The classification of noise-like signals comprised of quasi-periodic transients

Two algorithms based on principal component analysis of spectra of a point process derived from the turning points of the raw signal are particularly promising. The first method uses a database of all previously classified standards and a least squares technique to generate clusters and to observe deviation from these standards. This highly efficient means of data reduction results in two- or three-dimensional displays of good discriminating power. The second method splits up the set of understood instances into pre-defined groups. A matching process is then used to classify new instances on the basis of the smallest and whitest residual. There is evidence that this method may be more sensitive and provide even more powerful discrimination