XMM-Newton Users Handbook

3.3.9 EPIC photon pile-up

Photon pile-up, i.e., the arrival of more than one X-ray photon in one camera pixel or in adjacent pixels before it is read out, can affect both the PSF and the spectral response of EPIC.

Table 3 provides estimates of count rates for the different EPIC instrument modes for which pile-up should not be a problem. For the MOS full imaging mode, e.g., ca. 0.7 counts/s should not be exceeded. Note, that provided figures in fact are depending on the source spectrum and pile-up can become an important effect at even lower count rates for sources with very steep (i.e. soft) spectra. On the other hand at larger off-axis position pile-up is less likely.

The PSF is influenced by pile-up, because in the core of the PSF many photons arrive at almost the same time (within one readout frame), creating multi-pixel photon patterns which, for the MOS camera, are then rejected by the onboard event reconstruction software (which suppresses large size events, such as that induced by most of the cosmic rays). For the pn camera, event reconstruction is performed offline in the SAS. The pile-up effect leads, in the most extreme case, to a PSF with an artificial ``hole'' at its centre, as displayed in Fig. 38.

Figure 38: In-orbit observations performed with EPIC MOS showing the increase of pile-up with increasing photon count rate per frame. The panels are arranged clockwise, with the lowest count rate (and thus pile-up rate) in the upper left and the highest in the lower left. The observed count rates are 2, 5, 12 and 16 counts/frame, respectively.

The spectral response is compromised, because the charge deposited by more than one photon is added up before being read out, thus creating artificial ``hard'' photons where there have actually been two or more soft photons.

Pile-up effects on the spectral shape are very much reduced by selection of mono-pixel events. The effect of pile-up in this case is then essentially a loss of flux. Also, in case of very strong pile-up, building spectra and calculating absolute fluxes is still possible with a good accuracy by excising the heavily piled-up core (see below) from the analysis.

To get an idea on how important the pile-up effect can be for spectral analyses, a k$T$ = 1 keV Raymond-Smith model spectrum with $N_H\ = 3\times10^{20}$ cm$^{-2}$ has been fed into SciSim and the fit to the resulting ``observed'' spectrum compared with the input for EPIC MOS observations in the full window mode (Tab. 5).

Table 5: The effect of pile-up on spectral fits

Input flux$^1$	Output count rate [s$^{\bf -1}$]	Counts per MOS frame	N$_{\bf H}$ [10$^{\bf 20}$ cm$^{\bf -2}$]	kT [keV]	Norm./ expect
4.05	1.08	0.22	3.0	0.967	1
13.4	3.56	0.71	3.05	0.972	1
40.5	10.5	2.1	3.2	0.979	0.98
134	33.3	6.65	3.5	1.010	0.95
405	85.4	17.1	4.1	1.022	0.80

Notes to Table 5:
1) 0.15-10 keV flux in units of [10$^{-12}$ erg s$^{-1}$ cm$^{-2}$].

Figure 39: The best-fitting power law slope, $\alpha$, for an $\alpha = 1.7$ input spectrum into SciSim, with different input count rates, leading to different levels of pile-up.

One can see how the best-fitting $N_H$, k$T$ and normalisation vary as a function of input source flux (and thus count rate per frame). Note that, because of the photon pile-up, one apparently loses soft photons (whose charge combines and is then seen at higher energies), thus requiring a higher $N_H$ and also a higher k$T$ to reach the minimum $\chi^2$ for the fit. The peculiarity that for low fluxes the fitted k$T$ is a bit lower than the input value is explained by the fact that the SciSim simulation was performed without event reconstruction. What is important to note is the relative change in k$T$. A similar change is also caused in the best-fitting slope of a power law, as displayed in Fig. 39. The effect of pile-up on the PSF has been taken into account in these simulations by choosing an appropriately large photon extraction region for the spectral analysis.

Note: No error bars are provided in Tab. 5, because the input spectrum is a numerical model spectrum without noise. Differences in the output ``fit'' spectrum with respect to the noise-free input occur only because of re-grouping of photons due to pile-up. The relevance of the effect depends on the scientific goal of the observing programme. For up to 2 counts per MOS frame, the error in N$_H$ stays below 10%, that of k$T$ and the normalisation at a level of 2%, which in many cases will be below the uncertainty of the spectral fit anyway.

The SAS task epatplot can be used as a diagnostic tool for pile-up in the EPIC cameras. It utilizes the relative ratios of single- and double-pixel events which deviate from standard values in case of pile-up. Fig. 40 shows the produced plot for a source with a countrate of $\sim14$ counts/s in pn full frame mode.

Figure 40: Plot of the pn pattern distribution with energy as produced by the SAS task epatplot (see text for further details).

As input an eventfile was used which contains all events extracted from a circle with radius 1 arcmin around the source position. In the upper panel the spectral distributions as function of PI channel for single-, double-, triple- and quadruple-events (these are the four types of valid events which can be created by an X-ray photon) are plotted. Invalid patterns are also shown to indicate how much energy is lost by pattern pile-up.

Doubles can only be produced when the energy of both events is above the event threshold. Similarly triples and quadruples start at 3 and 4 times the threshold, respectively. In the lower panel the fraction of the four valid event types are plotted (relative to the sum). The ratios are energy dependent with higher energy photons producing less singles due to the increasing size of the electron cloud. As long as the count rate is below the pile-up limit the energy dependence follows standard curves. These are plotted for singles (1.0 at energies below twice the threshold and then decreasing with energy), doubles and singles+doubles. The curves are also used in the detector response matrices to correct for the pattern selection used in the spectrum to be analyzed. In the case of pile-up, the ratios deviate from the standard curves as the example shows. Due to `pattern pile-up' more doubles (which are actually two neighboring singles) are produced.

A quantitative comparison of both types of EPIC camera with the Chandra ACIS-I instrument in this respect is provided in § 3.7.1.3.

Users are referred to the following papers: Ballet (2003, AdSpR, 32, 2077) and Molendi & Sembay, (2003; http://xmm2.esac.esa.int/docs/documents/CAL-TN-0036-1-0.ps.gz) for further information. In particular, the former includes an alternative method for pile-up correction.