Predictive Algorithm Based Low Complexity 2D FastICA

Description/Abstract

Abstract—FastICA (FICA) is a popular algorithm in solving Blind Source Separation (BSS) problem. It mainly involves solving an adaptive iterative equation for computing the constituent vectors of the unmixing matrix to be estimated. Each of these iterations includes adaptive equation computation, normalization and orthogonalization which need significant amount of complex arithmetic operations such as multiplications, divisions and square rooting. For commonly used 2-dimensional FICA, the unmixing matrix consists of two vectors each of which is computed by running such FICA Iteration several times until convergence is achieved. In this paper we propose an algorithm that predicts the second vector of the unmixing matrix from the first one without any numerical calculation and thereby reduces the overall arithmetic computation complexity significantly by removing one complete stage of FICA Iteration. Subsequently analysis is carried out in terms of computational complexity and delay with respect to varying framelength, wordlength and number of iteration of convergence and compared with the previously reported 2D FICA architectures.