Skip to Main content Skip to Navigation

Egalisation aveugle de systèmes multi-utilisateurs

Abstract : The present thesis introduces three algorithms for BSS. The first algorithm, called PAJOD (Partial Approximate Joint Diagonalization), assumes that the channel impulse response matrix is para-unitary. This hypothesis is verified when the observations have been previously whitened. The problem can then be reduced to a joint partial (thus approximate) diagonalization of a matrix set containing the sensor-output higher-order cumulants at different time lags. In a bid to reduce complexity, the PAJOD algorithm is only designed to find a semi-unitary matrix and not para-unitary. This motivates the development of the second algorithm, so-called PAFA (PAra-unitary FActorization), which is also based on the pre-whitening assumption. Contrary to PAJOD, PAFA exploits an exact parameterization of para-unitary matrices to obtain the equalizer impulse response matrix with the appropriate structure. Like PAJOD, the criterion for para-unitary factorization exploits the observation higher-order cumulants at different lags. The last algorithm, APF (Alphabet Polynomial Fitting), is different from the other two in that it does not make the para-unitary channel assumption and requires prior knowledge of the input modulations. This knowledge enables the extraction of input signals with known modulation in an instantaneous or convolutive mixture. Indeed, if all emitted signals employ different modulations, it is possible to extract the signals one by one. To this end, the algorithm carries out a deflation-based extraction driven by the maximization of a contrast criterion adapted to each modulation.
Document type :
Complete list of metadata
Contributor : Monique Oger <>
Submitted on : Tuesday, December 19, 2006 - 3:14:37 PM
Last modification on : Wednesday, October 14, 2020 - 4:24:04 AM
Long-term archiving on: : Tuesday, April 6, 2010 - 9:18:00 PM


  • HAL Id : tel-00121138, version 1



Ludwig Rota. Egalisation aveugle de systèmes multi-utilisateurs. Autre [cs.OH]. Université Nice Sophia Antipolis, 2004. Français. ⟨tel-00121138⟩



Record views


Files downloads