Adaptive Blind Signal and Image Processing: Learning Algorithms and Applications
Adaptive Blind Signal and Image Processing: Learning Algorithms and Applications
Blind Source Separation Using Temporal Predictability
Neural Computation
An instantaneous formulation of mixtures for blind separation of propagating waves
IEEE Transactions on Signal Processing
| Hi-index | 0.00 |
We propose a novel approach to blind signal deconvolution. It is based on the approximation of the source signal by Taylor series expansion and use of a filter bank-like transform to obtain multichannel representation of the observed signal. Currently, as an ad hoc choice a wavelet packets filter bank has been used for that purpose. This leads to multi-channel instantaneous linear mixture model (LMM) of the observed signal and its temporal derivatives converting single channel blind deconvolution (BD) problem into instantaneous blind source separation (BSS) problem with statistically dependent sources. The source signal is recovered provided it is a non-Gaussian, non-stationary and non-independent identically distributed (i.i.d.) process. The important property of the proposed approach is that order of the channel filter does not have to be known or estimated. We demonstrate viability of the proposed concept by blind deconvolution of the speech and music signals passed through a linear low-pass channel.