Artefacts Removal in Signals Acquired with Wearable Systems using Blind Separation of Convolutive Mixtures

Wearable systems are designed to monitor vital signs in real life environments. Detected signals are often affected by several artefacts, like movement related ones. This work focuses on the application of a multivariate approach that can take advantage of the multichannel acquisition in order to improve the detection of signals and to perform effective removal of artefacts. The approach we propose is based on the blind separation of convolved mixtures by means of independent component analysis in frequency domain. The use of convolved mixtures allows taking into account the delay in signal propagation and some complex physiological effects in signal generation. The proposed methodology was applied to real data acquired by mean of a wearable system.