In order to investigate the detailed structure of the five-minute spectrum, it is necessary to combine residuals from many days and from more than one station. The first stage in this process is to produce time series of equally spaced data points. Because we changed our data-point interval from 42 to 40 seconds at different dates for different stations, and also because, even where the blocks are all of the same length, the data from different stations are not all synchronous, interpolation is sometimes necessary. Interpolation uses a simple function, of very low order compared with the number of data points, to estimate the value of a function between existing sample points. Interpolation, by definition, involves some smoothing, which can be expected to multiply the computed spectrum by factors which change both the amplitude and phase of the Fourier components. We have tested three functions: a simple parabola, a cubic spline, and a sinc function truncated to six points on either side of the central maximum. We find that only the truncated sinc function adequately preserves the higher frequencies in the spectrum. Details are given by Elsworth (1991). The high-precision timing synchronisation needs for the network sites are handled by GPS receivers at Tenerife, Carnarvon, Sutherland, Las Campanas and Narrabri, and a WWV receiver at Mount Wilson. (The timing signals at Tenerife, Sutherland and Mount Wilson are provided by the host establishments.)
In choosing which data to include in the time series, it is
convenient to quantify the quality of each day. A useful measure is
the ratio of the power in the five-minute band (1.5 to 
) to the power above . Typical values of this
"figure of merit'' are: around 1.5 to 3 for Mt. Wilson, and for
Carnarvon prior to a recent refurbishment; 5 for good data from
Tenerife; 6 to 8 for Carnarvon after refurbishment; and between 30
and 50 for the best data from the new stations. This reflects the
much improved high-frequency noise performance of the new
instruments. This measure of data quality can be used both to exclude
very poor data from the spectrum and to choose between stations when
data from more than one are present at the same time. In general,
data with a figure-of-merit less than 1 are worthless and are
excluded at the daily fitting stage.
The use of data where there are substantial overlaps between stations has become important now that we have six stations operating. The total fractional time for which we had overlap coverage during the calender year 1995 was 0.45. Overlaps of several hours can occur between stations; shorter three-station and (more rarely) four-station overlaps also occur (see later (Fig. 2 (click here))). The existence of such overlapping data allows us to investigate the solar noise by cross-correlation techniques, as described by Elsworth et al. (1994a). It also means that we can check, and where necessary correct, the time keeping at the stations by looking for obvious lags between the residuals. This is possible only when the data from each site are of reasonable quality. With the highest-quality data, we can check the timing to within a few seconds by interpolating to find the maximum cross-correlation: where the data quality is not quite so high the timing can be checked only to the nearest sample. Historically, we simply chose the best set of data (that is, the one with the highest figure of merit) at each overlap for inclusion in the time series: we now go on to discuss a straightforward method of merging residuals from different sites, that takes advantage of the extra statistical precision afforded, over the 5-minute region, by the overlap data.