A comparative study of deconvolution methods for gammaray spectra.
Abstract
Estimating an incident spectrum from observed count rates is generally a complex problem, because the solution is unstable with respect to the measurement errors. When information about the signal is available, one can simply proceed by "modelfitting"; the spectrum is modelled, leading to a solution which depends on a few parameters only. In general, this method does not allow to introduce the number of degrees of freedom required for a complete description of the entire structure in the signal. Because the solution does not have the appropriate number of degrees of freedom, this simplification leads to false confidence limits on the signal. That is why deconvolution methods must be used, and these methods differ from the previous ones in the way that each point of the unknown incident spectrum is a parameter. We compare the performances of various spectral deconvolution methods for data measured by Xand gammaray detectors. In our study, the data are the count rates, related to an incident spectrum by a complex transfer function. The signaltonoise ratios are small, and the dynamic range of the incident spectrum is very large leading to a tricky problem. Most deconvolution methods work best when the signaltonoise ratios are fairly large, e.g. 3σ detection in each energy channel. When detection per channel becomes small, only methods which impose some constraints on the range and shape of the spectrum remain performant.
 Publication:

Astronomy and Astrophysics Supplement Series
 Pub Date:
 October 1995
 Bibcode:
 1995A&AS..113..167B
 Keywords:

 GAMMA RAYS: OBSERVATIONS;
 XRAYS: GENERAL;
 METHODS: DATA ANALYSIS