February 24, 2015
Duration 13 weeks.
Thursday 8:30 - 11:45
International labs link
The usability of
adaptive algorithm methods to the
solution of real problems is extensive and represents a paradigm for
many strategic applications.
Example can be found in
multimodal and multimedia communications, the biological and biomedical
areas, economic model, environmental sciences, acoustics,
telecommunications, remote sensing, monitoring, and, in general,
modeling and prediction of complex physical phenomena.
processing methods are also used in economic and financial sciences, in
engineering and social sciences, neuroscience, and in many other
areas of high strategic interest.
Adaptive signal processing is also a very active field of
study and research, that, for a thorough understanding, requires
advanced interdisciplinary knowledge.
Objectives of the Course
aim of this course is to provide students advanced theoretical and
practical tools for the study and determination of circuit structures,
and robust adaptive algorithms, in different application scenarios. In
particular, in addition to presenting the fundamental theoretical base
concepts, the most important adaptive algorithms are introduced, while
also providing tools to evaluate the algorithms’ performance.
student, in addition to acquiring the basic theories, will be able to
design and implement the algorithms and evaluate their performance for
specific applications even in the presence of parallel and distributed
Brief review of basic concepts of the
nonlinear programming: fundamental concepts of the unconstrained and
the constrained optimization methods.
Wiener optimal filtering theory.
The normal equations and the optimal Wiener filter in
discrete time. Type 1, 2 and 3 multi-channel notations, and its
multi-input-output optimal filter generalization are presented. Are
also discussed corollaries, and presented some applications related
to the random sequences prediction and estimation.
Principle of least squares
The normal equations
in the Yule-Walker formulation. Minimum variance optimal estimators;
the normal equations weighing techniques, the regularization LS
approach, the linearly constrained and the nonlinear LS techniques.
Methods of matrix decomposition for solving the LS systems in the
cases and of over/under-determined case. Singular value
decomposition in the solution of the LS systems.
of Lyapunov attractor for the iterative LS solution.
Total least squares (TLS),
Matching pursuit algorithms for underdetermined sparse LS systems.
Stochastic gradient paradigm and LMS algorithm.
performance evaluation of adaptation algorithms: convergence speed
and tracking analysis.
algorithm variants, (NLMS,
multi-channel, delayed learning algorithms, filtered-x LMS the
method of the adjoint network).
order LS algorithms: the Newton’s method, the recursive least
squares (RLS), the affine projection algorithm (APA), the
General adaptation low based on natural gradient
approach in presence of sparsity constraints, LASSO.
adaptive algorithms. Frequency domain adaptive filters (FDAF).
Partitioned FDAF Transformed domain adaptive filtering. Multirate
methods and the subband adaptive filters (SAF).
Forward and backward
linear prediction. Order recursive algorithms. Implementative issues
with particular robustness and efficiency properties. In connection
with this last aspect, the subject of the filter circuit structure
and the adaptation algorithm is introduced, in relation to the
problems of noise control, scaling and efficient computation, and
effects due to coefficients quantization.
adaptive filtering. Anechoic and echoic wave propagation model,
sensors directivity functions, array signal model, steering
vectors of some typical array geometries. Characteristics of noise
field, array quality indices. Conventional LS beamforming. Super
directive methods. Adaptive on-line beamforming operating non
stationary signal condition. Time-delay estimation (TDE), direction
of arrival (DOA) estimation in the case of free-field narrow-band
signals and in the case broadband signals in reverberant environment.
Fundamentals of parallel distributed processing (PDP) model.
Computational and Biological inspired PDP model. Supervised and
Unsupervised Learning Algorithm. Learning in distributed environment.
Machine Learning for Signal
In addition to the theoretical part, is
provided for the implementation of some of the described
algorithms on parallel and distributed architecture.
Aurelio Uncini, "Fundamentals of
Adaptive Signal Processing" - Springer, ISBN
978-3-319-02806-4, Febbraio 2015.
Aurelio Uncini, “Algoritmi
Adattativi per l'Elaborazione dei Segnali”, Ed.
Esculapio, ISBN 978-88-7488-840-5, Febbraio 2015..
Aurelio Uncini, “Algoritmi
adattativi per circuiti intelligenti”, dispense disponibili
presso i centri fotocopie.
S. Scardapane, D. Comminniello,
M. Scarpiniti, A. Uncini, “Designing Lerge Machine
Learning Simulations Using the Lynx Toolbox".
Other recommended reading
Ian Foster, "Designing &
Building Parallel Programs: Concepts & Tools for Parallel
Software Engineering," Addison-Wesley, 1995, online:
Dimitri P. Bertsekas and John N.
Tsitsiklis, Parallel and Distributed Computation: Numerical
Methods, ISBN 1-886529-01-9
Rumelhart, D.E., Hinton, G.E., &
McClelland, J.L. (1986). A General Framework for Parallel
Distributed Processing. In Rumelhart, D.E., & McClelland,
J.L. and the PDP Research Group (1986) Eds. Parallel
Distributed Processing: Explorations in the Microstructure
of Cognition. Volume 1: Foundations. MIT Press: Cambridge,