Course detail

Advanced methods of signal processing

FEKT-MMZSAcad. year: 2015/2016

Formalised optimum filtering and signal restoration in unified view: Wiener filter in clasical formulation and generalised discrete Wiener-Levinson filter, Kalman filtering; source modelling and signal restoration, further approaches. Adaptive filtering and identification, algorithms of adaptation, classification of typical applications of adaptive filtering. Neural networks - error-backpropagation networks, feed-back networks, self-organising networks, and their application in signal processing and classification. Non-linear filtering - polynomial and ranking filters, homomorphic filtering and deconvolution, non-linear matched filters. Typical applications of the above methods.

Language of instruction

Czech

Number of ECTS credits

6

Mode of study

Not applicable.

Learning outcomes of the course unit

The graduate of the course is capable of:
- understanding principles of advanced signal processing methods and their relations,
- choosing a suitable method for a specific practical purpose,
- implementing the chosen method in a computing environment as a commercial or individually developed software,
- properly interpreting the results of the analyses.
-

Prerequisites

The knowledge on the Bachelor´s degree level is requested, namely on digital signal processing

Co-requisites

Not applicable.

Planned learning activities and teaching methods

Teaching methods depend on the type of course unit as specified in the article 7 of BUT Rules for Studies and Examinations. Techning methods include lectures and computer laboratories. Course is taking advantage of e-learning (Moodle) system.

Assesment methods and criteria linked to learning outcomes

Requirements for completion of a course are elaborated by the lecturer responsible for the course every year;
basically:
- obtaining at least 12 points (out of 24 as course-unit credit based on active presence in demonstration exercises),
- successful passing of final written exam (up to 76 points)

Course curriculum

1. Identification of stochastic signals. Introduction to signal restoration, formalised optimum LMS signal restoration in unified presentation
2. Wiener filter in classical and generalised discrete representation
3. Scalar and vector Kalman filtering, modelling of signal sources
4. Principles of adaptive filtering, algorithms of adaptation
5. Applications of adaptive filtering, classifying applications
6. Introduction to non-linear filtering – polynomial and ranking filters, homomorphic filtering and deconvolution, nonlinear matched filters
7. Introduction to neural networks, individual neuron and its learning
8. Feedforward layered networks learning by error back propagation, radial base networks
9. Feedback networks: Hopfield and Boltzmann nets, competing and Jordan networks
10. Self-organising networks, Kohonen maps
11. Applications of neural networks in signal processing and analysis
12. Principal component analysis in signal processing
13. Independent component analysis in signal processing

Work placements

Not applicable.

Aims

The goal of the course is to provide insight into principles of advanced signal processing methods and their relations, and demonstrating some practical applications.

Specification of controlled education, way of implementation and compensation for absences

Delimitation of controlled teaching and its procedures are specified by a regulation issued by the lecturer responsible for the course and updated for every year (see Rozvrhové jednotky).
Basically:
- obligatory computer-lab tutorial
- voluntary lecture

Recommended optional programme components

Not applicable.

Prerequisites and corequisites

Not applicable.

Basic literature

Jan, J.: Číslicová filtrace, analýza a restaurace signálů. Vutium Brno, 2002
J.Jan: Digital Signal Filtering, Analysis and Restoration. IEE Publishing, London, UK, 2000

Recommended reading

B.Mulgrew, P.M.Grant J.S.Thompson: Digital Signal Processing, Concepts and Applications, Mac-Millan Pres Ltd.1999
B. Kosko (ed.): Neural Networks for signal processing. Prentice Hall 1992

Classification of course in study plans

  • Programme EEKR-M Master's

    branch M-BEI , 1. year of study, summer semester, optional specialized

  • Programme EEKR-M1 Master's

    branch M1-BEI , 1. year of study, summer semester, optional specialized
    branch M1-EST , 1. year of study, summer semester, optional interdisciplinary

  • Programme EEKR-M Master's

    branch M-EST , 1. year of study, summer semester, optional interdisciplinary

  • Programme EEKR-CZV lifelong learning

    branch ET-CZV , 1. year of study, summer semester, optional specialized

Type of course unit

 

Lecture

39 hours, optionally

Teacher / Lecturer

Syllabus

More profound view of linear filtering, state models, design methods of FIR and IIR filters
Multirate systems, banks of decimation and interpolation filters
Non-linear filtering, polynomial filters, generalised and adaptive median filter, homomorphic filtering, non-linear matched filters
Classical and modern methods of statistical characteristics identification of stochastic signals
Unifying approach to methods of formalised signal restoration. Discrete Wiener filter as a golden standard
Kalman filtering, stationary and non-stationary, aaplication in signal restoration and in modelling of signal sources
Restoration via frequency domain. Constrained deconvolution, deconvolution via optimisation of impulse response
Concept of adaptive filtering, filter with recursive optimum adaptation, filter with stochastic gradient adaptation
Classification of adaptive filtering applications: system identification and modelling, channel equalisation, adaptive linear prediction, adaptive noise adn interference cancelling
Introduction to architecture and properties of neural networks: feed-forward networks, learning, knowledge generalisation; feed-back networks; self-organising maps.
Neural-network based signal processing: learned and adaptive neural filter, formalised restoration by feed-back networks
Typical applications of the above methods in communication, speech and acoustic signal processing
Typical applications of the above methods in processing of measurement and diagnostic signals, system identification and in biomedical applications

Exercise in computer lab

26 hours, compulsory

Teacher / Lecturer

Syllabus

Becoming acquainted with MATLAB - Signal Processing Toolbox and DSP Blockset environment
Design and verification of an FIR or IIR filter
Application of adaptive median- or homomorphic filtering
Identification of statistical properties of given stochastic signals
Design and application of a discrete Wiener filter
Kalman filtering, aaplication in modelling of signal sources
Restoration by a modified inverse filter via frequency domain
Experiment with an adaptive filter with stochastic gradient adaptation
Adaptive cancelling of given interference
Experimenting with a feed-forward network, learning and knowledge generalisation
Signal processing by a learned neural filter
Applications of given methods in acoustic signal processing
Applications of the above methods in processing of given measurement and diagnostic signals