Adaptive blind source separation of multiple-input multiple-output linearly time-varying FIR system

A new adaptive blind separation scheme for sources mixed by a multiple-input multiple-output (MIMO) linearly time-varying (LTV) FIR system is proposed. First, by dividing measured samples into a series of short segments, time-varying coefficients of the mixing system are approximated by polynomials in time over each segment. Then, a two-step BSS scheme is presented for the approximated system. The first step is to estimate the time variation and convolution effects of the mixing system, and reduce the LTV-FIR mixing system to a linearly time-invariant (LTI) instantaneous system using the conventional input/output system identification scheme. The second step uses the mutual independence knowledge of the sources to further separate the sources from the LTI instantaneous system. The theoretical and experimental studies show that the new BSS scheme has an improved performance in separating sources mixed by an LTV-FIR system.