1st Edition

Unsupervised Signal Processing Channel Equalization and Source Separation

    340 Pages 97 B/W Illustrations
    by CRC Press

    Unsupervised Signal Processing: Channel Equalization and Source Separation provides a unified, systematic, and synthetic presentation of the theory of unsupervised signal processing. Always maintaining the focus on a signal processing-oriented approach, this book describes how the subject has evolved and assumed a wider scope that covers several topics, from well-established blind equalization and source separation methods to novel approaches based on machine learning and bio-inspired algorithms.

    From the foundations of statistical and adaptive signal processing, the authors explore and elaborate on emerging tools, such as machine learning-based solutions and bio-inspired methods. With a fresh take on this exciting area of study, this book:

    • Provides a solid background on the statistical characterization of signals and systems and on linear filtering theory
    • Emphasizes the link between supervised and unsupervised processing from the perspective of linear prediction and constrained filtering theory
    • Addresses key issues concerning equilibrium solutions and equivalence relationships in the context of unsupervised equalization criteria
    • Provides a systematic presentation of source separation and independent component analysis
    • Discusses some instigating connections between the filtering problem and computational intelligence approaches.

    Building on more than a decade of the authors’ work at DSPCom laboratory, this book applies a fresh conceptual treatment and mathematical formalism to important existing topics. The result is perhaps the first unified presentation of unsupervised signal processing techniques—one that addresses areas including digital filters, adaptive methods, and statistical signal processing. With its remarkable synthesis of the field, this book provides a new vision to stimulate progress and contribute to the advent of more useful, efficient, and friendly intelligent systems.

    Introduction

    Channel Equalization

    Source Separation

    Organization and Contents

    Statistical Characterization of Signals and Systems

    Signals and Systems

    Digital Signal Processing

    Probability Theory and Randomness

    Stochastic Processes

    Estimation Theory

    Linear Optimal and Adaptive Filtering

    Supervised Linear Filtering

    Wiener Filtering

    The Steepest-Descent Algorithm

    The Least Mean Square Algorithm

    The Method of Least Squares

    A Few Remarks Concerning Structural Extensions

    Linear Filtering without a Reference Signal

    Linear Prediction Revisited

    Unsupervised Channel Equalization

    The Unsupervised Deconvolution Problem

    Fundamental Theorems

    Bussgang Algorithms

    The Shalvi–Weinstein Algorithm

    The Super-Exponential Algorithm

    Analysis of the Equilibrium Solutions of Unsupervised Criteria

    Relationships between Equalization Criteria

    Unsupervised Multichannel Equalization

    Systems withMultiple Inputs and/orMultiple Outputs

    SIMO Channel Equalization

    Methods for Blind SIMO Equalization

    MIMO Channels and Multiuser Processing

    Blind Source Separation

    The Problem of Blind Source Separation

    Independent Component Analysis

    Algorithms for Independent Component Analysis

    Other Approaches for Blind Source Separation

    Convolutive Mixtures

    Nonlinear Mixtures

    Nonlinear Filtering and Machine Learning

    Decision-Feedback Equalizers

    Volterra Filters

    Equalization as a Classification Task

    Artificial Neural Network

    Bio-Inspired Optimization Methods

    Why Bio-Inspired Computing?

    Genetic Algorithms

    Artificial Immune Systems

    Particle Swarm Optimization

    Appendix A: Some Properties of the Correlation Matrix

    Appendix B: Kalman Filter

    References

    Index

    Biography

    João Marcos Travassos Romano is a professor at the University of Campinas (UNICAMP), Campinas, Sao Paulo, Brazil. He received his BS and MS in electrical engineering from UNICAMP in 1981 and 1984, respectively. In 1987, he received his Ph.D from the University of Paris–XI, Orsay. He has been an invited professor at CNAM, Paris; at University of Paris–Descartes; and at ENS, Cachan. He is the coordinator of the DSPCom Laboratory at UNICAMP, and his research interests include adaptive filtering, unsupervised signal processing, and applications in communication systems.

    Romis Ribeiro de Faissol Attux is an assistant professor at the University of Campinas (UNICAMP), Campinas, Sao Paulo, Brazil. He received his BS, MS, and Ph.D in electrical engineering from UNICAMP in 1999, 2001, and 2005, respectively. He is a researcher in the DSPCom Laboratory. His research interests include blind signal processing, independent component analysis (ICA), nonlinear adaptive filtering, information-theoretic learning, neural networks, bio-inspired computing, dynamical systems, and chaos.

    Charles Casimiro Cavalcante is an assistant professor at the Federal University of Ceará (UFC), Fortaleza, Ceara, Brazil. He received his BSc and MSc in electrical engineering from UFC in 1999 and 2001, respectively, and his Ph.D from the University of Campinas, Campinas, Sao Paulo, Brazil, in 2004. He is a researcher in the Wireless Telecommunications Research Group (GTEL), where he leads research on signal processing for communications, blind source separation, wireless communications, and statistical signal processing.

    Ricardo Suyama is an assistant professor at the Federal University of ABC (UFABC), Santo Andre, Sao Paulo, Brazil. He received his BS, MS, and Ph.D in electrical engineering from the University of Campinas, Campinas, Sao Paulo, Brazil in 2001, 2003, and 2007, respectively. He is a researcher in the DSPCom Laboratory at UNICAMP. His research interests include adaptive filtering, source separation, and applications in communication systems.