STScI Logo

Hubble Space Telescope
Obtaining COS FUV spectra with a S/N > 100/1 per resolution element

Because the COS FUV detectors are subject to position and time dependent fixed pattern noise, we do not foresee producing routine pipeline processed data with a S/N much larger than 30. However, S/N > 100 (per resolution element which is about 6 pixels) spectra can be achieved by using the iterative FPPOS technique (e.g., Bagnulo & Gies, 1991, ApJ, 376, 266, and Lambert et al., 1994, ApJ, 470, 756) as long as the source spectrum and the observational sequence meet certain conditions.

The observations must meet the following criteria:

  1. The spectrum must contain enough counts in the region of interest to infer a Poisson S/N that is 10-20% larger than the desired S/N.
  2. The exposure must be distributed among all four PPPOS positions and, preferably, multiple adjoining CENWAVE settings.
  3. The individual exposures at each CENWAVE/FPPOS combination should not be too long (roughly 1/4 of an orbit), to avoid smearing the spectrum on the detector as a result of Doppler shifts and OSM drifts. Several short exposures are better than fewer long ones.

The source spectrum should have the following properties:

  1. The spectrum cannot be too "busy". The iterative technique, which is needed to attain high S/N, cannot constrain the power in spatial frequencies that are harmonics of the FPPOS displacements (see, Jenkins, 2002, ApJ, 580, 938). For COS, this effect can be quite strong because all of its FPPOS settings are roughly the same size.
  2. The feature of interest should not be near the edge of the detector. This is because the detector edges are subject to large amplitude, unstable fixed pattern effects, and are not reliable. Furthermore, they will not be contained in all of the exposures.
  3. When searching for weak lines, one must also keep in mind that the COS line spread function has large wings which can reduce the ability to detect weak features (see, sec. 3 of COS ISR 2009-01(v1)).

When the preceding criteria are fulfilled, extremely high S/N results are possible. For the observation shown in the following plots, the ratio of the Poisson to the actual S/N varied from nearly 1 to about 1.2. The exact ratio depended on the total counts and the detector location, so there is no simple conversion. Nevertheless, for an uncomplicated spectrum, exposing to a Poisson S/N of 10-20% more than desired, should be adequate.

Over the region 1220-1250 Ang in Figure 1, the RMS scatter of 3-pixel bins about the fit shown in the figure gives a S/N of 89, compared to the expected Poisson S/N of 100 (see Figure 2). For 6-pixel bins, the Poisson S/N is, as expected, 141, while the actual S/N is 122 -- better, but less than a root 2 improvement.


Figure 1 shows a portion of the net spectrum (counts/s/pixel) of the DB wd WD0308-565 obtained as part of the calibration 12426. The spectrum is the output of the iterative algorithm, contains all of the data for that region, and is binned by 3 pixels (half a resolution element). Also shown are the positions of 3 S II interstellar lines, and a quadratic fit to the line free region 1220 < lambda < 1250 Å.

Figure 2 shows the square root of the total number of counts in each 3 pixel bin (the expected Poisson S/N), over the region and a scaled version of the square root of the fit shown in Figure 1. Features in this plot correspond to lines in the spectrum or grid wire shadows, where the total counts decreased.

Figure 3 shows the normalized spectrum in the region of the S II lines and demonstrates how well weak lines can be identified.