A back-propagation network for analog signal separation in high noise environments

A backpropagation network is compared with principal components regression and prefiltered linear regression to demonstrate its ability to separate overlapped analog signals in high noise environments. Specifically, the signals tested were synthetically generated chemical mixture spectra that simulate the type of data obtained from chromatography and photospectrometry. The individual spectra are heavily overlapped, and thirty percent random noise and a random DC has been added to them. The comparisons were made for data sets composed of two, three, and four overlapping spectra. >