Extraction of Uncorrelated Sparse Sources from Signal Mixtures using a Clustering Method

TL;DR

This paper presents a clustering-based blind source separation method that exploits sparsity and uncorrelatedness of sources, demonstrating superior performance on clean signals but with sensitivity to parameters and noise.

Contribution

The paper introduces a novel clustering approach for blind source separation that does not assume source distributions and effectively handles sparse, uncorrelated sources.

Findings

Proposed method outperforms Fast-ICA and Clusterwise PCA on clean signals with proper parameters.

Performance is sensitive to input parameters and noise levels.

Method effectively isolates sources during segments where only one source is active.

Abstract

A blind source separation method is described to extract sources from data mixtures where the underlying sources are assumed to be sparse and uncorrelated. The approach used is to detect and analyse segments of time where one source exists on its own. Information from these segments is combined to counteract the effects of noise and small random correlations between the sources that would occur in practice. This combined information can then be used to estimate the sources one at a time using a deflationary method. Probability density functions are not assumed for any of the sources. A comparison is made between the proposed method, the Minimum Heading Change method, Fast-ICA and Clusterwise PCA. It is shown, for the dataset used in this paper, that the proposed method has the best performance for clean signals if the input parameters are chosen correctly. However the performance of this method can be very sensitive to these input parameters and can also be more sensitive to noise than the Fast-ICA and Clusterwise methods.