Blind source separation using sub band based JADE algorithm

The goal of blind source separation is to recover independent source signals from a set of mixed signals without having a prior knowledge of source signals. This technique is widely used in filed of communications', signal processing and applied to medical and astronomical images. Several methods have been developed to find a suitable linear transformation, these include principal component analysis, factor analysis, projection pursuit, independent component analysis. Recently, a particular method, called independent component analysis (ICA) has gained wide spread recognition. Independent component analysis (ICA) can be defined as maximizing the degree of statistical independence among the estimated signals using contrast functions.;The main goal of this thesis is to implement the joint approximate diagonalization of eigen matrices (JADE) using sub band analysis to blind source separation problem. A mixed signal can be described as mixture of the independent source signals mixed with random weights. To make this implementation suitable for real time, multipath implementation is suggested. This multipath implementation is done by convolving the unknown independent signals with the known channel taps.;The objective of this thesis is to show that higher-order correlations can be efficiently exploited to reveal independent components.;In subband JADE algorithm, the mixed signals are segregated into multiple subbands which are analyzed using power analysis to construct a full band weight vector using JADE algorithm. In this subband implementation weight vector with maximum power is choosen at regular time intervals contrary to full band implementation, where the mixed signals are fed to algorithm to estimate full band weight vector. The advantage of subband JADE implementation to fullband JADE is confirmed from simulation results in terms of performance and computational complexity.