AN ENGINEERING GUIDE TO SPECTRAL ANALYSIS.

The document is aimed at providing insight into the various aspects of spectral analysis of samples drawn from populations generally representing oceanographic and geophysical events. Hypothetical spectra, with more or less typical shapes, are developed and from these are generated a series of time histories. The records are subjected to spectral analysis with systematic variation of number of points, sampling interval and number of lags. From the results of computing several hundred spectra, it is seen that estimation of the true spectrum is affected in various ways by length of record, resolution, aliasing, spectral window and confidence bounds. Since it is impossible to specify a set of rules for every conceivable spectral density calculation, this paper shows how consideration of the elements mentioned above leads to design of the analysis experiment most likely to extract maximum reliable information from the data. In general, the work described here applies to stationary processes, but the treatment of non-stationary events is also discussed briefly. Author