SAS watchout page

Can I estimate the pile-up level in my observation?

Pile-up (the accumulation of more than one photon in the same pixel or in neighboring pixels during the time a CCD frame is exposed) can lead to flux losses and spectral distortion. It is very important to estimate the pile-up fraction in your own data, before you perform any spectral analysis. Pile-up fractions lower than 5% have normally a negligible impact on the spectral results.

Within SAS, a tool is available to check the pile-up fraction in the pn data: epatplot. This tool compares the observed pattern distributions (derived from a pn calibrated and concatenated even list) with the calibrated models. The latter are instrument mode dependent.

An example of the epatplot output (as in the SASv5.3 implementation) is shown in Fig.1

Fig.1: the SAS v5.3 epatplot output in the case of a strong pile-uped source

The upper panel shows the energy spectra. Different colors correspond to different pattern choices: singles (red), doubles (blue), triples (green), quadruples (ultramarine). The lower panel is the key element for the evaluation of the pile-up fraction: it shows the observed distribution of pattern as a function of energy (PI channel in Analog-Digital Units), using the same color code. A continuous line, superimposed to the histograms, represents the expected distribution. In the example above, the remarkable difference between observation and model indicate that the pn data are substantially affected by pile-up along the whole on energy bandpass.

The only way to reduce pile-up in a given dataset is to excise the core of the PSF, up to a radius where the pile-up fraction becomes negligible. The effect of a 4 arc-minutes excision on the same observation is shown in Fig.2. The agreement between observed and theoretical distribution is much better. Spectra extracted from annuli around the source, excluding the innermost 4' are pile-up free.

Fig.2: the same as Fig.1, if 4' around the PSF core are excised.

The user must be aware that differences between the observed and expected pile-up distribution can occur both in the softest and hardest energy band, due to the effects unrelated to pile-up:

• high background levels may yield a noise peak at low energies.
• increased level of Minimum Ionizing Particles (MIPs), which are not fully suppressed by the on board software algorithm, may yield a deviation from model at high energies (approximately >7.5 keV).

As a baseline, the user shall use in a spectrum accumulation events selected in an energy range where the agreement between the observed and the theoretical pattern distribution is good, no matter what the origin of the deviation is. The users are recommended to test several extraction radii, to find the best compromise between signal-to-noise ratio and pile-up minimization in their own data.

Two more caveats shall be born in mind:

• the running version of epatplot must be in line with the version of the Charge Transfer Efficiency correction. In other words, the epatplot version in use shall match the e[mp]chain/e[mp]events version. This condition is automatically fulfilled, if public SAS versions are employed
• the dependence of the pattern distribution as a function of the source distance from the readout node was not implemented in SAS v5.2. The algorithm was optimized for a source located at the nominal boresight position (i.e.: RAWY=190 for pn).
• models for pn modes different from Full Frame are available only starting from SASv5.3.