Adaptive Filters and Neural Networks

Objective

Presentation and comprehension of the fundamental theory and the most important methods and applications of adaptive filters (AF) and neural networks (NN) for signal processing, with emphasis on methodology, the derivation of fundamentals, and application. Further information can be found at:http://www.isi.ee.ethz.ch/education/lectures/index.de.html

Content

Introduction to adaptive filters (AF) and overview of the most important applications. Identification, inverse modeling, prediction, interference canceling. Algebraic fundamentals, properties of the correlation matrix, role of eigenvalues and eigenvectors, eigenvalue spread. Minimization of the mean-squared error (MSE), orthogonality principle, Wiener filter. Adaptation algorithms for FIR adaptive filters. Newton and gradient method, time and frequency domain least-mean-square algorithms (LMS). Convergence properties, learning curves, misadjustment, excess MSE. Recursive least squares algorithm (RLS), computational complexity Introduction to neural networks (NN). Nonlinear function approximation. Artificial neurons, feedforward architectures, mulitlayer perceptrons (MLP). Backpropagation algorithm (BPA), scaled conjugate gradient algorithm (SCG), statistical alternatives, error functions, practical considerations, rules of thumb. Applications: function approximation, classification of patterns, time series, system modeling, control and filtering. Illustrative MATLAB-exercises (program frame provided).

Resources