A Deflation Procedure for Subspace Decomposition

Reza Sameni (1) , Christian Jutten (2) , Mohammad B. Shamsollahi (3)

1 College of electrical and computer engineering
2 GIPSA-VIBS - GIPSA - Vision and Brain Signal Processing
3 BiSIPL - Biomedical Signal and Image Processing Laboratory [Teheran]

A general deflation framework is described for the separation of a desired signal subspace of arbitrary dimensions from noisy multichannel observations. The method simultaneously uses single and multichannel priors to split the desired and undesired subspaces, even for coplanar (intersecting) subspaces. By appropriate use of signal priors, it can even extract signals from degenerate mixtures of signals and noise recorded from a few number of channels in low SNR scenarios, without the reduction of the data dimensions. As a case study, the performance of the proposed method is studied for the problem of extracting fetal cardiac signals from maternal abdominal recordings, over simulated and real data. A second case study deals with the degenerate problem of extracting diaphragmatic electromyogram from electrocardiograph artifacts. A provisional patent application based on this method has been filed.

Mots clés: signal denoising subspace; Deflation; nonlinear filtering; nonstationarity; periodicity; prior; rank preservation; Rayleigh quotient; signal denoising subspace.

Domaines: Traitement du signal et de l'image [eess.SP]; Traitement du signal et de l'image [eess.SP]

Reza Sameni, Christian Jutten, Mohammad B. Shamsollahi. A Deflation Procedure for Subspace Decomposition. IEEE Transactions on Signal Processing, 2010, 58 (4), pp.2363-2374. ⟨10.1109/TSP.2009.2037353⟩. ⟨hal-00466435⟩