Hardware based blind source separation for realtime signal processing

Otto Frandsen

AbstractThe topic of this thesis is Blind Source Separation in context of hardware design. The demand for realtime processing enhances development of both sophisticated algorithms and advanced digital circuits. This project is based on Molgedey and Shusters time delay decorrelation algorithm. The gradient descent method is utilized to obtain independent components, and properties of the algorithm is investigated. Issues involved by converting floating point algorithms in Matlab to the fixed point domain of Hardware Description Languages is an aspect of the thesis. Synthizable HDL code is developed from a Simulink model by the System Generator toolbox.

In Danish:

Dette eksamensprojekt omhandler Blind Signal Separation i sammenhæng med hardware design. Krav om realtids eksekvering fremmer udviklingen af både sofistikerede algoritmer som avancerede digitale kredsløb. Dette project er baseret på Molgedey og Shusters tidsforsinkede korrelations algoritme. Gradient nedstignings metoden anvendes til at opnå uafhængige komponenter, og algoritmens egenskaber er undersøgt. Problemstillinger omkring konvertering af floating point algoritmer i Matlab til fixed point domænet i hardware beskrivende sprog, er en af aspekterne i dette afgangs projekt. Syntiserbar HDL kode er udviklet fra en Simlink model med værktøjet SystemGenerator.
KeywordsBlind source separation, Temporal correlation, Joint diagonalization of matrices, Optimization by gradient descent, Singular mixing matrices, Limitations by fixpoint realization, Algorithm/system modelling, Test methods
TypeMaster's thesis [Academic thesis]
Year2003
PublisherInformatics and Mathematical Modelling, Technical University of Denmark, DTU
AddressRichard Petersens Plads, Building 321, DK-2800 Kgs. Lyngby
SeriesIMM-THESIS-2003-17
Electronic version(s)[zip]
BibTeX data [bibtex]
IMM Group(s)Intelligent Signal Processing


Back  ::  IMM Publications