Deflation parallele avec des contrastes APF pour l'extraction aveugle de sources

The present paper addresses the problem of discrete source separation, particularly in the scenario where there exist less sensors than sources (underdetermined case). Alphabet polynomial fitting (APF) contrasts exploit the prior knowledge of digital modulation alphabets to perform the space-time equalization. Source extraction methods can be developed from such contrasts. The feature that each APF criterion targets a specific modulation enables the simultaneous operation of deflation algorithms to extract the signals of each modulation in parallel. Parallel deflation allows the reduction of the diversity necessary for the extraction of all sources from the mixture. Hence, this new approach can improve the estimation performance and the computational complexity of classical deflation.