An improved method to reduce noise in nuclear spectral data (paper no.D7)

Reduction of noise is important in nuclear spectra processing as it tends to get amplified during deconvolution. The noise power has unfortunately dependence on the signal amplitude in a spectrum. This paper describes a method with formulation that by applying square root function to the spectrum after adding a dc of value 3/8 makes noise very nearly independent of the signal amplitude. It is suggested that a popular Savitzky's FIR filter should be used with the preprocessed data. It is also suggested that a method should be tested over a large ensemble of independent data sets to establish its superiority. The above method and two other methods were applied on two spectral peaks. For each peak an ensemble of 50 computer generated data sets were used. It was observed that the described method showed improved performance and the improvement was more marked for the more developed peak. (author). 8 refs., 1 fig