Unsupervised Adaptive Filtering, Volume 1: Blind Source Separation
A complete, one-stop reference on the state of the act of unsupervised adaptive filtering While unsupervised adaptive filtering has its roots in the 1960s, more recent advances in signal processing, information theory, imaging, and remote sensing have made this a hot area for research in several diverse fields. This book brings together cutting-edge information previously available only in disparate papers and articles, presenting a thorough and integrated treatment of the two major classes of algorithms used in the field, namely, blind signal separation and blind channel equalization algorithms. Divided into two volumes for ease of presentation, this important work shows how these algorithms, although developed independently, are closely related foundations of unsupervised adaptive filtering. Through contributions by the foremost experts on the subject, the book provides an up-to-date account of research findings, explains the underlying theory, and discusses potential applications in diverse fields. More than 100 illustrations as well as case studies, appendices, and references further enhance this excellent resource. Topics in Volume I include:
* Neural and information-theoretic approaches to blind signal separation
* Models, concepts, algorithms, and performance of blind source separation
* Blind separation of delayed and convolved sources
* Blind deconvolution of multipath mixtures
* Applications of blind source separation
Volume II: Blind Deconvolution continues coverage with blind channel equalization and its relationship to blind source separation.
Why Read This Book
You should read this book if you want a unified, mathematically rigorous treatment of blind source separation (BSS) and blind channel equalization from the perspective of unsupervised adaptive filtering. You will get both algorithmic recipes (Infomax, natural gradient, ICA variants) and the theoretical performance and convergence considerations that help turn those recipes into reliable implementations.
Who Will Benefit
Advanced DSP engineers, researchers, and graduate students working on source separation, adaptive filters, or communications who need a deep understanding of BSS/ICA algorithms and their theoretical foundations.
Level: Advanced — Prerequisites: Linear algebra (eigenanalysis, matrix decompositions), probability and stochastic processes, basic information theory, and graduate-level familiarity with adaptive filters and spectral analysis.
Key Takeaways
- Formulate blind source separation and blind channel equalization problems in information-theoretic and statistical terms
- Derive and implement core ICA/BSS algorithms such as Infomax, natural-gradient methods, and FastICA-style approaches
- Analyze convergence, stability, and performance of unsupervised adaptive filters using statistical measures (e.g., mutual information, higher-order statistics)
- Relate blind equalization methods used in communications to BSS techniques used in audio and sensor processing
- Apply practical considerations for algorithm selection, preconditioning, and estimation when facing real-world mixtures and noise
Topics Covered
- Preface and Historical Background to Unsupervised Adaptive Filtering
- Problem Formulation: Models for Blind Source Separation and Blind Equalization
- Information-Theoretic Criteria: Entropy, Mutual Information, and Contrast Functions
- Independent Component Analysis: Principles and Identifiability
- Algorithmic Approaches I: Infomax, Maximum Likelihood, and Natural Gradient
- Algorithmic Approaches II: FastICA, Decorrelating Methods, and Joint Diagonalization
- Higher-Order Statistics and Cumulant-Based Methods
- Performance Analysis: Convergence, Cramer-Rao Bounds, and Robustness
- Practical Issues: Preprocessing, Whitening, Scaling, and Permutation Problems
- Applications: Audio/ Speech Separation, Communications Equalization, Remote Sensing
- Simulations, Implementation Notes, and Case Studies
- Summary, Open Problems, and Directions for Volume 2
Languages, Platforms & Tools
How It Compares
Covers the same BSS/ICA landscape as Hyv rinen, Karhunen & Oja's Independent Component Analysis (which is more focused on ICA theory and algorithms), but Haykin places stronger emphasis on unifying blind equalization and adaptive-filter perspectives.
