XMM-Newton - EPIC PN - Background correction

The conventional way to create a background spectrum is to select a region from the same observation where there is no emission from the source. Additionally, the region should be close to the source. This way, the spectral background should have the same characteristics as the background at the source region. For moderately extended sources that do not cover the entire field of view, a region which suffices the first criterion may be found at the border of the field of view. The second criterion, however, may not be satisfied. The background region may show a different detector background, and additionally the vignetting may be different.

In the case, where we are interested in determining the characteristics of emission with low surface brightness, which extends over a large region, where the background is (probably) the dominant component, we need a very accurate estimate of the background.

Here we describe a method to use a local estimate of the sky background that takes properly into account vignetting and detector background issues. This method has been used to analyse the diffuse emission in the Starburst Galaxy NGC 253 (Bauer, M. et al 2007, astro-ph/0711.3182).

To remove the detector background, archival observations are used which were taken in the same mode as the source observation, but where the filter wheel was closed. To avoid effects due to changes in the detector settings, or changes of the detector performance due to other reasons, a closed observation should be chosen to be as close as possible in time to the source observation. To ensure, that there are as little as possible differences between the source observation and the closed observation, bad columns and bad pixels should be removed both in the source and closed observation. Additionally, the closed observations may have been taken when the spacecraft was exposed to a different particle radiation level than the one present during the source observations. The XMM-Newton house keeping file (<revolution number>_<obs id>_PNX00000PMH.FIT, part of the ODFs) contains information on how many CCD rows per time unit were rejected due to a possible minimum ionising particle (MIP) event, which is a direct estimator of the average radiation level. These values can be used to rescale the count rate of the closed observation.

Out-of-Time spectra from the source and background region are used to correct for contribution from Out-of-Time events. When one subtracts a closed observation spectrum from a Out-of-Time corrected spectrum, one actually removes the Out-of-Time spectrum of the detector background twice. This is corrected in our method by adding again the Out-of-Time spectra of the detector background.

The background region spectrum is corrected for Out-of-Time events and the detector background, and the vignetting correction is applied in each energy bin as a function of off-axis angle of the source and background spectrum. This gives the sky background spectrum.

In all of the above steps, different exposure times and areas in the extraction regions have been accounted for. Since some of the components in the final background spectrum do have low number statistics, the conservative approximation to Poissonian errors σ_N≈ 1+√0.75+N (Gehrels, N. 1986, ApJ, 303, 336) is used. To avoid unjustified large errors, the spectrum is binned before calculating errors. The resulting background subtracted spectrum then has a significance in each bin of a least 3σ. The errors are propagated properly and are included in the file with the final background spectrum. This spectrum can be used with XSPEC as a background spectrum.

The whole method can be summarised by Eq. 1

(1)

with the following symbols:

B(E) is the counts at energy E in the background spectrum
B_obs(E) is the counts in the source observation
S_det(E) is the counts from the detector background spectrum in the source region
B_det(E) is the counts in the detector background spectrum in the background region
S^OOT_obs(E) are the counts in the Out-of-Time spectra in the source region
B^OOT_obs(E) are the counts in the Out-of-Time spectra in the background region
S^OOT_det(E) are the counts in the Out-of-Time detector background spectra in the source region
B^OOT_det(E) are the counts in the Out-of-Time detector background spectra in the background region
t_obs is the exposure time in the source observation
t_det is the exposure time in the closed observation
R_obs is the rejected line counter values in the source observation
R_det is the rejected line counter values in the closed observation
A_S is the area in the source region
A_B is the area in the background region
V(E,θ_S) is the vignetting value in the source region, depending on the offset angle θ and the energy E
V(E,θ_B) is the vignetting value in the background region, depending on the offset angle θ and the energy E
f is the fraction of Out-of-Time events in the corresponding mode of the observation

A comparison between this new method and the conventional method, that does not use the vignetting correction nor the closed observations, is shown in Fig. 1 for two example spectra from NGC 253. Region 7 covers a low surface brightness area in the northwestern halo. Region 14 is located in a high surface brightness region in the disk of NGC 253. The single background components in the source and background region are shown in Fig. 2. All figures show counts integrated over the extraction region. The counts in the background region were rescaled to the source region area to be able to compare them to the source spectrum. Also, the counts in the closed observation were rescaled to the exposure time and radiation level in the source observation.

Figure 1: Comparison between the background substraction on two examples: the low surface brightness NGC 253 region 7 (left) and the high surface brightness NGC 253 region 14 (right). The top panel shows the new method, as described in this work, the bottom panel shows the conventional method, where the raw background spectrum is used, and a correction for Out-of-Time events has been applied.

The differences between the new and the conventional method in terms of the resulting best fits are the following: In the majority of the tested cases, an additional power law component with Γ≈0 is required for the fit in the spectrum, obtained with the conventional method. The temperatures are consistent between both methods, but the resulting flux levels in the conventional method are higher. Differences in total flux values can be up to ~22%. The effect between the two methods is highest in regions with low surface brightness. Here the background dominates and a correct treatment is crucial. As an example, the difference in flux in region 7 (low surface brightness) is 22%. Whereas in region 14 (high surface brightness), the difference is 3%.

Figure 2: The single components that are part of the total background spectrum compared to the source spectrum. (left): Region 7, (right): Region 14, (top): components from the source region, (bottom): components from the background region. The single components were corrected for areas, exposure time, and radiation level, with respect to the source spectrum in the source region, but no vignetting correction was applied yet.

Available scripts: PNbackground.tgz
For the usage of this idl program see the documentation.txt in the PNbackground.tgz archive.