Non-orthogonal Gabor representation of biological signals

A new technique for modelling biological signals as a linear combination of non-orthogonal Gabor logons is described. The technique has been applied to two types of signals, event-related potentials (ERPs) and temporomandibular joint (TMJ) clicks. Examination of time-frequency representations of these signals revealed that they appear to consist of a small number of localized energy concentrations. Attempts to capture this apparent low dimensionality with the standard orthogonal Gabor expansion and the standard wavelet transform were unsuccessful. However, the non-orthogonal Gabor decomposition method described in this paper provides a compact, accurate signal representation and the parameters provide a good basis for ERP category and TMJ click classification