Z3 = R − δR . With this double
sensing principle the output voltage becomes:

1 R + δR δR
Vo = Vp − = −Vp (8.24)
2 2R 2R

This equation is fully linear with a double sensitivity, when compared to

the single sensing principle and also the thermal effects are compensated
as both sensing elements are located at the measurement site.
It should be noted that this linearity is limited to the optimal working point
with all resistors approximately equal.
At first glance the second sensor could also be located in the other branch
at Z2 . With very small impedance changes this indeed would give approx-
imately the same effect on the differential voltage but the non-linearity
564 Chapter 8. Measurement in mechatronic systems

would not be cancelled as can be seen by filling in the data:


R − δR R + δR 2R δR δR
Vo = Vp − = −Vp ≈ −Vp (8.25)
2R − δR 2R + δR 2
4R − 4R δR + (δR )2 2R

In this case the thermal effects are still compensated as these work equal in
both branches but the non- linear terms in the not approximated equation
still underline the optimal location of both sensors in the same branch.
The negative effect of using two sensors in different branches is also demon-
strated in the third configuration of Figure 8.9. This configuration is unfor-
tunately the only option, when the second sensing element has the same
sign for its sensitivity as the first sensing element. In that case for instance
Z1 and Z4 are replaced by the two equally acting sensing impedances R + δR .
The calculation of the output voltage gives the following result for this
situation:
δR δR
Vo = Vp ≈ Vp (8.26)
2R + δR 2R
Although the sensitivity is also approximately twice the value of the single
sensing principle, the non-linearity is not cancelled. The main disadvantage
is however that the thermal problem is not solved, because of the different
location of the impedances in their branch. A simultaneous equal change
in the impedance by the temperature will cause a differential interfering
voltage adding to the voltage of the useful measurement signal.
It should be noted that the same result but with a different sign would be
obtained when Z2 and Z3 are replaced by the two sensing elements instead
of Z1 and Z4 .
Thinking further on this path, it appears that the most optimal configuration
is created, when all four impedances of the Wheatstone bridge would be
replaced by sensing elements. This results in a total signal of:
R − δR R + δR δR
Vo = Vp − = −Vp (8.27)
R + δR + R − δR R + δR + R − δR R
In this configuration, both the temperature effects and the impact of the
noise of the source is minimised, the sensitivity is maximum, the sensor is
linear around the working point and only 4 connections are needed to the
measurement electronics.
It will be shown in Section 8.7.1 how this can be accomplished in an inte-
grated strain gage element.
Remark on non-linearity: The argumentation regarding non-linearity
has become less important over the years with the introduction of digital
8.3. Sensitive signals in measurement systems 565

data processing. In principle non-linearity is a deterministic error and it

can be compensated with the described model. It is certainly true to say
that compensation is as good as the model is described and nothing is ideal
so “prevention is still always better than curing”, but in many situations a
less optimal bridge configuration can be compensated to an acceptable error
level.

8.3.3 Electronic interconnection of sensitive signals

Like mentioned before, in most cases the electrical signals that are gener-
ated in a sensing element, are still susceptible to interfering signals from
external sources of energy, because of their small amplitude and high source
impedance. The sensitivity for these disturbances demands a careful con-
sideration of the interconnection from the sensing element to the signal
conditioning element. Although this interconnection is the most critical in
a measurement system also the other connections can cause problems but
first the purely analogue most sensitive interconnection will be examined.
All elements have to be interconnected with minimal two wires, because
electricity flows in a loop, and both wires are susceptible to interference.
Figure 8.10 gives an overview of the interconnection between two elements
of a measurement system with the different external disturbance sources
that can act on this interconnection.

8.3.3.1 Magnetic disturbances

The first source of disturbances is related to voltages that are induced in a

closed electric loop by changing magnetic fields from power transformers and
high currents in nearby electronic circuits. For both wires these voltages
have mostly a different value. The reason is that their corresponding loop
+ −
encloses a different surface, resulting in a voltage difference Vn,s − Vn,s that
is added to the output voltage of the first element. When these voltages
would be equal, the differential voltage would be zero and only a common-
mode disturbance input voltage would be present at the next element. This
common-mode voltage was defined in Chapter 6 with the operational ampli-
fier and represents the average value of both input voltages. This impact
of a common-mode voltage can be reduced by using a differential amplifier
with a high “common-mode rejection ratio” in the following element.
Unfortunately it is impossible to have the two wires follow exactly the same
path as then they would make contact and create a short circuit. It is also
566 Chapter 8. Measurement in mechatronic systems

+
_
Vn,c
Capacive interference Magnec interference

Sensing Vn,s+
Signal condioning

+
_
Zs Z c1 Zo

+
d's

+
+

dt Vi
Vs Z4 GVi
_ _
Signal loop
_

+
_
Z c2
Vn,s-
Zg1 d'g Zg2
Ground loop
dt

1 Ground 2
_ Vn,g
+

Figure 8.10: Overview of different sources of interfering signals acting on the con-
nection between two elements in a measurement system. Changing
electric fields insert a capacitive current in the connecting cables and
changing magnetic fields induce a voltage in closed loops. Grounding
at different places includes the voltage differences of these places in
the measurement.

not possible to isolate a sensitive system from magnetic fields as no mate-

rials exist with a relative magnetic permeability μr of zero that would be
needed to lead magnetic fields away from the sensitive electronics. Magnetic
fields can only be reduced to some extent by fully closed thick layers of fer-
romagnetic material for low-frequency magnetic fields and by a fully closed
conductive shielding for high-frequency magnetic fields (eddy-currents), but
that solution is very unpractical in many situations.
To solve this magnetic interference in a more practical way, two methods are
used. For less critical situations the Unshielded Twisted Pair (UTP) cable
can be used as shown in the upper drawing of Figure 8.11. A homogeneous
changing magnetic field will induce an electromotive force in each loop
that encloses a part of this field, according to Faraday’s law. As long as
every twist encloses the same magnetic flux an equal electromotive force
is induced. Every twist is succeeded with a next twist where the wires are
exchanged from position resulting in an exchanging positive and negative
electromotive force in each twist for both wires. These electromotive forces
8.3. Sensitive signals in measurement systems 567

+ + + + +

+ + + +
d'
Unshielded twisted pair (UTP) dt
~!
Shielding, return signal or ground
Signal

Coaxial cable
Grounded shielding

Shielded twisted pair (STP)

Figure 8.11: Interference by changing magnetic fields can be reduced by twisting

the two wires of a signal cable. For each wire the electromotive force
generated in one twist of the cable is directed opposite to the electro-
motive force generated in the adjacent twist, as shown for the blue
wire. Because of the series connections these electromotive forces
result in a net potential difference of ≈ 0 V over the total twisted pair.
This beneficial effect is maximum when the magnetic field is uniform.
Even better is the coaxial cable as shown in the middle as then at all
places both a positive and negative loop is present. When capacitive
disturbances and ground loops cause problems, the shielded twisted
pair is the preferred solution.

result in a total potential difference at the end of the wires of approximately

zero Volt, because of the series connection.
Unfortunately often the magnetic fields are not really homogeneous and
it is also difficult to keep all twists equal. For that reason the coaxial
cable is frequently used as shown in the lower drawing of Figure 8.11.
In principle the return path of the signal is a tube around the central
conductor that carries the signal voltage relative to the grounded tube. A
magnetically induced electromotive force at one side of the coaxial cable will
be compensated by the induced electromotive force at the opposite side and
as long as the resistance in the shield is low this compensation is almost
perfect. The word almost is due to the possibility that the magnetic field
might not be homogeneous over the cross section of the coaxial cable. A
thinner cable is better in that respect but as a side effect the capacitance of
the cable would be larger.
568 Chapter 8. Measurement in mechatronic systems

š
'
"

+
_
Vn,c

–!   


Vn,s+

+
_
Zs Z c1

+
Zi

+
+

Vs Vi Z4 GVi
_ _
Signal loop
_

+
_
Z c2

Sensing 




Figure 8.12: Capacitive disturbances can be reduced by electrostatic shielding. A

Faraday shield surrounds the entire sensitive part with conductive
material that is either grounded or connected to another trusted con-
stant voltage source. The multitude of parasitic capacitances between
the shielding and the instrument will then hardly cause problems,
because of the very low relative potential difference.

8.3.3.2 Capacitive disturbances

The second source of disturbances is capacitive coupling of electric fields,

originating from for instance mains power supply lines and high-frequency
circuits of electronic equipment. In principle always some capacitive cou-
pling exists between any electrical wire that carries an alternating voltage
and a wire that carries a sensitive signal. Even with small distances and
large distance a voltage of 230 V @ 50 Hz creates detectable disturbance
values in the mV to μV range in a sensitive, high-impedance electronic
circuit as can be observed by touching the input of an audio amplifier by
hand.
In principle the capacitive coupling can be approximated as a current source
because of the high impedance of the small capacitance value between the
source and the sensitive element. This means that the level of the resulting
voltages depends on this capacitive coupling with the source impedance of
the sensing element and the input impedance of the next element. Often
these impedances are both quite high.
Likewise with the magnetic coupling, the capacitive coupling creates volt-
ages in both wires with a common-mode disturbance part that can be can-
celled by using a differential amplifier in the next element. Also in this
8.3. Sensitive signals in measurement systems 569

case the situation for both wires is often different because of the different
locations of the wires.
Fortunately it is very well possible to shield the sensitive element by means
of a conductive layer, creating a Faraday shield around the element that
blocks out electric fields as shown in Figure 8.12. This shield can be con-
tinuous but also small perforations are often allowed for cooling purposes
depending on the disturbance frequency. Very high frequencies can however
even penetrate in very small holes, depending on their wavelength!
Capacitive coupling can cause a problem with coaxial cables. In principle
the signal wire is well shielded by the return wire but this return wire
receives all capacitive disturbance signals and the resulting voltage appears
at the corresponding input of the next element. A first solution is often to
choose that input as the grounded input of the measurement system, short
circuiting these disturbance currents to a constant reference voltage. This
is mostly sufficient for less critical situations but sometimes the impedance
Zg of this zero Volt ground point is not low enough, resulting in a small
detectable disturbance on the input.

8.3.3.3 Ground loops

An even worse situation occurs when the two sides of the connecting cable
are both grounded. At first sight one might think that this will short circuit
the capacitive disturbances better but this additional connection creates a
ground loop with a very low impedance, where magnetic disturbances will
create a voltage that induces significant currents in the loop impedance.
Reducing Zg,1 would even make things worse, because the increased current
by the reduced loop impedance will cause an increased voltage over Zg,2 .
These negative effects are even present inside an electronic instrument,
where the loops are small. Experienced electronic designers are very keen
on avoiding ground loops even on printed circuit boards, by grounding all
electronics at one location.
Especially when the distances between the elements in a measurement
system become large like in a chemical plant, with cables up to hundreds of
metres another problem becomes prominent, the difference in potential of
the ground at different locations. This potential difference can be caused by
leakage currents of large grounded high-power electrical systems, lightning
and magnetic fields.
For these reasons double grounding should in principle always be avoided.
Safety regulations however often require the housing of mains fed electronic
570 Chapter 8. Measurement in mechatronic systems

Zs Z c1
Signal
Z4
Sensing condioning

+
_
Z c2
Vn,s-
Zg1 d'g Zg2
Ground loop
dt

1 Ground 2
_ Vn,g

+
Figure 8.13: Grounding more than one element of a measurement system will cre-
ate ground loops that introduce common-mode magnetic disturbances
and voltages due to other sources. Even though differential amplifiers
can reduce the problem, the related voltage levels can be very high,
especially with large distances, so ground loops should be avoided.

systems to be connected to the safety ground. In those cases the internal

electronics are preferably not connected to this grounded casing but often
this connection is present for practical reasons. And even without an internal
hard wired connection, the capacitance between the electronics and the case
creates a capacitive grounding for high frequencies.
To overcome these problems a combination of the coaxial cable and the
twisted pairs can be used, the Shielded Twisted Pair (STP). The two wires
of the twisted pairs conduct the signal including the return path, like with
the unshielded version. By very small and well controlled twists, firmly
embedded in the insulation, the magnetic disturbances are minimised, while
the shielding has no other function than to cancel the electric fields.
In Section 8.5.4 the optical fibre is presented that can be used when the
signals are conditioned to a higher signal energy and digitised to a number.
The use of an optical fibre prevents any galvanic connection with the related
ground loops currents.
8.4. Signal conditioning 571

8.4 Signal conditioning

It is preferred to first add robustness to the electrical signal by placing a

signal conditioning element close to the sensing element because of the
often very high sensitivity of the sensing element. When possible, the very
best solution is created when this element is integrated with the sensor.
As mentioned in the previous section, signal conditioning elements prefer-
ably consist of an ideal differential amplifier with an infinite common-mode
rejection ratio without offering a load to the sensing element. Furthermore,
the signal conditioning element can also contain filters and modulation
principles to reduce the impact of interference.
In this section first the instrumentation amplifier will be presented as a
true differential amplifier, followed by a presentation of filtering with a focus
on modulation techniques.

8.4.1 Instrumentation amplifier

Generally the input of a signal conditioning element consists of one or more

operational amplifiers. In case of a less critical situation a non-inverting
amplifier can be sufficient, when only the impedance of the source is too
high for transporting the signal.
When the sensing element acts like a current source as will be shown
later with a light sensitive diode, an inverting amplifier with a low input
impedance can be applied to convert the signal to a voltage with a low output
impedance.
As explained in the previous section however, in most cases a high common-
mode disturbance voltage is present in the signal, due to capacitive coupling
and ground-loops. In those cases a true differential amplifier is necessary to
cancel this common-mode signal. Also in case of a sensor with a Wheatstone
bridge a differential amplifier is needed. Unfortunately however the simple
differential amplifier that is based on a single operational amplifier as shown
in Figure 8.14 has the disadvantage of a difference of loading of the two
inputs. The non-inverting input has a simple resistive input impedance, but
the current in the inverting input depends both on R1 and on the voltage at
the + terminal of the operational amplifier. This voltage at the + terminal
is equal to the non-inverting input voltage and as a result, the current at
the inverting input depends on the voltage at the non-inverting input. This
would be no problem when both Vi(1) and Vi(2) are delivered by a source with
a source impedance that is significantly lower than R1 . In many sensor
572 Chapter 8. Measurement in mechatronic systems

I2
R2
I1
Inverng R1 _

+
G ()

+
Non +
R1
inverng

+
Vi,1 Vi,2 R2 Vo

_ _ _

Figure 8.14: A simple differential amplifier with one operational amplifier suf-
fers from different loading of the input signals. The current of the
non-inverting input is determined by R1 + R2 , while the current at
the inverting input is determined by R1 and the voltage at the non-
inverting input.

elements this is however not the case and the input impedance has to be
increased by adding additional buffer amplifiers before the actual differential
amplifier.
Another reason for introducing additional amplifiers is the need to apply
only small resistor values, because of noise. The noise in a resistor depends
on the resistor value and the temperature as was shown in Section 8.2. Also
the input-current noise of an operational amplifier results in a noise voltage
at its output that is determined by the feedback resistor value. In practice
both inputs of an operational amplifier should “see” a resistive impedance
of less than 1 kΩ to achieve the minimum noise level and such a low value
of the load impedance is not preferred for many sensing elements.
A third reason to adapt the simple differential amplifier configuration is
based on the common-mode rejection ratio. In practice it is very expensive
to apply resistors with a better tolerance than 0.1 %. As a consequence,
with these resistor tolerances, the amplification of the inverting and non-
inverting input of the differential amplifier will also differ approximately
this same amount. As a result. the common-mode amplification will be
only a factor 103 below the differential-mode amplification, which is equal
to only −60 dB. This is far worse than the standard CMRR of an operational
amplifier, which is often better than −100 dB.
Figure 8.15 shows an amplifier that uses two additional operational ampli-
fiers at its inputs to solve these problems. This configuration is called an
8.4. Signal conditioning 573

I
R1,2

_ I
_ G ()
+
A2
+

R3

RG I
Vi(d) R2 _

+
I
Vb(2) G ()

+
Vi(2) R1,1 R2 +

+
I
A3
_
Vo
+ +
G () Vb(1) R3

A1
+

Vi(1)
_ _ _ _ _

Figure 8.15: An instrumentation amplifier consists of two non-inverting amplifiers

with a standard differential amplifier and one resistor RG to set the
differential gain. The indicated current direction corresponds with
the indicated sign of the input voltage Vi(d) .

instrumentation amplifier, a name based on its main application in sensitive

measuring instruments.
Instead of simply using two voltage follower buffer amplifiers, the input
amplifiers are configured as high-gain non-inverting voltage amplifiers in
order to improve the common-mode rejection ratio. The second difference
is the resistor RG between the negative inputs of the two non-inverting
amplifiers that replaces the normally used resistors from the negative input
of each operational amplifier to ground.
The instrumentation amplifier has the following working principle.
All operational amplifiers have negative feedback. Under normal conditions,
the operational amplifier will do whatever it can do to adapt its output
such that the minus input of the matches the plus input. As a consequence
the voltage over RG will match the differential input voltage Vi(d) of the
two non-inverting amplifiers. When the input currents of the operational
amplifier are neglected, the outputs of both amplifiers will deliver a current
through RG , R1,1 and R1,2 , equal to I = Vi(d)/RG . When R1,1 = R1,2 = R1 , the
574 Chapter 8. Measurement in mechatronic systems

intermediate voltages after the non-inverting amplifiers equal:

R1
Vb(1) = Vi(1) + IR 1 = Vi(1) + Vi(d) (8.28)
RG

and:
R1
Vb(2) = Vi(2) − IR 1 = Vi(2) − Vi(d) (8.29)
RG

This means that the common-mode voltage of Vb(1) and Vb(2) is equal to
the common-mode voltage of Vi(1) and Vi(2) while the differential voltage
Vi(d) = Vi(1) − Vi(2) is amplified with a factor R G/R 1 for both Vb(1) and Vb(2) .
As a consequence the common-mode rejection ratio of this combination is
improved with this factor RG/R1 . For that reason most of the gain of an
instrumentation amplifier should be realised in these non-inverting input
amplifiers, while the resistors of the differential amplifier part are often
chosen to be equal resulting in a unity gain of the second part. With this
equal resistor setting the output voltage becomes equal to:

R1 R1
Vo = Vb(1) − Vb(2) = Vi(1) − Vi(2) + 2Vi(d) = Vi(d) 1 + 2 (8.30)
RG RG

Fully integrated versions of such instrumentation amplifiers exist, where

only the external gain-setting resistor RG must be added to realise a fully
robust signal conditioning amplifier.

8.4.2 Filtering and modulation

Even with a careful design it is not always possible to reduce all sources
of random errors to an sufficiently low level. In that case it is preferred
that these errors are first reduced by filtering, before the signal is further
processed in the measurement system. Simple filtering can be very use-
ful if the disturbance signal has a different frequency spectrum than the
useful signal. In that case the active filter configurations can be used like
described in Section 6.2.7 of Chapter 6. Examples are the reduction of
the high-frequency part of wide-band noise of electronic components and
single frequency disturbances can be filtered by a selective band reject filter.
The second-order “taming” low-pass filter in the PID-controller that was
presented in Chapter 4 also serves this same purpose.
Unfortunately often the disturbance signal occurs in the same frequency
band as the measurement signal. This is especially the case with low-
frequency excess noise because most mechanical measurements have also
8.4. Signal conditioning 575

Synchronous square wave carrier signal

Modulator Disturbance Demodulator

+
”
< ' ”
<

+
+

 


+



+
_ _

Vi Va Vg Vo
_ _ _ _

Vi
Va

Vg

Vo
t1 t2 t3 t4 t5 t6 t7 t8 t9 t10
Time

Figure 8.16: Synchronous modulation and demodulation is achieved by alternat-

ingly switching the signal input Vi between the inverting and non-
inverting input of a differential amplifier. The resulting high-frequency
signal Va is amplified by a selective amplifier, resulting in Vg . A de-
modulator, switching synchronous with the input switch, inverts the
amplified signal again into a low-frequency signal Vo . The disturbance
is cancelled in the selective amplifier.

a strong low-frequent component in their spectrum. This can be solved by

combining the measurement signal with a high-frequency carrier signal
and only amplify the resulting high-frequency combination. This combining
action is called modulation and several types of modulation are applied. For
measurement systems, amplitude modulation (AM) is frequently used. It is
well-known from radio transmission and will be explained more in detail.
Other principles include frequency modulation that is also used in radio
transmission, pulse-width modulation that was introduced in Chapter 6 for
power amplifiers and phase modulation that will be presented in Section 8.8
with incremental optical measurement systems.

8.4.2.1 AM with square wave carrier

One principle of amplitude modulation is shown in Figure 8.16. The high-

frequency carrier signal is a square waveform, alternatingly switching the
input signal between the inverting and non-inverting input of a differen-
576 Chapter 8. Measurement in mechatronic systems

tial amplifier. The output of the modulator is a high-frequency amplitude

modulated square wave signal and can be filtered below the carrier wave
frequency without losing information. This filtering is done by means of
a selective amplifier with a high-pass frequency transfer function. As a
result, disturbances at the connection between the differential amplifier and
the selective amplifier will not arrive at the input of the demodulator. The
demodulator consists of a similar combination of a switch and a differential
amplifier as the modulator, restoring the original sign of the signal input
by switching synchronous with the modulator. This special demodulation
method is called synchronous demodulation because of the synchronous
switching of the modulator and demodulator.
Because of the sharp transients of the modulated square wave signal, the
amplifier has to be able to preserve the very high frequencies of the square
wave spectrum. For that reason, the use of this version of amplitude modu-
lation is restricted to relatively low-frequency measurement signals.
For higher frequencies a sine wave carrier frequency is preferred. Instead
of switching, the modulation with a sine wave carrier frequency is achieved
by multiplication. In fact also the switching modulator acts like a multi-
plier of the momentary magnitude of the signal input with the momentary
magnitude of a square wave signal.

8.4.2.2 AM with sinusoidal carrier

In radio transmission the sine wave carrier signal is multiplied with a com-
bination of a DC bias voltage and the music signal, as shown in Figure 8.17.
This method has the advantage that demodulation is fairly easy and for that
reason it is also useful in measurement systems, when the measurement
signal does not contain a DC value.
As a first step, the measurement signal is added to the DC bias voltage,
large enough to keep the voltage always unidirectional. By multiplying this
combined DC+AC signal voltage with the carrier signal, the resulting high-
frequency signal will never get a zero amplitude. This modulated signal can
be amplified with a selective amplifier with a band-pass transfer function
around the carrier frequency.
Demodulation is achieved by simple rectification, followed by a second band-
pass filter around the measurement signal frequency. This filter cancels the
DC bias voltage and the high-frequency remains of the carrier frequency.
To determine the transfer function of the selective amplifier it is useful to
look at the simple mathematics around this modulation principle. When the
8.4. Signal conditioning 577

Modulator Interference Demodulator

!
 ' … 
x 
 

+
+

+
+

+

Vi Va Vg Vr Vo
Vc
_ _ _ _ _

Vi
Vc

Va
Vg

Vr
Vo



Figure 8.17: Amplitude modulation by multiplying the input signal with the carrier
signal. The input signal consists of the measurement signal and a
DC bias voltage. As a result, the amplitude of the multiplier output
will never be zero. After amplifying and filtering, the demodulation is
done with a simple rectifier and a band-pass filter.

input signal with the DC bias voltage equals Vi = V̂i (1+sin ωi t) = Vi (1+sin 2π f i t)
and the carrier signal equals Vc = V̂c sin(ωc t) = V̂c sin(2π f c t) the multiplication
results in:

Va = Vc Vi = V̂c V̂i sin(2π f c t)(1 + sin(2π f i t))


cos(2π( f c − f i ) t) − cos(( f c + f i ) t)
= V̂c V̂i sin(2π f c t) + (8.31)
2

This means that the resulting signal contains three frequencies, the carrier
signal frequency f c and two frequencies equally spaced at both sides of the
carrier frequency at a frequency difference equal to f i . These frequencies
are called the side-bands around the sampling frequency and they both
contain the information of the measurement signal at f i .
With this result it can be reasoned that an input signal that contains differ-
ent frequencies up to f max requires the bandwidth of the selective amplifier
to range from f c − f max to f c + f max . In principle one of the two side-bands
could be filtered out and restored with special measures afterwards, but
that procedure is too specialistic for the purpose of this book.
A drawback of this simple scheme with rectifier demodulation is the fre-
quency limitation to AC signals. The synchronous demodulation of the first
578 Chapter 8. Measurement in mechatronic systems

Interference
Modulator Demodulator
!
 ' !
 
x 
 x

+
+

+
+

+

Vi Va Vg Vr Vo
Vc Vc
_ _ _ _ _

Vi
Vc
Va
Vg

Vr
Vo



Figure 8.18: Synchronous modulation and demodulation with a sinusoidal carrier

signal enables the application of an input signal without a DC bias
voltage. The sign of the amplified measurement signal is retrieved
by a second multiplication in the demodulator with the same carrier
frequency.

example with a square wave signal did not have that drawback. Fortunately
it is also possible to apply synchronous demodulation with a sine wave
carrier signal, as shown in Figure 8.18. It results in the amplification of
frequencies from 0 Hz and above.
The input signal is not added to a DC bias voltage, because also any DC
content of the input signal should be maintained. The resulting signal after
modulation has a zero amplitude at Vi = 0 and shows a phase reversal, when
Vi < 0. Demodulation by rectifying would result in a frequency doubling
with a DC offset, so the demodulation is done by multiplication with the
same carrier signal.
The average of the resulting signal is equal in shape as the input signal
and can be recovered by a low-pass filter. This method is also called phase
selective detection or homodyne detection and will return later in this chapter
with different sensing principles.
8.5. Signal processing 579

8.5 Signal processing

After the signal conditioning element, a strong and robust signal is present
that can be used in an analogue control system. Modern control electronics
work predominantly in the digital domain, which makes it necessary to
convert the momentary values of the electrical signal into a sequence of
data that sufficiently represents the measured value.
Digital systems are characterised by the fact that they work with very
strictly timed events, synchronised by means of an accurate clock frequency
generator, mostly based on a quartz crystal oscillator.
When only the frequency of a periodic signal is important, the analogue
signals can be directly converted into the digital domain. This frequency-
to-digital conversion process is achieved by means of a comparator, a one
bit bistable element that outputs either a “zero” or a “one” depending on
the voltage level at the input. In chapter 6, a special version of such a
comparator, the Schmitt trigger, was used in the pulse-width modulator of
a switched mode power amplifier. In measurement systems the Schmitt
trigger is used to suppress disturbances.
With real analogue-to-digital conversion the magnitude of a value needs to be
digitised and that is done by taking a momentary value of the measurement
signal and generate a number to the magnitude of that sample. In the
following sections, first the Schmitt trigger will be presented, followed by
the process of creating sampled numerical data from the magnitude of a
measurement signal.

8.5.1 Schmitt trigger

A comparator is a differential amplifier with an infinite gain. The output

is either a fixed positive value when the plus-input is at a higher voltage
than the minus-input or a fixed negative value in the other case. A Schmitt
trigger is a comparator with a well defined hysteresis level.
Figure 8.19 shows the functionality of a Schmitt trigger with different
signals. When first examining the upper graph, without any hysteresis, the
output voltage Vo changes its value any time the input signal crosses the
0 V reference level. This configuration without hysteresis is only useable,
when the input signal would be without any noise, which is never reality.
The effect of noise is shown after t0 , where a noise signal is added. The
noise will cause additional output transitions at the zero crossing with a
higher frequency than the input signal. When examining the lower graph,
580 Chapter 8. Measurement in mechatronic systems

Vo +
Vi
Vo
Schmi trigger 0
t0 t
Vo
+
Vr
Vi Vo -
Vr - Vo +
Vi Vr +
Vo 0
Vr t0 t
Vr -
Vo -

Figure 8.19: A “Schmitt trigger” is a comparator with a defined level of hysteresis

that converts a periodic signal into a square wave with a reduced sen-
sitivity for noise. The upper right graph shows the different voltages
when no hysteresis is present, while the lower right graph shows the
voltages with a hysteresis level of Vr . The input signal is sinusoidal
and after t0 noise is added to show the difference. Note the phase shift
due to the hysteresis.

the influence of the hysteresis is clearly observed. Starting at a situation

where Vo is negative, the input signal rises until it surpasses the value of
Vr+ . At that moment Vo will transit to the positive side. Only when the
input signal drops below the now negative value of Vr− , Vo will transit to the
negative side again. This hysteresis has two effects. Firstly an additional
phase shift occurs between Vo and Vi depending on the amplitude of Vi and
the hysteresis level but the secondly effect is more important because the
frequency of Vo will not be influenced by the noise as long as the peak to
peak value of the noise remains below the hysteresis level.
Comparators in both inverting and non-inverting configurations are fre-
quently applied and even when not specifically mentioned, they mostly are
provided with some kind of hysteresis. Only when the hysteresis level is
clearly defined the comparator deserves its name as Schmitt trigger.

8.5.2 Analogue-to-Digital conversion

Digital values are represented in a binary format, because computers work

with binary logic circuits. The elementary digits of a binary number are
called bits, with only two values, zero and one. A binary number consists
of a summation of the value of n bits, with n being an integer > 0. In the
8.5. Signal processing 581

Digital output
Digital output

LSB

Analogue input Analogue input


€!‚•JQ\…„@
€!‚•Q…„

Figure 8.20: The incremental nature of digitisation introduces a quantisation error

that can range between maximum 0.5 to 1 times the least significant
bit depending on the method of quantisation.

mathematical sense, the value of bit k, where 0 ≤ k < n, is equal to 2k .

The bit with k = 0 is called the Least Significant Bit (LSB) and the bit with
k = n − 1 represents the Most Significant Bit (MSB).
When digitising measurement values, the magnitude represented by the
most significant bit is equal to half the total range of the measurement
value. The magnitude of the least significant bit represents the smallest
increment that can be distinguished between different measurement values.
This incremental property of digitisation determines the first source of
errors in a measurement system, the quantisation error as shown in Fig-
ure 8.20. Depending on the quantisation method, the error can amount up
to a maximum of one least significant bit (LSB).

8.5.2.1 Gray code

Although digital numbers offer in principle an unambiguous representation

of a value, in practice its application can cause problems. This is due to
the fact that with several incremental steps the change of one increment is
represented by a change of more than one bit. This effect is most significant
in a situation, where the value is changed from just below half the range to
just above half the range or the other way around. With for instance an eight
bit number, increasing with one increment this would mean a change from
01111111 to 10000000. If for some reason not all bits switch at the same
moment, this can lead to erroneous data. A well-known configuration, where
this situation can occur, is the direct digital measurement of an angular
582 Chapter 8. Measurement in mechatronic systems

Lig
‡
 
Decimal Gray Biary
šllimatr ls
0 0000 0000
1 0001 0001
2 0011 0010
š


3 0010 0011
4 0110 0100
5 0111 0101
6 0101 0110
7 0100 0111
 8 1100 1000
9 1101 1001
10 1111 1010
LSB 11 1110 1011
12 1010 1100
13 1011 1101
14 1001 1110
MSB 15 1000 1111
Pht tectr

Figure 8.21: With direct measurement of digital data, a binary code can cause
problems by not synchronised changing of all related bits. This is
solved by using a Gray code that switches only one bit at any increment,
like in this example of an absolute rotary optical encoder. At the right
a 4-bit coding is shown as example.
(Courtesy of Heidenhain)

or linear position by means of a multitude of light beams through holes

in a coding disk that represent the different bits. Figure 8.21 shows an
example of such a rotary measurement system. Instead of a binary code this
measurement system uses a Gray code, named after the American physicist
Frank Gray. The Gray code is designed such that only one bit is switched at
any incremental change. As shown in the example at the right of the figure
the least significant bit switches at all uneven decimal numbers, so once in
every two increments. The next bit switches once in every four increments
starting from zero to one at the second decimal value, the third bit switches
on every eight increments, starting at the fourth decimal value and so on.
The use of a Gray code prevents erroneous data due to the tolerances in the
alignment of the slots on the coding disk with the photo sensitive detectors.
The example of the rotating disk shows the effect of a position alignment
error but also errors can occur, when not all bits change at the same time
due to a different switching speed of the bits. This timing problem can be
solved by a latch, a memory that holds a digital value for a certain time and
can be read out at the moment that all data is stable. This is applicable
in systems that work with a well defined clock frequency, where all digital
data are synchronised.
Most converters that will be presented in the following sections, need several
8.5. Signal processing 583

Vi Sampling Hold Vh
_

Vi S0 S1 S2 S3 S4
Vh S5 S6 S7 S8 S9

Figure 8.22: Sampling a signal is done with a switch, while a capacitor will hold
the value constant until a new sample is taken. A buffer amplifier
before the sampling switch prevents excessive loading of the input
signal by the hold capacitor.

clock cycles to determine all bits of the digital value of an analogue sample.
During this process each determined bit is stored in the latch and only when
all bits are found, this number is transferred to the digital processor in the
last clock cycle.

8.5.2.2 Sampling of analogue values

The incremental nature of digital signals is both related to the value and
to the time. Numbers can not change gradual and are only valid over a
certain period. This fact implies that digital numbers represent samples
of a momentary value of the analogue signal. Depending on the method
of analogue to digital conversion, the sampling frequency can be constant,
synchronous with the clock frequency of the digital controller or variable
depending on the signal value. The latter can be synchronised afterwards
by means of a memory where the data are written asynchronous but the
readout is synchronised.
Samples are by definition only exact at the moment of sampling and this
automatically means that intermediate information is lost. In case of a
varying analogue signal, it seems logical that the sampling frequency should
at least be larger than the largest frequency in the signal.
It is important to see, what the limitation really is.
584 Chapter 8. Measurement in mechatronic systems

8.5.2.3 Nyquist-Shannon theorem

The low limit of the sampling frequency in a perfect analogue to digital

conversion is determined by the following key rule of any conversion process:

Any conversion is only perfect, when it can be reversed to repro-

duce the original input signal.

With this rule in mind, first the sampling process itself is considered. Fig-
ure 8.22 shows how sampling can be done in practice. The samples of the
signal are taken with a switch that shortly connects a capacitor to the signal
voltage at the output of a buffer amplifier. During this sampling moment
the capacitor is charged to the momentary signal voltage. After opening
the switch, the sampled voltage remains stored in the capacitor until the
next sample is taken. As will be shown with the A to D converters further
in this chapter, this sample-and-hold functionality is necessary for those
converters that need the input signal to be constant during the conversion
process. Other converters do not have that requirement as they average the
input signal over each sample.
One observation that can be made from the figure is the time delay that is
caused by the “hold” functionality. In principle this delay corresponds with
the fact that information is lost between the samples. The average time
delay by sampling is half the sampling period and unfortunately this delay
can never be restored! This in itself is an important determining factor for
the minimum sampling frequency when the measurement is used to control
a process.
Even more important is the effect of sampling on the frequency spectrum.
In principle, without the hold function, sampling alone is equal to amplitude
modulation of a repeated impulse signal, while the holding action results in
a kind of square wave modulation. With reference to the previous section on
signal conditioning with amplitude modulation, the frequency spectrum of a
sampled signal also shows “side bands” around the sampling frequency. With
a sampled impulse signal this pattern is repeated at n-times the sampling
frequency due to the wide frequency spectrum of an impulse.
This phenomenon is illustrated in Figure 8.23, where also the effect of the
"hold" functionality is shown with a low-pass filter character, reducing the
magnitude of the spectrum at higher harmonics of the sample frequency.
The upper graph in the figure shows the frequency spectrum in case the
sampling frequency is much higher than the maximum signal frequency.
The original signal spectrum could be simply recovered by filtering away all
8.5. Signal processing 585

Magnitude
Filter

Signal fs 2fs Frequency 3fs

spectrum
Magnitude

Filter

Signal fs 3fs 2fs 4fs Frequency

spectrum

Figure 8.23: Sampling a signal will create mirroring side bands of the frequency
spectrum of the signal around n−times the sample frequency, where
n is an integer > 0. When f s is smaller than two times the maximum
signal frequency, overlap between the spectra will occur, impairing
the possibility to reproduce the original signal. The effect of the “hold”
action at higher frequencies is shown in red.

sampling related signals above the maximum signal frequency.

The lower graph shows the effect when the sampling frequency is exactly
two times the maximum signal frequency. The lowest frequency of the first
sideband around the sampling frequency is equal to the maximum signal
frequency and only an infinitely sharp filter could prevent that nearby higher
frequencies from the side band are introduced in the original signal. With
an even lower sample frequency this can no longer be avoided and it becomes
not possible anymore to recover the original signal.

Aliasing

Figure 8.24 shows what happens when the sampling frequency is for instance
equal to the maximum frequency of the input signal spectrum. The side
bands around the sampling frequency overlap with the frequency spectrum
of the analogue input signal and any frequency of this spectrum will get
a corresponding alias frequency that is lower than the original frequency,
equal to the frequency difference f d = f s − f i . When using these samples for
reconstruction two frequencies would result, the original and its alias, which
has given the name to this effect. In case the sample frequency is higher
than two times the signal frequency this difference frequency remains larger
than the maximum signal frequency and when all frequencies above the
original signal frequency are filtered out, only the original frequency would
remain.
586 Chapter 8. Measurement in mechatronic systems

Magnitude
Alias of fi

fs - fi fi fs fs + fi

Figure 8.24: An alias frequency is generated when sampling with a frequency f s

that is lower than twice the maximum signal frequency. The fre-
quency component f i of the analogue input spectrum combines with
the sampling frequency into a much lower frequency f s − f i , the alias
frequency. This aliasing effect is also demonstrated in the lower two
graphs, where the same samples correspond with two different ana-
logue signal frequencies.

Based on the occurrence of aliasing, the Nyquist criterion for sampling

is postulated by the same Harry Nyquist, who created the Nyquist plot.
This criterion was later improved by the American mathematician Claude
Elwood Shannon (1916 – 2001) into the Nyquist-Shannon sampling theorem
by incorporating the effects of noise. The theorem defines the Nyquist
frequency f N as half the sampling frequency and the theorem reads as
follows:

When sampling, the signal frequency spectrum is not allowed to

contain frequencies above the Nyquist frequency.

This theorem corresponds with the following expressions:

fs
f i,max < f N = ⇐⇒ f s > 2 f i,max (8.32)
2
8.5. Signal processing 587

8.5.2.4 Filtering to prevent aliasing

In most cases, signals in a measurement system consist of a large spectrum

of frequencies. When f i,max is the maximum frequency of interest in this
spectrum, the sampling frequency f s must at least comply to the Nyquist-
Shannon theorem and remain above f N .
In practical systems it is however better to take some more margin especially
in feedback controlled motion systems, where the phase margin determines
the dynamic performance. This is caused by the need to apply filters twice in
the loop. The first filter, the anti aliasing filter, is used to prevent that non-
relevant frequencies higher than f N enter the sampling process. This filter
can never be infinitely steep so it is better to keep the sampling frequency
a little bit above the 2 f i,max . The other filter is used when the analogue
signal needs to be reconstructed, before the analogue amplifier part after
the controller. The higher frequency components in the sampled signal
must be filtered away as otherwise these frequencies would cause unwanted
effects in the amplifier and actuator. All low-pass filters show a phase delay
that increases with their steepness, especially when overshoot is not allowed
(high damping). For these reasons it is preferred to choose a higher sampling
frequency with a lower-order filter.
A nice example of the application of the Nyquist-Shannon theorem is the
digitisation of audio signals for a Compact Disc. The sample frequency of
the first CD-players and recordings was set at 44 kHz, only with a small
margin of 10 % above two times the maximum audible frequency of 20
kHz. This caused a lot of criticism at the introduction of the CD, because it
resulted in a “harsh” sound quality induced by the sharp higher-order filters
needed to prevent sound artifacts by aliasing. In present days, the increased
capability of digital electronics enabled even in a simple CD-player a digital
over-sampling algorithm by higher-order interpolation to a multiple of the
original sample frequency, like 96 or even 192 kHz. When reconstructing
the original signal this high sampling frequency allows the application of
simple second-order Bessel filters with a constant phase delay, which is far
more pleasant to listen to.

8.5.3 Analogue-to-digital converters

In this section three different analogue-to-digital converters (ADC) will be

presented. First the most simple version is shown, the dual-slope ADC
that generates sampled values of a variable unidirectional input voltage,
asynchronous with the heart beat of the further digital processing. It is used
588 Chapter 8. Measurement in mechatronic systems

-Vr
R 2 3
C

1 _
R Set
Integrator
+
Dual-level Micro- Data
+ processor
Comparator
_ V
i
Reset System
0V clock
-Vr

1
0

2 -Vr

3
0

Figure 8.25: Dual-slope analogue-to-digital converter. A microprocessor measures

the time in which the output voltage of an integrator changes from
0 V to a negative reference voltage level Vr . This time is proportional
to the positive input voltage and is calibrated with the integrator time
to 0V caused by switching the input of the integrator to the negative
reference voltage Vr .

in less critical situations like with controllers for central heating of buildings.
This converter does not require a separate sampling and hold function and
can be realised with only few components around a microprocessor at a
very low cost. The second converter is the frequently applied successive-
approximation AD converter. It combines a high precision with an average
speed and has long been a standard in A-to-D conversion. This converter is
not as fast as the third converter, because it requires a separate sampling
and hold function and the quantisation error is maximum 1 LSB. The third
converter is the Delta-Sigma converter. Its principle is derived from the
dual-slope ADC, but adapted such that it creates synchronous samples of
alternating voltages at a very high sampling frequency.
8.5. Signal processing 589

8.5.3.1 Dual-slope ADC

The working principle of a dual-slope ADC is shown in Figure 8.25.

This converter is only suitable for signals with a single polarity. A switch
at the input of an inverting integrator connects the input of this integrator
to either the positive analogue input signal or to a fixed negative reference
signal with a known value. A dual-level comparator is used to detect the
precise moment that the output voltage of the integrator passes either below
a negative reference voltage Vr or above a positive voltage of 0 V. As soon as
one of these events occurs, the comparator will give a set- or a reset-pulse to
a microprocessor. A set-pulse corresponds to the moment of reaching 0 V
and after that event the microprocessor will command the switch at the
input of the integrator to connect to the input voltage of the AD converter.
The output voltage of the integrator will then fall at a proportional speed
of the analogue input voltage and the microprocessor measures the time
for the integrator to reach the reference value, which triggers a reset-pulse
from the comparator. As soon as this reset pulse is received, the micropro-
cessor will command the switch at the input of the integrator to connect to
the reference voltage and starts measuring the corresponding time for the
integrator to rise to 0 V again.
The ratio between the two obtained time slots is equal to the ratio between
the absolute values of the reference voltage and the average input voltage
during the integration. With the known value of the reference voltage this
ratio can be converted by the microprocessor into a voltage value.
This dual-slope converter is the most simple of all configurations as it can
be made with only a few electronic parts. It is often used in microprocessor-
controlled devices, like the controller of a central heating system, and fre-
quently the integrator is replaced by an RC-network while the inputs of the
microprocessor are often used as the comparator inputs. In that case the
microprocessor is programmed to compensate the resulting non-linearity
and other anomalies that are related to this low-cost solution.
The resolution of this converter is limited by the amount of clock pulses
of the microprocessor in one integrator cycle, which reduces its use to less
time-critical applications. If for instance the clock-frequency of the micro-
processor is 1 MHz, a 1 ms integrator-cycle would contain only 103 pulses.
Also with an input voltage close to 0V the cycle time increases, which makes
this converter less suitable for feedback controlled motion systems that need
a low and constant phase delay.
590 Chapter 8. Measurement in mechatronic systems

8.5.3.2 Successive-approximation ADC

A more precise and relatively fast conversion procedure is achieved with the
successive-approximation analogue-to-digital converter. With this method,
the output of a digital-to-analogue converter (DAC) is compared with the
input signal, while an efficient algorithm is used to get the right digital
number in the least possible steps. Figure 8.26 shows this working principle
with a DAC based on a R-2R ladder-network.
The R − 2R ladder-network is best understood when starting at the first sec-
tion with the least significant bit (LSB) switch S1 . The source impedance at
point (1) is two times 2R in parallel, which results in a Thevenin impedance
of RTh = R . This is indifferent of the setting of switch S1 . With both settings
a low impedance path to ground is created at the switch, as a voltage source
has a zero source impedance.
The Thevenin voltage Vth(1) of this first section at point (1) equals either zero
or 0.5Vr , depending on the setting of S1 . The Thevenin impedance of the
first section is loaded by the second section and creates a voltage divider
with the resistances R and 2R of the second section, where the 2R resistor
is connected to ground at S2 , either directly or via the voltage source. This
voltage divider results in an attenuation of the Thevenin voltage of the first
section with a factor two, so either 0 V or 0.25Vr .
The equivalent Thevenin source impedance of this first and second section
combined also equals R as two 2R resistors are connected in parallel to
ground like with the first section. When switching S2 to Vr the resulting
Thevenin voltage at point (2) will be the sum of the voltage from S1 (0 or
0.25 ×Vr ) and the voltage from S2 (0 or 0.5 ×Vr ) resulting in a total voltage
of 0, 0.25, 0.5 or 0.75 ×Vr , depending of the settings of S1 and S2 .
This reasoning can be repeated for the next stages to get the total voltage at
point (3) and point (4). While in the figure only 4 bits are shown, an accurate
multi-bit DA-converter can be designed with only two values of resistors,
which is very suitable for realisation in a precision integrated circuit.
The logic circuit for the AD-converter works as follows:

• First the most significant bit (MSB) is switched on with an output

voltage of the DAC of half the full scale. When this results in a higher
output voltage than the analogue input voltage, the output of the
comparator will go from 1 to 0 and the processor decides to reset the
MSB back to 0. Otherwise the processor decides to keep the MSB to 1.

• Then the second significant bit is switched on and the same decision
process takes place: “Keep bit setting if output of DAC is smaller than
8.5. Signal processing 591

Comparator
S1
+ 1/0 S2
+ Logic processor
_ S3
_ V
i S4

RTh R RTh R RTh R

R R R
1 2 3 4

2R 2R 2R 2R 2R

+
S1 1 S2 1 S3 1 S4 1
0 0 0 0
LSB MSB _ V
r

Analogue voltage

reset
reset
reset

keep

keep

keep
keep

keep

set

set

set
set

set

set
16 voltage levels

reset

reset
keep

keep
set

set

set

set

set
set

1 0 1 0 0 1 0 1 0 1 1 1
MSB (8) LSB (1) MSB LSB MSB LSB
Time (8 clock pulses) Time Time

Figure 8.26: A 4-bit Analogue-to-Digital converter, based on the principle of suc-

cessive approximation, with an R-2R DA converter, a comparator and
a logic processor. The lower graph shows in three examples how the
four bits are sequentially determined in eight set and decide (keep or
reset) steps.

input, reject when larger”.

• This step is continued for every bit in sequence of lower significance

until the least significant bit is defined.

The main advantage of the successive-approximation ADC is its absolute

resolution and inherent linearity, which is mainly determined by the applied
DAC. As long as the DAC has enough time to stabilise after each step, the
additional errors can remain below the value of the least significant bit. Note
592 Chapter 8. Measurement in mechatronic systems

that this stabilisation of the voltage at the DAC is always necessary because
of any parasitic capacitances that are always present in any electronic circuit.
This is indicated in the figure by the gradual slope after the (re-)setting of
any bit. This timing requirement inherently leads to a the drawback of this
method as it requires many steps for a high number of bits. During these
steps the input voltage has to remain constant until the entire procedure
is finished. This means, as was stated before, that after the moment of
sampling the momentary value of the input voltage has to be “remembered”
as long as necessary with a sample and hold circuit. Also this process
needs time as the conversion process can only start, when the hold value is
stabilised.
The consequence of this reasoning is that very high speed electronics are
needed, which in practice always implies an increase in cost.
8.5. Signal processing 593

8.5.3.3 Delta-Sigma ADC

The last example of an ADC, the Delta-Sigma (ΔΣ) analogue-to-digital con-

verter, is based on another principle that combines the cost benefits of the
dual slope converter with a much higher speed and makes use of extreme
oversampling. Oversampling means that the signal is sampled with a fre-
quency that is significantly higher than necessary according to the “Nyquist”
criterion of two times the maximum signal frequency. This allows the use of
rather simple electronics, no sample and hold and only a one bit DAC with
just two values, Vr− and Vr+ . Figure 8.27 shows the basic circuit diagram of
this converter and the signals with a sinusoidal input voltage.
The core of a ΔΣ-ADC is of a ΔΣ-modulator that converts the input voltage
in a “bitstream” consisting in a timed stream of 1 bit coded signals. A digital
processor, the “decimator” converts this bitstream in a stream of multi-bit
decimal data. The ΔΣ-modulator is a variation on the pulse-width modulator
that was described in Section 6.3.3 for a switched-mode power amplifier.
The block diagram of Figure 8.27 shows the main components of a ΔΣ-
modulator:

• A differential amplifier, shown as a circle, that gives the difference

between the analogue input voltage and the output of the 1-bit DAC.

• A non-inverting integrator that consists of an inverting integrator and

an inverting amplifier with a gain of one.

• A non-inverting comparator with a very small hysteresis.

• A latch that transfers the momentary value at its input tot the output
at the positive edge of the clock.

• A digital processor that creates digital data from the output of the
latch.

• A clock that generates the timing sequence of the conversion process.

• A one bit AD-converter with two output states.

The non-inverting integrator integrates (Σ) the difference (Δ) between the
momentary value of the input signal and the output of the 1-bit DA-converter.
The output of the non-inverting integrator will rise when this difference is
positive or fall when this difference is negative. Like with the pulse-width
modulator, the comparator is used to detect the moment that the output
of the integrator crosses a reference value, but in this case the hysteresis
594 Chapter 8. Measurement in mechatronic systems

is very small and the reference voltage is 0 V. Another difference is that

the moment of switching does not immediately take place after a change
in the output of the comparator but the decision moments are based on
the clock with a latch that transfers the output of the comparator to the
digital processor at a positive edge of the clock. The output of the latch will
set the 1-bit DAC to give Vr+ when the output of the comparator gives a 1,
which corresponds to a positive voltage at the output of the integrator. The
1-bit DAC will be set at Vr− when the comparator gives a 0 at a negative
output voltage of the integrator. The resulting output of the latch is a
so called bitstream, a continuous stream of bits at a very high frequency.
This bitstream can be converted in normal digital numbers by a suitable
decimation algorithm giving multi-bit decimal data at fixed intervals.
In principle the loop with the integrator-inverter-comparator-latch-DAC-
integrator acts like a negative feedback loop with three times an inversion.
As a result of the integration, the steady state error is zero and the average
value of the output of the 1- bit DAC will be equal to the inverted value of
the input voltage. With a little bit of imagination this is visible in trace (6)
in Figure 8.27. As a consequence, the reconstruction of the original signal
is possible by simple low-pass filtering of this bitstream. This is comparable
with the output filtering in a switched mode PWM power amplifier.
This direct reconstruction principle is applied in the ultimate audio format,
“Super Audio CD” with “Direct Stream Digital” registering maximum 22 kHz
audio with 64 times oversampling which gives 64 · 44 · 103 = 2.8224 MHz clock
sampling. It achieves a dynamic range of 120 dB in the audible frequency
range, while it is allowed to register frequencies up till 100 kHz at a reduced
dynamic range without any aliasing effects because of the high sampling
frequency.
8.5. Signal processing 595

™*
'
 šm 5 Bim
_
 2 _
3
+ _ +
+
+ 4 Dii D
+

 _
L
P
_ V
i
7
6 iD@š
Vr+/ Vr* š
+

Vr

 Vi
~!'

_
Vr
+

2 2Vr
”
<
!

!*
”@š
_
2Vr
3 ¦J
!!
~ §J

4 
!!
 0

5 
!!
 0
+

6 Vr
!!
*
”@š _
Vr

7
š

Figure 8.27: Delta-Sigma analogue to digital converter. A non-inverting integrator

integrates the difference between the input signal and the output
signal of a one bit DA converter. A positive output of the integrator
results in a one and a negative output in a zero at any clock cycle. The
resulting “stream” of bits has an average analogue value equal to the
analogue input signal.
(Courtesy of Uwe Beis)
596 Chapter 8. Measurement in mechatronic systems

L L L L L L

C C C C C C

Figure 8.28: A cable with two wires, like a coaxial cable, is equivalent to a series
of small inductors and capacitors. This combination represents a
characteristic resistive impedance, equal to the required resistive load
to get Q = 1.

8.5.4 Connecting the less sensitive elements

Even though the signal conditioning element serves to create a robust signal,
it is necessary to transport this signal with care in a coaxial or twisted pair
cable. Also the digital data after the A to D conversion should be transported
with a careful choice of the connecting cables. Next to the reason to prevent
interference, especially with very long distances, another reason for a well
defined connection are the high-frequency properties of a connecting cable.

8.5.4.1 Characteristic impedance

Any combination of two wires will have a certain total capacitance and
inductance value. When examining a connecting cable with two wires more
in detail, such a cable can be modelled as an infinite series of infinitesimal
LC -filter elements. Such a combination behaves like a transmission-line
or wave-guide, an electrical equivalent to the travelling wave in the rope
as was presented in Chapter 2. Just like with the mechanical example, a
signal at the input of the cable will arrive with some delay at the output,
but what is more important is that the signal will be reflected back towards
the input, where it will again be reflected. This can cause standing waves
and resonances in the signal.
It was shown in Chapter 6 that an LC -filter can be damped by means of
a resistor and in the same way the reflection of a connecting wire can be
prevented when it is terminated by means of a resistor with a value that
would result in a quality factor Q = 1. With Equation (6.40) the value of this
8.5. Signal processing 597

characteristic impedance can be calculated:

 
C L
Q = Rc =1 =⇒ Rc = (8.33)
L C

This value is equal for any length of the same cable as both L and C depend
proportional to this length. It can be mathematically proven that a cable
with infinite length represents an input impedance that is equal to this
characteristic impedance. By terminating a non-infinite cable with its char-
acteristic impedance, the cable will behave dynamically like it is infinitely
long. This means that an optimal connection is achieved, when both the
output impedance of the preceding element and the input impedance of the
succeeding element are chosen equal to the characteristic impedance of the
cable.
When that condition is not met, a short cable might possibly not raise much
problems as it will only give an undamped response at a very high natural
frequency. For example a coaxial cable with a characteristic impedance of
50 Ω has a capacitance of approximately 50 pF per metre. Based on these
values the inductance over one metre would be equal to L = R c2 C = 125 nH.
When all capacitance is approximated to be concentrated at the end of the
cable and all inductance is in series, this combination would act as an LC

filter with a natural frequency of ω0 = 1/LC = 4 · 108 , which is approximately
60 MHz. No reflections will occur as long as the signals that are transported
do not contain any frequencies at or above ω0 . Actual wired data connec-
tions, however, already transmit frequencies above 1 Gbit per second so
in that case even one metre of cable should be terminated with the right
impedance in order to prevent reflections and high-frequency attenuation.
Longer cables will increase the problem by lowering the natural frequency
proportional to the length of the cable.
The full derivation of the related equations is beyond the scope of this book
but some observations are interesting enough to be added here.
At first sight the equal output and input impedance will result in a factor
two attenuation of the signal. This however is not bad at all as first this
level corresponds with the maximum power that the preceding element can
deliver, as was shown in Chapter 6. This is optimal from an disturbance
point of view as that is related to the interfering energy relative to the
energy in the useful signal. Furthermore, because of the transmission line
character, even frequencies above the original natural frequency will be
transmitted equally well because at any position in the cable the signal
“sees” only the resistive characteristic impedance in two directions and the
resistance is not frequency dependent.
598 Chapter 8. Measurement in mechatronic systems

Ii +
Vp

Io


‡
   ?


Figure 8.29: Transfer of a digital signal with a fibre coupling prevents electric
disturbances and errors by ground loops. It consists of a light emit-
ting diode that transmits light through the optical fibre towards a
phototransistor.

8.5.4.2 Non-galvanic connection

Although well defined wired connections are very suitable for many ap-
plications it still suffers from the problems with ground loops and other
disturbances, that were previously presented. A real solution of all these
problems is based on the transfer of information by means of light. In Sec-
tion 6.3.3.6 of Chapter 6 the opto-coupler was introduced, consisting of a
light emitting diode (LED) and a light sensitive phototransistor that are
electrically insulated from each other. A current in the LED will give a pro-
portional light output that creates a proportional current in the transistor.
The current amplification ratio of the transistor results in a current level in
the same order of magnitude as the current in the LED. Also light sensitive
diodes can be used. They have a higher switching speed but also a reduced
current because they fail the current amplification of the phototransistor. A
clear advantage of the opto-coupler is the galvanic insulation that completely
cancels any kind of ground loops, although at very high frequencies some
capacitive coupling still remains. Because of the significant temperature
dependent non-linearity, the application of opto-couplers is mainly limited to
the digital domain, so after the signal processing element. A special version
of the opto-coupler is the fibre coupling of Figure 8.29. The insertion of an
optical fibre between the LED and the phototransistor completely cancels
the capacitive coupling but more important is the possibility to transport
the signal over very long distances without any electromagnetic interference.
When necessary even a strong semiconductor laser can used as the light
source. This technology has become the de-facto standard for data transfer
over very long distances like with internet.
8.6. Short-range motion sensors 599

8.6 Short-range motion sensors

The first category of position and motion sensors are those sensors that are
used to determine the relative position of two objects at short ranges and
these sensors are often called a proximity detector. Typical applications
include the boundary detection for long-range actuated systems. Also the
sensors for fixating a position by means of a servo system like in a magnetic
bearing belong to this category. Often only the accuracy at the zero position
is relevant while the sensitivity over the total measurement range is relevant
for the gain of the servo-loop.
The second category consists of those sensors that register motion like
velocity and acceleration. They are applied in many different fields like in
geophysics to detect earthquakes and in precision equipment for vibration
control.
Because of this wealth of possibilities only a few examples are selected as
these are representative for the entire field

8.6.1 Optical sensors

The first example of a proximity detector is shown in Figure 8.30. It con-

sists of a light emitting diode and a photo-transistor sensor in one housing
“looking” in the same direction. The working principle of this optical proxim-
ity detector is based on the difference in irradiance levels detected at the
photo-transistor as a function of a distance of the object. It is interesting to
see that there are two areas that can be used with an approximately linear
relationship between the position and the current. This is due to the fact
that at zero distance the object will obscure the LED sensor combination
and at infinite distance no light is reflected anymore. Somewhere in between
there is a maximum of captured light at the detector with two slopes at each
side. The slope at the near side has the highest sensitivity. 5
In order to achieve some level of accuracy both modifying and interfering
error sources of this system must be determined. First of all the current
level of the LED has influence as it is directly proportional to the amount of
emitted light. The overall shape of the curve of Figure 8.30 will be unaffected
but the magnitude is linearly proportional to the amount of light from the
5 If this detector is applied in a servo system it is necessary to be aware that the two slopes

work with an opposite sign. This means that a stable system working on one slope will be
unstable due to phase sign reversal at the other slope. This necessitates special precautions,
when the system is out of range and the stable point has to be retrieved.
600 Chapter 8. Measurement in mechatronic systems

Detector ›

(Photo-Transistor)

?!
z

›
‚›”„
Sensor 0 ˜


Figure 8.30: An optical proximity detector consisting of a Light Emitting Diode

with a sensor that determines the irradiance of the reflected light
from an object. The relation between the current of the sensor and
the distance of the object (blue line) shows two opposite approximately
linear slopes (dashed red line) that are both applicable for position
sensing.
(Courtesy of Jasper Wesselingh)

LED. This means that the current in the LED is both a modifying input as
it changes the slope (sensitivity) and an interfering input as it changes the
value at any position. The effect of this can be reduced by comparing the
current with a reference current that is proportional to the current in the
LED. Also supplying the LED with a constant current source as presented
in Chapter 6 is a suitable method to reduce the influence of the LED.
Other error sources include the reflectivity of the object, the angle of the
object, noise in the detector and ambient light that is detected by the photo-
transistor. The influence of the ambient light can be reduced by using an
optical filter in order to only detect the wavelength of the LED light. To limit
the noise of the detector an amplifier with a low input noise level must be
located as close as possible near the sensor. Figure 8.31 shows an example
of such a circuit which is also optimised for high-frequency behaviour as the
transistor is directly connected to the virtual ground at the minus input of
the amplifier. This means that the current from the transistor is transformed
into a voltage at the output of the amplifier without causing a voltage change
over the transistor itself, thus avoiding high-frequency current from the
collector to the base by the internal parasitic Miller capacitor.

8.6.1.1 Position sensitive detectors

Instead of a single photo-transistor or a photo-diode also a sensor consisting

of a multitude of sensing elements can be used, giving both information on
the irradiance and on the position of the light spot that is directed to the
8.6. Short-range motion sensors 601

I
_

G
+

_
Vo
Vb
+

Figure 8.31: A photo-transistor works optimal with reduced current noise, when it
is supplied with a constant bias voltage Vb . The virtual ground at the
minus input of the operational amplifier cancels the AC voltage over
the transistor which gives a better high-frequency behaviour.
(Courtesy of Jasper Wesselingh)

device. An example of this principle is the four-quadrant detector with 4

photo-diodes, also called a quad-cell, as shown at the left side of Figure 8.32.
Also an analogue version of this four quadrant detector exists as shown on
the right side. The total area of this detector is sensitive to the incident
light but the resulting current is transported to the four external contacts
by means of a material with a high resistivity. This causes the current of
each contact point to be inverse proportional to the distance to the incident
spot of the light.
With both types of PSDs the ratio between the current of each output is
ideally related to the position of a light spot that is exposed on these sensors
according to the following equations:
(I2 + I4) − (I1 + I3)
x=
I1 + I2 + I3 + I4
(I1 + I2) − (I3 + I4)
y= (8.34)
I1 + I2 + I3 + I4

By dividing the difference of the currents by the sum of all four currents
the resulting value is independent of the irradiance of the light beam which
would otherwise be a modifying input. In reality, the relation between the
position and the found values also depends on the size of the light beam,
the distribution of the light inside the beam and the character of the sensor.
602 Chapter 8. Measurement in mechatronic systems

Beam of Beam of
light light
y
I1 I2 I1 I2

I3 Quad-cell I4 I3 Proporonal I4
Figure 8.32: The four quadrant position sensitive detector either consists of 4 sepa-
rate photo diodes as shown at the left side or of one large photosensitive
area with high resistive paths to the output wires as shown at the
right.

Generally a four-quadrant detector is used as a zero-sensor in a feedback

system like in the Optical pick-up unit of a CD-player, as around zero the
errors due to diameter and light distribution are minimal. A proportional
sensor is less sensitive for these deviations but it shows more noise due to
the high resistivity.
A more elaborate example of such a sensor is the CCD sensor as used in
digital cameras and the Shack-Hartmann wavefront sensor from Chapter 7.
In order to avoid a large number of wires a CCD sensor has a dedicated elec-
tronic circuit, the charge coupled device, that makes it possible to read out
a row of photo-diodes in series by sequentially transferring the accumulated
charge of each photo-diode (the integral of the current over the sampling
time) to its neighbour like buckets of water that are transferred by people
in a row to empty a well.
8.6. Short-range motion sensors 603

Laser Posion Sensive Detector (PSD)

dd

Mirror surface
ds

Figure 8.33: Surface displacement measurement with a position sensitive detector

by means of triangulation. A shift of the surface (ds ) will cause a shift
( dd ) of the laser beam on the PSD.

8.6.1.2 Optical deflectometer

A position sensitive detector can be used to measure several mechanical

properties. By deflecting a beam of light off a surface and projecting it
on the PSD it is possible to determine the angle and/ or the position of
this object as shown in Figure 8.33. This method is called triangulation
because the real displacement depends on the angles and can be calculated
by trigonometry. In this simple set-up the angle can not be distinguished
from the displacement of the surface. This can be solved by focusing the
beam on the surface and imaging that focal spot by a lens on the PSD. In
that case the angle will have no influence anymore.
An example where imaging is used to detect the angle of a surface instead
of the height is the Deflectometer of Figure 8.34. This instrument is used
to measure the surface shape of a reflective surface. It works as follows:
A laser beam is directed to a mirror that oscillates around the point of
incidence of the laser beam. Due to this oscillation, the laser beam will
be reflected in different directions between the outer positions as shown
in the picture causing a scanning light spot on the scan lens. This lens is
positioned such that the rotating point of the mirror is in its focal plane
which means that the rays from the laser beam after the lens are always
running parallel. The angles of these rays are only determined by the
position of the oscillating mirror relative to the optical axis so they are
independent of the momentary angle of the mirror. The surface that must
be measured reflects these parallel rays back to the lens where they will
be focused again to create an image of the original spot on the rotating
mirror. Because the rotating mirror is positioned just left of the optical axis
604 Chapter 8. Measurement in mechatronic systems

Oscillang
mirror
Laser
PSD

Scan
lens

Surface under test Scan track

z y
Movement direcon
x

Figure 8.34: Optical deflectometer that measures the shape of a surface by a scan-
ning beam and suitable optics that images the spot of the oscillating
mirror to the PSD via the surface. Local angular variations of the
surface will result in position differences of the spot on the PSD.

the image will be positioned at the same distance6 right of the optical axis,
where the PSD sensor is located. Because the spot of the rotating mirror
is imaged on the PSD, the position of this image is not influenced by the
angle of the mirror as long as the surface that must be measured is perfectly
flat. When that is not the case the parallel incident rays on the surface
will not be reflected parallel anymore and an angular deviation at a certain
location on the surface will result in a change of position of the spot on the
PSD. When the corresponding electrical signal is synchronised with the
oscillation of the mirror, this position signal of the PSD gives information
about the slopes at different locations on the surface and by calculation the
shape of that surface can be derived.

6 This is a 1:1 optical system as both image and object are in the focal plane and a positive

lens inverts the sign of the position relative to the optical axis.
8.6. Short-range motion sensors 605

I (~)
w x x
Metal plate
+ A d
x
d 1 2
_ V (~) ε d
s l l l l dH
Distance §¨© dx ·¸¹ Area §¨© dx ·¸¹ Dielectric §¨© dx ·¸¹
dd dA
Dielectric

Figure 8.35: Three types of capacitive position sensing by changing the position of
two plates of a capacitor in two directions or changing the dielectric
properties as function of the displacement.

8.6.2 Capacitive position sensors

In Section 6.1.2.2 of Chapter 6 it was shown that a capacitor consists of two

parallel plates, separated by an electrically non-conductive dielectric layer
with a capacitance of:
εA
C= (8.35)
d
With ε = ε0 εr and

ε0 ≈ 8.8541878176 · 10−12 (8.36)

For air εr equals approximately one but as will be shown further in this
chapter, this assumption is a cause for errors.
This inverse relationship of the capacitance with the distance is the reason
that this physical property is used to determine displacements between
two objects that are positioned close together. An AC source is necessary
to determine the capacitive value with the complex version of Ohm’s law
(V = Z · I ). In Figure 8.35 it is shown that not only the mutual distance
between the plates can be used to measure displacements but also the
other two terms in the simple relation can be made variable corresponding
with a displacement. In case the displacement of one of the plates is in
its own plane the distance remains the same but in this case the shared
surface is changed while in the third example a separate part with another
dielectric value is inserted between the plates which also gives a change in
the capacitance according to the position of this part relative to the plates.
Each solution has its own advantages. A variation of the distance gives
generally a higher sensitivity than the parallel movement of the plates but
the latter has a better linearity which can be concluded from the respective
606 Chapter 8. Measurement in mechatronic systems

E1
C1
V1  V2 _ dx E2 x

+
Vo

+
Vp (~)

+
+
_ dx E3
V1 V2 C2

_ _

Figure 8.36: A differential capacitive sensor has a linear relation between the
displacement x and the voltage Vo .

transfer functions of these three examples:

εA
1: Variable-distance: C= (8.37)
d+x
ε
2: Variable-area: C = ( A − wx) (8.38)
d
ε0 w
3: Variable-dielectric: C= (ε2 l − (ε2 − ε1 ) x) (8.39)
d
From these three different sensing principles it can be concluded that a
measurement according to one principle is also influenced by the other effects.
Especially the distance has a very large influence on the sensor in case of the
variable-surface and variable-dielectric sensor. With these principles the
distance acts as a modifying error source. In case of the variable distance
sensor the lateral displacement has less influence especially if one of the
electrodes is larger than the other. Because of its high sensitivity, the
variable-distance measurement capacitive sensor is preferred, but the non-
linearity still is a factor that preferably is avoided. This can be achieved
by means of a differential sensor and the previously presented Wheatstone
bridge as shown in Figure 8.36.

8.6.2.1 Linearising by differential measurement

The differential capacitive sensor consists of two stationary electrodes E 1 , E 3

with a moving electrode E 2 in between. The capacitance C1 is the combina-
tion of E 1 with E 2 and the capacitance C2 is the combination of E 2 with E 3 .
When the surface of the electrodes is equal to A the capacitance becomes:
Aε Aε
C1 = , C2 = (8.40)
d−x d+x
8.6. Short-range motion sensors 607

The voltage Vo = V1 − V2 . When applying the rules of a voltage divider this

voltage equals:

Z2 Z4
Vo = Vp − (8.41)
Z1 + Z2 Z3 + Z4

In order to get a positive output voltage with a movement in the positive

x−direction C 1 can be located at Z1 and C 2 at Z2 to give an increase of V1 .
Another option would be when C1 is located at Z4 and C2 at Z3 as in that
case a positive movement in the x−direction gives a decrease of V2 . In this
example Z1 and Z2 are chosen and Z3 = Z4 are two equal resistors to balance
the bridge.
With this choice the following equation gives the voltage as function of the
displacement:
⎛ 1 ⎞

⎜ j ωC 2 1⎟ d+x 1 Vp
Vo = Vp ⎜
⎝ − ⎟⎠ = Vp − =x (8.42)
1 1 2 d−x+d+x 2 2d
+
j ωC 1 j ωC 2

This is a linear and frequency independent relation.

8.6.2.2 Accuracy limits and improvements

From the previous part it can be concluded that the errors in capacitive
measurement sensors are caused by many external influences, summarising:

• The amplitude and frequency of the source signal.

• The noise of the sensing electronics.

• Crosstalk of movements in other directions.

• Thermal effects due to change in size of the electrodes.

• Changes in the dielectric.

• Inhomogeneity in the electric field.

• External electric fields.

It is clear that the electric source has a large impact if the values are not
exactly known. Fortunately these electrical parameters can be controlled
to high levels of precision and the measurement result can in most case be
compensated. This is less the case with the error by the noise of the sensing
608 Chapter 8. Measurement in mechatronic systems

electronics. In practice a capacitive sensor has a capacitance in the range of

a few pico-Farad which requires a very high excitation frequency to create a
sufficiently high current level for an acceptable signal to noise ratio. This
sensitivity can be illustrated by using the data of one of the best operational
amplifiers on the market regarding#noise, the AD797 from Analog Devices. #
It shows a voltage noise of 0.9 nV/ Hz and a current noise of 10 pA/ Hz.
With a 1 kHz bandwidth this is ≈ 30 nV and ≈ 0.3 nA of noise. When this
amplifier is used with a capacitive sensor of 6 mm diameter active area
giving a surface area A ≈ 30mm2 ) and a distance d of 300 μm, this results
in a capacitance of ≈ 1 pF. A voltage source of 10V @ 1 MHz will create a
current of ≈ 0.1 mA, a factor 3 · 105 above the 0.3 nA noise of the amplifier.
With a distance of 300 μm this noise level corresponds with around 1 nm
error. When a lower error value is needed it is only possible to decrease the
working distance and hence limit the range.
The factors on thermal effects and crosstalk of other movements are directly
related to the mechanical construction of the sensor in combination with
its surrounding parts. An isothermal measurement condition is required
to avoid the thermal errors but in principle it is preferred to design the
mechanics around the sensor in such a way that the so called thermal-
centre is located inside the sensor. The thermal centre in a construction is
defined as the place where no local movement is observed at a change of
temperature. When the capacitive sensor is located at the thermal centre of
the measurement system, a temperature rise does not affect the distance of
the plates. This is not always possible to achieve as it requires a thermally
symmetrical construction around the sensor where all other parts move
away from the thermal centre at a temperature rise. Sometimes the use of
materials with a different thermal expansion coefficient can help to achieve
this goal. Aside from a linear movement by temperature rotations and other
movements between the plates have to be avoided by a stable construction.
Changes in the dielectric are present in case the sensor is not used in
vacuum which is almost always the case. The relative permittivity of air
is a function of its composition and humidity and pressure variations can
be observed in the measurement value. Although this relation is expected
to be quite straightforward, the humidity level has a more direct effect
that is not yet completely understood. It is based on the occurrence of thin
layers of water on any object that is in contact with air that contains water.
The layer thickness depends on the partial pressure of the water vapour
(level of saturation). This layer has a far higher permittivity than air and
can in principle be approximated as a continuation of the electrode. This
means that the air gap is decreased with the same amount as the water-
8.6. Short-range motion sensors 609

Transmi‡ng electrode (Moving part)

+
Vp (~) Io
_
R

Guard Ring Sensing electrode Io

_

G () Vo(~) _

+
+

Figure 8.37: By using a guard ring the electric field at the sensing electrode is
homogenised. The capacitance is measured by virtually grounding the
sensing electrode and measuring the current with a transimpedance
amplifier.

layer thickness. Much research is spent on this subject with measured

thickness’s ranging from 1 nm to almost a μm but no concluding results are
found to sufficiently model this phenomenon so that it can be compensated.
Though this phenomenon is influenced by the hydrophilic properties of the
electrodes, the effect occurs even with a PTFE layer. For extreme precision
measurements this means that these have to be done in dry air or vacuum.
Short term measurements are less critical as a change in the water layer
by the environmental humidity is generally rather slow, in the range of
minutes, depending on the geometry and diffusion speed.
Inhomogeneity in the electric field is caused by edge effects as the field
lines are bent at the edges of the sensor. This can be reduced by a special
configuration as shown in Figure 8.37 for a single ended sensor. It should
be noted that this sensor does not work with a Wheatstone bridge but in
this case the current of the sensing electrode is measured when grounding
this electrode at the virtual ground of a transimpedance amplifier.
At the upper side a large electrode is supplied by the electrical supply with
a high-frequency AC voltage relative to ground. The opposing electrode, the
sensing electrode is much smaller and is surrounded by a guard ring that is
kept to zero volt ground level. The sensing electrode is also kept at ground
level by the transimpedance amplifier which means the total lower plane of
the sensor is at ground level and the field lines will only bend at the edges
of the guard ring and not at the sense electrode.
The last item of errors is caused by the extremely small measurement
current levels as mentioned before. This makes this sensor type extremely
sensitive for interfering electromagnetic fields even when only the excitation
610 Chapter 8. Measurement in mechatronic systems

Moving part
Io
R
_
+
Vp (~) Transmi‡ng Guard Ring Sensing electrode
_ electrode G ()

+
Vo(~) _
+

Figure 8.38: By using two electrodes and a guard ring, the distance to a moving
conducting object can be measured without the need to directly contact
the moving part. In this case the guard ring reduces the current that
flows directly over the capacitance between the transmitting electrode
and the sensing electrode.

frequency is filtered out with a selective filter. In case this sensor is applied
in a high bandwidth real time feedback controlled system, all measures for
shielding must be applied, while avoiding loops in the wiring. Furthermore,
the sensing electronics should be located as close as possible to the sensor.
With extreme critical applications this might even requires measures for
cooling the heat that relates to the often high power consumption of the
required low noise amplifiers in order to avoid thermal errors.

8.6.2.3 Sensing to conductive moving plate

The guard ring method can in principle also be applied in a differential

sensor to achieve linearity but in some cases it is even necessary to measure
the distance to an object that can not be supplied directly with any electric
current by means of a connected wire. As an example one can think of a
continuously rotating shaft or a fast moving long-range linear positioning
system. In that case it is possible to use a third electrode at the stationary
part as shown in Figure 8.38. In this configuration a series of two capacitive
sensors is obtained that both depend on the distance to the moving part.
When the total active surface area is unchanged, the capacitance of this
sensor is about a quarter of a normal sensor because the surface of the
electrodes is half and the distance is passed twice. This implies a very small
working distance in the order of 100 μm or less for real precision measure-
ments. In this case the guard ring is used to separate the transmitting
electrode from the sensing electrode. This will reduce the currents that
directly flows between both electrodes. When the transmitting electrode is
much larger than the sensing electrode the field on the sensing electrode
will determine the total current and in that case the more homogeneous
8.6. Short-range motion sensors 611

n windings
Ac

Ac,1
Ac Ag d
Ac,2 Ag

Figure 8.39: Distance and velocity sensor based on the variable reluctance of an
air gap.

field at the sensing electrode will be also useful.

Also this sensor is used with virtual grounding by a transimpedance ampli-
fier.
As a conclusion, a capacitive sensor has the ability to give information
about the relative position of an object without any moving parts that are
susceptible to wear. This means that these sensors can be used where a
high reliability over an extended period of time is required.

8.6.3 Inductive position sensors

In Section 5.1.2 of Chapter 5 the inductor was introduced, consisting of a coil

wound with n windings around a ferromagnetic core creating a magnetic
flux according to the following equation for a homogeneous core:
nI A c μ0 μr
Φw = = nI (8.43)
ℜ c
When the magnetic path consists of both a ferromagnetic core and an air
gap, the total reluctance is the sum of the respective reluctances. For the
situation as shown in Figure 8.39 the total reluctance equals:
c g 2
ℜ= + = ℜ0 + c 1 d with: c 1 = (8.44)
A c μ0 μr A g μ0 A g μ0

Note that the flux has to pass the air gap twice and that c consists of two
parts, one part is the core with the windings and one part is the return path
to the object of which the distance d is measured.
The self inductance was defined as follows:
Φw,t nΦw n2
L= = = (8.45)
I I ℜ
612 Chapter 8. Measurement in mechatronic systems

L1

V1  V2 _ dx x

+
Vo

+
Vp (~)

+
_ dx

V1 V2
_ _
L2

Figure 8.40: A differential reluctance sensor has a linear relation between the
displacement x and the voltage Vo .

where Φw,t equals the sum of the flux Φw of each winding. It is further
assumed that Φw is equal for all windings. Using this definition, the self
inductance becomes:
n2 L0 c1 n2
L= = with: c 2 = and: L 0 = (8.46)
ℜ0 + c 1 d 1 + c 2 d ℜ0 ℜ0
The voltage over a self inductor is proportional to the change of the total
flux Φw,t over time. This leads to the following relation:
dΦw,t d dI dL
V= = LI = L +I (8.47)
dt dt dt dt
This means that an inductive sensor can use two sensing principles: In
case of an AC current through the coil the voltage is directly related to
the position. With a DC current the voltage is proportional to the relative
velocity. This second principle is hardly used for measurement as using
permanent magnets is far more efficient to detect a relative velocity in a coil
as will be explained later. For that reason this principle will not be pursued
any further.
The first term however is important as the self inductance is inversely
proportional to the distance, similar to a single capacitive sensor. The
related non-linearity can be solved in the same way as with a capacitive
sensor by using a Wheatstone bridge and two sensors in a balanced bridge
as shown in Figure 8.40. The impedance of an inductor is proportional to
the frequency ( Z = j ωL). For that reason the other branch of the Wheatstone
bridge is taken as with the capacitive sensor in order to get a positive output
voltage with a movement in the positive x−direction. The voltage can be
calculated with Equation (8.46) in the same way as with the differential
capacitive sensor:
L0 L0
L1 = , L2 = (8.48)
1 + c 2 ( d − x) 1 + c 2 ( d + x)
8.6. Short-range motion sensors 613

50
0 100

Di<erenal
transformer

Amplier ?' 

Probe

Figure 8.41: A Linear Variable Differential Transformer can measure very accu-
rately small displacements around the centre position.

The voltage Vo = V1 −V2 becomes, when applying the rules of a voltage divider:


Z2 Z4
Vo = Vp − (8.49)
Z1 + Z2 Z3 + Z4

If Z2 = Z1 , being just simple resistors in most cases and replacing Z3 and Z4

by the impedance of respectively L 1 and L 2 , the following equation gives the
voltage as function of the displacement:
⎛ ⎞

1 j ωL 2 ⎜1 1 ⎟
Vo = Vp − = Vp ⎜
⎝ − ⎟=
⎠
2 j ωL 1 + j ωL 2 2 1 + c 2 ( d + x)
+1
1 + c 2 ( d − x)

1 1 + c2 d − c2 x c2 x
= Vp − = Vp (8.50)
2 2(1 + c 2 d ) 2(1 + c 2 d )

Also this relation is linear and frequency independent.

8.6.3.1 Linear variable differential transformer

In laboratory measurements and workshops the Linear Variable Differential

Transformer (LVDT) as shown in Figure 8.41 is a frequently used instru-
ment for measuring small displacements. It is based on the principle of a
coupled alternating magnetic field like with the electrical transformer as
was presented in Section 5.5. In Figure 8.42 the working principle is shown
in more detail. The transformer consists of three coils where the middle
coil is supplied with an AC voltage. The two secondary coils are positioned
on both sides in line with the primary coil. Other than with the normal
614 Chapter 8. Measurement in mechatronic systems

+
Alternang V1 Phase sensive detector
magnec ux _

+
+ Amplitude of V1  V2
Vs(~) Value Vo
_ Vo

(V1  V2 ) u Vs Sign Vo
_

+
V2
_

Top

Middle

Boom

Vs V1 V2 V1  V2 (V1  V2 ) u Vs Vo

Figure 8.42: Signals in an LVDT. The secondary coils in an LVDT partly share the
flux of the primary coil. In the mid position the induced voltages are
equal. A position change of the core will increase the flux at one coil
while decreasing the voltage in the other. The difference voltage is
then proportional to the displacement while the phase corresponds
with the direction.

transformer the core is not closed but consists only of a small ferromagnetic
rod with about the same size as the primary coil. In the mid position the
ferromagnetic rod is located just inside the primary coil. The generated
magnetic flux from the primary voltage will only partly be coupled to the
secondary coils. The induced voltage in these coils therefore is small and
equal for both coils. The subtraction of these voltages gives a value of 0 V,
corresponding with this mid-position. When however the core is moved in
the direction of either one of the secondary coils, the coupling with that
coil will increase while the coupling with the other coil will decrease with a
corresponding increase and decrease of the induced voltage. In that case
the subtraction of both voltages results in a voltage with an amplitude that
is proportional to the displacement. Depending on the direction, whether
8.6. Short-range motion sensors 615

Alternang magnec ux

AC primary current

Ie
Induced AC eddy-current

Figure 8.43: Measuring the distance to a subject by means of an eddy-current sensor.

The magnetic field induces a voltage over the conductive object. This
creates a current that will suppress the magnetic field resulting in an
increased current in the primary coil.

moving towards coil A or B, the phase of the resulting signal will be inverted
or non-inverted giving the direction of the position. By using a phase se-
lective detector a DC voltage is obtained with a sign that corresponds to
the direction of the displacement. Phase selective detection can be achieved
by taking the sign of the signal obtained by multiplying the input with
the output and use the rectified and low-pass filtered output signal V1 − V2
to get the amplitude. This principle was also explained in Section 8.4.2.2
regarding synchronous demodulation of an amplitude modulated signal.

8.6.3.2 Eddy-current sensors

As a last example of an inductive proximity detector the eddy-current sensor

is presented here as it is frequently used in industrial systems to detect
the proximity to an electrically conducting object. Like with the LVDT,
its working principle is based on the interaction of two windings by their
coupled magnetic field however in this case one of the windings is the object
itself. When looking at Figure 8.43 the sensor consists of a single primary
coil wound around a ferromagnetic open core. This primary coil is supplied
with an AC voltage. The resulting magnetic field exits the core on one end
and enters it at the other. As soon as the sensor approaches the electrically
conducting surface of an object, the alternating magnetic field will induce an
alternating voltage in this surface comparable with the secondary winding
in a transformer. This voltage is however short circuited as this secondary
winding is a solid piece of metal resulting in an eddy-current that will
616 Chapter 8. Measurement in mechatronic systems

suppress the magnetic field. This causes the current in the primary coil to
increase just like with a regular transformer and by measuring this current,
a value for the distance to an object is obtained. Information on the distance
is available both in the amplitude and in the phase of the primary current
relative to the primary voltage which is explained as follows. At a large
distance from the conductive object the coil acts as an ideal self inductance
giving a current with a low amplitude that is 90◦ delayed. As soon as the
distance is smaller the eddy-current will induce a real primary current that
adds to the imaginary current level because the eddy-current is in phase
with the secondary voltage and consequently with the magnetic flux and the
primary voltage. This also means that the consumed power of the sensor
increases and the total current will be less than 90◦ out of phase.
Eddy-current sensors are not very precise but measurements with μm accu-
racy around a fixed setpoint are possible like in magnetic bearings.
8.7. Dynamic measurements of mechanical quantities 617

8.7 Dynamic measurements of mechanical quan-

tities

Mechatronic systems are ruled by dynamic, time varying signals that are
determined by forces acting on the system. The measurement of these forces
and the related velocity and acceleration of a positioning system is most
important in controlling a positioning system for all states. Also in the
dynamic modal analysis of a mechanical part inside a mechatronic system
these dynamic sensors are frequently applied.

8.7.1 Measurement of force and strain

Force is a physical concept that is only indirectly observed at two different

phenomena, strain and acceleration. The force we “feel” as a human being
is the result of strain as our body deforms under the influence of a external
force and we need to control our muscles to compensate that deformation.
Generally, even with acceleration, force is measured by measuring the strain
in a material that is exposed to a force. The strain gage has become the
standard method for this measurement but lately also a very sophisticated
modern optical method has been developed, using a Fibre Bragg Grating
that will be presented after the strain gage.

F
A
B

Figure 8.44: A force sensing bending beam with different locations for the strain
gages. The gage at location A is elongated while the gages at locations
B and C are compressed when applying a force F in the downward
direction.
618 Chapter 8. Measurement in mechatronic systems

Vi+

Z1 Z3

Vb + Vb-

Z2 Z4

Vi-

Figure 8.45: A full bridge strain gage consisting of 4 strain sensitive resistors.

8.7.1.1 Strain gages

A strain gage is a piece of thin conductive material that changes its resistance
as function of a change of its dimensions. An elongation in the direction of
the current flow will increase the resistance both by the longer trajectory
and by the corresponding contraction of its diameter by the stretching effect.
These sensors are mainly used in non-destructive measurements with very
limited strain levels. As a consequence the resistive change is also very
limited in the order of less than 1 %.
A strain gage is a passive sensor and needs an external source of electricity
to function. As explained before, the Wheatstone bridge is the best choice
for this kind of sensor and in the following the Wheatstone bridge is taken
as example. It was shown that the temperature effect in the Wheatstone
bridge can be minimised by using a thermally balanced branch with two
sensors with an opposite sign in the sensitivity. In Figure 8.44 it is shown
how this can be realised with strain gages by mounting the second strain
gage close to the first strain gage on a location that either shows an opposite
strain like the reverse side of the bending beam or on the same side but
orthogonal to the first strain gage where the poisson ratio of the material
will cause the surface to show an opposite strain.
The most optimal configuration of a strain gage where both temperature
and source noise is minimised relative to the sensitivity is the full Wheat-
stone bridge with four strain gages. This can be realised on one monolithic
part when using the Poisson ratio effect according to the configuration of
8.7. Dynamic measurements of mechanical quantities 619

Table 8.1: K factors of strain gage materials.

Material K factor
Platinum-Iridium (Pt 95 %, Ir 5 %) 5.1
Platinum-Tungsten (Pt 92 %, W 8 %) 4.0
Isoelastic (Fe 55.5 %, Ni 36 % Cr 8 %, Mn 0.5 %) 3.6
Constantan / Advance / Copel (Ni 45 %, Cu 55 %) 2.1
Nichrome V (Ni 80 %, Cr 20 %) 2.1
Karma (Ni 74 %, Cr 20 %, Al 3 %, Fe 3 %) 2.0
Advance (Cu 54 %, Mn 1 %, Ni 44 %) 7 2.0
Armour D (Fe 70 %, Cr 20 %, Al 10 %) 2.0
Monel (Ni 67 %, Cu 33 %) 1.9
Manganin (Cu 84 %, Mn 12 %, Ni 4 %) 0.47
N or P type Silicon 120-175!

Figure 8.45. Only 4 connections are needed to the measurement electronics.

This configuration is frequently used in many force sensors (load-cells) and
is proven to be very reliable.
Table 8.1 shows several materials that are used in strain gages, each with
a different strain sensitivity. While K for most metals ranges between 0.5
and 5 the piezoresistivity effect in semiconductors gives a sensitivity that
far outweighs the simple straightforward effect by the volumetric changes
of a material under stress like in metals. This effect is due to the special
conduction mechanism in semiconductors with holes and electrons which is
also influenced by strain resulting in gage factors of 100 or more. This has
enabled its use in fully integrated MEMS accelerometers that are frequently
used in cameras and cell-phones for motion detection and sensing of the
gravitational forces of the earth in order to detect the orientation. With
these materials the full bridge configuration is even more necessary because
of the also inherent very large temperature sensitivity of Silicon.
620 Chapter 8. Measurement in mechatronic systems

d n B
Broadband
light  z B

B

0
10
Intensity (abs)


Transmission
>^
10

>
10

1570 \{^ \{ 1576 1578

Wavelength [nm]

Figure 8.46: A fibre Bragg grating is created inside an optical fibre by means of an
area with a longitudinal periodic variation of the index of refraction.
A standing wave will occur at a wavelength λB of which an integer
number just fits inside the grating period d . The light corresponding
with that standing wave is reflected.
(courtesy of Technobis Fibre Technologies)

8.7.1.2 Fibre Bragg grating strain measurement

The fibre Bragg grating principle is discovered by the British physicist

William Lawrence Bragg (1890 – 1971) by research on X-ray diffraction at
crystalline materials, for which he won the Nobel prize. The observed effect,
the reflection and refraction of certain wavelengths in non homogeneous
material, is closely related to the multiple Fabry-Perot interferometer effect
of optical coatings as explained in Chapter 7.
An optical fibre Bragg grating is created inside a transparent optical fibre by
introducing small areas of approximately 3 mm with a periodic longitudinal
variation of the index of refraction. When the periodicity is constant, the
grating will cause a reflection of those light wavelengths λB that fit in the
grating period d according to the expression d = nλB , where n is an integer
and in most applications equal to one.
The working principle is illustrated in Figure 8.46.
When light from wide-band light source is inserted in the fibre, only the light
8.7. Dynamic measurements of mechanical quantities 621

G1 G2 G3 Gn
Analiser
λ1 λ2 λ3 λn

λ1 λ2 λ3 λ1 λ2 λ3
~! 
  
#'
'

!•


Input pulse Re ected pulses

I

t0 t1 t2 t3 tn


'

!•


Figure 8.47: A fibre Bragg grating can be used to measure strain at long distances
simultaneous at different locations on the fibre, distinguished by the
wavelength. Also the temporal response of the different reflections
gives information of the strain.
(courtesy of Technobis Fibre Technologies)

with a wavelength equal to λB will be reflected. An elongation of the fibre

will result in a change in the reflected frequency because of this relation,
which makes this frequency a measure for the strain of the fibre. By locating
the fibre on or inside a solid object, the strain can be determined at the
location of the grating.
In principle the grating can be located at a very long distance, but even more
interesting is the fact that more then one grating can be applied on one fibre,
as shown in Figure 8.47. With Wavelength Division Multiplexing (WDM) the
different gratings are distinguished by giving them each a slightly different
grating period as long as the spectral range of the frequencies that fit in
each grating period do not have an overlap.
With some real numbers this becomes more clear.
In practice a light emitting diode (LED) can be used as a light source. In this
example the LED has a wavelength of 850 nm and a spectral bandwidth of
50 nm. This bandwidth is divided in ten equal frequency ranges separated
with 5 nm spectral difference with the other areas. Rejecting the outer areas
with a lower irradiance results in practice in eight different sensing areas
per fibre.
The grating period is ≈1 μm due to the average refractive index of the fibre
giving 3000 periods in the 3 mm measurement area.
622 Chapter 8. Measurement in mechatronic systems

The frequency spectrum is measured by means of a diffraction grating

where the wavelength determines the diffraction angle. The diffracted beam
exposes a CCD photo detector at a fixed distance from the grating and the
position of incidence of the diffracted light is measured. This system has a
maximum resolution of a wavelength change of one picometre, corresponding
with 1 μm per metre strain while the repeatability is better than three micro-
strain.
The electronics are so fast that the dynamic bandwidth equals 20 kHz with
a fixed latency of 50 μs.
Next to this spectral difference, the distance between the gratings can be
measured by determining the return time of the reflections by a pulsed light
source. This is called Time Division Multiplexing (TDM).
Like with almost all sensing methods also this principle suffers from temper-
ature as an interfering error source. The index of refraction of any material
is temperature dependent to a certain degree but it can be compensated
and even used by combining different fibers with different optical and ther-
mal properties and that enables to measure both temperature and strain
simultaneously.
With both WDM and TDM methods combined, large objects like the wings
of an airplane, the vanes of a windmill and the cables of a hanging bridge
can be measured, but also for measuring less large object this method offers
ample possibilities.
8.7. Dynamic measurements of mechanical quantities 623

8.7.2 Velocity measurement

The measurement of velocity has long relied on the interaction of the mag-
netic field of a permanent magnet with a moving conductive element. Before
the onset of the electronic era the standard speedometer in cars and mo-
torbikes all worked according to the principle that a moving conductive
disk inside a magnetic field experiences a force that is proportional to the
velocity.
Presently still the majority of cars are equipped with speedometers with
needle pointers but they are no longer driven by moving disks in a magnetic
field but by a moving-coil Lorentz type rotating actuatorthat is controlled
by electronics that receive the velocity information from different sensors.
These sensors often consist of systems with an incremental character, re-
sulting in a frequency proportional to the velocity.
Incremental velocity sensors can be based on many principles. One example
is using a light source, a photo sensor and a wheel with holes that intermit-
tently blocks the light as function of the angle.
Another example is a rotating permanent magnet that induces an AC voltage
in a stationary coil with a frequency and amplitude that is proportional to
the speed according to the law of Faraday.
These principles will be presented more in depth in Section 8.8 with long
range optical encoders and this subsection will focus on the velocity sensing
that is needed in feedback systems to control damping.
It is often not necessary to use separate velocity sensors in precision posi-
tioning systems for damping in a full state feedback configuration. In most
cases damping is achieved with PD-control by differentiating the position
measurement signal as was explained in Chapter 4. In certain situations,
however, it is better to use a separate sensor, especially when the reference
for the damping action is undefined. This is the case with active vibration
isolation systems like will be presented in Chapter 9. In those systems a
damper connected between the isolated part and the vibrating environment
introduces an increase in the transmissibility of the vibrations at higher
frequencies. By using an active control system with a low stiffness actuator
and a velocity sensor, damping can be created without an increase of the
transmissibility as long as the sensor measures the velocity relative to a
reference that is completely free from vibrations.
Such a quiet reference is realised with an inertial velocity sensor like the
geophone that was originally designed to measure low-frequency vibrations
like earthquakes (Geo-phone = sound of earth).
624 Chapter 8. Measurement in mechatronic systems

Casing

N
Permanent Coils
Magnet =
Seismic
S
mass

Membrane

Figure 8.48: A geophone has a moving coil system supported by a compliant mem-
brane around a permanent magnet that induces a magnetic flux
through the coil. The movement of the coil will create an electro-
motive force, proportional to the velocity of the movement.
(©Aaronia AG, www.aaronia.com)

8.7.2.1 Geophone

The operating principle of a geophone is shown in Figure 8.48. A moving

mass, also called seismic mass, consists of a coil in two sections that is
supported inside a round ferromagnetic casing by means of a very compliant
spring membrane, resulting in a very low resonance frequency in the order
of 1 – 10 Hz.
The coil moves inside a magnetic field that is created by a permanent magnet
inside the coil, connected to the casing. The magnetic field enters one of the
coil sections from the inside and returns via the casing through the other
section from the outside. The coil sections are wound in opposite directions
and as a result their induced voltage will have the same sign. When added,
the total voltage from the coil is proportional to the its relative velocity to
the stationary magnetic field, according to Faraday’s law.
A geophone is used to measure velocity vibrations of an object by mounting
it to the moving object. This means that the output voltage is determined by
the transfer function from the velocity of the casing to the velocity difference
between the casing and the coils as the latter is proportional to the voltage
of the sensor.
It can be reasoned that under static conditions, at a frequency below the
first resonance frequency, the seismic mass of the coils will follow the casing
8.7. Dynamic measurements of mechanical quantities 625

Seismic mass mm xm

c ' x xc  xm k xc

Case mc

Figure 8.49: The dynamic model of a geophone as used for deriving the transfer
function.

and no voltage will be induced. Above the resonance frequency the seismic
mass will decouple and its amplitude will rapidly drop resulting in a velocity
difference and a measurable voltage. With the standard motion equations,
the influence of the internal damping on this system can be determined.
This damping is caused by eddy-currents in the coil former that often is
made from solid copper for increased mass and behaves like a short circuited
coil.
The relation between the velocity difference Δ ẋ = ẋc − ẋm and the velocity of
the casing ẋc is defined by the following equation using the rigid body model
from Figure 8.49:
d
( xc − xm )
Δ ẋ ẋc − ẋm dt s( xc − xm ) xc − xm
= = = = (8.51)
ẋc ẋc d sxc xc
xc
dt
With the second law of Newton the force balance on the moving mass m m is
defined:

m m s2 xm = c( sxc − sxm ) + kxc − kxm = ( cs + k)( xc − xm ) (8.52)

From the theory on transmissibility from Chapter 3 it is known that:

cs + k
xm = xc (8.53)
m m s2 + cs + k
Which combines with the previous expression into:
cs + k
m m s2 xc = ( cs + k)( xc − xm ) (8.54)
m m s2 + cs + k
And leads to the following concluding equation:
Δ ẋ ẋc − ẋm xc − xm m m s2
= = = (8.55)
ẋc ẋc xc m m s2 + cs + k
626 Chapter 8. Measurement in mechatronic systems

^
10

0
10

Magnitude (abs)
>^
10

>
10 Ns/m
c †`
c = 30 /mNs
>`
10
180

Z\
Phase [deg]

90

\

0
>^ > 0 1 ^
10 10 10 10 10
Frequency [Hz]

Figure 8.50: The Bode-plot of a geophone with a moving mass m m = 1 kg, a spring
with a stiffness of k = 1 · 102 N/m and two values of the damping co-
efficient c = 6 Ns/m (green) and c = 30 Ns/m (blue) shows the effect of
damping on phase and amplitude and the low-frequency bandwidth
limitation.

When the voltage sensitivity of the moving coil equals Vg = c g ( ẋc − ẋm ), where
c g in Vs/m equals the sensitivity constant of the geophone, the total transfer
function becomes:
Vg ẋc − ẋm m m s2
= cg = cg (8.56)
ẋc ẋc m m s2 + cs + k

This equation is a combination of the compliance response of a second-order

mass-spring system with a second-order differentiator that act together as
a second-order high-pass filter.
In this sensor the damping only influences the overshoot at the resonance
frequency and can be chosen such that the overshoot is zero without impact-
ing the high-frequency response.
In Figure 8.50 a Bode-plot of the transfer function of a typical geophone
with a moving mass m m = 1 [kg] and a spring with a stiffness of k = 1 · 102
[N/m] is shown. The figure indicates that the overshoot at the resonance
frequency is reduced with increasing damping but also the phase is changed.
8.7. Dynamic measurements of mechanical quantities 627

Depending on the application, these two aspects can be optimised such that
a flat response above the resonance frequency is obtained with an acceptable
phase response.
It is clear from this example that the values of the mass and stiffness be-
come impractical when the bandwidth is extended to even lower frequencies.
Under gravity conditions this spring would already need to elongate with
0.1 m to compensate the static gravity-force on the mass, which means that
a pre-stressed spring must be used.
This practical consideration limits the use of this geophone to a fixed orien-
tation in respect to the earth gravity as otherwise a very long coil would be
necessary, that would not be preferable because of the increased noise by
the increased resistance.
In practice the LF-bandwidth limitation can be compensated by a suitable
filter with integrator characteristics to increase the LF-gain, but this is
limited because of the resulting amplification of LF noise (drift).
The main benefit of this geophone for the application in vibration isolation
systems is the reference of the velocity to the inertial reference of a seismic
mass as the inertial universe is by definition free from vibrations.
An alternative method to obtain this inertial velocity signal is based on
using a sensor that even more directly refers to this inert reference by
measuring acceleration. The velocity can then be derived by integration of
the acceleration signal.
628 Chapter 8. Measurement in mechatronic systems

8.7.3 Accelerometers

Measurement of acceleration is in principle achieved by measuring force.

The relation between acceleration and force is given by the well-known
second law of Newton:
dv
F=m = m v̇ = ma (8.57)
dt
By using the force as an intermediate term, the acceleration is related to the
distortion of the spring which creates the possibility to measure acceleration
by using a combination of a seismic mass with a deforming support.
Three types of acceleration sensors or accelerometers will be presented
based on this principle.

• The closed loop feedback accelerometer.

• The piezoelectric accelerometer.

• The MEMS accelerometer.

8.7.3.1 Closed-loop feedback accelerometer

One way to measure a force is by compensating that force with the same force
by means of an actuator. When the force to current ratio of that actuator
is known, the current that is necessary for the compensating force is a

Posion
sensor

I
Controller Amplier R _ N
+

Permanent
Vs Magnet
S
Seismic
mass

Figure 8.51: Closed loop feedback accelerometer. The controller suppresses the
movement of the seismic mass relative to the casing with a current
proportional to the acceleration.
8.7. Dynamic measurements of mechanical quantities 629

reliable measure for this force. In Figure 8.51 this principle is used in the
closed-loop feedback accelerometer. It is build with the same components as
the geophone with the addition of a position sensor and a control system.
In this configuration the coils function as a linear Lorentz actuator instead
of a sensing element. The position sensor measures the position of the
moving coils relative to the casing. This position sensor can be any reliable
proximity detector as previously presented. The controller and the amplifier
are tuned to keep the displacement as small as possible by supplying a
current to the coils. This position feedback system creates an additional
stiffness by the proportional gain. The necessary current for that action is
proportional to the force on the mass due to the acceleration. By measuring
the current, a value is obtained for the acceleration
In Chapter 4 it was shown how such a position controller should be made in
order to remain stable by placing the poles of the system on the right location
in the complex plane with an additional D-control action. The system is a
servo-feedback system as the controller has to control the seismic mass to
follow the case just like with the optical pick-up unit of the CD player and
the same PID-control principle can be used here.
Because of the fact that this servo controlled accelerometer is a zeroing
measurement system, the position sensor inside the accelerometer only
needs to be accurate around zero. This implies that a relatively simple
sensor can be used with only a low level of the signal drift around zero
when static accelerations over a longer period must be measured. A second
benefit of this configuration is the additional virtual control-stiffness of the
spring due to the feedback that reduces the influence of any non-linearity
in the stiffness of the support structure in the frequency area where the
closed-loop gain is high.
The feedback principle with a high gain at low frequencies makes this
closed-loop feedback accelerometer especially suitable for the measurement
of static or low-frequency accelerations like gravity and the slow motion of
ships or air planes.
It is also a suitable option for low-frequency damping in vibration isolation
systems when the sensor is combined with an integrator to obtain the
inertial velocity. The complete sensor is however quite costly when compared
to the two other accelerometers so the application of a closed-loop feedback
accelerometer is limited to professional use at very low frequencies.
630 Chapter 8. Measurement in mechatronic systems

8.7.3.2 Piezoelectric accelerometer

In Section 5.6 of Chapter 5 the direct piezoelectric effect was explained.

The internal charge distribution of a piezoelectric material will displace
when deformed because of a special crystal structure with atoms that are
constrained in the lattice. The charge displacement manifests itself as a
voltage on the electrodes of the material.
Depending on the crystal orientation and the polarisation direction, the ma-
terial is sensitive for shear strain or compressive strain and both principles
are used.
Figure 8.52 shows three different configurations of a piezoelectric accelerom-
eter.
The shear design is built with the seismic mass around the piezoelectric
crystal. Similar to the shown example with one piezoelectric element also
other versions exist of this principle with for instance an assembly of three
flat piezoelectric plates in a triangular configuration. The main benefit
of this configuration is its low sensitivity for residual strain in the base
that can be caused by the mechanical mounting and by thermal induced
deformations. The rotational symmetric configuration reduces also the
sensitivity for accelerations in other directions.
The drawback of the shear design is a more complicated assembly process,
especially when a very small conical angle is needed. In assembly, excessive
stress must be avoided, which requires careful handling.
The compression design is quite straightforward. The seismic mass is
pressed onto a flat piezoelectric ring with a bolt of which the force is con-
trolled by a flexible washer.
The compression design is more sensitive for thermal transients, base strain
and accelerations in shear directions.
The bending design uses a special piezoelectric composition that consists
of two layers, called a bimorph. A bending force from the seismic mass will
create compressive strain in one layer and strething strain in the other layer.
Both layers are polarised in the opposite direction. As a result the voltages
of both layers add together to one voltage between the two flat sides of the
bimorph.
The main advantage of this principle is its high sensitivity-to-mass ratio.
Its main drawback is its relatively low resonance frequency that limits the
application to low frequencies as will become clear in the following section.
8.7. Dynamic measurements of mechanical quantities 631

Shear Design:

Cover
Seismic mass
Piezo ceramics
Post

Socket
Shear force
Base

Compression Design:
Cover
Spring
Seismic mass
Piezo ceramics
Bolt

Socket
Compression force
Base

Bending Design:

Fricon coupling
Cover
Piezo ceramics

Spring
Seismic mass
Bending force and damping piston
Base

Figure 8.52: Three different mechanical designs of piezoelectric accelerometers.

The shear design is the most modern version with better dynamic
performance.
(Courtesy of “Metra Mess- und Frequenztechnik Radebeul”)
632 Chapter 8. Measurement in mechatronic systems

Seismic mass mm xm

'x xc  xm xc
c k
Piezo xc
Case mc

Figure 8.53: The mechanical dynamic model of a piezoelectric accelerometer as

used for deriving the transfer function. The voltage is proportional to
the deformation Δ x of the piezoelectric sensing element and relates to
the acceleration ẍc of the case.

Mechanical frequency limitation

Mechanically a piezoelectric accelerometer behaves like a mass-spring sys-

tem with a resonance frequency that is determined by the seismic mass
and the stiffness of the piezoelectric crystal. The transfer function can be
derived with the help of Figure 8.53. The voltage Va of the accelerometer is
proportional to the displacement Δ x of the seismic mass that is caused by
the acceleration a = ẍc of the case.
Va c d ( xc − xm ) c d ( xc − xm )
= cv = = (8.58)
ẍc ẍc s2 xc
2
where c v in Vs /m equals the nominal voltage sensitivity of the accelerometer
and c d in V/m equals the voltage sensitivity of the piezoelectric crystal as
function for its deformation.
The equation is equal to Equation (8.56) of the geophone, divided by s2 . This
means that the mechanical transfer function of a piezoelectric accelerometer
is simply derived from that equation:
mm
cd
Va c d ( xc − xm ) cd mm k
= cv = = = m c (8.59)
ẍc s2 xc m m s2 + cs + k m 2
s + s+1
k k
For low frequencies where s is very small the voltage sensitivity equals:
mm
cv = cd (8.60)
k

The transfer function of Equation (8.59) is similar to the compliance re-

sponse of a mass-spring system. Below the resonance frequency the output
voltage is frequency independent. Around and above the resonance fre-
quency the transfer function deviates from the constant value depending on
8.7. Dynamic measurements of mechanical quantities 633

Seismic mass mm Seismic mass mm

+
IN Va VTh Va
C _ C
Piezo Piezo
Case mc Case mc

Norton Thevenin

Figure 8.54: The electrical model of a piezoelectric accelerometer is basically a

capacitor with a variable charge. This corresponds with the Norton
equivalent model where the charge equals the integral of the frequency
dependent current I N over time. The Thevenin equivalent model
is more practical to use because the voltage VTh is not frequency
dependent.

the damping. In practice the damping is quite low which means that these
mechanical dynamics limit the high-frequency bandwidth of a piezoelectric
accelerometer to around 0.3× the resonance frequency. For the shear and
compression design the support stiffness is rather high, giving practical
values for the usable maximum frequency of several kHz, depending in the
size. The bending design is more compliant. With the same size and mass
as with the other design principles the application of the bending design is
limited to high-sensitive low-frequency measurements until an approximate
maximum of a few hundred Hertz.
For situations where only a very low mass is allowed at the measurement
location, the bending design can still be a better alternative because with a
very small size the resonance frequency can become acceptably high again.

Electrical frequency limitation

Electrically a piezoelectric accelerometer behaves like a capacitor with a vari-

able charge that is determined by the acceleration ẍc . The change of charge
is equivalent to a current which means that the sensor can be modelled
according to Norton as a current source with a parallel capacitor. Unfortu-
nately the current I N is determined both by the acceleration amplitude and
by the frequency. This is explained as follows: The location of the charge
is determined by the force that is proportional to the acceleration. Only a
change in the charge location will cause a current. This means that only a
changing acceleration will cause the force, deformation and charge location
634 Chapter 8. Measurement in mechatronic systems

Sensor Cable …!<

+
amplier
1nF
100pF
+ Vo

Rp
VTh Va

Ri
_
_

Figure 8.55: The parasitic capacitance and resistance of a cable and the input
resistance of a buffer amplifier affect the sensitivity and frequency
response of a piezoelectric sensor.

to change with a corresponding current.

With c i being the charge sensitivity for acceleration, the Norton source
current I N becomes:
dQ d ẍc
IN = = ci = c i s ẍc = c i j ω ẍc [A] (8.61)
dt dt
This means that the Norton model is less practical due to this frequency
dependence. The Thevenin equivalent voltage source VTh can be derived
from the Norton equivalent by taking the voltage of the sensor when no
other load is applied than the internal capacitance:
c i j ω ẍc c i ẍc
VTh = I N Z = = = c v ẍc [V] (8.62)
j ωC C

This is a frequency independent voltage and can be detected by a non-

inverting amplifier. This amplifier would need to have a very high input
impedance because the source impedance is capacitive. The high output
voltage of a piezoelectric accelerometer allows the application of a unity-gain
buffer amplifier as signal conditioning element when the capacitive nature
of the sensor would be the only problem.

Charge amplifier

A simple non-inverting buffer amplifier does not solve all problems as is

indicated by the example of Figure 8.55 with some realistic values of the
components.
When the sensor is connected to an amplifier, a well shielded coaxial cable is
necessary to prevent external interference. This cable will have a parasitic
8.7. Dynamic measurements of mechanical quantities 635

10nF

100 M

_
1nF G

+
100pF
+

Rp
VTh
_ Vo

Figure 8.56: A piezoelectric accelerometer with a charge amplifier as signal con-

ditioning element. The charge amplifier reduces the influence of the
cable capacitance and other parasitic impedances by virtually ground-
ing the output of the sensor.

capacitance that acts as a load to the sensor and gives a proportional reduc-
tion of the sensitivity, depending on the ratio between the cable capacitance
and the sensor capacitance. Although this is a not frequency dependent
attenuation, these values should be exactly known in order to take this
attenuation into account with accurate measurements. This is only possible
up to a certain level due to the temperature dependence of these values.
Another problem is related to a possible parasitic resistance over the con-
nector that might be caused by humidity in combination with salt residues
due to manual handling.
Together with the input impedance of the amplifier and the capacitive source
impedance of the sensor this circuit forms a first-order high-pass filter. When
for this example the input resistance Ri equals 50 MΩ, this low-frequency
limit would be:
ωLP 1 1
f LP = = = ≈3 [Hz] (8.63)
2π 2πRC 2π · 5 · 107 · 1.1 · 10−9

because the total capacitance equals the impedance of the 1 nF and 100 pF
capacitor in parallel.
A second problem is caused by the inherent non-linearity of a piezoelectric
crystal.
As was shown with piezoelectric actuators these crystals exhibit a certain
level of hysteresis between the voltage and the deformation. Fortunately
this hysteresis is far less present in the relation between the charge and the
636 Chapter 8. Measurement in mechatronic systems

deformation as the hysteresis is caused by the capacitance. This means that

an optimal way to detect the signal of a piezoelectric accelerometer would
be to directly determine the charge. This can be achieved by “tapping” the
current from the charge displacement by means of the charge amplifier of
Figure 8.56, before it can cause a change in the voltage. This solution is
especially useful because the charge amplifier also solves the first problem
of the sensitivity for the cable capacitance and the parasitic resistance.
A charge amplifier is an operational amplifier in an inverting amplifier
configuration with only a feedback impedance that is mainly determined by
a capacitor. The sensor is directly connected to the inverting input of the
operational amplifier.
For the explanation of the principle first the effect of the very large 100 MΩ
feedback resistor can be neglected. The feedback capacitor for this example
has a value of 10 nF. The rules of an operational amplifier from Chapter 6 are
used to determine the gain of the inverting amplifier. The input impedance
Z1 is equal to the source impedance of the sensor with a capacitor value
of 1 nF and the feedback impedance Z2 equals the impedance of the 10 nF
capacitor that was chosen for this example. The transfer function of this
amplifier can now be written as follows
1
V0 Z2 j ω · 10−8
G= =− =− = −0.1 (8.64)
VTh Z1 1
j ω · 10−9

Although this does not look impressive from a magnitude point of view, the
effect of the parasitic impedances is cancelled as the amplifier will keep the
voltage at the minus input equal to 0 V, being the voltage at the grounded
plus input. This means that the sensor is virtually grounded and no voltage
is present over the parasitic impedances, while all current of the sensor
flows into the feedback impedance.
As this impedance is merely capacitive the voltage at the output is equal to
the stored charge in the feedback capacitor, divided by its capacitance value.
This charge is equal to the displaced charge of the piezoelectric accelerometer
for which reason this configuration is called a charge amplifier. This might
seem strange as the charge itself is not amplified and a “charge-to-voltage
converter” would probably have been a better name.
The reduction of the voltage gain to 0.1 is the direct result of a design
choice which is related to the maximum attainable value of the parallel
resistor over the feedback capacitor. In theory this resistor is not needed but
the bias current of the operational amplifier requires a non-infinite source
8.7. Dynamic measurements of mechanical quantities 637

^
10

Magnitude (abs)
0
10

>^
10

>
10
|J
Phase [deg]

>|J

>}J
>^ > 0 1 ^ 3  5
10 10 10 10 10 10 10 10
Frequency [Hz]

Figure 8.57: Typical bode-plot of a piezoelectric accelerometer with a seismic mass

of 20 · 10−3 kg and a support stiffness of 1 · 107 N/m showing both the
HF bandwidth limitation by the mechanics and the LF limitation
by the electronics. The amplitude plot is normalised to a value of
one in the pass-band, the response of the blue line corresponds with
a non-inverting buffer amplifier and the green line with the charge
amplifier.

impedance at 0 Hz. The bias current level determines the maximum value
of the parallel resistor together with the allowable DC offset error voltage at
the output. Special amplifiers with MOSFET inputs are developed to achieve
extremely low bias currents in the order of 1 pA which would give a DC error
of 0.1 mV with the indicated 100 MΩ resistor. Compared with the voltage
amplifier situation parasitic impedances parallel to this resistor can more
easily be avoided by careful assembly of the circuit, avoiding contamination
in handling and when necessary, even a protective coating can be applied,
to keep humidity away from the sensitive input of the amplifier. In this case
the low-frequency limitation of the bandwidth is only determined by the
feedback impedances resulting in the following:
ωLP 1 1
f LP = = = ≈ 0.16 [Hz] (8.65)
2π 2πRC 2π · 1 · 10 · 10 · 10−9
8

This is a factor 20 better than with the buffer amplifier. The sacrifice of
the attenuation is easily solved by an additional voltage amplifier after the
638 Chapter 8. Measurement in mechatronic systems

charge amplifier. Furthermore, High-frequency voltage noise is no problem

because of the capacitive feedback of the charge amplifier. The inherent low
impedance for high frequencies results in a low gain for this kind of noise.
In Figure 8.57 a typical Bode-plot of a piezoelectric accelerometer with the
two types of signal conditioning amplifiers is shown for comparison reasons.
The data correspond with the examples of Figure 8.55 and Figure 8.56.
The mechanical data of the corresponding sensor are a seismic mass of
20 · 10−3 kg and a support stiffness of 1 · 107 N/m giving a resonance frequency
of approximately 3.5 kHz. When using a charge amplifier, the maximum
usable bandwidth of this example ranges from ≈ 0.2 – 1000 Hz.
Because of the inherent limitation of the bandwidth on both the HF and the
LF side the piezoelectric accelerometer is used in situations where vibra-
tions in specific frequency bands are investigated like in the dynamic modal
analysis of mechanical structures.
It is a frequently applied instrument in determining improvements in me-
chanics used in precision positioning systems like stages in wafer scanners.
An additional benefit over other accelerometers is the very large dynamic
range which is also due to the low HF noise level originating from its capac-
itive nature. The only noise that still can cause problems is low-frequency
excess noise as that will be amplified by the feedback capacitor. Together
with the low-frequency limitation of the sensing principle itself, this low-
frequency noise is one of the problems when using this accelerator as velocity
detector with an additional integration.

8.7.3.3 MEMS accelerometer

Micro Electro-Mechanical Systems or MEMS are based on the application

of semiconductor production technology in the realisation of miniature
mechanical designs.
By using materials like Silicon and Germanium that not only have very
special electrical properties but also superior mechanical properties due
to their mono crystalline structure, dynamic systems with for instance
a very high Q (resonators and actuators) and an excellent linearity and
sensitivity (sensors) can be realised. The possibility to directly integrate the
signal conditioning electronics with the sensor itself on the same material
guarantee a relatively low sensitivity for external magnetic and electrostatic
fields.
For measuring purposes, the MEMS accelerometer, of which a very basic
principle is shown in Figure 8.58, might be the most widely used MEMS
8.7. Dynamic measurements of mechanical quantities 639

Strain gage

Leaf spring

Seismic mass

Cutout

Figure 8.58: MEMS accelerometer that consists of a seismic mass supported by 4

leaf springs provided with semiconductor strain gages. In spite of the
low mass this sensor has a very high sensitivity from 0Hz up till very
high frequencies. (Courtesy of Fredric Creemer, DIMES,TUD).

_
+

+
+

+
_ _ _

Figure 8.59: Vibrations in multiple directions can be measured with a MEMS

accelerometer by using two strain gages per flexure.

device ever. It is applied in safety devices like the air bag for a car, in
photographic cameras and Personal Digital Assistants (PDAs) to determine
the orientation towards gravity.
Like with the other accelerometers, the MEMS accelerometer applies a
seismic mass that is supported by a compliant mechanism, consisting of
several flexure leaf springs. The position of the mass can be measured both
with capacitive sensors and with integrated piezoresistive strain gages on
the supporting springs. The capacitive sensing principle is frequently used
in MEMS devices because of the simplicity. With comb structures the capac-
itance sensitivity can be increased and manufacturing these structures is
relatively easy. With capacitive sensing linear accelerations can be detected
in three orthogonal directions.
Measuring strain with integrated piezoresistive strain gages is an alterna-
tive that requires a bit more processing but it gives a very high sensitivity
640 Chapter 8. Measurement in mechatronic systems

and enables to sense even in six directions injcluding rotational accelera-

tions.
The principle is shown in Figure 8.59.

A representative example of a MEMS accelerometer is the ADXL345 from

Analog Devices of which the functional diagram is shown in Figure 8.60.
This fully integrated sensor works with the capacitive sensing principle
and can measure accelerations in 3 directions to a maximum acceleration
of 160 m/s2 with a resolution of 0.04 m/s2 , a non-linearity of 1 % of the full
scale, an offset of less than 1 m/s2 and a maximum 13 bit wide data-rate of
3.2 kHz.
It is to be expected that these accuracy and noise levels will continuously
improve over time which means that MEMS accelerometers will ultimately
replace most other acceleration sensors because of their versatility, reliabil-
ity, frequency range and low cost.

VS DD I/O

A D XL 3 4 5 Power
Management

Control INT1
Sense ADC and
Electronics Digital Interrupt
3- Axis Filter Logic INT2
Sensor

SDA/ SDI / SDI O

32 Level Serial I/O SDO/ ALT
FIFO Address
SCL/SCLK

GND CS

Figure 8.60: Functional diagram of the three directional MEMS accelerometer

ADXL345 showing the full integration of the sensor, the signal con-
ditioning and signal processing part that directly outputs the digital
measurement data.
(Courtesy of Analog Devices).
8.8. Optical long-range incremental position sensors 641

8.8 Optical long-range incremental position

sensors

Analogue measurement of distances over a long range is limited by the

signal to noise ratio of the applied sensor. A long range inherently conflicts
with low noise that is needed for a sufficient resolution. For that reason long
distances have always been measured in an incremental way for instance
by counting the number of revolutions of a wheel or the number of stripes
that were passed on a scale. In the age of electronics, incremental counting
is not done anymore with rotating numbered disks interconnected by gear
wheels but with digital logical circuits and microprocessors. When applying
electronics, the only way to realise a long range high resolution position
measurement system, is to provide a reliable signal with a sufficiently fine
spaced spatial frequency.
This signal can be generated in different ways. Take for example Figure 8.61
with a permanent magnet inside a coil that will give a sinusoidal voltage as
a function of the velocity and the angle of a toothed wheel that creates an
alternating reluctance path for the magnetic field. This sinusoidal signal
can be converted into a one bit periodic signal by means of a Schmitt trigger
before counting the pulses. This principle however only works at a certain
velocity, as at stand still no variable reluctance nor flux is observed. This

Vc
ω,ϑ

Vc

Figure 8.61: Incremental inductive rotation sensor giving a sinusoidal output volt-
age with a frequency depending on the rotation speed and the number
of teeth.
642 Chapter 8. Measurement in mechatronic systems

lamp

ω,ϑ
Ir

ϑ
Figure 8.62: Optical version of the incremental rotation sensor. The magnitude
of the irradiance I r from the lamp is modulated by the teeth of the
rotating wheel.

can be solved by using light instead of magnetism, as shown in Figure 8.62.

By just using a sufficiently small lamp and a detector at the opposite sides
of the rotating disk the detector will receive an alternating irradiance signal
as function of the angle of rotation. This principle is refined in the optical
encoders of the next section while the generation of an incremental signal
by interference is used in the laser interferometers of the last section of this
chapter. Both principles have presently achieved such high standards of
measurement that they even can compete in resolution with many short
range analogue sensors while still maintaining long measurement ranges,

8.8.1 Linear optical encoders

Optical encoders are widely applied both for angular and linear measure-
ments. This section will restrict to the linear versions, because in precision
mechatronic systems direct linear measurements are most commonly ap-
plied. Nevertheless, the same principles are applied in angular measure-
ment systems.
The most basic example of a linear optical encoder is shown in Figure 8.638 .
The working principle is originally based on the idea to compare the position
8 This figure, together with several of the other figures and the background information in

this section on optical encoders is gracefully provided by Heidenhain, a leading company in

high precision optical encoders.
8.8. Optical long-range incremental position sensors 643



‡
  Collimator lens

–


Scale

Scanning


Reference mark
Photo detector




Current
level
‘Dark’ Current

Grang Period osion

Figure 8.63: Basic configuration of a linear optical encoder. A scanning reticle with
a grating is illuminated with parallel rays of light. The shadow of the
reticle is projected on a scale with an equal grating peiod connected to
the moving object. A photo detector detects the resulting irradiance,
that alters periodically as function of the displacement.

of one grating with well defined equidistant slits with the position of another
grating by means of shadow projection.
A parallel beam of light, originating from a light emitting diode and a colli-
mator lens, is projected on the first grating. The shadow of this grating is
projected on the second grating which has equal periodicity and is designed
such that it obscures the light when shifted with half a period in respect
to the first grating while transmitting the light when the gratings are in
phase. This causes the irradiance to be modulated as function of the position
644 Chapter 8. Measurement in mechatronic systems

Collimator lens


‡
  –



Scale




Reference mark F!

phot etector

Figure 8.64: Four field incremental optical encoder giving robustness, directional
and absolute position information with high resolution by interpola-
tion.

with an essentially triangular spatial waveform. As will be explained later,

a more sinusoidal spatial waveform is preferred. This can be created by
spatial low-pass filtering of the triangular waveform. In Chapter 7 the effect
of low resolution imaging optics was presented to be equivalent to low-pass
spatial filtering. A shadow is only sharp under two conditions. Firstly the
light must be created by a point source. Secondly the details need to be
large relative to the wavelength of the light in order to prevent diffraction
effects. With a larger surface of the light source a small grating period
pattern becomes blurred, resulting in a more sinusoidal spatial waveform.
With high resolution encoders this diffraction character determines the
functionality, but for simple encoders this is only a side effect.
One typical property of incremental measurement systems is the relative
character of the data. At a certain known position the counter is set to zero
and, as long as all periods are counted during movement, the new position
is determined relative to the zero position. In case periods are lost due to
noise, the absolute position has to be traced back again to reset the counter.
Several possibilities to create such an absolute position reference are applied
in linear encoders. All options are based on an additional separate reference
mark that is simultaneously exposed and can range from one simple set of
8.8. Optical long-range incremental position sensors 645

4 1 2 3 4 1
A C Scanning
D recle
B A
1
Measuring
scale

B
2

Reference
mark
C
3

D
4

Figure 8.65: Sensor signals in the four field encoder as function of the relative posi-
tion of the scales. At the left side the four positions of the measuring
scale are shown relative to the scanning reticle, corresponding with
the numbered vertical lines in the signal plot at the right side.

slits to a binary coded pattern that is read out by a CCD camera. As long
as a transition of the reference mark uniquely points into a specific period
of the incremental encoder signal, the absolute position can be deduced by
setting the incremental counter accordingly.
The single field encoder has several problems that need to be addressed.
First of all it is impossible to determine the direction of movement. Further-
more, the influence of electrical noise (dark current) and contamination of
the gratings becomes large at the low signal level positions.
Figure 8.64 shows an enhanced version of this principle that solves these
issues by the following measures:

• Redundancy by using two sections per phase for the same information
but working in counter phase to have high signal levels at all positions.

• Two pairs of segments shifted 90◦ in spatial phase to give directional

information.
646 Chapter 8. Measurement in mechatronic systems

I90° I90°

M I0°
0° 90° 180° 270° 360°

I0°

I90q
M arctan
I 0q

Figure 8.66: Interpolation with two sine functions at 90◦ phase shift shown graphi-
cally by means of a Lissajous figure. The angle ϕ is proportional to the
position within one cycle and can be calculated from both coordinate
values.

The beneficial effect of the redundancy is obvious. To better explain the

effect of the phase shifted sections, Figure 8.65 shows the signal levels of the
different fields, corresponding with the four positions where one of the fields
is at its maximum irradiance level. It is shown that signal A will decrease
while signal C has a positive sign when moving to the right from position 1
to position 2. A decrease of A with a negative sign of C (or more reliably a
positive sign of D) corresponds with a movement from 3 to 2 which is in the
left direction. This method gives a reliable direction information, but the
total system can even be improved by a vectorial combination of all signals.

8.8.1.1 Interpolation

In principle the value of the position can be determined by digitising the

current value of the field sensor by means of a comparator but then the
resolution would not be better than one grating period.
The spatial sinusoidal character of the signal creates the possibility to make
a reliable estimation of the intermediate positions. This estimation process
is called interpolation and its principle is shown in Figure 8.66.
From the signals of the four field encoder two signals are created, one signal
is the difference of A and B and the other is the difference from C and
D. These signals can be plotted in a four quadrant vectorial representa-
tion, called a Lissajous plot after the French mathematician Jules Antoine
8.8. Optical long-range incremental position sensors 647

Signal deviaon Lissajous Interpolaon error

gure

O<set I 90 o

O<set I0 o

Amplitude error
I0 o / I90 o

Phase angle
error

Signal wave
error

Figure 8.67: Deviations of the signals in their shape, magnitude and average value
cause errors in the interpolation.

Lissajous (1822 – 1880). The phase angle ϕ appears to be equal to:

I 90◦
ϕ = arctan (8.66)
I 0◦

where I 0◦ equals the combined signal current of the sensor parts with their
extreme values at 0◦ + n · 180◦ while I 90◦ equals the combined signal current
of the sensor parts with their extreme values at 90◦ + n · 180◦ .
Originally this calculation was done in the analogue domain by electronic
addition of both signals with different amplification ratios, followed by a
comparator. Nowadays the interpolation is far more easily achieved in the
digital domain by using high resolution AD converters for digitising the
signals follwed by a direct calculation of the angle.
Under the condition of a fairly well defined sinusoidal shape of the signal
an interpolation factor of 1024 until 4096 can be achieved.
Figure 8.67 shows a qualitative overview of the errors that are caused by
deviations from the ideal waveforms. A low interpolation error can only be
achieved by means of careful mounting and positioning of all the optical
parts in the measurement system.
648 Chapter 8. Measurement in mechatronic systems

Structured
Scale Window photo detector

Scanning
 Condenser lens

–



‡
 

Figure 8.68: The same properties can be obtained with a reflective scale as with a
transparent scale. This alternative allows to use a very long flexible
metal band scale that can be glued to the moving object. Directional
information is obtained at the photo detector by a slight difference in
periodicity of the gratings, causing a Vernier effect.

8.8.1.2 Vernier resolution enhancement

The previously described configurations have two important drawbacks. The

first is the need for a transparent scale that requires mounting space of
the sensor parts at both sides while complicating the connection with the
moving object. The second drawback is the need for multiple sensor fields to
realise both robustness and directional information. The first issue can be
solved quite simply by using the reflective grating as shown in Figure 8.68.
The configuration looks quite similar to the one field transparent grating
and adds the benefit of the integration of the illumination and sensor part.
This principle can be used with thin metal reflective scales that can be glued
to the moving object. The inherent problem of the one field sensor regarding
failing directional information and lack of ruggedness to contamination has
been solved in a different way. By using two scales with a slightly different
period a Vernier effect occurs, named after the French mathematician Pierre
Vernier (1580 – 1637) who invented the principle. Like with a Vernier
8.8. Optical long-range incremental position sensors 649

Recle 1: 40 Lines in 0.8 mm

Recle 2: 41 Lines in 0.8 mm Two posions shi$ed half a grang period

Vernier period ( 800 μm)
o o o o o o o o o o o o
0 90 180 270 0 90 180 270 0 90 180 270

Mulple vernier periods

Figure 8.69: Using Vernier gratings to obtain both directional and interpolation
information with a one field CCD camera sensor.

calliper, this effect increases the position sensitivity and by reading out
the signal with a structured photo detector like a CCD camera sensor, both
directional and interpolation data are obtained.
The Vernier effect is illustrated in Figure 8.69. The upper half of the figure
shows the effect of two gratings that differ one line over the total width of
the grating of 40 lines. The corresponding Vernier ratio is 40 with a Vernier
period that equals the shown size of the grating. The upper right image
shows the visual effect when the two gratings are on top of each other in
two situations that differ in a relative shift with half a grating period. It
is clear that the dark and light zones have moved half the Vernier period
giving a factor 40 more sensitivity over the movement of the grating. The
moving direction of the dark and light zones is related to the relative moving
direction of the gratings which enables to determine that direction from the
signal of the CCD camera sensor.
Several Vernier regions can be combined to increase the robustness against
contamination as shown in the lower half of Figure 8.69. As the entire
surface of a sensor can only be used once, the designer can decide either for
a very large Vernier ratio or for more Vernier regions for robustness. This
650 Chapter 8. Measurement in mechatronic systems

is an optimisation where the outcome depends on the application.

With the resolution enhancement by the Vernier effect also an additional
interpolation of the position can be done to a factor that depends on the
resolution and noise of the applied CCD camera sensor.

8.8.1.3 Interferometric optical encoder

The search for a continuous improvement in resolution, accuracy and robust-

ness has resulted in manufacturing improvements to continuously decrease
the grating period of the applied scales and eventually the grating period
approached the wavelength of the light. As a consequence, simple shadow
projection methods were no longer applicable and diffraction and interfer-
ence effects had to determine the functionality.
The last example of a linear optical encoder is fully based on this effect of
interference of light, using the phase gratings as described in Chapter 7
Section 7.4.3.2. This interferometric optical encoder enables resolutions into
the sub-nanometre region and is applied in precision machines like wafer
scanners.
The explanation of this more complex encoder principle will be done in
three steps. First a global overview of the system is given. The second step
explains the phase shift at the moving scale as function of the movement. In
the third step all phase relations are determined that create the interference
pattern on the photo detectors.
For the first step, Figure 8.70 is used to show the path of the light in three
dimensions. The light source must be monochromatic with a high radiance
to guarantee a clear phase relation of the light at the source. The coherence
length does not need to be very long because the optical path length of all
beams in the system is designed to remain equal. Small path differences
may be induced by mechanical tolerances. Therefore the light source may
have a rather short coherence length as long as these tolerances. A light
emitting diode with a coherence length of ≈ 15 μm is often sufficient but
for more critical applications a solid state laser can be used. The diverging
light from this point source is collimated into parallel rays by a positive lens
and directed to the scanning reticle. The scanning reticle consists of a phase
grating with a depth such that 75 % of the light is equally diffracted in three
orders. The three orders are directed to the scale on the moving part with
a reflective phase grating. The depth of this second phase grating is such
that most of the light is diffracted into the 1st −order directions. This has
different consequences for the diffracted orders from the scanning reticle.
First of all the 0th −order from the scanning reticle will be reflected to a
8.8. Optical long-range incremental position sensors 651

 "
<
-1 0 +1 '
Scale
Lig
‡
diode or
Collimatorlens solid state laser

–




 3 * Photo
detectors

Figure 8.70: A phase grating with a periodicity in the order of the wavelength of
the light result in a very high resolution and a three phase position
signal because of the interference at the detectors.

+1st − and −1st −order at the scale. Secondly the ±1st −order directions from
the scanning reticle will be reflected back orthogonal to the surface of the
scale according to the fact that light, entering under an angle corresponding
with the 1st −order, will be diffracted in the direction of the 0th −order.
Because of the symmetry and the equal diffraction angles, the resulting four
beams are recombined at the scanning reticle where they will undergo a
third diffraction with equal 0th - and ±1st −orders. Only two of these will be
used further in the system. The third is not shown in the drawing and will
in reality be projected by the collimator lens on an area without a sensor.
The collimator lens that originally created the parallel rays from the light
source will create one focused spot from all returning parallel rays on a posi-
tion that is determined by the angle of these rays according to the principles
of optical imaging. Depending on the angles the rays will come together at
three photo detectors.
In the previous chapter on optics, Equation (7.26) showed that the radiance
of the interference of two beams with an equal irradiance is proportional
to 1 + cos ϕ with ϕ being the phase difference of the two beams. This means
that the phase differences between all beams at the point of interference
need to be determined.
652 Chapter 8. Measurement in mechatronic systems

-1st +1st
order order

1 λ Phase shi$ @
1 period displacement
?
'
displacement
x
λ/4 Grang period x
p ™' Posive
phase shi$ φm(x) phase shi$ φm(x)

Figure 8.71: A displacement of the phase grating causes a phase shift of the
diffracted 1st and −1st −orders with a sign equal to the sign of the
order. One grating period displacement gives one wavelength phase
shift of the diffracted orders.

For this reason in the second step of reasoning the effect of a displacement of
the reflective phase grating on the phase of the reflected 1st - and −1st -order
light is determined with the help of Figure 8.71.
During the displacement of the grating with one period the phase of the
diffracted orders have also shifted exactly one period of the light. This means
that the incremental phase shift in the refracted orders as function of an
incremental displacement Δ x is equal to:
Δx
Δϕm ( x) = 360 [deg], or:
p
Δx
= 2π [rad], (8.67)
p

with a sign depending on the direction of movement relative to the direction

of the order. With the indicated directions in Figure 8.71, the sign is equal
to the sign of the order.
The third step uses the graphical representation of Figure 8.72 where a few
simplifications are applied in respect to the real system to enable a more easy
understanding of the total phase relationships between the different beams.
First the collimator lens is left away. This is allowed as the collimator lens
only creates the parallel beams and concentrates them again on the sensors.
The phase impact of this lens, if any, is equal for all beams. Secondly the
reflective grating is drawn as if it is a transparent grating. This is only for
8.8. Optical long-range incremental position sensors 653

graphical purposes as otherwise the reflected and incoming beams would

overlap. The part right from the moving scale must be seen as mirrored9
from the left and the second scanning reticle is in fact the same element
as the first. The phase relations shown correspond with the real reflective
grating. The third simplification is the omission of the phase shift due to the
optical path length. Also this is allowed as the optical path length between
the gratings is equal for all beams. The last simplification is the omission
of the diffraction orders that are not used. This has some consequences as
it might seem that not all orders will keep the same radiance level. Where
necessary the effect is separately mentioned.
Starting at the left side, first the incoming light from the light source is
diffracted by the scanning reticle into three orders with an equal irradiance
level of 25 % each. The remainder of the light is emitted in higher orders.
It was shown in Section 7.4.3.2 of Chapter 7 that such a phase grating will
cause a phase shift of +60◦ at both 1st -orders and a phase shift of −60◦ for
the 0th -order. Like presented before, the moving scale is a 50 % diffracting
scale that does not introduce a phase shift by the diffraction but only by
the incremental movement Δ x. This incremental phase shift is −Δϕm ( x)
for a diffraction angle opposite to the motion direction and +Δϕm ( x) for a
diffraction angle in the motion direction. As shown in step one, the orig-
inal 1st -orders from the first transition through the scanning reticle will
be diffracted by the moving scale into the 0th -order direction orthogonal
to the scale and another order that is not used. The orthogonal order is
recombined at the scanning reticle with both 1st -orders that originated from
the 0th -order of the first transition at the scanning reticle.
The real important phase shifts happen at the second transition through
the scanning reticle. All beams will be diffracted in three orders of which
only two are used. Depending on the diffraction angle there will be an
additional +60◦ phase shift when diffracted under an angle and a −60◦ phase
shift when running in the same direction as the incoming beam. After the
second transition of the scanning reticle four beams are obtained that each
consist of two beams with a different phase, each determined by the addition
of all phase shifts over their respective optical path. The phase difference
between both beams in a pair gives an irradiance value at the photo detec-
tors with a cosine function of double the phase shift by the moving scale.
This doubling is a useful increase of the sensitivity. The three irradiance
signals of the detectors have a 120◦ phase relation with each other because
of the different 60◦ phase shifts. This three phase measurement signal is
9 This method for graphical presentation of an optical system is often applied in situations

where mirrors are applied.

654 Chapter 8. Measurement in mechatronic systems

even more optimal for interpolation, than the 90◦ shifted signals of the four
field encoder. There is even less need for inverse sensors to overcome the
noise at the low irradiance areas because at any position sufficient signal
with a high irradiance level is available to increase the signal to noise ratio.
There is also no need for complicated structured sensors with Vernier en-
hancement. Furthermore the sensitivity for contamination is acceptable,
as contamination will only cause a reduction of the irradiance by scattering
of light but the phase relationship is not affected. The irradiance levels
of the different beams will however influence the interpolation and a very
strict adherence to narrow tolerances is required when the interferometric
encoder is used at extreme levels of interpolation.

8.8.1.4 Concluding remarks on linear encoders

Only a limited overview is presented of the numerous methods to achieve

incremental position information by means of a grating scale with dedicated
sensing optics.
An interesting method that for instance not has been mentioned is the two
dimensional sensing principle of the optical computer mouse. This inexpen-
sive system uses a CCD sensor to determine the displacement of a surface
by means of its surface structure. This structure fulfils the role of grating
and the pattern can be of any shape. Improvements with a laser source to
enhance contrast have made these sensors quite reliable and sensitive up
to sub-micrometre levels. Unfortunately this method is very sensitive for
distance variations, which could be solved with telecentric optics but that
will dramatically increase the cost. Nonetheless, it surely is an option for
future sensors in more advanced applications.
An important benefit of optical encoders when used in high precision po-
sitioning systems is the small distance between the sensor and the scale.
This largely reduces the sensitivity for changes in the index of refraction of
the intermediate air. This problem is more prominently present in the laser
interferometers that are presented in the following section.
With a further enhancement of the interferometric encoder principle it is
even possible to measure in more directions with one integrated sensor and
a two dimensional grating. In Chapter 9 such a sensor is used to measure
the position of the wafer stage in a wafer scanner.
Optical encoder measurement systems can be applied to measure very large
displacements with a high resolution as long as the scale can be mounted on
the moving object and the sensor can keep track of the scale. This limitation
is not present with measurement systems that are based on the determina-
8.8. Optical long-range incremental position sensors 655



Photo detector (R)
 r,R v1+cos(2(
I m
(x)-120))
3 orders 25 %
o +60- m(x)+60
Ž«*`J
$  =2( m(x)-120)
x -60+ m(x)-60

+60- m(x) +60- m(x)-60

-60+ m(x)+60
Plane
wavefront +60  =2 m(x) x

-60+ m(x)
-60
Photo detector (S)
-60- m(x) Ir,Sv1+cos(2 m(x))

 =2 m(x)
+60 

-60- m(x)+60

$
+60+ (x) +60+ m(x)-60
m

-60- m(x)-60
'

' +60+ m(x)+60  =2( m(x)+120)
+/- 1 order 50 %
r,T v1+cos(2( m(x)+120))
™
$ I
Photo detector (T)

Figure 8.72: Graphical representation of the phase of an interferometric optical

encoder. The phase shift ϕm caused by the moving scale gives an
irradiance pattern I r at the photo detectors that gives a signal of two
spatial periods for one period movement of the scale. The 60◦ phase
shifts at the transparent grating create 120◦ spatial phase difference
between the three sensors, resulting in a three phase measurement
signal (R,S,T).

tion of the optical path length differences by means of laser interferometers.

656 Chapter 8. Measurement in mechatronic systems

8.8.2 Laser interferometer measurement systems

In the previous section it was demonstrated that incremental information

on the position can be obtained by interference. In laser interferometry the
interference is not caused by combining diffracted beams of light but by
combining beams of light that passed through optical paths with a different
length. Laser interferometry is an extremely wide field of expertise with
countless configurations. Two quite different application areas can be dis-
tinguished for geometric measurement.
First of all laser interferometry is the “de facto” standard for the mea-
surement of the surface shape topology of optical elements and precisely
machined objects. By comparison of the phase of a wavefront of light that is
reflected from the measured surface with the phase of the wavefront that
is reflected from a known surface, deviations of up to several picometres
can be determined. These measurements are in principle static and require
extreme precautions for vibrations and stability.
For mechatronic systems that deal with controlled motion, the second appli-
cation area that focuses on displacement measurement of solid objects is
more important. For this reason this section presents only displacement
measurements of moving objects with even only the most frequently used
principle, the interferometer based on the Michelson configuration. Though
only one of many possible configurations, its understanding gives sufficient
background knowledge to understand also the physical principles of the
other interferometers.
Figure 8.73 shows an overview of the main components of a laser interfer-
ometer displacement measurement system10 . A laser source is used with a
large coherence length, because of the interference principle. The phase re-
lationship of the light needs to remain coherent over the entire optical path
which is far longer than with the interferometric encoder of the previous
section. The real measurement of the distance to a moving object takes place
at the interferometer itself. The source of light is split in the interferometer
in two beams with a known phase and frequency relationship. One beam
travels to a reference reflector and the other to a reflector to the object of
which the distance is to be measured. After returning to the interferometer
both beams are mixed and interfere. The resulting interference pattern
is an irradiance-modulated light signal as function of the phase difference
between the interfering beams. This interference signal is less critical in
10 Several of the used figures and background information in this section are gracefully

provided by Agilent Technologies, a leading company in extreme precision multi-axis laser

interferometer displacement measurement systems.
8.8. Optical long-range incremental position sensors 657

Beam delivery

Interferometer
Laser source
Moving object

Reference
Signal cable
Plane
Opcal Fibre Opcal
mirror
Part
Receiver
Electrical
Part
Receiver

Measurement Signal Posion

Signal cable Processing inf

Figure 8.73: The main components of a laser interferometer measurement system.

The reference signal cable is only used with the heterodyne principle.

respect to wavefront errors so it can be transported to a photo detector via

an optical fibre preceded by an optical part that focuses the light into the
fibre. The fibre is connected to the receiver electronics that converts the
irradiance of the light into an electrical signal and sends it to the signal
processing element.

8.8.2.1 Homodyne distance interferometry

The most simple version, the homodyne interferometer is shown in Fig-

ure 8.74 where the term “homodyne” refers to the use of light with only one
frequency. The monochromatic coherent light of a laser enters the system
from the left and is split in two parts with equal irradiance by a 50 % re-
flecting mirror. The first part, the reflected light, goes towards the reference
mirror, where it is reflected back to the 50 % mirror that will again split
the light in two equal amounts. One half of the light is reflected back to
the source and not used anymore. The remaining light with an irradiance
level of 25 % of the level at the input of the system, will pass the mirror and
reach the photo detector.
The second part, the transmitted light from the first encounter with the
50 % mirror, goes towards the measurement mirror and likewise with the
first part, this beam will be reflected back to the 50 % mirror and again half
658 Chapter 8. Measurement in mechatronic systems

Reference mirror
parallel to wavefront
Measurement mirror
Plane wavefront parallel to wavefront
Incident light
50 %
100 %

50 %
50 % mirror x
225 % Fringes
Photo detector

Figure 8.74: Basic homodyne Michelson Interferometer with a 50 % reflecting

mirror, a reference mirror and a measurement mirror. The reflected
light after the reference and measurement mirror is dashed to show
the paths. At the right, the irradiance pattern is shown with fringes
that occur at the photo detector plane when the wavefronts are not
running parallel. Note the phase jumps of 0.5λ at the mirrors when
reflecting from low refractive index to high refractive index material.

of the light is lost and transmitted back to the source with only 25 % of the
light left to be reflected at the mirror and reach the photo detector. When
all optics are well aligned, the wavefronts of both beams are still parallel
and interference takes place on the photo detector, either constructive when
both beams are in phase or destructive when both beams are out of phase
or partial in the situation in between. The reference mirror is stationary so
only the phase of the reflected beam from the measurement mirror is mod-
ulated by the distance. A displacement results in a sinusoidal irradiance
modulation of the interference at the photo detector according to the 1 + cosϕ
relation from Equation (7.26) in the previous chapter on interference.
Because of the double-passing of the trajectory from the interferometer to
the measurement mirror and back the incremental phase shift related to
an incremental movement Δ x of the measurement mirror equals:
2π N 2π f p N n
Δϕm ( x) = Δx = Δx [rad ], (8.68)
λ c
where N equals the interferometer constant defined as the number of single
trajectories that the measurement beam passes to and from the object, in
this case N = 2. λ, f p and c are respectively the wavelength, the frequency
and the propagation velocity of the light in vacuum while n equals the
refractive index.
During a movement, the irradiance at the photo detector equals zero in
8.8. Optical long-range incremental position sensors 659

every period of the modulated interference. Apparently the energy from the
source has disappeared at those positions. In reality the law of conservation
of energy abides. It appears that the light of both beams that is sent back
to the source shows constructive interference when the light at the detector
shows destructive interference and the other way around. This can be
checked when realising that reflection at a surface from a low refractive
index medium to a high refractive index material introduces a sign reversal
of the phase. This happens at the reference mirror, the measurement
mirror but only at one side of the 50 % reflecting mirror, the side on
the mirror glass with the reflective aluminium coating. When counting
the phase sign reversals at the mirrors, tracing both beams, it shows that
the beam at the photo detector that came from the reference mirror has
undergone two sign reversals while the beam from the measurement mirror
had undergone one sign reversal. This means that the sign of the phase
difference between both beams due to the displacement is reversed.
The beams that are reflected back to the source have either one sign reversal
from the measurement mirror while the beam from the reference mirror got
a total of 1.5 sign reversals. This means that both beams have the same sign
reversal and as a consequence the sign of the phase difference between
both beams due to the displacement is not reversed. This means that the
phase difference in the beams to the sensor and to the source differ exactly
one sign reversal. As a consequence, constructive interference in one of the
two beams corresponds with destructive interference in the other beam.

Directional information and interpolation

The described basic homodyne interferometer is comparable with the basic

optical encoder with only two gratings and one sensor. At standstill the sig-
nal from the detector is a constant value and with a movement a sinusoidal
frequency is present that is proportional to the velocity. Unfortunately, like
with the basic encoder, there is no directional information and also at low
irradiance levels the noise impairs the accuracy.
One method to solve these problems is comparable with the previously pre-
sented Vernier encoder method with a structured sensor. One could say
that the Vernier effect is caused by an imperfection in the periodicity of the
gratings. Along the same line of thoughts the solution in the interferometer
is based on an imperfection in the orientation of the mirrors.
A perfect constructive or destructive interference happens only when the
wavefronts are perfectly flat and aligned. As soon as one of the mirrors is
slightly tilted, fringes are observed at the photo detector as shown at the
660 Chapter 8. Measurement in mechatronic systems

right side of Figure 8.74. These are caused by differences in path length
due to the tilt11 and their spatial frequency is proportional to the tilt angle.
With a single photo detector these fringes would result in a constant average
irradiance over the surface of the sensor when the fringe period is smaller
than the diameter of the sensor. In a normal interferometer this would
require a smaller sensor at the sacrifice of sensitivity but it is far better
to use a structured sensor like a CCD camera. In that case the movement
direction can be detected as these fringes will move over the surface of the
detector when the measurement mirror moves in the x-direction. A dis-
placement of Δ x = λ/N will cause a shift of the fringes with one fringe period.
This enables a good resolution with sufficient interpolation capability of the
spatial sinusoidal signal. An interpolation factor of hundred already yields
a resolution of 3 nm with a Helium-Neon laser of 632 nm.
Although with further improvements on interpolation even better results
are obtainable, this principle still has some additional drawbacks that have
challenged people into different concepts to solve these. First of all there is
the sensitivity for the tilt in the mirrors and secondly the signal frequency
ranges from 0 Hz until very high frequencies depending on the movement
velocity.

Sensitivity for tilt

Sensitivity for the tilt of the reference mirror can be avoided by a very stable
mounting of this mirror in the interferometer. Tilt of the measurement
mirror will, however, directly influence the observed effect on the CCD
sensor. This would limit its use to systems with only one degree of freedom
on a perfect straight guiding mechanism that does not introduce angular
movements.
This problem can be avoided by replacing the measurement mirror by a
cube-corner retro-reflector, also called just a cube-corner or corner-cube. Also
the cats-eye can serve this purpose. Both optical systems have in common
that they reflect the light back (retro) in the same direction as where it came
from without any optical path difference for parallel rays, thus keeping the
reflected wavefront flat and parallel to the incident light. Figure 8.75 shows
both optical systems.
A cats-eye consists of a positive lens with a spherical mirror in its focal point.
The radius of the spherical mirror is equal to the focal length of the mirror
11 This effect of these fringes is primarily used when measuring surfaces with interferometers.

The fringe shape and distance represents the value of the surface slope relative to the reference.
8.8. Optical long-range incremental position sensors 661

R=f

Cats-eye Cube-corner

Figure 8.75: A cats-eye and a cube-corner retro-reflector. A cats-eye consists in its

basic shape of a positive lens with a spherical mirror in the focal point
of the lens. The cube-corner is a prism with three mutually orthogonal
reflecting surfaces. With both optical systems, the light rays will be
reflected in the same direction as where they came from but shifted to
an opposite position of the chief ray through the central point of the
optic entry pupil. This point symmetry is shown in the photographic
picture from three cube-corners with a different orientation where the
reflection of the lens always has the same orientation to the camera.
The text on the lens, however, is mirrored.

causing the chief ray through the centre of the lens to always be reflected
back in the same direction. As the reflecting surface is positioned in the
focal point of the lens all parallel rays will be focused on one point on the
spherical surface and reflected back under the same angle with the chief
ray which is always orthogonal to the spherical surface.
After the lens the reflected rays will be re-collimated parallel to the incident
rays. Their position is however mirrored in respect to the point of incidence
of the chief ray. For this reason this system is called point-symmetric. The
main drawback of a cats-eye is the fact that a simple two element cats-eye
can only be used with small apertures and approximately paraxial rays as
otherwise spherical aberration and coma will cause wavefront errors. With
aspheric optics or multiple optical elements these errors can be avoided at
662 Chapter 8. Measurement in mechatronic systems

increased cost.
The cube-corner has less problems in that respect because it does not apply
curved surfaces. It consists only of a monolithic piece of transparent material
with three mutually perpendicular mirror surfaces that internally reflect
all rays back in the same direction. Like the cats-eye, the cube-corner is a
point-symmetric retro-reflector.

Wide frequency range

The wide frequency range of the interference signal is typically problematic

for low signal level situations as noise can more easily be filtered out when
a limited frequency range is used. This low signal-level situation is typical
the case in a multi-axis measurement system where one laser source has to
supply a multitude of interferometers. Encoders often consist of a complete
system per direction, including the light source and sensor, and in that case
the noise level is less a problem.
It is especially important to avoid a DC value of the signal as all DC measure-
ments show some level of DC drift over time. In Section 8.4.2 it was shown
that by modulating a low-frequency signal, the spectrum of the measurement
signal can be shifted to a higher frequency. With a homodyne interferometer
this modulation can be realised by adding a small high-frequency motion
oscillation to the reference mirror, resulting in a modulated signal at the
detector. This can be synchronously demodulated after amplification.
It is also possible to use polarisers and polarisation dependent optical ele-
ments to create two signals with 90◦ phase difference to get the directional
information by quadrature detection, a method that enables a plotting of
the signal as a vector in a four-quadrant plane like the Lissajous plot of
Figure 8.66 that was introduced with the theory on interpolation of encoder
signals.
Ultimately, heterodyne interferometry appeared to be a better solution for
high precision displacement measurements with laser interferometry in
actively controlled positioning systems.

8.8.2.2 Heterodyne distance interferometry

The heterodyne distance interferometer solves the above mentioned prob-

lems by using a laser source with two different beams, one for the reference
and one for the measurement path that have a different frequency and a
mutually orthogonal linear polarisation direction. In most cases the beams
are combined into one beam at the input of the system but the principle can
8.8. Optical long-range incremental position sensors 663

also work with fully separated beams as long as they have a known phase
and frequency relationship.
The term “heterodyne” refers to the frequency difference. This split-frequency
f s can be in the order of several Megahertz. One method to generate a laser
source with such a dual frequency spectrum is based on the Zeeman effect,
named after the Dutch physicist and Nobel prize winner Pieter Zeeman
(1865 – 1943) who discovered the effect. When the laser cavity of a Helium
Neon laser is inserted in a permanent magnetic field, the normally single
laser-frequency radiation is split in two radiation parts with an equal irra-
diance but with two different frequencies and polarisation states. The two
frequencies are symmetrically spaced around the original frequency, called
the centre-frequency f c . The split-frequency is proportional to the magnetic
flux density. The polarisation direction of both radiation components is
circular with opposite directions. This circular polarised light is converted
into two orthogonal linear-polarised radiation components by a birefringent
plate with one quarter of a wavelength delay between the orthogonal polari-
sation directions. This quarter wave plate or λ/4-plate plays an important
role in the further explanation of the heterodyne interferometer.
A disadvantage of the Zeeman method to create a laser beam with two
frequencies is the limited optical power. This is due to the bandwidth of the
Fabry-Perot cavity from the laser, which is too small to optimally resonate
at both frequencies. This effect is especially prominent when a very high
split frequency is needed. As will be shown later, measuring displacements
with a very high velocity demands such a high frequency.
In case more power is needed at such a high split-frequency, a suitable
light source can be created by modulation of the light of a single-frequency
laser by means of an acousto-optic modulator. An acousto-optic modulator
contains a piezoelectric actuator that creates running sound waves in a
material like quartz. The resulting density differences act like a modulated
moving refractive-index phase grating. Depending on the intensity of the
modulation, the magnitude of the light in the diffraction orders is modulated
in a similar way as was presented on modulation by the depth of a phase
grating. More important is the frequency shift that is caused by a Doppler
effect at the moving sound waves that creates different frequencies for the
different diffraction orders. This frequency difference equals the frequency
of the sound times the order number.
Figure 8.76 shows the principle of the heterodyne interferometer. The dif-
ferent polarisation directions of the two-frequency beam components are
used to guide the light to the measurement and reference optical path in
the following way:
664 Chapter 8. Measurement in mechatronic systems

Referene mirror
parallel to wavefront
Measurement mirror
p - polarised parallel to wavefront
‚
 „ /4-plates Interferene
f1

pol f2 x+y

x
y
f2 x
pol f1
s - polarised Polarising
Polariser Polariser
‚ „ beam splier
Photo detetor

Figure 8.76: Heterodyne Michelson interferometer with a two frequency orthogonal

polarised laser where the polarisation is used to separate the reference
and measurement path by means of a polarising beam splitter and
two λ/4 plates. At the right side the effect of the polariser is shown
enabling interference that otherwise would not occur between orthog-
onal polarised beams.
Note that the phase steps of 180◦ at the reflecting surfaces are not
shown as this is not relevant for this principle.

The combined beam enters the system from the left. The beam component
with frequency f 1 is p-polarised in the plane of incidence of the polarising
beam splitter and the beam component with frequency f 2 is s-polarised.
S-polarised light will be reflected by the polarising beam splitter and p-
polarised light will be transmitted. As a consequence the s-polarised light
will serve as reference beam. It is reflected towards the reference mirror and
passes a quarter wave plate consisting of a birefringent material. When ori-
ented in the right way to the polarisation direction of the beam, this λ/4 plate
creates a phase lag of 90◦ in one of the polarisation directions relative to
the other direction as explained in Section 7.4.1.1 in the previous chapter
and changes the polarisation from linear into circular. After reflection at
the reference mirror the second pass through the λ/4 plate will convert the
circular polarised light again into linear polarised light but orthogonal to
the original direction. As a consequence the now p-polarised beam will be
transmitted through the polarising beam splitter towards the polariser and
the photo detector.
The p-polarised component of the light from the laser source serves as mea-
surement beam. It is first transmitted by the polarising beam splitter and
after a passing through a second λ/4 plate, reflection at the measurement
8.8. Optical long-range incremental position sensors 665

E( f1)

1 0+n·2π

Ir
E( f1)+E( f2)

Phase
0 π+n·2π

2π+n·2π
E( f2)
Phase
Field
Irradiance Time

Figure 8.77: Two interfering beams with different frequencies result in an irradi-
ance signal that is proportional to one plus the cosine of 2π( f 1 − f 2 ).
The phase between both frequencies shifts linear over time.

mirror and a second pass through the λ/4 plate it has become s-polarised and
is reflected towards the polariser and the photo detector where it recombines
with the reference beam.
The polariser fulfils an important requirement as the vectorial addition of
orthogonal polarised electromagnetic fields will not show destructive or con-
structive interference. With an accurately oriented polariser under 45◦ with
both polarisation directions only the component of each beam in the polari-
sation direction of the polariser will pass as shown in the right drawing on
Figure 8.76. This means that after the polariser the fields of both beams
have an equal direction and the vectorial addition will show interference at

the expense of a factor 2 loss of amplitude which corresponds with a factor
two of the radiance. It can be reasoned that also in this case constructive
interference at one location corresponds with destructive interference at
another place, hence corresponding to the physical law of conservation of
energy. In case x and y are in phase in the direction of the polariser as
drawn, the components of pol f 1 and pol f 2 in the orthogonal direction are in
counter phase.
666 Chapter 8. Measurement in mechatronic systems

Interference at the detector

The resulting interference field at the detector consists of the two fields with
different frequencies, like shown in Figure 8.77, that add according to the
following12 equation:

E t = E 1 + E 2 = Ê (sin(2π f 1 t) + sin(2π f 2 t)) (8.69)

where Ê is the equal field amplitude of both fields.

Using a trigonometric identity and cancellation of the factors two in the
numerator and denominator yields the following equation:
 
2π( f 1 + f 2 ) t 2π ( f 1 − f 2 ) t
E t = 2Ê sin cos
2 2
= 2Ê sin (π( f 1 + f 2 ) t) cos (π( f 1 − f 2 ) t) = 2Ê sin (2π f c t) cos (π f s t) , (8.70)

From this equation the sine term is the very high-frequent centre frequency.
The cosine term is the much smaller frequency difference and can be seen
as the amplitude modulation of the high-frequency term.
The irradiance is proportional13 to the field magnitude squared. Because
the high-frequency term is unmeasurably high it becomes a constant RMS

value of 2 that can be left out in the proportional equation, resulting in
the following expression for the irradiance:

I r ∝ Ê 2 cos2 (π f s t)

1 + cos(2π f s t)
∝ Ê 2 ∝ Ê 2 (1 + cos(2π f s t)) (8.71)
2

The resulting signal after the photo detector is a combination of a DC voltage

and a sinusoidal voltage with a frequency that is equal to the frequency
difference of the two beam components that interfered at the sensor.
When comparing this equation with the 1+ cosϕ relation from Equation (7.26)
that is used with the homodyne interferometer it is clear that the two
heterodyne frequency components represent a continuous phase shift:

ϕs = 2π f s t (8.72)

The phase shift is linear proportional with time at standstill.

It will be shown in the following that this phase relation is changes by a
12 The following mathematical analysis including the lock-in quadrature detection is derived

from the Phd thesis of Jonathan Ellis from our laboratory in Delft. His thesis can be downloaded
for more details.
13 The constant cn is avoided in these expressions as it does not contribute to the under-
0
standing of the principle.
8.8. Optical long-range incremental position sensors 667

change in the optical path difference between the measurement and refer-
ence beam which enables the measurement of moovements with only the AC
part of the signal, thereby avoiding induced errors by low-frequency noise.

Velocity and position detection by Doppler shift

When the measuring mirror in a heterodyne interferometer moves along

the direction of the beams, the measured frequency difference after inter-
ference f m becomes different from the frequency difference f s at the input
of the interferometer. This phenomenon is called the Doppler shift of the
measurement beam, caused by the Doppler effect that was discovered in
1842 by the Austrian physicist Christian Andreas Doppler (1803 – 1853).
The Doppler effect is related to the constant wavelength and propagation
velocity of a travelling wave. At a fixed position a travelling wave with prop-
agation velocity vp and wavelength λ will be observed as a fixed temporal
frequency f = vp/λ.
When the position of the observer is changing, the observed frequency is
decreased when moving in the same direction as the wave propagation
and increased when moving in the opposite direction. In fact the spatial
frequency of a wave, as determined by the wavelength, is converted into a
temporal frequency by the movement of the observer. Depending on the
movement direction, this temporal frequency is added to or subtracted from
the temporal frequency of the wave that would be observed at stand still.
A light wave with temporal frequency f c and a propagation velocity vp = c/n,
with n being the refractive index, experiences the Doppler shift in the fre-
quency of the measurement beam by a motion velocity vm equal to the
following expression:
 
vp c/n
fd = fc 1 − = fc 1 −
vp ± Nvm c/n ± Nvm
 
c/n ± Nvm − c/n N nvm
= fc ≈ ± fc (8.73)
c/n ± Nvm c

The approximation is valid when the motion velocity is very small in respect
to the propagation velocity of the wave which is in practice true for the
speed of light. N equals the previously defined interferometer constant and
presents itself as a multiplication factor for the Doppler effect, correspond-
ing with the number of times that the measurement beam travels the path
to the moving mirror. Furthermore the centre frequency f c is in most cases
allowed to be taken as reference as the split frequency is with even 5 MHz
only a small part ≈ 10−8 of f c . If necessary this small deterministic factor
668 Chapter 8. Measurement in mechatronic systems

p1

Reference
0V
signal

p2

Measurement
signal 0V

π
?
<
(rad) 0

Figure 8.78: The velocity and position of the measurement mirror can be detected
by measuring the difference in the timing period of the reference and
measurement irradiation signal or by measuring the relative phase.

can be taken into account. More important is the refractive index n that
shows to be a factor that directly influences the observed frequency shift. It
will be shown later that the refractive index induces a serious measurement
uncertainty when measuring in air with extreme requirements on precision.
The Doppler shift is equally present as an offset in the frequency difference
at the measurement sensor f m = f s + f d . This means that f d can be detected
by comparing the detector signal from the interferometer with a reference
signal equal to the split frequency f s of the beam components at the entry of
the interferometer. With a Zeeman laser this difference signal is obtained
by creating an interference signal from a small portion of the light via a
beam splitter, a polariser and a detector, similar to the detection after the
interferometer. With an acousto-optic modulated laser the frequency of the
modulator can directly be used as reference.
The detection can be done in several ways. One method uses a comparator
like a Schmitt trigger to detect the zero voltage crossings of the AC part of
the detector signal and the reference signal. When the timing period be-
tween the zero crossings of both signals is precisely measured, as indicated
in Figure 8.78, the difference between both signal periods is an accurate
measure for the frequencies and the corresponding velocity.
When this measurement is done at regular intervals, the incremental po-
sition change over that interval is determined by multiplying the interval
8.8. Optical long-range incremental position sensors 669

period with the found velocity.

In principle the sampling of the zero crossings is equal to a phase measure-
ment as with different frequencies their phase relation shifts proportional
with the time as function of the velocity. When applying Equation (8.72) for
the measured frequency f m , the following relation for the phase is obtained:

ϕt = 2π f m t = 2π( f s + f d ) t = ϕs + ϕd (8.74)

The phase term ϕs = 2π f s t is equally present in the reference signal which

means that, when the measurement and reference frequencies are compared
in phase, only the Doppler phase shift ϕd remains. With Equation (8.73)
and neglecting the small error due to the velocity difference between the
movement and the speed of light this becomes:

N nvm t
ϕd = 2π f d t = ±2π f c (8.75)
c

Within an incremental time period Δ t the incremental displacement equals

Δ x = vm Δ t and the incremental phase shift due to this displacement becomes:

2π f c N n
Δϕm ( x) = ± Δx (8.76)
c
This relation is equal to the phase relation that was found with the homo-
dyne interferometer. This is not without logic as the heterodyne frequency
difference only introduces an additional phase shift to the phase shift caused
by the displacement. For this reason the second detection method measures
directly the incremental phase shift as function of an incremental displace-
ment by means of quadrature detection with a lock-in amplifier.

Lock-in quadrature detection

Figure 8.79 shows a schematic of a typical lock-in amplifier used to measure

the phase between two signals. Both the measurement ( M ) and reference
(R ) signals are bandpass filtered (BPF) around the frequency f s to remove
their nominal offset, eliminate the sensitivity to optical power fluctuations,
and provide anti-aliasing filtering for the digital signal processing. After
filtering, the reference signal is sent to a phase-locked loop (PLL). A phase-
locked loop is a feedback controlled variable oscillator of which the frequency
is controlled with a phase detector to track the input frequency with a
next-to-zero phase difference. The reason to apply a phase-locked loop is
670 Chapter 8. Measurement in mechatronic systems

I LPF
M BPF
×
atan unwrap gain x
Rc
R BPF PLL
Rs ×Q
LPF

Figure 8.79: Schematic of a lock-in amplifier for phase measurements. Two signals
are detected and initially filtered. The reference signal R is sent to a
phase-locked loop (PLL) to generate matched sine and cosine signals.
Those are then multiplied with the filtered measurement signal M
to produce in-phase ( I ) and quadrature (Q ) outputs. Those are then
low-pass filtered and sent to an arctangent function. The phase is
then unwrapped and a gain is applied to determine the displacement
x from the unwrapped phase.
(Courtesy of Jonathan Ellis)

its possibility to generate two signals with a nominal phase offset of 90◦ .
Similar to the Lissajous plot shown with the spatial signals of a decoder
with four sensors, two temporal signals with 90◦ phase difference enable
signal interpolation and directional information. The name quadrature
detection for this method comes from radio transmission practice and is
related to the four quadrant vectorial representation defined by the phase
difference of the signals.
The two outputs with 90◦ phase difference from the phase-locked loop, Rc
and Rs , are multiplied with the measurement signal to generate two signals,
the in-phase I and the quadrature Q signals. These signals relate in the
following way to a phase shift ϕm ( x) related to the difference in optical path
lengths between the reference and measurement beams:

I = R c M = cos(2π f s t) cos(2π f s t + ϕm ( x)) and (8.77)

& π'
Q = R s M = cos 2π f s t + cos(2π f s t + ϕm ( x)), (8.78)
2
where ϕm ( x) is the Doppler phase change of the measurement signal, rela-
tive to the reference signal as a function of the displacement. Applying a
trigonometric identity yields:
1 1
I= cos(4π f s t + ϕm ( x)) + cos(ϕm ( x)) and (8.79)
2 2
1 & π ' 1 &π '
Q = cos 4π f s t + + ϕm ( x) + cos + ϕm ( x) (8.80)
2 2 2 2
8.8. Optical long-range incremental position sensors 671

If a low-pass filter is used after both multipliers to provide sufficient at-

tenuation to the signal at a frequency of 2 f s , then the remaining signals
are
1
I= cos ϕm ( x) and (8.81)
2
1 &π '
Q = cos + ϕm ( x) (8.82)
2 2
The phase shift due to the optical path differences is then equal to:

Q
ϕm ( x) = arctan , (8.83)
I

In order to derive a correct incremental phase shift Δϕm ( x), related to an

incremental displacement Δ x one has to ensure that the sign of the input
values is taken into consideration in order to place the angle in the proper
quadrant at the start and the end of the increment. The last step needed is an
unwrapping function which properly adds or subtracts 2π for each successive
2π phase jump that has passed during the incremental displacement. The
phase can then be scaled to the displacement value with Equation (8.76)
by knowing the refractive index n, the interferometer constant N and the
centre laser frequency f c .
Important remark:
These requirements, regarding the right quadrant and perfectly counting
all 2π phase jumps, point clearly to the fully incremental nature of practical
laser interferometer systems. They all require an initialisation step to set
the position counters and phase pointers to zero.
It is not possible to measure absolute positions along the total trajectory,
like with the separate marks of encoders. To solve this problem, research
is done on interferometry with multiple frequencies, so called frequency
combs that are based on a laser with short pulses of only a few hundred
femtoseconds (10−15 ). This research has not yet resulted in a practical
commercial implementation.

Velocity limitation of heterodyne interferometry

An important limitation to the applicability of heterodyne interferometry is

due to the maximum value that can reliably be realised for the split frequency
of the applied laser beam. When moving in a direction that increases the
frequency difference at the measurement detector no real problem will
occur as long as the detector can handle the increased frequency level. A
movement in the opposite direction, however, will first reduce the frequency
672 Chapter 8. Measurement in mechatronic systems

difference to a lower frequency depending on the velocity. When increasing

the velocity, at some point the measured frequency difference becomes equal
to zero. Above that velocity, the frequency difference will increase again
but with a different sign. It is clear that in such a situation all directional
information is lost and the measurement system should be reset again.
The minimum split frequency is determined by Equation (8.73):

N nvm
fs > fc (8.84)
c

The interferometer constant N was equal to two in the previous example

but in the following section, examples will be shown with N = 4. Although a
higher interferometer constant will increase the sensitivity and accuracy it
requires a corresponding increase of the split frequency.
This property of the heterodyne interferometer forces a choice between
accuracy and maximum velocity and has stimulated the development of
acousto-optic laser sources with a large frequency difference combined with
high-speed signal conditioning and signal processing electronics.

8.8.2.3 Measurement uncertainty

There are two major sources of uncertainty in laser interferometer position

measurement systems. The first is related to the question whether the
interferometer detects only the desired displacement changes. The second
uncertainty source is related to the phase measurement and whether it
converts only the desired phase shift into a measurement signal.
There is a subtle, yet important difference between these two uncertainty
sources. The former pertains to the interferometer design and direct com-
pensation or cancellation by balancing the system whereas the latter deals
more with the laser system employed, measurement environment, phase-to-
displacement conversion, and sample properties.
The uncertainty factors related with mechanics will be presented in Sec-
tion 8.8.3 at the end of this chapter. In the following, first the internal errors
of the interferometer will get attention:

• Periodic errors, causing a repetitive modifying additional measure-

ment error at the nanometre and sub-nanometre levels.

• Frequency stability, that acts as a modifying error.

• Wavefront errors and pointing stability due to non-ideal optical

parts.
8.8. Optical long-range incremental position sensors 673

1.6 6
4
Displacement [μm]
1.2
2

Error [nm]
0.8 0
>^
0.4
>
No PNL
0 With PNL >`
0 1 2 3 4 5 0 1 2 3 4 5
Fringe Fringe

Figure 8.80: Comparison between a linear signal and a signal with periodic nonlin-
earity, highly exaggerated for illustrative purposes. The second graph
shows a linear displacement with 5 nm of periodic errors with the
nominal slope removed. The periodic errors typically have a first-order
component with one cycle per fringe and a second-order component
with two cycles per fringe. Higher-orders may appear from additional
ghost reflections or stray signals.
(Courtesy of Jonathan Ellis)

• Refractive index changes due to the properties of the medium be-

tween the interferometer and the moving object.

Periodic errors

The working principle of a heterodyne interferometer is based on the capa-

bility to fully separate the measurement and reference beam by means of
their polarisation. Unfortunately the polarisation direction is never fully
orthogonal due to imperfections in the source or the polarising optics. This
means that there is always a small part of the reference beam that trav-
els along with the measurement beam and the other way around. This
unwanted effect is called polarisation mixing and causes small changes in
the measured phase between the measurement interference signal and the
reference signal that is a periodic function of the displacement in relation to
the wavelength expressed in fringes as were introduced with the homodyne
interferometer. One fringe is equal to a 2π phase shift corresponding to
a displacement of x = λ/N . These phase differences induce a deterministic
periodic error as shown in Figure 8.80 that can in principle be compensated
by means of software and calibration as long as the conditions don’t change.
This compensation can only be as good as the stability of the system and in
some cases it is better to avoid them by fully separating both beams directly
from the beginning.
674 Chapter 8. Measurement in mechatronic systems

Cube corner


Right angle
prism

Photo detector 1

f Photo detector 2
2

f1
\J¬


Figure 8.81: Heterodyne interferometer without periodic errors. Two parallel

beams from the optical source ( f 1 and f 2 ) travel to the beam split-
ter and the reflected beams travel toward the right angle prism. The
transmitted beam travels to the retro-reflector. Then, the reference
and measurement beams can be recombined by the beam splitter to
create an interference with opposite phase directions, detected by the
photo detectors.
(courtesy of Ki-Nam Joo)

Figure 8.81 shows an example of such a system with one degree of freedom.
Two separate laser beams with a fixed split frequency, originating from an
acousto-optic modulated laser source are inserted in a 50 % beam splitter,
similar as the one used in a homodyne interferometer. One half of both
beams is reflected as two separate reference beams by a right angle prism,
back towards the beam splitter and the other half is transmitted as two sep-
arate measurement beams towards the movable cube-corner retro-reflector.
The right angle prism has line symmetry which means that the rays are
reflected in the same plane orthogonal to the edge line of the prism as where
they entered. As a consequence the lower reference beam f 1 will stay at the
lower side of the interferometer and the higher reference beam f 2 will stay at
the high side. The point symmetry of the cube-corner causes both beams to
change position and the measurement beam f 1 arrives at the beam splitter
at the same location as the reference beam f 2 while the measurement beam
f 2 arrives at the same place as the reference beam f 1 .
These two beams both interfere at the beam splitter and their interference
signals will show an opposite phase shift as function of the measured dis-
8.8. Optical long-range incremental position sensors 675

tance between the interferometer and the moving retro-reflector.

The first advantage of this configuration is that no polarisation mixing can
occur before the interference takes place as periodic errors are related to the
changing trajectory of the measurement beams to the moving retro-reflector.
The second advantage is the double sensitivity with interferometer constant
N = 4 because of the fact that four beams are travelling the same trajectory
to the moving retro-reflector. The last advantage is that no use has been
made from the polarisation direction. This enables its application in more
complex interferometers where the polarisation direction can be used for
other reasons, like for creating a plane-mirror version of this interferometer
along the line of thoughts that are presented in the next section.

Frequency stability

The uncertainty of the interferometer measurement itself is fully based on

the phase measurement that is equal to Equation (8.76):
2π f c N n
Δφx = ± Δx (8.85)
c
The first source of uncertainty is the frequency of the light f c . In principle
lasers can be made with a stable frequency up to 10−11 , or 10 pm per metre,
by thermal stabilisation and other methods, which is sufficient for most
measurements. Further all factors are constant with the exception of the
refractive index n that is related to the propagation speed in the environment
where the measurement takes place, which is mostly in air.

Wavefront errors and pointing stability

Interference of the reference and measurement beam takes place over the
total overlapping area of both beams. As long as both beams have a flat and
parallel wavefront the interference is identical over the total surface but any
deviation to that ideal flat wavefront will cause a phase shift and reduction
of the total observed interference at the photo detector. With the exception of
the purposely applied tilt in the wavefront of the homodyne interferometer
to create moving fringes, normally all wavefronts in a laser-interferometer
measurement system need to be parallel and without any exception always
the wavefronts need to be flat. This poses extreme requirements to both
the spatial coherence of the laser source, the flatness of the optical surfaces
and perfection in the retro-reflectors. In principle optical flats are very
difficult to manufacture with an exponential cost increase as function of
676 Chapter 8. Measurement in mechatronic systems

size and maximum allowable surface topology deviation. The deviations can
be expressed using the Zernike modes as described in Section 7.5.2.1 and
have to remain at least below λ/20 to get an acceptable interference signal
but when lateral beam displacements are present, the deviations in the
surface need to remain below the required measurement accuracy of the
interferometer system.
Next to non-parallelism, a beam that is not pointing to the right direction
will induce a so called cosine error that is proportional to the cosine of the
angle of the beam with the reflecting surface of the measurement mirror.
This error is minimal when the direction is orthogonal. When initially all
interferometers are mounted such that the measurement beam is as good as
possible aligned, any change in the pointing due to temperature or to other
mechanical instabilities will have an influence on the measurement error.

Refractive index changes

The refractive index of air is influenced both by pressure and by tempera-

ture and the related measurement uncertainty is a more difficult error to
overcome. This is in a large part due to the systematic uncertainty in the
equations used to calculate the refractive index. Even with ideal environ-
mental parameter measurements, there still is an uncertainty due to the
equations used and the empirical data from which they are based. The typi-
cal calculation for refractive index is done with the modified Edlén equation,
named after the Swedish physicist Bengt Edlén (1906 – 1993. After his
original definition it was improved by several scientists in an international
comparison of interference air refractometers with participants like Piet
Schellekens of the Eindhoven university of technology and Jo Spronck from
Van Swinden laboratories who presently is active in metrology at our Uni-
versity laboratories in Delft. Other participants included G.Wilkening and
F Reinboth from PTB Germany and K.P. Birch and M.J. Downs of NPL in
England. This highly accurate experimental work resulted in a publication
in Metrologia 1986,22 and a successor paper by the two scientists of NPL
in Metrologia 1993,30 from which the following equation is taken for the
refractive index difference of air relative to vacuum:
P ( n − 1)σ 1 + 10−8 (0.601 − 0.00972T )P
( n − 1)T,P = (8.86)
96095.43 1 + 0.003661T
The dispersion factor (n − 1)σ is equal to:

2406294 15999
( n − 1)σ = 8343.05 + + · 10−8 (8.87)
130 − σ2 38.9 − σ2
8.8. Optical long-range incremental position sensors 677

The wave number σ equals the inverse of the wavelength in vacuum (1/λ).
These equations give the refractive index n T,P for dry air as a function of
the wave number σ in μm−1 , the temperature T in degrees Celsius and the
pressure P in Pascals for a wavelength in the range of 350 – 650 nm.
The refractive index for non-dry air n T,P,Pv is calculated by:

n T,P,Pv − n T,P = −Pv (3.7345 − 0.0401λ−2 ) · 10−10 (8.88)

where the vapour pressure Pv is in Pascals.

Using these equations, the traceable uncertainty in the calculated refrac-
tive index is not better then approximately one part in 108 because of the
uncertainty in the measurements of pressure, temperature and humidity.
This value is equal to a traceable uncertainty of 10 nm over a metre, re-
lated to agreed standards with long term stability demands. Fortunately
this uncertainty contains unknown but rather constant errors. When only
short term relative measurements have to be done in the order of minutes
or seconds like is often the case in mechatronic positioning systems, the
uncertainty can be as low as 100 pm.
By taking the partial derivative of the modified Edlén equation to tempera-
ture, pressure and humidity the individual sensitivity for these modifying
error inputs can be estimated in a certain environmental condition setting.
With an atmospheric pressure of 105 Pa, a temperature around 20 ◦ C and
a wavelength of 600 nm, the equations show the following approximated
partial derivatives in [◦ C−1 ] and [Pa−1 ]:
∂ n T,P,Pv ∂ n T,P,Pv ∂ n T,P,Pv
≈ 1 · 10−6 , ≈ 2.8 · 10−9 , and ≈ 3.6 · 10−10 (8.89)
∂T ∂P ∂Pv
This means that the traceable uncertainty of one part in 108 approximately
corresponds with environmental changes of 3.7 Pa, 10 mK and 27 Pa. A par-
tial vapour pressure of 27 Pa is about 1 % of the saturation vapour pressure
of water of 2.4 · 103 Pa. It is clear that of these factors the temperature has
the largest effect and a temperature difference of only 1 mK already gives
an error of 1 nm over 1 metre of optical path length.
With slow moving systems in a thermally controlled environment it is often
possible to determine the refractive index of air at only one location inside
the machine by means of a wavelength tracker. This element consists of
a separate interferometer with a measurement retro-reflector at a fixed
mechanical position relative to the interferometer. Any measured phase
changes of this interferometer can then only be caused by refractive index
changes of the air by the temperature, pressure and humidity.
Fast moving machines however create so much turbulence inside the ma-
chine that next to an accurate thermal control, also the airflow around the
678 Chapter 8. Measurement in mechatronic systems

f Referene Retr
2

f f f1
1 2
From Laser

To  
f1 ± f f f1 ± f
2

f1 Measurement
Polarising Beam Splier 
f
2

Figure 8.82: Basic single-pass heterodyne interferometer with two cube-corner

retro-reflectors to shift the beams and prevent sensitivity for tilt.
Note: The rays are shown shifted and the positions are only indicative.
The polariser before the detector is not shown.

moving object needs to be controlled. The air needs to flow faster than the
maximum velocity of the moving object in order to avoid mixing of air with
different temperatures due to warm objects like actuators.
This airflow necessitates fast blowing air showers with temperature sta-
bilisation of better then 1 mK for nanometre range errors. The additional
complication of these measures, on top of other requirements like the maxi-
mum allowable vibration forces exerted by the turbulent air, is one of the
reasons for applying encoders instead of laser interferometers in these ap-
plications.
When working in vacuum, these problems do not occur and under that condi-
tion laser interferometers are well suited for the most critical requirements,
for instance with stages for EUV lithography wafer scanners. Also in space
instrumentation distance measurements with laser interferometers is often
the best solution, especially when long distances between satellites have to
be measured.

8.8.2.4 Different configurations

Several practical configurations are applied in laser interferometry. The

examples shown all apply to heterodyne interferometry because of their
application in complex multi-axis measurement systems and the possibility
to direct the beams by virtue of their different polarisation.
The first example from Figure 8.82 is the most simple heterodyne interfer-
ometer configuration possible. It is a single-axis single-pass interferometer
and instead of applying λ/4 plates to direct the light to a detector at another
8.8. Optical long-range incremental position sensors 679

location the detector is placed aside of the laser source. To prevent colliding
laser beams it uses two cube-corner retro-reflectors.
The light from the source enters at the upper half of the polarising beam
splitter where one of the frequencies is reflected and one is transmitted
based in their polarisation direction. One cube-corner serves as reference
mirror and shifts the beam to the other half of the polarising beam splitter.
The other cube-corner serves as measurement mirror. It prevents wavefront
errors that would otherwise be caused by angular movements and also shifts
the measurement beam to the other half of the polarising beam splitter. In
this way the return path does not collide with the entry path and a detector
can be placed next to the laser source.
It should be noted that instead of a cube-corner also a right angle prism
can be used in this configuration for the reference beam. The main benefit
of a cube-corner in this configuration is its insensitivity for rotation in any
direction when mounting the element. A right angle prism needs to be
mounted sufficiently well that the reference beam will be pointed at the
same location as the measurement return beam.
The measurement retro-reflector always needs to be a cube-corner to avoid
shifting of the beam and the creation of non-coplanar wavefronts due to
angular motions.
680 Chapter 8. Measurement in mechatronic systems

Reference Retr
f
2

f f
1 2
From Laser
f1 ± f
To photo detector
f1 ±2 f f f1 ±2 f
2
Measurement
Polarising Beam Splier plane mirror
Quarter wave plate
f1
f ! Tilt
2
f1 

›<"!

Figure 8.83: Basic dual-pass heterodyne interferometer with two cube-corner retro-
reflectors to shift the beams and a plane-mirror as moving element.
One λ/4 plate changes the polarisation of the measurement beam,
causing it to be reflected at the polarising beam splitter and follow a
second trajectory to the moving mirror. The lower image shows the
effect of tilting of the measurement mirror on the trajectory of the
measurement beam. Also here the polariser before the detector is not
shown.

Dual-pass plane-mirror interferometer

The dual-pass plane-mirror interferometer as shown in Figure 8.83 has

been developed to enable measurements in more directions with multiple
interferometers by replacing the cube-corner with a plane mirror. When the
mirror is large, this configuration allows a displacement of the mirror in a
perpendicular direction to the measurement direction without affecting the
measurement. This configuration uses one λ/4 plate and two cube-corners
at the interferometer. The reference beam follows the same path as with the
single-pass interferometer but the measurement beam follows a different
path. After the polarising beam splitter its polarisation is changed from
linear to circular by the λ/4 plate and after reflection to the moving mirror its
polarisation is converted again in a linear polarisation by the λ/4 plate but
now in the orthogonal direction. As a consequence, the measurement beam
8.8. Optical long-range incremental position sensors 681

…#<


Ġ|

100
90
 !Ÿ


80
70
60
50
40
30
20
10
0
0
0.45

1.35

2.25

3.15

4.05

4.95

5.85

6.75

7.65

8.55
0.9

1.8

2.7

3.6

4.5

5.4

6.3

7.2

8.1

9
…#<’

Figure 8.84: Tilting of the measurement mirror in a dual-pass plane-mirror inter-

ferometer causes the measurement beam to shift. As a consequence
the measurement beam and the reference beam will not completely
overlap, called “beam walkoff”. Interference takes place only in the
overlapping regions.

will be reflected by the polarising beam splitter and the second cube-corner
and is directed a second time towards the moving mirror. In this second
trajectory, its polarisation will be converted via circular to the original
linear polarisation direction enabling it to pass through the polarising beam
splitter and interfere with the reference beam.
The dual-passing means that a movement of the mirror gives a double
phase shift and double Doppler effect on the measurement beam. The
related interferometer constant equals four, which is confusing with the
general wording of this interferometer as a dual-pass type. The “dual-pass”
refers to the fact that always at least the distance between the object and
interferometer has to be passed twice in both directions. Because of the
interferometer constant N = 4, the measurement signal will hence be double
sensitive to the movement and as mentioned in the previous section this
also requires a higher split frequency to be able to perform a high velocity
measurement. Next to this increased higher sensitivity for movements in
the measurement direction and insensitivity for large movement in the
orthogonal directions, also the sensitivity for tilt of the measurement mirror
682 Chapter 8. Measurement in mechatronic systems

is zero for small rotations. This can be seen by following the dashed rays in
the lower drawing of Figure 8.83. The ray, returning after the first reflection
will be tilted with double the tilt angle of the measuring mirror.
The four reflections on the polarising beam splitter and the cube-corner will
return the beam for the second trajectory with the same double tilt angle as
after the first reflection and the second reflection will add that same amount
of tilt but now in the opposite direction. The result is an unchanged optical
path length of the measurement beam.
Although the optical path length is not changed, the measurement beam
to the photo detector shows a lateral displacement of the rays, called beam
walkoff (BWO).
With the tilt angle ϑt in radians [rad] and the measurement mirror at
a distance dm from the interferometer the beam walk off is equal to the
following expression:

BWO = 4 · d m ϑt , (8.90)

Because of this beam walkoff, tilting is only allowed as long as the rays are
all kept inside the interferometer and as long as the reference beam and
the measurement beam will sufficiently overlap to induce interference. Fig-
ure 8.84 shows the effect of the beam walkoff on the interference efficiency.
The part of the beams that do not overlap will give a DC irradiance signal
to the sensor and with increased beam walkoff the AC interference part will
become relatively smaller.
By increasing the diameter of the beams a larger tolerance on tilt is allowed,
forcing the application of large optical parts. Because of the optical require-
ments on flatness and wavefront errors these parts are however extremely
expensive.

Multi-axis laser interferometers

It was mentioned that plane-mirror interferometers allow movements in

the orthogonal direction of the measurement direction. This enables their
use in measurements of solid objects like a wafer stage of a wafer scanner
in more directions.
Figure 8.85 shows the lay out of the three degree of freedom wafer stage
from the front page of this book. One interferometer is used for the y dis-
placements and two interferometers are used for the x displacement.
The difference between both x measurements is a measure for the rotation
around the Rz axis. This rotation can only be small in view of the remarks
made previously regarding beam walkoff but still it is sufficient for its use
8.8. Optical long-range incremental position sensors 683

x1-interferometer
To signal processing x - y - Rz
stage
x2 -interferometer
Ž\J¬


To signal processing x-mirror

33 - 66 % y-mirror
Laser …


y-interferometer
To signal processing

Figure 8.85: With three plane-mirror interferometers measurements can be done

in three directions while x1 and x2 define both the x displacement and
a rotation Rz around the z-axis.

in wafer scanners. By expanding the measurement mirror in the z direction

a full 6-axis measurement system can be realised as will be presented in
Chapter 9.
As a last example, the 3-axis x − y − z measurement system from Figure 8.86
shows the possibilities to direct the light to any measurement location. It
also incorporates a fourth measurement branch with a wavelength tracker
that enables to determine the refractive index changes of the air by temper-
ature or pressure.
684 Chapter 8. Measurement in mechatronic systems

Receiver
z-interferometer y-interferometer
Beam bender
Plane mirror
 50 %


Plane mirror 33 %
 

Receiver Beam bender
A
Receiver

B Wavelength
tracker

Beam bender Beam bender

Plane mirror

x-interferometer
Laser source Receiver
^\¬


Figure 8.86: A three degree of freedom x − y − z measurement system on a gantry

type of machining centre uses different optical components to direct the
light to the measurement locations. The wavelength tracker enables
to compensate for refractive index variations. The metrology loop from
A to B is very long and indirect.

8.8.3 Mechanical aspects

Position measurement is a relative measurement to a known reference.

With precision long-range measurements, the reference is often made from
thermally solid material mounted on a vibration free environment to prevent
any measurement errors that can occur by deformation of the reference. The
frame holding this stable reference environment is called a metrology frame
and this frame represents the “sacred world” of the precision metrologist.
In Chapter 9 on wafer scanners, this metrology frame is shown to be solidly
connected to the lens in order to guarantee the positioning accuracy of the
image to the wafer. Next to this extreme precision application example, also
in other applications the metrology frame should be well connected to the
reference location, like the position of a work piece in a precision machining
centre. With single-axis positioning systems, this metrology frame can
be positioned quite close to the measurement location but with multi-axis
measurement systems like the gantry type configuration of Figure 8.86 this
8.8. Optical long-range incremental position sensors 685

is no longer possible.
With this configuration, the metrology frame is not a separate frame but
is consists of the base and the complete gantry. The relative position mea-
surement from the moving part A to the solid table B is done in a very
indirect way, described by means of the metrology loop. The metrology loop
is defined by the shortest path that carries information about the relative
position of two or more measurement locations and consists of a series of
solid objects measurably connected by position measurement sensors or a
calibrated sliding mechanism. In the example from the figure, the metrology
loop runs from Position B via the table to the x-interferometer that connects
its position information to the gantry. The metrology loop runs further
through the gantry to the z-interferometer that connects the position of the
gantry to the moving part A. All the parts in the loop contribute to the total
measurement and as long as all properties of all parts in the entire metrol-
ogy loop are exactly known, no errors will happen. Unfortunately reality is
far from ideal and for instance imperfections in the sliding mechanism and
thermal effects on the solid material influence the measurement, often in
an unknown and undesired way.

8.8.3.1 Abbe error

One important source of errors is based on angular movements combined

with long arms. It was the German physicist Ernst Karl Abbe (1840 – 1905)
who introduced the term Abbe error for these errors. He stated the following:

If errors in parallax are to be avoided, the measuring system must

be placed coaxially with the axis along which the displacement
is to be measured on the workpiece.

Based on this general statement it was concluded that either the displace-
ment measuring system should be in line with the object of which the
displacement is to be measured or the angular motions should be measured
separately in order to compensate the effects. An example of such a solution
is shown in Figure 8.87. In this case the position of a point A on top of the
measurement mirror must be measured but it is not possible to direct the
laser beams coaxial with this position as the mirror needs to be moved also
in the z direction. In that case additional measurement beams can provide
information on the rotation angle around the x-axis.
It is extremely important to be aware that any error in this angular mea-
surement will be converted into a proportional error in the z measurement
686 Chapter 8. Measurement in mechatronic systems

To receivers Cube corners

Measurement
Input beam beams, three axes

z
~!
 Quarter y ya
wave plates A
x

e 3
1 2

Figure 8.87: By measuring with multiple interferometer beams, the rotation

around the x, y and z axes can be measured. The Abbe error, caused by
the non-coaxial measurement with the surface plane e and a rotation
around the x-axis, can be compensated by the angular measurement
based on the distance d of two of the interferometer beams. A “mono-
lithic” three axis interferometer of Agilent Technologies is shown that
can accomplish this measurement in one unit. It integrates three inde-
pendent plane-mirror interferometers with the high relative stability
that is needed for a reliable angular measurement.

direction linearly depending on the distance ya to the point of interest. For

this reason the requirements on stability are even more extreme than for
the linear movements and it is always better, when possible, to measure as
good as possible in Abbe, as it is called in metrology terms.
The interferometer unit of Agilent Technologies clearly demonstrates the
complexity of such an interferometer as all optical parts are fully integrated
to guarantee a stable relation between the three further independent plane-
mirror interferometers. This integration of multiple optical components can
be done by means of adhesives but also direct optical bonding between the
polished surfaces is possible when these surfaces are sufficiently flat. Only
by such a high level of integration the necessary accuracy in the angular
measurement can be achieved.
These principles are applied to the extreme in the wafer scanners of ASML
that are presented as the closing application case of this book.
Chapter 9

Precision positioning in
wafer scanners

In Chapter 1 the invention of the first wafer stepper, the Silicon Repeater
of Philips Electronics was memorised. This machine was designed for the
purpose of realising the complicated structural lay-out of an integrated
circuit and its successors at ASML1 became an important factor in the
worldwide proliferation of electronics, because of its capability to expose an
image of a reticle on a wafer with very small details.

9.1 Introduction
The smallest details in an integrated circuit are called features and their
minimum size is mainly determined by the resolution of the optical exposure
system as indicated by the Critical Dimension (CD). In Chapter 7 it was
explained that the resolution of an optical system is determined by the
wavelength λ of the light, the Numerical Aperture (NA) of the lens and a
k 1 -factor that mainly is determined by the illumination system. With these
variables the critical dimension was shown to be equal to:
λ
CD = k 1 (9.1)
NA
This resolution is a theoretical value that is influenced by the surrounding
pattern. The wave character of the light creates an airy-disk field profile
1 Several of the figures in this chapter and the background information is gracefully provided

by ASML, market leader in lithographic exposure systems for the semiconductor industry.

687
688 Chapter 9. Precision positioning in wafer scanners

of the diffraction limited feature and the “ringing” around the peak will
add vectorial to the airy-disk field profile of neighboring features. As a
result the CD of isolated features is different from the CD of densely packed
features. Several definitions for the CD are used for this reason of which
one is chosen to be used further in this chapter, the half-pitch CD. Other
than with isolated features, the half-pitch CD refers to a regular grating of
lines and spaces of which the width of a line equals half the period (pitch).
Based on the basic formula for the CD, research and development on these
machines has always focused on the reduction of the wavelength λ and the
k-factor and on maximising the NA. The need for a short wavelength has
resulted in a continuous development of light sources with ever smaller
wavelengths, starting with Mercury arc discharge lamps with 436, 405 and
365 nm wavelength, followed by excimer lasers that produce light with a
wavelength of 248 and 193 nm. The next step will be an extreme ultraviolet
(EUV) light source with 13 nm wavelength. With EUV light only catoptric
optics with mirrors can be used, because no transparent material exists
for this short wavelength. The improvements to the k-factor have resulted
in the development of several optical techniques to direct the illumination
in such a way that as many as possible diffractive orders of the image are
captured in the aperture of the lens as was explained in Section 7.4.4.
It was the required high numerical aperture that originally forced the indus-
try to cover the wafer in parts and not as a whole. Exposing a large surface
with a high numerical aperture would only be feasible with extremely expen-
sive lenses with a very large diameter. For that reason it was assumed more
affordable to use a smaller lens and expose the wafer in steps. Each separate
exposed area is called a die and one die can contain a multiple of integrated
circuits. In practice the size of a die is determined by the projection lens and
with a wafer stepper its size was approximately 18 × 18 mm square. Still for
practical reasons the numerical aperture in air is maximised to a value of
about 0.93 and because of that limitation, most modern exposure systems
use water as intermediate medium with a refractive index of ≈ 1.44 at 193
nm. This has increased the practical value of the numerical aperture to
≈ 1.35.
Wafer steppers have remained the main method for the most critical layers
with the finest details until circa 1998 when the wafer scanner entered the
arena.
The principle of the wafer scanner was introduced much earlier by the
American company Perkin Elmer that applied a scanning principle in an
exposure system with a magnification of one from reticle to wafer. Although
it required a reticle with the same size as the wafer it was a good solution
9.1. Introduction 689

for the contamination and damage problems of the original exposure princi-
ple at that time that used a reticle in direct contact with the wafer. Like
with a document scanner, a light stripe exposes the reticle in a scanning
motion simultaneous with the wafer. Only in this case both wafer and reticle
are moving relative to the stationary light stripe. The wafer scanners of
ASML were based on that same scanning principle while they applied the
demagnifying lens of the wafer steppers to achieve a better resolution.

9.1.1 The wafer scanner

Figure 9.1 shows in a very schematic way how the exposure of a silicon wafer
by a wafer scanner takes place. The illuminator defines a light stripe, called
the slit with a certain width that is stationary with respect to the projection
lens. Instead of using a stationary reticle, in a wafer scanner the reticle
is placed on a reticle stage that performs a scanning motion, where the
illumination system illuminates only one part of the reticle simultaneously.
The projection lens images the reticle on the silicon wafer that also is placed
on a scanning positioning system, the wafer stage that moves synchronously
with the reticle stage. The wafer is firmly clamped to a wafer table on the
wafer stage by means of vacuum. The wafer table is also often called the
wafer chuck or just “the chuck” because it clamps the wafer tightly to the
moving stage. Smaller details can be defined by a relatively large reticle
when compared with the scanners of Perkin Elmer, because of the four times
demagnification of the projection lens and the exposure of only one die simul-
taneously. The disadvantage of this demagnification is that the reticle has
to move four times as fast with corresponding high acceleration levels and
reaction forces. In spite of these dynamic drawbacks this demagnification is
necessary because of the complexity to create reticles with sufficient quality.
The four times larger details enable the electron beam pattern generators
to realise the pattern including the assist features2 , small details that are
added to the IC pattern to tweak the imaging and enhance the resolution
around corners and in area’s where a neighbouring detail is very close by.
The scanning exposure of the reticle has several advantages over the expo-
sure of a stationary reticle as is done in a wafer stepper.
The first advantage of scanning versus stepping is the averaging of small
image position errors. Every point on the reticle is exposed over the full
2 It is beyond the scope of this book to enter deeper in the physics behind these assist

features that work on the wave character of the exposed light. The theory on these resolution
enhancement methods is a real field of experts working with highly advanced modelling
software.
690 Chapter 9. Precision positioning in wafer scanners

Exposure light from illuminator

Scan

~ 

 } Stepper


ƒ
Scanner


ƒ
?˜
height 32
~6

Slit

^`
Lens error
averaging
Slit movement
Die

Step Silicon wafer on waferstage

Scan

Figure 9.1: Basic principle of a wafer scanner exposure system. A small “slit” of
light exposes a pattern on a transparent reticle and the pattern is
imaged on the wafer by a four times demagnifying projection lens. A
simultaneous movement of reticle and wafer results in a full image
of the pattern on the wafer. To cover the entire wafer, the scanning
motion is repeated with an intermediate stepping motion to the next
die position. The size of the die is increased by utilising the full width
of the circular image field of the lens and by maximising the scanning
stroke.

width of the exposure slit by the scanning motion and small image position
errors by the projection lens, present in one part of the slit, will compensate
an opposite error in on another part of the slit. If the errors are random
they average out.
The second advantage is related to the maximum size of the image field.
A lens made of round lens elements will have a circular image field where
all parts can be exposed with the same numerical aperture. Exposing a
wafer in circular regions is not very practical because in that case parts of
the wafer will not be exposed. Although hexagonal dies could fill the entire
9.1. Introduction 691

surface this method is highly unpractical when such a die has to be filled
with rectangular integrated circuits. For these reasons the die always has
a rectangular shape. With a wafer stepper the rectangle has to fit in the
circular field and a square die will be the maximum surface that fits.
A wafer scanner that uses a slit with a limited height can utilise a larger part
of the width of the image field to approximately 26 mm, without increasing
the size and cost of the lens. It even allows optimising the optical image
quality in the slit area. Further, in theory, the size of the image in the
scanning direction could be made infinitely long but the required large
reticle would become extremely expensive. For that reason the die size in
the scanning direction is limited to approximately 32 mm.

9.1.2 Requirements on precision

The critical dimension is one of the main drivers of precision in a wafer

scanner. The continuous reduction of the critical dimension is almost like a
law of nature determined by Moore’s law.
The American chemist and physicist Gordon Earle Moore (1929), co-founder
of Intel, defined his law in 1965 based on the observation that the number
of transistors on an Intel processor doubled every two years. Although
not a real law of nature, this postulation has predicted the continuous
exponential growth rate over almost four decades, with a corresponding
“shrink” of the smallest details. The industry has organised itself so well in
this respect that they even use a jointly defined roadmap, the International
Technology Roadmap for Semiconductors ITRS from International Sematech
that describes in full detail the expected developments of the different
parameters that rule this market for a 15 years forecast.
Next to the critical dimension, the overlay is the second main driver for
precision. Overlay is the relative position of any pattern layer in respect to
the other layers and it should be as small as possible. As can be observed in
Figure 9.2, all layers need to be connected by a multitude of small conductive
pillars, the vias.
Errors in the overlay will impact both the electrical properties of the contacts
and the insulation and might even create short circuits.
The maximum allowed overlay value is strongly related to the critical di-
mension. Originally a maximum overlay in the order of 30 % of the critical
dimension was sufficient but recently the overlay requirements were aggra-
vated to a value below approximately 15 % of the CD.
This more severe relative overlay requirement is related to recently intro-
692 Chapter 9. Precision positioning in wafer scanners

Wire
Wire

Via
Via

Metal 1

Metal 1 Pitch

Figure 9.2: Cross section of a modern integrated circuit illustrating the multitude
of interconnecting layers that have to be positioned with strict
requirements on overlay.
(courtesy of Semiconductor Industry Association. The International
Technology Roadmap for Semiconductors, 2009 Edition. International
SEMATECH:Austin, TX, 2009.)
The right image is a CMOS logic microprocessor IC of IBM.
(courtesy of IBM Research)

duced special exposure methods like double-patterning or double-exposure,

where lines are exposed in between the lines from a previous exposure cycle.
By slightly underexposing a pattern, that can just be imaged by the lens,
the developed pattern shows smaller lines than spacings and it becomes
possible to expose another set of lines in between. These methods enable IC
manufacturers to extend the use of 193 nm wavelength light to ever smaller
dimensions, but it has simultaneously resulted in a strong increase in the
overlay requirements beyond Moore’s law, as can be seen in Figure 9.3.
The third factor of importance for precision is the productivity. The high cost
of these machines, mainly driven by the optics, requires an ever increasing
speed of operation in order to retain an acceptable return on investment for
the IC manufacturer.
The productivity is defined in different ways. The throughput of the machine
is the easiest measurable item and tells something about how many wafers
can be exposed in one hour. On the other hand the productivity tells nothing
about how successful the exposures are and so for that reason the number of
good exposures per unit of time would be more meaningful for the customer.
9.1. Introduction 693

30
NAND
20 Flash
Logic

Process Overlay Requirement [nm] 10 DRAM

2
100 80 70 60 50 40 30 20

Resoluon Half Pitch [nm]

Figure 9.3: The requirements on overlay follow a trend that outpaces Moore’s law.
Soon all layers in an integrated circuit need to be positioned well within
a few nanometres.

Still for practical reasons the throughput is most frequently used during
the design of the machine.
In 2010 a throughput of more than 180 wafers per hour was the state of the
art. This value corresponds to 20 seconds per wafer, including loading and
unloading. The related extreme velocities and accelerations of the stages
and other robotic motion systems cause strong reaction forces with vibration
levels that easily could impair the critical dimensions by fading, lack of
contrast by vibrations during exposure.
All these factors result in a large set of requirements for the wafer scan-
ner. The list in Table 9.1 only gives an overview of a selection of the most
important requirements that directly determine the precision positioning-
systems, the wafer stage and the reticle stage. The values are approximated
as they are only meant to give an idea of the order of magnitude. Note
that the wafer stage position error is divided in a low-frequency part, the
moving average (MA), causing overlay errors and the high-frequency part,
the moving standard deviation (MSD) that causes image fading.
694 Chapter 9. Precision positioning in wafer scanners

Based on these requirements, this chapter focuses on the following important

aspects that determine these extreme performance levels and will pose new
challenges for future developments.

• Dynamic architecture, preventing disturbing vibrations of the dynam-

ical sensitive parts.

• Zero stiffness stage actuation with Lorentz actuators to fulfill the

demands on dynamic performance.

• Indirect relative position measurement between image and wafer with

alignment marks and long- range sensors.

• Motion control, with both feedforward and PID-feedback control with

a high level of predictability of the plant dynamics.

Table 9.1: Main requirements on precision of the positioning systems in a wafer

scanner.

Requirement Approximate
value

Critical dimension (CD). < 40 nm

Overlay. < 4 nm
Wafer stage velocity, stepping. > 2 m/s
Wafer stage velocity, scanning. > 0.5 m/s
Reticle stage velocity. > 2 m/s
Wafer stage acceleration. > 30 m/s2
Reticle stage acceleration. > 120 m/s2
Wafer stage metrology error. < 0.5 nm over 20 s
Wafer stage MA in-plane position error (overlay). < 1 nm
Wafer stage MSD in-plane position error (fading). < 10 nm
Focus error. < 100 nm
Settle time. < 10 ms
9.2. Dynamic architecture 695

9.2 Dynamic architecture

One major rule in precision engineering is to first prevent problems by fully

mastering the open-loop dynamics of the mechatronic system before at-
tempting to control its behaviour. For this reason, the dynamic architecture
of a wafer scanner, as shown in Figure 9.4, is focused on keeping all non
deterministic dynamic disturbances as good as possible separated from the
optical imaging system.
The first sources of disturbances originates within the wafer scanner itself
due to the large accelerations of the stages. Another important source of
disturbing forces are vibrations that are caused by other equipment like the
large air conditioning, purifying and processing equipment in the wafer fab,
the usual name for a semiconductor factory.
These vibrations are reduced by a well designed dynamic structure consist-
ing of the mechanical frames, the use of a balance mass to absorb reaction
forces and a vibration isolation system to protect the most sensitive parts.
The basic mechanical structure of a waferscanner consists of several parts.
The first part is the pedestal, a heavy and stiff structure that connects
the wafer scanner with the floor of the wafer fab. This floor is typically
made from large steel bars and is not extremely stiff relative to the mass of
the wafer scanner. The pedestal is often made from solid concrete and is
located as low as possible on the fab floor. In this way the pedestal effectively
grounds the wafer scanner on the compliant floor structure and reduces the
impact of vibrations from inside the machine to the wafer fab.
All the modules of the wafer scanner itself are built on a rigid base frame
that is made of steel. This main structure of the wafer scanner is directly
mounted on the pedestal without additional vibration isolation measures.
The projection lens is firmly held in the right position by the metrology frame
that defines the measurement reference of the stages to the image of the
die. The metrology frame is connected to the base frame by means of three
air mounts, air cushion springs with a very low compliance that serve to
reduce the transmission of base-frame vibrations to the metrology frame.
The stages are moving in six degrees of freedom and are supported by the
base frame, either on air-bearings or by means of an active magnetic support.
The support of the wafer stage in the vertical direction often consists of a
large air bearing, the air foot, that floats on the flattened surface of a large
granite stone. Granite is a very stable material and often used for this
purpose in precision machines.
696 Chapter 9. Precision positioning in wafer scanners

3000 N
Recle stage Balance
Reacon mass
forces

?˜

Metrology frame

Base
frame Balance
Air mounts Wafer stage mass
Granite stone

Machine pedestal

Figure 9.4: The dynamic architecture of a wafer scanner is based on two principles.
The first is to create a vibration free environment for the projection
lens and metrology frame and the second is to reduce the vibrations to
this “sacred reference” by balance masses and a well-tuned vibration
isolation system. In reality the balance mass of the wafer stage includes
the granite stone that supports the wafer stage.

9.2.1 Balance masses

The horizontal directions need to move with very high accelerations while
the vertical direction with the rotations have less stringent requirements
on acceleration. These high horizontal acceleration levels of the heavy
stages create reaction forces in the order of several kilo Newton. In the first
generations of wafer scanners, these reaction forces were directed towards
a heavy force frame that was mounted on the base frame separately from
the other more sensitive parts of the machine.
Although in that way it was expected that the mass of the force frame would
absorb these forces, the coupling of the resulting movements to the base
frame had caused many problems in practice. The remaining vibrations
sometimes excited the dynamics of the fab floor and could disturb any other
equipment that was located nearby and they still reached the sensitive
parts inside the machine by transmission via the base frame through the
air mounts.
For that reason, it was decided in more recent generations of wafer scanners
9.2. Dynamic architecture 697

that the reaction forces of the high acceleration x and y movements should
be absorbed by very heavy balance masses that are not rigidly connected
to the base frame but horizontally guided on air bearings or compliant
mechanisms.
The balance masses consist of a seismic mass that is directly connected to
the stator of the linear actuator that drives the stage. The actuation force
from the actuator drives both the stage and the balance mass in opposite
directions and causes the relative movement between the mover and the
stator of the actuator to become larger than would be the case when the
stator was connected to the force frame.
The first consequence of this increased relative movement is an increase of
the induced EMF over the motor coils, requiring a higher maximum voltage
of the power amplifiers. The second consequence is the need for a longer
movement range of the linear actuator with the associated cost of larger
coils or magnet assembly.
By increasing the mass of the balance mass its maximum displacement
decreases proportionally and for that reason the mass is chosen much larger
than the mass of the stage. With a stage of 80 kg the balance mass can
have a mass of 500 kg or more. A large mass is also large in volume and
measures should be taken to guarantee that the balance mass will not touch
any other part of the machine during its movements. To guarantee the
consistent free movement of the balance mass measures are taken to keep
its average position in the middle of the moving range, compensating the
effects of gravity and other non alternating forces working on the balance
mass. These measures either consist of a separate actuator with a low
stiffness kp feedback control or of a passive spring with a sufficiently low
compliance to prevent excess transmission of vibrations to the base frame.
The method to absorbing reaction forces in a balance mass requires the forces
to act as good as possible in a horizontal plane through the centre of mass.
Any deviation from this ideal situation will result in a proportional torque
around the horizontal axes through the centre of mass with corresponding
vibration forces in the vertical direction to the base frame.
A well designed structure with balance masses has helped keeping the trans-
mission of internal vibrations to such a moderate level that the vibration
isolation system is able to further reduce these vibrations to the level that
is required to achieve the maximum imaging quality of the wafer scanner.
698 Chapter 9. Precision positioning in wafer scanners

?˜
with internal
<! 


Metrology frame
Lens
Air mount Air mount

Granite stone
Base frame
<
Pedestal
š
"!!

Figure 9.5: The frames and springs that determine the sensitivity for external
vibrations. The transmission of vibrations from the fab floor takes
place via three mass-spring systems in series. The stiffness of the air
mounts and the mass of the metrology frame and lens determine the
low-frequency limitation of the main vibration isolation element. The
stiffness of the connection between the lens and the metrology frame
determines a resonating eigenmode that will increase the sensitivity
for vibrations at the corresponding eigenfrequency.

9.2.2 Vibration isolation

Figure 9.5 shows a rigid body diagram of the parts of the wafer scanner that
determine the sensitivity for external and internal vibrations.
The main line of defence against disturbing vibrations is determined by the
transmissibility transfer function of the three air mounts with the combined
mass of the metrology frame and the lens. In principle the air mounts consist
of air cylinders with a large volume of which the pistons are guided by means
of air bearings. These bearings have a sufficiently small air gap that it is not
necessary to apply the usual rubber bellows that would otherwise introduce
an increased transmissibility at higher frequencies by the intrinsic damping
of rubber. The stiffness value of the three air mounts together can be in
the order of 105 N/m. With a combined mass of the metrology frame and
the projection lens of approximately 2500 kg, the natural frequency with
these air mounts is in the order of 1 Hz. The relatively low stiffness causes
a practical problem in relation to changes in the mass of the isolated body.
The total static sag of the spring by the mass of 2500 kg would be equal
to a quarter of a metre and a small variation of this mass could lead to an
unacceptable change in the position of the projection lens. For that reason
9.2. Dynamic architecture 699

@! .
'
  xm

Velocity

'  
' 
to Force
with mass m with mass m xm
control loop

c ­<
k ! F k
xf

Floor Floor
?
' 
 @'®®¯¯ 


Figure 9.6: Active “skyhook” damping in a vibration isolation system avoids the
transmission of external vibrations through the damper. An absolute
velocity detector measures the velocity of the sensitive body relative to
an inert seismic mass and its output is used to exert a force opposite to
the velocity.

this position in the vertical direction is actively controlled by adapting the

amount of air in the air cylinders to the situation. An increased mass would
result in a reduced volume with an increased pressure in the air mount.
The volume can be restored by adding sufficient air from an external source
by means of an air valve that opens and closes as function of the vertical
position of the metrology frame.
The next problem with this system is the need for damping. At the nat-
ural frequency of 1 Hz the system will resonate, when excited with that
frequency by a force or a movement of the floor. As explained in Chapter 3 on
transmissibility, a normal viscous damper like a rubber bellows introduces
an unwanted connection, increasing the transmissibility of these vibrations
at frequencies above the natural frequency. For that reason a skyhook active
damper is applied. Figure 9.6 shows the principle that is based on the mea-
surement of the “absolute” velocity, relative to a real quiet reference as if
the system is ”hooked to the sky“. In practice this quiet reference consists of
an elastically suspended seismic mass inside the velocity sensor. A suitable
sensor for this principle is the geophone as was described in the previous
chapter on measurement, that measures the velocity relative to the seismic
mass, but the velocity signal can also be obtained by integrating the signal
of an accelerometer. The velocity signal is used to create a proportional
damping force, opposite to the velocity, by means of a Lorentz actuator of
which the force is only determined by the current and not by the position.
As a consequence this actuator has no stiffness that would otherwise add
700 Chapter 9. Precision positioning in wafer scanners

^
10
Low passive damping
1
Transmissibility [abs]
10 High passive damping
@' 

0
10

>
10

>^
10

>Z
10

>
10
> 0 1 ^ Z
10 10 10 10 10
Frequency [Hz]

Figure 9.7: The transfer function of the active skyhook damping shows an increased
attenuation of the vibrations in a frequency band starting at the eigen-
frequency of the mass-spring system.

to the stiffness of the supporting mechanical spring k and increase the

transmissibility.
When writing down the equations of motion the transfer function of the
transmissibility can be determined:

d2 xm d
m 2
= − c s ( xm ) + k( xf − xm ) (9.2)
dt dt
where c s equals the control gain of the velocity loop.
With the Laplace transform of the differentiation, the transfer function of xf
to xm becomes:
xm k 1
= = m cs (9.3)
xf ms2 + cs + k 2
s + +1
k k
With the known terms for the eigenfrequency ω0 and damping ratio ζ the
equation becomes:
xm 1
= 2 (9.4)
xf s s
+ 2ζ +1
ω20 ω0

This transfer function does not show the differentiating term in the numera-
tor that represented the increased transmissibility in the transfer function
of a passive damper and as a consequence the Bode-plot of Figure 9.7 shows
9.2. Dynamic architecture 701

clearly an increase in attenuation of the external vibrations, approaching

the level as would be obtained without damping.
Although this method of active damping is often used, its performance is
limited by the noise of the sensor. In principle the sensor needs to detect only
very small movements and any noise source will insert an disturbing signal
in the damping loop. In practice this reduces the positive effect of the active
damping at low frequencies. In reality the power spectral density of the
noise is not equal for all frequencies. A piezoelectric accelerometer with a
charge amplifier requires an additional integrator to create an inertial based
velocity signal. An integrator will amplify the noise at very low frequencies
more than at high frequencies due to the feedback capacitor of the amplifier.
As was explained in Chapter 8 this is especially problematic with excess
noise (1/ f ). Even though a direct inertial velocity sensor like the geophone
does not need an integrator, additional amplification at low frequencies
might be necessary to achieve a flat velocity response, sufficiently below the
natural frequency of the vibration isolation system.
At several research institutes investigations show promising results in the
optimisation of the the active part of vibration isolation systems. These
investigations include the active reduction of the stiffness of the connection
and a virtual connection of the position to a quiet reference. In spite of these
developments it is in practice still preferred to base the main part of the
vibration isolation based on a heavy mass and a compliant spring because
of the low-frequency noise and other dynamic problems in the sensors like
cross-coupling.

9.2.2.1 Eigendynamics of the sensitive parts

A large mechanical structure, like the metrology frame with the lens, in-
evitably shows several dynamic eigenmodes with their related eigenfre-
quencies. Figure 9.5 shows springs between the lens and the metrology
frame that represent the stiffness of the mutual connection. Even though
this connection is very stiff, it still results in a significant resonance with
the heavy lens. Also the optical elements inside the lens with their com-
pliant mechanical mounting determine eigenmodes that might be excited
by external vibrations. The related eigenfrequencies are often in a rather
low-frequency range around 50 – 100 Hz because of the large masses and
for that reason the vibration isolation system should work especially well in
that range.
The low amount of damping that generally is present in mechanical mounts
normally creates resonances with high Q levels that increasing the vibra-
702 Chapter 9. Precision positioning in wafer scanners

Flexure hinges Actuator sensor stack

Compliant parallel
guiding mechanism

Figure 9.8: An piezoelectric active lens mount with “smart-disks” that consist of a
combined piezoelectric actuator and sensor. By measuring the defor-
mation a damping force can be generated in an active feedback loop.
(Courtesy of Jan Holterman UT)

tions with a factor thirty or more. Even with a well controlled vibration
isolation system it is necessary to create additional damping to reduce the
amplitude at these eigenfrequencies. Damping in a mechanical mounting
structure is difficult to achieve with passive means because of the high
stiffness of the connection. With the related small movements it is almost
impossible to dissipate much energy in viscosity.

Smart disk

Damping in stiff connections can be created actively with actuators and

a suitable control scheme. The Piezoelectric Active Lens Mount (PALM)
of Figure 9.8 is an example that is used to dampen the eigenmode of the
connection of the projection lens with the metrology frame. Three of these
systems are required to create damping in six degrees of freedom.
The basic design of this system is the outcome of a research project at the
University of Twente by Jan Holterman under the guidance of Rien Koster
who invented the smart-disk that fulfils the key role in this system.
A smart-disk consists of a combination of a piezoelectric actuator and sensor
stacked on top of each other. In principle the sensor measures the forces
that act on the stack and the actuator creates a displacement as function of
the applied voltage.
A feedback loop can be created from the sensor to the actuator. Integration
9.2. Dynamic architecture 703

of the measurement signal in the loop (I-control) creates an effect as if a

damper is inserted in series with the smart-disk according to the following
reasoning.
For damping the force needs to relate to the velocity of the vibrating lens.
When Fd (s) equals the force signal from the sensor in the Laplace domain,
I-control will generate a displacement xd by the actuator:
ki sxd 1
xd = Fd ( s) =⇒ = ki = (9.5)
s Fd ( s) c

The force by the movement of the lens will generate a proportional velocity
of the actuator which is the behaviour of a damper with damping coefficient
c = 1/k i in series with the controlled smart-disk.

Anti-resonator or tuned-mass damper

A second method to reduce vibrations due to eigenmodes is by connecting

an anti-resonator, also called a tuned-mass damper to the sensitive body.
An anti-resonator is a tuned mass-spring system that is attached to the
sensitive body. The decoupling phenomenon of this additional mass-spring
system causes a typical combination of an anti-resonance and a resonance
in the transfer function from the excitation force to the sensitive body as
described in Section 3.3.1 of Chapter 3. The frequency where the transfer
function equals zero was shown to be:

1 ka
fa = [Hz] (9.6)
2π ma

where ka and m a are equal to the stiffness and mass of the anti-resonator.
The anti-resonance frequency f a is chosen equal to the undamped eigen-
frequency of the lens with its mounting stiffness. The combined transfer
function of the anti-resonator and the regular transmissibility for vibration
forces from the floor to the movement of the lens via the air mounts is shown
in Figure 9.9. The Bode-plot has been derived using the standard equations
of motion from Chapter 3 and it shows that the original resonance at 100 Hz
of this example is replaced by two resonances with a smaller magnitude
depending on the damping of the tuned mass.
The observed shifting of the resonance frequencies is especially useful when
the undamped system resonates in a frequency area where external distur-
bance peaks occur like for instance the 100 Hz hum from high power mains
supply units.
704 Chapter 9. Precision positioning in wafer scanners

c1 c^

F m1 m^

k1 x k^

>^
10

>Z
Compliance [m/N]

10

>
10

>\ Undamped eigenmode

10 Low damping
Medium damping
High damping
>`
10
ZJ `J 100 ^JJ
Frequency [Hz]

Figure 9.9: An anti-resonating tuned-mass damper replaces the original undamped

eigenmode of the lens with its support by two resonances and an anti-
resonance. At a certain value of c 2 the magnitude of the resonances is
reduced in comparison with the situation without the anti-resonator.

In those cases where the resonance frequency is not fixed it is also possible
to create an actively controlled anti-resonator. By changing the stiffness
with proportional feedback, the tuned-mass damper system can be adapted
to the right resonance frequency.
9.3. Zero stiffness stage actuation 705

9.3 Zero stiffness stage actuation

With the explanation of active vibration isolation a Lorentz actuator was

used to exert the damping forces. This actuator was chosen because a
Lorentz principle shows an almost ideal zero-stiffness behaviour, without
increasing the transmissibility for vibrations from the vibrating base frame
to the sensitive metrology frame.
In a similar way, also the requirements on the accuracy of the stages, with
their high level of velocity and acceleration, require the use of these zero-
stiffness actuators in order to completely avoid any elastic connection be-
tween the moving stage and the surrounding machine parts.
In a fundamental way, the positioning principle of the wafer- and reticle
stage is purely based on the second law of Newton, stating that the acceler-
ation of a body is proportional to the force and inversely proportional to the
mass of the body. As long as the forces are known and controlled, a body
with a known mass can be accurately positioned by a feedback controller as
was explained in Chapter 4 with the PD-control positioning of the optical
disc readout unit.
In the design of a stage of a wafer scanner most of the effort is invested in
full deterministic control of the forces acting on the moving body, consisting
either of the wafer or of the reticle on their supporting table. As long as
these bodies behave like a rigid body with predictable and reproducible
dynamics, the movement of the body is only determined by the forces acting
on it. With this reasoning, the term “zero-stiffness” is only related to the
mechanical connection to sources of disturbing vibrations and it does not
refer the control stiffness. When zero-stiffness of the actuation is achieved,
the body will behave like an ideal inertial mass and the task of the control
system is to connect the stages as stiff as possible to the optical system.
More precisely stated, the control system needs to connect the reticle stage
as stiff as possible to a planned scanning trajectory, relative to the optical
system. Simultaneously the control system needs to connect the wafer stage
as stiff as possible to the scanning image of the moving reticle.
The wafer stage is mostly taken as example in the following sections but the
design principles are equally true for the reticle stage. The main difference
is the higher velocity and acceleration with reduced accuracy of the reticle
stage in respect to the wafer stage.
706 Chapter 9. Precision positioning in wafer scanners

9.3.1 The wafer stage actuation concept

From all possible actuator types, only the mentioned Lorentz actuator almost
ideally complies with the zero-stiffness criterion. Piezoelectric actuators are
stiff by definition and reluctance based actuators have a negative stiffness
that induces a comparable, however inverted, transmission of vibrations as
a positive stiffness.
Unfortunately a simple Lorentz actuator offers only a small linear range.
In order to solve this limitation, electronic commutation was introduced in
Section 5.2.5 of Chapter 5. The shown example of Figure 5.23 had a long
set of three moving coils with a large magnet assembly and in spite of the
electronic commutation the range was still rather limited because the coil
set and magnet assembly both needed to have the same length in the motion
direction.

9.3.1.1 Wafer stepper long-range Lorentz actuator

With the design of the first wafer stepper an alternative two phase elec-
tronic commutated Lorentz actuator was designed by one of the authors of
this book at Philips Research laboratories. This extended-range Lorentz
actuator was able to work without the above mentioned limitation and was
applied in the wafer stages of all wafer steppers until the introduction of
the wafer scanner that required a better solution.
Figure 9.10 shows the thinking steps that have led to this design. It starts
with a standard loudspeaker type Lorentz actuator with the permanent
magnet system as moving part.
The first transformation step (1) is to divide the moving ferromagnetic yoke
in a moving outer part that slides over the cylindrical inner part that holds
the coil. This step reduces the moving mass and is allowed as still the
permanent magnet field is moving in respect to the coil, creating a dΦ/d x.
The sliding could be an air bearing but in the next steps the slider will be
avoided.
Step (2) is the addition of an equal permanent magnet-coil set around the
common central yoke. This step increases the force at a certain current level
and becomes even more useful when the direction of both the permanent
magnetic flux and the current in the coil of one of the actuators is reversed
in step (3).
The sliders are deleted in the step (4) because after step three the flux of
both permanent magnets can be shared. This step is essential as now there
is a symmetrical set of two permanent magnets connected by one simple
9.3. Zero stiffness stage actuation 707

Slider
1: Coil connected with a sliding centre yoke. 2: Take two to create symmetry.

3: Reverse  4: Connect the stator.

5: Enlarge the coils. 6: !"•

Q

1 2 1 2 1 2 1 1 2 1 2 1 2 1

1 2 1 2 1 2 1 1 2 1 2 1 2 1

7: Short interleaved coils reduce reluctance force. 8: š!"

^ Q

1 2 1 2 1 2 1 1 2 1 2 1 2 1

1+ 2+ 1- 2- 1+ 2+ 1- 1 2 1 2 1 2 1

9: Current in coil 2 is commutated. 10: Further moon is possible.

Current directed
away from reader
Current directed
towards reader

11: Increased force with two permanent magnet secons.

Figure 9.10: Design of a zero-stiffness Lorentz type actuator with long linear-motion
range by electronic commutation. The direction of the current in the
coils corresponds with a force on the moving part to the right. The
curved arrows indicate the direction of the permanent magnetic flux.
708 Chapter 9. Precision positioning in wafer scanners

Rz
y

Figure 9.11: The usual way of creating a three-axis positioning system is done by
stacking three single-axis stages on top of each other. The driving
forces are not directed to the centre of mass, resulting in unwanted
torques and rotations. For that reason this configuration is not suitable
for fast stages with a high precision.

moving yoke that moves above two coils that are enlarged in the step (5) to
give a larger range.
Unfortunately the large coil will induce a reluctance force acting on the
moving part of the yoke (6), driving it in the outer positions where the reluc-
tance for the magnetic field of the coil is the least. This problem is largely
reduced by choosing shorter coils and applying electronic commutation as
shown in step (7).
The windings around the ferromagnetic yoke consist of two coils that are
interlaced with an alternating winding direction. This means that the first
section of coil one is wound clockwise, followed by a clockwise wound first
section of coil two, again followed by the counterclockwise wound second
section of coil one, and so on.
With commutation , reversing the current direction at any coil, the sections
of that coil that are wound clockwise give a counter clockwise current and
the other way around. This is shown in steps (8 – 10) for a movement to
the right. In principle the stator with the coils and the inner ferromagnetic
yoke can be made infinitely long by adding more sections to each coil and in
step eleven also the moving part is shown to be multiplied to increase the
force.
The main problem of this configuration is the efficiency, as the current has
to run through the entire coil, including those coil sections that are outside
the permanent magnetic field. This disadvantage could be solved by feeding
each coil section with a separate power amplifier but that was not done in
the real system in order to avoid disturbing forces due to switching the coils
on and off. It also would require a lot of separate wires while otherwise one
actuator only needs four wires with two amplifiers to control the current in
a smooth way without disturbing switching moments.
9.3. Zero stiffness stage actuation 709

Y1
X
Roller
bearings Y2

Figure 9.12: Original 3-D sketch and photographic image of the first wafer stage
with electric linear actuators for a wafer stepper in a H-configuration.
The magnetic mover is supported by roller bearings that run on rails
alongside of the coil sections while an air-bearing is used to support
the wafer chuck and mirror block, floating over a flat granite stone.

The pulse-width modulated design principle of the applied amplifiers was

presented in Figure 6.69 of Chapter 6.

9.3.1.2 Multi-axis positioning

The different directions in a wafer stepper and scanner are defined in a

metrological coordinate system that is oriented relative to the scanning
direction and the projection lens. The horizontal direction x equals the
stepping direction of the waferstage while y equals the scanning direction
in both wafer and reticle stage. The vertical direction z is directed according
to the optical axis of the projection lens while the rotations around the
translational axes Rx , Ry and Rz determine the remaining directions. In the
first wafer stepper, the described extended-range linear Lorentz actuator
only needed to provide movements in three directions in the horizontal
plane as the wafer was oriented parallel to the image on the wafer chuck by
mechanical means.
The standard method at that time for multi-axis positioning was to stack
different single-axis motion systems on top of each other like shown in
Figure 9.11. A three-axis stage, working in the x − y − Rz direction, generally
consisted of two linear drives with a rotation table on top. Due to the
stacking, the driving force could never be directed towards the centre of
mass of the moving system. The first actuator had to drive also the other
actuators while only the last actuator was close to the centre of mass of the
moving stage.
710 Chapter 9. Precision positioning in wafer scanners

Short-stroke Lorentz actuators

y1 Wafer y2
chuck
x
Long-stroke actuators

Figure 9.13: Schematic presentation of a 3-axis long-stroke, short-stroke wafer

stage in a H-configuration. Both long-stroke and short-stroke actua-
tors work in the same plane through the centre of mass of the wafer
table. The rotation along the z−axis is achieved by the difference in
positioning of the y actuators.

For this reason a different configuration was chosen, named after the letter
“H” of the alphabet with one linear actuator for the linear movement x and
two linear actuators y1 and y2 for the linear movement in the y direction,
allowing a limited rotation Rz by the difference of y1 and y2 . Figure 9.12
shows this configuration where the bearings consisted both of roller bearings
to support the permanent magnetic movers on the coil stator and the air foot,
an air bearing floating on a flat granite stone surface to support the wafer
stage itself. With later versions of the wafer stepper many improvements
were added to this basic system with for instance three separate actuators
to position the wafer in the three remaining directions, z − Rx − Ry that were
needed in order to guarantee the vertical position and flatness of the wafer
relative to the focal plane of the lens.

9.3.1.3 Long- and short-stroke actuation

The single-stroke extended-range Lorentz actuator from the previous section

was sufficiently accurate to serve in a wafer stepper that only had to stand
completely still at exposure.
With the wafer scanner, however, the exposure takes place during a move-
ment at a constant velocity. Furthermore the increased requirements on
resolution and overlay posed more severe requirements on precision and
acceleration. This necessitated the design of a wafer stage without the in-
9.3. Zero stiffness stage actuation 711

R S T R S T

Long-stroke high-power actuator

Mirror block with wafer chuck

Air foot on granite stone

Figure 9.14: The wafer stage of a wafer scanner. The three-phase coil section with
ferromagnetic yoke of the powerful long-stroke actuator runs over a
permanent magnet stator. The mirror block is used to measure the
position of the wafer stage with a laser interferometer.

herent drawbacks of the linear actuator that was used in the wafer stepper,
like the remaining parasitic forces by the commutation and the friction
variations of the applied roller bearings.
The solution was to use a non-commutated short-stroke Lorentz actuator
in cooperation with a second actuator, the long-stroke actuator, as shown
schematically in Figure 9.13.
This long-stroke actuator is connected to the stationary part of the short-
stroke actuator and only needs to position this part in such a way that
the permanent magnet system of the short-stroke actuator remains in the
±0.1 mm range around the centre of its range with the least stiffness value.
This configuration allows the long-stroke actuator to operate at a reduced
precision of only this same level of ±0.1 mm, even though the precision of
the short-stroke actuator needs to be on sub-nanometre levels.
The movements and vibrations of the long-stroke actuator are hardly trans-
ferred to the sensitive wafer position because of the low stiffness of the
short-stroke actuator. For that reason it is allowed to use strong electroni-
cally commutated three phase actuators with ferromagnetic yokes in the
long-stroke actuation system as shown in Figure 9.14.
This actuator is optimised to force rather than to precision positioning and
zero-stiffness with a ferromagnetic yoke around the coils. This yoke will
cause cogging forces towards those positions where the reluctance for the
permanent magnet field is smallest. Cogging forces are perceived as bumps
when moving the mover by hand over the total stroke.
712 Chapter 9. Precision positioning in wafer scanners

…*'



*

z,Rx ,Ry-axis
short-stroke
actuators
x-axis long-stroke actuator

x,y,Rz-axis
short-stroke
z
actuators x
Rz Rx
Ry
Top-view wafer stage with
y
*


Figure 9.15: The short-stroke actuators are integrated within the mirror block.
The actuation in the horizontal plane is done by flat Lorentz actua-
tors while the out-of-plane movements are provided by loudspeaker
type Lorentz actuators without a ferromagnetic yoke to prevent non-
linearity by reluctance forces.

The advantage of the ferromagnetic yoke around the coils is the low reluc-
tance path for the magnetic field that increases the force to current ratio of
the actuator. This actuator can easily transport the required cables, wires
and water cooling tubes for both the short-stroke coils and its own coil sec-
tion.
Fortunately only the in plane motions are large in a wafer scanner, so only
those motions require a corresponding long-stroke actuator. The out-of-
plane motions, needed to keep the surface of the wafer in the focal plane
of the lens are realised by separate short-stroke actuators only, as shown
in Figure 9.15. These loudspeaker type Lorentz actuators are designed
without a ferromagnetic yoke in order to prevent reluctance forces as was
explained in Section 5.3 of Chapter 5.
9.3. Zero stiffness stage actuation 713

N S N S N S N

N S N S N S

S N S N
F S N S

S N S
B N S N

N F S F N S N I S N

S N S I N S N S

S N S N S N

N S N S N S N

Figure 9.16: The forces acting on a coil, positioned in the magnetic field above an
array of alternating permanent magnets. Depending on the position
in the field, the forces can be in-plane, perpendicular or a combination
of both.

9.3.2 Full magnetic levitation

The guiding of the long-stroke part of the H- configuration wafer stage is

mainly realised by means of air bearings. The air foot provides the support of
the heavy centre piece with the short-stroke actuator coils and is preloaded
by a closed area with vacuum between the air foot and the granite stone
to achieve a sufficient stiffness. The long-stroke actuators themselves have
integrated air bearings that are preloaded by the attracting magnetic forces
between the mover and the stator.
In future wafer scanners with EUV light of 13 nm, air is not allowed in
the light path because it absorbs the light.For that reason these machines
are operated in vacuum. Although it is in principle possible to realise air
bearings in a vacuum chamber, by adding special measures to locally extract
the air that escapes from the side of the air bearings, it is better to avoid
them. For that reason a long-stroke stage has been developed that uses
an electromagnetic planar actuator, working in six degrees of freedom. Its
working principle is shown in Figure 9.16. A permanent magnetic plat-
form, consisting of a plurality of alternating permanent magnets, creates a
magnetic field with different field directions. A current-carrying coil will
experience a force of which the direction depends on its position on the
magnetic table and the direction of the current. By using different coils at
different locations a configuration can be created that is able to exert forces
and torques in all six degrees of freedom at any position of a free moving
stage, when combined with a suitable electronic amplifier system.
714 Chapter 9. Precision positioning in wafer scanners

@
'‚«„
2
600

500

400

300

200

100

0
3
@ 2.5 800
! 2

 600 „
« 1.5
‚™«
@ 1 400
"
!
 0.5 200  

 
 @!

Figure 9.17: All electric actuators have a limitation in the amount of force per unit
of moving mass of the actuator itself, the specific peak force. With a
given mass that has to be actuated, an increase of the acceleration
would require an increase of the mass of the actuator. The relative
added acceleration flattens out at a high ratio between the mass of the
actuator and the actuated mass with an asymptote at the maximum
specific peak force.

9.3.3 Limits in acceleration of reticle stage

The continuous increase of the acceleration levels of reticle stages poses

several challenges that are increasingly not trivial. The main issue is that
both the long-stroke and short-stroke actuator are essentially placed in
series, which means that they both need to deliver the same force. This high
force is especially problematic for the Lorentz actuator.
Like any other actuator also a Lorentz actuator has thermally determined
limitations in its force and power capability. The high power dissipation
in the resistance of the windings by the motor current must be removed
by means of a closed water circuit. At a certain level of heat per unit of
volume of the windings, the temperature of the windings will reach the
maximum value that the insulation can sustain, determined by the heat
conductivity of the windings and the cooling circuit. The volume of the
windings is proportional to the size of the moving permanent magnetic part
of the actuator which leads to a maximum in the amount of force per unit
of moving mass of the actuator itself. This specific peak force of a Lorentz
actuator is approximately 650 N/kg and corresponds with the maximum
acceleration that the actuator could realise without the mass of the actuated
9.3. Zero stiffness stage actuation 715

body.
A direct consequence of this relation is that the total moving mass of the
stage becomes noticeable larger than only the mass of the actuated body
that carries the wafer. For the required maximum acceleration of 120 m/s2
as mentioned in Table 9.1, the mass of the actuator is still below 50 %
of the mass of the actuated body and this can as yet be provided without
much problems, but when more acceleration is needed in the future, the
total mass will rapidly increase. For this reason, present developments on
the stages focus on the reduction of the moving mass by the application of
hollow structures with thin plates and reinforced plastics. Unfortunately
this mass reduction does not only have benefits as the sensitivity for external
disturbing forces by transmission is increased. The counterbalancing effects
of the mass with on the one hand the need to decrease the mass in order
to increase the maximum acceleration and on the other hand the need to
increase the mass in order to reduce the sensitivity for external disturbing
forces is called the mass dilemma of precision positioning systems. In other
words: ”A reduced mass requires improved system dynamics that enable
a higher control bandwidth to compensate for the increased sensitivity for
external vibrations“.
716 Chapter 9. Precision positioning in wafer scanners

9.4 Position measurement

Several position measurement systems are applied in wafer scanners. The

following list mentions the most important ones:

• The alignment system that measures the position of the wafer relative
to the wafer stage with sub-nanometre accuracy.

• The level sensors that measures the surface profile of the wafer relative
to the wafer stage with nanometre accuracy.

• Long-range laser interferometers or encoders for real-time incremental

position measurement of the stages relative to the metrology frame
with sub-nanometre accuracy.

• Capacitive sensors for small relative measuring distances and ranges

like the vertical position of the reticle stage relative to the metrology
frame with nanometre accuracy.

• Optical proximity detectors for surface measurement of the wafer with

nanometre accuracy.

• Conventional encoders for less critical measurements like the in-

ternal position measurement of the long-stroke actuators with sub-
micrometre accuracy.

• State-of-the-art encoders for extreme-precision measurements of very

small displacements with an accuracy of several picometres that are
used to measure the relative position of optical elements inside the
projection lens.

The most important and complicated measurement in a wafer scanner is

related to the overlay, the position of the previously exposed layers on the
wafer relative to the image of the reticle. The complication in this metrology
loop from image to previous layer is based on the fact that this measurement
consists of several relative measurements of which some are not possible
in real-time. The main cause for this problem is that the projection lens
closely covers the part of the wafer that is exposed, thereby preventing any
direct measurement of the image and the previous layer at that location.
The only alternative to a direct measurement is an indirect measurement
where the position of the previous layers on the wafer and the image location
are separately measured relative to a sufficiently stable reference with a
well-known position in the wafer scanner. This reference is the real-time
9.4. Position measurement 717

measured wafer stage. The 6-axis incremental position measurement of

the stages is continuously available and defines the connection between the
stages and the metrology frame, the central reference that is connected to
the lens. This real-time measurement system is also used for the closed-loop
feedback control and synchronisation of the reticle stage and wafer stage
with strict requirements on latency.
A consequence of this indirect measurement is the addition of all mea-
surement errors that occur during the different measurement steps. This
addition can be calculated by means of the “root of the sum of squares” of the
errors, as explained in the Chapter 8 on measuring, but it requires that the
accuracy of all measurements have to remain well better than the overlay
specification, in practice below one nanometre.
Fortunately the requirements on overlay of a few nanometres are only strictly
relative. There is no traceable relation with the standard metre other than
the order of magnitude. Traceability with a proven uncertainty level over
a long period of time is not important. All calibrations are related to the
previous layers on the wafer and only need to remain stable between the
moment that the previous layer is measured relative to the wafer stage
and that the exposure is finished. Furthermore underlining the relative
nature of the measurements is the fact that the wafer itself is not stable as
during the chemical processing at high temperatures the previously defined
structures can be moved by deformation of the wafer. By measuring these
deformations both the magnification of the lens can be adapted and the
positioning of the stages can be made to accommodate to these deformations.
The following sequence of different steps describes more in detail the total
measurement process from the location of the previous layer to the image
position:

1. The wafer is pre-aligned in the wafer handler with optical sensors to

measure the location of the flat edge or the notch at one location on the
side of the wafer that determines the basic orientation of the wafer.

2. A robot brings the wafer from the pre-alignment chuck to the wafer
stage. The position of the vacuum gripper of the robot is alternately
referenced to the pre-alignment chuck and the waferstage by a kine-
matically determined mechanical connection. The total resulting po-
sitioning error of the wafer on the waferstage is approximately ten
micrometres.

3. The position of the previous layers on the wafer is measured by means

of alignment marks that are defined during the exposure of the first
718 Chapter 9. Precision positioning in wafer scanners

layer. The number of these reference marks can range from two till
the total number of dies. Similar alignment marks, called fiducials,
are also stably attached to the wafer stage, close to the wafer. The
alignment system measures the position of the alignment marks on
the wafer relative to the position of the fiducials.

4. The waferstage with the aligned wafer moves to the projection lens
under real-time measurement relative to the metrology frame. The
position and focal plane of the image is measured relative to the wafer
stage with a special image sensor. It is assumed that this information
remains stable during one or several exposure cycles.

5. The fiducials of the wafer stage are measured either to a reference

inside the lens or relative to alignment marks on the reticle through
the projection lens.

In the following subsections the alignment sensing principle, the measure-

ment of the focal plane and the real-time metrology loop with the long-range
incremental measurement systems will be presented.

9.4.0.1 The alignment sensor

The principle of alignment with the alignment marks as reference for the
pattern on the wafer is shown in Figure 9.18. The alignment marks are
phase gratings and the measurement is comparable with the interferometric
encoder of Section 8.8.1.3.
The grating is illuminated with a laser and the phase steps with a height of
0.25λ create diffraction orders mainly in the two first orders. Both orders are
recombined at the photo detector after interfering with a second reference
grating. The interference irradiance depends on the phase relationship of
these orders. As was explained with the interferometric encoder, the phase
shift by the movement in the +1st order is opposite and equal in magnitude
to the phase shift of the −1st order, resulting in a doubling of the combined
differential phase shift, thereby increasing the measurement resolution
with a factor two.
The alignment is done by a scanning movement of the wafer stage through
the range where optimal alignment should take place and the zero crossings
of the irradiance signal are measured. These zero crossings repeat with
every half grating period due to the doubling effect of the two diffraction
orders which limits the range of this alignment system to half a period of
9.4. Position measurement 719

8.8 Pm 8 Pm Photo detector

17.6 Pm
Detector
output
Lens
posion
Mutual zero crossing
is aligned posion Reference
Projecon
lens

16 Pm Alignment
Wafer laser beam
O/4
-1 +1
O/4 Wafer
Markers

Fiducial on waferstage

Figure 9.18: The alignment marks are reflective phase gratings with different
periodicity, to increase the capture range. One way to measure the
markers is shown at the right, where both first orders are imaged on a
reference transparent phase grating and their disturbance is detected
by a sensor. This configuration increases the position sensitivity with
a factor two. Note the alignment mark in the ASML logo.

the alignment grating, being 8 μm.

This value is too small for the wafer handler to reliably position the wafer
inside the capture range of the alignment system. A misalignment of 8 μm
would mean that the entire exposure becomes erroneous.
To avoid this problem and increase the capture range, a Vernier principle is
applied by simultaneously measuring a second grating with a 10 % different
grating period. The zero crossing of the irradiance by the two gratings will
only coincide in one on every ten periods of the irradiance signal of one
grating, increasing the range with a factor ten. In practice a capture range
of approximately 40 μm is adhered in order to avoid errors by outliers in
the measurement.
An interesting effect of the phase grating is that the phase shift by the
displacement remains the same irrespective of the height of the grating.
As was shown with the interferometric encoder, the phase shift is only
determined by the ratio between the displacement and the grating period.
This implies that the position detection is not affected by the height of the
grating. On the other hand, the irradiance in both 1st orders does depend
on the height of the grating with a maximum at a step height of 0.25 · λ. At
720 Chapter 9. Precision positioning in wafer scanners

250

Depth of focus (nm)

~
~
#"
200
150
100
50
0
100 90 80 70 65 57 50 45 38 32 26
š


"
‚„

Figure 9.19: The required depth of focus (DOF) as function of the critical dimen-
sion. After the introduction of immersion, the requirements were less
stringent.

a certain moment Chemical Mechanical Polishing (CMP) of the metal layers

was introduced in IC manufacturing to flatten the surface in order to achieve
a better focusing performance and it was feared at the time that this CMP-
process would erase the alignment marks. Fortunately it appeared that a
much shallower grating than the optimal value still provides just sufficient
signal for alignment. With some additional measures on sensitivity of the
optical sensors the remaining phase differences over the flattened gratings
are still detectable with sufficient accuracy.

9.4.1 Keeping the wafer in focus

The high numerical aperture that is necessary for a better resolution au-
tomatically implies a decreased depth of focus (DOF) as was explained in
Chapter 7. The three dimensional structures on the surface of a wafer, that
are created during the many manufacturing process steps, have made it
increasingly difficult to guarantee a sufficiently flat surface to accommodate
this continuously decreasing depth of focus of the projection lens. Figure 9.19
shows the relation between the depth of focus and the critical dimension. A
focal error of less than 100 nm over a die of 25 mm is a challenging require-
ment. To achieve such extreme specifications, this problem is approached
from two sides. The first step is to guarantee a certain flatness of the wafer.
As previously explained the deposited metal layers for the wiring are often
treated by a CMP-process to keep the surface non-planarity locally within
acceptable levels to less than 50 nm over a die.
The second step to achieve a sufficiently small focus error is to actively
position the wafer within the focal area at the region of the exposed die.
This process is also called the levelling of the wafer and with the first gen-
erations of the wafer scanner the levelling was done by means of real-time
9.4. Position measurement 721

measurement of the wafer surface ahead of the lens during scanning. This
measurement was done by means of an optical triangulation method like ex-
plained in Section 8.6.1.2 of the previous chapter on measurement. A servo
tracking feedback control system kept the surface approximately parallel
to the image in the focal plane. Unfortunately feedback control can never
prevent some remaining focal error due to the positioning system dynamics
and the limited control gain. A better solution was found to determine the
height profile of the wafer prior to exposure and use that information in
feedforward control at exposure.
With a single wafer stage a separate measurement of the surface topology of
the wafer would require a decrease in the productivity due to the additional
time for the measurement. The dual wafer stage with parallel measurement
and expose cycles was designed to solve that problem.
722 Chapter 9. Precision positioning in wafer scanners

Map
Align wafer
wafer

Load Stage
Stage
wafer Measurement Exposure
cycle Swap
Swap cycle
Expose
Unload Wafer
Wafer
Align
Recle

Figure 9.20: The process steps in a dual-stage wafer scanner between loading and
unloading consist of two cycles. In the measurement cycle the wafer is
aligned to the wafer stage and the surface is mapped for focus. In the
exposure cycle the wafer stage is aligned to the reticle and the wafer
is exposed.

9.4.2 Dual-stage measurement and exposure

Productivity has been the main driver for the decision to design a wafer
scanner with parallel processing of the measurement and the exposure cycle.
Next to the possibility to precisely measure all dimensional details of a
specific wafer without sacrificing throughput, the increased utilisation of
the expensive lens is another important reason for this duplication of the
wafer stage. With a dual wafer stage the lens can expose one wafer during
the time that the other wafer is measured and as a consequence the lens is
continuously utilised. Figure 9.20 shows the process steps of a dual-stage
wafer stage. The measurement cycle of one wafer is executed synchronous
with the exposure cycle of another wafer. As a result of the measurement,
the exposure starts with a wafer that is fully known in respect to the position
of the alignment markers to the wafer stage and the height profile of the
surface. This enables a maximum use of feedforward control.
9.4. Position measurement 723

z1-axis
Rz Ry y
x-Rz-axis reference mirror
Rx
x
z2-axis
reference mirror

x-axis

z-axis
o
45 mirror
y-Rz-Rx-axis

Figure 9.21: A mirror block for 6-axis position measurement by means of laser
interferometry. The z-axis is realised with a mirror under 45◦ that
reflects the light upwards to a reference mirror. All axes are dual
path plane mirror interferometers and Ry is determined from two z
interferometers.

9.4.3 Long-range incremental measurement system

The long-range measurement systems in the wafer- and reticle stage have to
be able to track the position with the specified accuracy and velocity levels
without adding latency which would reduce the accuracy of the synchroni-
sation between the wafer stage and reticle stage.
Until the end of the last century only laser interferometer measurement
systems were capable of measuring large distances with nanometre level
resolution and a latency timing that neither limited the bandwidth of the
feedback control nor interfered with the synchronisation of the stages. Sam-
ple rates far above 20 kHz are presently state of the art and with suitable
synchronised phase detection electronics a large number of measurement
axes could be realised that all worked in parallel.
Figure 9.21 shows the mirror block of a wafer stage, showing the laser beams
that are used to realise a six-axis real-time measurement system.
A special configuration is used to derive the z-axis measurement signal with
a 45◦ mirror to direct the light upwards from the wafer stage. The light
is reflected back by a stationary reference mirror that is connected to the
metrology frame and returns via the 45◦ mirror to the interferometer.
724 Chapter 9. Precision positioning in wafer scanners

With this configuration, the total optical path length of this beam is de-
termined both by the x and the z distance. By subtracting the directly
measured incremental x movement from another laser beam, the z position
is obtained.
The accuracy of laser interferometer measurement systems is strongly in-
fluenced by the refractive index of air according to the Edlén equation
(8.86) that was presented in Section 8.8.2.3 of the previous chapter on mea-
surement. The main problem with interferometer distance measurement
systems is that they measure over a long distance through air. At large
distances, small variations in the light velocity in air, that are induced by
the temperature and/or the pressure, can have a significant influence on
the measurement. To keep the accuracy of the measurement on accept-
able levels, air showers are used to condition the air over the full optical
measurement path. The temperature of the air is required to be controlled
within about 1 mK stability over an exposure cycle. Furthermore the air
has to flow at a higher speed than the maximum speed of the stages in
order to prevent mixing of the conditioned air from the air showers with
air coming from elsewhere at a different temperature. These requirements
have resulted in significant developments that always have just met the
required specifications. Still it is assumed that on the long run this method
will no longer be adequate and that also the high speed of the conditioned
air might excite dynamic resonances in the sensitive parts.
Fortunately another solution is found in a planar reflective-encoder system
with a phase grating that measures in all six directions with a sufficient
resolution to be applied as the long-range measurement system of a wafer
scanner. This alternative has two advantages over the laser interferometer.
First of all the thermally induced refractive index changes have far less
effect because of the short path through air of less than 15 mm. The thermal
sensitivity is reduced with a factor 100 to ΔL/L = 2 · 10−8 per degree Kelvin.
The second advantage has to do with the stability of the real-time metrology
loop as is explained with the help of Figure 9.22.

9.4.3.1 Real-time metrology loop

The real-time metrology loop is a subset of the full metrology loop that was
defined at the start of this section and contained also measurements that
were not performed in real-time. The real-time metrology loop assumes that
all alignment steps are done well and remain stable during the exposure.
This means that the real-time metrology loop connects the lens via the
metrology frame to the wafer on the waferstage.
9.4. Position measurement 725

?˜ ?˜

Metrology frame Metrology frame

~" –

< 15 mm
Mirror 

#" #"

100-500 mm

Figure 9.22: Two methods for long-range incremental position measurement. The
laser interferometer, shown at the left has long been the standard but
is partly replaced by the plane encoder system as shown at the right.
The difference is seen in the length and stability of the metrology loop
and a reduction of thermally induced instability by refractive index
changes of air.

Figure 9.23: The combination of the H-configuration with the laser interfero-meter
measurement system on the top and the plane encoder with the planar
actuation system at the bottom clearly illustrate the differences.
726 Chapter 9. Precision positioning in wafer scanners

The first measurement in this real-time metrology loop is the position of

the lens with respect to the metrology frame. As was shown previously
the connection between the lens and the metrology frame is not stiff and
can even contain active dampers to control vibrations. Because of the low
stiffness, separate optical or capacitive sensors are applied to measure any
small displacements of the lens relative to the metrology frame.
The second measurement in the real-time metrology loop is the position
of the wafer stage with respect to the metrology frame by the long-range
measurement system. With laser interferometers, the interferometer part
is firmly connected to the metrology frame at a distance from the optical
axis that is determined by the maximum movement range of the wafer
stage. This distance requires the metrology frame to be stable over the total
distance from the lens to the interferometer. In principle this problem can
be solved by adding a reference measurement axis from the interferometer
to the lens but that would increase the induced errors by the refractive index
variations of air due to this additional measurement.
With the planar encoder, the measurement grating, called the grid plate is
stationary mounted on the metrology frame and consists of thermally inert
material like Zerodur, a glass ceramic made by Schott with zero thermal
expansion in a limited temperature range. This approach both reduces
the length of the metrology loop through air and decreases the stability
requirements of the metrology frame.
The last part of the metrology loop within the wafer stage from the measure-
ment mirror to the wafer is quite simple and straightforward for the laser
interferometer option. The mirror block forms a “monolithic” structure with
the wafer chuck that holds the wafer firmly attached on the wafer table by
vacuum and this monolithic structure is also made from a thermally inert
material like Zerodur.
The plane encoder version has a more complicated last part of the loop to
the wafer as the sensor part is located on the moving part of the wafer
stage. The potential problems regarding stability of this part are solved
by both fully integrating all optical parts in one unit and by continuous
calibration with redundant measurement axes. In principle four sensors
are used that each determine more degrees of freedom. The superfluous
degrees of freedom are used to compare and calibrate all sensors at every
exposure cycle. Figure 9.23 shows the different drive and measurement
configurations for comparison.
9.5. Motion control 727

9.5 Motion control

Figure 9.24 shows the basic principle of the motion control of the wafer stage.
A modern control expert will miss elements like Multiple Input Multiple
Output (MIMO) controllers, observers and other more recently developed
control approaches.
It is true that the total controller itself consists only of six separate Single
Input Single Output (SISO) controllers for all directions with feedforward
and feedback paths. The feedback controller uses a well tuned classical
Proportional, Differentiating and Integrating PID-control algorithm.
The six-axis machine coordinate system for the position measurement and
the controller has its orientation at the centre of the image position. This is
not the place where the centre of mass of the wafer stage, nor the actuators
of the wafer stage are located. For that reason the six control forces Fc are
first transformed by a Gain Scheduling matrix operation into the forces Fcom
that should be exerted at the centre of mass of the positioning stages. The
gain scheduling matrix is constantly adapted to the actual position of the
stage. A second transformation is applied to determine the corresponding
forces of the actuators Fa by means of a Gain Balancing matrix operation.
The gain balancing matrix has to be well tuned to the physical hardware of
the wafer stage in order to compensate for performance deviations due to
tolerances in the different functional parts.

Feedforward
control

–<
Posion Stage
setpoint +
+ Fc Fcom F 

+ GS GB
-
–°

Feedback ActuatorSystem Mechanics

control

Figure 9.24: The position control of the wafer stage is a SISO 6-axis PID-control
system. The Gain Scheduling matrix (GS) transforms the control
forces (Fc ) in six degrees of freedom, according to the machine co-
ordinate system into the forces (Fcom ) of the wafer stage coordinate
system around the centre of mass. The Gain Balancing matrix (GB)
transforms these forces into the actuator forces (Fa ).
728 Chapter 9. Precision positioning in wafer scanners

16 40

12 30

8 20

Wafer stage error [nm]

Accprole [m/s2]
4 10

0 0

-4 -10

-8 -20

-12 -30

-16 -40
0 0.05 0.1 0.15 0.2
Time [s]

Figure 9.25: The wafer stage error as function of time induced by a mismatch of
0.1% of the feedforward force when a well-tuned feedback is applied.

The reason for this rather simple and straightforward approach in control
is twofold. First of all this methodology allows a more direct investigation of
observed errors in the system. This shortens the time for trouble shooting
at the customer site, when the system has failed, or when parts have been
exchanged and the system should be tuned again. The second reason is
that the feedback part only plays a role in the robustness of the control to
deviations of the system parameters from the ideally modelled situation as
will be further explained in the following section.

9.5.1 Feedforward and feedback control

A property of feedback control is the fact that it needs an error to act anyway,
so it is allowed to say that Feedback is always (too) late. The related
delay manifests itself in the settle time that is the time, needed to reduce
the effects of an disturbing event to the specified maximum error.
Feedback should only be applied to correct errors that could not be avoided
by feedforward control due to random disturbances or uncertainties in the
properties of different elements of the system. In the design of a wafer scan-
ner the focus is on creating a dynamic system that behaves as deterministic
as possible, mainly because of the impact of settle time on throughput.
A small example can place this requirement in a realistic perspective.
Imagine a wafer stage with the specifications as mentioned in the list of
Table 9.1, a mass of 20 kg, accelerating with 30 m/s2 to a maximum speed of
0.5 m/s. With these numbers the required force during acceleration becomes
9.5. Motion control 729

600 N and the acceleration time takes ≈ 17 ms.

Assume that this situation is the result of a pure open-loop inertia-based
feedforward controlled system with a maximum error in the force of only
0.6 N (=0.1 %!). Then a small calculation gives the following position error
after the 17 ms acceleration time.
0. 6 ( )2
ε = 0.5at2 = 0.5 · 17 · 10−3 ≈ 4 · 10−6 m (9.7)
20
This is still a factor 400 above the required 10 nm level. Improving the
feedforward force to the required level would mean a total error in the
order of 1 ppm (1 : 106 ) and those are levels that can only be realised in
electronics. This all clearly indicates the need for an additional feedback
control action to reduce the remaining error and Figure 9.25 shows a realistic
graph of the modelled performance of a waferstage with a well-tuned PID
feedback controller that illustrates the strong improvement that is achieved.
The feedback control starts almost immediately after the beginning of the
acceleration profile and keeps the overall error at just more than 11 nm
which almost meets the specifications.
This proof of the value of feedback should however not be seen as a reason
to reduce the requirements for the feedforward control. When the accuracy
of the feedforward control action would for instance be reduced with a factor
three that the graph shows that the settle time to reduce the corresponding
higher peak error with a factor three is almost 10 ms.
This example reduction of the feedforward accuracy with a factor three has
the following consequences for the throughput of the machine. An average
300 mm wafer has 100 dies so 100 times acceleration and deceleration.
These 10 ms settle time per scan would mean approximately 1 second loss
from the 20 seconds total exposure time of the machine. This 5 % loss
reduces directly 5 % of the customer value of the wafer scanner!
When the requirements in future become more strict there are in principle
only two ways to reduce the settle time. The first is to increase the control
bandwidth frequency. This however requires further improvements on the
mechanical dynamics that were already assumed to be optimal and it is
not expected that this can be easily improved without increasing cost. The
only other solution that remains is an improved accuracy of the feedforward
control system. The errors in the feedforward model are mainly determined
by changing properties of the actuators due to temperature and in practice
such a reduced error level can only be maintained by continuously calibrating
the total system.
730 Chapter 9. Precision positioning in wafer scanners

9.5.2 The mass dilemma

The above example did not include errors caused by external disturbances.
With all measures taken to its maximum, in practice still the last errors
have to be corrected in less than a just allowable 10 ms settle time and there
the mass dilemma returns in its full magnitude!
A reduced moving mass is preferred in order to reduce the electrical power
that is necessary for the required high levels of acceleration. A reduced
mass will also limit the reaction forces and will indirectly lead to reduced
errors by nonlinearities and dynamic disturbances in the actuators.
Unfortunately a reduced mass will proportionally increase the sensitivity
for external disturbance forces through elastic and damping elements. This
means that an increased feedback control gain would be needed to correct for
these errors with a corresponding higher bandwidth frequency. One might
think that a low mass construction will show higher eigenfrequencies that
enable to achieve this increased bandwidth. Unfortunately this is not a fully
correct assumption as a reduced mass without reduced overall dimensions
can only be realised by a combination of thin structures. Those structures
are not always sufficiently stiff in all directions to guarantee the required
high eigenfrequencies.
Research is done at several places to compensate these potentially harmful
eigenmodes by means of overactuation. Overactuation implies the use of
more actuators than the number of coordinate directions that have to be
controlled. By optimising the placement of the actuators in respect to the
mode shapes of interest, the excitation of the lower-frequency eigenmodes
can be avoided. As soon as these techniques are really applied, the simple
control configuration of Figure 9.24 is no longer possible as the actuation
axes are dynamically well-coupled giving a space matrix with many numbers
beside the diagonal.
9.6. Main design rules for precision 731

9.6 Main design rules for precision

Next to the purpose of pin-pointing the important value of mechatronics for

a major technological development of this era, this chapter was written to
emphasise some important universal design rules. Following these rules
will help to achieve a high precision when designing high-speed production
systems like wafer scanners:

1. Know your problem. Only start designing when the requirements

and specifications are clear.

2. Don’t create your own problem. Potential disturbance sources

should be solved at the source.

3. Know your hardware. The first design goal is to define an extremely

well-balanced hardware concept that offers the best possible determin-
istic dynamic behaviour of all systems combined. Mechatronics is an
integral approach on design.

4. Do not think that software solves your problem. Changes in

hardware are very expensive and time consuming. In most cases soft-
ware is only capable to camouflage flaws in the hardware by introducing
longer waiting times.

5. Feedback only for robustness. Make maximum use of feedforward

control and use feedback only for robustness against real undetermined
error sources.

6. Measuring is knowing. Anything can only be precise when it can

be determined that it is precise. This means that all aspects of the
metrology loop need to be fully mastered.

7. Timing is essential. Be aware of the timing of all elements in the

control loop. The latency in digital measurement systems can be
detrimental for a stable feedback loop.

Especially the rule on reliable knowledge of the hardware in relation to

software is of paramount importance. There is a growing belief especially
with younger people that software will ultimately solve all our problems.
It is indeed true that, although virtually absent in this book, software has
proved to be crucial for the integration and control of complex high-tech
equipment. The flexibility of software in combination with the extreme high
computing speed of computers has tremendous power. It should, however,
732 Chapter 9. Precision positioning in wafer scanners

never be forgotten that a bad mechanical design can never be mended by

software. A loose bolt will stay loose as no force can be transferred by
software.
Of course far more subtle aspects play a role in this perception but it is
safe to state that the ultimate of performance of a complex mechatronic
system will always require the ultimate performance from all parts simul-
taneously. In this chain of performance the hardware elements will remain
the limiting factor for the maximum attainable performance of most high
performance mechatronic systems due to their intrinsic complexity and
physical constraints.
Appendix

Recommended other books

The following books are recommended for further reading, when more in
depth knowledge of the different disciplines within mechatronics is required.
The first two books are a useful addition to the theory on mechanics of
mechatronic systems, because they present mechanisms, transmissions,
bearings and several other aspects of mechanical engineering that are not
presented in this book. The first book is based on the practical experience
within the Engineering departments of Philips Electronics on the dynamics
of mechanisms. Earlier versions were written in Dutch, the original by
professor Wim van der Hoek and refined by professor Rien Koster, both from
the Eindhoven University of Technology. Recently a fully re-written version
is published in English by professor Herman Soemers from the University
of Twente:
Design principles for precision mechanisms
H.M.J.R.Soemers
Publisher: T-Point Print VoF (2010)
ISBN: 978-9036531030

The second book on mechanical design is written by Anton van Beek from
Delft University of technology with an emphasis on reliability, friction
phenomena and bearings:
Advanced Engineering design – lifetime performance and reliability
Anton van Beek
Publisher: Tribos (2009)
ISBN-13: 978-9081040617

733
734 Recommended other books

Regarding motion control, the following four books provide ample infor-
mation that directly connects with the described methodology in Chapter 4,
including the relations with the physical aspects of the controlled plant:
Feedback Systems: An Introduction for Scientists and Engineers
Karl Johan Åström, Richard M. Murray
Publisher: Princeton Univ Press (2008)
ISBN-13: 978-0691135762
Digital Control of Dynamic Systems
Gene Franklin, J. David Powell, Michael L. Workman
Publisher: Addison Wesley (1998)
ISBN-13: 978-0201820546
Computer-Controlled Systems: Theory and Design
Karl Johan Åström, Björn Wittenmark
Publisher: Prentice Hall (1996)
ISBN-13: 978-0133148992
Advanced PID Control
Karl Johan Åström, Tore Hägglund
Publisher: Isa (2005)
ISBN-13: 978-1556179426

For optics the following two books are recognised as being the prime source
of reference on this subject:
Principles of Optics:
Electromagnetic Theory of Propagation, Interference and Diffraction of light
Max Born, Emil Wolf
Cambridge University Press (1999)
ISBN-13: 978-0521639217
Optics
Eugene Hecht
Publisher: Addison Wesley (2002)
ISBN-13: 978-0805385663
Recommended other books 735

On physics in general, including light and waves, the many books from
Richard Feynman are all worthwhile reading. The following two examples
give a nice overview of his ability to teach the difficult material in clear
manner, not only in the classroom where his style of lecturing became famous,
but also by the examples and understandable structure of the material.
The first and most comprehensive work is based on the “Feynman lectures
on physics”, that were given to undergraduate students at the California
Institute of Technology. The textbook was originally written by the staff of
Caltech in 1964 based on recordings of the lectures and is now available in
a newly edited full set:
The Feynman Lectures on Physics, boxed set: The New Millenium Edi-
tion
Richard P. Feynman, Robert B. Leighton, Matthew Sands
Publisher: Basic Books (2011)
ISBN-13: 978-0465023820
The second example is a much smaller book that was written to explain the
principles of the theory on quantum-electrodynamics to people from outside
the particle-physics community. This very readable and even sometimes
amusing booklet is referred to, when explaining photons:
QED: The Strange Theory of Light and Matter
Richard P. Feynman, A. Zee (intro)
Publisher: Princeton University Press (2006)
ISBN-13: 978-0691125756

For measurement, the following book has often been used in our lectures
at the university:
Principles of Measurement Systems
John P. Bentley
Publisher: Prentice Hall (2005)
ISBN-13: 978-0130430281
736 Recommended other books

When more knowledge is needed on the physical theory and application

of electromechanics, dealing with all electromagnetic forces that work
on objects, the following books gives these relations with examples from
engineering. The first book has more emphasis on physics while the sec-
ond and third book are more focused on the application in a mechatronic
environment including power amplifiers.
Continuum Electromechanics
James R. Melcher
Publisher: The MIT Press (1981)
ISBN-13: 978-0262131650
Electromechanical Systems and Devices
Sergey Edward Lyshevski
Publisher: CRC Press (2008)
ISBN-13: 978-1420069723
Electric machinery
A.E. Fitzgerald, Charles Kingsley Jr., Stephen Umans
Publisher: McGraw-Hill (2002)
ISBN-13: 978-0073660097

For electronics the following recognised reference covers both analogue

and digital electronic circuit design:
The art of electronics
Paul Horowitz, Winfried Hill
Publisher: Cambridge University Press (1989)
ISBN-13: 978-0521370950
Nomenclature 737

Nomenclature and abbreviations

Nomenclature

a [m/s2 ] Acceleration.
A [-] Amplitude.
A [m2 ] Surface of Cross Section (SCS).
Ac [m2 ] SCS of a coil.
A c,w [m2 ] SCS of the windings in a coil.
Ag [m2 ] SCS of an air gap.
Ai [m2 ] SCS of the flux path inside a coil.
Am [m2 ] SCS of a permanent magnet.
Ao [m2 ] SCS of the flux path outside a coil.
Ay [m2 ] SCS of the flux path inside a yoke.
b [-] Wien’s constant.
B [T] Magnetic field.
B [T] Magnetic flux density (MFD).
Bg [T] MFD in the air gap.
Bg,c [T] MFD in the air gap induced by current.
Bg,m [T] MFD in the air gap induced by magnet.
Br [T] Remnant MFD.
Bs [T] Saturation MFD.
Bw [T] MFD inside a coil winding.
By [T] MFD inside a ferromagnetic yoke.
c [-] Dimensionless constant or factor.
c [m/s] Velocity of light.
c [Ns/m] Damping coefficient.
cd [V/m] Deformation constant piezo accelerometer.
2
ci [Is /m] Current sensitivity piezo accelerometer.
cs [-] Sensitivity factor in Wheatstone bridge.
2
cv [Vs /m] Voltage sensitivity piezo accelerometer.
C [F] Capacitor value.
C, C [m/N] Compliance, Transfer Function (TF).
C fb ( s) [-] TF of a feedback controller.
C pd ( s) [-] TF of a PD-type feedback controller.
C pid ( s) [-] TF of a PID-type feedback controller.
C ff ( s) [-] TF of a feedforward controller.
Cm [m/N] Mass compliance.
Cs [m/N] Spring compliance.
738 Nomenclature

Ct [m/N] System compliance.

Cd [m/N] Damper compliance.
d [m/V] Piezoelectric coefficient matrix.
D, D [C/m2 ][N/Vm] Electric flux density or displacement.
E, E [V/m], [N/C] Electric field.
Ef [V/m], [N/C] Induced electric field by change of flux.
E [J] Energy.
EC [J] Energy contained in a capacitor.
EL [J] Energy contained in an inductor.
f [Hz] Spatial or temporal frequency.
f [m] Focal length.
f ( f ), F ( f ) [-] TF in the temporal frequency domain.
f ( s ), F ( s ) [-] TF in the Laplace domain.
f ( t), F ( t) [-] TF in the time domain.
f (ω), F (ω) [-] TF in the angular frequency domain.
f0 [Hz] Temporal natural or corner frequency.
fa [Hz] Anti-resonance frequency of decoupling body.
fc [Hz] Unity-gain cross-over frequency.
fc [Hz] Centre frequency of dual laser beam.
fd [Hz] Doppler shift frequency.
fd [Hz] Differentiating corner frequency of D-control.
fi [Hz] Integrating corner frequency of I-control.
fm [Hz] Measured frequency.
fN [Hz] Nyquist frequency limit of sampling.
fp [Hz] Frequency of a photon.
fs [Hz] Switching frequency of SMPA.
fs [Hz] Split frequency of dual laser beam.
ft [Hz] Taming corner frequency of D-control.
F [N] Force.
F̂ [N] Amplitude of force.
Fa [N] Force of actuator.
Fd [N] Force due to deformation or displacement.
Fc [N] Control force.
Fcom [N] Force at centre of mass.
F ( s) [-] Transfer function of a filter.
Fe [V] Electromotive force.
Fm [A] Magnetomotive force.
G [m2 sr] Etendue.
G [-] Gain of amplifier.
Ga [-] Closed-loop amplifier TF (gain).
Nomenclature 739

Gf [-] TF of operational amplifier feedback loop.

Go [-] Open-loop TF of operational amplifier.
Gv [-] Voltage gain of transistor.
G ( s) [-] Transfer function of the plant.
G t,ff ( s) [-] Total TF of plant with feedforward controller.
h [-] Planck’s constant.
hc [m] Winding height of a coil.
h c,g [m] Overlap winding height of a coil and air gap.
hg [m] Height of air gap in the motion direction.
H [A/m] Magnetizing field.
H [A/m] Magnetic field strength.
Hc [A/m] Coercitive magnetic field strength.
Hw [A/m] Magnetic field strength inside a coil winding.
I [A] Electric current.
I [W/m2 ] Intensity.
Î [A] Amplitude of alternating electric current.
Ib [A] Base current of a transistor.
Ic [A] Collector current of a transistor.
Id [A] Diode current.
Ie [A] Emitter current of a transistor.
Ig [A] Gate current of a MOSFET.
Ii [A] Input current of an electronic circuit.
Il [A] Current in a load.
Im [A] Induced current by movement of actuator.
IN [A] Norton Equivalent current.
Io [A] Output current of an electronic circuit.
Ir [W/m2 ] Irradiance.
I r,c [W/m2 ] Combined Irradiance.
I r,i [W/m2 ] Irradiance of incident radiation.
I r,r [W/m2 ] Irradiance of reflected radiation.
Is [A] Electric current caused by the source.
It [A] Total parallel current in an electric circuit.
J [A/m2 ] Electric current density.
k [N/m] Stiffness.
ka [N/m] Stiffness of anti-resonator.
kd [-] Differentiating gain factor of D-control.
kd [-] Differentiating gain factor of D-control.
kf [-] Stiffness of flexure spring.
ki [-] Integrating gain factor of I-control.
kn [N/m] Negative stiffness of plant.
740 Nomenclature

kp [-] Proportional gain factor of P-control.

k pz [-] Stiffness of piezo-actuator.
kt [-] Total gain over a series of elements.
Ki [-] Interfering input gain.
Km [-] Modifying input gain.
 [m] Length.
g [m] Length of the air gap.
i [m] Length of the flux path inside a coil.
m [m] Length of the magnet.
o [m] Length of the flux path outside a coil.
w [m] Wire length of a winding.
w,t [m] Total wire length of a coil in PM field.
w,a [m] Active part of the wire in a coil.
w,p [m] Passive part of the wire in a coil.
y [m] Length of the flux path inside a yoke.
L [H] Self inductance.
La [H] Self inductance of actuator.
L [W/(m2 sr] Radiance of a light source.
L eff [W/(m2 sr] Effective radiance of a light source.
m [kg] Mass.
ma [kg] Mass of anti-resonator.
mm [kg] Mass of mover.
M [-] Magnification.
M [W/m2 ] Radiant emittance.
M ( s) [-] Transfer function of measurement system.
n [-] Number of, order of TF (integer).
n [-] Index of refraction.
N [-] Order number or interferometer constant.
2
NE [V /Hz] Excess, flicker or 1/ f noise .
2
NS [A /Hz] Shot noise.
2
NT [V /Hz] Thermal noise.
n̂ [-] Normal vector on defined surface.
p [-] Probability.
P [Pa] Pressure.
P [W] Power, Intensity.
PC [W] Power needed for a voltage over a capacitor.
Pe [W] Electrical Power.
Pi [W] Illumination Power of a light source.
Pi,c [W] Total captured illumination power.
Pi,t [W] Total radiated illumination power.
Nomenclature 741

Pm [W] Mechanical Power.

Pl [W] Dissipated Power loss.
Ps [-] Signal power, squared amplitude.
PL [W] Power to create a current in an inductor.
Pv [Pa] Vapour pressure.
Q [-] Quality factor.
Qm [-] Figure of merit.
q [C] Electric charge.
qe [C] Charge of an electron.
R [Ω ] Resistance.
Ra [Ω ] Resistance of the actuator.
Rc [Ω ] Characteristic impedance of a cable.
Rs [Ω ] Output resistance of the source.
R s,c [Ω ] Output resistance of the collector source.
R s,e [Ω ] Output resistance of the emitter source.
R x,y,z [rad] Rotation around x, y, z axis.
ℜ [A/Wb] Magnetic reluctance.
ℜg [A/Wb] Magnetic reluctance of the air gap.
ℜi [A/Wb] Magnetic reluctance inside a coil.
ℜm [A/Wb] Magnetic reluctance of the permanent magnet.
ℜo [A/Wb] Magnetic reluctance outside a coil.
ℜt [A/Wb] Total magnetic reluctance of the flux path.
ℜy [A/Wb] Magnetic reluctance of the ferromagnetic yoke.
s [-] Laplace variable.
S, S [-] Mechanical strain.
t [s] Time.
T [◦ C] Temperature.
T, T [N/m2 ] Mechanical stress.
Tc [◦ C] Curie temperature.
T [s] Period.
v [m/s] Velocity.
vp [m/s] Wave propagation velocity.
vm [m/s] Motion velocity.
V [m3 ] Volume.
V [V] Potential difference, voltage.
V̂ [V] Amplitude of alternating voltage.
Va [V] Voltage over an actuator.
Vb [V] Voltage at the bridge between two switches.
Vc [V] Voltage of carrier frequency.
Vb−e [V] Transistor base-emitter voltage.
742 Nomenclature

Vc−e [V] Transistor collector-emitter voltage.

Vcs [V] Current sensing voltage.
Vc,a [m3 ] Active volume of a coil in a magnetic field.
Vd [V] Voltage over a diode.
Vg [V] Voltage of a geophone.
VC [V] Voltage by charge displacement in capacitor.
VL [V] Voltage by current change in inductor.
Vi [V] Input voltage of an electronic circuit.
Vm [V] Voltage by the movement of an actuator.
Vo [V] Output voltage of an electronic circuit.
Vp [V] Power supply voltage.
VR [V] Induced voltage by a current in a resistor.
Vs [V] Voltage from the source.
Vt [V] Total series voltage in an electronic circuit.
Vth [V] Base-emitter threshold voltage of transistor.
VTh [V] Thevenin equivalent voltage.
x [m] Position, distance.
xd [m] Displacement.
x̂ [m] Amplitude of alternating position.
Z [-] Impedance.
β [-] Current amplification ratio of transistor.
γ [-] Effective fill factor of coil windings.
 [-] Control error.
0 [As/Vm] Electric Permittivity (EP) in vacuum.
r [-] Relative EP, dielectric constant.
ζ [-] Damping ratio.
ϑ [rad or ◦ ] (Capture) angle.
ϑi [rad or ◦ ] Angle of incidence.
ϑr [rad or ◦ ] Angle of reflection.
ϑt [rad or ◦ ] Angle of refraction.
λ [-] Loss factor for magnetic stray flux.
λ [m] Wavelength.
λB [m] Bragg grating wavelength.
μ0 [Vs/Am] Magnetic permeability of vacuum.
μr [-] Relative magnetic permeability.
ϕ [rad or ◦ ] Phase angle.
ϕm [rad or ◦ ] Phase angle due to movement.
Φ [Wb] Magnetic flux.
Φe [C] Electric charge flux.
Φm [Wb] Magnetic flux of magnet.
Nomenclature 743

Φw [Wb] Magnetic flux inside coil winding.

Φw,t [Wb] Total magnetic flux over all windings.
Φy [Wb] Magnetic flux inside a yoke.
ρr [Ωm] Resistivity.
ρq [C/m3 ] Electric charge density.
σ [1/m] Wave number.
σx [nb] Standard deviation of variable x.
τ [s] Time constant, RC time.
τd [s] Differentiating time constant (D-control).
τe [s] Electrical time constant of actuator.
τi [s] Integrating time constant (I-control).
τt [s] Taming time constant (D-control).
ω [rad/s] Angular frequency.
ω0 [rad/s] Angular natural or corner frequency.
ωc [rad/s] Angular unity-gain cross-over frequency.
ωd [rad/s] Differentiating corner frequency of D-control.
ωi [rad/s] Integrating corner frequency of I-control.
ωt [rad/s] Taming corner frequency of D-control.
Ω [sr] Solid angle.
Ω [V/A] Resistance.
744 Abbreviations

Abbreviations

ADC Analogue-to-Digital Converter

AFM Atomic Force Microscopy
AO Adaptive Optics
BIPM Bureau International des Poids et Mesures
BWO Beam Walk Off
CAD Computer Aided Design
CAS Cumulative Amplitude Spectrum
CD Critical Dimension
CCD Charge Coupled Device (Camera sensor)
CMMR Common Mode Rejection Ratio
CMP Chemical Mechanical Polishing
CPF Cumulative Probability Function
CPS Cumulative Power Spectrum
DAC Digital-to-Analogue Converter
DEB Dynamic Error Budgeting
DFT Discrete Fourier Transform
DOF Depth Of Focus
EMC Electro-Magnetic Compatibility
EP Electric permittivity
ESD Electrostatic Discharge
ESO European Southern Observatory
FBG Fibre Bragg Grating
FEM Finite Element Method
FET Field Effect Transistor
FFT Fast Fourier Transform
FRF Frequency Response Function
FWHM Full Width Half Maximum
GB Gain Balancing
GS Gain Scheduling
GUM Guide to the expression of Uncertainty in Measurement
HF High Frequency
HP High Pass (filter)
IC Integrated Circuit
IGBT Insulated Gate Bipolar Transistor
JCGM Joint Committee for Guides in Metrology
LF Low Frequency
LP Low Pass (filter)
Abbreviations 745

LSB Least Significant Bit

LVDT Linear Variable Differential Transformer
MA Moving Average
MEMS Micro Electro Mechanical System
MFD Magnetic Flux Density
MIMO Multiple Input Multiple Output (control)
MOS Metal Oxide Semiconductor (FET)
MSB Most Significant Bit
MSD Moving Standard Deviation
NA Numerical Aperture
OPD Optical Path Difference
OPL Optical Path Length
PALM Piezoelectric Active Lens Mount
PCB Printed Circuit Board
PDA Personal Digital Assistant
PDF Probability Density Function
PLL Phase Locked Loop
PLZT Lead Lanthanum Zirconate Titanate (Piezo)
PM Permanent Magnet
PSD Power Spectral Density
PSD Position Sensitive Detector
PSF Point Spread Function
PSRR Power Supply Rejection Ratio
PVDF Polyvinylidene Fluoride (Piezo)
PTFE Polytetrafluoroethylene, Teflon
PZT Lead Zirconate Titanate (Piezo)
PWM Pulse Width Modulation
RMS Root Mean Square
SCS Surface Cross Section
SISO Single Input Single Output (control)
SMPA Switched Mode Power Amplifier
SNR Signal to Noise Ratio
SPM Scanning Probe Microscopy
STM Scanning Tunnelling Microscopy
STP Shielded Twisted Pair
TDM Time Division Multiplexing
TF Transfer Function
UTP Unshielded Twisted Pair
VCO Voltage Controlled Oscillator
VGM Vector Gain Margin
746 Index

WDM Wavelength Division Multiplexing

Index
RC -time, 336 bandwidth, 91, 93, 114, 171
“Hooke – Newton” law, 87 barrel, 486
0 dB level, 83 base frame, 695
batteries, 44
Abbe error, 685
BBN criteria, 552
absolute standards, 540
beam splitter, 533
AC, 49
beam walkoff, 682
acceleration, 95
bias, 402
achromatic doublet, 488
biasing, 402
adaptive feedforward control, 155, 208
bimorph, 630
adaptive optics, 525
birefringence, 499
aging, 303
bistable, 579
air foot, 695
bits, 580
air mounts, 695
bitstream, 594
air showers, 678, 724 black body, 452
air-gap, 235 black-box system identification, 205
amplitude grating, 506 Bode Sensitivity Integral, 181
amplitude modulation, 575 Bode-plot, 79
analogue-to-digital converters, 587 body, 95
angular frequency, 49 branches, 560
annular illumination, 521 Brewster’s angle, 499
anti-resonance, 121 bridge, 428
anti-resonator, 703 bridge-rectifier, 356
anti-windup control, 212 brightness, 461
aperiodic, 110 Butterworth, 394
aperture plane, 485
carrier, 575
aperture stop, 492
catadioptric, 473
asphericity, 482
catoptric, 473, 688
assist features, 689
cats-eye, 660
Astigmatism, 481
cavity, 504
asymptotically stable, 175
centre-frequency, 663
at equal amplitude, 76
channel, 360
attenuation-band, 335
characteristic impedance, 325, 597
autotransformer, 291
characteristic polynomial, 199
azimuthal order, 529
charge carriers, 353
balance mass, 695 charge control, 311
balanced bridge, 560 charge-pumping, 432, 434
Index 747

Chebyshev, 394 Cumulative Probability Function, 544

chief ray, 493 Curie temperature, 296
chip, 351 current source, 44
chromatic aberration, 485 cut-off frequency, 335
chromatic dispersion, 487 cylinder, 483
circular polarised, 500
damper, 95
clipping, 209, 363
damper-line, 97
closed-loop, 146
Darlington pairs, 411
closed-loop feedback accelerometer,
DC, 49
629
de-polarisation, 311
coaxial cable, 567
deci-Bel, 81
coercive force, 233
decoupling, 120
cogging, 711
Deflectometer, 603
coherence, 456
Delta-Sigma converter, 588
coherence length, 459
demagnetisation graph, 233
coherence time, 458 departure, 482
collector follower, 365 depletion layer, 354
Coma, 481 depletion-mode, 360
common collector, 362 depth of field, 522
common emitter, 365 depth of focus, 522, 720
common-mode rejection ratio, 370, destructive interference, 502
404 diamagnetic, 228
common-mode signal, 367 dichroic coating, 505
Compact Disc player, 6 die, 688
complementary sensitivity function, dielectric constant, 329
169 differential-mode, 368
completer and finisher, 26 diffraction, 450
compliance, 86, 95 diffraction limited, 518
constructive interference, 502 digital-to-analogue converter, 590
controllability, 131, 205 dioptre, 473
converse piezoelectric effect, 296 dioptric, 473
corkscrew, 244 direct band-gap, 455
corner-frequency, 335 direct piezoelectric effect, 296
cosine error, 676 discrete components, 350, 409
creep, 301 discrete Fourier transform, 71
critical angle, 469 displacement, 95
Critical Dimension, 521, 687 distortion, 485
cross-coupling, 195 divergence, 219
cross-over distortion, 364 Doppler shift, 667
Crown glass, 488 Dose-control, 522
cube-corner retro-reflector, 660 double logarithmic, 80
Cumulative Amplitude Spectrum, 551 double-telecentricity, 493
Cumulative Power Spectrum, 548 Drain, 359
748 Index

drift, 402 finesse, 505

dual-ended, 435 first-order, 108
dual-slope ADC, 587 flare, 481
Dutch school of mechatronics, 20 flexure-scanner, 309
duty-cycle, 424 flicker noise, 554
Dynamic Error Budgeting, 52, 544 Flint glass, 488
floating, 382, 432
eddy-current, 258
floor vibrations, 548
eigendynamics, 109
focal plane, 485, 710
eigenfrequencies, 121
force frame, 696
eigenmodes, 118, 125
four quadrant, 417
eigenvalues, 109, 125, 199
four-quadrant detector, 601
electric load, 324
Fourier transform, 65
electric motor, 217
fourth-order, 125
electric permittivity, 329
free electrons, 353
electric permittivity in vacuum, 39
frequency combs, 671
electric signals, 48
frequency domain, 52, 72
Electro Magnetic Compatibility, 409
frequency response function, 101
electrodes, 43
frequency spectrum, 52
electromagnetic actuator, 217
frequency-to-digital conversion, 579
Electromotive Force, 44
fringes, 659
electronically commutated, 251
Full-width at half-maximum, 457
electrostatic force, 39
elliptical polarised, 500 Gain Balancing, 727
emitter follower, 362 gain margin, 171
enhancement-mode, 360 Gain Scheduling, 727
entry-pupil, 492 gain-bandwidth product, 401
etendue, 464 gang of four, 169
exit-pupil, 492 gang of six, 168
Gate, 359
f-number, 519
Gaussian distribution, 545
Fabry-Perot interferometer, 504
geometric optics, 450
fading, 693
geophone, 623
far field, 507
Gray code, 582
Faraday shield, 569
grey-box system identification, 205
Fast Fourier Transform, 71
grid plate, 726
features, 687
ground loop, 569
feedback, 139
group delay, 395
feedforward, 139
Fibre Bragg Grating, 617 H-bridge, 435
fiducials, 718 half-pitch, 688
field curvature, 485 Hall effect, 444
field lines, 39 hard-switching, 438
field plane, 485 Hartmann sensor, 533
figure of merit, 254 Heisenberg uncertainty principle, 541
Index 749

heterodyne interferometry, 662 Laplace domain, 73

holes, 353 Laplace plane, 106
homo-polar, 277 laser, 456
homodyne detection, 578 latch, 582
homodyne interferometer, 657 LCR-filter, 340
hot electrons, 358 lead-lag compensation, 163
hydraulic linear motors, 13 lead-network, 162
Hysteresis, 302 Least Significant Bit, 581
hysteresis operators, 302 lensmakers equation, 478
level-shifter, 411
I-control, 165, 174
levelling, 720
idle current, 362
limit-cycling, 212, 411
illumination power, 460
linear polarised, 499
image-space telecentric, 495
Lissajous plot, 646
impedance, 54
long tailed pair, 369
in Abbe, 686
long-lead items, 32
in parallel, 272
long-stroke, 19, 711
in series, 274
loop shaping design, 170
independent, 126
Lorentz force, 222
indirect band-gap, 455
low-pass filter, 189
inductor, 260
inertial velocity sensor, 623 machining, 19
innovator and creator, 26 magnetic bearing, 141
input-shaping, 152 magnetic interference, 566
instrumentation amplifier, 573 magnetomotive force, 226
Insulated Gate Bipolar Transistor, majority carriers, 358
422 mass dilemma, 715
Integrated Circuits, 9 measurand, 539
integrator windup, 212 measurement accuracy, 540
intensity, 59 measurement error, 539
interference, 450, 501 measurement precision, 540
interfering input, 542 measurement resolution, 541
interferometer, 503 measurement uncertainty, 540
interferometer constant, 658 Mechanical amplification, 308
intermediate image, 490 meridional plane, 484
inverse piezoelectric effect, 296 metrology, 537
iris diaphragm, 492 metrology frame, 684, 695
irradiance, 60 metrology loop, 685, 716
iterative learning control, 155 Miller capacitor, 397
minimal realisation, 205
jerk, 279
minority carriers, 358
k-one factor, 521 mirage, 526
Kalman-filter, 205 modal analysis, 129
Kalman-gain, 207 mode-shapes, 125
750 Index

model-based, 149 parasitic, 327

modified Edlén, 676 paraxial rays, 475
modifying input, 542 pass-band, 335
modulation, 575 PD-control, 156, 161
Moore’s law, 691 pedestal, 695
Most Significant Bit, 581 pentafoil, 531
mover, 265 periodic error, 673
multi-lens array, 533 Perovskite structure, 296
multi-sines, 78 phase, 50
Multiple Input Multiple Output, 195 phase grating, 508
N-material, 353 phase margin, 171
negative stiffness, 140, 270 phase selective detection, 578
noise, 49 photons, 451
normal dispersion, 487 physical optics, 450
normal distribution, 545 PID-control, 156, 165
notch filter, 151 piezo-gain, 300
NTC, 558 Piezo-stack actuators, 306
Numerical Aperture, 519 Piezoelectric Active Lens Mount, 702
Nyquist frequency, 586 piezoelectric scanner, 309
Nyquist plot, 83 Piezoelectric tube actuators, 305
object-space telecentric, 495 piezoresistivity, 619
observability, 131, 205 Pincushion, 486
observer, 205 pink noise, 52
off-the-shelf, 33 plant, 138
offset, 401 platforms, 30
one-over-f (1/ f ) noise, 554 point source, 461
open-loop, 81, 140, 159 point-symmetric, 661
open-loop control, 143 polarisability, 329
Operational Amplifiers, 361 polarisation, 57, 296, 497
optical axis, 474 polarisation mixing, 673
optical flats, 675 polariser, 499
Optical path length, 467 polarising beam splitter, 664
optical pick-up unit, 5 pole pieces, 235
optical throughput, 464 pole-splitting, 400
optimal control, 192 pole-zero cancellation, 144
order and delivery, 33 poles, 106, 139
over-constrained, 88 poles and zeros, 73
over-hung, 250 potential difference, 41
overactuation, 730 potentiometer, 323
overlay, 691 Power Spectral Density, 548
P-control, 159 power supply rejection ratio, 404
parallel-resonant, 341 Pre-loading, 307
paramagnetic, 228 Precision Engineering, 19
Index 751

Preisach model, 302 requirement budgeting, 28

primary windings, 292 Resistivity, 38
printed circuit board, 328 resonance, 64
prism, 530 resonant-mode, 438
Probability Density Function, 544 reticle, 10, 522
product creation process, 26 reticle stage, 689
product manager, 27 reverse-recovery time, 439
propagation, 57 right hand, 244
proportional, 158 road mapping, 30
proportional control, 159 robustness, 147
Proportional Differential control, 161 roll-off, 99
proximity detector, 599 roll-off frequency, 93, 165
pulse-width modulation, 425 root of the sum of squares, 547
pupil, 491 rotation, 219
pupil plane, 485, 494
purple fringing, 487 safety ground, 570
Push Pull class AB, 364 sagittal plane, 484
Push-Pull class B, 364 sagittal rays, 484
Sallen-Key, 391
quad-cell, 601 sample-and-hold, 584
quadrafoil, 531 samples, 583
quadrature detection, 669 saturation, 209
quality factor, 106 scanner bow, 309
quantisation error, 581 Schmitt trigger, 372, 425
quarter wave plate, 663 Schottky diodes, 439
R-2R ladder-network, 590 second-order, 102
race track coil, 248 secondary breakdown, 422
radial order, 529 secondary trefoil, 531
radiance, 456, 461 secondary windings, 292
radiant emittance, 461 seismic mass, 624
radiant flux, 460 selective amplifier, 576
radiant intensity, 60 self-inductance, 260
random errors, 540 sensitivity function, 169
ray tracing, 479 serial scanner, 309
RC-filter, 335 series-resonant, 344
reactive, 56, 330 servo-system, 6, 156
reactive impedance, 328 servo-valves, 13
real value, 539 settle time, 728
recombination, 354 Shack-Hartmann wavefront sensor,
reflection, 61, 449 533
refraction, 449 shadow mask, 12
Reluctance, 226 Shielded Twisted Pair, 570
reluctance force, 257 short-stroke, 711
remnant flux density, 233 side-bands, 577
752 Index

Silicon Repeater, 2, 9 synchronous demodulation, 576

simulated, 166 System Identification, 204
Single Input Single Output, 143 systematic effects, 52
single-ended, 428 systematic errors, 540
single-ended Class A, 363 systems engineer, 27
single-phase, 437 tacho-generator, 213
skyhook, 699 tamed PD-control, 163
slew-rate, 209, 401 tangential plane, 483
slit, 689 tangential rays, 484
smart-disk, 702 temporal coherence, 458
snap, 282 temporal frequency, 49
solenoid, 333 temporal or spatial variability, 48
solid-state lamps, 455 terminals, 43
Source, 359 thermal-centre, 608
space, 195 thin-lens equation, 478
spatial coherence, 458 three-phase, 251
spatial frequency, 516 throughput, 692
specific peak force, 714 Time Division Multiplexing, 622
speckle, 457 time domain, 52, 72
Spectrum Analyser, 71 total quality, 32
Spherical Aberration, 481 trace, 479
split-frequency, 663 traceability, 537, 717
spring, 95 traceable, 540
spring-line, 96 Trait é de la lumi ère, 505
stability, 140 transconductance, 381, 413
stable, 140 transfer function, 101
standard deviation, 545 transformer core, 291
star configuration, 446 transimpedance, 380
state feedback, 202 transmissibility, 86, 95
state-space, 137, 156, 195 transmission-line, 596
trefoil, 531
state-variables, 195
triangulation, 603
stator, 265
tripod-scanner, 309
stimulus, 74
tuned-mass damper, 703
stops, 519
two degree of freedom control, 214
strain gage, 558
stray flux, 237 ultimately, 112
Strehl ratio, 519 unambiguously measurable, 24
successive-approximation, 588 unbound, 218
super-capacitors, 332 under-hung, 251
super-spring, 201 unity-gain cross-over frequency, 83,
Surface Mount Devices, 328 158
switched-mode power amplifiers, 418 Unshielded Twisted Pair, 566
switched-mode power supply, 295 V-model of systems engineering, 23
Index 753

vector margin, 171 wafer stage, 13, 689

velocity, 95 wafer table, 689, 726
Vernier, 648, 719 waterbed effect, 180
vias, 691 wave-guide, 596
vibration isolation, 115 wavelength, 57
vibration isolation system, 548, 695 Wavelength Division Multiplexing,
Video Long Play Disk, 2 621
vignetting, 496 wavelength tracker, 677
virgin curve, 232 Weiss domains, 231, 297
virtual, 477 Wheatstone bridge, 559
virtual ground, 377 white noise, 52, 78, 553
voltage controlled, 311
Zernike polynomials, 528
voltage source, 43
zero-stiffness, 19, 705
wafer chuck, 689 zeros, 139
wafer fab, 695 zoom lens, 490
This page intentionally left blank