This article describes the flat-fielding phase of data processing. The FIMS and SPEAR teams both followed a similar procedure for this step.
Differential non-linearities (DNL) are evident in raw FIMS-SPEAR data. These features do not indicate regions of higher and lower quantum detector efficiency (QDE). Rather, they were primarily caused by oscillations in the time-to-analog stage of the time-to-digital converter, manifesting as vertical and horizontal bands of higher and lower counts. When the constant current switched into charging the capacitor, there was ringing (signal oscillation around equilibrium due to a sudden change in input). Accordingly, these striations damp out toward the high-delay end of the anode. A secondary effect involved the crossed delay line anodes employed by the detectors which sampled the electron charge cloud emanating from the back of the micro channel plates; events were not lost, but were shifted slightly in position depending on where the cloud centroid landed relative to the anode wire centers.
Figure 1 - Raw L-Band Sky Survey. X and Y refer to the axes of the anode, which are tilted by 15 degrees with respect to the dispersion axis.
Because no long exposure flat-field detector data were obtained during ground calibration, dividing by a true flat-field was not possible for FIMS-SPEAR data. Instead, localized flat-field images were created by adding up all the data from the sky survey. The resulting images contain particle events, scattered light, stellar continua spectra, and emission lines. To reduce the effects of the stellar continua, the data were limited to periods of low countrate < 2.5*median(counts/s). Next, to eliminate the remaining spectral features, the images were first temporarily rotated by 15 degrees such the stellar continua and emission lines are parallel to the X and Y axes, respectively. Then each image row was divided by the median value of the row, and finally, each column was divided by the median of the column.
The resulting image is a localized flat-field with an average value equal to 1, depicting the small scale differences in sensitivity between adjacent parts of the detector caused by the anode wires (Figure 2). Hotspots, deadspots, and other small scale features such as reflections of light off the filter frames are also retained in the flat-field images, which means their effects were removed (to first order) when the flat-fields are applied to the science data. Large scale gradients (such as vignetting and the QDE as a function of wavelength) are not accounted for by these localized flat-fields, but were dealt with later by applying the effective area curves to integrated spectra.
Figure 2 - L-Band Flat-Field
Each photon event was assigned a weight based on its corresponding XY location on the flat-field image. Since the mean value of the flat-field image is one, no net counts were lost or gained by applying the weights of the flat-field. Typical weights were about 10% RMS.