Abstract:

The FFT is a successful steady-state analysis tool, but falls short when temporal analysis is needed. A class of filters designed around a Gaussian windowed DFT is described that, when applied to signal data sets in a way similar to that of a wavelet analysis, yields high-definition, fractional octave bandwidth, temporal analysis. The filter set's desired properties are temporal resolution that is inversely proportional to bandwidth and directly proportional to frequency, no assumption of signal periodicity, and the trade-off of accuracy against temporal window width. The main undesirable property is the computation needed to perform an analysis. Examples are chosen to show the need of a good display technique. They include analysis of a mathematical impulse, a hearing aid output as driven by a transient tone burst, and a room reflection with a broadband chirp signal. The author wishes to acknowledge Mike Day and Ilker Gungor of Frye Electronics for help in basic concepts and software programming.