The IVM flat fields are taken at several times during the day to insure that the data is corrected, since it is known that there are variations in the flat fields. This note discusses those variations.
The flat field variation in the IVM is presumed to be a product of the pixel-to-pixel gain variations in the CCD times the vignetting and transmission variations of the optics. Several of the optical elements appear to generate fringes that are both spatial and spectral in appeance. Various of the patterns also appear to vary with time, as parts of the IVM drift in temperature and perhaps alignment.
The flat field sequence consists of 6 spectral scans, each taken at a slightly different pointing. The pointings are not truly random, but are in a grid pattern. The total sequence requires about 3 to 4 minutes, therefore the solar fine structure varies as well. The telescope is not greatly defocused, so the solar structure is visible.
The 7 flat field sequences for 1999 August 17 were processed on September 28 2000. I will examine the inverted flats, as they are processed to eliminate various systematic.
The average flat for the entire day represents a summation of 1260 images. Black/white is [0.9,1.15].
The dominant upper-right to lower-left fringes are modulation sensitive and are believed to be formed deep in the system.
The individual images are superficially indistinguishable from the daily average. The averages over spectral scan of the individual sequences are ratioed to the daily average to show the deviations. Black/white is [0.98,1.02]. The sequence times are 16:25,16:58,18:07,19:17,20:13,21:18,and 22:12 UT.
Three patterns appear to be present.
The spectral fringing can be emphasized by ratioing the individual images to the spectral average. Black/white is [0.98,1.02].
There is a clear residual pattern that is nearly orthogonal to the fringes that dominate the daily or sequence averages. The pattern is repetitive but wavelength dependent. To isolate the spectral variation, a principal components analysis of the residual images from a flat sequence was performed.
The fraction of the variance removed by the principal components drops rapidly after the first few. The first 2 components remove 54% of the variance; the next 3 a further 15%. The appearance of the principal components confirms this rapid decline into noise. The following image shows the first 6 components, with the last appearing virtually noise-like.
The spectral variation that these components shows a simple pattern.
The first 5 coefficient vectors are plotted here, starting at the top, with a offset that declines from 4. to 0. The first components are basically a sine and cosine over the width of the scan, while the next 3 add higher spectral frequencies.
The source of this fringe pattern is not clear. It is not simply caused by a spatial shift of the pattern, as the following image shows.
The top images are taken at wavelength steps 2 and 26 (of 30) The lower left is their difference. The lower right is the difference of a shifted and unshifted copy of the step 2 image. The spectral fringe is just that, a true variation of transmission with wavelength.
Barry LaBonte