Once the time series has been prepared, standard Fast Fourier
Transform algorithms are used to obtain the power spectrum. Much of
our work (Elsworth et al. 1990a,b, 1991, 1993, 1994b) has been
based on two-month time series. This is a convenient length for
several reasons: it gives a frequency resolution of about 0.2
Hz, which is suitable for studying modes with a natural width of
the order of 1 Hz; the use of relatively short time periods also
allows us to follow solar cycle trends, and to estimate the
uncertainties of peak parameters from the scatter of many
measurements. (For data obtained prior to the automation of the
network, this was the longest suitable interval over which an
adequate duty cycle could be obtained.)
We are now in a position to generate spectra over much longer periods with reasonable duty cycles, and these have enabled us to study the low-frequency end of the five-minute spectrum where the modes have small amplitudes and long lifetimes (Elsworth et al. 1995b; Chaplin et al. 1996d).