On the detection of the number of signals in multi-lead ECGs

New methods which have been developed in the field of signal processing to detect the number of intrinsic source signals in a set of multiple signals have been studied in their application to the analysis of multilead ECGs (body surface maps). All these methods are based on the analysis of the eigenvalues of the covariance matrix related to the observed data. By using artificial multilead ECGs in which the properties of the signal and noise parts were known, the effect of taking into account these properties was investigated. It was found that the method based on information-theoretic criteria was superior in detecting the number of signal components, in the case that the noise in the data is non-white, as compared to a simple threshold method.