Background subtraction for low-loss transmission electron energy-loss spectroscopy.
We present a technique for removing the zero-loss background from electron energy-loss spectra at very low energies (down to approximately 2eV), generating results that are superior in a number of ways to the results of standard Fourier deconvolution techniques. Our technique is based on a separately measured background spectrum which is spline-interpolated and matched to the zero-loss peak in the low-loss spectrum using curve-fit techniques. The data points are weighted with the use of a semi-empirical model of the random error in the data produced by a spectrometer. We demonstrate in tests on real-world data that this model accounts for the random error within the energy range of interest. We discuss practical details of implementation and present detailed comparisons of the results of various algorithms on a piece of test data obtained from a carbon nanotube sample. Compared to the standard techniques, our algorithm tends to be more consistent, less dependent on arbitrary parameters, and better able to quantify spectral features with small signal-to-noise ratios, particularly those at very low energies.