Seyrek temsile dayalı olarak sınırlı kaynakların kör ayrılması

In this thesis, a new class of novel Bounded Component Analysis (BCA) algorithms based on sparsity assumption, and their applications on some practical problems are introduced. As an application, the BCA algorithm proposed by Erdogan is demonstrated to separate the direct and reflected echoes which can improve the performance of classical direction of arrival estimation methods based on free space propagation theory. Following, we propose a new BCA framework for the separation of the instantaneous mixtures of sparse and bounded sources. Based on this framework, our fi rst proposed algorithm is named Sparse Bounded Component Analysis (SBCA) which is derived from a geometric objective function defi ned over a completely deterministic setting. Since the framework is not related to statistical properties of source signal model, it is applicable to sources which can be statistically independent or dependent in both spatial and temporal domains. Then, SBCA framework is extended to the convolutive mixtures of sparse sources, and time and frequency domain convolutive signal separation algorithms are proposed. Finally, a space-time analysis tool is provided to detect and identify brain activity signals that have a non-stationary nature. This tool mainly relies on the short time convergence property of our SBCA framework.