Real-time Independent Component Analysis Implementation and applications

Abstract

A common problem in disciplines such as high energy physics, biomedicine and acoustic signal processing is finding a suitable representation of multivariate data. Independent Component Analysis (ICA) is a recently developed mathematical tool that can recover independent source signals and is now mature enough to be implemented in real-time applications such as photomultipliers signal processing, magnetic resonance imaging and acoustic arrays. This technique is based on the assumption that signals from different sources are statistically independent and statistically independent signals can be extracted from mixture signals. ICA defines a model for the observed data that requires a large number of samples in order to establish the necessary statistics. The model assumes that the data variables are linear combination of unknown variables, the unknown variables are assumed to be non-Gaussian and independent. The goal then becomes to find a transformation in which the components are as statistical independent as possible from each other. This technique is related with methods such as principal component analysis and factor analysis. The ICA algorithm is computing intensive since it must accumulate and go through the signal samples performing complex operations. Efficient versions of the algorithm have being already deployed using different techniques such as the FastICA that can be implemented efficiently in hardware platforms such as DSP processors and FPGA's. In this paper, we present the ICA principles, implementation and current applications.