1. COS FUV Flat Fields and Signal-to-Noise Characteristics
Tom Ake
(STScI/CSC)
and the COS STScI and IDT Teams
2010 STScI Calibration Workshop
21 July 2010

2. Status of COS FUV Flat Fields
SMOV Program and Results
CALCOS Creation of X1DSUM Spectra
Generation of 1-D Flats Through Spectral Iteration
1-D Flat Evaluation and S/N Achieved
Conclusions and Plans

3. SMOV Program and Results
Prelaunch analyses indicated an FUV flat field could be made more easily using an external target
Program SMOV mapped science region of detector with WD
Exposures taken at 5 cross-dispersion positions with G130M, and 2 each with G160M and G140L
Different cenwaves and FP-POS settings used to separate spectral and detector features
2-D flats were made for each grating and segment
Flats removed prominent dips due to grid wire shadowing, but induced some structure due to low S/N
Extracting a 1-D flat and dividing it into the extracted spectrum is somewhat better than flat fielding the image prior to spectral extraction
Decided not to install the 2-D flat field files in CALCOS at this time, but to change merging of FP-POS exposures into X1DSUM spectra

4. CALCOS Creation of X1DSUM Spectra
Grid wire shadows are the largest FPN features (20% deep, every 840 pixels)
Originally, when CALCOS coadded FP-POS exposures into an X1DSUM spectrum, wire shadows were reduced in depth, but appeared in more places
For 4 FP-POS steps, 30% of pixels
were affected by grid wires
Until we have a flat field, CALCOS
SDQFLAG keyword changed to ignore
grid wires when creating X1DSUM
spectra
Cycle 18 GOs are required to use at
least two FP-POS or cenwave settings
to minimize chance that spectral features
fall on bad spots and to smooth out FPN

5. Generation of 1-D Flats Through Spectral Iteration
Since 1-D correction appeared to be better than 2-D, investigated spectral iteration of X1D extracted spectra to create 1-D flats
Technique had been developed for GHRS
Requires data taken at different grating settings (cenwave and/or FP-POS position)
Iterate between wavelength and pixel space in merging and correcting data sets
Solves simultaneously for the stellar spectrum and underlying fixed pattern noise (FPN)
Started analysis of G130M and G160M exposures to see if FPN residuals could be made into general 1-D flat fields
Each grating still has to be processed separately since spectra fall at different cross-dispersion locations

6. Spectral Iteration Results Example

7. Final 1-D Flat Fields - G130M

8. Flat Field Evaluation
Consistency check performed by dividing final 1-D flat into each contributing flat
Grid wire shadows are nicely corrected
Detector dead spots leave residuals since spectra were taken at different Y position. These regions were never expected to be correctable and are flagged by CALCOS
Long wavelength end of segment A
(X>11000) shows either misalignment
of flats or low S/N effects

9. Signal-to-Noise Achieved
Maximum S/N for single grating setting (~20 per resel) reached at ~1500 counts/resel (red line)
Current CALCOS FP-POS summing while ignoring grid wires improves global S/N by smoothing FPN (violet line)
S/N from flat fielding approaches Poisson noise for single exposures (open blue circles). With 4 FP-POS steps, S/N=100 per resel possible (blue line)
Distribution of P-flat variations give maximum S/N without a flat field
Histogram of variations in each NUV stripe fit with Gaussian profile
Widths indicate S/N (= 1/) ~ 50 per resel can be obtained without a flat
Caveat - using same data to evaluate as what went into the flats, although different targets, various cenwaves, and multiple FP-POS steps were averaged

10. Conclusions and Plans
1-D flats show promise. Need to check against more data
New flats cannot be used to correct data with the current flux calibration since sensitivity curves were created without flat fielding. A self-consistent flux calibration is being worked
Need to investigate why long wavelength side of segment A is so noisy
G140L still to be studied. Criteria for iteration convergence may need revision since spectrum covers only part of the detector segments
More high S/N observations are being obtained to characterize the flats at other cenwave settings