Abstract: Principal features are inventively derived from a non-stationary time series signal, and can be used to classify the attributes of the signal. Intrinsic to the invention is its recognition of the value of information contained within singular vectors. The invention therefore seeks to render singular vectors as proper density functions for purposes of formulating at least one feature of a signal, which is or has been sensed. To this end, the signal is typically evaluated thusly according to the invention: a time-frequency distribution (e.g., positive time-frequency distribution) matrix is generated; the time-frequency distribution matrix is decomposed; an element-by-element square of singular vectors is performed; the non-principal singular values are sorted and truncated; density functions are obtained; the density functions are normalized; and, at least one feature from the normalized density functions is formulated.