A Statistical Method for Identifying Lines in Wideband Spectra
Abstract
We present a method and model results for determining the probability that a spectral line is present within a set of data. In wideband spectroscopy the noise across a spectrum varies significantly with frequency, complicating the usual methods for calculating noise. As an alternative to determining the noise across frequency bins in an averaged spectrum, it is possible to analyze the time sequence of signals to estimate the mean and uncertainty in the mean for each bin. A detection is then defined as a channel or set of channels whose mean is statistically significantly higher than those of its neighbors, making it possible to construct a plot of the confidence that a line is detected versus frequency. Calculating amplitude uncertainties for each channel also establishes errors on line intensities regardless of gain and noise variation with frequency. This method of spectral analysis is useful for wideband spectrometers such as the Zpectrometer (Harris, Baker, & Jewell 2004), which will be installed on the Green Bank Telescope next year.
 Publication:

From ZMachines to ALMA: (Sub)Millimeter Spectroscopy of Galaxies
 Pub Date:
 October 2007
 Bibcode:
 2007ASPC..375..299Z