Frequency domain ICA based signal restoration from non-linearly distorted acoustic signals

Abstract

Independent Component Analysis (ICA) is a class of blind source separation which can be successfully used for extracting unknown independent source signals from a set of signal mixtures. In this study, we introduce a new method for separation of acoustic source signals using frequency domain complex valued ICA. Although the conventional time domain ICA algorithms can be effectively used for extraction of the source signals from a set of linearly mixed signal mixtures, such algorithms fail under non-linear mixing conditions. The proposed method is capable of extracting acoustic source signals from several non-linearly distorted and corrupted audio signals. The results show that the frequency domain ICA has a superior performance compared to the conventional real valued time domain ICA algorithm under non-linear conditions.