Blind signal separation and its application to long-term bio-medical monitoring

There is a real clinical need to detect, separate, classify and ultimately monitor bio-medical signals over long periods of time. This would permit such applications as monitoring operator state using electroencephalogram (EEG) and determining foetal well-being in labour non-invasively. The long-term monitoring problem is, however, very challenging as bio-medical signals can be masked by much stronger interference, occur intermittently and the analysis may have to be performed in real-time (i.e. on-line) and over periods lasting several hours if not days. In this paper, to help overcome these problems, blind signal separation (BSS) methods that are capable of long-term monitoring are presented. It is established that block-based BSS approaches can be developed into efficient moving (exponentially fading) window approaches where the permutation problem in each window is overcome and subsequently the statistics of the signals can be updated and tracked. This approach is called the tracking blind signal separation (BLISS) algorithm. To provide a demonstration of concept, separation results are presented for EEG and twin foetal electrocardiogram (ECG) analysis.