FPGA implementation of blind signal separation by estimating mixing channels

Abstract

The goal of blind signal separation (BSS) is to recover the original independent sources given only sensor observations that are linear mixtures of independent source signals. The term blind indicates that both the source signals and the way the signals were mixed are unknown. Blind signal separation based on independent component analysis has been successfully used many fields, it still needs to satisfy more strict requirements to extend its applications. Information maximization algorithm is used to maximize the entropy at the outputs. Blind signal separation could use two kinds of structure. The one is feedback structure and the other is feed-forward structure. These two structures are used for many kinds of simulations. The advantages of using a feedback system is that a parsimony of parameters may be sufficient to approximate the inverse system. A disadvantage is that the full filter architectures whiten the original sources and one possible way to prevent this is to approximate the inverse system with only cross filters. Another aspect is the problem of finding correct time-delays for natural signals since they are correlated over time. An inaccurate time-delay estimation may result into an incorrect system inverse and therefore the time-delay learning rule for the feed-forward system was omitted and the time-delays were incorporated simply as part of the FIR filter. Also, since feedback systems are limited to minimum-phase systems a feed-forward system is proposed to give a more general inverse system. The blind signal separation for convolutive mixtures estimates the finite unmixing matrix which represents time delays, that is number of filter taps. However, the estimated unmixing matrix same as approximation of IIR filter, therefore the error term between FIR and IIR filter more increased. As if the blind signal separation estimate the mixing matrix instead of unmixing matrix, then each component of matrix can be represented as FIR filters. E...