Title Details

Danilo Mandic, Department of Electrical and Electronic Engineering, Imperial College London, London
Dr Mandic is currently a Reader in Signal Processing at Imperial College, London. He is an experienced author, having written the book Recurrent Neural Networks for Prediction: Learning Algorithms, Architectures and Stability (Wiley, 2001), and more than 150 published journal and conference papers on signal and image processing. His research interests include nonlinear adaptive signal processing, multimodal signal processing and nonlinear dynamics, and he is an Associate Editor for the journals IEEE Transactions on Circuits and Systems and the International Journal of Mathematical Modelling and Algorithms. Dr Mandic is also on the IEEE Technical Committee on Machine Learning for Signal Processing, and he has produced award winning papers and products resulting from his collaboration with industry.

Su-Lee Goh, Royal Dutch Shell plc, Holland
Dr Goh is currently working as a Reservoir Imaging Geophysicist at Shell in Holland. Her research interests include nonlinear signal processing, adaptive filters, complex-valued analysis, and imaging and forecasting. She received her PhD in nonlinear adaptive signal processing from Imperial College, London and is a member of the IEEE and the Society of Exploration Geophysicists.

Preface xiii

Acknowledgements xvii

1 The Magic of Complex Numbers 1

1.1 History of Complex Numbers 2

1.2 History of Mathematical Notation 8

1.3 Development of Complex Valued Adaptive Signal Processing 9

2 Why Signal Processing in the Complex Domain? 13

2.1 Some Examples of Complex Valued Signal Processing 13

2.2 Modelling in C is Not Only Convenient But Also Natural 19

2.3 Why Complex Modelling of Real Valued Processes? 20

2.4 Exploiting the Phase Information 23

2.5 Other Applications of Complex Domain Processing of Real Valued Signals 26

2.6 Additional Benefits of Complex Domain Processing 29

3 Adaptive Filtering Architectures 33

3.1 Linear and Nonlinear Stochastic Models 34

3.2 Linear and Nonlinear Adaptive Filtering Architectures 35

3.3 State Space Representation and Canonical Forms 39

4 Complex Nonlinear Activation Functions 43

4.1 Properties of Complex Functions 43

4.2 Universal Function Approximation 46

4.3 Nonlinear Activation Functions for Complex Neural Networks 48

4.4 Generalised Splitting Activation Functions (GSAF) 53

4.5 Summary: Choice of the Complex Activation Function 54

5 Elements of CR Calculus 55

5.1 Continuous Complex Functions 56

5.2 The Cauchy–Riemann Equations 56

5.3 Generalised Derivatives of Functions of Complex Variable 57

5.4 CR-derivatives of Cost Functions 62

6 Complex Valued Adaptive Filters 69

6.1 Adaptive Filtering Configurations 70

6.2 The Complex Least Mean Square Algorithm 73

6.3 Nonlinear Feedforward Complex Adaptive Filters 80

6.4 Normalisation of Learning Algorithms 85

6.5 Performance of Feedforward Nonlinear Adaptive Filters 87

6.6 Summary: Choice of a Nonlinear Adaptive Filter 89

7 Adaptive Filters with Feedback 91

7.1 Training of IIR Adaptive Filters 92

7.2 Nonlinear Adaptive IIR Filters: Recurrent Perceptron 97

7.3 Training of Recurrent Neural Networks 99

7.4 Simulation Examples 102

8 Filters with an Adaptive Stepsize 107

8.1 Benveniste Type Variable Stepsize Algorithms 108

8.2 Complex Valued GNGD Algorithms 110

8.3 Simulation Examples 113

9 Filters with an Adaptive Amplitude of Nonlinearity 119

9.1 Dynamical Range Reduction 119

9.2 FIR Adaptive Filters with an Adaptive Nonlinearity 121

9.3 Recurrent Neural Networks with Trainable Amplitude of Activation Functions 122

9.4 Simulation Results 124

10 Data-reusing Algorithms for Complex Valued Adaptive Filters 129

10.1 The Data-reusing Complex Valued Least Mean Square (DRCLMS) Algorithm 129

10.2 Data-reusing Complex Nonlinear Adaptive Filters 131

10.3 Data-reusing Algorithms for Complex RNNs 134

11 Complex Mappings and M¨obius Transformations 137

11.1 Matrix Representation of a Complex Number 137

11.2 The M¨obius Transformation 140

11.3 Activation Functions and M¨obius Transformations 142

11.4 All-pass Systems as M¨obius Transformations 146

11.5 Fractional Delay Filters 147

12 Augmented Complex Statistics 151

12.1 Complex Random Variables (CRV) 152

12.2 Complex Circular Random Variables 158

12.3 Complex Signals 159

12.4 Second-order Characterisation of Complex Signals 161

13 Widely Linear Estimation and Augmented CLMS (ACLMS) 169

13.1 Minimum Mean Square Error (MMSE) Estimation in C 169

13.2 Complex White Noise 172

13.3 Autoregressive Modelling in C 173

13.4 The Augmented Complex LMS (ACLMS) Algorithm 175

13.5 Adaptive Prediction Based on ACLMS 178

14 Duality Between Complex Valued and Real Valued Filters 183

14.1 A Dual Channel Real Valued Adaptive Filter 184

14.2 Duality Between Real and Complex Valued Filters 186

14.3 Simulations 188

15 Widely Linear Filters with Feedback 191

15.1 The Widely Linear ARMA (WL-ARMA) Model 192

15.2 Widely Linear Adaptive Filters with Feedback 192

15.4 The Augmented Kalman Filter Algorithm for RNNs 198

15.5 Augmented Complex Unscented Kalman Filter (ACUKF) 200

15.6 Simulation Examples 203

16 Collaborative Adaptive Filtering 207

16.1 Parametric Signal Modality Characterisation 207

16.2 Standard Hybrid Filtering in R 209

16.3 Tracking the Linear/Nonlinear Nature of Complex Valued Signals 210

16.4 Split vs Fully Complex Signal Natures 214

16.5 Online Assessment of the Nature of Wind Signal 216

16.6 Collaborative Filters for General Complex Signals 217

17 Adaptive Filtering Based on EMD 221

17.1 The Empirical Mode Decomposition Algorithm 222

17.2 Complex Extensions of Empirical Mode Decomposition 226

17.3 Addressing the Problem of Uniqueness 230

17.4 Applications of Complex Extensions of EMD 230

18 Validation of Complex Representations – Is This Worthwhile? 233

18.1 Signal Modality Characterisation in R 234

18.2 Testing for the Validity of Complex Representation 239

18.3 Quantifying Benefits of Complex Valued Representation 243

Appendix A: Some Distinctive Properties of Calculus in C 245

Appendix B: Liouville's Theorem 251

Appendix C: Hypercomplex and Clifford Algebras 253

Appendix D: Real Valued Activation Functions 257

Appendix E: Elementary Transcendental Functions (ETF) 259

Appendix F: The O Notation and Standard Vector and Matrix Differentiation 263

Appendix G: Notions From Learning Theory 265

Appendix H: Notions from Approximation Theory 269

Appendix I: Terminology Used in the Field of Neural Networks 273

Appendix J: Complex Valued Pipelined Recurrent Neural Network (CPRNN) 275

Appendix K: Gradient Adaptive Step Size (GASS) Algorithms in R 279

Appendix L: Derivation of Partial Derivatives from Chapter 8 283

Appendix M: A Posteriori Learning 287

Appendix N: Notions from Stability Theory 291

Appendix O: Linear Relaxation 293

Appendix P: Contraction Mappings, Fixed Point Iteration and Fractals 299

References 309

Index 321