Parallel ICA algorithms

Blind source separation by Independent Component Analysis (ICA) has recently received attention because of its potential applications in signal processing applications. The separation time of the most well-known instantaneous Blind Source Separation (BSS) algorithms derived from ICA, kurtosis, Negentropy, and the Maximum Likelihood (MLE), isan application dependent. Furthermore, the performance of these algorithms should be assessed and their merits should be addressed to be able for a particular application to choose the most applicable algorithm. To address these issues, this paper focuses on the parallelization of the ICA algorithms based on SCILAB that uses a Parallel Virtual Machine (PVM). Also, we evaluate the performance of parallel ICA algorithms. Furthermore, the paper presents a new hybrid algorithm that combines MLE and Kurtosis. Extensive simulations on audio signals have been performed to demonstrate the evaluation of these algorithms. The achieved results show that the Maximum Likelihood (MLE) outperforms in terms of source to distortion ratio, source to interference ratio, source to noise ratio, and source to artifacts ratio, however, the kurtosis is the fastest algorithm only at low number of processors. ISCA