[Previous] | [Session 16] | [Next]
Y. Yu (Dept. of Statistics, Harvard Univ.), D. Van Dyk (Dept. of Statistics, UC Irvine), A. Siemiginowska, P. Freeman, A. Zezas, V. Kashyap (Smithsonian Astrophysical Observatory, Harvard Univ.), A. Connors (Eureka Scientific)
CCD detectors such as Chandra's ACIS suffer from the problem of event pileup when observing bright sources. When multiple photons arrive at the same location on the detector within a frame readout time, the detector electronics cannot distinguish them as separate events, and consequently they are counted as a single event of higher energy, or discarded because of a non-standard charge pattern, or lost altogether if the total energy goes above the on-board discriminators. Thus, for bright sources pileup can seriously distort both the count rate and the spectrum in an irreversible way. However, while the changes are non-deterministic, they can be dealt with within a statistical framework. We have developed a highly structured statistical model that directly incorporates the mechanism of the instrument that results in pileup. We adopt the Bayesian approach and use state-of-the-art computing algorithms such as MCMC (Markov chain Monte Carlo) algorithms. This formulation allows us to incorporate the pileup model in a general spectral model that takes into account various source features, such as continuum, emission lines, interstellar absorption, as well as the effective area of the instrument, instrument response, and background contamination. We can also handle the spatial dimension of the problem by simultaneously fitting counts from several detection islands that suffer varying degrees of pileup. This strategy allows us to include data from regions of the detector with low pileup in our analysis. Even when the counts in these regions are relatively low, these data can be quite informative as to the shape of the spectrum, and thus, as to the amount of pileup in more heavily piled regions of the detector. We present results of our method for spectral analysis of the X-ray point source PKS2126. Residual plots for counts in several detection islands are used to check the self-consistency of the analysis.
The author(s) of this abstract have provided an email address for comments about the abstract: yu@stat.harvard.edu
[Previous] | [Session 16] | [Next]
Bulletin of the American Astronomical Society, 36 #3
© 2004. The American Astronomical Soceity.