A New Algorithm for Blind Source Separation Using Parallel Processing

Abstract

In this dissertation the digital filters were used to solve the Convolutive Blind Source Separation within the parallel computing context, A revision of the strategy of tackling BSS and CBSS problems via parallel computing was proposed. In this dissertation a convolutive Blind Source Separation algorithm was developed within the context of parallel computing. The algorithm finds the demixing system by simultaneously jointly diagonalizeing a set of time-lagged correlation matrices by using parallel form of Jacobi plane transformations. The parallel computing approach was applied on some existing BSS and CBSS algorithms. As an example of the iterative CBSS algorithms Lambert and Bell work was selected and modified so as to apply parallel computing approach. Very good results were obtained compared with Lambert and Bell results. A very important improvement is that the developed equalizing system converges very fast. As an example of the diagonalization algorithms the JADE algorithm was modified to apply parallel computing approach. The results of the simulations showed that applying parallel computing approach on this type of algorithms is easier if done on shared memory systems.